THE PLANNING AND OPTIMISATION OF DVB-H RADIO NETWORK

Transkriptio

1 Helsinki University of Technology Faculty of Electronics, Communications and Automation Department of Communications and Networking Teknillinen korkeakoulu Elektroniikan, tietoliikenteen ja automaation tiedekunta Tietoliikenne- ja tietoverkkotekniikan laitos THE PLANNING AND OPTIMISATION OF DVB-H RADIO NETWORK Jyrki T.J. Penttinen Dissertation for the degree of Doctor of Science in Technology to be presented with due permission for public examination and debate in Auditorium S4 at Helsinki University of Technology (Espoo, Finland) on the xxth of xxxx, 2011, at xx o'clock noon.

2 Distribution: Helsinki University of Technology Department of Communications and Networking P.O.Box 3000 FIN TKK Tel Fax Jyrki T.J. Penttinen ISBN xxx ISBN xxx (pdf) ISSN xxx ISSN xxx (pdf) Multiprint Oy Espoo 2011

3 AB ABSTRACT OF DOCTORAL DISSERTATION Author Jyrki Teppo Juho PENTTINEN HELSINKI UNIVERSITY OF TECHNLOGY P.O. BOX 1000, FI TKK Name of the dissertation The Planning and Optimisation of DVB-H Radio Network Manuscript submitted 21 May 2010 Manuscript revised 31 October 2010 Date of the defence Monograph Faculty Department Field of research Opponent(s) Supervisor Instructor Article dissertation (summary + original articles) Electronics, communications and automation Communications and networking Telecommunications Engineering Prof. Sven-Gustav Häggman Dr. Jukka Henriksson Abstract There are details in DVB-H (Digital Video Broadcasting Handheld) radio network planning that lack of the final consensus in the scientific field. The aim of this doctorate thesis is thus to investigate advanced DVB-H radio network planning and optimisation. The outcome includes results about measurement techniques, network coverage and quality estimation, technological and economical optimisation, and methodologies related to the error correction and single frequency network aspects. The results provide a set of DVB-H radio network planning and optimisation principles that can be applied for the further investigation of detailed parameter values of the radio link budget. There are also case studies presented that show the functionality of methods and typical performance values. Based on the comparative investigations, a process chart was created for the DVB-H radio network planning and optimisation. The identified process blocks can be applied to a typical DVB-H network deployment in the initial high-level phase as well as in detailed network planning and optimisation phase. Based on the process, most relevant items were selected for in-depth studies. The related investigations are presented in annexed Publications. The rest of the process steps were revised via comparative literature studies. The structure of the thesis follows thus the designed process charts. The main focus of this thesis is in the development of DVB-H radio network planning methodologies. One of the goals was to investigate the radio interface measurement data post-processing and analysis. This can be used for the correct selection of the related radio parameter values as function of the radio channel type. Another topic is related to the controlled handling of over-sized single frequency network areas by balancing the elevated SFN interference level with the SFN gain. A radio path loss simulator was created as a basis for these studies, and case results are presented as function of the relevant radio parameter values, transmitter power levels and site antenna heights both in theoretical as well as realistic network layouts. In addition to the above mentioned topics, this thesis also investigates selected EMC and human exposure safety zone calculations, radio coverage estimations and the balancing of technical radio parameters and network costs in order to complete the planning process steps. Keywords Mobile TV, digital video broadcasting, radio network design, SFN, MPE-FEC, single frequency network ISBN (printed) ISBN (pdf) Language ISSN (printed) ISSN (pdf) Number of pages Publisher Print distribution The dissertation can be read at

4

5 AB VÄITÖSKIRJAN TIIVISTELMÄ Tekijä Jyrki Teppo Juho PENTTINEN TEKNILLINEN KORKEAKOULU PL 1000, TKK Väitöskirjan nimi DVB-H verkon suunnittelu ja optimointi Käsikirjoituksen päivämäärä Korjatun käsikirjoituksen päivämäärä Väitöstilaisuuden ajankohta Monografia Tiedekunta Laitos Tutkimusala Vastaväittäjä(t) Työn valvoja Työn ohjaaja Yhdistelmäväitöskirja (yhteenveto + erillisartikkelit) Elektroniikan, tietoliikenteen ja automation tiedekunta Tietoliikenne- ja tietoverkkotekniikan laitos Telecommunications Engineering Prof. Sven-Gustav Häggman TkT Jukka Henriksson Tiivistelmä DVB-H (Digital Video Broadcasting Handheld) radioverkon suunnitteluperiaatteissa on vielä yksityiskohtia, jotka eivät ole täysin selkeitä tieteellisessä yhteisössä. Tämän väitöskirjan tavoitteena on siksi kehittää DVB-H verkon suunnitteluja optimointimenetelmiä. Työ sisältää tutkimuksia mittaustekniikoista, radioverkon peitto- ja laatuarvioinneista, teknistaloudellisista optimointiperiaatteista sekä virheenkorjauksen ja yksitaajuusverkon toimivuudesta. Työn tuloksena on tarkennettuja DVB-H radioverkon suunnittelu- ja optimointimenetelmiä, joita voidaan soveltaa yksityiskohtaisessa radiolinkkibudjetin tutkinnassa ja parametriarvojen valinnassa. Väitöskirjassa on myös esimerkkitutkimuksia, jotka osoittavat esitettyjen menetelmien käytettävyyden ja mallituloksia tyypillisiä parametriarvoja käytettäessä. Työn pohjaksi on luotu DVB-H radioverkon suunnittelu- ja optimointiprosessikaavio vertailevaan aihepiirin selvitykseen perustuen. Prosessin osioita voidaan soveltaa tyypillisessä DVB-H verkon käyttöönotossa niin alustavassa kuin tarkennetussa verkon suunnittelussa ja optimoinnissa. Prosessista valittiin merkittävimmät osiot työn tutkimuskohteiksi. Niihin liittyvät osajulkaisut on esitetty väitöskirjan liitteinä. Prosessikaavion muita lohkoja on selvennetty vertailevilla kirjallisuustutkimuksilla. Väitöskirjan rakenne seuraa siten esitettyä prosessikaaviota. Työn pääasiallinen näkökulma liittyy DVB-H radioverkon suunnittelumenetelmien kehitykseen. Ensimmäinen päätutkimusaiheista liittyy radiorajapinnan mittauksiin, mittaustulosten jatkokäsittelyyn ja analyysimenetelmiin. Tätä tietoa voidaan käytää radiokanavaspesifisiä parametriarvoja valittaessa. Toinen pääaihe liittyy hallittuun yksitaajuusverkon (SFN, Single Frequency Network) rajojen kasvattamiseen tasapainotellen SFN-häiriöitä ja SFN-vahvistusarvoja. Tämän osion perustaksi työssä luotiin radiorajapinnan yhteysvaimennuksen arvioiva SFN-simulaattori, jonka avulla tutkittiin esimerkkitapauksia varioiden lähetinasemien tehotasoa ja antennikorkeutta niin teoreettisissa kuin käytännön ympäristöissä tyypillisillä radioverkon parametriarvoilla. Suunnitteluprosessin täydentämiseksi väitöskirjassa esitetään myös häiriö- ja turvaetäisyyden laskennan menetelmiä, vertaillaan radioverkon peittoalue-ennusteiden toimivuutta sekä tutkitaan teknisten radioparametrien ja verkkokustannusten tasapainottelua. Asiasanat Matkaviestin-TV, digitaalinen videolähete, radioverkkosuunnittelu, SFN, MPE-FEC, yksitaajuusverkko ISBN (painettu) ISBN (pdf) Kieli ISSN (painettu) ISSN (pdf) Sivumäärä Julkaisija Painetun väitöskirjan jakelu Luettavissa verkossa osoitteessa

6

7 ix Preface The idea of this thesis took a concrete form after I had been working in various DVB-H related projects, including pre-finpilot in Helsinki, Finland with TeliaSonera during 2004, and technical DVB-H trialling in USA with Nokia and Nokia Siemens Networks during 2007, among the other smaller scale DVB-H radio network planning tasks in international level. I also had chance to participate in the specification work of DVB-H Technical Module in As a result of these activities, I started to consider selected study items about radio network planning and optimisation of DVB-H. As the main idea was to carry out an overall revision of the DVB- H planning, it was logical to divide the tasks into separate study items, process the work in the format of sub-publications and write the thesis as an article dissertation. The purpose of this thesis is to identify and study the most relevant items that have effect on the radio planning of DVB-H. For the basis of the revision, I designed a planning and optimisation process chart. The actual investigation work is based on the DVB-H field tests, simulations and literature studies about the radio network planning. I express my warmest gratitude to my supervisor Professor Sven-Gustav Häggman and my instructor Dr. Jukka Henriksson for the most fruitful discussions and comments related to the work. I thank my reviewers Professor Zander and Professor Jukka Lempiäinen for the prompt reviews and most useful comments. I thank my colleagues at Nokia, Nokia Siemens Networks and TeliaSonera for the discussions related to various details of the DVB-H. I want to express my special thanks to Pekka H. K. Talmola, Juha Viinikainen, Jouni Oksanen, Pekka Pesari, Anssi Korkeakoski, Francesco Calabrese and Paul Hartman, among all my colleagues that have kindly shared professional experiences and ideas with me during this work across the organizations. I would like to express my gratitude to Nokia Siemens Networks for the support which was more than helpful as I made the actual thesis and related sub-publication set in my spare time. I give a special thanks to the Foundation of the Finnish Society of Electronics Engineers (Elektroniikkainsinöörien säätiö), Ulla Tuomisen säätiö (the Foundation of Ulla Tuominen) and Nokia Foundation for the scholarships during the most hectic and economically challenging time of the thesis. Finally, I want to thank my close family for all the support during the work. Madrid, Spain, 31 October 2010 Jyrki T.J. Penttinen

8 x

9 xi Contents Preface... ix Contents... xi List of Publications... xv Author's contribution... xvii List of Abbreviations... xix List of Symbols... xxv List of Figures... xxxi List of Tables... xxxvii 1 Introduction DVB-H scene DVB-H radio network planning Problem identification Radio network planning process Cost impact Single frequency network performance Field test and analysis methodology Coverage planning Objectives of the thesis Publications and contributions Thesis outline Principles of DVB-H Description DVB-H network... 16

10 xii Architecture Functionality Elements Radio Functionality Radio transmitter Terminal Frequency OFDM Parameters Error recovery Time Slicing SFN / MFN Interaction channel Initial radio network planning process High-level network dimensioning process Capacity Planning Coverage and QoS Planning [Publications V, VII] Link budget Propagation models Safety distance [Publication VI] Cost prediction [Publication I] Detailed radio network design Identifying the planning items Detailed network planning process Capacity planning Coverage planning [Publications V, VII] Effect of site locations Site cell range predictions Local measurements [Publications II, VIII, IX] Coverage area... 66

11 xiii Error correction Accuracy of error correction analysis Effect of SFN [Publications III, IV, X] Non-interfered network Interfered network Conclusion Radiation limitations [Publication VI] Cost prediction [Publication I] Optimisation Site parameters Controlled SFN interference [Publications III, X] SFN Gain vs. Interferences [Publications IV, VII, X] MPE-FEC Gain [Publication VIII] Antenna down-tilt [Publication V] User Experience Related Parameters Cost Optimization [Publication I] Cost optimization in the non-interfered network Cost optimization in interfered SFN network Summary and conclusions Main results of this thesis Usability of the results Further study items References Errata Appendix A: SFN Simulator Appendix B: Publications I X

12 xiv

13 xv List of Publications This thesis contains a summary of the following Publications which are referred to in the text by their Roman numerals. These Publications were presented in conferences or journals during I II III IV V VI VII VIII CAPEX and OPEX optimisation as function of DVB-H transmitter power. The Third International Conference on Digital Telecommunications ICDT International Academy, Research, and Industry Association (IARIA). June 29 July 5, Bucharest, Romania. 6 p. Field measurement and data analysis method for DVB-H mobile devices. The Third International Conference on Digital Telecommunications ICDT International Academy, Research, and Industry Association (IARIA). June 29 July 5, Bucharest, Romania. 6 p. The simulation of the interference levels in extended DVB-H SFN areas. The Fourth International Conference on Wireless and Mobile Communications ICWMC International Academy, Research, and Industry Association (IARIA). July 27 August 1, Athens, Greece. 6 p. The SFN gain in non-interfered and interfered DVB-H networks. The Fourth International Conference on Wireless and Mobile Communications ICWMC International Academy, Research, and Industry Association (IARIA). July 27 August 1, Athens, Greece. 6 p. DVB-H coverage estimation in highly populated urban area. The 58th Annual IEEE Broadcast Symposium, October Alexandria, VA, USA. 6 p. DVB-H radiation aspects. IEEE International Symposium on Wireless Communication Systems ISWCS 2008, October Reykjavik, Iceland. 5 p. DVB-H performance simulations in dense urban area. The Third International Conference on Digital Society ICDS International Academy, Research, and Industry Association (IARIA). February 1 7, Cancun, Mexico. 6 p. MPE-FEC performance as function of the terminal speed in typical DVB-H radio channels. IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB) May. 2009, Bilbao, Spain. 6 p. The author's share of this publication is 50 %.

14 xvi IX X Field measurement and data analysis method for DVB-H mobile devices. International Journal on Advances in Systems and Measurements. International Academy, Research, and Industry Association (IARIA). ISSN x, vol. 2, no. 1, year 2009, pages The SFN gain in non-interfered and interfered DVB-H networks. International Journal on Advances in Internet Technology. International Academy, Research, and Industry Association (IARIA). ISSN , vol. 2, no. 1, 2009, pages

15 xvii Author's contribution The author has contributed solely all the contents of the Publications I VII and IX X, including the creation of the related simulator explained in Appendix 1. For Publication VIII, the author contributed 50 % of the contents which included the theoretical revision, laboratory and indoor live network measurements and data collection, as well as the respective analysis and conclusions. The summary of the above mentioned Publications is presented in Chapters 4 6 and all the Publications are annexed in originally presented format. The contributions of this thesis can be divided into the following high-level areas: Studies about the DVB-H radio network planning and optimisation. The study items consist of the theoretical DVB-H coverage area calculations with comparisons of the results obtained via operational network planning tool. There are also studies about the EMC (Electro-Magnetic Compatibility) and safety distances presented. This part clarifies the DVB-H network planning methods compared to the typically presented like [Bro02], [Gre06], and [Ung06], and creates a general basis for the proposed network planning process charts. Clarified method for the field measurements with hand-held DVB-H terminal and respective method for the analysis in order to obtain sufficiently in-depth information about the quality level of the DVB-H radio network. This part deepens the principles and confirms the results for the MPE-FEC that can be found in the most relevant related sources [Dvb09], [Apa06a], and [Apa06b]. Based on the presented measurements and analysis, the radio link budget can be adjusted accordingly depending on the special characteristics of the investigated environment. Creation of the SFN simulator based on the Monte-Carlo method and execution of physical radio propagation related simulations for the investigations of the performance of the single frequency network in order to obtain the geographical and cumulative distribution of the SFN interference and SFN gain levels of DVB-H. The simulations presented in the thesis provide more detailed information about the suitable parameter settings of the radio network, including the transmitter antenna heights and transmitter power levels, compared to the other related theoretical or practical studies identified by the production of the Publications III, IV and X. [Bac04] [Bee07] [Ble09] [Sil06] The presented simulations can be utilized in the theoretical studies for the approximation of the optimal parameter settings in uniform site cell layout cases, as well as in in-depth phase of the network planning with varying site locations, power levels and antenna heights.

16 xviii The outcome of the work is an overall revision of the DVB-H radio network planning, with a selected set of details that clarifies the radio network planning processes by presenting investigation methods and case performance values that either confirm or gives new point of view to the already existing information.

17 xix List of Abbreviations 16-QAM 2D 2K 3D 3G 4K 64-QAM 8K AAA Quadrature Amplitude Modulation (with 16 constellation points) Two-dimensional FFT mode of OFDM consisting of Three-dimensional Third Generation FFT mode of OFDM Quadrature Amplitude Modulation (with 64 constellation points) FFT mode of OFDM Authentication, Authorisation and Accounting AAC ADSL ADT AWGN BMCOFORUM CAPEX CBMS CDF CELTIC CINR COFDM Application Data Table Additive White Gaussian Noise Broadcast Mobile Convergence Forum Capital Expenses Convergence of Broadcasting and Mobile Services Cumulative Distribution Function Services to Wireless, Integrated, Nomadic, GPRS, UMTS & TV handheld terminals. Carrier and Interference to Noise Ratio, C/(N+I) Coded Orthogonal Frequency Division Multiplex

18 xx CR CRC DTTV DVB DVB-C DVB-H DVB-NGH DVB-S DVB-T EIRP EMC ERP ES ESG ETSI EU FEC FER FFT GI GPS GSM HP Code Rate Cyclic Redundancy Check Digital Terrestrial TV Digital Video Broadcasting Digital Video Broadcasting, Cable Digital Video Broadcasting, Handheld Digital Video Broadcasting, Next Generation Handheld Digital Video Broadcasting, Satellite Digital Video Broadcasting, Terrestrial Effective Isotropic Radiated Power Electro-Magnetic Compatibility Effective Radiated Power Elementary Stream Electronic Service Guide European Telecommunications Standardisation Institute European Union Forward Error Correction Frame Error Rate Fast Fourier Transform Guard Interval Global Positioning System Global System for Mobile communications High Priority stream

19 xxi HW I/Q IMSI IP IPDC IPE ITU-R LAN LOS LP LTE MBMS MER MFER MFN MMS MIMO Motivate MPE MPE-FEC Hardware In phase (I) and Quadrature (Q) signal components International Mobile Subscriber Identity Internet Protocol IP Datacast IP Encapsulator International Telecommunications Union, Radio section Local Area Network Line-of-Sight Low Priority stream Long TERM Evolution Mobile Broadcast / Multicast Service Modulation Error Rate Frame Error Rate after MPE-FEC Multi Frequency Network Multimedia Messaging Service Multiple In Multiple Out (antenna) Mobile Television and Innovative Receivers project Multi-Protocol Encapsulation Multi-Protocol Encapsulation Forward Error Correction MPEG-2 Moving Pictures Expert Group 2 MS MUX Mobile Station Multiplexer

20 xxii N/A N-LOS OFDM OMS OPEX PDF PID Non-applicable Non-Line-of-Sight Orthogonal Frequency Division Multiplex Operations and Management System Operating Expenses Probability Density Function Program Identifier PSI PTM PTP QEF QoS QPSK RF RS RSSI RX SER SFN SI SIM SMS Stdev Point-to-Multipoint Point-to-Point Quasi Error-Free Quality of Service Quadrature Phase Shift Keying (with four constellation points) Radio Frequency Reed-Solomon coding Received Signal Strength Indicator Receiver SFN Error Rate Single Frequency Network Service Information Subscriber Identity Module Short Message Service Standard Deviation

21 xxiii TDMA TPS TS TX UHF UMTS VHF Time Division Multiple Access Transmitter Parameter Signalling Transport Stream Transmitter Ultra High Frequency band Universal Mobile Telecommunications System Very High Frequency band WiMAX WLAN Wireless Local Area Network

22

23 xxv List of Symbols α α σ Alpha parameter expressing the hierarchical constellation shape Angle of the position of the DVB-H terminal (rad) Standard deviation (db) φ k (t) OFDM sub-carrier in time domain π ϖ k Pi (constant) Frequency of the orthogonal sub-carrier k (Hz) A Area (km 2 ) A tot Area, whole investigated (km 2 ) A cell Area, site cell coverage (km 2 ) a(h MS ) LC a(h MS ) SMC B B ch Area type factor in Okumura-Hata path loss prediction model for large city Area type factor in Okumura-Hata path loss prediction model for small and medium city Receiver noise bandwidth (Hz) Bandwidth, channel (MHz) C common Costs of the site, common fixed items (currency unit) C site Costs, total of the DVB-H network (currency unit) C tot Costs, total of the DVB-H site (currency unit) C C tot Costs, total of the CAPEX (currency unit) O tot year C / Costs, total of OPEX per year (currency unit) C var iable Site dependent costs (currency unit) (cn) min Carrier to noise, minimum C/N requirement (db)

24 xxvi C C/(N+I) C/N (C/N) min C 1 C tot c e d D eff delta-t D eff D SFN Carrier level (db) Carrier to noise and interference level (db) Carrier to noise level (db) Carrier to noise level, minimum functional (db) Reference carrier (db) Sum of all carriers (db) Cost of electricity (currency unit / kwh) Maximum distance (km) Difference of the arriving signals (effective distance, km) Time for the beginning of next time sliced burst (s) Effective distance of the DVB-H signal in SFN Maximum allowed inter-site distance (diameter) within SFN area (km) d cell Radius of the site cell (km 2 ) E E min(in) E min(out) E in_average E out_average F F err_am F err_bm F r f Number of errors (positive integer) Field strength, minimum in indoors (dbµv/m) Field strength, minimum in outdoors (dbµv/m) Field strength, average in indoors (dbµv/m) Field strength, average in outdoors (dbµv/m) Noise figure (db) Erroneous frames after MPE-FEC (positive integer) Erroneous frames before MPE-FEC (positive integer) Total number of received frames (positive integer) Frequency (MHz)

25 xxvii f k Frequency of the sub-carrier number k f N OFDM carrier (order number N) FER Frame Error Rate (%) G ant G MPE-FEC G RX G SFN G TX h(t) h BS h MS I I tot k k k K l L ant L b L cc L GSM L lv Gain, mobile station antenna (dbi) Gain, MPE-FEC (db) Gain, receiver antenna (dbi) Gain, SFN (db) Gain, transmitter antenna (dbi) OFDM channel in time domain Height of the transmitter antenna of DVB-H site (m) Height of the DVB-H receiver (m) Interference level (db) Sum of all interferences (db) Boltzmann s constant (J/K) Number of data symbols per block (positive integer) Positive integer in SFN reuse pattern Formula (variable) Reuse factor in hexagonal site cell layout (positive integer) Positive integer in SFN reuse pattern Formula (variable) Loss, antenna (db) Loss, building (db) Loss, cable and connectors (db) Loss, GSM filter (db) Location variation (db)

26 xxviii L max L maxinterference L norm L open L pl(in) L pl(out) L ps L sub-urban L Rayleigh LND Loss, maximum path for the carrier (db) Loss, maximum path for the interfering signal (db) Loss, fading caused by the long-term variations (db) Loss, Okumura-Hata path loss prediction for open area Path loss, maximum in indoors (db) Path loss, maximum in outdoors (db) Loss, power splitter (db) Loss, Okumura-Hata path loss prediction for sub-urban area Loss, fading caused by the short-term variations (db) Distribution for the long-term fading (table variable) Locprob Location probability, area (%) MFER Frame Error Rate after MPE-FEC (%) m n n N ND Noisefloor N sites N cells N y P c P i Number of bits in symbol (positive integer) Total number of m-bit symbols in the encoded blocks (positive integer) Amount of samples (positive integer) Number of carriers in OFDM (positive integer) Distribution for fast fading (table variable) Noise floor including noise limit and receiver s noise figure (dbm) Number of the sites in SFN (positive integer) Number of the site cells of DVB-H network (positive integer) Number of the years (positive real number) Electrical power consumption (W) Power, isotropic (dbm)

27 xxix P min(in) Power, minimum required in indoors (dbm) P min(out) Power, minimum required in outdoors (dbm) P n Noise floor (dbm) P RX Received power level (dbm) P RXmin Sensitivity, receiver (dbm) P RXref Reference power level (dbm) P RX (C) Received useful carrier power level (dbm) tot P C Useful total received power (W) tot P I Total received power of the interfering components (W) P RX (I) Received interfering power level (dbm) P TX Transmitter output power (dbm) P TX,opt Optimal radiating power as a result of CAPEX / OPEX analysis (dbm) P TX,reg Maximum radiating power as a result of regulatory limits (dbm) P TX,safety Maximum radiating power as a result of EMC / radiation safety analysis (dbm) effective P TX Transmitter power after the filter loss (dbm) P tx Power level, radiating power in EIRP (dbm) R Sampling resolution (%) r Site cell radius of the carrier C (km) r interference Site cell radius of the interfering site (km) s(t) S T OFDM signal in time domain Number of erasures in the block (positive integer) Temperature (Kelvin)

28 xxx t t T F T GI T S T U x x x k Parameter in RS-coding (variable) Time (s) Time, frame duration (s) Time, Guard Interval duration (s) Symbol duration of OFDM (s) Time, useful symbol duration (s) Simulation area, x-coordinate point (km) Loss, average (db) Data symbol number k x ol Overlapping portion of the site cells in x-axis (%) x1km x2km xkm y y1km y2km ykm Simulation area, x-coordinate, upper left corner point (km) Simulation area, x-coordinate, lower right corner point (km) Simulation area, total length of x-axis (km) Simulation area, y-coordinate point (km) Simulation area, y-coordinate, upper left corner point (km) Simulation area, y-coordinate, lower right corner point (km) Simulation area, total length of y-axis (km) y ol Overlapping portion of the site cells in y-axis (%)

29 xxxi List of Figures Figure 2-1. The DVB-CBMS architecture as interpreted from [Dvb09] Figure 2-2. The complete DVB-H delivery chain can be divided into core network (DVB-IPDC) and radio network (DVB-H) Figure 2-3. It is possible to multiplex the DVB-H IP streams with the DVB-T MPEGstreams. The streams can be delivered via the common infrastructure for the DVB-H and DVB-T terminals Figure 2-4. The high-level block diagram of DVB-H receiver. The RF signal is directed to the input of the demodulator, and output is further handled by the DVB-H terminal.18 Figure 2-5. The DVB-H reference receiver Figure 2-6. The frame structure of DVB-H Figure 2-7. OFDM is based on the orthogonal sub-carriers containing the useful data (marked with light grey colour) as well as pilot signals (marked with dark grey colour) Figure 2-8. The principle of the DVB-H constellations Figure 2-9. The DVB-H error protection scheme Figure The MPE-FEC frame consists of application data table for IP datagrams and RS data table for RS parity bytes Figure The principle of Time Slicing functionality Figure The correlation and combination of the separate multi-path signals can be done whilst all the components occur within the time delay determined by GI Figure The principle of the overlapping GSM/UMTS coverage areas Figure 3-1. The high-level cross-relations of the most relevant DVB-H radio network planning items Figure 3-2. The radio network planning process in the initial phase. The process contains high-level estimations of the capacity and coverage taking into account the economical and regulatory limitations Figure 3-3. A case example of typical building loss in format of cumulative RSSI histogram, which is obtained by the difference of the indoor and outdoor received power levels. The case represents deep building type of B2. The Figure is a result of postprocessing of the Figure 25 of Publication IX Figure 3-4. Example of the DVB-H ranges (radius) for selected cases, when 16-QAM, CR 1/2 and MPE-FEC 1/2 are applied. No SFN gain nor MPE-FEC gain are included Figure 3-5. An example of the transmitter cost depending on the power level... 54

30 xxxii Figure 4-1. The cross relations of the items affecting on the DVB-H radio dimensioning. In this diagram, the final aim is the balancing of the capacity and coverage by taking into account the restrictions and enhancements related to the technology, commercial and regulatory items Figure 4-2. The detailed planning phase contains in-depth analysis of the effects of the performance parameters. The steps can be highly iterative. In the operational phase of the network, a continuous optimisation may take place via the field testing and customer feedback Figure 4-3. Comparison of the RSSI display of 3 different DVB-H terminals used in Publication II. Cumulative distribution of the laboratory measurement with 90 samples per terminal Figure 4-4. An example of the collected received power levels by moving the receiver close to the functional cell edge. The received power level samples and corresponding geographical locations can be observed by post-processing the measurement data to a map format which indicates the expected coverage limits Figure 4-5. The first step of the proposed methodology arranges the occurred samples as function of the received power level, showing the amount of error-free instances, occurred instances with frame errors that could be corrected with MPE-FEC, and remaining erroneous frames after MPE-FEC Figure 4-6. The method shows the normalized number of occurred events per each RSSI value Figure 4-7. An amplified view to the 5% FER / MFER level Figure 4-8. An example of the collected data distribution over the RSSI scale Figure 4-9. Amplified view to the case result graph showing the FER and MFER curves with respective error margin calculated for the absolute values of the samples for each RSSI category individually Figure An example of individually received signal levels in the same area, and their total power based on the direct power summing. Same terminal was utilized in these measurements Figure CDF of the 3 individual and summed signals presented in Figure Figure An example of simulated area with SFN reuse pattern size K= Figure The SFN gain obtained via the simulations of Publication IV for the noninterfered network (QPSK, FFT 8K, GI 1/4). The comparable results in studied references are also presented Figure There are no interferences when the sites are within the SFN area even if the receiver drifts outside of the SFN area. The reception within this outer zone is thus possible when the minimum required C/N can still be reached Figure When the sites exceed the SFN area, there will be interferences in those areas where D eff > D SFN. Figure shows the interference zone that applies for D eff > D SFN

31 xxxiii everywhere with the I-component greater than the noise floor. In this case, P tx was +70 dbm and site antenna height 80 m. The area is 45 km 45 km Figure The distribution of the destructive interference. It can be seen that the useful field is sufficiently high close to the nearest site cell to avoid the destructive interferences, but otherwise this type of interference may occur anywhere within the interference zone Figure The geographical distribution of C/I when P tx = +70 dbm Figure The effect of the interference can be seen when the power level is set to +80 dbm in this specific case. The reduced C/I-level can be seen clearly behind the sites within the interference zone on left and right hand sides Figure PDF of the 2-site simulations (P tx = +70 dbm) Figure CDF of the 2-site simulations for P tx of +70 dbm. Also a comparison results of +60 dbm and +80 dbm sites are included Figure 5-1. The outcome of the simulations of the extended SFN area interferences Figure 5-2. An example of the balancing of SFN gain and SFN interference level in long-term fading channel. K represents the SFN reuse pattern size, i.e. the number of sites within the SFN area Figure 5-3. The simulation results of the SFN gain and interference balancing for the 16-QAM cases Figure 5-4. An example of the simulation results in Mexico City layout with parameter set of B=6MHz, QPSK modulation, CR=1/2, MPE-FEC=1/2, GI=1/32 and FFT=8K in a combined long-term and Rayleigh fading channel. The map in left hand site represents the carrier distribution with noise level as reference, and the right one the interference distribution. The site circles indicate the height of the antennas Figure 5-5. The C/(N+I) distribution of the previous example Figure 5-6. An example of the cumulative distribution of occurred frame errors as function of the received power level Figure 5-7. An example of the error recovery in case of the impulse noise Figure 5-8. Summary of FER5 and MFER5 analysis in a single site cell case Figure 5-9. Cost optimization process in non-interfered DVB-H network Figure Cost optimization in interfered DVB-H network Figure The cost effect of DVB-H network when the antenna height is varied. This case includes relatively high transmission costs as utilized in Publication I Figure An example of the cost effect as function of the antenna height when the transmission costs are assumed to be low Figure 6-1. PDF and CDF of the Rayleigh distribution Figure 6-2. The PDF of the comparative analysis of the power summing in squared and direct manner

32 xxxiv Figure 6-3. The CDF of the power summing analysis Figure 6-4. The comparative simulation results for the SFN gain in non-interfered environment Figure 6-5. The high-level simulator block diagram...a 1 Figure 6-6. Simulator s initiation phase....a 3 Figure 6-7. Principle of the combination of fast and long-term fading in the receiving end....a 4 Figure 6-8. PDF and CDF of the normal distribution representing the variations of longterm loss when the standard deviation is set to 5.5 db...a 5 Figure 6-9. Example of the transmitter site locations the simulator has generated...a 7 Figure The active transmitter sites are selected from the 2-dimensional site cell matrix with the individual numbers of the sites....a 8 Figure The x and y coordinates for the calculation of the site locations....a 8 Figure The geometrical characteristics of the hexagonal model used in the simulator....a 9 Figure The idea of the SFN reuse pattern. In this case, K=9. The colours represents different frequencies, each forming a single SFN area with 9 sites...a 10 Figure The simulation phase....a 12 Figure The terminal is placed in the map varying the x and y coordinates, and the distance from the site antenna is calculated in 3D space...a 13 Figure The principle of the snap-shot of each simulation round...a 13 Figure An amplified view to the outage probability of 10% (area location probability of 90 %) with respective C/(N+I) values for SFN reuse pattern size of K= A 18 Figure The formed curve for the distances of 1 20 km of the mountain site is close to the logarithmic form, as used in Publication VII....A 21 Figure The interpolated curve for the distance of km can be formed linearly...a 21 Figure The error margin of the estimated path loss values obtained via the original tabulated values of the model ITU-R P and the respective regression curves is within db up to 60 km of distance. There is also a peak up to +0.8 db in close distance which does not have significance in this case...a 22 Figure The uniformity of the coordinate distribution can be analyzed by slicing the x-axis and y-axis of the investigated area in 10 sub-areas and revise the percentage of the occurred locations in each slice....a 24 Figure The graphical presentation of occurred x- and y-coordinates in the subslices of the area....a 25

33 xxxv Figure Comparison of the effect of the number of the simulation rounds per complete simulation. A total of 5 simulations are presented in each case. A part of the S- curve is shown around the 5 % FER point....a 27 Figure Cumulative presentation of the standard deviation classes per different lengths of the simulation....a 28

34

35 xxxvii List of Tables Table 2-1. The relation between GI and FFT size as for the maximum tolerable signal delay and respective longest distance of the extreme transmitters in SFN Table 3-1. The summary of total DVB-H bitrates as function of the parameter values as presented in [Dvb09]. The values for 5 MHz band can be extrapolated from the other values Table 3-2. An example of the DVB-H link budget Table 3-3. The relationship of the area location probability in the site cell edge and over the whole site cell area for the mobile reception when the standard deviation is 5.5 db, according to [Dvb09] Table 3-4. The estimated site cell radius for different transmitter antenna heights h BS and transmitter power levels P TX by applying the Okumura-Hata model for the urban area.55 Table 3-5. The number of sites in the planned area Table 3-6. An example of the unit cost of the sites with different power categories Table 3-7. The normalized cost for different parameter values in order to cover km Table 4-1. An MPE-FEC error analysis for the example shown above. (* indicates too low sampling rate) Table 4-2. Numerical values of individual signals of the example Table 4-3. Values of the summed signals of the example. G SFN is calculated by comparing the average and 50%-ile level with the strongest individual signal of each case Table 4-4. Comparison of the SFN gain values obtained from different references Table 5-1. The MPE-FEC effect for the investigated parameter values in a single site cell case Table 5-2. Comparison of indoor pedestrian MPE-FEC results obtained form Publication VIII and [Apa06b] Table 5-3. The summary of pros and cons of DVB-H transmitter power levels Table 5-4. The most relevant CAPEX items and relations Table 5-5. Cable types and main characteristics utilized in the CAPEX/OPEX analysis of this thesis Table 5-6. An example of the site cell radius obtained by varying the cable type and transmitter power levels. N/A is shown if the cable is not suitable for the respective power level

36 xxxviii Table 5-7. The most relevant OPEX items and relations Table 6-1. Comparison of the squared and direct power summing Table 6-2. An example of the simulation results, C/(N+I) as function of the cumulative area location probability. A total of 7 complete simulations are presented, for the SFN reuse patterns K of 1, 3, 4, 7, 9, 12 and A 17 Table 6-3. Comparison of the values obtained by interpolating the tabulated form and the regression curves....a 22 Table 6-4. The percentage of the samples in each sliced area of x-axis. The ideal distribution would result 10 % of samples in each slice. The table presents a total of 5 complete simulations with 60,001 rounds each...a 24 Table 6-5. The percentile values as function of y-axis slices. A total of five simulation rounds are presented....a 24 Table 6-6. The analysis of the probability variations when C/I = 8.5 db reference point is observed in each simulation. The Table shows the respective variations in probability values...a 26 Table 6-7. The analysis of the variations of the interpretation of C/I value when 5% probability point is observed....a 26 Table 6-8. The accuracy analysis of the simulation cases...a 28

37 1 1 Introduction 1.1 DVB-H scene The time and location independent mobile communications has changed our living style globally. It is thus not surprising that the evolution of the mobile technologies contains ever advanced solutions. In addition to the convergence evolution of mobile communications with terminals capable of handling various simultaneous services and bearers like GSM, 3G and wireless LANs, also the importance of the broadcast based services is increasing. Mobile broadcast is a method that provides the users with high channel capacity and interactivity. It brings new aspects in the personal information handling, whether it is about leisure time with entertaining TV programs or real-time news, economics and business related information delivery. This interactive mobile multimedia is one of the key ideas in the evolution of mobile communications. The public service broadcasters deliver television and audio programs over terrestrial, satellite and cable transmission increasingly in digital mode. It is foreseen that this program offer will be enhanced for reception on mobile devices [Dvb06c, p. 21], meaning that the contents and format of the information source will be adapted to the small screen of the terminal and moving environment. This thesis concentrates on the Digital Video Broadcasting in Handheld environment (DVB-H). The advantage of the broadcast system over the mobile network's Point-to-Point streaming is that it provides the services in large areas without capacity limitations, i.e. the amount of the actively receiving users does not limit the transmitting capacity. The site cell coverage area of DVB-H is normally between that of traditional cellular systems and TV broadcast networks. DVB-H has adopted ideas from both TV broadcast world as well as from cellular solutions. It is especially suitable for the mobile environment because it takes into account the varying radio conditions in both indoors and outdoors, it has a small screen size, relatively long battery life and it is able to cope with the varying velocities of the terminal. Various DVB-H element vendors, service providers and broadcasters are actively working in the field as [Dvb09] shows. The system has been standardised in DVB-H sub-groups working under ETSI (European Telecommunications Standardisation Institute). The first complete DVB-H specification was published in 2004 which generated various trials and pilots. One of the first trial setups called Finpilot was initiated in Finland during 2004, and Italy was the first country to launch the commercial DVB-H service in The system is also supported by European Union. Nevertheless, DVB-H is not European specific, but it has been evaluated as a strong candidate in other continents like North and Latin America as well as in Asia.

38 2 DVB-H belongs to the DVB family, other variations being DVB-T (terrestrial), DVB-C (cable) and DVB-S (satellite). DVB-H is based on the DVB-T, but DVB-H is designed specifically for the moving environment with mobile equipment. Within a DVB-H frequency band, which would be useful only for a single DVB-T channel at the time, it is possible to transmit several DVB-H sub-channels containing video and/or audio. This is a result of relatively low bit stream capacity demand as the screen size and thus the resolution is only a fraction of the typical DVB- T solution. The evolved version of the DVB-H system is called DVB-NGH (Next Generation Handheld), and the main idea of it is to provide more capacity by optimising the coverage area. Also a satellite version of DVB-H is being standardized, called as DVB-SH. It can be expected that DVB-H with its variations will be part of a complete set of multimedia systems. Other bearer layer networks that can be compared technically with DVB-H are DAB/T-DMB (extended version of Digital Audio Broadcasting), MBMS (Mobile Broadcast Multicast Service of 3G), and MediaFLO (Forward Link Only) [Bmc07c, p. 41]. Even if the focus of this thesis is in the radio network planning of DVB-H, it can be assumed that the methods presented are applicable with minor modifications also for the evolved versions of DVB-H and for other, parallel mobile TV systems in the broadcast environment. As the DVB-H is based on the OFDM (Orthogonal Frequency Division Multiplex), the principles presented in this thesis could also be applied at some extend to other systems utilizing the same technology, e.g. the MBMS (Mobile Broadcast / Multicast Service) of LTE (Long Term Evolution). 1.2 DVB-H radio network planning The basic planning of DVB-H radio network is in principle a straightforward task as there is only a limited set of parameters available compared to the actual mobile communications systems like GSM and UMTS. It can be assumed that the initial DVB-H radio network is thus possible to deploy by applying the environment-dependant default parameter values, according to the expected terminal speed and the theoretical maximum size of the single frequency network. The optimal values obviously depend on the more specific regional aspects like terrain topology and radio signal propagation conditions, as well as on the expected local use cases, including the proportion of the indoor, outdoor and vehicle users. This means that the detailed radio network planning and optimisation should consider well the regional differences. The general performance of the ODFM that DVB-H utilizes in the radio interface gives good assets in the initial DVB-H radio network planning phase in order to select the suitable first parameter settings, and also to eliminate the least feasible settings in the early phase of the planning. The general OFDM performance as a function of the radio parameter values that is

39 3 relevant in the nominal network planning phase were identified in [Bee02], [Bee98], [Dvb09], [Far04], [Far05], [Far05b], [Fis08], [Law01], and [Pos05]. For this thesis, publicly available information about the estimated performance of the DVB-H radio parameter values was identified for different planning items. By writing this thesis, there were already real field tests reports available based on DVB-H trials and pilots, as can be seen in [Apa06a], [Apa06b], [Avo06], [Bou06], [Bov09], [Fan06a], [Far05c], [Dvb09], [Kru05], [Mäk05], [Mil06], and [San05]. There were also various studies available about the DVB-H link budget studies as described in [Bmc09], [Sci07], and [Zyr98]. The coverage and frequency planning issues, either based on the DVB-T, DVB-H, other comparable OFDM systems, or on the applicable general radio propagation theories, were found in [Bac04], [Bmc07b], [Bro02], [Cha06], [Ecc04], [Fan06a], [Fan06b], [Fis08], [Goe02], [Gre06], [Hat80], [Itu07], [Jeo01], [Jos07], [Öst06], [Pal08], [Tun05], [Ung06], [Voj05], and [Zha08]. In this physical radio frequency layer in question, also the EMC (Electro-Magnetic Compatibility) and RF exposure limit analysis was identified as an important part of the radio network planning. These studies were indentified in [Fcc98], [Min99], and [Sci07]. For the more in-depth phase of the radio network planning, the DVB-H error correction performance in the link layer is one of the important topics. There were various studies identified about the error correction behaviour of DVB-H as can be seen in [Bou08], [Gar07], [Gom07], [Goz08], [Him06], [Him09], [Ili08], [Jok05], and [Jok06]. Another important detailed radio network planning related topic was noted to be the functionality and performance of the SFN which is presented in [Bee07], [Ebu05], [Pit09], [Ple08], [Ple09], [Ple09b], [Sil06], [Ung08], and [Zir00]. The radio network planning would not be complete if the cost effect prediction were missing in the deployment processes. In fact, this is essential part in the DVB-H business modelling which, after all, determines if the DVB-H network deployment is feasible in the first place. The deployment cost and business model related studies were identified in [Bal07], [Bmc07], [Sat06], and [Sci06]. It was noted that the complete technical and economical joint analysis were not found too many by the writing of this thesis. References [Had07] and [Sil06] were the most relevant techno-economical DVB-H studies that were identified. Also [Lig99a] and [Lig99c] were noted as highly relevant sources even if they describe the techno-economical planning for DVB-T, i.e. for higher power levels and antenna heights by default than DVB-H utilises. It is obvious that in the deployment of the DVB-H, a sufficiently complete network planning procedure is needed. In the most functional format, the planning process should take into account the inter-dependencies between the technical, economical and regulatory issues. There are various cross-relations between the parameter values and their effects on the deployment and operating expenses of the DVB-H. As an example, the QPSK modulation scheme provides

40 4 the largest site cell coverage areas, i.e. lowest amount of sites and thus a base for the most economical network. On the other hand, the resulting channel capacity is considerably lower than the one the 16-QAM or 64-QAM modulation schemes provide, but via smaller (and thus more expensive) coverage areas. The techno-economic optimisation is thus needed in the early phase of the planning. 1.3 Problem identification Radio network planning process There are various references that describe the DVB-H radio network planning items as shown in Chapter 1.2. Nevertheless, based on these sources, it can be concluded that the complete and detailed level DVB-H specific radio network planning process was not possible to identify, or only selected parts of the process were explained at the time. Furthermore, by writing this thesis, no in-depth descriptions about the complete radio network planning were found. The intention of this work was to seek for a complete and proven description that would provide as detailed information about the inter-dependencies of the technical items as possible, and which would take into account also the economical impact in a short and long term of the DVB- H network operation. The conclusion was that there would be room for a more complete and detailed network planning process. As a result of this observation, the process for an initial and detailed radio network planning was created as a basis for this thesis Cost impact Cost optimization is one of the most essential tasks of the DVB-H operators. If the sufficiently in-depth methodology or confirming results are not possible to obtain, it can result a decision to even reject the whole deployment project. Despite of this fact, surprisingly low amount of the related studies were identified by the initiation of this thesis. Furthermore, even relevant, these studies were either not DVB-H specific [Lig99a] and [Lig99c], or they could still be enhanced for the realistic deployment purposes [Sil06]. As a result, CAPEX and OPEX optimization as a function of technical parameters was selected as a starting point for the complete radio network planning studies of this thesis Single frequency network performance The planning of the DVB-H radio network is relatively flexible due to the possibility to select the Multi Frequency Network (MFN) or Single Frequency Network (SFN) variants. The theoretical functionality of these is straightforward. MFN is based on the handovers between the

41 5 neighbouring DVB-H sites that utilize different frequencies, whilst SFN is based on the highly synchronized functionality of a set of sites. For the MFN, the planning problematic relates merely on the handover success rate. As this is a relatively limited item and can be solved by provisioning of the sufficient overlapping site cell areas, the MFN was not studied further in this thesis. As for the SFN, it is commonly understood that the additional benefit of it is a performance gain that is a results of the summing of received radio signals [Ebu05]. As Chapter 1.2 indicates, there are several references available that aims to estimate the level of the SFN gain. Nevertheless, the presented studies in these references gave an impression that the definition of the SFN gain is not well harmonized. In fact, due to a relatively large variation of the publicly available estimates of the gain, no values are possible to apply for the radio planning purposes implicitly. According to the identified reference list, the value for the SFN gain seem to be either negative due to the increased modulation error rate [Bov09], or vary between about 0 db [Bmc09, p. 15], [Ple09b] and over 6 db [Apa06b, p. 163], [Apa06b, p. 133]. In average, the value seems to be in order of 2 3 db [Ple09], [Apa06b, p. 30, 32, 35], [Apa06b, p. 116]. It is worth noting that all of these results have been obtained by assuming or measuring a low amount of, i.e. 2 3 DVB-H transmitters. Due to this inconsistency of the publicly available information, and because there was obviously not coherent definition for the SFN gain available for the link budget purposes, in-depth studies of the SFN in both normal-sized as well as in interfered, i.e. over-sized SFN was selected for one of the study items of this thesis Field test and analysis methodology Based on the typical operation of mobile communications networks in general, as well as on the experiences of the trial and pilot phases of the DVB-H activities, it can be assumed that one of the essential tasks in the DVB-H network planning, performance evaluation and operation is the execution of the field tests. By carrying out the field tests, the operators can evaluate the performance and quality of the service, as well as make sure that the network is constructed according to the plans. The measurements also serve as a base for the in-depth network planning. Furthermore, due to the lack of the report delivery mechanism of DVB-H, the test measurements provide the only reliable manner to do the fault management in the radio interface, i.e. to make sure that the antenna system and radio equipment works correctly. The applied criteria for the field test results are typically the received power level and relevant performance indicators that quantify the received signal quality in numeric values or in the form of coverage maps. The drive tests can be executed throughout the life cycle of the network. In the beginning of the deployment, the measurements are needed for the service functionality revisions and for the preliminary parameter adjustments. In the more mature phase of the

42 6 network, the measurements are done for the optimization purposes, quality revisions for the fault management (as the return channel is missing, radio quality measurements are not available in operations and maintenance systems), and to audit the effects of the parameter value changes. The drive testing typically requires dedicated radio measurement equipment. The monitoring of the effect of the parameter values is important especially in the early phase of the network, in order to select correctly the values. As a one of the most important examples, the effect of the MPE-FEC (Multi Protocol Encapsulation Forward Error Correction) might give additional performance gain in certain locations and situations. In order to make sure the correct value is selected in each area, the measurement is essential. By the initiation of this thesis, the measurement equipment, e.g. for the MPE-FEC evaluation, was very limited. The methodology of the MPE-FEC gain evaluation was also still in an early stage. Some results were already available via the trial and pilot DVB-H network measurements which showed typical performance of the selected DVB-H functions, including MPE-FEC as described in [Apa06a], [Apa06b], and [Mil06]. Nevertheless, the measurement setup was varying depending on the sub-activities of these network tests, and it was not always clear in a deep level how the respective measurements and post-processing had been done. In order to clarify the field measurement methodology and post-processing, a case study was done for collecting and analysing of the MPE-FEC performance. The method is based on the mobile receiver with already embedded DVB-H field test software. The aim of these studies is to clarify the test methodology and post-processing, with typical performance values included Coverage planning The most useful models for the initial and in-depth coverage area and transmitter antenna safety distance estimation are not necessarily clear in DVB-H specific environment. It was noted that there is room especially for the investigations of the interference distribution and their significance in over-sized DVB-H SFN network as this topic is not clearly presented in the scientific publications. The combined studies of the useful coverage areas and SFN that contains interferences was thus identified as an important study item. The coverage area planning also requires the EMC and radiation exposure limit estimates, which was not straightforward to find specifically for the DVB-H. The evaluation of the radio propagation models in theoretical and practical DVB-H planning was thus decided to be performed together with the presentation of the methodology of EMC and RF exposure estimation in typical DVB-H scenarios.

43 7 1.4 Objectives of the thesis The objective of this thesis is to find clarifications and solutions for the identified set of problems described in Chapter 1.3. The initial aim was to clarify the DVB-H planning and optimization process, and to identify the items that need to be clarified due to the lack of consensus in the scientific field. The more specific objective was to plan and utilise scientific investigation methods in order to clarify the unclear topics. As for the methods and tools, the concrete aim was to apply both simulations and field tests with related analysis as well as to carry out general radio network planning related studies. In addition to the developed methods, also examples of the parameter values are presented in order to estimate the functionality of the models and the performance of the DVB-H radio network in the investigated cases. When possible, they are compared with the other published results. The complete system can be divided into the radio part (which is called the actual DVB-H network) and the core network (including the head-end, delivery and management blocks of the system, forming the IP Datacast part of the network). This thesis concentrates on the radio part of the system. In general, the main idea of the thesis is to present methodologies that can be applied in the planning and optimisation of the DVB-H radio network. As the network designing values have inter-dependencies as function of the radio propagation environment and other technical and commercial assumptions, the presented parameter values are meant for case examples rather than for final link budget guidelines. The focus of this thesis is thus to present investigation methodologies as a basis for the local investigations and adjustments of the DVB- H radio networks. As a concrete outcome, these studies provide a useful tool for the cost efficient deployment and operation of DVB-H. 1.5 Publications and contributions Publication I contains a techno-economic optimisation principle for DVB-H network planning. The main contribution of this Publication is a complete method for the optimisation of the technical parameter values and costs of the DVB-H network in the deployment and operational phases. Although the DVB-H network cost calculations are important in the efficient network deployment, there were no analyses found from public sources that would present methods for the techno-economic network optimisation that takes into account sufficiently the relationship between the technical items and related expenses in the network deployment phase as well as during the longer-term operation of the network. The closest identified source was [Sil06, p. 48], which studied the cost-effect as function of DVB-H cell size in Finnish environment. The objective of Publication I was to create a complete and in-depth method that can be used as an iterative step in the cost-efficient network planning process. This publication clarifies the topic and identifies the relevant cost items. The outcome indicates the relationship between the

44 8 CAPEX and OPEX of DVB-H network as function of the transmitter power level and antenna height. Publication II shows a method that can be applied for the fast revisions of the network performance with respective data collection, post-processing and analysis. The main contribution of this Publication is a clarified method to analyse the performance of MPE-FEC, and a confirmation of the behaviour of the DVB-H error correction in mobile outdoor environment. The case examples and their results show the importance of the parameter adjustments as function of the radio channel type. The presented method for the post-processing of the measurement data as well as for the respective analysis and case results are based on the commercial data collection software of the DVB-H terminals. The post-processing method and the analysis had not been used in this specific format by the writing of this thesis. The closest methods are found in the WingTV field measurement documentation [Bou06], [Apa06a] and [Apa06b] by the publication of the papers. Due to the developed method, Publication II gives added value for the optimization process of the DVB-H radio network planning. This thesis also clarifies the accuracy of the respective measurements of MPE-FEC gain compared to the principles presented in [Dvb06] and [Dvb09]. Publication III presents a method that can be applied for the simulations of the interference levels in over-sized SFN area. The main contribution of this Publication is the method to estimate the interference levels as function of the geographical interference distribution by applying simulation code that is relatively straightforward to implement to the typical network planning tools as an additional module. A radio interface path loss prediction based simulator was developed, and case studies were carried out with some of the most logical radio parameter values. The simulator is based on the Monte-Carlo method. Unlike typically in coverage planning tools, the developed simulator does not divide the investigated area into smaller regions. Instead, the proposed method simulates the whole investigated area and produces CDF of the carrier, interference and C/(N+I) levels. This speeds up considerably the simulation time, yet producing sufficiently accurate results for the case comparison purposes. The results clarify the effect of the antenna height and transmitter power level on the errors. The method can thus be used for the estimation of the severity of the errors, and it provides information about the optimal setting of the antenna heights and transmitter power levels that still fulfils the final radio reception quality requirement even if the theoretical SFN limits are exceeded. The method is straightforward to implement in any other simulation platforms. The interference analysis results were not found in this format from the other references by the publication of the paper. Nevertheless, the effects can be compared with the identified SFN study references. Publication IV presents further development of the SFN interference simulator of Publication III. The main contribution of this Publication is the extended functionality of the simulation method, providing possibility to estimate the balance between the SFN gain and self-

45 9 interference levels. This version utilizes a hexagonal model layout that is comparable to the reuse pattern concept of TDMA networks. The applied type of the reuse pattern, which can be called an "SFN reuse pattern", applies to the variable sizes of the SFN area, consisting of site cells that functions in the same frequency and that forms a similar pattern than, e.g. the GSM cells. Different SFN areas form thus MFN areas. The respective simulations show the C/I distribution as function of the SFN reuse pattern size K. This gives added value for the DVB-H radio network planning in cases where the theoretical SFN limits are exceeded, e.g. due to the lack of available frequencies. The method gives more detailed information about the balancing of SFN gain and SFN interferences than was found in other references by the writing of this thesis. Publication V presents case studies about the DVB-H coverage area predictions. The main contribution of the Publication is a set of study results which indicates the feasibility of the generic radio propagation models in the initial phase of the DVB-H network planning. It compares the correlation of the Okumura-Hata model in theory and when it is embedded in an operational radio network planning software. The selected environment represents a dense urban area type. The outcome gives useful information for the initial DVB-H radio network planning phase, and shows that the first-hand estimate for the number of the required sites can be done sufficiently accurately by simplifying the area for different cluster types and by applying the Okumura-Hata. The findings indicate the issues of the practical environment which should be taken into account in the detailed radio network planning phase. Publication V also creates a base for the third phase SFN simulations presented in Publication VII, so these documents can be utilized as a pair for the complete coverage and respective interference analysis. A set of other references about the interference analysis were found by the publication of these papers. Nevertheless, the studies presented in this thesis show the expected interference behaviour from new perspective with clarifying results. The papers confirm the previously published phenomena in case of the over-sized SFN network, giving more detailed information about the cumulative interference levels. Publication VI presents case studies about the radiation levels of DVB-H radio network. The main contribution of this Publication is to present the feasibility of the generic safety zone estimation principles on the DVB-H antenna location planning, with clarified method to estimate the radiation as function of the vertical radiation pattern. In the radio network planning, the limits of the radiation should be taken into account when selecting the optimal parameter values. The limits are determined by the international and national legislation, as well as in the tolerance levels of the other systems for the interfering electro-magnetic fields. Publication VI shows the calculations adjusted to specifically DVB-H environments. The findings can be utilized as a feedback loop module in the complete techno-economic radio network planning process for the final adjustment of the parameter values.

46 10 Publication VII presents a third version of the SFN interference simulator developed in Publications III and IV. The main contribution of this Publication is an adjusted simulation method that can be applied in a realistic outdoor environment for the prediction of the radio performance and interference distribution as function of the parameter set that determinates the SFN size. This version of the simulator is adjusted for the practical environment that represents a dense urban area. The network layout of the simulator was selected the same as was studied in Publication V, i.e. the site locations, antenna heights and transmitter power levels were equivalent. Okumura-Hata and ITU-R P radio propagation models were applied in the analysis. The outcome indicates the C/I distribution over the whole area as function of the all parameter values that have effect on the SFN size. The results show the severance of the interferences of each case, which gives possibility to select the optimal parameter values. Publication VII indicates that the presented SFN simulation method can be applied in practical environments. The presented technique is relatively straightforward to implement to existing network planning tools as an additional module that requires only low amount of processing power and memory capacity, which shows novelty in the related field of development. Publication VIII presents analysis of the MPE-FEC performance in indoor and outdoor environments, both in laboratory premises as well as in live network coverage areas. The main contribution of this Publication is the confirmation of the MPE-FEC performance levels in indoor and outdoor environment, by applying the method developed in Publication II. The undersigned carried out the pedestrian cases in indoors and outdoors by applying the analysis methodology presented in Publication II. Publication VIII confirms the general understanding about the MPE-FEC performance that has been presented, e.g. in the WingTV field study documentation. Nevertheless, Publication VIII shows selected cases in more detailed level as function of the relevant parameter values. The results show that the effect of MPE-FEC is most significant in the cell edge area, i.e. in city areas where the C/I level is high by default, the effect of MPE-FEC is not necessarily significant. Publication VIII identified also practical cases in the live network which indicated the presence of the impulse noise in a high field. By applying the presented method for the field measurement analysis, it was shown that even low MPE-FEC rate can recover the signal. This proves that there is a benefit of the usage of MPE-FEC even in the city environments, but only with low MPE-FEC rate in order to optimize the offered channel capacity when the network coverage is sufficiently good. Publication IX is an extended version of Publication II, describing the field test method and MPE-FEC gain analysis in journal format. The main contribution of this Publication is the description of the presented field test analysis method in a more in-depth way, with deeper analysis of the performance values. In Publication IX, there is a more detailed analysis done for different radio channel types by breaking the original test drive results into segments. The results indicate the logical behaviour of the MPE-FEC gain as function of the parameter values.

47 11 This Publication gives new and more detailed information about the topic compared to related studies presented, e.g. in WingTV documentation. Publication X is an extended version of Publication IV. The main contribution of this Publication is the more in-depth description of the SFN interference simulations and case results that confirm and deepens the already existing results. It presents the complete SFN simulator method with respective case studies carried out in Publications III, IV and VII. Appendix I presents the SFN simulator developed in Publications III, IV and VII. The main contribution of this Annex is the description of the simulator. The block diagram of the simulator is presented with respective models and main functionality, which provides a possibility to implement the method to other platforms, including operational network planning programs as an additional module. There is also a performance analysis of the method presented with selected investigations of the accuracy of the simulator. 1.6 Thesis outline This thesis is divided into the following chapters: introduction, overview of the DVB-H architecture and functionality, identification and investigation of the planning and optimisation related items of DVB-H, and conclusions. In addition, the developed SFN simulator description is annexed. The first part of the thesis contains an introduction to the DVB-H system and the most relevant background information about the presented study items. Chapter 2 explains the functionality and architecture of the network identifying the relevant items for the investigation. The main references that are used in this part are found from the IEEE Xplorer document library, DVB specification groups, WingTV CELTIC field test and simulation documentation as well as from the ETSI digital libraries. The second part of this thesis identifies the relevant items for the planning and optimisation of the DVB-H radio network. For the complete radio network planning and optimisation description, the initial and in-depth planning processes are presented in Chapters 3 and 4, and optimisation investigations are presented in Chapter 5. This is the core part of the thesis as it describes the investigations of selected items based on the publications of the thesis. The studies include literature comparisons, creation of methodology for the field measurement and analysis with the respective case results, and simulations with the results that were obtained by applying typical parameter settings. Finally, the conclusions are presented in Chapter 6, followed by the annexed simulator description and Publications.

48 12

49 13 2 Principles of DVB-H 2.1 Description DVB-H (Digital Video Broadcasting, Handheld) is based on the terrestrial digital television standard DVB-T. DVB-T system is designed for a static environment where a rooftop mounted receiver antenna is installed providing line-of-sight (LOS) or nearly-los with the transmitting site. Although it is possible to use the DVB-T receiver to some extend in a mobile reception, according to the related experiences, the use of DVB-T is not optimal in an environment where multipath propagation, impulse noise and Doppler shift are present [Far06, p. 194]. The initiation of the DVB-H standardisation work was a result of the identified need for sufficiently high-quality mobile TV reception when using small portable or mobile terminals. An EU-sponsored Mobile Television and Innovative Receivers (Motivate) project studied the item and stated in the conclusion in 2000 that although DVB-T could be used in the mobile environment, it was not an optimal solution. During the standardisation of DVB-T, TV subscribers had not yet used the service significantly in mobile environment. Even if DVB-T contains also definitions for the mobile channel, the experience has shown that the usage of home set-top box solutions in the mobile reception is not feasible in practice due to the constantly changing radio conditions [Pek05, p. 36]. Among the general mobile communications development, the need for optimized broadcast system for mobile environment increased and finally triggered the standardisation work of the DVB-H [Far06, p. 194]. There is a set of special aspects in TV reception when the terminal is moving. Main differences between fixed TV and mobile TV are related to the characteristics of the radio interface. Whilst DVB-T utilizes a fixed roof-mounted high-gain directional antenna, the mobile receiver antenna is normally an internal one. The mobile users are typically on street level where the radio wave propagation conditions are challenging as the radio interface consists of time and space dependent variations both in outdoor and indoor environments [Mar05]. In addition, the power consumption is an important issue with the hand-held devices. There is a considerable power saving identified in the receiver functionality of DVB-H [Far06, p. 199] which provides sufficiently long usage time in the typical moving environment compared to the receiver that is switched on permanently. It can be noted that TV is developing towards the use of multiple niche channels [Hum09, p. 18]. DVB-H would be suitable platform for this as it contains several audio and video channels with varying quality settings.

50 14 The first version of the DVB-H standard with the system name of Digital Video Broadcasting Handheld, was published as an ETSI standard number EN in November The standard actually compiles the existing DVB-T standards in such a way that the DVB-H system can be created based on those. In practice, the DVB-H includes the fixed and in-car standards of DVB-T, adding new functionalities that take into account the above mentioned specialties in the mobile environment. The most important additional parts in the link layer include the Time Slicing and MPE-FEC (Multi-Protocol Encapsulation Forward Error Correction) functionalities as described in [Dvb09]. The DVB-T is defined in ETSI EN , and includes now an Annex for DVB-H. The DVB Data is defined in ETSI EN , which embeds the encapsulation mechanism to DVB-H data. The DVB service information related information is defined in ETSI EN , with update of the signalling definitions for the DVB-H handheld terminals [Far04, p. 2-3]. The Time Slicing functionality provides a possibility to transmit data within cyclic, highcapacity bursts. After the reception of a single burst, the terminal switches its receiver to a sleep-mode until the next burst is transmitted. According to the DVB-H implementation guidelines [Dvb09], Time Slicing reduces the average power consumption of the DVB-H receiver front-end approximately 90 95%. Nevertheless, as the terminal includes also other functionality in addition to the DVB-H receiver (video processing capability with related audio and video player, separate mobile phone functionality of GSM and/or 3G as well as other functionalities, each reserving part of the processing power), the overall saving of the power consumption is logically lower than single savings in the receiver sleep-mode indicates, but it anyway extends the battery life in mobile environment. In addition to the receiver s power saving, Time Slicing can be used for the smooth frequency handover functionality when moving from one site cell area to another. The DVB-H includes the forward error correction (FEC) mechanism of DVB-T. There is also additional error correction functionality in DVB-H, i.e. MPE-FEC. It improves the performance under impulse interference and the Doppler shift. The MPE-FEC is defined as an optional functionality in DVB-H. As Time Slicing and MPE-FEC are defined in the link layer, the already existing DVB-T receivers are not disturbed due to the existence of DVB-H. On the other hand, DVB-H is completely backwards compatible with DVB-T, so both of these broadcasting methods can be multiplexed into a single transmitter antenna. Also the physical layer of DVB-H has some important additions. According to [Dvb09], these are transmitter parameter signalling (TPS), 4K OFDM (Orthogonal Frequency Division Multiplex) mode, new interleaving depths and additional bandwidth of 5 MHz.

51 15 The DVB-T Transmission Parameter Signalling (TPS) is upgraded for DVB-H. The DVB system s TPS includes thus two additional bits in DVB-H that indicates the presence of DVB-H services and the possible MPE-FEC. The signalling enhances the service discovery process speed. TPS bits also carry the cell identifier in order to support the fast scanning of the mobile receiver and to perform a frequency handover. TPS is defined as mandatory in DVB-H. In addition to the already existing FFT modes 2K and 8K of DVB-T, a new 4K OFDM mode is specified in order to optimize the mobility of the terminal and the size of the single frequency network (SFN). This allows single-antenna reception in medium-sized SFN with relatively high mobile speed. 4K mode is not mandatory for DVB-H, though. The DVB-H standard defines new symbol interleaving options. In DVB-T, the 8K mode has a so called native interleaver in order to spread the bits over time domain and minimize the bursty errors over a complete symbol. In DVB-H, it is possible to interleave the data also in 2K mode, which results the interleaving over four OFDM symbols. In case of 4K mode, the interleaving is done accordingly over two OFDM symbols. This DVB-H functionality is called in-depth interleaver. The addition reduces the effects of impulse noise up to the level that is possible to obtain with the native interleaver of 8K mode. It is worth noting that neither the 4K mode nor in-depth interleaver is mandatory for DVB-H, but TPS is obligatory as well as Time Slicing and cell identifier. As DVB-H is backwards compatible with DVB-T, all of its modulation schemes, i.e. QPSK, 16-QAM and 64-QAM, are possible to use also in DVB-H. In addition to the already existing bandwidth definition for 6, 7 and 8 MHz of DVB-T, the DVB-H standard defines also a new 5 MHz bandwidth for the areas where this value is possible to utilize according to the regulations. It can be generalized that the DVB-H defines the radio functionality of the system whilst the remaining part of the network from the encoders up to the IP Encapsulator (IPE which is the interface towards DVB-H) is called DVB-IPDC (DVB IP Datacast) network. IPDC delivers the content in form of data packets by applying standard routing principles of Internet. The benefit of the use of IP is the possibility to use standard components and protocols for the content transmission, storage and manipulation [Dig05]. In addition, the DVB-CBMS (Convergence of Broadcasting and Mobile Services) defines the video and audio formats, Electronic Service Guide (ESG) and content protection on top of the DVB-H.

52 DVB-H network Architecture Figure 2-1 shows the functional architecture of DVB-H as interpreted from [Dvb09]. The interfaces that are specified in DVB-CBMS are CBMS-2 (for audio/video streams and files), CBMS-3 (delivery of ESG metadata and Point-to-Multipoint, i.e. PTM), CBMS-1 (PSI/SI), CBMS-4 (access control to service applications, ESG metadata and PTP delivery), and CBMS-5 (PTP, i.e. Point-to-Point transport services, i.e. SMS, MMS and IP connectivity). The interfaces that are not in the scope of DVB-CBMS are X-1, X-2 and X-3. The elements that are in scope of DVB-CBMS are marked in grey colour. DVB-H network s air interface Content creation X-1 Service application CBMS-5 CBMS-7 CBMS-2 CBMS-6 Broadcast network Service management X-3 Interactive network CBMS-1 CBMS-3 CBMS-4 X-2 Terminal Mobile network s air interface Figure 2-1. The DVB-CBMS architecture as interpreted from [Dvb09]. The DVB-H refers to the radio, i.e. broadcast network of the system, including the transmitters, modulators, antenna systems and connections from the IP Encapsulators. As [Dvb07b] states, the standard ETSI EN defines the DVB-H radio transmission, whereas the ETSI TS defines the DVB-H transmitter s and receiver s system-level signalling. The DVB IPDC network refers to the rest of the system, including DVB-H Head-End, transport of the contents and signalling up to IP Encapsulator and related management systems. In other terms, this part can be referred as DVB-H core network.

53 17 Figure 2-2 shows the complete DVB-H delivery chain, including the core and radio parts [Pen09, p. 39]. The focus area of this thesis is the DVB-H radio network. The DVB-H core network, i.e. the part from encoders until the IP Encapsulator, consists of the encoded program source stream, data handling and management elements, interconnection towards the return channel system with its billing system, and connection to the DVB-H radio network. Between the elements, there is also an IP network with respective adapters. Operations and management Mobile network operator Content provider Broadcaster DVB-IPDC core network DVB-H radio network Head-end IP distribution network Broadcast network Service protection Advertisement insertion IPE network management Satellite transmission Encoding IP encapsulation Terrestrial transmission Modulator and transmitter ESG server Figure 2-2. The complete DVB-H delivery chain can be divided into core network (DVB-IPDC) and radio network (DVB-H) Functionality The idea of the high-level functional DVB-H core and radio network principle is presented in Figure 2-3 [Dvb09], [Hen05]. As can be seen, both MPEG-2 streams of DVB-T as well as DVB-H specific IP streams can be multiplexed and distributed via the same radio network infrastructure. The DVB-T transmitters carry the original 2K or 8K FFT mode that are meant for the full sized DTTV (Digital Terrestrial Television) equipment or set top boxes that converts the signal for the analogue TV. The same site can deliver the additional DVB-H specific signalling for the reception with DVB-H terminals. The terminal has a DVB-T demodulator and IP Encapsulation for the recovery of the original bit stream. The DVB-H terminal finds the proper signalling via the TPS, and it can optionally use the MPE-FEC functionality. If the latter is not used in the terminal, it can anyway receive the stream with the basic coding scheme, but with reduced

54 18 quality as it is not optimized for the mobile environment. The use of Time Slicing functionality is obligatory for DVB-H terminals as the delivery is based on the shared channels in the bandwidth, saving the battery life accordingly [Far06]. DVB-H transmitter IP network IP stream MPEG-2 stream MPEG-2 stream MPEG-2 stream IP Encapsulator FEC DVB-H specific: MPE-FEC Time Slicing MUX TS DVB-T modulator Modulation: QPSK 16-QAM 64-QAM FFT 8K / 2K DVB-H specific: FFT 4K DVB-H TPS RF channel Figure 2-3. It is possible to multiplex the DVB-H IP streams with the DVB-T MPEG-streams. The streams can be delivered via the common infrastructure for the DVB-H and DVB-T terminals. In the downlink of the DVB-H radio interface, the DVB-H receiver contains the demodulator as shown in Figure 2-4. The DVB-H demodulator gets the DVB-T signal via the RF input either from the internal or optional external antenna. The DVB-H demodulator block is a DVB-T demodulator with additional DVB-H specific 4K and TPS functionality. The demodulator block also contains Time Slicing with power control functionality, as well as optional MPE-FEC. DVB-H Demodulator RF input DVB-T demodulator 4K TPS Time Slicing MPE-FEC IP datagrams TS packets Figure 2-4. The high-level block diagram of DVB-H receiver. The RF signal is directed to the input of the demodulator, and output is further handled by the DVB-H terminal. The DVB-IPDC network includes the coding of the source contents and the delivery of the streams (audio/video or files) via IP network towards the radio network. There is multicast IP delivery used within the DVB-H core network. The Head-End consists of IP encapsulation and related functionalities which are ESG creation, service protection, advertisement insertion and the encoding of the source code. There is also an IPE manager located in the Head-End.

55 Elements The IPE Manager takes care of the setting up of the sessions with respective timing of the beginning and end of the sessions, and creates the ESG (Electronic Service Guide) which contains program related information. The DVB-H encoders produce DVB-H IP streams. The source signal's video and audio is coded separately, but delivered via the same physical bit pipe. The typical bit rate for audio is 64 kb/s (2 32 kb/s) via AAC. Typical video bit rates vary from 128 to 384 kb/s, and can be differentiated in channel-basis. The content is streamed to the DVB-H terminal which includes local streamer that is capable of decoding and presenting the original contents of the encoder. The IP Encapsulator of DVB-H takes care of the protection of the stream by applying FEC (derived from DVB-T) and DVB-H specific MPE-FEC, which brings additional protection against impulse noise as well as for the fast fading radio channels especially in environments that contain multi-propagated radio components. There is possibility to encrypt the contents via a service protection server and insert advertisements to the IP stream via an advertisement server. There can be a separate Operations and Management System (OMS) connected to Head-End. It takes care of the DVB-H / DVB-IPDC performance monitoring and fault management, backup and restore, inventory management and other typical IP network management functions. The OMS includes means for different parties to handle the services. There are thus OMS connections to Head-End s IPE network management from the mobile network operator, content provider and broadcaster. Although DVB-H as such is only a broadcast system for the delivery of the contents in the unidirectional downlink radio channel, also the interactive type of services can be provided via a separate return channel and respective elements based on GSM or UMTS networks. The transport from the Head-End to the radio network can be done physically in all known methods via the distribution network, including LAN, satellite link, fibre optics, as the bit pipe is based on IP, as long as the capacity and quality is complied. The stream delivery within the core network, i.e. from the encoders of Head-End up to IP Encapsulators between the core and radio network, is done via IP Multicast. The idea of the multicast is relatively mature as it was presented already in 1985 [Che85]. There are means to optimize the performance of the multicast in the IP datacast network, e.g. FEC mechanisms can be utilized [Lun06]. The actual functionality of the content delivery within the IPDC network is vendor dependent, though.

56 Radio Functionality The audio / video stream or data file is delivered from the DVB-IPDC network to the DVB-H radio network, IP Encapsulator being the interface. The signal is delivered to the DVB-H modulator, DVB-H transmitter and finally to the radio interface via the antenna system Radio transmitter A typical DVB-H transmitter output power is in range of some hundreds of watts up to some thousands of watts. For the DVB-H radio transmission, the transmitter includes DVB-H specific functionality, although also separate DVB-H modulator can be used with the standard DVB-T transmitter. The DVB-H modulator can also be connected in front of an analogue transmitter if the respective power amplifier functions in desired frequency and bandwidth. Due to the physical obstacles like buildings and variations of the terrain height, outages can occur in the coverage area of the site cell. The coverage of these problematic areas can be enhanced by installing separate gap-fillers nearby, which are either passive or active repeaters. The latter can be either direct amplifier or regenerating one, which decodes, amplifies and codes again the bit stream. The output power of gap-fillers varies typically from few watts to some hundreds of watts. Gap-fillers may also perform a frequency translation. In practice, this variation is not useful in typical DVB-H networks as multiple channels are not likely to be available in reduced areas. Even the gap-fillers could be used relatively freely to fill in the coverage holes, the location of the elements should be taken into the network planning in order to maximize the isolation between the receiving and transmitting end to avoid the uncontrolled oscillation Terminal Figure 2-5 shows a high-level block diagram of the DVB-H receiver [Dvb09]. The reception of the Transport Stream (TS) is compatible with DVB-T system, and the demodulation is thus done with the same principles also in DVB-H. The additional DVB-H specific functionality consists of Time Sliced burst handling, MPE-FEC module and the DVB-H de-encapsulation. As can be seen in Figure 2-5, the FER information, i.e. frame errors before MPE-FEC specific functionality, is obtained after the Time Slicing process, and the MFER (remaining FER after MPE-FEC functionality) is obtained after the MPE-FEC module. The IP level information is obtained after the demodulation procedure.

57 21 DVB-H specific functionality DVB-T Demodulator DVB-H Time Slicing RF reference TS reference DVB-H MPE-FEC DVB-H De-encapsulation FER reference MFER reference IP reference IP output Figure 2-5. The DVB-H reference receiver. The measurement point for the received RF power level is found after the antenna element and the optional GSM interference filter. There might also be an optional external antenna connector implemented in the terminal before the RF reference point. The presence of the filter and antenna connector has thus a frequency-dependent loss effect on the measured received power level in the RF point. It should be noted that the receiver antenna diversity can enhance the performance of DVB-H lowering the required C/N value with several db [Far01, p. 11] Frequency The DVB-H specifications define the system for VHF III ( MHz), UHF IV ( MHz) and UHF V ( MHz). There is also a possibility to adopt DVB-H to other frequencies like in L-band. As an example, DVB-H has been tested in USA in the 1.6 GHz band. In practice, the terminal s DVB-H engine might contain a limited band, e.g MHz as stated in [Nok05] because the upper frequencies of UHF are close to GSM 850 system which might cause interferences to the reception of DVB-H. On the other hand, despite of the good propagation characteristics, the lowest identified frequencies in VHF band are demanding for the receiver s physical antenna dimensions as the wavelength is relatively large [Far05]. In Europe, the UHF is identified as the primary band for DVB-H [Bmc07b, p.3]. The frequency bandwidth for the DVB-H has been defined to 6, 7 and 8 MHz. There is also an additional bandwidth of 5 MHz included to the specification. These bandwidth values of DVB- H correspond to the analogue TV channel division in different countries. The modulation bandwidth of DVB-H is mapped into these frequency bandwidth values in such a way that it is the difference between the last (N) and first carrier of OFDM symbol being equal to (N-1)/T S,

58 22 where T S is the symbol length. This results in a modulation bandwidth of 4.76 MHz, 5.71 MHz, 6.66 MHz and 7.61 MHz, respectively, for the frequency bands of 5, 6, 7 and 8 MHz. Figure 2-6 shows the frame structure of DVB-H. The basic unit is the symbol with duration of T S. There are 68 symbols in a single frame which has duration of T F. The frames are repeated in superframe cycles Frame (duration: T F ) Symbol (duration: T S ) T GI T U Guard Interval Useful symbol Figure 2-6. The frame structure of DVB-H. The DVB-H symbol consists of Guard Interval (GI) and useful part that delivers the actual data. GI is needed in order to cope with multipath propagation with a delay spread less than the GI. The useful part contains the OFDM carriers of f 1 f N. The carrier spacing is 1/T U for the separation of two consecutive carriers. The channel bandwidth B ch = (f N -f 1 ) is thus (N-1)/T U. The amount of carriers (that are transferring pilot and data information) within a single symbol depends on the FFT mode. In 2K mode the carrier number is 1705, in 4K mode 3409, and in 8K mode it is consecutive symbols (0 to 67) form a superframe which is repeated in cycles [Sci07] OFDM Parameters DVB-H is based on OFDM (orthogonal frequency division multiplexing). It is a spread spectrum technique that delivers the high speed bit stream via several subcarriers each containing lower speed bit streams. It is used in various systems like DVB-T, LTE, ADSL and WiMAX. The transmission is parallel, and subcarriers are separated by different frequencies as shown in Figure 2-7 [Far07]. The information is distributed in interleaved way to the multiple sub-carriers with the error protection, resulting COFDM (Coded Orthogonal Frequency Division Multiplex) [Fis08, p. 316], [Wan03, p. 952].

59 23 Amplitude Time Orthogonal carriers Frequency Frequency division multiplex GI Data period Figure 2-7. OFDM is based on the orthogonal sub-carriers containing the useful data (marked with light grey colour) as well as pilot signals (marked with dark grey colour). The signal s(t), which is present when multicarrier symbol is combining n sub-symbols s k in a channel h(t), is the following [Rei05, p. 175]: n 1 = s( t) s k = 0 k h ( t) e t jωk t (2-1) A special case of the multicarrier technique is the OFDM, which is based on requirement of orthogonal sub-carrier frequencies ω k 2πkf 0. In OFDM, the signal has N orthogonal subcarriers. They are modulated by N parallel data streams. The sub-carrier can thus be expressed [Bee98, p. 27]: k j 2πf kt φ ( t) = e, (2-2) where f k represents the frequency of the sub-carrier number k. Furthermore, a single OFDM symbol in its basic form multiplexes N modulated sub-carriers in the following way: N 1 k = 0 1 s( t) = x k φ k ( t), (2-3) N where x k is the data symbol number k. The formula is valid for the values t that are shorter than the symbol length. In order to keep the sub-carrier orthogonal in this region, the f t should equal

60 24 to k divided by the symbol length. In this formula, the noise and fading are ignored. The symbols can be then demodulated via DFT. Relatively easily implemented FFT is used in ODFM [Pit09, p. 33]. The DVB-H can use QPSK, 16-QAM and 64-QAM modulation schemes as shown in Figure 2-8. In addition to the constellation plots, there is also continuous pilot signal. Channel estimation of OFDM is usually done with the aid of pilot symbols. The channel type for each individual OFDM subcarrier is flat fading. The pilot-symbol assisted modulation on flat fading channels involves the sparse insertion of known pilot symbols in a stream of data symbols [Bee02]. There is also the TPS carrier found in the DVB-H constellation diagram [Fis08, p. 337, 342]. The QPSK provides the largest coverage areas but least capacity per bandwidth. 64-QAM results in smaller coverage, but offers more capacity. In practice, 64-QAM has been noted to be sensible for the errors in mobile environment as the results in Publication II indicates together with [Mil06, p. 20]. According to this information, 64-QAM would suite for limited parts of the network, e.g. in indoor environments with static or slow pedestrian usage. 64-QAM would thus be suitable for offering high-quality resolution and large channel capacity, e.g. in shopping centres. 16-QAM is an intermediate solution combining relatively high capacity in larger coverage area that 64-QAM provides, but in smaller area than QPSK. QPSK 16-QAM 64-QAM Figure 2-8. The principle of the DVB-H constellations. In addition to the single modulation scheme, DVB-H also can use hierarchical modulation. It divides the RF channel in such a way that two simultaneously set of transport streams can be sent over the single modulation, which is divided into two different sub-parts of the modulated signal. The hierarchical modulation provides possibility to interpret the modulation constellation in different way than single constellations presents via QPSK (i.e. 4-QAM), 16-QAM and 64- QAM. An additional α parameter can be utilized in hierarchical modulation in order to separate farther away the low bit-rate areas of the I/Q-constellation from each others.

61 25 The essential OFDM parameter for DVB-H is FFT size representing the amount of modulated sub-carriers. It can be some of the original DVB-T values of 2K and 8K, as well as the DVB-H specific 4K. The 4K size is a compromise which results a balanced performance between the terminal speed and SFN size. 8K provides with the largest SFN areas (without interferences) but with the cost of the maximum terminal speed. 2K mode offers reversed benefits, i.e. highest terminal speeds but with the cost of reduced SFN size. Guard Interval (GI) is also a relevant parameter for DVB-H. It can have values of 1/4, 1/8, 1/16 and 1/32, representing the proportion of the signal that is not used for the delivery of the data. GI affects directly on the SFN size. GI is important to reduce the negative effects of multipath radio propagation, including the paths caused by the environment itself (e.g. buildings) as well as co-channel interferences in case of SFN network. The combination of FFT and GI parameter values determinates the maximum distance between physical DVB-H transmitters [Dvb09], [Mil06] as shown in Table 2-1. Table 2-1. The relation between GI and FFT size as for the maximum tolerable signal delay and respective longest distance of the extreme transmitters in SFN. Guard interval time / Maximum SFN diameter FFT mode GI=1/4 GI=1/8 GI=1/16 GI=1/32 FFT=2K 56 µs / 16.8 km 28 µs / 8.4 km 14 µs / 4.2 km 7 µs / 2.1 km FFT=4K 112 µs / 33.6 km 56 µs / km 28 µs / 8.4 km 14 µs / 4.2 km FFT=8K 224 µs / 67.2 km 112 µs / 33.6 km 56 µs / 16.8 km 28 µs / 8.4km As can be seen, longer guard interval means higher immunity to inter-ofdm symbol interferences. In practice, the longest GI values do have most significant benefits in static channel types like AWGN-channel [Mil06]. According to the simulations and practical field tests [Mil06], the shorter GI values do have better Doppler tolerance compared to the longer ones. According to [Mil06], the maximum Doppler performance enhances linearly about 25% when GI=1/4 is changed to GI=1/ Error recovery Error control coding is an essential part of mobile communication systems. The error detection and recovery of DVB-H is based on the inner forward error correction (FEC) that is also used in DVB-T. As the receiving end can recover the occurred errors up to certain parameter dependent limits, it is especially suitable for the uni-directional broadcast systems like DVB. Both DVB-H and DVB-T defines five levels for the respective code rate (CR), i.e. 1/2, 2/3, 3/4, 5/6 and 7/8. There is also an optional in-depth interleaver defined for DVB-H. It can be used to increase the

62 26 robustness of 2K and 4K modes as they can be used to further widen the depth of the native 8K interleaver. This functionality enhances the performance of DVB-H especially in fading channels that creates bursty errors in the reception [Mil06], [Dvb09]. The MPE-FEC error correction scheme in DVB-H combines FEC and interleaving functionalities. It is defined as optional in DVB-H terminals. If the terminal lacks MPE-FEC functionality, it can still use the DVB-T compatible FEC for the basic error correction in radio interface. Otherwise, it gives additional benefit for the C/N and maximum Doppler tolerance. The error protection solution in DVB-H combines the already existing inner FEC and outer FEC of DVB-T system, combined with the DVB-H specific MPE-FEC. The inner FEC uses punctured convolutional code and pseudo-random interleaving that is based on OFDM symbols [Bro08, p 6], whereas the outer FEC uses shortened RS(204, 188, t=8) code and convolutional byte interleaving [Rei06, p 10]. The MPE-FEC uses RS(255, 191, t=32) code with erasure decoding as well as time and block interleaving. Figure 2-9 clarifies the position of the FEC and MPE-FEC blocks. IP IP Encapsulator DVB-H services DVB-T services (optional) Time Slicing MPE-FEC MPE MUX DVBT/H transmitter Outer FEC Inner FEC OFDM mod. RF signal Figure 2-9. The DVB-H error protection scheme. In order to carry the IP datagrams of the MPEG-2 Transport Stream (TS), a Multi Protocol Encapsulator (MPE) is defined for DVB-H. Each IP datagram is encapsulated into single MPE section. Elementary Stream (ES) takes care of the transporting of these MPE sections. Elementary Stream is thus a stream of the MPEG-2 Transport Stream packets with a respective Program Identifier (PID) [Ger06, p 198]. The MPE section consists of 12 byte header, 4 byte CRC-32 (Cyclic Redundancy Check), as well as a tail and payload length [Jok05], [Him06]. MPE-FEC table provides virtual time interleaving because datagrams are located in the MPE table column-wise, and correction data is calculated row-wise [Bou08].

63 27 The main idea of MPE-FEC is to protect the IP datagrams of the time sliced burst with Reed- Solomon (RS) parity data. The RS data is encapsulated into the same MPE-FEC sections of the burst with the actual data. The RS part of the burst belongs to the same elementary stream (MPE section), but they have different table identifications. The benefit of this solution is that the receiver can distinguish between these sections, and if the terminal does not have the capability to use the DVB-H specific FEC, it can anyway decode the bursts although with lower quality when being in difficult radio conditions. The part of the MPE-FEC frame that includes the IP datagrams is called Application Data Table (ADT). ADT has a total of 191 columns. In case the IP datagrams does not fill completely the ADT field, the remaining part is padded with zeros. The division between ADT and RS table is shown in Figure The number of the RS rows can be selected from the values of 256, 512, 768 and The amount of rows is indicated in Service Information (SI). RS data has a total of 64 columns. For each row, the 191 IP datagram bytes are used for calculating the 64 parity bytes of RS rows. Also in this case, if the row is not filled completely, padding is used. The result is a relative deep interleaving as the application data is distributed for the whole burst. 1st byte 191 data columns 64 RS columns 2nd byte IP datagrams... P A D D E D S P A C E P A D D E D S P A C E P A D D E D S P A C E... RS parity bytes... P U N C T U R E D C O L U M N S 256, 512, 768 or 1024 rows Figure The MPE-FEC frame consists of application data table for IP datagrams and RS data table for RS parity bytes. The Reed-Solomon code is based on the polynomial correction method. The polynomial is encoded for the transmission over the air interface. If the data is corrupted during the transmission, the receiving end can calculate the expected values of the data within the certain setting specific limits.

64 28 The RS data of DVB-H is sent in encoded blocks with the total number of of m-bit symbols in the encoded block is n = 2 m 1 [Pos05, p. 22]. With 8-bit symbols the amount of symbols per block is n = = 255. This is thus the total size of the DVB-H frame. The actual user data inside the frame is defined as a parameter value k, indicating the number of data symbols per block. The normal value of k is 223 and the number of parity symbols is 32 (with 8 bits per symbol). The universal format of presenting these values is (n, k) = (255, 223). In this case, the code is capable of correcting up to 16 symbol errors per block. RS can correct the errors depending on the redundancy of the block. For the erroneous symbols which location is not known in advance, the RS code is capable of correcting up to (n-k)/2 symbols. This means that RS can correct half as many errors as the amount of redundancy symbols is added to the block. If the location of the errors is known (erasures), then RS can correct twice as many erasures as errors. If E is the number of errors and S the number of erasures in the block, the combination of error correction capability is according to the formula 2E + S < n. The characteristic of RS error correction is well suited to the environment with high probability of errors occurring in bursts, like in radio interface. This is because it does not matter how many bits in the symbol are erroneous. If multiple errors occur in byte, it is considered as one single error. It is possible to use also other block sizes. The shortening can be done by padding the remaining (empty) part of the block (bytes). These padded bytes are not transmitted, but the receiving end fills in automatically the empty space. FEC consists of block coding and convolutional coding. RS is an example of block coding, where the blocks or packets of bits (symbols) are of fixed size whereas convolutional coding is based on bit or symbol lengths. In practice, the block and convolutional codes are combined in concatenated coding schemes; the convolutional coding does the major part of the process whilst the block code, e.g. RS, carries out the recovery of the remaining errors as much as possible after the convolutional coding. In mobile communications, the convolutional codes are mostly decoded with the Viterbi algorithm. Viterbi algorithm is an error-correction scheme for noisy digital communication links. It is widely used, e.g. in GSM, dial-up modems, satellite communications and in LANs. The performance of the different RS decoding schemes has been investigated in [Jok05] and [Him06]. The MPE-FEC error detection and correction can be done based on conventional nonerasure RS decoding, where a maximum of 32 erroneous RS symbols, or bytes, are allowed on each row of the MPE-FEC frame. Frames of one or more rows with more than 32 errors are interpreted as erroneous. Another option for the MPE-FEC is to use erasure decoding, as has been selected for the DVB-H. In that case, a complete section is marked as unreliable if it contains an error. One section erasure leads to one column erasure in the MPE-FEC frame, when the IP datagram length equals the number of rows in the MPE-FEC frame. For erasure decoding a maximum of 64 erasures are allowed on one row in the MPE-FEC frame. Frames

65 29 consisting of at least one row with more than 64 erasures are considered erroneous. The performance of different decoding schemes depends on the radio channel type, AWGN channel in open areas being more controlled than the multipath channel of city centres. The simulation results of [Jok05] show that non-erasure decoding is substantially stronger in AWGN than multipath cases. The performance of non-erasure decoding is furthermore better in bursty error cases if the errors are uniformly distributed. The transmitted data contains a checksum, which is typically created via CRC-32, but the standard leaves the selection of the terminal s decoding method for the receiver designers. The use of CRC-32 in the receiver is thus optional. Probable reason for the selection of erasure decoding in DVB-H has been the need to reduce the computational complexity of conventional Reed-Solomon decoding algorithms, even if the over-head due to CRC-32 increases regardless of it s use in terminal s side. It is shown in [Him06] that the decoding methods inserting also erroneous data into the MPE- FEC frame are in fact more efficient than erasure decoding methods suggested in the DVB-H standard. The gain seem to exceed 1 db in favour of so called hierarchical transport stream decoding, when compared to the pure section erasure decoding. The results of [Him06] indicates that the end-user s video quality can be increased significantly when allowing erroneous data to be used for decoding and passed to the application layer rather than using erasure decoding methods, where erroneous IP packets or even MPE-FEC frames are discarded. Furthermore, [Jok06] claims that all other investigated decoding methods including the TSE decoding (transport stream erasure derived from the transport stream headers) that ignore the CRC-32 information would perform better than the CRC Erasure decoding. These findings indicate that the optimal setting for the decoding method can improve significantly the subjective DVB-H reception quality level that the end-user experiences Time Slicing The basic idea of Time Slicing functionality is to send the DVB-H specific elementary stream data in bursts with higher data rate than the actual average bit rate of the stream would be. This allows the terminal to get a certain data stream portion in advance and switch off the receiver whilst the terminal buffers the data and presents the video and/or audio content. The obvious advantage of the functionality is the battery saving as the power consumption of the receiver gets considerably lower. According to [Far06], the normal power saving is typically %. The significance of the final power saving of the terminal lowers, though, as there are applications and functionalities, e.g. video streamer and mobile communications transceiver which take their part of the processing power. Time Slicing also provides a method for the seamless handover between the frequencies.

66 30 Figure 2-11 shows the principle of the Time Slicing functionality. The burst window can consist of several time sliced bursts, but the terminal needs to activate itself only when the contents of the own channel is transmitted. The correct dimensioning of the Time Slicing functionality is important as it affects directly on the waiting period when the user changes the time sliced channel. The interval of the time sliced burst can be calculated when the peak bit rate, average ES bit rate and the complete burst size is known. Assuming the average bit rate is 500 kb/s, the peak bit rate 10 Mb/s, and the burst size is 2 Mb, the burst cycle is 4 seconds. In this case, it can be estimated intuitively that the average waiting time of the switching that the user experiences is approximately 2 seconds (an average of the extreme values). Taking the typical DVB-T channel switching time as a reference, it can be estimated that the channel switching time of less than 2 seconds is still tolerable from the user s point of view. Peak bit rate of single burst Single frame Burst Rate (kb/s) On Off Window for the bursts Average bit rate Time (s) Figure The principle of Time Slicing functionality. The header of each MPE section contains a delta-t parameter. It indicates the time for the beginning of the next time sliced burst. In this way it is not necessary to synchronize the receiver and transmitter separately as the timing for the next reception is indicated in each burst SFN / MFN DVB-H can be defined as a single frequency network (SFN) or multi frequency network (MFN). In practice, the network can consist of both modes. One of the feasible strategies is to cover single cities with SFN isles, and the remaining part of the network as MFN in order to inter-connect the isles. When the DVB-H terminal moves from the coverage area of one DVB-H cell to another, the Time Slicing functionality provides the fluent frequency handover in MFN mode. This is a result of the off-time periods of the time sliced bursts, as the terminal can scan the other

67 31 frequencies during this time. The terminal evaluates the best frequency, and when the order change, the terminal executes the handover process during the off-time period. The scanning procedure is implementation dependent as stated in [Dvb07], which also describes various use cases for the handover. Time Slicing provides seamless MFN-handover without disturbing the fluent following of the received contents. The handover process requires though sufficiently good overlapping of the cell coverage areas. In case of the same content delivery via different adjacent cells, the synchronization of the site transmitters is important. This gives the possibility to provide totally transparent delivery of the contents to the users. SFN provides the fluent implementation of the transmitters inside the theoretical limits of the SFN network area. New transmitters can be added to the same frequency basically without extra planning efforts whenever the SFN limits are not exceeded and the correct setting of the transmitter synchronization is taken care of (with GPS timing). In this case, the overlapping parts of the coverage provide additional SFN gain. The functionality and gain due to the combination of the separate signals, e.g. echoes from the individual transmitter as well as signals from separate transmitters, is based on the correlation of the received elements. If the elements are within the window determined by the guard interval, the correlation can be done as shown in Figure 2-12 [Rei05, p. 178]. 10 T GI T U 0 Main channel -10 Echo 1-20 Echo 2-30 Co-channel -40 sum -50 T S -60 Figure The correlation and combination of the separate multi-path signals can be done whilst all the components occur within the time delay determined by GI.

68 32 The COFDM bit stream is distributed over several individual carriers, i.e. the transmission is carried out by spreading the baseband bit stream. A set of carriers form a COFDM symbol. The processing of the symbols is parallel. Before the receiver starts to evaluate the individual symbol, it waits a time window determined by T GI, i.e. during the guard interval in order to collect all the echoes. When the echoes occur within the GI window, they can be combined in order to add energy for the original signal [Rei98, p. 4]. This enhances the carrier to noise and interference ratio of the received signal, resulting gain in the cell coverage and/or capacity. In practice, part of the symbol is copied from the beginning of the symbol to the end, which increases its duration determined by the guard interval [Greg06, p. 26]. The evaluation of the symbol takes place inside the FFT window determined by T U, and the orthogonal criterion is also considered within this original symbol duration, without considering the extended part of the signal. The FFT window position is then selected depending on the received radio components. If the SFN limits are exceeded, i.e., there are sites outside the theoretical SFN area, they start acting as interfering sources. Whilst the level of the combined useful carrier is high enough compared to the total interference level, as Publications III, IV, VII and X show, the exceeding of the SFN limits can be done in controlled way, and it is possible to find optimal parameter sets by balancing the SFN gain and SFN interferences. The assumption of the above mentioned publications is that the total contribution of interfering components, as well as the useful carrier components can be calculated, i.e. the power level of the useful carriers can be summed as well as the interfering components. In the high-level considerations this is a valid methodology that gives possibility to utilize the minimum required carrier-per-noise, and thus C/(N+I) ratio as an investigation criteria by applying these levels according to [Dvb09, p96]. The values depend on the channel type, and in the most accurate considerations the values are also frequency selective over the considered bandwidth. Nevertheless, as a higher-level value set is shown in [Dvb09], the respective C/N and C/(N+I) was not investigated in more detailed level in Publications of this thesis Interaction channel The DVB-H system does not define the uplink communications for the interactions. Nevertheless, the uplink can be included, e.g. via the GSM or UMTS networks. The management of the chargeable channels can also be done by using the local terminal functionality. One possibility for the DVB-H stand-alone terminal can be based on separate codes that are delivered in form of scratch cards. The interaction channel provides means for the opening of the ciphered contents, and it can be used, e.g. for the real time voting type of activities among the program. In theory, there are no

69 33 obstacles in using any other radio interfaces for the interactions, like WLAN. In any case, the most logical combination is the DVB-H and GSM/UMTS modules integrated physically into the same terminal, as the coverage areas of the 2G and 3G mobile networks can be assumed to usually overlap with DVB-H. In a typical DVB-H network planning process, it can be assumed that DVB-H and GSM/UMTS networks are overlapping. The overlapping is most perfect if the DVB-H can be co-sited with mobile communication networks. As commented in [Ung06], in order to build a DVB-H network, position and size of site cells should be selected to maximize the benefit of a hybrid network. The implementation of the DVB-H network on top of the existing 3G and DVB-T network infrastructure is attractive also because it lowers the initial network investments [Had07]. Nevertheless, in practice, there might be locations where the coverage areas of either DVB-H or mobile communications networks are present as shown in Figure In these cases, the interaction channel does not work at that specific moment and/or location, and it is thus impossible to initiate the opening process of a channel that requires separate signalling in order to be used. If the initiation has been done earlier, the channel can be used without the presence of an interaction channel until the possible expiration time of the channel deciphering and scrambling key validity is reached. DVB-H coverage GSM/UMTS coverage Figure The principle of the overlapping GSM/UMTS coverage areas. In case of in-depth radio analysis, it might be interesting to take into account even the effects of packet switched data on the GSM/3G performance if it is used for DVB-H interactions. As concluded in [Pen99], [Pen99b] and [Pir99], the amount of the GSM traffic load can have effect on the useful coverage areas creating a small scale UMTS-type cell breathing in the very edge areas of the network s non-bcch frequencies. In practice, though, the C/I of typical GSM mobile networks is dimensioned high enough in order to provide sufficient overlapping even in the presence of co-channel interferences which minimizes the presence of the outages within the planned coverage areas. Also UMTS is assumed to be dimensioned sufficiently robust for the

70 34 highest load cases. For this reason, the effect of the variations in the return channel availability can be discharged in case of the interactions with DVB-H if overlapping areas are found.

71 35 3 Initial radio network planning process 3.1 High-level network dimensioning process The dimensioning of the high level DVB-H radio network can be summarized by presenting three main variables, i.e. channel capacity, coverage area and quality of the service, which all together have effect on the total cost of the network as illustrated in Figure 3-1. The network with poor capacity, coverage and QoS (Quality of Service) has a minimum cost, but the revenue per customer would also be low in that case due to the unsatisfactory service. On the other hand, the highly overlapping and high-capacity network with excellent outdoor and indoor coverage in large area is excellent, but the cost for building and operating the network might be too high in order to recover the expenses as there is a practical limit for the user fees. Capacity Cost Coverage QoS Figure 3-1. The high-level cross-relations of the most relevant DVB-H radio network planning items. The task is to design a sufficiently high quality network with initial and operating costs that can be recovered in planned time period, e.g. via monthly fees. It can be estimated that the quality of the network is on correct level when the customers are willing to use the service and accepts the technical performance of the services as well as the related usage fee. In the complete network design, it is thus essential to take into account both technical issues as well as their costs, and seek for their balance in order to make sure that the return of investments are on acceptable level. As an example of the cross-relation between the values, by keeping the site number the same, 16-QAM modulation would offer high-capacity with low coverage, whilst QPSK offers less capacity but in larger area with a certain location probability for the coverage.

72 36 A process chart shown in Figure 3-2 was created as a part of this thesis as a basis for the nominal network planning tasks in order to identify the relevant items of the planning and optimisation of the DVB-H radio network in the initial phase. The process chart provides a rough estimation of the site numbers with uniform planning assumptions for all the sites. For the next level of the planning, a detailed analysis is needed that takes into account the realistic site locations and topological variations, different antenna heights and power levels. General nominal radio network assumptions Initial options Select transmitter type (power levels), antenna types (gains), general assumptions for the environment type (% of dense urban/urban/sub-urban, rural) Capacity planning Input: total capacity and/or number of channels with possibly requirement for kb/s. Set the parameters: modulation, FFT, GI that complies with the requirement. Output: set of parameter values that provides the channels. Coverage planning Input: quality level (area location probability). Apply link budget and prediction models, tune antenna height, transmitter power level (max: regulatory limit) Output: radius per cell (generic) No Is the coverage per site good enough? Yes Change the coverage parameter values and/or quality level requirement. Change the capacity requirement Change the transmitter type (maximum power levels), antenna types (gains) yes no Is it possible to tune the coverage related parameters? Is the EMC and radiation safety ok? No Yes Cost optimization Calculate the total network cost (CAPEX, OPEX per x year) Acceptable level of costs vs. technical performance? No Yes Go to detailed planning process Figure 3-2. The radio network planning process in the initial phase. The process contains high-level estimations of the capacity and coverage taking into account the economical and regulatory limitations.

73 Capacity Planning In the initial phase of the DVB-H network planning, the usual procedure is to decide first the offered capacity of the system. The total capacity in certain DVB-H bandwidth defined as 5, 6, 7, or 8 MHz affects also on the size of the coverage area. The dimensioning process is thus iterative, with the aim to find a balance between the capacity, coverage and the cost of the network. The capacity can be varied by adjusting the modulation, guard interval, code rate and channel bandwidth. As an example, the parameter set of QPSK, GI 1/4, code rate 1/2 and channel bandwidth 8 MHz provides a total capacity of 4.98 Mb/s, which can be divided into one or more electronic service guides (ESG) and various audio/video sub-channels with typically around kb/s bit stream dedicated for each. The capacity does not depend on the number of carriers (FFT mode) but the selected FFT affects though on the Doppler shift tolerance. As a comparison, the parameter set of 16-QAM, GI 1/32, code rate 7/8 and channel bandwidth of 8 MHz provides a total capacity of 21.1 Mb/s. It should be noted, though, that the latter parameter set is not practical due to the clearly increased C/N requirement which reduces considerably the useful coverage area and makes the reception very sensible for the variations in the radio interface. Table 3-1 shows the reachable capacity in Mbit/s per total DVB-H channel as function of the relevant parameter values. This useful bit rate shown in the Table would be reduced accordingly with the direct proportion of MPE-FEC rate when it is present, i.e. MPE-FEC rate of 3/4 (about 35 %) results 3/4 from the original value for the user capacity. In this sense, e.g. combination of CR of 1/2 and MPE-FEC 1/2 results in the highest protection over the radio transmission but with lowest capacity. Table 3-1. The summary of total DVB-H bitrates as function of the parameter values as presented in [Dvb09]. The values for 5 MHz band can be extrapolated from the other values. Modul. CR GI=1/4 GI=1/8 GI=1/16 GI=1/32 6MHz 7MHz 8MHz 6MHz 7MHz 8MHz 6MHz 7MHz 8MHz 6MHz 7MHz 8MHz 1/ / QPSK 3/ QAM 64- QAM 5/ / / / / / / / / / / /

74 38 The first task of the initial capacity planning is thus to select the whole parameter value set that complies with the target capacity value. As an example, if the target capacity value is (minimum of) 8 Mb/s, and the bandwidth is 8 MHz, Table 3-1 indicates the compliant parameter values for QPSK that are {GI=1/4, CR=5/6}, {GI=1/4, CR=7/8}, {GI=1/8, CR=3/4}, {GI=1/8, CR=5/6}, {GI=1/8, CR=7/8}, {GI=1/16, CR=3/4}, {GI=1/16, CR=5/6}, {GI=1/16, CR=7/8}, {GI=1/32, CR=2/3}, {GI=1/32, CR=3/4}, {GI=1/32, CR=5/6} and {GI=1/32, CR=7/8}. For the 16-QAM and 64-QAM modulations, all the GI and CR combinations complies with the minimum requirement of 8 Mb/s in this case. Knowing that the MPE-FEC rate will reduce the final useful data with direct proportion, the set of the compliant parameters is reduced in following way when the MPE-FEC is added: For MPE-FEC 7/8: {QPSK, GI=1/8, (CR=5/6, 7/8)}, {QPSK, GI=1/16, (CR=5/6, 7/8)}, {QPSK, GI=1/32, (CR=5/6, 7/8)}, {16-QAM, GI=all, CR=all}, {64-QAM, GI=all, CR=all} For MPE-FEC 5/6: {QPSK, GI=1/8, CR=7/8}, {QPSK, GI=1/16, (CR=5/6, 7/8)}, {QPSK, GI=1/32, (CR=5/6, 7/8)}, {16-QAM, GI=all, CR=all}, {64-QAM, GI=all, CR=all} For MPE-FEC 3/4: {16-QAM, GI=1/4, (CR=2/3, 3/4, 5/6, 7/8)}, {16-QAM, (GI=1/8, 1/16, 1/32), CR=all}, {64-QAM, GI=all, CR=all} For MPE-FEC 2/3: {16-QAM, GI=1/4, (CR=2/3, 3/4, 5/6, 7/8)}, {16-QAM, GI=1/8, (CR=2/3, 3/4, 5/6, 7/8)},{16-QAM, GI=1/16, (CR=2/3, 3/4, 5/6, 7/8)}, {16-QAM, GI=1/32, CR=all}, {64-QAM, GI=all, CR=all} For MPE-FEC 1/2: {16-QAM, GI=1/4, (CR=5/6, 7/8)}, {16-QAM, GI=1/8, (CR=3/4, 5/6, 7/8)}, {16-QAM, GI=1/16, (CR=3/4, 5/6, 7/8)}, {16-QAM, GI=1/32, (CR=2/3, 3/4, 5/6, 7/8)}, {64-QAM, GI=1/4, (CR=2/3, 3/4, 5/6, 7/8)},{64-QAM, (GI=1/8, 1/16, 1/32), CR=all} As the increased capacity reduces respectively the radio coverage, the task is to find a parameter combination that complies with the original capacity requirement with a possible margin that should be decided beforehand. The division for the respective categories can be decided in such a way that the excess of the capacity of, e.g % is still acceptable and complies with the target capacity dimensioning. If the excess is, e.g %, it can be called as slightly overdimensioned capacity, % can be categorized as clearly over-dimensioned, and >50% can be considered as heavily over-dimensioned. In this example, the parameter values that comply with the target value of Mb/s are the following:

75 39 For MPE-FEC off: {QPSK, GI=1/4, CR=5/6}, {QPSK, GI=1/4, CR=7/8}, {QPSK, GI=1/8, CR=3/4}, {QPSK, GI=1/16, CR=3/4}, {QPSK, GI=1/32, CR=2/3} For MPE-FEC 7/8: {QPSK, GI=1/8, CR=5/6}, {QPSK, GI=1/8, CR=7/8}, {QPSK, GI=1/16, CR=5/6}, {QPSK, GI=1/32, CR=5/6}, {16-QAM, GI=1/4, CR=1/2} For MPE-FEC 5/6: {QPSK, GI=1/8, CR=7/8}, {QPSK, GI=1/16, (CR=5/6, 7/8)}, {QPSK, GI=1/32, (CR=5/6, 7/8)} For MPE-FEC 3/4: {16-QAM, GI=1/8, CR=1/2}, {QPSK, GI=1/16, CR=1/2} For MPE-FEC 2/3: {16-QAM, GI=1/32, CR=1/2} For MPE-FEC 1/2: {16-QAM, GI=1/4, (CR=5/6, 7/8)}, {16-QAM, GI=1/8, CR=3/4}, {16-QAM, GI=1/16, CR=3/4}, {16-QAM, GI=1/32, CR=2/3}, {64-QAM, GI=1/8, CR=1/2}, {64-QAM, GI=1/16, CR=1/2} When the parameter value set is selected for the coverage planning, some high-level rules should be already known about the effects of the MPE-FEC, modulation, GI and CR. As an example, the optimal performance of MPE-FEC depends on the environment, and it functions best when the field strength is low enough, and impulse noise is present. Publication II shows that the MPE-FEC does have a clear benefit in the extending of the coverage areas. Furthermore, it can be estimated, according to the results of Publication VIII, that in environments with good field strength, MPE-FEC does not bring added value in outdoor environment. In indoors, the effect of impulse noise can be lowered and the frame error rate in the low field in site cell edges inside the buildings can be lowered within a relatively small window. The selection of MPE-FEC depends thus on various aspects, but based on this information, the strongest MPE-FEC rates are not recommendable to be used in network parts with sufficiently high field as they waste the capacity. It has been identified also in [Apa06a, p. 4] that the combination of high code rate and low MPE-FEC rate gives better balance between the capacity and coverage compared to the low code rate and high MPE-FEC rate. On the other hand, as concluded in Publication VIII, it is not recommendable to switch off the MPE-FEC as it reduces the impulse noise and helps to extend the useful coverage when the terminal is located in the edge area of the site cell. QPSK is the most robust in the DVB-H set of modulations. It provides largest coverage areas, but lowest capacity. As [Rei05, p. 172] and [Law01, p. 65] shows, the bit error rate of for QPSK (4-QAM) requires about 8.2 db E b /N o, whereas the requirement for the 16-QAM is about 12.1 db and for the 64-QAM about 16.4 db. This indicates that the 16-QAM increases the path loss approximately 4 db compared to QPSK when applied as such on the radio link budget. On the other hand, 16-QAM provides a double capacity compared to QPSK. 64-QAM further

76 40 decreases the coverage with an additional 4 db, approximately. Based on the practical field tests carried out in Publication II, 64-QAM is shown to be very sensitive to the varying radio channel conditions and is thus not recommendable as the primary choice for large areas. When investigating further the case results of Publication II, Tables I III, the QEF point of BER, i.e , is obtained typically with about 7 8 db stronger C/N values for 16-QAM compared to QPSK. According to the case results of Publication II, the QEF point in case of the 64-QAM seem to require typically more than 20 db compared to QPSK, which indicates strong practical challenges for 64-QAM in mixed radio channel types although the results in this specific case were obtained by collecting the data with a prototype terminal. Even if the practical 64-QAM performance might require higher C/N than indicated in theory, there are locations where 64- QAM can be used in isolated pedestrian environment, like airports and shopping centres with a clearly isolated SFN or MFN area. If SFN mode is used, the smallest GI values provide small functional area as can be seen in Table 2-1. It means that when using only one or two frequencies and there is need to cover large areas, GI-values of 1/32 and 1/16 are not recommendable. The largest SFN area can be obtained by using the GI value of 1/4. On the other hand, the small GI values provide more Doppler tolerance. According to the above information, in this specific case, it would be logical to select the following settings as primary option for the first iteration of the capacity planning phase: {16- QAM, GI=1/4, CR=1/2, MPE-FEC 7/8}, {16-QAM, GI=1/8, CR=1/2, MPE-FEC 3/4} or {QPSK, GI=1/16, CR=1/2, MPE-FEC 3/4}. The outcome of this first step is then a compliant set of parameters for the capacity demand. If the forthcoming analysis of coverage does not produce desired plan, the initial capacity target should be changed and the process repeated. If the first parameter selection is not considered feasible, the effecting parameters should be revised until the wanted capacity can be achieved with required coverage and quality of the network. 3.3 Coverage and QoS Planning [Publications V, VII] When the capacity requirement and respective radio parameter value set is known, the next step of the DVB-H network dimensioning is to estimate the coverage of the site cells. The major items for the first hand coverage estimation are: coordinates of the transmitter, radiated power, frequency and antenna pattern [Goe02]. In addition to the antenna height, radiating power and radio path loss in different environments, the area depends on the required quality level of the reception. The estimation of the radio channel type is important in this phase as it includes the fading profile and has thus effect on the link budget and Doppler tolerance. In order to have the first estimation of the number of the transmitters, the nominal plan can be carried out by

77 41 assuming an ideal distribution of the sites and most probable channel type. The practical radio network is always somewhat non-ideal as for the site locations, so the final plan must be tuned accordingly, by using possibly various different power levels and antenna heights. The coverage holes, e.g. in street canyons and indoors, can be further enhanced by using separate DVB-H gap-fillers. As there is time and location dependent fluctuation in the received power, the dimensioning is done by estimating the probability for the reception of the sufficiently high-level signal, i.e. the task is to design the quality target of the coverage. This margin is presented by the location variation (db) in the link budget. The margin is estimated for the coverage area over the whole site cell Link budget When the coverage criteria are known, the site cell radius can be estimated by applying the link budget calculation. As the DVB-H is a broadcast system, the link budget is calculated only for downlink. For the possible interaction channel, the respective downlink and uplink path losses can be estimated by applying the separate link budget of the used system for the interactions (e.g. GSM/GPRS or UMTS). In the normal planning case, though, it can be assumed that the coverage area of the interaction channel is present where DVB-H is found. The generic principle of the DVB-H link budget can be seen in Table 3-2. The calculation shows an example of the transmitter output power level of 2,400 W, with the quality value of 90% for the area location. There are 4 different cases shown in the table, varying the modulation and MPE-FEC rate. The SFN gain has assumed to be 0 db in these cases. According to the link budget, the outdoor reception of this specific case with 2,400 W transmitter yields a successful reception when the radio path loss is equal or less than db. The principles of Table 3-2 have been used throughout of the publications of this thesis, including the SFN simulator presented in Annex A. The interpretation of the quality of the coverage area depends on the agreed area location probability level. In general, the location variation is considered to follow a log-normal distribution [Bee07], meaning that the logarithm of the signal level follows a normal or Gaussian distribution. The statistical distribution should be applied in the respective quality level estimations. The mean value basically means that 50 % of the samples are above this value and the other half below. In case of any other percentage for the coverage quality criteria, the relationship between the mean value and standard deviation should be known. The standard deviation of 5.5 db is normally used in the typical sub-urban type of environments of DVB-H. The standard deviation is commonly used as a basis for the mobile communications coverage predictions, informing about the confidence in statistical conclusions.

78 42 Table 3-2. An example of the DVB-H link budget. Case: Modulation: QPSK QPSK 16-QAM 16-QAM DVB-H Link Budget CR: 1/2 1/2 1/2 1/2 MPE-FEC: 1/2 2/3 1/2 2/3 General parameters Variable Unit Frequency f MHz Noise floor for 8 MHz bandwidth P n dbm RX noise figure F db TX Transmitter output power P TX W Transmitter output power P TX dbm Cable and connector loss L cc db Power splitter loss L ps db Antenna gain G TX dbi Antenna gain G TX dbd Eff. Isotropic radiating power EIRP dbm EIRP W Eff. Radiating power ERP dbm ERP W RX Min C/N for the used mode (C/N) min db Sensitivity P RXmin dbm Antenna gain, isotropic ref G RX dbi Antenna gain, 1/2 wavelength dipole G RX dbd Isotropic power P i dbm Location variation for 95% area prob L lv db SFN gain G SFN db MPE-FEC gain G MPE-FEC db Building loss L b db GSM filter loss L GSM db Min required received power outdoors P min(out) dbm Min required received power indoors P min(in) dbm Min required field strength outdoors E min(out) dbuv/m Min required field strength indoors E min(in) dbuv/m Maximum path loss, outdoors L pl(out) db Maximum path loss, indoors L pl(in) db The relationship of the area location probability and the additional margin that should be taken into account in DVB-H link budget can be thus derived from the characteristics of the normal and log-normal distribution, e.g. by observing the attenuation points (db) in cumulative scale that fulfils the required percentage of the area location (in the whole area). Furthermore, it can be decided that 90 % of area location probability indicates a decent outdoor coverage, whilst 95 % is considered good and 99 % is providing excellent quality. Table 3-3 summarizes the

79 43 mapping of the typical values that can be used in the DVB-H planning for mobile reception, with standard deviation of 5.5 db [Dvb09, p. 95]. In addition to the standard deviation, the criteria vary depending on the environment, i.e. on the propagation slope. Slope of 2 (i.e. 20 db/decade) represents line of sight in free space. The slope of 3.5 (i.e. 35 db/decade) is used in Table 3-3, representing typical urban environment. Table 3-3. The relationship of the area location probability in the site cell edge and over the whole site cell area for the mobile reception when the standard deviation is 5.5 db, according to [Dvb09]. Area location prob. (minimum coverage target) Loc probability in site cell edge Location correction factor Subjective quality description 90 % 70 % 7 db Fair outdoor 95 % 90 % 9 db Good outdoor, fair indoor 99 % 95 % 13 db Excellent outdoor, good indoor [Mil06, p. 31] presents the correction factors as function of the reception environment: pedestrian 90% location area probability results 7.1 db, pedestrian 95 % location 9.0 db, indoor 90 % location 10.4 db, indoor 95 % location 13.3 db, mobile 90 % location 9.0 db and mobile 99 % location 12.8 %. In [Bmc09], the area types have been further divided into different classes, i.e. outdoor pedestrian (A), light indoor (B1), deep indoor (B2), mobile roof-top (C) and mobile in-car (D). [Bmc09] proposes that for class A and B, a good coverage quality corresponds to 95 %, and acceptable to 70 % area location probability whilst class C and D corresponds to values 99 % and 90 %, respectively. It is thus important to clarify the level of the quality in such a way that no misinterpretations may occur in the requirement levels. In addition to the quality related terminology, it should be noted that in DVB-H, cell coverage refers to the coverage area of one or more sites belonging to a certain SFN. In this work, the term site cell refers to a coverage area of an antenna system of a single transmitter. The antenna can be omni-radiating or a set of directional antennas. The more comprehensive definition of the cell can be found in EN [4]. In the DVB-H system, the cell_frequency_link_descriptor indicates the frequencies that are used for the different cells of the network. The frequencies (and thus cells) are furthermore mapped with Transport Streams. The cell_list_descriptor contains the information of the coverage area of the cells. Physically, cell is defined as a geographical area covered by the signals that contain one or more transport streams, which can be done with one or more transmitters. In the simulations of this thesis, hexagonal model with omni-radiating antennas are used for the coverage estimate. In Publication V, though, sectorized site cells are used with realistic topo-

80 44 logical information about the surrounding areas in order to estimate the DVB-H coverage areas in dense urban area. Path loss The maximum possible path loss is the difference between the radiating transmitter power and the received power: L db) = P ( db) P min( ( dbm) (3-1) ( EIRP out) In this formula, the radiating power P EIRP is: P ( dbm) = P L L + G (3-2) EIRP TX cc The minimum received power level P min(out) is: ps TX P min( + L (3-3) out) = Pi + Llv GSFN GMPE FEC GSM where P i is the isotropic received power, L lv is the location variation for a certain area probability (that can be obtained from Table 3-3), G SFN is the SFN gain, G MPE_FEC is the MPE- FEC gain, and L GSM is the GSM filter loss (in order to isolate the DVB-H receiver and GSM transmitter). The isotropic received power is obtained from: P = P min G (3-4) i RX RX where P RXmin is the receiver sensitivity, or the minimum power level the receiver functions, and G RX is the receiver's antenna gain. The latter depends on the frequency. Based on the information of [Dvb09, p. 87] a linear interpolation may be utilized: 5 f f 2370 G RX [ dbi] = 10 = 10 = f (3-5) The minimum required receiver's power level P RXmin is obtained by: P RX min Pn + ( C / N) min = (3-6) where P n is the receiver noise input power level, and (C/N) min is the minimum functional C/N (which depends on the used modulation, CR and MPE-FEC rate). The receiver noise input power can be presented further by:

81 45 [ dbw ] F 10 log( ktb) P n = + (3-7) where F is the receiver's noise figure (component dependent) and P t =10 log(ktb) is the thermal noise level, k is Boltzmann's constant J/K, T is the temperature in Kelvin scale (290 K is normally used as an average value) and B is the receiver's noise bandwidth (Hz). In DVB-H, the value for B is 7.6 MHz in case of the 8 MHz variant. As an example, for the 8MHz band, the thermal noise floor is dbm. Combined with the terminal's noise figure of 5 db (this depends on the quality of the receiver's components, and it has frequency dependency), the P n would be dbm. There is still one important item that should be taken into account when estimating the final maximum allowed path loss. This is the loss that the transmitter filter absorbs from the radiating power. In the general calculations, it can be estimated as 10% of the radiating power level (W). The effect can be taken into account when the transmitter power level is presented in dbm: ( P [ W ] 0.9) effective TX PTX [ dbm] = 10 log (3-8) The complete formula for the maximum path loss is thus: L( db) = 10 log + ( P [ ] ) TX W 0.9 [ dbm ] L [ db ] L [ db ] + G [ dbi ] NF [ dbm ] ( log( ktb) [ dbm] ) ( C / N) min[ db] + ( f [ MHz] )[ db] Llv[ db] G [ db] + G [ db] L [ db] SFN 1 10 MPE FEC 3 GSM cc ps TX + (3-9) Building loss The building loss, or building penetration loss, can be estimated in general level as an average value in different environment types, and it can thus be considered as a fixed radio link budget value in respective area types. The DVB-H implementation guideline recommends 11 db of median value and standard deviation of 6 db to be used for the building loss [Dvb09, p. 94]. In addition to the building penetration loss, i.e. the ratio of the average powers measured outside and inside the building with a fixed transmitter, also building floor loss can be important to understand in detailed network planning. Reference [Jos07, p 3008] has concluded that the floor loss can be in certain cases approximately between 30 and 40 db. In [Bmc07], it is stated that the building loss is frequency dependent. This is logical as the radio wave penetrates into the buildings depending on the conditions. As an example, a typical urban building normally contains metallic supports that might create a Faraday cage. Depending on the wavelength of the signal and the hole-size of the supporting metal the respective attenuation varies. Reference [Bmc09, p. 12] includes typical building penetration losses for the general use of the radio link budget. They have been adjusted from the previous [Bmc07] link budget document. The updated table shows values of 7 db for class D (in-car) for all the bands VHF,

82 46 UHF, L-band and S-band with no standard deviation. For the class B1 (light indoor), the penetration loss is given as 9 db (with standard deviation σ p of 4.5) for VHF, 11 db (σ p = 5 db) for UHF, 13 db (σ p = 5 db) for L-band and 14 db (σ p = 5 db) for S-band. For the class B2 (heavy indoor), the respective values are: 15 db (σ p = 5 db), 17 db (σ p = 6 db), 19 db (σ p = 6 db) and 19 db (σ p = 6 db). The snap-shot measurements carried out during the measurements of Publication IX correlates with the above mentioned information. There were two building types investigated, first case (A) being an 8-floor hotel with open centre area representing deep indoor (class B2), and the second (B) being lighter 1-floor construction (class B1). In both cases, the received power level was stored in a slow-moving pedestrian radio channel type by walking outside of the building on the side of the transmitter antenna, and then repeating the round in ground floor inside the centre of the building. A UHF frequency of 701 MHz was used in the tests. The buildings were located in a typical sub-urban area. It should be noted, that the building loss might vary considerably depending on the environment and building material. Figure 3-3 shows the cumulative normalized histogram of RSSI values for the first case (A). This format gives the 50- percentile of the indoor and outdoor. Cumulative percentage Building penetration loss Indoors Outdoors db Figure 3-3. A case example of typical building loss in format of cumulative RSSI histogram, which is obtained by the difference of the indoor and outdoor received power levels. The case represents deep building type of B2. The Figure is a result of post-processing of the Figure 25 of Publication IX. The results show that the difference of the average RSSI values of indoor and outdoor, i.e. building loss, for the first case A representing class B2 is 16.1 db with a standard deviation of 5.8 db in indoor and 3.9 db in outdoor. The respective value set of [Bmc09, p. 12] is very close to this result, the difference being about 1 db. Comparing the building loss via the 50-percentile values the result would be about 18 db. For the second case (B) representing the class B1, the building loss calculated via the differences of the average values is 14.1 db, with standard

83 47 deviation of 6.1 db in outdoor and 2.9 db in indoor. The difference with the [Bmc09] is now about 3 db. The respective 50-percentile comparison indicates about 13 db building loss. As both cases show differences between the average and respective 50-percentile values, it is important to define which the used method is for the building loss. In [Bmc09, p. 11], the average values are used, and the building penetration loss L P is obtained by comparing the signal level distributions E inside and outside of the building: L P = E E (3-10) out _ average in _ average The examples presented in this thesis show that sufficiently amount of field tests clarifies the typical building loss values that can be used for the local adjustment of the link budget. Although only two snap-shot cases were investigated in Publication IX (shown in the Publication s Figure 25), they correlate with [Bmc09, p. 12] indicating that the respective building loss values can be used as default ones in the radio link budget until possible more indepth local modification is made. Effect of antenna height: Receiver The link budget items that can be varied are the requirements for the location variation (i.e. area location probability). This has a direct impact on the expected quality level of the radio network within a planned coverage area. The mobile environment affects on the coverage area of DVB-H compared to the DVB-T. The planning assumption of the DVB-H link budget is outdoors with 1.5 meters of terminal height with typically N-LOS in the city area. This causes a penalty for the DVB-H link budget compared to the DVB-T that is typically based on the fixed rooftop antenna with LOS [Far07]. [Mil06, p. 32] indicates that the receiver antenna height loss can be 11 db for rural area, 16 db in suburban and 22 db in urban area in Band IV. For the band V, the respective values are 13, 18 and 24 db. Effect of antenna height: Transmitter As Figure 3-4 and Publications I and V indicate, the height of the DVB-H transmitter site antenna has a key role in the coverage area of the DVB-H site cell. By observing Figure 3-4, the doubling of radiating power level, i.e. adding 3 db to the radio link budget might raise the radius of the site cell by about 30% in the typical DVB-H antenna heights, meaning that the coverage area would enhance around 70%. This could happen, e.g. by changing the 2400 W transmitter model to 4700 W model and using the antenna in 60 m height. On the other hand, if the transmitter antenna height is moved from 60 m to 85 m (40% higher) but using the same 2,400 W transmitter, the effect of the coverage area would be the same as doubling the power.

84 48 SFN gain One possible item in the link budget enhancement is the SFN gain that can enhance the performance in the overlapping areas or provide the theoretical possibility to construct the sites further away from each others as the coverage area of each site cell rises. The interpretation of the benefit of SFN thus varies. The implementation guidelines [Dvb09, p. 78] mentions that there is a potential SFN diversity gain though without specifying more concrete values. The SFN gain in general has been noted as useful in [Zir00] and [Cha06]. On the other hand, [Bmc09, p. 15] recommends that SFN gain would not be taken into account in the radio link budget. Nevertheless, the simulations carried out in Publication III show that the SFN gain is present in an SFN network that does not contain interfering sites (i.e. the distance of the extreme sites is inside the distance determined by GI). Furthermore, Publication IV has concluded that as the number of the sites grows, the SFN gain grows up to about 6 db level in the theoretical case where the absolute signal power levels are summed in direct power sum format in a noninterfered environment. With the parameter sets that provide smaller SFN sizes, as the number of sites grows, part of the sites starts to add interference reducing the SFN gain. The interference can be inter-symbol or noise type. Depending on the radio parameter set (FFT size and GI), the balance can still be achieved by adjusting the transmitter antenna heights and power levels, but some of the parameter values leads to the highly interfered network as shown via the simulations in Figures 6 7, and of Publication VII. The challenge of using SFN gain in the link budget is that there is no single definition of the SFN gain. It could be interpreted as the difference between the received power levels in db, comparing a single stream with a varying number of streams as has been presented in [Ple08]. The information could be used to map the C/(N+I) distribution behaviour in more in-depth way of analyzing the gain, as has been presented in the simulations of Publications III, IV and X. Minimum C/N The minimum C/N ratio that is required depends on the combination of the modulation, code rate and MPE-FEC, and also the radio channel type has clear effect. The information about the modulation and code rate dependency can be seen in [Dvb09] and [Bou06, p. 27]. The respective values have been used throughout in the Publications of this thesis. The values might not be final, though, but it can be assumed that the values are close to the reality for different channel types. MPE-FEC The MPE-FEC has been designed to DVB-H in order to provide additional protection level for the radio transmission, enhancing the quality of the signal. The MPE-FEC rate can vary between 15 50% (with option of no use at all). According to [Dvb09, p. 14] it is suitable for the improvement in C/N performance in mobile radio channels. It also gives additional protection against impulse noise, and enhances the performance of fast moving terminals by adding the

85 49 Doppler shift resistance. The MPE-FEC gains depend on the environment. In general, the closer to AWGN type the radio channel is, the less benefits MPE-FEC offers. On the other hand, the MPE-FEC gain is more notable in Rayleigh type of fast fading channel as the OFDM can utilize the separate radio components inside of the SFN area providing this additional gain. Among various others, [Him09] indicates a clear advantage in the use of MPE-FEC. The effect might be in order of several db. In addition to the streaming services, MPE-FEC is also useful for the file-cast mode of DVB-H. According to [Gom07, p. 5], the needed time for the file transfer is considerably reduced, and more content can thus be delivered with the same infrastructure. On the other hand, if the transmission time is kept the same, the area coverage for reliable reception is enlarged. Publications II, VIII and X have identified behaviour of MPE- FEC with different parameter combinations and area types that is in align with the common understanding about the benefits of its functionality. According to these results, depending of the parameter settings and radio channel type, the MPE-FEC can move the 5 % frame error rate point up to 7 db in RSSI scale, indicating that in the best case the additional error correction can enhance considerably the link budget. Other effects The seasonal conditions might cause low-level yearly fluctuations in the radio propagation due to the variations of the moisture level of vegetation and weather conditions (e.g. causing occasional tunnelling effects in ionosphere). As an example, [Apa06a, p. 64] has informed that the field measurement results do have deviation due to rain. It can be assumed though that in a typical link budget calculation, the effect of vegetation and rain can be estimated to be minimal for DVB-H in VHF/UHF bands. In case of the 1.6 GHz version, the effect might be more considerable but according to the more sensitive behaviour of high frequencies. In any case, the item can be considered as a minor detail in practical radio link budget calculations, and due to the challenges in the seasonal adjustment of broadcast type of network, it is not necessary to take into account in the possible variations. As an additional item, the reception antenna diversity could be utilized to exploit the multipath propagation. According to [Bmc09, p. 15], this feature may not be implemented on all devices, so the effect is not to be considered in the link budget until the penetration of the terminals containing possible receiver diversity or MIMO type of functionality is sufficiently high Propagation models The most important radio related task in the nominal as well as in detailed planning is to estimate the DVB-H coverage area with the given parameters. There are various models available that are based on the radio propagation theories and experiments. There are also interpolation methods presented [Bac04]. The outcome is typically a method that can be applied

86 50 for the mathematical calculation of the estimated site cell radius. Some of the widely used experimental models are Okumura-Hata [Hat80], Cost 231-Hata [Cos99] and ITU-R [Itu07] based methods. These type of models divide the formulation depending on the rough division of the area type, e.g. in urban, sun-urban and open areas. Cost 231-Walfisch-Ikegami is a slightly different model as it tends to quantify the propagation environment. Typically after the initial presentation of the models, there have been various validation rounds that have been confirming the functionality, or have adjusted the models closer to the reality. As an example, the original Cost 231-Walfisch-Ikegami had a minor error in the initial presentation which was found later. Furthermore, the functionality of the model has been investigated, e.g. in [Jeo01], which concluded that the results of the model are relatively close to the ones obtained from the Okumura-Hata based models especially when the building group height is close to the value of half of the street width. The original Okumura-Hata propagation model [Hat80] is useful in the approximate coverage estimation of DVB-H in many cases especially in the nominal radio network planning phase. As an example, the estimated path loss L (db) in large city type can be obtained by: L( db) = lg( f ) 13.82lg( h + [ lg( h )] lg( d) BS BS ) a( h MS ) i (3-11) where h BS is the height of the DVB-H transmitter antenna (in range of m), h MS is the height of the receiver, and d is the distance between the antennas. For the frequency range of ( MHz), the area type factor for the large city is: ( log( )) a ( h MS ) LC = 3.2 hms (3-12) The maximum distance up to 20 km between the site antenna and mobile terminal can now be obtained: L( db) [ log( f ) lg( h ) a( h ) ] log( hbs ) BS MS i d = 10 (3-13) For medium-small city type, the correction factor is: ( 1.1log f 0.7) h ( 1.56 log 0.8) a( hms ) SMC = m f (3-14) For the sub-urban area type, the path loss L of Formula 4-1 is used as a basis with the following: 2 f Lsub urban = L 2 log 5.4 (3-15) 28 Finally, the open area can be obtained with the following correction:

87 51 2 ( log f ) log L open = L 4.78 f (3-16) Figure 3-4 presents the estimated site cell range of the example calculated with the large city model and by varying the transmitter antenna height and power level. As can be noted, the antenna height has the major impact on the site cell radius compared to the transmitter power level. Other suitable model for basically all of the DVB-H environments is ITU-R P.1546 [Itu07]. The model is based on pre-defined curves for the frequency range of 30 MHz to 3,000 MHz and for maximum antenna heights of 3,000 m from the surrounding ground level. The model is valid for the terminal distances of 1 to 1,000 km from the base station over terrestrial and sea levels, or in the combination of these Cell radius d (km) P=1500W P=2400W P=3400W P=4700W Antenna height (m) Figure 3-4. Example of the DVB-H ranges (radius) for selected cases, when 16-QAM, CR 1/2 and MPE-FEC 1/2 are applied. No SFN gain nor MPE-FEC gain are included. If the investigated frequency and antenna height does not coincide with the pre-defined curves, the correct values can be obtained by interpolating or extrapolating the pre-defined values according to the annexes of [Itu07] and applying the methods presented in its Annex 5. The presented curves represent field strength values for 1 kw effective radiated power level (ERP), and the curves have been produced for 100 MHz, 600 MHz and 2 GHz. The curves are based on the empirical studies about the propagation in certain conditions. In addition to the graphical curve format, the values can be obtained also in tabulated numerical format. The ITU-R P.1546 method has been evaluated in different sources. Reference [Öst06] has concluded that in the rural area of Australia, P and P provide better overall predic-

88 52 tion of path loss compared to traditional models like Okumura-Hata model. The comparison also shows that P on average underestimates the field strength by more than 10 db in that area type. Nevertheless, it was shown that P improves the standard deviation of the prediction error compared to previous versions of the ITU-R P This result correlates with [Tun05] which has concluded that the accuracy of the ITU-R P.1546 is consistent with Okumura-Hata model up to about 20 km for urban areas. For rural areas, predicted field values of ITU-R P-1546 model differ from the reference solution more than those of the older ITU models do. At the moment, the latest version of the model is ITU-R P [Itu07]. It was used in the simulation of the performance of a mountain site in Publication VII. It can be assumed that the basic and extended version of Okumura-Hata as well as ITU-R P models provide first-hand estimate for the DVB-H coverage areas and respective capacity and quality levels in the initial network planning phase. These models have been designed for environments with antenna heights and site cell distances that fall into the typical assumptions of DVB-H networks. The [Mil06] identifies several other models, including raytracing type of estimates for the dense city centres. These models require more detailed map data with respective height and cluster attenuations. In the most advanced versions, a vectorbased 3D map is needed. It logically has a cost effect on the planning but increases considerably the accuracy of the coverage estimate. It can further be enhanced via reference measurements by adjusting the estimate respectively. Even with the cost effect, it would be thus justified to use 3D models in the advanced phase of the radio network planning in the most important areas. 3.4 Safety distance [Publication VI] Preliminary calculation about the power levels should be done already in the initial radio network planning. In this stage, regulatory rules, rough estimate about EMC and safety zones dictate the guidelines for the minimum distance of the DVB-H antennas and population or antennas of other systems. In the initial phase, it is sufficient to identify the high level regulatory limits for the radiation. Typical maximum allowed value might be in order of 50 kw (EIRP), with additional rules as function of the antenna height and environment. The regulation provides first-hand information about the limitations that should be taken into account in the nominal plan. The radiation level might need to be limited further depending on the area type (urban or open), the antenna height, and the frequency. The practical limitations might mean that the highest power class transmitters and high-gain directional antennas can not be used in the implementation but the level should be revised case basis. The detailed safety zone and EMC limit calculations can be carried out when the concrete site locations are known. The antenna distances depend on the rooftop and tower installation, or

89 53 indoor antenna installation. In a typical broadcast tower case, the main task is to calculate the EMC limitations, i.e. the interferences the DVB-H causes to others and vice versa. In the rooftop installations, also the safety limits for the human exposure should be taken care with related safety zone markings for the occasional maintenance visits etc. Publication VI shows selected examples about the safety distance calculation principles especially in DVB-H environment in typical cases of rooftops or telecom or broadcast towers. A simple but functional model that is suitable for the DVB-H deployment is presented for the safety distance calculations. Although the safety distance calculations have been applied for the previous mobile communications systems, no other references were identified that would present the interpretation of the EMC and human exposure limit estimation is the same manner as in Publication VI. In both of the cases the electrical field is important to calculate in relevant directions. When the field is calculated right above or below the antenna, the attenuation factor of the vertical radiation pattern should be taken into account accordingly. The methodology applies in the close distance of the site, and the outcome is to minimize the interference level caused by DVB-H to the other systems nearby, as well as to make sure the human exposure limits are not exceeded in sites where people might be present nearby. Although the DVB-H frequency usage is regulated, there might also be need to calculate some special cases in longer distances, like safety zones in the same area where space signal reception stations or military bases are present. In these cases, the related safety zone calculation is straightforward and the methodologies of, e.g. [Chu00] can be utilized. In the site installations, it can be expected that the total exposure increases clearly close to the sites due to the DVB-H installations. Field measurements presented in [Ple09, p. 332] show that at most of the locations around the transmitter site, DVB-H is the dominating source of radiation. The methodology presented in Publication VI gives an estimate of the effect, and field tests can be performed in selected locations especially at the rooftop sites in order to fine-tune the calculations in practice. The outcome of the safety distance calculations in the nominal planning phase is to reject the radiation levels from the planning assumptions that exceed the regulatory limits, in order to minimize unnecessary planning with the above mentioned levels as they cannot be applied after all in the deployment phase. 3.5 Cost prediction [Publication I] In the initial phase of the cost estimation, the strategy is to estimate only roughly the capital expense level. This gives an idea about the general amount of costs varying the most important parameters and identifying only the main items. The cost effect might be challenging to perform due to the lack of information as stated in [Had07, p. 11], but especially in the initial phase of

90 54 the network planning, a rough estimation can be utilized for the DVB-H specific components if no market data are available. For the rest of the costs, e.g. as for the site construction, antenna and feeders and transmission, typical mobile communications solutions can be used as a basis for the cost estimates. As an example, the possible DVB-H transmitter list could be limited to the models with output power levels of W. The cost of each transmitter includes common parts as well as additional ones. The common parts are the equipment shield and transmission module. Depending on the model, there is a varying amount of power amplifier units that might fit into the same shield, but the higher power levels might require liquid cooling instead of air cooling. This means, that the price of the transmitters does not grow linearly as function of the power level, but there is a technical as well as marketing related dependency between the models and their cost effects. Figure 3-5 shows an example about the transmitter price level behaviour when the price of the power level (W) per transmitter type is investigated. The price level has been normalized taking the 500 W-transmitter as a reference (normalized to 100 %). The next step is to identify the other common and variable costs for the individual site setup. The main elements might include the cost of the installation, civil works, antenna system (antenna elements, cables and other related material) Relative TX cost (ref: 500-W TX) Transmitter power (W) Figure 3-5. An example of the transmitter cost depending on the power level. The exercise can be done for the initial parameter set, i.e. according to the wanted capacity and the estimated average antenna heights. The coverage can be obtained for an individual site according to the suitable radio path loss prediction model. The cost of the network can thus be estimated by multiplying the expenses of a single site and the number of the site cells according to the single site cell radius.

91 55 As an example, the initial capacity target could be 5 Mb/s, which can be divided into program guide (about 300 kb/s) and 10 channels consisting about 450 kb/s each for the combination of audio and video streams. This capacity requirement can be complied with QPSK modulation, code rate 1/2 and MPE-FEC rate of 2/3. Assuming the mode requires C/N of 11.5 db, and that the environment is urban with the coverage quality target for the location probability of 70% in the site cell edge (90% in the site cell area), Table 3-4 can be created by applying Okumura- Hata prediction model and hexagonal site cell layout. Table 3-4. The estimated site cell radius for different transmitter antenna heights h BS and transmitter power levels P TX by applying the Okumura-Hata model for the urban area. Radius of the site cell (km) P TX (W) h BS =30m h BS =60m h BS =90m h BS =120m After the capacity and coverage analysis, the estimation of the expenses can be done. The cost of the single site can be calculated with the following Formula: C site = C + C (3-17) common var iable C common represents the fixed costs and it consists of the site civil and installation work as well as other costs that are constant independent of the power class used. C variable consists of the transmitter cost (depending on the power level), feeder cost (which depends on the feeder length and on the type of the feeder which is selected based on the maximum supported power), and other directly related material that depends on the cable type/length and transmitter power level. When taking the P TX = 500 W-case as a reference as shown in Figure 3-5, the cost for the transmitters with 500, 750, 1500 and 2800 W is now 1, 0.6, 0.5 and 0.45, respectively. The unit cost per meter of the antenna feeder is estimated in this analysis comparing it to the normalized cost of 500W transmitter. In a practical case, the costs can logically be expressed in absolute commercial values of each item. It is now possible to estimate the approximate cost for each site combination in order to build a sufficient amount of sites in a certain area. The total cost is thus: C tot = C N (3-18) site A In this Formula, the N A is the total amount of sites in the area A.

92 56 As an example, the A could be selected as km 2. The radius of the site cell for, e.g. 500 W-case and antenna height of 30 m, is 2.4 km. The ideal overlapping in the hexagonal model can be taken into account as shown in the network layout calculation of Annex I. Based on this information it is possible to calculate the number of the partially overlapping sites in the investigated area as shown in Table 3-5. Table 3-5. The number of sites in the planned area. P TX (W) h BS =30m h BS =60m h BS =90m h BS =120m Investigating further the cases, the unit cost of the sites can be obtained per transmitter power level category as shown in Table 3-6. The information represents a snap-shot example of certain transmitter vendor's different models when they are compared with the 500 W-model of this specific provider. It should be noted that the relative comparison depends on each case. Table 3-6. An example of the unit cost of the sites with different power categories. P TX (W) Normalized cost In this analysis, an assumption about the impact of the cable length on the costs can be rejected. Then, the normalized cost per transmitter category can be obtained. Table 3-7 shows the cost of the solutions as function of the antenna height in the investigated area. In this analysis, the cost effect of the towers is not considered as the assumption is to use already existing ones. Table 3-7. The normalized cost for different parameter values in order to cover km 2. P TX (W) h BS =30m h BS =60m h BS =90m h BS =120m

93 57 As can be noted from Table 3-7, by using only the basic parameters, we can note the cost effect of the antenna height compared to the transmitter power category. As can be seen in Table 3-7, the 1500 W transmitter solution with the antenna installed in 120 m comes more attractive than 2800 W transmitter with the antenna installed to 60 m height. As the CAPEX analysis shows, the strategy for the antenna heights and transmitter power category should be done in sufficiently in-depth level in order to make sure the optimal combinations of the parameters vs. costs. At the initial planning phase, though, a rough estimation is sufficient in order to understand the cost-effect of different solutions as presented above. The high-level cost estimate is important to include already in the nominal planning phase in order to reject the least feasible parameter values and materials. In the detailed network dimensioning phase, the CAPEX analysis should be more detailed. In that phase, also the OPEX should be taken into account, as the importance of the operating costs of the network might rise considerably during the operational years of the network. In this phase, also the more in-depth investigation of the items that have cost effect should be carried out. The respective methodology is presented in Publication I and in the detailed radio network design chapter.

94

95 59 4 Detailed radio network design 4.1 Identifying the planning items In the detailed network dimensioning, the aim is to find the optimal balance of all the relevant items that have effect on the network performance and costs. There are various different pieces of information and investigations available identifying the possible sub-tasks of DVB-H network planning, including technical parameter effects, economical and regulation aspects as shown in [Dvb09], [Gre06], [Apa06a], [Apa06b], [Bee07], [Far06], [Had07], [Him09], [Mil06], [Ung08], [Apt06], [Ecc04], [Bro02] and [Min99]. Based on the above mentioned sources as well as the field tests, analysis and simulations carried out in Publications I X, a complete picture of the planning aspects and their relations was formed as an initial part of this thesis. Figure 4-1 shows the outcome of the investigations, i.e., the main topics to be considered as a basis for the detailed DVB-H network planning phase. The balancing of the radio network performance can be done by investigating and dimensioning in deep level the cross-relations of technical, economical and regulatory items. The ultimate goal of the detailed DVB-H radio network dimensioning is to find the cost-efficient balance between the coverage, capacity and the quality by taking into account all the relevant parameters. There might be several optimal points for the balance, i.e. several different combinations of the parameter values could result in the same optimal solution technologically. Whilst the sufficiently good balance is found, it does not matter what is the theoretical set of solutions. The practical restrictions can determine though the perfect solution. As an example, even if different number of sites could produce the same optimal solution by varying several parameters, the lower amount of sites could be more practical as the identifying of the site locations is not straightforward in practice. As an example, it might not be always possible to buy or rent a site in the technically most suitable location due to the practical reasons that are not always predictable in the network design. The ideal network dimensioning takes into account the future enhancement needs. It is thus important to make sure that the service areas are not reduced in the later phases. As an example, if the coverage area is done with QPSK providing a large operational areas but with low capacity, it should be made sure that if the modulation is switched to 16-QAM providing more capacity but lower coverage per site, there are sufficient amount of gap-fillers and additional

96 60 main sites operating in the critical areas, that the customers are not experiencing the sudden lack of the services they have got used to. The mature network operation phase requires technical optimisation, which might be a continuous process even in a stable network. The optimisation can be done by carrying out field measurements in selected, most relevant areas of the network. The correct measurement methodology as well as right interpretation of the measurement data is important. Publication I Economical limitations Site performance Publication VI Regulatory limitations CAPEX Antenna height EMC/Safety distance OPEX TX power Frequency coord. Publications III, IV, X Publications V, VII Effect of network type Link budget Standard SFN Capacity SFN gain level -overlapping Coverage Publication VIII Correction capability MPE-FEC Code rate Over-sized SFN SFN gain / SFN interferences -FFT, GI MFN MFN functionality -signalling speed, overlapping Performance Measurements Analysis Fault Management User experiences CH switching time Reception quality Bit/frame rates Publications II, IX Figure 4-1. The cross relations of the items affecting on the DVB-H radio dimensioning. In this diagram, the final aim is the balancing of the capacity and coverage by taking into account the restrictions and enhancements related to the technology, commercial and regulatory items. 4.2 Detailed network planning process Figure 4-2 presents the steps in the detailed DVB-H radio network planning, by identifying the topics from the planning process. The detailed process was developed as a part of this thesis. It is a continuum of the nominal planning phases, and it is meant to adjust the higher level

97 61 planning assumptions by identifying the rest of the relevant items that may change the coverage, capacity and quality of the radio network. The model can be considered as an enhanced version compared to the typical processes due to the presentation of the cost planning and optimization modules. As an example, [Ebu05] and [Dvb09] describe in detailed level different aspects of the network planning, but they do not show complete planning processes. By the writing of this thesis, no references were identified that would present the in-depth initial and detailed planning processes. Initial planning process Capacity planning Revise / confirm the capacity requirement per SFN area => tune parameters if needed Decision of the site locations Site candidates e.g. in rooftops, BC or telecom towers => confirm / adjust the preferable set Coverage planning I Radio planning program: Adjust antenna heights, transmitter powers per site => estimate the radius per site cell Coverage planning II SFN interference analysis to adjust coverage accordingly. Is the interference and radiation level sufficiently low? No Yes Coverage planning III MFN analysis to adjust coverage accordingly. Is there sufficient overlapping for handovers? No Yes Optimisation Measurements, user experiences Coverage planning IV Revise that coverage fulfils the requirements (with x % area/location probability) and sufficient overlapping exists No Yes Cost planning Adjust the cost prediction according to the detailed coverage plan (detailed CAPEX/OPEX optimization). Acceptable level of costs? No Yes End of planning process Figure 4-2. The detailed planning phase contains in-depth analysis of the effects of the performance parameters. The steps can be highly iterative. In the operational phase of the network, a continuous optimisation may take place via the field testing and customer feedback.

98 Capacity planning In the detailed phase of the radio network dimensioning, the capacity planning methodology does not change much compared to the initial phase. The principle remains the same, but the balancing of the capacity and coverage (i.e. number of the sites and related costs) might need several iteration rounds. The provided capacity does have a direct relation with the cash flow of the operational network. The end-user requires normally several channels to choose from, and the quality of each one should be in acceptable level, although taking into account the contents of each channel. The balancing of the number of channels and the bit rate of each one is relatively flexible in DVB-H. There can be several different bitrates defined for different channels, i.e. there could be low bit rate channels for audio news type of service, whilst the highest resolution and audio quality might be wanted, e.g. for music video channels. The upper limit for the dimensioning is determined by the bandwidth, which can be divided for several sub-channels. 4.4 Coverage planning [Publications V, VII] Effect of site locations In the initial phase of the coverage planning, a rough estimation of the number of the sites is needed. The estimation can be done by carrying out a theoretical estimation of the overall path loss to different environment, e.g. urban and dense urban, sub-urban and rural/open areas. The estimation can be done by filling the respective areas, e.g. by hexagonal type of site cells. Although this approach is theoretical, it gives a first estimation about the needed sites. If the average site antenna height and the transmitter power level are close to reality, the prediction is sufficiently good for the nominal planning purposes. In the detailed planning phase, the more accurate site locations should be known. This is the most challenging part of the coverage planning, as the practical sites can rarely be found in the ideal locations according to the uniform type of initial network layout. In addition, there might be restrictions in the antenna height and transmitter power level usage. As an example, the already existing broadcast or mobile communications towers are normally equipped with other antennas of various systems, as GSM, UMTS, radio, TV and link antennas. The variation of the practical antenna heights and transmitter power levels affects on the optimal CAPEX and OPEX calculations, so the respective analysis should be done accordingly in order to find the proper balance for the radio parameters and economical aspects.

99 63 In order to provide positive user experiences, the coverage areas should be overlapping. In the SFN network, this provides additional SFN gain, which does have effect on the performance via the link budget enhancement [Ebu05]. In case of the MFN, the sufficiently overlapping areas provide the handover functionality without problems Site cell range predictions When the final site locations are known, the detailed coverage estimation can be done. The selection of the suitable radio propagation prediction model is important, and in detailed analysis, a local adjustment of the propagation model parameters is recommendable. The latter can be done by carrying out field measurements and correlating the results with the predictions, correcting the propagation model s parameter set (e.g. clutter attenuation values). In the detailed phase of the network coverage planning, the possible SFN interference level should be investigated thoroughly. If the theoretical SFN limits are not exceeded, it is straightforward to assume that there are no interferences present. If reflections are expected from distant obstacles, they increase the effective delay of the radio components and the probability of interferences. In case of interferences, the path loss prediction method, i.e. method to estimate the carrier levels as function of the geographical location should include the interference analysis. One possibility is to integrate the method shown in Publications III, IV, VII and X into the traditional radio planning program. As for the sole coverage within non-interfered SFN area, Publication V shows examples about the commercial planning tool's (NetAct Planner) prediction model usage in urban and dense urban environment. The model of the tool is based on the Okumura-Hata [Hat80], taking into account the local clutter attenuation factors. As the used map resolution is in order of 30 meters, it does not provide an accurate coverage prediction, e.g. inside the street canyons. Nevertheless, the model is sufficiently accurate for planning purposes especially in sub-urban areas, like [Pal08, p. 64] has been concluded. NetAct Planner is typically utilized in the 2G and 3G mobile network coverage predictions, and can be adapted to the basic DVB-H coverage prediction by applying the same principles. Various different models can be utilized as a base of the predictions, including the more accurate ray-tracing based 3-dimensional models. If the Okumura-Hata model is utilized, the basic outcome of the tool's propagation model is further enhanced by taking into account additional attenuation values that depend on a separate cluster map of the investigated area. Several different cluster types can be defined, e.g., for the water areas and forests with different vegetation densities. The cluster values can be further adjusted by comparing the prediction of different area types with the field test results. If the attenuation values of the clusters are

100 64 estimated well enough, this method results more realistic predictions compared to the pure Okumura-Hata model that generalizes the prediction solely over different city and rural environments. Nevertheless, as Publication V concludes, the basic Okumura-Hata model as such indicates the needed number of the sites per area type sufficiently accurately for the nominal planning purposes. In practice, according to [Mil06, p. 17], there might be well over 20 taps in the radio channel. This should be taken into account in the deep level radio network planning and analysis. The recommendation of [Mil06, p. 17] is to use at least 12 taps for the respective simulations although the practical terminal deployment would not take the full advantage of all these components. The basic coverage area is studied in [Mil06, p.66]. The outcome correlates with the analysis done in Publication V, although the studies represent different densities of the urban area types. The balance between the economical aspects and technical solutions depends highly on the wanted quality of service level. It is thus most fluent to decide an initial target for the coverage and capacity, and to investigate with what parameter combinations the target could be achieved. The detailed network planning might require several iteration rounds depending on the outcome, i.e. if the network cost in initial and longer term turns out to be expensive even in the optimal point. The main idea of this thesis is to investigate the useful coverage area that is a result of the basic link budget and area dependent propagation models, added by the possible MPE-FEC and SFNgain and which is reduced by the possible SFN interference in the over-sized SFN. The identified sources typically do not consider this type of complete vision at the same time but concentrates on the selected details. [Bac04] [Bmc09] [Bro02] [Ecc04] [Fan06a] [Fan06b] [Far01] [Fcc07] [Goe02] [Gre06] [Hum09] [Mar05] [Pal08] [Sil06] [Tun05] [Ung06] [Wan03] [Zha06] [Zha08] [Zyr98] 4.5 Local measurements [Publications II, VIII, IX] The field measurements are needed for the revisions of the performance level of DVB-H. As the system is broadcast type without uplink channel, the terminals themselves cannot be used directly as reporting devices and investigated in the OSS of the core network. The field measurements provide data for both performance analyses as well as for the fault management. Other usage of the field measurements is related to the radio propagation prediction model adjustment of DVB-H. The received power level combined with the location information can be

101 65 utilised in planning programs in order to correlate and correct the model parameter values, e.g. the cluster attenuation values. The correction can be made either manually or automatically. When the field measurements are initiated, it is important to calibrate the equipment. Publication IX, Figure 21 shows an example of the differences in the channel display when three different DVB-H terminal units are used as measurement equipment. Figure 4-3 shows the further processed cumulative presentation of the differences of the channel displays. As can be seen, the 50%-ile point results in 57.3 dbm, 58.8 dbm and 61.2 dbm for each terminal in this specific case CDF, P(RX) of 3 terminal displays MS1 MS2 MS RSSI (dbm) Figure 4-3. Comparison of the RSSI display of 3 different DVB-H terminals used in Publication II. Cumulative distribution of the laboratory measurement with 90 samples per terminal. In the field measurements described above, the terminal s field test program displays the received power level based on the interpretation of the gain values of the automatic gain control (AGC). The minimum step size of the AGC tends to be typically in order of 1 db. The accuracy of this method is approximately db depending on the received power range [Aur09]. If a power meter is used instead, the accuracy is roughly in the same order. In addition, the quality of the calibration affects on the final estimate of the received power level. The importance of the calibration has been noted also in [Apa06a, p. 20] which describes the challenge in the interpretation of field tests when high accuracy is required. In order to minimize the inaccuracy of the channel display, the corresponding calibration can be carried out, e.g. by installing a coaxial cable directly to the input of the DVB-H receiver and by connecting a TS generator to it. If instead more advanced field or laboratory equipment like spectrum analyzer is used as in [Jos07], the accuracy of the averaged samples is logically more reliable. There are also other measurement criteria like error vector that can be utilized in the signal-to-noise ratio calculations as presented in [Fan06a, p. 33].

102 Coverage area The basic coverage area can be revised by observing the received power level P RX of the site cell around the functional coverage limit as presented in Figure 4-4. The receiving of the radio signal can be done with a scanner using respective bandwidth and averaging settings, or with a terminal capable of showing the radio parameter values as explained in Publications II, VIII and IX. In the analysis, it is important to understand the effect of the averaging of the samples during the measurement. If a real audio / video stream can be transmitted, a normal DVB-H receiver can be used as parallel equipment for the subjective quality check. This method would be sufficient in the very initial phase of the planning, in order to understand the required field strength for the most probable parameter set Received power level Sample # -70 RSSI (db) Figure 4-4. An example of the collected received power levels by moving the receiver close to the functional cell edge. The received power level samples and corresponding geographical locations can be observed by postprocessing the measurement data to a map format which indicates the expected coverage limits. One of the aims of the radio coverage measurements is to revise the accuracy of the applied prediction models in the investigated area type. Based on the received power level distribution, the respective model adjustment can be made by adjusting the outcome of the prediction with the offset the field measurements provide. Depending on the model the area cluster attenuation factors can be modified as it is one of the main items that affect on the accuracy of the predictions. It should be noted that the accuracy of the received power level measurement depends on the equipment. As shown in Publication II, the DVB-H terminals can be used well for the fast revision of the basic coverage, but the respective accuracy of the analysis suffers about 2 db additional margin due to the lack of calibration of the field strength displays.

103 67 For the more in-depth revision of the achieved quality levels as function of the received power level requires deeper analysis which is presented in the following Chapter. The method is based on the data collection from the field and on the respective post-processing Error correction Publications II, VIII and IX show the methodology for the field measurements and their respective analysis in order to find the detailed level information about the performance of the network. In order to optimize the network performance, basic field strength measurements are important to carry out in an early phase of the network deployment, observing the functional limits via measurement equipment with a GPS producing the location information for the measurement results. The Publication IX describes the more in-depth quality measurements and respective results for identifying the QEF, FER and MFER breaking points for different operational modes of DVB-H. In the method presented in Publications II, VIII and IX, the frame error rate has been identified as useful criterion when the basic performance indicators can be collected from the air interface. This is logical as the end-users experiences the quality level variations directly depending on the correctness of the received frames. The method shows that the quality can be analyzed by arranging the collected FER and MFER occurrences as function of the received power level. The MPE-FEC gain can be obtained when the results are converted into a cumulative distribution format. As the accuracy of the presented equipment is 1 db for the received power level, the corresponding proportion of error-free frames, frames that can be corrected with MPE-FEC, and the ones that remains erroneous can be normalized per each RSSI category. In this way, the breaking point of the observed criterion of MPE-FEC and FEC is possible to obtain. When this method is utilised, it is important to collect a sufficient amount of measurement data for the statistical reliability. According to [Dvb09, p. 83] the accuracy of the frame error rate would be sufficient if at least 100 samples are collected. These samples include both error-less and erroneous MPE-FEC frames. The FER information, i.e. frame errors before MPE-FEC specific analysis, is obtained after the Time Slicing process, and the MFER is obtained after the MPE-FEC module. If the data after MPE-FEC is free of errors, the respective data frame is de-encapsulated correctly and the IP output stream can be observed without disturbances. As the erroneous frame indicates directly the user perception of the quality, the frame error rate before and after MPE-FEC are useful indicators for the performance studies. As stated in [Dvb09, p. 83], an erroneous frame destroys the service reception for the whole interval between the time-sliced bursts. According to [Dvb09, p. 83] the 5 % FER / MFER can be considered as a breaking point for the acceptable quality level. The respective point can be studied in various ways. One method is to collect in certain spots sufficient amount of samples, e.g. in area of 5 5 meters by moving

104 68 slowly the terminal within the area and thus rasterizing the area. By interpreting in more ample way the statement of the [Dvb09], the respective FER and MFER error rate can now be obtained by applying the following formulas: Ferr _ bm FER(%) = 100 (4-1) F r Ferr _ am MFER(%) = 100 (4-2) F r F err_bm represents the erroneous frames before MPE-FEC and F err_am is the number of residual erroneous frames after MPE-FEC. F r indicates the total number of received frames. The presented methodology in Publications II, VIII and IX proposes the arranging of the occurred samples as function of received power levels as shown in Figure Count of samples per Prx level FER0, MFER0 # FER1, MFER0 # FER1, MFER1 # 120 Counts dbm Figure 4-5. The first step of the proposed methodology arranges the occurred samples as function of the received power level, showing the amount of error-free instances, occurred instances with frame errors that could be corrected with MPE-FEC, and remaining erroneous frames after MPE-FEC. The method is based on the normalizing of the occurred events per each RSSI category individually as shown in Figure 4-6. This presentation is valid when other non-rssi related effecting components are not present. These components might include impulse-noise type of interference which is spread over a wide RSSI scale or the effects that occur after exceeding the Doppler shift limits determined by the used parameter value combination. The format shows now directly the proportion of the frame errors that could be corrected. It is thus possible to have a closer look on the breaking point of interest, e.g. in 5 % FER/MFER. Figure 4-6 shows an example of the breaking point. The difference between the FER and MFER can be interpreted directly as the MPE-FEC gain in db value. This can be interpreted in such a

105 69 way that the coverage area of the site cell is limited to the received power level corresponding to FER 5 %, whilst the RSSI level is enhanced by P TX when MPE-FEC is applied. 100% 80% Count of samples per Prx level, normalized FER0, MFER0 # FER1, MFER0 # FER1, MFER1 # 60% Counts 40% 20% 0% dbm Figure 4-6. The method shows the normalized number of occurred events per each RSSI value. It is straightforward to observe the respective performance in terms of received power level from Figure 4-7 which shows FER of 74.9 dbm and MFER of 78.0 dbm resulting P TX, i.e. MPE-FEC gain of 3.1 db in this case FER and MFER (results in scale 0-20 % per Prx value) FER% MFER % 10% criteria 5% criteria % 10 5 MPE-FEC gain dbm Figure 4-7. An amplified view to the 5% FER / MFER level Accuracy of error correction analysis The applied FER and MFER comparison methodology requires sufficient amount of collected data per RSSI level especially around the observed breaking point of FER and MFER. The

106 70 sufficient count number per RSSI level can be investigated in practice by collecting different amount of samples and by comparing the variations of the results especially in the FER / MFER 5 % breaking point. Reference [Dvb09, p. 83] states that in order to provide sufficient accuracy, it is necessary to analyze at least 100 frames. This statement is relatively loose though, although in practice the MFER has a clear breaking point with a small increment of C/N. The statement does not explain more thoroughly about the requirement for the number of related RSSI values. In the analysis presented in this thesis, the accuracy of the samples is investigated individually for each RSSI category in order to achieve an accurate error margin estimate. As a starting point, for the comparison of the FER and MFER graphs, the resolution of the outage-% scale (y) is important. When the number of samples n is known, i.e. the number of occurred instances ( no errors, errors that could be corrected with MPE-FEC, and remaining errors ), the sampling resolution R in normalized outage % scale can be obtained via the following Formula: 100 R = % (4-3) n If the amount of the samples n is 200 per certain RSSI value, the resolution would be 0.5 %- units indicating the maximum accuracy that can be obtained with that number of samples. In order to make a more thorough error estimate, according to the simple random sample from a large set of values comparable to the polling error estimates as the randomness of the measurement is fulfilled the maximum error is a single re-expression of the sample size n as stated in common sources like [web03]. According to the rule of thumb, whilst the sampling fraction is less than 5 %, the margin of error can be estimated via the simple random sample principle. It assumes that the "population", which refers in this case the entities that can be either erroneous or free-of-errors, is infinite. This assumption can be taken as a basis for the previously described method. As an example, the margin of error at 95 % confidence for the simple random sample principle with n samples is: 0.98 Err[ 95 %] = (4-4) n By applying this principle, an example of 100 samples per RSSI category would produce a margin of error of 9.8 % whilst a case with 50 samples produces a value of 13.8 %. This estimate can be applied to the values of non-erroneous events, erroneous events before MPE- FEC and erroneous events after MPE-FEC when the total amount of samples is known. The accuracy of the individual y-axis values of a single curve (FER or MFER) can be estimated via the above mentioned principle. Then, the MPE-FEC gain, which is the difference between the indicated points of FER and MFER curves in 5% line, can be further estimated by observing the extreme values of the accuracy of the single curve.

107 71 As an example, a measurement result of the Publication IX can be analysed. Let s select a case of 16-QAM, CR 2/3, MPE-FEC 2/3, GI 1/4, FFT 4K, which produced the sample distribution shown in Figure Samples per Prx level # of samples FER0, MFER0 # FER1, MFER0 # FER1, MFER1 # RSSI (dbm) Figure 4-8. An example of the collected data distribution over the RSSI scale. Table 4-1 summarises the occurred total amount of samples per RSSI category and shows the respective margin of error for 95 % confidence by applying the above mentioned formulas. As can be seen, the accuracy of the RSSI levels of 70 dbm and 71 dbm is not sufficient, but Figure 4-8 indicates that even if the respective number of total samples is low for these RSSI categories, the respective received power level has been clearly sufficient for error-free functioning. It can be seen from the margin error analysis that the MFER5 can be found between RSSI levels of 76 dbm and 77 dbm when the confidence level of 95 % is applied. Let s observe more thoroughly the RSSI scale of dbm, analyzing individually the error margin for each RSSI category in such a way that the total number of occurred events, i.e. error-free (FER0, MFER0), FEC resulting error but MPE-FEC has corrected (FER1, MFER0) and error after MPE-FEC (FER1, MFER1) is taken into account for the error margin calculation. Figure 4-9 shows the result of the analysis with respective marginal for each respective FER and MFER value. When the FER5 and MFER5 is observed, FER5 results values of P RXmin = 71.9 dbm, P RX = 72.1 dbm and P RXmax = 72.3 dbm. MFER results P RXmin = 76.3 dbm, P RX = 76.4 dbm and P RXmax = 76.5 dbm dbm. The respective MPE-FEC gain can thus be informed as 76.4 ( 72.1) dbm = 4.3 db with value range of [4.0, 4.6] taking into account the error margin. This example represents a typical situation for the measurements presented in this thesis

108 72 and the error margin can thus be assumed to reside within % for individual FER5 and MFER5 curves, and the respective MPE-FEC gain would have typically a maximum of 10% error margin. Table 4-1. An MPE-FEC error analysis for the example shown above. (* indicates too low sampling rate). RSSI Sampling (db) fraction Number of samples Margin of error (95 %) Amount of FER1, MFER 0 Amount of MFER 1 (erroneous samples) FER and margin of FER (%- units) MFER and margin of MFER (%- units) * 0.0 ± ± * 0.0 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 1.9 It should be noted that also the RSSI scale has error margin that depends on the calibration of terminal the (±2 db) and on the mapping of the AGC process to the RSSI value (±1 db). If the accuracy of the above mentioned measurement methodology should be enhanced, more samples would need to be collected per RSSI category. It is obvious that sufficiently good field strength provides with reliable results (non-erroneous samples) even with low amount of samples, whereas erroneous frames are produced relatively often in the lowest field. It is thus sufficient to collect the samples around the critical RSSI scale around the 5% breaking point of FER and MFER. On the other hand, the idea of the above mentioned method is to collect fast snap-shot data which means that the task could be carried out in relatively short time period per investigated area. Typically, the storing of each measurement sample (frame) could take about 1 second, and assuming that in sufficiently accurate analysis ±5 RSSI values, i.e. about 11 RSSI categories need to be measured, e.g. with 300 samples per RSSI category producing margin of error of 5.7 %, this would require s = 3300 s, i.e. 55 minutes would be needed. In typical measurement routines of an operator, this might be in the practical limit of the used work time for a single measurement, although the error margin would provide with better accuracy.

109 73 Outage-% CDF, FER and MFER with error margins FER MFER FER min FER max MFER min MFER max RSSI (dbm) Figure 4-9. Amplified view to the case result graph showing the FER and MFER curves with respective error margin calculated for the absolute values of the samples for each RSSI category individually. According to the results, the analysis described in Publication II might produce relatively smooth FER and MFER curves, which results a straightforward way of comparison of FER5 and MFER5 points and indicates thus the respective MPE-FEC gain with relatively high confidence level. In some cases, though, the curve behaviour is more "aggressive", producing high peaks along the RSSI scale an making the FER5 and MFER5 interpretation challenging. A part of these peaks can be explained by impulse noise, and some can be due to the lack of samples. Nevertheless, the most important RSSI scale is in the range corresponding the FER5 and MFER5, and normally the phenomena of the correction capability can be seen graphically even if the FER5 and MFER5 breaking points are not implicitly possible to interpret. One way of trying to get the approximate value for the breaking point in this type of cases would be to interpolate regression curves for the FER and MFER graphs. The error margin could be estimated by observing the difference between the regression curve and the original deviating points. Nevertheless, the method would not reflect totally the reality as the errors do occur in non-predictable way along the RSSI scale in these cases making the reception unstable. The deeper analysis has thus not been made for this type of results in this thesis. 4.6 Effect of SFN [Publications III, IV, X] The SFN has advantages in normal operation as it produces gain due to the multipath propagated separate radio signals. [Bmc09, p. 15], though, has concluded that as SFN gain is not persistent across all locations of the DVB-H network, it is not considered in the radio link budget. In the theoretical case, if the carrier level is considered useful only if the minimum

110 74 functional level of the investigated parameter settings is achieved, the SFN gain can be obtained only in the overlapping area with C/N complying with the minimum functional value. Nevertheless, the C/N values just below the functional limit combined with the C/N value of other signals below the functional limit might set the sum of these signals above the functional level which enhances the coverage area between the sites. The more carrier components (above the noise limit) there is present in a certain spot, the more probably the sum of the signals exceed the minimum C/N limit Non-interfered network Publications IV and X show the respective principle and simulation method in order to estimate the SFN gain by varying the relevant radio parameters. The method is based on the gain that is achieved in the physical level, i.e. in the receiving of the total signal with stronger level than an individual stream provides. Although the method is based on the pure received power levels in a single cell case compared to the n number of combined signals, it gives an enhanced estimate of the increased probability of receiving the signal correctly compared to the other theoretical studies found about the topic [Sil06] [Bee98] [Cha06] [Dvb09] [Law01] [Ung08]. The practical field test based references are, on the other hand, limited as for the complete effect of the SFN gain as they are typically based on only few sites [Ple08] [Ple09] [Ple09b]. The presented simulation method can take into account basically all the sites that possibly contributes either to the useful received carrier power or to the received interfering power. The basic principle is based on the summing of the received power levels [ P RX ( C) ] k in absolute values (W) as shown in Formula: n ( RX k )( W ) [ P ( C) ] ( W ) [ P ( C) ] RX tot = (4-5) k = 1 The principle has been presented in [Ebu05] and it has been used in the received power level estimations, e.g. in [Sil06]. In the annexed Publications, the received power level is observed originally in dbm. For the power summing, the absolute value is thus first obtained by: 10 P RX ( W ) = (4-6) 1000 [ ( C) ] P RX ( dbm) 10 In order to compare the SFN gain value, the reference level of received power level, at given time and location, is the strongest possible from the sites that can be received above the noise floor: RXref {[ P ( C) ],[ P ( C) ],...,[ P ( C ] } P = ) (4-7) max RX 1 RX 2 RX n

111 75 Now, the SFN gain G SFN is the difference between the summed and reference levels in dbm values: [ P ( C) ] G [ P ( C) ] RX ( W ) RX ( dbm) = 10log 3 (4-8) 1 10 SFN [ P ( C) ] ( dbm) [ P ( C) ] ( ) ( db) = dbm (4-9) RX tot RX ref Publications III, IV, VII and X assumes that the whole OFDM channels, i.e. multi-propagated signals, can be summed without taking into account the frequency selectivity within each bandwidth. It is obvious that especially in case of Rayleigh fading, some of the sub-carriers of a single OFDM bandwidth experiences higher attenuation than the others. As the overall effect is a combination of some destroyed sub-channels via the Rayleigh fading, and corrected subchannels via the error recovery properties of MPE-FEC, it can be estimated that the significance of the frequency selectivity is low for the calculation of the SFN gain as presented in this thesis. The practical observations during DVB-H field tests supports this reasoning [Tal10]. The principle of the general power level summing for the high-level simulation purposes is shown in Figure 4-10 which shows four non-correlating individual signals with approximately same average level as an example. These data is derived from the individual measurements of a single site carried out in Publication III, and represents a mix of channel types including longterm and fast fading. These data are averaged by the terminal to 1 second resolution, so the highest Rayleigh attenuation peaks can not be shown. -65 Individual signals Summed signals Signal 1 Signal 2 Signal Sum (1-2) Sum (1-3) Figure An example of individually received signal levels in the same area, and their total power based on the direct power summing. Same terminal was utilized in these measurements. Figure 4-10 presents also the sum of these signals by applying SFN expressions. As can be noted, the resulting signals, in addition to the increased total received power level,

112 76 have smaller variations. Figure 4-11 shows the CDF of the individual and combined signal levels. Table 4-2 and Table 4-3 summarise the behaviour in numerical values Signal 1, original Signal 2, original Signal 3, original Signals 1-2, summed Signals 1-3, summed CDF Figure CDF of the 3 individual and summed signals presented in Figure Table 4-2. Numerical values of individual signals of the example. Average RSSI Signal # Stdev (db) (dbm) 50%-ile of CDF (dbm) Table 4-3. Values of the summed signals of the example. G SFN is calculated by comparing the average and 50%-ile level with the strongest individual signal of each case. Summed signals Average RSSI (dbm) Stdev (db) 50%-ile of CDF (dbm) 1, , 2, G SFN / reference signal # +1.8 db (averages) +1.3 db (50%-iles) Ref: Signal # db (averages) +3.3 db (50%-iles) Ref: Signal #1

113 77 The above presented example shows the principle of the summing of power levels. The same principle is applied in the simulator presented in Annex 1. There are also analytical methods developed for the fast fading signal combining as presented in [Fus08], and SFN coverage probability estimation as presented in [Lig99b] and [Lig99c]. Nevertheless, this thesis uses the power summing of the carriers and interfering components only via Monte-Carlo simulations. The non-interfered case of the simulation results presented in IV can be used for the SFN gain estimations. The suitable principle is based on the filtered coverage area of a single cell, which gives the reference level for the cumulative C/N. Then, more cells are added according to the SFN reuse pattern method presented in Figure 3 of Publication IV, and the C/N distribution is simulated for each case. The G SFN can be obtained by comparing the 10 % outage probability level (which equals to 90 % location probability) of the CDF of each case. Figure An example of simulated area with SFN reuse pattern size K=7. Table 4-4 presents comparisons of the SFN gain studies presented in the related references. According to Table 4-4, the SFN gain value varies largely even with low number of transmitters (2-3), the minimum being db and maximum over 6 db, depending on the source. The gain obtained from Publications IV and X represents a typical value range thus correlating with major part of the presented references when the number of sites is up to 3. Unlike the other presented results, [Bov09] has concluded that the SFN gain would actually be negative in two-transmitter case due to the significant degradation of the signal quality indicated by modulation error rate (MER). This outcome clearly differs from the other references. The source does not present cause for the effect, but the explanation might be that the presented test setup includes feedback loop interferences between the input and output of the gap filler.

114 78 Table 4-4. Comparison of the SFN gain values obtained from different references. Reference Method Sites SFN gain Publications IV, X Ref 1/1: [Ple09] Ref 1/2: [Ple09b] Ref 2: [Bov09] Ref 3: [Bmc09, p. 15] Ref 4: [Apa06a, p. 43] Ref 5: [Apa06b, p 133] Ref 6: [Apa06b, p 30, 32, 35] Ref 7: [Apa06b, p. 116] Ref 8: [Apa06b, p. 163] Path loss based simulations, hexagonal model, non-interfering network. Results obtained over the whole area. Field tests, Electrical field measurements and mathematical formula for the gain calculation. Field tests, electrical field measurements with mathematical modelling and CINR estimate. Field tests, main transmitter (TX) and gap filler (GF), Modulation error rate observations. Link budget investigation sites: 2.8 db 4 sites: 3.4 db 7 sites: 4.6 db sites: db, see Figure 4-13 Note: QPSK, CR=1/2, MPE-FEC=1/2, GI=1/4, FFT=8K. For 16-QAM, the gain was up to 6.4 db (19 sites) (stdev ) db (electric field method) 0.3 db (CINR method) 1 TX + 1 GF N/A db ( 52.3 db 54.5), but MER decreased about db (20 28), resulting negative SFN gain. SFN gain not recommended for radio link budget. Field tests. 2 SFN gain was evident, but no exact value presented. Indoor field tests, Barcelona, Spain. Finnish field tests, outdoor. Finnish field tests, outdoor and indoor. Common outdoor field tests, Barcelona, Spain. 2 SFN gain was calculated and treated as a statistical variable. The result was obtained by measuring the average gain in terms of field strength due to the contribution of the gap filler. The variable showed a log-normal behaviour with a mean value of 6.1 db ( db) and 6.4 db of stdev. The range of the values was very wide; study did not thus recommend using SFN gain values in link budget. 2 Case 1: db, closer to 1.5 db Case 2: 3 db Case 3: 5.4 db All results are related to the higher field strength. 2 One transmitter was used in outdoor pedestrian and indoor corridor measurements. The results show better MFER at a certain signal level when using two transmitters. The SFN gain was db (in average 1.5 db) 2 The histogram of the accumulated SFN gain values resulted a mean of 6.08 db and a stdev of 6.36 db.

115 79 The reference [Mil06, p. 71] shows that as the gain margin of the gap filler gets close to zero the ripple increases considerably. At 0 db gain margin point the system is already unstable. As the gap filler setup is critical in the SFN analysis, the most reliable results might be derived from the test cases where only main transmitters are switched on and off as explained in [Ble08] and [Ble09]. As an overall observation, the field tests that are related to SFN gain have normally been carried out with only two transmitters. In [Ble08], a method of adding the electrical field strengths from a total of three main transmitters resulted 2.3 db. Nevertheless, in a further analysis related to the same network, [Ble09] has found that the CINR results in only 0.3 db gain. This indicates that SFN gain should be estimated case-by-case by investigating what is the effect of the interference level, e.g. for the MER degradation, not only the sum of electrical field strength. Figure 4-13 shows a summary of the SFN gain results obtained via the simulations of Publication IV, Figure 12. The originally presented non-uniform SFN reuse pattern size scale has been converted linear and a logarithmic regression curve has been added to the resulting plots. The reference results from Table 4-4 have also been marked in the graph for comparison purposes. As can be observed, the SFN gain values of the identified references vary considerably. The main reason is probably caused by the differences in the practical setup, which might not be explained detailed enough in the references in order to indicate the size of the coverage area of a single site compared to the coverage produced by the gap-filler or secondary transmitter. 7.0 SFN gain, no interference, QPSK 1/4 8k SFN gain (db) ref8 ref2 ref7 ref5 ref6/3 ref6/2 ref1/1 ref6/1 ref1/2 y = Ln(x) Reuse pattern size Figure The SFN gain obtained via the simulations of Publication IV for the non-interfered network (QPSK, FFT 8K, GI 1/4). The comparable results in studied references are also presented.

116 80 Although the results of Publications IV and X are based on simulations, the setup was made in controlled way by selecting the overlapping proportions of the site cells according to the hexagonal layout model, i.e. there were sufficiently overlapping parts present but without excess. The results are derived always from filtered cell areas, i.e. within the dimensioned cell based on the given area location probability. Furthermore, the simulations are performed for varying amount of transmitters up to 21. The presented methodology for the comparison of the C/N distributions of a single cell and multiple cell cases creates a logical and controlled environment for the SFN gain estimation. Furthermore, the presented simulation method brings novel aspect for understanding the complete behaviour of the SFN gain even in over-sized SFN area where partial interference is present. The model does not explain the practical phenomena of lower MER levels that were identified in [Bov09]. The degradation can be assumed to arise from the non-optimal performance of the terminal or test setup Interfered network SFN simulations In practice, there occur outages for single streams within the planned cell area which can possibly be recovered by the reception of other streams. In this sense, the SFN gain is not limited to the outer border edge areas of the cell. Furthermore, when the theoretical maximum limit of SFN is exceeded, there might appear interferences as the sites outside the area determined by the maximum allowed guard interval distance converts to interferers instead of sources of useful signals. Publications III, IV and X shows the principle of the mechanism via respective simulations. According to the results of Publications IV and X, it is possible to find a functional balance between the raised interference level and the compensating SFN gain even in over-sized SFN network. The simulation method presented in Publication IV takes into account the SFN gain in interfered DVB-H networks, i.e. when the parameters affecting on the SFN size are selected in such a way that there are sites found outside the theoretical SFN area (as shown in Table 2-1) causing interferences. The level of the interference can first be investigated in high level by observing the interfered links between the sites that are exceeding the SFN limits. When the number of these potential interference sources is compared with the total number of links between all the sites, the severity of the interference in the investigated area can be identified. As can be observed from Tables 3 and 4 of Publication IV, there are a relatively clearly identified parameter sets which results in interference level zero or 100 %, There is also a set of parameters that results only partial interference levels. As shown in Figure 6 of Publication X, the interference level is not constant, though, but tends to be switched on and off depending on the location of the terminal

117 81 in the field. This is due to the fact that the difference between the arriving signals is the one that determinates weather the signal is interfered or not. When all the sites are within the SFN area, i.e. when the distance between of any of the sites is equal or less than the theoretical maximum SFN diameter distance D SFN, there will be no interferences. This is the case also outside the circle defined by D SFN. This can be shown by investigating the distance between the receiver and the dominating site compared to the distance of the receiver and any of the other sites. The difference between these distances defines the effective distance D eff of the signal. This distance can be investigated by making the receiver drift away of the SFN circle towards any direction as presented in Figure The absence of the interferences can be investigated by observing the angle α of Figure It can be seen that D eff is maximum when cos α = 1. In that case, D eff = D SFN, i.e. D eff never exceeds D SFN when no sites are located outside of the SFN area. SFN area Delay 1 α Delay 2 TX1 Max SFN diameter = D SFN TX2 Figure There are no interferences when the sites are within the SFN area even if the receiver drifts outside of the SFN area. The reception within this outer zone is thus possible when the minimum required C/N can still be reached. When the distance between any of the two sites using the same frequency is longer than the D SFN, there will be interferences in the locations where the D eff > D SFN. The interference distribution can be observed by defining two sites in the simulator that was used in Publications III, IV and VII. By selecting a parameter set of GI=1/4 and FFT = 2K, the D SFN would be 16.8 km. The simulation of two sites in a km 2 area, with x/y-coordinates of (14.9, 22.5) and (30.1, 22.5) shows that there are no interferences present at any point as the inter-site distance is 16.8 km. The site antenna height was then set to 80 m, EIRP to +70 dbm, and the first site was moved to the coordinate (10.0, 22.5), i.e. the inter-site distance was extended to 20.1 km. The standard deviation of the long-term fading was maintained in 5.5 db and the area location probability as 70 %. When the maximum D SFN is exceeded by 4.9 km, the interference is present in areas where D eff > D SFN as can be seen via the simulation shown in Figure 4-15.

118 82 Figure When the sites exceed the SFN area, there will be interferences in those areas where D eff > D SFN. Figure shows the interference zone that applies for D eff > D SFN everywhere with the I-component greater than the noise floor. In this case, P tx was +70 dbm and site antenna height 80 m. The area is 45 km 45 km. Even if the interference starts being present, the reception is destructed only when the received C/I < minimum defined C/N for the used parameter set. For this reason, we can introduce a term "destructive interference", which can be defined as an interference level in given time and location that lowers the otherwise useful C/N ratio in such a way that the reception gets below the minimum C/N requirement of the investigated parameter set. By continuing the analysis of the previously presented simulation, Figure 4-16 shows the distribution of the destructive interferences within the area where D eff > D SFN. Nevertheless, when observing the minimum C/N ratio of 8.5 db in this specific case, the effect of these destructive interference instances is not significant because their percentage of the total events is relatively low, so the remaining C/(I+N) ratio provide still almost as large geographical cell coverage area as the situation would be without interferences. This effect can be observed in Figure When the power level rises to +80 dbm, the interference zone starts to be affected as can be seen in Figure 4-18.

119 83 Figure The distribution of the destructive interference. It can be seen that the useful field is sufficiently high close to the nearest site cell to avoid the destructive interferences, but otherwise this type of interference may occur anywhere within the interference zone. Figure The geographical distribution of C/I when P tx = +70 dbm.

120 84 Figure The effect of the interference can be seen when the power level is set to +80 dbm in this specific case. The reduced C/I-level can be seen clearly behind the sites within the interference zone on left and right hand sides C/I, Ptx=70 dbm C distribution I distribution # of samples db Figure PDF of the 2-site simulations (P tx = +70 dbm).

121 85 1 CDF, C/I C/I, Ptx=60 dbm C/I, Ptx=70 dbm C/I, Ptx=80 dbm db Figure CDF of the 2-site simulations for P tx of +70 dbm. Also a comparison results of +60 dbm and +80 dbm sites are included. The geographical analysis presented above can be analysed in PDF and CDF of C/N, I/N and C/(N+I) over the whole geographical area of km 2 for the comparison purposes. Figure 4-19 and Figure 4-20 shows the PDF and CDF, respectively. Figure 4-20 visualises the previous geographical observation that shows somewhat decent interferences in case of +70 dbm. This indicates that it is possible to construct a larger SFN cell than the theoretical D SFN dictates, depending on the radio propagation conditions, site antenna height and transmitter power level. In order to understand more detailed the impact of the parameter values on the level of interferences, a set of simulations was carried out in Publication III. A fixed area of km 2 was used as a basis for the investigations. The area was filled with partially overlapping cells according to the hexagonal model. As the number of non-correlating interfering components increases from the above presented 2-site case, the power summing of the received interference components I k can be applied: n ( RX k ) [ P I) ] = [ P ( I) ] RX ( (4-10) tot k = 1 The C/I ratio of each simulation round in investigated location is thus:

122 86 C I [ P ( C) ] ( dbm) [ P ( I) ] ( ) ( db) = dbm (4-11) RX tot RX tot The results show that by varying the antenna height and relevant parameters affecting on the SFN size, the C/I distribution over the whole area clearly varies. The results also indicate that when the parameter values that cause minimal interference levels in the over-sized SFN network are utilized together with maximum site antenna height and power level, it is possible to construct a large DVB-H network with only single frequency by accepting a certain outage. The extension of the SFN size is possible especially with GI 1/4 and FFT mode 8K. On the other hand, when applying too demanding parameter values, the network quality might collapse. The optimization chapter describes in detail the respective simulation principle Conclusion As Publication IV, Figures 12 and 13 shows, the SFN gain over the whole investigated network area depends on the GI and FFT modes and the interference level within the SFN area. In the non-interfered SFN network, using GI 1/4 and FFT 8K in a uniform site cell layout with ideally overlapping areas according to the hexagonal model, a maximum of 6 db of SFN gain can be expected in the best case in average within the whole investigated area. This result clarifies the SFN gain behaviour. It can thus be assumed that this result is more realistic than the information of [Bmc09, p. 15] that recommends the use of 0 db as SFN gain value in DVB-H radio link budget. On the other hand, the definition of [Bmc09] is not very accurate. Instead, it informs that SFN gain is the reduction of number of transmitters to cover a given area using synchronised transmitters compared to the number of independent MFN transmitters. [Bmc09] concludes that depending on the network topology, this transmitter savings may be carefully translated into an average increase of coverage. However, the SFN gain is not persistent across all points of the network, so it will not be considered in the link budget. The definition assumes that the only benefit of SFN could be obtained in the site cell edges, unlike what Publication IV shows. Based on the simulations of Publication IV, SFN gain of 0 6 db could be applied to the DVB-H radio link budget depending on the parameter settings and the total number of the transmitter sites per SFN area. For the local adjustments of the SFN value of the link budget, a respective simulation method can be applied by varying the GI, FFT, code rate, power level and antenna height settings. In a realistic network, the SFN gain could be taken into account when the site locations and other parameters are selected. The SFN gain can be simulated over the planned area, and its effect can be taken into account both in non-interfered and over-sized SFN cell by reducing the transmitter power levels and/or lowering the assumption of the antenna height, which gives more freedom in the optimal deployment if there are problems in using the values for the power or antenna height that the original plan indicates. If the SFN gain can be simulated in early phase of the first deployment plan, it can influence on the optimal site selections.

123 87 As an outcome of the SFN gain related simulations presented in this thesis, the definition of the SFN gain has been clarified. The method takes into account the combination of the gain and interference within the whole investigated network, not only non-interfered environment and in limited locations as the identified other publications typically show. 4.7 Radiation limitations [Publication VI] The installation of the DVB-H sites must comply with the local and international regulations of the EMC limits and the safety zones related to the limitations of exposure of the public to nonionizing radiation of the signals. On the other hand, the installation of DVB-H antennas should be designed in such a way that the other systems located nearby will not cause interferences to the DVB-H transmission, and vice versa. The general limitations of exposure should be investigated via the national and international laws. European Union dictates the limits for the member countries. As an example, the regulation in Finland covers the frequency band GHz [Reg02]. The format and values of the limits varies though depending on the regulator. As another example, FCC informs the limitations for cellular and broadcast environment in USA as function of the antenna height and installation type as indicated in [Fcc98] and [Fcc07]. The allowed limit in 700 MHz band is maximally 50 kw ERP, but decreases as the antenna height gets lower. When initiating the planning process, the general maximum power limit should be taken into account. When the plan advances, also the EMC and safety zones should be estimated as accurately as the planning phase allows. When the final plan is under work, the site specific safety zones should be estimated. Publication VI describes the principles for the dimensioning of the radiation levels and how the safety zone can be taken into account in rooftop and tower installations of DVB-H. The calculations are based on the formulas shown in [Min99]. The respective level should also be taken into account in the CAPEX/OPEX optimisation as the maximum allowed power level may change the optimal transmitter power depending on the case. Publication VI shows case examples about the estimation of the levels in a typical DVB-H antenna installation setup. In a typical scenario, the DVB-H site consists of directional antennas either in towers or rooftops. The radiation pattern should be taken into account accordingly in vertical and horizontal layers. The radiation safety distance calculations should be carried out for different sides of the antenna. Publication VI proposes a method that takes into account the radiation in back lobe by estimating a maximum value over the complete half of the hemisphere. For the field below the antenna, the estimation can be made based on the vector's angle in vertical plane and respective amplitude that depends on the vertical radiation pattern of the

124 88 antenna. This method is relatively simple but functional in the typical DVB-H deployment. The method was not found in this specific format in the related references [Chu00] [Fcc98] [Min99] [Reg02] [Sci07]. As a summary, the regulatory limits dictates the maximum radiating power P TX,reg and EMC / radiation safety analysis results in the maximum radiating power in site-basis P TX,safety, the optimal radiating power P TX,opt that is obtained via the CAPEX / OPEX optimisation can thus move based on the formula: { P, P P } P TX min TX, opt TX, reg, TX, safety = (4-12) The selection of the final radiating power level P TX, might thus require changes of either the antenna heights, transmitter power levels or both, reducing the site cell coverage. The respective analysis should thus be taken into account in general level in the nominal radio network planning (the correct selection of the uniform power levels) and in the detailed radio network planning (by applying the analysis in individual site basis). The final selection of PTX might also require various iteration rounds in order to identify the optimal point of technical and economical parameters. It should be noted that the radiation analysis method presented in Publication VI utilizes directional antennas as a basis for the examples. Nevertheless, only omni-radiating antennas were utilized in the SFN simulations described in Publications III and IV. The selection of the omni-directional antennas in the simulations is due to the practical point of view that the DVB- H site cell is typically done as omni-directional, either by utilizing the omni-radiating antenna (pole), or a set of directional antennas in order to create the omni-radiating coverage. The utilization of the directional antennas in the simulations has been identified as a potential future study, with the analysis of the effect of the antenna down-tilt on the radio performance (for maximizing the SFN gain and minimizing the SFN interference level). The directional antenna has been studied in the radiation related publication because this provides a practical methodology and useful examples of the expected outcome for the installation of the antenna elements, with safety distance estimations below and on the sides of the antennas. The outcome gives a realistic estimation of the radiation for the antenna installations in roof-tops, in roof edge parts, where the omni-directional element would not be logical solution. Nevertheless, the method is valid also for the omni-radiating elements. 4.8 Cost prediction [Publication I] In the complete network planning, the cost effect needs to be taken into account in every phase. In the initial phase of the network planning, a rough estimation of the average site cost as well

125 89 as the total cost of the area of interest is sufficient. The total channel capacity requirement depends on the number of the channels and their respective quality level. The demand for the services and thus additional capacity can be expected to grow as function of time as stated in [Ski06, p.2]. The detailed plan should already contain more in-depth analysis of the network costs as for the CAPEX and OPEX. In this phase, it is possible to take into account different solutions, including the transmitter type, transmission, preferable and available antenna heights, antenna types, site solutions in general etc. It is important to balance the costs with the technical solutions, i.e. the hardware, software and related technical quality, coverage and capacity of the network. The cost effect is such an important item that it should be taken into account preferable in an iterative way, i.e. in order to seek the optimal point of technical and commercial balance, possibly several planning scenarios should be carried out. The more detailed cost optimisation is explained in Chapter 5.3. It can be assumed that in the initial planning, uniform assumptions can be applied for the site costs whilst in the detailed planning, site-specific assumptions should be utilized for the more in-depth cost optimisation.

126

127 91 5 Optimisation 5.1 Site parameters Controlled SFN interference [Publications III, X] In some cases, there might be a need to extend the maximum theoretical SFN area by building more sites in wider areas that use the same frequency. This might be relevant, e.g. in very large cities with a single network topology. The division of the area can be made by selecting two or more frequencies for MFN and creating separate SFN isles. The handover between the isles provide a fluent continuum of the service. Nevertheless, there might be lack of frequencies, which possibly requires the use of large SFN areas. Publication III presents a method for the simulations of the over-sized SFN area, including case results that indicate how the C/(N+I) ratio behaves over the whole investigated area. The simulations are based on the hexagonal model, and the idea is to select a rectangular area with x- and y-coordinates which is filled in with partially overlapping cells. A set of parameters was studied by varying the modulation scheme, FFT and antenna height of all the sites, i.e. a uniform network was considered in each simulation case. The radius of the cells was calculated in the initial phase of each simulation in order to create a perfectly dimensioned radio network. The C/(N+I) ratio of 8.5 db was utilised as a limit for QPSK, and 14.5 db for 16-QAM. The radiating power level of the sites was maintained constantly in +60 dbm. The GI was fixed to 1/4, CR to 1/2 and MPE-FEC rate to 1/2. The studied area was km 2 for each case, i.e. depending of the antenna height and modulation, the total amount of the site cells varied according to the cell radius. The power summing for the useful carriers as well as for the interfering signals was applied in the simulations as indicated in [Ebu05, p. 55]. The developed simulator is based in the Monte- Carlo method. As a difference for the typical solution mentioned in [Ebu05, p. 55], the percentage of the useful signal levels was simulated in such a way, that the complete area was not divided into smaller coverage sub-areas. This is because the coverage area as such was not investigated, but the difference of the CDF of C/(N+I) for different parameter values. It can be estimated that this method gives an accurate result for this purpose in much faster time frame compared to the separate simulations of each sub-area. Figure 5-1 shows the outcome of the simulations for the QPSK and 16-QAM modulation schemes. Each plot represents an individual simulation of 60,001 snap-shot rounds. Figure 5-1

128 92 indicates the maximal useful antenna height with certain outage percentage over the whole area. The outage refers to the percentage of the locations in the whole planned area that do not provide the minimal C/(N+I) ratio for the successful reception. In this case, the outage value of 10 % indicates the area location probability of 90 % over the whole investigated coverage area. It should be noted that this graph is a corrected version of the one showed in Publication III, which utilized a squared power sum method. Figure 5-1 is based on the direct power sum for both useful carriers and interfering signals. It can be seen that in this specific case, the outage probability over the whole investigated area of 10 % or less can be obtained with all the antenna heights of m when FFT 8K is applied. According to this figure, the FFT 4K provides an outage level of 10 % or less with antenna heights up to 60 m and 45 m for QPSK and 16-QAM, respectively. For the FFT 2K mode, the interference level grows fast as a function of the antenna height. For the QPSK case, the maximum antenna height of 20 m seems to provide still slightly less than 10 % outage. For the 16-QAM case, the interference level is far beyond the acceptable level for all the antenna heights. 50% 45% Effect of the antenna height 16-QAM,2K Outage 40% 35% 30% 25% 20% 15% 10% 5% 0% 16-QAM,4K QPSK,2K QPSK,4K 16-QAM,8K QPSK,8K Antenna height (m) Figure 5-1. The outcome of the simulations of the extended SFN area interferences. The simulation method shows the behavior of the interference when the site cell parameters are varied in over-sized SFN network, and provides an estimate for the maximum useful antenna height in theoretical DVB-H site distribution as a function of the radio parameter values.

129 93 The simulation result of Figure 5-1 is interesting as it indicates that the DVB-H network can be build by exceeding the theoretical SFN size if QPSK or 16-QAM and {FFT=8K, GI=1/4} is utilized together with all the antenna heights of at least up to 200 m. Furthermore, {FFT=8K, GI=1/4} provides a functional QPSK or 16-QAM network for the typical mobile network systems antenna heights. By writing this thesis, only relatively limited information about the over-sized SFN was found. As an example, [Ebu09] mentions the possibility to utilize nation-wide SFN deployment for the DVB-T, if the synchronization of the sites is planned carefully (e.g., by utilizing varying delays), but no analysis is presented. Nevertheless, the reference mentions that "For DVB-T, only the most rugged system variants allow for a national extension of the SFN coverage area (for larger countries). These rugged system variants, however, provide only restricted data capacity (typically, 5 6 Mbit/s). For more practical system variants, with a data capacity between 13 and 24 Mbit/s, the size of the SFN coverage area is restricted to a diameter of km." The [Ebu05] shows an example about the effect of the slightly over-sized SFN, by investigating the complete area location probability in geographical format. A fixed antenna height of 150 m, and GI 1/4 and FFT 8K was selected for the study. The study mentions that the inter-transmitter distance in an SFN should not exceed by too much the distance equivalent to the guard interval length. The result shows the minimum coverage probability as a function of the transmitter power. It was concluded that the coverage probability did not fall below the designed value of 95% in any part of the network regardless of the over-sized SFN. It should be noted that the study was made for a relatively small amount of sites (7), and the setup was terrestrial digital television system specific. The information about the more specific effect of the transmitter antenna height on the quality and utilization of the SFN especially in typical DVB-H environment was still missing from the identified references. Even the simulations presented on this thesis are based on a fixed DVB-H transmitter power, the results in anyway align with the observation of [Ebu09] about the usefulness of the most robust variants of the DVB-T that works as a nationwide solution, i.e. in an over-sized SFN. Publication III confirms thus the possibility to extend the SFN by having investigated a large amount of the DVB-H sites with a typical DVB-H transmitter power that can be assumed to be between the power levels typically utilized in the DVB-T and mobile communications, and shows the impact of the antenna installation solution in DVB-H specific environment.

130 SFN Gain vs. Interferences [Publications IV, VII, X] In addition to the interferences in the over-dimensioned SFN, there are also overlapping carriers producing SFN gain. Publications IV and X indicate that the balancing of the SFN interferences and SFN gain can be done in controlled way. The presented studies show a methodology to simulate the quality level of the over-sized SFN by using the theoretical hexagonal model as well as practical site locations. The modelling of the radio interface was done by applying the long term and fast fading profiles in snap-shot simulation rounds. The simulation was repeated 60,001 times in order to achieve statistical reliability. It should be noted that for the simulations that used Rayleigh fading (that is applied only in Publication VII), the modelling of the fading was simplified and considered over the whole OFDM channel per simulation round. In the more detailed simulation, the attenuation peaks of the Rayleigh fading should affect only on part of the OFDM sub-carriers. As this investigation was done in higher level, and due to the practical limitations of the simulator, the sub-carrier specific behaviour was not considered and flat fading was applied even if the channel requires frequency selective approach. This is justified as the DVB-H Implementation Guidelines document [Dvb09] considers the respective performance limits as function of the overall C/N values. Nevertheless, the more in-depth modelling of the fast fading in wide-band DVB-H signal can be identified as a further study item. In case of a non-interfered network, i.e. when the SFN area is dimensioned in such a way that the common distance of the sites never exceeds the maximum SFN limit, the SFN gain can be estimated by simulating and observing the cumulative presentation of the sum of the carrier power. If the SFN limit is exceeded, part of the sites starts acting as interfering sources. Simulations were carried out in this thesis in order to investigate the behaviour of the balance of the SFN gain of the useful carriers, and the negative SFN gain that the interferers cause. Figure 5-2 shows an example of the balance that can be obtained with QPSK and the investigated radio parameter values presented in Publication IV (Figure 12). Publication IV estimated the maximum SFN gain in a slightly more pessimistic way due to the difference in the power summing whilst Figure 5-2 presents the outcome of the direct power summing. The analysis shows that when the radio parameter values presented in Publication IV are utilized, the parameter set of {QPSK, GI=1/4 and FFT=8K or 4K} produces an interference-free network for sites. In this analysis, the sites are located according to the hexagonal network layout and are overlapping ideally. It can be seen from the figure that as the amount of the sites grows, the SFN gain (sum of the received power levels compared to a single site) reaches the saturation point which in this case is about 6.0 db. The observation has been done inside the calculated site cell areas, by using a limit of the radius of the site cell that complies with the designing value of 10% outage probability in the site cell area. The parameter set of {QPSK,

131 95 GI=1/4, FFT=2K} is still capable of producing SFN gain of about 3 db in a large SFN, even the interference starts taking place after K=7 is exceeded. The parameter set of {QPSK, GI=1/8, FFT=2K} produces SFN gain in a small SFN, but the gain starts decreasing when K grows. With K value between 9 and 12, the SFN gain is 0 db. 8.0 SFN gain (db), QPSK K=1 K=3 K=4 K=7 K=9 K=12 K=16 K=19 K= GI1/4,8K GI1/4,4K GI1/4,2K GI1/8,2K GI1/16,2K GI1/32,2K Figure 5-2. An example of the balancing of SFN gain and SFN interference level in long-term fading channel. K represents the SFN reuse pattern size, i.e. the number of sites within the SFN area. Figure 5-3 shows the simulation result for the 16-QAM mode, when otherwise the same parameter values were applied. The figure indicates that about 6.3 db value can be achieved with the parameter set of {16-QAM, GI=1/4, FFT=4K or 8K} in a large SFN network at least 19 sites. When utilizing the applied radio parameter values of Publication IV, these provide thus a non-interfering SFN network. The parameter set of {16-QAM, GI=1/8, FFT=2K} follows the performance of the latter ones until K grows to 16, resulting an optimal SFN gain of about 6.0 db. When K grows more, small interference takes place reducing slightly the gain. Nevertheless, with this parameter set the SFN gain is between 5.5 and 6.0 db in the range of K= For the parameter set of {16-QAM, GI=1/8, FFT=2K}, the SFN gain first grows along the K, but starts soon decreasing due to the growing proportion of interferences. According to the simulations, this mode still provides an SFN gain of 0 db between K values of 9 and 12. The figure also indicates that the rest of the parameter values create heavy interference for all the multi-site cases.

132 SFN gain (db), 16-QAM K=1 K=3 K=4 K=7 K=9 K=12 K=16 K=19 K= GI1/4,8K GI1/4,4K GI1/4,2K GI1/8,2K GI1/16,2K GI1/32,2K Figure 5-3. The simulation results of the SFN gain and interference balancing for the 16-QAM cases. The results of Publication IV indicate that whenever the maximum SFN is exceeded, the related interference level tends to accumulate primarily to the outer boundaries of the service area. Even if the interference is present nearer to the centre areas, the serving sites produce sufficiently high carrier level in order to cope with the interference levels. The smaller the SFN diameter is, the more the interference points accumulate to the boundaries, finally filling the rest of the areas also inside the network centre. Figure 5-4. An example of the simulation results in Mexico City layout with parameter set of B=6MHz, QPSK modulation, CR=1/2, MPE-FEC=1/2, GI=1/32 and FFT=8K in a combined long-term and Rayleigh fading channel. The map in left hand site represents the carrier distribution with noise level as reference, and the right one the interference distribution. The site circles indicate the height of the antennas.

133 97 Figure 5-5. The C/(N+I) distribution of the previous example. Publication VII clarifies the phenomena. Figure 5-4 and Figure 5-5 summarise the behaviour of the interference in a practical environment presented in Publication VII. This example shows the effect of the smallest defined GI. The results of Publications III, IV, VII and X correlate with the reference [Sil06]. The outcome of the reference shows that the interferences, which are a result of the long distance paths of SFN, tend to accumulate to the outer parts of the investigated network. Only few references for the overall effect of the SFN gain were found by writing this thesis. The most relevant is [Ebu05] which is considering the SFN gain over the whole network, not only in certain limited locations as the rest of the identified references study. In particular, the reference has concluded that the additive network gain is not constant and is an increasing function of the percentage location probability of interest. This indicates that the simulations presented in this thesis are more realistic for estimating the SFN gain over the whole area than typical, somewhat filed tests with a limited drive route, or theoretical calculations of only few sites. As an additional note, it has been shown in [Bov09] that even in good field the Modulation Error Rate (MER) can increase and thus destroy the potential SFN gain in some of the locations. Nevertheless, based on [Bov09], it is not fully clear, what is causing the increased MER in a relatively good field. The problem might be caused by the non-ideal synchronization (as the reference utilized gap-filler for the secondary source) or practical limitations in the receiver performance. In general, the constructive SFN gain can be assumed to be best in the border areas of the cells where no clearly dominating signals are found MPE-FEC Gain [Publication VIII] The detailed network planning might require regional adjustment of the parameters. As an example, the MPE-FEC rate can be selected lower in the areas that clearly have a good

134 98 coverage. Even the MPE-FEC gain can be in order of several db as indicated in [Far05b, p. 24] and [Far05c, p. 11], MPE-FEC does not bring added value in areas with a good field strength as it does not wake up until the received power level reaches a critical level as has been shown in Publication VIII, except against the effects of the impulse noise. In order to enhance the capacity in such areas where impulse noise is present, low MPE-FEC rate can be used as has been concluded in Publication VIII. The completely removed MPE-FEC might be risky as even in good coverage there are probably spots with sufficiently low field, impulse noise and multipath propagated radio signals, where even light MPE-FEC rate does bring enhancement. Outdoor The field measurement guideline [Bou06] was taken as a basis for the field tests carried out in Publications II, VIII and IX. As the analysis in Publication VIII concludes, the MPE-FEC is normally useful only when the received power level is low enough. Figure 5-6 shows the basic principle of the functional scale of MPE-FEC via a case example presented in Publication VIII, Figure 4, indicating that the RSSI window where MPE-FEC takes effect and is able to correct the erroneous frames is relatively small and near the performance limits of the site cell. This means, that in mid-level and good radio field, MPE-FEC does not give added value even if there are multi-propagated radio components present. This applies for the indoor and outdoor pedestrian and vehicular environments. With the used parameter set, the Doppler shift was not the limiting factor in the city areas as the maximum velocity normally does not reach the limits of the maximum velocity FER/MFER%, cumulative over the route FER0, MFER0 cum% FER1, MFER1 cum% FER1, MFER0 cum% dbm Figure 5-6. An example of the cumulative distribution of occurred frame errors as function of the received power level. As the MPE-FEC reserves a part of capacity from the total radio channel bandwidth, the balancing of the capacity and the MPE-FEC gain should be planned according to the operational environment. Based on this analysis, it would be recommendable to limit the MPE-FEC rate to lower values of the scale in the dense city environments as the MPE-FEC reserves thus only

135 99 small proportion of the capacity but still corrects to some extent the occurred frame errors, including the ones caused by possible impulse noise as shown in Figure 5-7 (example from Publication VIII, Figure 6) FER and MFER (re sults in scale 0-20 % pe r Prx value ) FER% MFER % 10% criteria 5% criteria % dbm Figure 5-7. An example of the error recovery in case of the impulse noise. As can be seen, it is impossible to estimate the numerical value of the MPE-FEC gain in this case by comparing the 5 % cumulative occurrence point. Nevertheless, the MPE-FEC has cleaned the impulse interference peaks that the basic FEC could not handle. This specific interference occurred in a specific spot near a mechanical escalator in indoor environment, and was repeated systematically. The light MPE-FEC usage would be justified in indoor environment as the field tends to drop fast in the site cell edge according to the laboratory and indoor field measurements carried out in this study. On the other hand, in the typical urban and dense urban environment, the DVB-H radio coverage is normally good enough in order to provide sufficiently high quality indoor coverage. This means that the field in outdoor is so high that MPE-FEC is not normally needed, but recommendable due to multipath fading. Also sub-urban area within the DVB-H coverage seems to benefit only little from MPE-FEC. In the buildings, the gain seems to be even less than in urban areas due to the lower amount of multipath components. Nevertheless, approaching the service area edge, the field strength is sufficiently low in order to activate the MPE-FEC, which brings about 1 db MPE-FEC gain in normal motorway environment with the lowest MPE-FEC rate. When MPE-FEC is studied in lower received power level areas, i.e. near the site cell edge, the effect of MPE-FEC is obviously more remarkable. Publication II shows the MPE-FEC behaviour in typical vehicular environments in sub-urban area type. It shows that the MPE-FEC

136 100 performance depends clearly on the environment, and the variation of the effect can be seen in typical urban type of cases with sufficiently reflected multipath components. Publication IX shows the further post-processed analysis about the MPE-FEC behaviour for different radio channel types. The clearest result of MPE-FEC was obtained for the nearly LOS situation in the main lobe of the transmitter antenna. Figure 5-8 presents the post-processed values for the MPE-FEC gain with a varying QPSK, code rate, MPE-FEC rate and FFT size, the GI being 1/4 in all the cases. The analysis presented in Chapter was applied. Table 5-1summarizes the MPE-FEC values for each investigated parameter set for the main lobe case. The summary indicates that the MPE-FEC gain can be more than 7 db in the investigated radio channel type. It seems that in this type of channel, regardless of the modulation and CR, the MPE-FEC gain correlates with the FFT size, i.e. the lower the FFT size is, the less is the MPE-FEC gain also is in general except in the case of (QPSK, CR 2/3, MPE- FEC 1/2, FFT 2K) which has produced the highest MPE-FEC gain. For the (QPSK, CR 1/2, MPE-FEC 1/2, FFT 8K), the result is not possible to interpret although around 4 db gain could be seen when the strongly varying curves are averaged. In general, if there are too many variations in the FER and MFER curves, no accurate estimate about the breaking points of FER5 and/or MFER 5 can be made with the presented method. FER 5% and MFER 5%, RSSI per parameter set, terminal 1 QPSK, CR 1/2, MPE 1/2, FFT 8k 16-QAM, CR 2/3, MPE 2/3, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 2k 16-QAM, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 2/3, FFT 4k QPSK, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 1/2, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 1/2, FFT 8k QPSK, CR 2/3, MPE 1/2, FFT 2k dbm Figure 5-8. Summary of FER5 and MFER5 analysis in a single site cell case.

137 101 The measurement results were not always possible to interpret explicitly due to the large variation of FER and MFER curves. This indicates the insufficient amount of samples per RSSI category. In those cases, further measurements with a larger set of collected data would be needed. Nevertheless, the values that could be obtained are shown in Table 5-1. Table 5-1. The MPE-FEC effect for the investigated parameter values in a single site cell case. Modulation CR MPE-FEC rate FFT MPE-FEC gain 16-QAM 2/3 1/2 2K QAM 2/3 2/3 2K 1.5 QPSK 2/3 2/3 4K QAM 2/3 2/3 4K 2.8 QPSK 2/3 2/3 2K QAM 2/3 1/2 4K 4.3 QPSK 2/3 1/2 4K 6.2 QPSK 2/3 2/3 8K QAM 1/2 2/3 8K 6.7 QPSK 2/3 1/2 8K 7.4 QPSK 2/3 1/2 2K 7.7 The results are in align with the references [Far05b, p. 24] and [Far05c, p. 11] which have concluded that the MPE-FEC gain might be in order of 5 9 db in certain conditions. As a conclusion of the vehicular test in near-los main lobe, the MPE-FEC gain can be seen clearly when the field reaches the critical limit where FEC does not help but MPE-FEC still takes care of the frame errors, i.e. in the area where MPE-FEC wakes up. The level seems to be in order of 1 4 for FFT 2K, about 2 6 for FFT 4K, and about 6 7 for FFT 8K. Nevertheless, when sufficiently good overlapping is found, as concluded in Publication VIII the MPE-FEC does not improve the performance except for the occasional moments where impulse noise is present. In the latter case, MPE-FEC seems to clean the impulses quite efficiently in the entire received power level interval. Comparison of the outdoor results According to [Mil06, p. 20], with moderate Doppler values the MPE-FEC curve is very flat and has a constant gain of 6 to 7 db compared to the DVB-T with the same receiver. The effect of the code length is studied in [Apa06a, p. 28]. There is a relationship between the MPE-FEC performance and the number of MPE-FEC rows noted. In the studies of Publication II, the row number was varied but the effect was not analysed. The overall MPE-FEC gain results obtained

138 102 in [Apa06a, p. 23] concludes though that the MPE-FEC has positive effect in mobile reception. When the vehicle speed is well below the Doppler limit as also has been the case in Publication II and IX, [Apa06a, p. 23] has found that the effect can be seen to some extent. The analysis presented in [Apa06a] is comparable with Publications II and IX, presenting curves that indicate the frame error ratio after MPE-FEC. Even [Apa06a] has not specifically analysed the MFER5 point, it is possible to interpret the curves of [Apa06a, p 24] which indicates the breaking point of MFER5 for {QPSK, CR 2/3, MPE-FEC 2/3} in approximately 85 dbm region whereas the {QPSK, CR 1/2, MPE-FEC 1}, i.e. no MPE-FEC produces about 83 dbm and thus 2 db gain for MFER5. This does not give direct comparison though as the CR has been varied in the same graph. Nevertheless, [Apa06a, p 23] has concluded that the performance seem to be better, when main part of the overall redundancy is used by convolutional coding, i.e. it seems to be more efficient to combine CR 1/2 and MPE-FEC 2/3 rather than CR 2/3 and MPE-FEC 1/2. According to Publication IX, Figure 28, this phenomenon is not very clearly seen by varying the values of CR and MPE-FEC between 1/2 and 2/3. As an example, in the vehicular environment with almost LOS in suburban area, the combination of {QPSK, CR 2/3, MPE-FEC 1/2, FFT 8K, GI 1/4} has produced almost 7 db MPE-FEC gain, and {QPSK CR 1/2, MPE-FEC 1/2, FFT 8K, GI 1/4} slightly over 7 db. As another example, Publication IX, Table 1 and Table 2 show the results with a mixture of different radio channel types in vehicular environment in sub-urban area. For GI 1/4 and FFT 8K, the combination of (QPSK, CR 1/2, MPE-FEC 2/3) gives 4.0 db gain whereas (QPSK, CR 2/3, MPE-FEC 1/2) gives 6.5 db. Furthermore, (16-QAM, CR 1/2, MPE-FEC 2/3) gives 4.9 db and (16-QAM, CR 2/3, MPE-FEC 1/2) gives 0.3 db. These are naturally snap-shot examples with about 25-minute data collection period per each parameter setting, but they do not confirm the statement of strong CR and lighter MPE-FEC producing better performance compared to a strong MPE-FEC and lighter CR. As it seems that the [Apa06a] is not based on very complete field test data, further investigations with larger data collection would be needed in order to confirm this behaviour. There is also an evolution version of MPE-FEC developed for the satellite version of DVB-H, i.e. DVB-SH. It is backwards compatible with the link layer of the original MPE-FEC, but uses sliding mechanisms. According to [Goz08], the new version called MPE-iFEC improves the performance of MPE-FEC, and it could be used also in DVB-H. The comparison of the MPE- FEC and MPE-iFEC would be thus an interesting further study item. Indoor Publication VIII presents pedestrian field tests in indoor environment. A laboratory network with a single site cell was utilised with a radius of about 50 meters. The methodology presented

139 103 in Publication II and Chapter of this thesis was applied in the respective analysis, i.e. FER5 and MFER5 behaviour was observed as function of RSSI. The outcome of the laboratory tests shows that the MPE-FEC rate has an expected effect on the performance. A MPE-FEC rate of 15 % produced MPE-FEC gains close to zero whereas 25 % indicated about 0.5 db gain, and 35 % rate showed about 2 db gains. The respective environment was challenging for the MPE-FEC as the channel included only one dominating radio path. This 1-tap environment would not provide too much MPE-FEC gain. As Publication VIII shows, the indoor tests were carried out also in live network. There was only MPE-FEC rate 15 % in use. The comparative analysis showed that the laboratory results show pessimistic values for MPE-FEC 15 %, being about 0.2 db, whereas the live network results with the same MPE-FEC rate provided about 1 db MPE-FEC gains. The value depends though on the case; where low amount of multipath radio components are expected, the gain is respectively lower. It is interesting to note that in those cases where obviously impulse noise occurred, even low MPE-FEC rate indicated good performance. Comparison of the indoor results The reference [Apa06a, p. 25] presents indoor test results for the pedestrian environment. The presentation is slightly different from the one used in Publication VIII, but it is possible to interpret the general behaviour of the MPE-FEC in the 5 % cumulative point. Publication VIII uses 16-QAM, CR 1/2, FFT 8K and GI 1/8. FEC rows were 512 for 15 % and 25 % MPE-FEC, and 768 for MPE-FEC 35 %. In [Apa06a, p. 25], the settings have been QPSK, CR 1/2 and 512 FEC rows. Some of the MPE-FEC results can be obtained also via the impulse noise test case in discharging the impulse area as shown in [Apa06a, p. 28]. The setting was 16-QAM, CR 1/2, MPE-FEC 2/3 (35 %). [Apa06a] does not describe thoroughly the indoor environment or the test setup, but the comparative results of Table 5-2 seem to fall in the same range for the MPE-FEC % cases. Table 5-2. Comparison of indoor pedestrian MPE-FEC results obtained form Publication VIII and [Apa06b]. MPE-FEC Publication VIII [Apa06a] 15% (2 terminals) not possible to interpret implicitly 25% 0.4 (1 terminal) % (2 terminals) 1.5 (via impulse noise tests), 1.0 (via non-impulse noise tests)

140 Antenna down-tilt [Publication V] For the remaining site parameters, the down-tilting of the transmitter antenna is one of the most relevant optimisation items. The studies carried out in Publication V showed that the effect can be seen clearly when varying the down-tilt of the sites in order of 0 6 degrees. Its relevance has been studied in [Fan06b]. It concluded via propagation analysis using smooth-earth and Longley-Rice models that significant coverage improvements may be obtained by simply increasing the antenna beam tilt. It is worth noting that the down-tilt of the main beam raises at the same time the back beam of the antenna. Especially in the LOS environment, this might cause fragmented coverage spots in long distances in the back lobe of the antenna. When they do not exceed the SFN limits, they do not have harmful effects within SFN. On the other hand, they also can be interference sources outside the limits of the SFN. It is thus important to study the effect of the down-tilt in the detailed planning phase of DVB-H network. 5.2 User Experience Related Parameters The user experiences are important feedback for the operators in order to tune the radio and core network of DVB-H correctly. The experiences are normally collected already in the trial or pilot phase of the network. There is often a friendly user phase organized before the commercial launch of a DVB-H network, and selected customers can test the services and give feedback to the operator. This helps in understanding the technical limits for the adequate service level, preferred program types as well as the commercial aspects with cost structure of the provided services. The user experiences can also be utilized in the operational phase of the network. Depending on the technical abilities of the DVB-H network, it might be able to collect and send automatically the statistics of the usage via the interaction channel, i.e. GSM or UMTS packet service. One of the concrete parameters related to the user experiences is the time-slice as the average channel switching time depends directly on that. The correct dimensioning is important in order to create fluent user experiences when the channel is switched. It can be estimated that the average waiting period should not exceed considerably that of terrestrial DVB system in fixed use. On the other hand, the channel switching time has direct relation with the battery saving value. This is logical as faster the switching time is, more often the receiver has to be switched on. A short unpublished snap-shot test with the variation of the Time Slicing window size was carried out in the test setup aside the activities that are explained in Publication II. It can be

141 105 estimated by a subjective test, that four seconds is already too long waiting time, two seconds being in the limit of typical tolerance of the user. As the battery saving benefit lowers along the faster switching time, it would be important to carry out related user experience test cases with sufficiently large audience. The power saving at the receiver can be defined as the fraction of the time during which the front-end of the receiver is inactive. A typical battery saving value is at least 90 % [Hen05, p. 16]. The source [Gar07, p.5] has investigated the battery saving effect of the Time Slicing functionality. The results show that in a realistic scenario, the TS rate could be approximately 10 Mb/s, with a typical video service bit rate of 500 kb/s. Defining bursts of 2 Mb, each burst has duration of 200 ms and the inter-burst time is 4 seconds. The theoretical power saving is thus 96%. It should be noted, though, that in the complete DVB-H terminal the receiver end is only one part that consumes battery. As there are other functionalities and components like video stream player, display, GSM/UMTS terminal, among others, the percentage of the battery saving of the DVB-H is logically lower when the reference is the complete equipment. In this sense, the optimal balance of the channel switching time (i.e. user experience as the usability) and battery saving (i.e. user experience as the usage time without power outlet) can be found probably around 2 seconds area. Other important aspect is related to the video and audio quality. The bit rate as well as frame rate should thus be dimensioned correctly for each content type. 5.3 Cost Optimization [Publication I] The whole chain of the DVB-H network, including the IPDC part (core) and radio networks, includes many business parties [Sat06]. As each one of the delivery chain representatives has their costs due to the investments, they might face the cost optimisation issues. It is thus essential to take into account the cost effect of the technical solutions of the DVB-H network deployment as deeply as possible as there might be several parameter values producing a zeropoint in the derivative of the cost as function of the selected parameter values Cost optimization in the non-interfered network Publication I shows the methodology for the seeking of an optimal parameter set for the OPEX and the CAPEX of the network. It is based on the adjustment of the essential variables, i.e. antenna heights and transmitter power levels, in order to vary the size of the single site cells of the network assuming there are no interferences found in the planned area, i.e. either MFN or non-interfered SFN is used. Based on the methodology of Publication I, a more detailed version of the cost optimization module was formed in this thesis as shown in Figure 5-9.

142 106 Previous planning item Set general limits Maximum available transmitter power Maximum available antenna gains Maximum allowed radiating power level Adjust general parameter values Modulation, CR, MPE-FEC, FFT, GI Select parameter values for each site Power levels Antenna heights no Adjust parameter values for selected site Power levels Antenna heights yes Revision Is it possible to still tune site parameters? Cost planning Adjust the cost prediction according to the detailed coverage plan (detailed CAPEX/OPEX optimization). Acceptable level of costs? No Yes Next planning item Figure 5-9. Cost optimization process in non-interfered DVB-H network. The variables for the complete techno-economic network optimisation include the total cost of the site, radius of the site cell and radiating power level vs. the respective cost of the equipment. There are various sources of information as well as tools in order to identify the main aspects in the physical environment, including the geographical distribution of the population and economical levels, data bases of the site locations (existing towers, the height of the antennas), digital maps with the area types and respective propagation models etc. In the detailed network optimisation, unlike in the nominal plan, uniform power levels would not provide the best performance due to the practical differences of the areas and site solutions, so the power levels and antenna heights should be planned on site-by-site basis in this phase. In the initial phase of the planning, the area type can be assumed as uniform, or that there are only few different area types each one being uniform as a cluster class. As an example, the area inside of ring road of a large city could be assumed as dense city area type, whilst outside of the ring road, a sub-urban cluster could be applied. The first step of the cost estimation is to identify all the relevant items that do have cost effect on the network implementation. Then, the CAPEX and OPEX of a single site can be estimated, with a respective radio coverage area in that cluster type. In order to have a reliable estimate of

143 107 the cost effect of the whole network, a sufficiently large area can be selected and filled in with the sites according to the coverage area of each transmitter. In the nominal planning phase, a hexagonal model can be utilized whilst in the detailed planning, site-by-site adjustment and real map data should be applied. Then, by varying the essential parameters like antenna heights and gains, transmitter power levels etc., this process can be repeated. The combined CAPEX and OPEX curves now indicate what would be optimal solution as function of time. The CAPEX represents the initial one-time cost of the site. It should contain cost information about the investigated transmitter type, antenna system with related jumpers, connectors, feeders and power splitters, other material for the mounting etc, the work for site acquisition, legal and technical preparation, and finally the work for actual installation and commissioning of the site. The OPEX indicates the costs that are generated during the time after the initial installation of the site. The most important long-term cost items are related to the transmission type, maintenance of the equipment, possible tower and site rental agreements, and electrical power consumption. Publication I shows the analysis by applying an estimate for the transmitter costs as function of the power level. The cost information was obtained in a snap-shot way from the typical commercial transmitter models and the values were normalized taking the cost of the 500 W- transmitter as a reference. It is obvious that this cost information can vary vastly depending on the case, including the possible discounts for a large numbers of transmitters. Nevertheless, it indicates the cost behaviour as the power level varies. As a result, it is possible to create a power vs. cost -dependency graph, i.e. what is the cost of single watt as function of transmitter type. Next, the CAPEX and OPEX per site can be presented in graphical format by normalizing the values. In Publication I, the transmitter type producing 500 W is selected as a reference. For the CAPEX, the higher power transmitters are logically more expensive, they require more installation work, and the respective feeders are more expensive. As for the OPEX, the transmission and site rent are the major cost items. Assuming that the power consumption is about six times more than the produced output power, the energy consumption can be important for the highest power transmitter models. In the example shown in Publication I, Figure 4, the energy consumption represents about 25 % of the total operating costs of a single site. For the coverage estimation with Okumura-Hata propagation model, Paper I assumes that QPSK is used, area location probability of 90 %, shadowing margin 5.5 db, frequency 700 MHz, transmitter antenna gain 13 dbi, receiver antenna gain 7 dbi, CR 1/2, MPE-FEC 3/4. This results a received power level requirement of 87 dbm. For each transmitter power level case, the feeder was selected according to the output power requirements, i.e. less output power requires thinner feeder, which is more economical and easier to install, but on the other hand its loss is higher reducing the coverage area. As there is clear inter-dependency, this item requires additional iteration rounds in order to find the optimal balance between the total cost and

144 108 performance, i.e. the task is to select the feeder that complies with the maximum power requirement and that keeps the total cost on minimum level by balancing the cable type (attenuation) and its cost. In the analysis shown in Publication I, the cables were selected in such a way that power levels up to 3400 W use 1 5/8 feeder (with 1.9 db attenuation per 100 m) and power classes 4700 W and 9000 W used 3 feeders (with 1.5 db attenuation per 100 m but resulting about 30 % more costly material than the previous one). The cable cost for different cable models can vary several tens of percents. The portion of the cable cost reduces to only some percents when the overall site cost gets higher, i.e. when the transmitter power increases. Table 5-3. The summary of pros and cons of DVB-H transmitter power levels. Transmitter type Benefits Drawbacks Low power High power Low energy consumption. Economical and easy to install and maintain. The installation is possible also in rooftop sites due to the smaller safety zones. Low number of transmitters which results easier identification and obtaining of sites. The network can be build up relatively fast. Provides large coverage areas especially in relatively open area and if the antenna can be installed high. Single site coverage small which requires high number of sites. The obtaining of sites is not straightforward in practice. High transmission costs in case of terrestrial solution. Power consumption rises exponentially as the transmitter power is higher. Liquid cooled transmitter needs additional maintenance. The outages in coverage if obstacles (especially in urban areas) which requires additional gap fillers. A single site breakdown causes a large service outage. The hexagonal model used in the large network area analysis is theoretical as the realistic sites cannot normally be obtained in such a uniform way. Nevertheless, this approach gives possibility to compare different cases in the initial phase of the network planning. In addition, the overlapping areas can be utilized in the practical network for handover purposes of multi frequency network, or for obtaining SFN gain in single frequency network. As Publication I concluded, e.g. the operating cost effect of transmission and electrical power might be considerable. Table 5-3 summarises the most relevant cost related arguments in the detailed optimization.

145 Cost optimization in interfered SFN network When the over-sized SFN is applied, the cost optimization method presented in Chapter 4 can still be used, but a feed-back loop should be added to the process as proposed in Figure The SFN simulator method of Publications III, IV and X can be utilized in the feedback loop. The iteration level of the cost optimization thus rises, but in the in-depth network planning and optimization it is recommendable to carry out as thorough cost optimization as possible as it might have remarkable effects on the final savings in the network deployment and operating phases. Previous planning item Set general limits Maximum available transmitter power Maximum available antenna gains Maximum allowed radiating power level Adjust general parameter values Modulation, CR, MPE-FEC, FFT, GI Select parameter values for each site Power levels Antenna heights no Adjust parameter values for selected site Power levels Antenna heights yes Revision Is it possible to still tune site parameters? Cost planning Adjust the cost prediction according to the detailed coverage plan (detailed CAPEX/OPEX optimization). Acceptable level of costs? No Yes Simulation Analyse the interference level no Is the SFN intereference level < allowed outage% yes Next planning item Figure Cost optimization in interfered DVB-H network. The CAPEX/OPEX analysis of Publication I can be presented in an analytical format in the following way.

146 110 Path loss The initial task is to define the basic loss L (db). The L is obtained from the link budget as presented in Table 3-2. As an example, for the QPSK, CR 1/2 and MPE-FEC 2/3, the required C/N is 11.5 db according to [Dvb09]. With this combination, the maximum allowed path loss in outdoor is db, when the frequency f is 600 MHz, receiver's noise figure F is 5 db, transmitter output power P TX is 2400 W, cable and connector loss L cc is 3 db, power splitter loss L ps is 3 db, transmitter antenna gain G TX is 13.1 dbi, receiver antenna gain 8.4 dbi, and the location variation for 95% area probability is 7.0 db. In the cost analysis, the variables shall be the cable and connector loss L cc, the transmitter power P TX and the height of the transmitter antenna h BS. The receiver antenna can be assumed to be fixed to 1.5 meters. Following the principle of the link budget of Table 3-2, the initial L is estimated as shown in the link budget analysis based on Table 3-2. Cell radius When the maximum allowed path loss is known for the investigated radio parameter values, the next step of the CAPEX/OPEX analysis is to estimate the single cell radius. A uniform analysis can be utilized in the first phase of the analysis, as has been done in Publication I. This gives an overall estimate about the situation by presenting the global values in an ideally overlapping site setup which can be presented by hexagonal layout model. The drawback of this method is that in practice, it is not possible to build the sites according to the ideal distribution the model suggests. Nevertheless, the model provides a functional estimate about the feasible parameter value ranges as for the antenna heights and transmitter power levels. The method presented in Publication I can also be applied in a practical DVB-H network layout, e.g., by extending the coverage and interference analysis of Publication VII with the CAPEX/OPEX module. This uniform cost analysis is based on the estimate of a single cell radius. In the simplest format, the free path loss calculation could be applied with a rough estimate of the attenuation factor that represents different area types. Throughout this thesis, the basic Okumura-Hata model [Hat80] or ITU-R p [Itu07] are used as they provide a realistic estimate for the path loss, and they are commonly utilised in the coverage predictions. If the basic Okumura-Hata model is applied for the CAPEX/OPEX analysis, the cell radius d can be estimated as an example in medium and small city environment as presented in Formulas 3-13 and 3-14, and by utilizing the L as presented in Formula 3-9: [ log( f ) lg( h ) ( (1.1log f 0.7) h (1.56 log f 0.8) ] L BS log( hbs ) 10 m d = (5-1) where L is the outcome of Formula 3-9, f is the frequency in MHz, h bs is the transmitter antenna height, and h m is the mobile antenna height. As soon as the cell radius in known, the next task is

147 111 to model the cost items in order to understand the expenses of the parameter values. The items can be divided to the initial investments (CAPEX), and to the longer term yearly costs (OPEX). CAPEX items The initial network deployment costs include various cost items that impact on the CAPEX. Table 5-4 summarizes the most important variables for a single site. Table 5-4. The most relevant CAPEX items and relations. Variable OPEX item Observations C C Transmitter The transmitter cost depends on the power level P TX as presented 1 in Figure 3-5, and thus on the cell radius. The complexity of the transmitter has impact on the cost, e.g., depending on the air or oil cooling system. C C Antenna system The transmitter power level dictates the minimum power level 2 that the antenna elements should support, with a practical margin. The elements can be omni-directional or directional, latter constructed via the power splitter and separate antenna elements which can also be placed in a vertical array form for each sector, forming thus an omni-cell. C C Antenna feeder The antenna feeder must support the transmitter power level, and 3 certain margin should be taken into account. The antenna feeder has an impact on the L as the cable and connector losses decrease the cell radius. Higher power requires thicker antenna feeder, which is more expensive, but with also lower losses. C C 4 Installation, antenna system The installation cost depends on the height and weight of the antennas, height of the cables, as well as the length of the cable. C C Power splitter If various directional antennas are used, the power splitter is 5 utilized. The cost depends on the power levels and antenna ports. The power splitter has a direct impact on the cell radius. C C Cable brackets The cable mounting requires brackets. The amount and cost 6 depend on the antenna feeder type and antenna height. C C Installation Personnel's compensation of the work. 7 C C Other installation costs Special fees, e.g., due to the helicopter installation etc. 8 C C Site acquisition The identification and acquiring of the site, including personnel's 9 compensation. C C Planning, drawings The preparations of the site, personnel's compensation. 10 C C Miscellaneous costs Any other sufficiently significant site related cost. 11 C C 12 Tower / site building If the site and/or tower are purchased, this item triggers a one-time cost effect. It should be noted that the building costs might be exponential as a function of the tower height.

148 112 As can be seen from Table 5-4, there are various relevant items affecting on the CAPEX. The cost variables that are shown in the table have partial inter-dependencies between the transmitter power levels, antenna height, OFDM parameter values and cell radius. The complete CAPEX for each investigated option is thus: 12 C C C = C (5-2) tot i= 1 i Antenna feeder selection When the transmitter power level is selected for the investigation, the antenna feeder type selection depends on the respective maximum possible transmitter power level. Table 5-5 summarises a set of example values that indicates the relationship between the cable type and power requirements. This information was utilized as a basis in Publication I. Table 5-5. Cable types and main characteristics utilized in the CAPEX/OPEX analysis of this thesis. Type Attenuation at 500 MHz / 100 m P (const) kw / 500 MHz Attenuation at 700 MHz / 100 m P (const) kw / 700 MHz 1/2'' /8'' /8'' /4'' '' As Table 5-5 indicates, the most cost-effective solution (balance of the feeder cost and loss) might be possible to obtain even if the cable type is clearly over-dimensioned as for the maximum power because certain cables might increase clearly the cell radius due to the lower loss. As an example, 1/2'' cable with 100 m antenna height produces about 7.5 db loss whilst the 7/8'' produces 3.5 db. This 4 db difference is significant in the radio link budget, as the same effect can be achieved by, e.g., more than doubling the transmitter output power level. It should be noted that the tower installation for a certain antenna height requires typically additional cable for the equipment room. In the calculations of Publication I, 30 meters extra length is assumed. Assuming that only one piece of cable is utilized per site, the antenna feeder cost item is selected from the list of available types, e.g., in the following way in the presented example:

149 113 1/ 2'' c1/ 2'' / m, PTX < Pmax = 0.82kW ε 7 / 8'' c7 / 8'' / m, PTX < Pmax = 2.65kW ε C C = 1 5 / 8'' 3 c / m, PTX < P = 5.77kW ε (5-3) 1 5 / 8'' max 2 1/ 4'' c2 1/ 4'' / m, PTX < Pmax = 8.21kW ε 3'' c3'' / m, PTX < Pmax = 13.09kW ε It can be decided that the additional power safety margin ε is 10% of the maximum supported power level P max of each antenna feeder type. There are no technical restrictions in the use of any of the antenna feeder type whilst the power limits are taken into account. If the effective output power of the transmitter is 2.16 kw (2.4 kw with TX filter loss of 10%), the feeder type of 7/8'' supports 2.65 kw kw = 2.39 kw, which thus complies with the realistic C power of the 2.4 kw transmitter type. In this case, the antenna feeder and related cost C 3 could thus be selected from {c 7/8'', c 1-5/8'', c 2-1/4'', c 3'' }. The cost c of each antenna feeder should now be known via the available sources. In this analysis, the assumption is that the cost is constant per meter, although in practice there might be additional discounts due to the volume purchase. On the other hand, it should be noted that the feeder diameter has an effect on the installation cost C C 4 due to the more difficult handling of the material and additional weight of the feeder. When the transmitter power level is selected, as well as the transmitter antenna height, the cable type or different options for the cable type can be selected, and the respective cable and connector attenuation L cc as well as total cost can be calculated. The cable is clearly a CAPEX item. It is installed only once and is almost maintenance free, making the related OPEX practically minimal. When estimating the cost effect of different cable types, the balance can be found by evaluating the cost of material and work per cable type (heavier cable requires more work) as well as the effects of different cable losses (thicker cable produces lower losses). It should also be noted that the attenuation is frequency dependent. Table 5-6 summarises a case used in Publication I by presenting the site cell radius obtained with different cable types and transmitter power levels. The assumption is to use all the antennas in 100 m tower. This table indicates that the cable types of {1-5/8'', 2-1/4'', 3''} provide similar coverage areas, so the lowest cost type that is supported by each transmitter power level is an obvious selection in order to optimize the cost. Nevertheless, the table shows that there is a clear difference between {1/2'', 7/8'', 1-5/8''}, the thinner cable producing clearly lower coverage areas. By taking into account this in the cost analysis, i.e., by calculating the final cost of the network if thicker cable is utilized in these lower power cases, the outcome could be that the thicker cable, even the resulting CAPEX is higher than for the thin cables, can provide more cost-efficient OPEX.

150 114 Table 5-6. An example of the site cell radius obtained by varying the cable type and transmitter power levels. N/A is shown if the cable is not suitable for the respective power level. Cell range (km) for cable type of: P TX / W 1/2" 7/8" 1-5/8" 2-1/4" 3" N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A 22.6 OPEX items Table 5-7 summarizes the OPEX items that were utilized in the cost analysis of Publication I. As was the case in CAPEX item identification, also the OPEX items were selected according to the realistic deployment. Table 5-7. The most relevant OPEX items and relations. Variable OPEX item Observations O C Electricity The electricity consumption depends mainly on the transmitter 1 type (power level) as well as the power consumption of the related other site elements like modulator, site cooling system, routers etc. O C Maintenance The maintenance includes principally the transmitter maintenance 2 that depends on the complexity of the transmitter. As an example, air cooled results less site visits than oil cooled, which in turn also requires oil changes. O C Transmission The transmission can be either terrestrial or via satellite. Both 3 have cost effect, terrestrial triggering site specific costs whilst the costs of a single satellite usage can be divided between all the respective sites. O C Tower / site rent If no own site or tower has been obtained, the rental costs of the 4 equipment shelter and / or antenna system usage are added to the OPEX. O C Other expenses Any other item triggering regular costs like average level of the 5 general site maintenance.

151 115 The electrical power consumption P c of the transmitters can be estimated in a general level as a function of the transmitter output power level, unless there is no more accurate estimate available: P = 6 (5-4) c P TX The yearly cost of the transmitter with a power level of P TX is thus: O C1 = 6 PTX ce 24h 365 /1000 (5-5) where c e is the cost of electricity per kwh. Equally, the other OPEX items should be estimated in absolute values in yearly basis, which results the total OPEX per year: 5 O O C = C (5-6) tot / year i= 1 i The yearly costs might vary over the time, but in this level analysis it is sufficient to estimate an average cost per year. Combined CAPEX/OPEX of the whole network When the CAPEX and OPEX are calculated for all the investigated options as a function of transmitter power levels (transmitter types), antenna heights and OFDM parameters (which results the balance between the capacity and coverage), the combined network cost can be calculated. The single cell radius d, or the distance between the site and calculated cell edge with the given area location probability, indicates how many partially overlapping sites can be located within a certain area, which can be the whole planned network or part of that. By taking into account the overlapping share of the hexagonal model as presented in Figure 6-8 of the simulator Appendix, and in Figure 5 of Publication I, the number of the site cells N cells is: N cells = A A tot cell A = πd tot 2 cell 0.827, (5-7) where A tot is the whole investigated area (km 2 ) and A cell is the single site cell coverage area (km 2 ). This information is the key for the CAPEX/OPEX analysis in Publication I as it provides the possibility to compare the efficiency of the selected parameters in order to minimize the cost when filling the area with the full coverage area. The total cost of the network is thus:

152 116 C tot cells C O ( C + C N ) = N /, (5-8) tot tot year y where N y is the number of the years. The aim is now calculate as many options as are logical to investigate, by following the planning process from the beginning. The first task is to fix the maximum offered capacity, which results a certain limited set of OFDM parameter values for the modulation scheme, CR and MPE-FEC. The analysis can also be limited by selecting realistic general values for the possible maximum antenna heights in the investigated area. Also the regulatory radiation values might be needed to be taken into account when limiting the antenna heights and maximum possible radiating power levels. The most logical set of parameters can now be tested by calculating the maximum path loss L and respective cell radius d, which results the cost estimate for each case. The results of Publication I show that there is an optimal point in the combined CAPEX and OPEX curves when varying the transmitter powers. In the presented case analysis of Publication I, the optimal transmitter power class is found in mid-range models. A further analysis shows that the optimal point varies during the time, higher power transmitters becoming more costefficient solution in the longer run CAPEX and OPEX (4 years) h(ant)=50m h(ant)=75m h(ant)=100m h(ant)=150m h(ant)=200m h(ant)=250m h(ant)=300m 100W 200W 500W 750W 1500W 2800W 3400W 4700W 9000W 0,1 Figure The cost effect of DVB-H network when the antenna height is varied. This case includes relatively high transmission costs as utilized in Publication I.

153 117 There are many inter-dependencies between the variables which require various iteration rounds if the optimal set of parameter values are investigated thoroughly. As an example, by keeping the variables the same as in the presented example of Publication I, but varying the antenna height in a sub-urban area, Figure 5-11 can be obtained. The presented values are normalized to 500W-transmitter case and for the antenna height h ant of 100 meters. The analysis shows the combined CAPEX and OPEX after 4 years of operation. Figure 5-11 indicates that the higher the antenna is located the lower the cost of the network is. This is logical as the coverage areas get larger as function of the antenna height. It is interesting to note though, that in this specific case and in a range of about meters of antenna height, the optimal transmitter power level is in mid-range models whereas in lowest and highest antenna locations the high-power solution is the optimal in each antenna height category. For the high antenna cases this is understandable as the number of high-power transmitters is relatively low in order to achieve the same coverage area as with the lower power cases. The explanation for the low antenna height behaviour is that the number of the highpower sites in that specific situation is low enough to favour the high-power transmitters even if their relative cost of power consumption is clearly higher. CAPEX and OPEX (4 yers), low transmission costs 10 1 h(ant)=50m h(ant)=75m h(ant)=100m h(ant)=150m h(ant)=200m h(ant)=250m h(ant)=300m 100W 200W 500W 750W 1500W 2800W 3400W 4700W 9000W 0,1 Figure An example of the cost effect as function of the antenna height when the transmission costs are assumed to be low.

154 118 In the above mentioned examples, the transmission cost has been assumed to be based on leased lines with relatively high cost (in order of 150,000 euros per year per site). The tower rental and site cost has been assumed to be in order of 20,000 euros per year. If the transmission cost can be reduced considerably, e.g. down to 15,000 euros per year per site, the result changes. This is due to the fact that the relative proportion of the electricity consumption gets much higher in the total OPEX. Figure 5-12 shows a related analysis, with all the other parameters being the same as in case of Figure The reference for the normalized y-values (costs) has been kept the same in Figure 5-12 as in Figure 5-11, i.e. the analysis is based on the 500 W transmitter case and for 100 meters of antenna height. As can be noted, the relative cost of the network is considerably lower with low transmission costs. In this case, the highest power case of 9,000 W results now about 49 % of the OPEX out of the total operating expenses. This effect can be seen in Figure 5-12 as the optimal point per antenna height category is clearly indicating the mid-range power transmitters. It is interesting to note that the above mentioned analysis results in optimal transmitter power level and antenna height values, which in many cases indicates the mid-level range. As a comparison, [Mil06, p. 63] has identified the optimal point in mid antenna height in city area case, but the transmitter power seem to behave in linear way, i.e. the higher the power level is, the more economical the network cost is as the coverage areas of single site is large. The lack of clear optimal points indicates that the study might be missing of such detailed inputs that was used in Publication I. Furthermore, the operating costs were not taken into account in [Mil06, p. 63]. Cost optimization in varying sites The method shown in this thesis for the CAPEX/OPEX optimization is based on the uniform parameter values over the whole investigated network. Nevertheless, the method can also be applied for the varying site configurations, e.g., in the environment presented in Publication VII. Furthermore, the analysis can be combined to the coverage area study, as well as to the SFN gain and SFN interference balancing study by varying the radio parameters and by calculating the respective costs either in uniform case or site-dependent parameter values. Conclusions As a conclusion of the cost optimisation, it is important to identify all the relevant cost items case-by-case and carry out the related analysis. If the DVB-H transmitter antennas can be located relatively high, it helps in optimizing the coverage areas. There are limits though, the rise of power consumption being significant in highest power level transmitters. As Figure 5-11 and Figure 5-12 shows, the antenna height does have a considerable effect on CAPEX and OPEX and the optimal cost point can be identified.

155 119 On the other hand, the optimal height might depend also on the other aspects like on the finetuning of the maximum size of the single frequency network, i.e. when sufficiently low EIRP levels are used, the theoretical SFN diameter is not necessarily a limit in practice and the single frequency can be used in large areas as shown in Chapter An optimal solution would possibly include part of the antennas installed as high as possible to broadcast towers and others in telecom towers and rooftops due to the easier access. Furthermore, there might be possibilities to adopt transmit diversity for the DVB-H networks. It would improve the reception and QoS in site cell edges and outage areas within the site cell. The technique provides thus robust reception with improved QoS and can reduce the network costs by lowering the transmit power and the number of infrastructure elements as has been identified in [Zha08, p. 575]. As a conclusion of the cost optimization part of this thesis, a sufficiently detailed method was developed to be taken into account in typical DVB-H deployment. The other identified references presented the related aspects either from purely economical point of view [Bal07] [Sat07], or taking into account only partially the relationship between cost and technical items [Bmc07] [Sat06] [Ski06]. Instead, this thesis presents a complete set of CAPEX and OPEX items which were investigated as a function of transmitter antenna height and power levels, taking into account largely practical aspects of the equipment and planning environment.

156

157 121 6 Summary and conclusions 6.1 Main results of this thesis As a basis for this thesis, a radio network planning and optimisation process chart was created for the nominal as well as for the detailed phase of the planning. The process modules were designed based of the publicly available information, but enhancing the processes. Based on these process charts, the methods of the most relevant parts of the processes were identified and investigated. The outcome clarifies the DVB-H link budget usage. The results include the optimisation of the essential network parameter values taking into account the related network building and operation costs, as well as the regulatory limitations of the radiation powers. As for the second topic, an enhanced methodology for the field measurements and analysis with hand-held DVB-H terminal was created. The outcome of this part was the introduction of the procedure for the measuring and post-processing of the field test data. Selected cases were carried out, and the respective results can be used in order to fine-tune the radio link budget in similar locations the tests were performed. The closest similar analysis can be found in [Amp06a]. This thesis clarifies the way to present and interpret MPE-FEC gain, and gives a more detailed means for the estimation of the accuracy of the results. Thirdly, a simulation method for the SFN investigations was developed and respective case studies were carried out by utilizing a set of practical parameter values. The method is based on the physical radio propagation analysis assuming that the upper layer functionality is ideal. The respective carrier-to-noise and carrier-to-interference ratios are assumed to be fixed for the used parameters. The method is thus valid for obtaining information about the limits of the network performance and is thus sufficiently accurate to select the optimal parameter values, e.g. as a function of transmitter antenna height and power level. The initial version of the simulator is based on the hexagonal site cell layout, common antenna heights and power levels, but it was also shown that the tool can be utilized for more practical environments with individual site configurations. One of the outcomes of the simulations was the confirmation that the large (over-sized) SFN network can be applied when the essential parameters are tuned correctly like the height of the antennas and radiating power levels of the transmitters. The used analysis method is slightly different than the identified references because the comparative simulations are proposed to be done over the whole investigated area instead of individual sub-regions. The proposed method gives sufficiently accurate results for the comparison of different parameter settings, and the simulations require considerably less time than the sub-region method. Nevertheless, the results

158 122 correlate with the outcome of the available reference material. Furthermore, the results show acceptable quality when the balance between the SFN gain and SFN interference levels are taken into account. The performance of the network can be further optimized by using directional antennas, down-tilting and case-by-case adjustment of the antenna directions by utilizing the topology of the surrounding areas. At the same time, the simulator shows the overall cumulative distributions of C/N and C/(N+I) over the investigated area in numeric format as well as in visual format within the geographical area. The simulator also shows the level of the SFN gain in a non-interfered DVB-H network. Although [Bmc09] recommends that the SFN gain should be set to zero in the radio link budget assumption, the results of this thesis indicate that the SFN gain can be taken into account in a sufficiently overlapping site cell layout. As one of the outcomes of the SFN gain related simulations, the definition of the SFN gain has been clarified. The method takes into account the combination of the gain and interference within the whole investigated network, not only noninterfered environment and in limited locations as the identified other publications typically show. 6.2 Usability of the results The results of this thesis can be used for detailed DVB-H network planning and optimisation which benefits primarily the network operators. The main focus of the thesis was to develop methodologies that can be applied in arbitrary parameter value settings depending on the area and radio channel type. The numerical values are meant for examples, but as such they indicate that the adjustment of the DVB-H radio link budget accordingly is one of the most important tasks in the technical radio network planning. The results clearly show the importance of the selection of cost-effective assumptions in the network deployment, which is essential in the designing of the business models with efficient return of investment schemes. By applying the presented methodologies for the measurements and simulations, it is possible to adjust the parameter values in an optimal way in each environment of the commercial networks. 6.3 Further study items The transmitter diversity in DVB-H can increase the network performance with several decibels as has been concluded in [Zha06]. It would be interesting to take into account the balance of SFN gain and SFN interference level in over-dimensioned SFN networks that uses transmitter diversity. The SFN simulator presented in Publications III, and IV could thus be enhanced in order to obtain the optimal parameter settings, transmitter antenna height and transmitter power level.

159 123 The quality criterion of DVB-H has been studied in [Him09]. Even this thesis uses the typical error criteria presented in [Dvb09], the frame error rate before and after MPE-FEC being the most important, there might be parallel criteria that reflect the service quality better as stated in [Him06]. Based on the experiences during the field measurement with the methodology presented in Publication II, it seems that the BER before and after Viterbi are not reliable criteria especially in the edge of the coverage area as the algorithm for calculating the error rate in terminal or measurement equipment side is inaccurate due to the lack of information of correct and erroneous data. Furthermore, based on the observations of Publication II it seems justified to claim that frame error rate gives practical indication about the quality level that the end-user can observe. This is due to the fact that if the frame is erroneous, it is seen immediately in the terminal side as non-fluent streaming of audio and/or video. The MPE-FEC receiver is implementation dependent and decides what the streamer should do when erroneous frame is detected. In some cases, according to [Him06], it might be more beneficial to simply reject the erroneous frame and stream it in any case as the end-user might be less disturbed about the small occurred error than losing the whole frame. The practical limit of acceptable level of passing these erroneous frames is an interesting further study item in order to optimize the quality of the service in the terminal side. Other items for potential further study that are not much reported are listed below: The effect of hierarchical modulation on DVB-H. The hierarchical modulation has been designed for the DVB-T and might thus be considered as a relatively complicated item in DVB-H environment with varying radio conditions, but a further study could bring deeper understanding about any benefits hierarchical modulation could bring in certain network areas. Advanced measurement methodologies that include MER, i.e. vector error mapping to the quality level that the end-user experiences. The frequency response information could first be used in the measurements by identifying the number and characteristics of different components. This information could further be used for the MER / vector error or other information in order to get the perceived error effect. The clarified method could explain the degraded MER performance of SFN gain analysis that is presented in [Bov09]. A further techno-economic study by taking into account a realistic map (either realistic site distribution or hexagonal). The map could include a candidate site rings. The analysis could be carried out by switching on and off different sites according to the functional site combinations, and tune the complete set of parameters of each site separately. Different radio propagation models can be used individually for each site depending on the surrounding environment, cluster type, topography etc. By iterating,

160 124 and taking into account the radiation limitations (regulatory limits), and cost optimisation, it would be possible to search for parameter combinations that comply with the quality criteria and minimize costs over the investigated time period. As a continuum, a complete techno-economical optimization tool could be created that integrates an iterative CAPEX / OPEX analysis based on the technical parameter selection and SFN gain/interference simulator. More detailed level simulations in link level by applying the same physical level ideas as presented in this thesis. The idea would be to create complete DVB-H frames with varying error correction schemes. The frames could be created by utilizing real A/V contents that is encoded, or samples of the real frames. When the frames are distributed over the radio interface, a bit error rate can be applied depending on the radio conditions, i.e. fading profile, signal level and the presence of interferences. The error rate could be applied on bit-by-bit basis, and study the error correction capability in the receiving end. Furthermore, more realistic fading schemes could be applied in the simulator by taking into account the frequency selectivity in the Rayleigh channel, and by modelling real measured fading data from the investigated area type.

161 125 References [Apa06a] Maite Aparicio (editor). Wing TV. Services to Wireless, Integrated, Nomadic, GPRS, UMTS & TV handheld terminals. D6: Common field trials report. November p. [Apa06b] Maite Aparicio (editor). Wing TV. Services to Wireless, Integrated, Nomadic, GPRS, UMTS & TV handheld terminals. Wing Tv Country field trial report. November p. [Apt06] APT Recommendation on Guidelines for the Frequency Coordination for the Terrestrial Services at the Border Areas Between Administrations. No. APT/AWF/2. Edition: Asia-Pacific Telecommunity. The APT Wireless Forum February Approved By The 30th Session of the APT Management Committee September 2006 Maldives. [Aur09] Interview of Tommi Auranen, Nokia, Salo, Finland [Avo06] Peter MacAvock. The DVB-H Experience. 3GSM 06 Barcelona. 15 th February p. [Bac04] Wenche Backman. Signal Level Interpolation for Coverage Area Prediction /04. IEEE Pp [Bal07] [Bee02] [Bee07] [Bee98] [Bmc07] Gian Paolo Balboni. How Advertising Revenues Can Increase Profitability of DVB- H Services. Presentation, Deploying and Managing DVB-H, Gruppo Telecom Italia. London, September 27, p. J. J. van de Beek, P Ödling, S. K. Wilson, P. O. Börjesson. Orthogonal Frequency- Division Multiplexing (OFDM). International Union of Radio Science (URSI) Karina Beeke. Spectrum Planning Analysis of methods for the summation of Log-normal distributions. EBU technical review October p. Jan-Jaap van de Beek. Synchronization and Channel Estimation in OFDM Systems. PhD thesis, Lulea University of Technology, Division of Signal Processing, Lulea, Sweden p. Claus Sattler. Mobile Broadcast Business Models - A State of the Art Study. BMCOFORUM, November p. [Bmc07b] Spectrum Position for Mobile TV. BMCOFORUM, March p.

162 126 [Bmc07c] Mobile Broadcast Bearer Technologies A Comparison. BMCOFORUM, January p. [Bmc09] Link Budgets. Update 02/2009. BMCOFORUM, February p. [Bou06] [Bou08] [Bov09] [Bro02] [Bro08] [Cha06] [Che85] [Chu00] [Cos99] [Dig05] [Dvb04] Thibault Bouttevin (editor). D8 - Wing-TV Measurement Guidelines & Criteria. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS, UMTS & TV handheld terminals. June p. Imed Bouazizi, Miska M. Hannuksela, Lukasz Kondrad, Moncef Gabbouj. Efficient FEC Protection of Scalable Media Streams in DVB-H. BTS08, Alexandria, USA. 6 p. Jose Bóveda, Gorka Marcos, Jesús Maria Pérez, Sara Ponce, Ane Aranaz. MER degradation in a Broadcast Mobile Network. IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (Broadband Multimedia 2009). Bilbao, Spain, May, p. P. G. Brown, K. Tsioumparakis, M. Jordan, A. Chong. UK Planning Model for Digital Terrestrial Television Coverage. Esearch & Development. White Paper WHP 048. September p. Petter Brodal, Henning Gundersen, Gunn Kristin Klungsøyr. DVB-H technology overview. R&I N 24/2008. R&I Research Note. ISSN p. Joo Chan Kim, Jin Young Kim. Single Frequency Network Design of DVB-H (Digital Video Broadcasting Handheld) System. ICACT2006. February 20-22, Pp David R. Cheriton, Stephen E. Dewing. Host groups: A Multicast Extension for Datagram Internetworks. Report No. STAN-CS (also numbered CSL ). Department of Computer Science, Stanford University. July p. Arto Chubukjian, Hughes Nappert, Kirit Mehta, Andrk Legris. An example of efficient spectrum management: Army tactical radio operations in broadcasting bands. Industry Canada, Ottawa, Ontario, Canada IEEE, Pp Digital Mobile Radio Towards Future Generation Systems. COST 231 Final Report. COST/EC, Brussels, Belgium, p. Television on a handheld receiver broadcasting with DVB-H. Digital Terrestrial Television Action Group. Version 1.2, Grand-Saconnex, Geneva, Switzerland, p. Digital Video Broadcasting (DVB); Transmission System for Handheld Terminals (DVB-H). ETSI EN , V1.1.1 ( ). 14 p.

163 127 [Dvb06] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union, March p. [Dvb06b] Digital Video Broadcasting (DVB); IP Datacast over DVB-H: Architecture. ETSI TR V1.1.1 ( ). 28 p. [Dvb06c] Digital Video Broadcasting (DVB); IP Datacast over DVB-H: Use Cases and Services. ETSI TR V1.1.1 ( ). 29 p. [Dvb07] IP Datacast over DVB-H: Implementation Guidelines for Mobility. DVB Document A117 (BlueBook), July p. [Dvb07b] Digital Video Broadcasting (DVB); IP Datacast over DVB-H: Set of Specifications for Phase 1. ETSI TS V1.1.1 ( ). 8 p. [Dvb08] [Dvb09] [Ebu05] [Ebu09] [Ecc04] [Fan06a] Broadcasting to Handhelds. DVB Fact Sheet, August Produced by the DVB Project Office. 2 p. Digital Video Broadcasting (DVB); DVB-H Implementation Guidelines. ETSI TR V1.3.1 ( ). 114 p. BPN 066, issue 1.0, guide of SFN Frequency Planning and Network Implementation with regard to T-DAB and DVB-T, EBU technical department 2005, 124 p. General issues to be considered when planning SFNs. EBU Technical Media Technology & Innovation. 13 March p. ECC Report 49. Technical Criteria of Digital Video Broadcasting Terrestrial (DVB- T) and Terrestrial Digital Audio Broadcasting (T-DAB) Allotment Planning. Electronic Communications Committee (ECC) within the European Conference of Postal and Telecommunications Administrations (CEPT) Copenhagen, April p. Myron D. Fanton. Analysis and Measurement of RF System Reflections and DTV Transmission. Electronics Research, Inc. ERI Technical Series, Vol. 6, April p. [Fan06b] Myron D. Fanton. Analysis of Antenna Beam-tilt and Broadcast Coverage. Electronics Research, Inc. ERI Technical Series, Vol. 6, April p. [Far01] Gerard Faria. Mobile DVB-T using antenna diversity receivers. IBC Amsterdam, September p. [Far04] Gerard Faria. DVB-H to deliver Digital TV to Hand-Held Terminals. IBC September 2004, Amsterdam. 12 p.

164 128 [Far05] [Far05b] Gerard Faria. DVB-H: Digital TV in the hands. IBC 05, Amsterdam, September p. Gerard Faria. The handheld world outside DVB-H. Presentation material, DVB World 2005, 4 th March 2005, Dublin. 27 p. [Far05c] Gerard Faria. Validation Task Force Field Trials. Embedded Systems Conference 3G Cellular System Design Seminar. March p. [Far06] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. Proceedings of the IEEE, Vol. 94, No. 1, January Pp [Far07] Gerard Faria. From Digital TV to Mobile TV. CSTB 07 Media on the Move DVB tutorial. 5 th February p. [Fcc07] [Fcc98] Federal Communications Commission. Report and order and further notice of proposed rulemaking (FCC 07-72). Adopted: April 25, Released: April 27, p. Federal Communications Commission, Office of Engineering & Technology. Washington, D.C Information on Human Exposure to Radiofrequency Fields from Cellular and PCS Radio Transmitters. January [Fis08] R. Fuschini, H. El-Sallabi, V. Degli-Esposti, L. Vuokko, D. Guiducci, P. Vainikainen. Analysis of multipath propagation in urban environment through multidimensional measurements and advanced ray tracing simulation. IEEE Transactions on Antennas and Propagation, Vol. 56, no. 3, Pp [Fis08] [Gar07] [Goe02] [Gom07] [Goz08] Walter Fischer. Digital Video and Audio Broadcasting Technology. A Parctical Engineering Guide. ISBN , Spriner-Verlag Berlin Heidelberg. 580 p. G. Gardikis, H. Kokkinis, G. Kormentzas. Evaluation of the DVB-H Data Link Layer. European Wireless April 2007, Paris, France. 6 p. Roland Götz. Supporting Network Planning Tools II. Presentation, LS telcom AG, p. David Gómez-Baquero, Aurelian Bria. Forward Error Correction for File Delivery in DVB-H. Polytechnic University of Valencia, Spain, and Royal Institute of Technology, Sweden. Vehicular Technology Conference VTC 2007, Spring. Dublin, April p. David Gozálvez, David Gómez-Barquero, Narcis Cardona. Performance Evaluation of the MPE-iFEC Sliding RS Encoding for DVB-H Streaming Services.PIMRC p.

165 129 [Gre06] [Had07] [Hat80] Emmanuel Grenier. DVB-H radio planning aspects in ICS telecom. White paper, ATDI. July p. Kamel Haddad. DVB-H in Denmark. Technical and Economic aspects. Master s Thesis. Technical University of Denmark (DTU), Center for Information & Communication Technologies (CICT). September 10, p. Masaharu Hata. Empirical Formula for Propagation Loss in Land Mobile Radio Services. IEEE Transactions on Vehicular Technology, Vol. VT-29, No. 3, August p. [Hen05] Jukka Henriksson. DVB-H standard, principles and services. HUT seminar T , presentation. Helsinki, p. [Him06] [Him09] Heidi Himmanen, Ali Hazmi, Jarkko Paavola. Comparison of DVB-H link layer FEC decoding strategies in a mobile fading channel. The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06) /06. IEEE p. Heidi Himmanen. On Transmission System Design for Wireless Broadcasting. Academic dissertation. University of Turku Department of Information Technology Turku, Finland, p. [Hum09] The limits to terrestrial television s case for further spectrum. Human Capital. February p. [Ili08] [Itu07] [Jeo01] [Jok05] [Jok06] Teodor Iliev, Dimitar Radev, Izabella Lokshina, Georgi Hristov. Analysis and Evaluation of Reed Solomon Codes in Digital Video Broadcasting Systems. Wireless Telecommunications Symposium WTS April Pp Recommendation ITU-R P Method for point-to-area predictions for terrestrial services in the frequency range 30 MHz to 3000 MHz p. Minseok Jeong, Bomson Lee. Comparison between Path-Loss Prediction Models for Wireless Telecommunication System Design /01. IEEE Pp Heidi Joki, Jarkko Paavola, Valery Ipatov. Analysis of Reed-Solomon Coding Combined with Cyclic Redundancy Check in DVB-H link layer. ISWCS X/05. IEEE Pp Tero Jokela, Jarkko Paavola, Heidi Himmanen,Valery Ipatov. Performance of Different Reed-Solomon Erasure Decoding Strategies at the DVB-H Link Layer. The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 06) /06. IEEE p.

166 130 [Jos07] [Kru05] [Law01] [Lee86] Lig00] [Lig99a] [Lig99b] [Lig99c] [Lun06] Wout Joseph, Emmeric Tanghe, Daan Pareit, Luc Martens. Building penetration measurements for indoor coverage prediction of DVB-H systems /07. IEEE Pp Stefan Krueger. DVB-H Pilot Berlin. Presentation. T-Systems International GmbH Media&Broadcast, Broadcast Network & Services. September p. Eric Phillip Lawrey Be. Adaptive Techniques for Multiuser OFDM. Doctoral thesis. School of Engineering, James Cook University, p. William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp Agnes Ligeti. Outage Probability in the Presence of Correlated Lognormal Useful and Interfering Components. IEEE Communications Letter, Vol. 4, No. 1, January Pp Agnes Ligeti, Jens Zander. Minimal Cost Coverage Planning for Single Frequency Networks. IEEE Transactions on Broadcasting, Vol. 45, No. 1, March Pp Agnes Ligeti. Coverage Probability Estimation in Single Frequency Networks in Presence of Correlated Useful and Interfering Components. VTC Pp Agnes Ligeti. Single Frequency Network Planning. PhD thesis. Royal Institute of Technology p. Janne Lundberg. A Wireless Multicast Delivery Architecture for Mobile Terminals. Dissertation for the degree of Doctor of Science in Technology. Helsinki University of Technology (Espoo, Finland) on the 15th of May, p. [Mäk05] Juri Mäki. Finnish Mobile TV Pilot Results. August 30th, Research International Finland. 13 p. [Mar05] [Mil06] [Min99] Peter Marshall. Digital Television Project. Advanced Receiver Techniques with emphasis on Portable TV Reception. Issue 1.1. DTG Management Services Ltd, March p. Davide Milanesio (editor). Wing TV. Services to Wireless, Integrated, Nomadic, GPRS, UMTS & TV handheld terminals. D11 Wing TV Network Issues. May p. Limits of Human Exposure to Radiofrequency Electromagnetic Fields in the Frequency Range from 3 khz to 399 GHz. Safety Code 6. Environmental Health Directorate, Health Protection Branch. Publication 99-EHD-237. Minister of Public Works and Government Services, Canada ISBN p.

167 131 [Nok05] Technical data sheet of Nokia N-92. Nokia, p. [Öst06] [Paa06] [Pal08] [Pek05] [Pen09] [Pen99] [Pen99b] [Pir99] [Pit09] [Ple08] [Ple09] [Ple09b] Erik Östlin, Hajime Suzuki, Hans-Jürgen Zepernick. Evaluation of the Propagation Model Recommendation ITU-R P-1546 for Mobile Services in Rural Australia. 63rd IEEE Vehicular Technology Conference, Melbourne, Australia, May Paavola, J. (Editor); Kornfeld, M.; Rinne, J.; Björkqvist, J. Simulation Report. Deliverable D15, August 2006, EU CELTIC Project Wing TV. 67 p. Gema Roig Pallardó. On DVB-H Radio Frequency Planning: Adjustment of a Propagation Model Through Measurement Campaign Results. Master s Thesis, University of Gävle, March p. Stuart Pekowsky, Khaled Maalej. DVB-H architecture for mobile communications systems. RF Design, April Pp Jyrki T.J. Penttinen, Petri Jolma, Erkki Aaltonen, Jani Väre. The DVB-H Handbook. The Functioning and Planning of Digital Mobile TV. ISBN Wiley p. Jyrki Penttinen. The effects of the new data services on the GSM radio network planning. Licentiate thesis. Helsinki university of Technology Jyrki Penttinen. Effects of GPRS on GSM Radio Network. International Conference on Telecommunications (ICT). Acapulco, Mexico, p. Riku Pirhonen, Tapio Rautava, Jyrki Penttinen. TDMA Convergence for Packet Data Services. IEEE Personal Communications. June pp Renaud-Alexandre Pitaval. Relay Deployment in Single Frequency Network. Master's thesis. Helsinki University of Technology, Faculty of Electronics, Communications and Automation, Department of Signal Processing and Acoustics, p. David Plets, Leen Verloock, Wout Joseph, Luc Martens, Etienne Deventer, Hugo Gauderis. New Method to Determine the SFN Gain of a DVB-H Network with Multiple Transmitters. 58th Annual IEEE Broadcast Symposium October Alexandria, VA, USA. 4 p. David Plets, Leen Verloock, Wout Joseph, Luc Martens. Influence of DVB-H Network Deployment on Total RF Exposure. Proceedings, 20th Int. Zurich Symposium on EMC, Zurich Pp David Plets, Leen Verloock, Wout Joseph, Luc Martens, Etienne Deventer, Hugo Gauderis. Weighing the Benefits and Drawbacks of an SFN by Comparing Gain and Interference caused by SFN Operation. IEEE International Symposium on

168 132 Broadband Multimedia Systems and Broadcasting (Broadband Multimedia 2009). Bilbao, Spain, May, p. [Pos05] [Reg02] [Rei05] [Rei06] [San05] [Sat06] [Sat07] [Sci07] [Sil06] Renzo Posega. Advanced OFDM Systems for Terrestrial Multimedia Links. Doctoral thesis No. 3220/2005. École polytechnique fédérale de Lausanne. 194 p. Decree on the limitation of exposure of the public to non-ionizing radiation 294/2002, based on the law on radiation protection 592/1991, 43. Ulrich Reimers. DVB The Family of International Standards for Digital Video Broadcasting. Second Edition. ISBN X. Springler-Verlag Berlin Heiderberg p. Ulrich Reimers. DVB-H and IP Datacast: Technologies for Broadcasting to Hendheld Devices. Beijing, 22 November p. Finnish Mobile TV: Analysis on log file data. April June Press event p. Claus Sattler. Mobile Broadcast Business Models: A State of the Art Study. BMCOFORUM, November p. Claus Sattler. Mobile Broadcasting: Trends and Challenges. LS Summit. 20. June 2007, Lichtenau. BMCOFORUM. 23 p. Grzegorz Scibior. DVB-H Network Engineering Link Budget & Human Exposure to EM fields prelus 1.0. March Siemens Networks. Airi Silvennoinen. DVB-H Network Optimization under Finnish Conditions (DVB- H lähetysverkon optimointi Suomen olosuhteissa). Written in Finnish. Helsinki University of Technology, Department of Electrical and Communications Engineering. Master s Thesis. Espoo, Finland, p. [Ski06] Daniel Skiöld. An economic analysis of DAB & DVB-H. EBU Technical Review January p. [Tal10] Interviews of Pekka H. K. Talmola, Finland, March [Tun05] [Ung06] Celal Alp Tunc, Ayhan Altintas, Vakur B. Ertürk. Examination of Existent Propagation Models Over Large Inhomogeneous Terrain Profiles Using Fast Integral Equation Solution. IEEE Transactions on Antennas and Propagation, Vol. 53, No. 9, September Pp Peter Unger, Thomas Kürner. Radio Network Planning of DVB-H/UMTS Hybrid Mobile Communication Networks. Institut für Nachrichtentechnik (IfN), Technische Universität Braunschweig, Germany. European Transaction on Telecommunication, special issue ETT 17-2, March pp

169 133 [Ung08] [Voj05] [Wan03] [Zha06] Peter Unger, Thomas Kuerner. Optimizing the Local Service Areas in Single Frequency Networks. COST 2100 TD(08)467. Wroclaw, Poland, 2008/Febr/ p. Ondrej Vojtko, Florian Makan. Radio Signal in WLAN Networks. Journal of Electrical Engineering, Vol. 56, No , Pp Xianbin Wang, Yiyan Wu, Jean-Yves Chouinard, Sili Lu, and Bernard Caron. A Channel Characterization Technique Using Frequency Domain Pilot Time Domain Correlation Method for DVB-T Systems. IEEE Transactions on Consumer Electronics, Vol. 49, N0. 4, November Pp Yue Zhang, John Cosmas, Maurice Bard, Yong-Hua Song. Diversity Gain for DVB-H by Using Transmitter/Receiver Cyclic Delay Diversity. IEEE Transactions on Broadcasting, Vol. 52, No. 4, December Pp [Zha08] Yue Zhang, C. Zhang, J. Cosmas, K. K. Loo, T. Owens, R. D. Di Bari, Y. Lostanlen, M. Bard. Analysis of DVB-H Network Coverage With the Application of Transmit Diversity. IEEE Transactions on Broadcasting, Vol. 54, No. 3. September Pp [Zir00] [Zyr98] Single Frequency Network Concepts for cellular OFDM Radio Systems. Wolfgang Zirwas. Siemens AG, München. OFDM workshop, Hamburg, Germany, Sept p. Jim Zyren, Al Petrick. Tutorial on Basic Link Budget Analysis. Application note AN9804.1, Intersil, June p. Internet sources Reference Link Contents, date of link validity revision [web01] Official web site of DVB-H project office, including complete list of DVB-H networks that are launched or in trial stage [web02] Web page of DVB-H project office, containing complete list of DVB-H networks that are launched or in trial stage [web03] Wikipedia pages for "Margin of error" [web04] Broadcast Mobile Convergence Forum, [web05] USA regulations about the radiation safety limits

170 134

171 135 Errata 1 Publication IV, Figure 11, is missing the SFN reuse pattern size information for K=1 to K=9. The information is shown correctly in Publication X, Figure There is a wrong measured MPE-FEC 1/2 gain value presented in Publication II, Table II for the case of 16-QAM, FFT 4K, CR 2/3. The value should be 3.2 instead of 3.7. There is no impact of this inaccuracy in the respective analysis carried out in the Publication as the measured values have relatively large variations in any case. 3 An error was found from the simulator code after the related publications had been presented (Publication III, IV, VII and X). The error was in the Okumura-Hata formula of the medium and small city type as shown in bold letter in the Pascal code Function lu2 below. (* Okumura-Hata propagation prediction, medium and small city---*) Function lu2(fhb, fhm, ff, fd : real) : real; var oha, ohb, ohc, ohd, ohe : real; begin oha:=69.55; ohb:=26.16*log(ff); ohc:=13.82*log(fhb); ohd:=(1.1*log(ff)-0.7)*fhm-(1.56*log(ff-0.8)); ohe:=( *log(fhb))*log(fd); lu2:=oha+ohb-ohc-ohd+ohe; end; (* lu2 *) The correct form of the term ohd should be the following, as presented in [Hat80]: ohd:=(1.1*log(ff)-0.7)*fhm-(1.56*log(ff)-0.8);

172 136 An analysis shows that the direct effect of the faulty formula is db for the L, when 700 MHz frequency is used in small and medium city area. A snapshot investigation was made in order to evaluate the effect of the wrong formula. Five consecutive simulations were carried out with the erroneous formula, and other five simulations with the corrected version. The parameter values were selected in such a way that they are close to the previous simulation assumptions. An area of 100 km 100 km was created with base station antenna height of 100 m, MS height of 1.5 m, MS antenna gain of -7 dbi, frequency of 700 MHz, standard deviation of 5.5 db and TX power of +70 dbm. Rayleigh fading was not selected but large-scale fading was on. The area location probability was 90% in a small and medium city type with bandwidth of 8 MHz, modulation of QPSK, code rate of 1/2, MPE-FEC rate of 2/3, guard interval of 1/4 and FFT mode of 8K. The cumulative C/I distribution was investigated in such a way that the five simulation results were averaged per C/I (db) value for the correct version and for the erroneous version. By observing the mapping between the probability level and C/I, the following correction Tables can be formed to present the most important values. Table E-1. The C/I as function of the probability scale. Probability C/I, correct C/I, faulty C/I, difference, when correct is reference 5% 9.5 db 9.9 db +0.4 db 10% 13.5 db 14.7 db +1.2 db 50% 22.7 db 23.3 db +0.6 db Table E-2. The C/I as function of the probability scale. C/I Probability, correct Probability, faulty Probability, difference, correct as reference % 4.1 % 0.2 % units % 9.7 % 2.3 % units % 21.6 % 3.9 % units As can be noted, the investigated probability per C/I value, i.e. the minimum limit for certain C/I, has been slightly optimistic. This does not have major impact on the previously presented results, though. Publication IV takes into account the balance of the SFN gain due to the overlapping areas as well as for the loss of the gain due to the SFN interference. Slight difference in the mapping of absolute C/I values to the probability figures affects in both ways, so the relative balance of the gain and interference can be assumed to be practically unchanged. The impact of the error is not significant on the results of the Publication III either.

173 137 4 The fast (Rayleigh) fading is present in those environments where multi-path radio signals occurs, e.g. on the street level of cities. The Rayleigh fading modelling was not complete in Publication VII. It was aimed to be presented with the following PDF: L Rayleigh x = 2 δ e 2 x ( ) 2 2δ Figure 6-1 shows the PDF and CDF of the fast fading representing the variations of short-term loss when the standard deviation is set to 5.5 db. Fast fading was used only in Publication VII together with long-term fading as the environment was a large but dense urban city. It should be noted that in the simulations carried out in this thesis, the frequency selectivity of the wide band OFDM signal was not taken into account. In the correct analysis, the fast fading would result different C/N values for different sub-carriers within a single OFDM bandwidth. PDF PDF and CDF of log-normal distribution, stdev=5.5 PDF CDF db CDF Figure 6-1. PDF and CDF of the Rayleigh distribution. Nevertheless, the contribution of Publication VII is to present the functionality of the overall SFN interference simulation method, and the inclusion of the Rayleigh fading model was aimed to merely adjust the radio propagation channel to be more suitable for the dense urban environment. Based on the practical experiences of the field tests [Tal10], the modelling could contain only long-term fading model also in the urban environment. The Rayleigh fading module was thus removed from the simulator. An additional simulation shows that the geographical distribution of the interferences remains practically the same as presented in Publication VII when only long-term fading model with 5.5 db standard deviation value was included, and the overall C/I level rises as can be expected.

174 138 5 An error was found in the power sum formula of the simulator. The original version suggested that the multi-propagated components of the DVB-H radio signal was possible to estimate in SFN by summing the components in squared form, i.e., by presenting the useful received signal as shown in Publication III, Formula (1). Equally, in order to estimate the C/(N+I) level, the original simulator was based on the assumption that the total received interfering power would be calculated in a squared form as shown in Publication III, Formula (2). Nevertheless, the squared form is utilized in the summing of individual amplitudes of the signal, not for the power levels. The useful carrier signals between the multi-propagated components, as well as the interfering signals between each individual interferer, can be assumed to be noncoherent. A direct power summing should be applied for the received power calculation as indicated in [Ebu05]. This means that the useful total received power (W) is: tot C P = n i= 1 P C i (6-1) Equally, the total received power (W) of the interfering components is: P tot I = n i= 1 P I i (6-2) When comparing the squared and direct form, it can be seen that the error for the estimate of the received power level is highest when two or more signal components are received at the same level. As an example, a practical worst case can be assumed to occur in the overlapping area of three sites where the signal level is the same. Assuming the received power level is 90 dbm for a parameter set of {QPSK, CR=1/2, MPE-FEC=1/2, GI=1/4} in such a location, the received total power of these three signals would result the total power values by applying the squared and direct power summing as presented in Table 6-1. The received power level in dbm is converted to absolute powers (W) for the summing, and the result is converted back to dbm form by applying the formula P RX [dbm]=10log 10 (P RX [W]/1mW). Table 6-1. Comparison of the squared and direct power summing. P C ( dbm) P 1 C ( dbm) P 2 C ( dbm) 3 PC tot (dbm) Signal 1 Signal 2 Signal 3 Squared form PC tot (dbm) Direct form Error (Direct form squared form)

175 139 In order to estimate the effect of the error in the simulations presented in the publications of this thesis, comparative simulations were carried out. First comparative simulation was carried out for the analysis presented in Publication VII, i.e., the effect of the SFN parameter values for the useful coverage area in a large urban area. A parameter set was selected in such a way that both useful and interfering signals were present about equally, i.e., {16-QAM, CR=2/3, MPE-FEC=2/3, GI=1/8, FFT=4K}. Otherwise all the other settings, including the site locations and antenna heights, were the same as presented in Publication VII. Figure 6-2 summarizes the PDF analysis, and Figure 6-3 presents the CDF analysis of the C(N+I) distribution Pow er sum comparison, PDF C, direct sum I, direct sum C, squared sum I, squared sum # of samples C/(N+I), db Figure 6-2. The PDF of the comparative analysis of the power summing in squared and direct manner Power sum comparation, CDF C/(N+I), direct sum C/(N+I), squared sum C/(N+I), db Figure 6-3. The CDF of the power summing analysis.

176 140 As Figure 6-2 and Figure 6-3 show, the final effect of the power summing in a squared and direct way produces very similar results in this specific case. The selected parameter set means that the minimum C/(N+I) should be at 17.5 db. According to the simulations, the squared summing results an area coverage probability of 60.0 % whereas the direct summing results 60.8 %. The difference is thus 0.8 %-units which can be noted as insignificant in this analysis. The most important reason for the small difference between the power levels summing in this very case is that the received useful and interfering signals compensated largely each others. This case also represents the most practical manner to carry out the analysis as the investigated area is not limited to the expected site cell coverage areas. Due to the relatively high proportion of the coverage outages in the area (45 km 45 km) with only 7 sites, this analysis gives thus the most optimistic value (lower limit) to the power summing error. It can be thus concluded that the error does not affect on the examples of Publication VII. As a second part of the power summing error analysis, the simulation was carried out as presented in Publication IV for a selected set of the most relevant parameters. The upper limit for the power summing error can be found via the non-interfering cases when no compensation of the SFN interferences can be achieved. The worst case scenario for the error margin was thus selected by utilizing the most robust parameters for QPSK and 16-QAM. The SFN reuse pattern size was varied between 1 and 21. The other parameters were {CR=1/2, MPE-FEC=1/2, GI=1/4, FFT=8K}. Figure 6-4 summarises the comparative simulations. The results indicate that the SFN gain, as well as the absolute error value for the SFN gain estimate grows up to the SFN reuse pattern size of 19. After that, the gain saturates to approximately 6.0 db for QPSK and 6.3 db for 16- QAM when the correct version of the direct power summing is applied. The original, noncorrect squared power summing results the saturated SFN gain at 4.2 db for both QPSK and 16- QAM. This means that the upper limit for the occurred error of the SFN gain estimate is approximately 2 db. For the cases with 4 sites, i.e., when the SFN size is K=4, the SFN gain estimate error is 1.1 db and 1.3 db for QPSK and 16-QAM, respectively. The correct values of the SFN gain are presented in all related analysis of the summary of this thesis. Furthermore, the simulation of the effect of the SFN interference in a large DVB-H network (Publication III), where SFN limits are exceeded, was carried out for {QPSK, CR=1/2, MPE-FEC=1/2, GI=1/4} by varying the FFT size between 2K, 4K and 8K. The correct simulation result is shown in Figure 5-1. The outcome of Publication III is the estimate of the antenna height that provides a useful DVB- H coverage in a large, over-dimensioned SFN which contains interferences. Publication III concluded that the {QPSK, FFT=8K, GI=1/4} and {16-QAM, FFT=8K, GI=1/4} modes can be utilized with all the antenna heights of m when 10 % outage criterion is applied. The

177 141 correct power summing gave the same outcome for this, although the curves are different. Publication III also concluded that for the {QPSK, FFT=4K, GI=1/4} the maximum useful antenna height was noted to be 35 m, and for the {16-QAM, FFT=4K, GI=1/4} the value was 30 m. Based on the new simulations, these values are about 60 m and 45 m, respectively, as Figure 5-1 indicates Squared sum, QPSK Squared sum, 16-QAM Direct sum, QPSK Direct sum, 16-QAM SFN gain (db) K=1 K=3 K=4 K=7 K=12 K=16 K=19 K=21 Figure 6-4. The comparative simulation results for the SFN gain in non-interfered environment. The formulas of Publications that are wrongly presented are: Publication Formula Error level and actions III 1 Wrong carrier power sum. The correct formula is shown in (6-1) 2 Wrong interference signal sum. The correct formula is shown in (6-2) IX 5 Wrong carrier power sum. The correct formula is shown in (6-1) 6 Wrong carrier power sum. The correct formula is shown in (6-1) 7 Wrong interference signal sum. The correct formula is shown in (6-2) The figures of Publications III, IV, VII and X that are affected are listed in the following table.

178 142 Publication Figures Corrections III The C/I level of figures is shown in a pessimistic way due to the squared power summing. Figure 16 of Publication III summarizes these by showing the outage as a function of the antenna height. These simulations were repeated completely, and the correct version of Figure 16 is presented in Figure 5-1 in the optimisation chapter. IV 5 11 Examples of the simulation functionality. These figures are not utilized in the analysis and correction is thus not needed Simulation results for the SFN gain of QPSK and 16-QAM. These simulations were repeated by utilizing the correct power summing. The correct results are shown in Figure 5-2 and Figure 5-3. VII 2 20 Visual presentation of the interference distribution for different cases. The error analysis presented above shows that the wrong power sum method did not affect on these figures because the amount of the sites was low (7), and the investigated area contained a large outage portion. X 11, 14 18, Examples with minor impacts, not utilized in the analysis Simulation results for the SFN gain of QPSK and 16-QAM as presented in Publication IV. These simulations were repeated by utilizing the correct power summing. The correct results are shown in Figure 5-2 and Figure Case examples with a minor impact to the absolute values. The power sum method did not affect in this analysis as it is based on Publication VII. In addition, the comparison is done only in a relative manner.

179 A 1 Appendix A SFN Simulator A.1 The principle In order to identify the effects of the SFN mode of DVB-H on the network planning, simulation software was designed and written. Also a set of simulation cases was selected and carried out for the studies of the SFN gain and interference levels. The simulations are described in Publications III, IV, VII and X. Initial phase Start Parameter value input h bs, h ms, G ant, L ant, stdev of the distribution, location probability in cell edge, area size (x/y coordinates or reuse pattern size K), simulation area (inside the calculated cells or in the whole rectangular area), area type, bandwidth, modulation, code rate, MPE-FEC rate, GI, FFT. Table creation PDF and CDF of normal and log-normal cases. Initiate (clean) C and I tables. Simulation phase Simulation rounds Increase the round count and place the mobile station on map. Calculate the C/I levels and store to text file row together with location. Update the C/N, I/N and C/(N+I) tables. No 60,001 rounds? Yes Finalization phase Data storing Save the C/I PDF and CDF results ( db range, 0.1 db resolution). End Figure 6-5. The high-level simulator block diagram.

180 A 2 The simulator was coded with standard Pascal language. The basic functionality of the simulator does not require complicated procedures, which means that the code can be produced with major computer languages as long as the simulation results can be stored to one and two dimensional tables and produced as text files for the post processing purposes. The simulator represents thus methodology that provides high-level information about the relative parameter behaviour of DVB-H, and gives indication about the performance limits. More detailed modelling of different DVB-H protocol layers is presented in [Paa06]. The idea of the simulator is based on the collection of snap-shot values of the useful and interfering signal levels, the latter in case the theoretical SFN limits are exceeded. The DVB-H receiver is located randomly in the pre-defined area according to the Monte-Carlo method, which is either given as x and y coordinates of the corners of the area (in kilometres) or as a SFN reuse pattern size (by using reuse factor K, which fixes the number of the sites within the SFN area). In each simulation round, the calculation of the sum of carriers and interfering signals is done separately, and the results are stored both individually as well as in cumulative format for the post-processing. For each simulation round, the new receiver location is determined by random x and y values. The radio channel type is estimated by using slow and/or fast fading profile by applying relevant mathematical format. Even if the simulator does not include dynamic continuum of the terminal path with respective velocities, the fading profiles gives sufficiently good estimate of the performance assuming the maximum Doppler limit is not exceeded. Figure 6-5 shows the high level block diagram of the simulator. The diagram consists of the initial and simulation phases as described in next Chapter. A.2 Initiation phase Figure 6-6 shows the block diagram of the simulator in the initiation phase. When the simulator code is executed, the initiation of the case begins. First, the user input is delivered either by typing manually the parameter values, or by fetching the pre-defined values from a text file. The requested parameters are the following: Height of the transmitter site antenna h bs in m, height of the DVB-H receiver h ms in m, antenna gain of the mobile G ant in dbi, which is handled as antenna loss L ant = G ant in the simulator calculations. The DVB-H Implementation Guidelines suggests that the antenna gain can be estimated as 10 dbi in 474 MHz frequency and as 5 dbi in 858 MHz frequency [Dvb09]. In the simulations, in can be assumed that the value of the antenna gain can thus be interpolated or extrapolated linearly as function of the frequency by applying the following formula:

181 A G ant = f 5 10 (A-1) 384MHz 384 Other requested initial parameters are frequency f in MHz, standard deviation stdev of long-term fading that should be estimated depending on the area type, and radiated isotropic transmitter power level in dbm. According to [Dvb09], the typical standard deviation value is 5.5 db in sub-urban environment. The value is lower in open areas, and respectively higher in multi-path rich environments like dense city centres. Begin Inputs: h bs, h ms, L ant, f, stdev, P tx, LND, ND, locprob, area (xkm, ykm, x1km, y1km, x2km, y2km) or reuse pattern size (K), areatype, bw, modulation, CR, MPEFEC, GI, mode. Form tables: Probability table of log-normal distribution, probability table of normal distribution. Calculate maximum allowed path loss: L max, L maxinterference. Calculate radius: r, r interference Create cell layout: 1) Area filled with cells, or 2) reuse patterns, or 3) case specific layout Show simulation assumptions: Cell number, TX-TX links for C and I, cell coordinates, Calculate parameter values: Delay, D SFN, noisefloor, cnmin. Assign and open log files. Simulation Figure 6-6. Simulator s initiation phase. The standard deviation and area location probability has an important inter-dependency. When estimating the coverage areas, it is essential to utilize realistic values for the standard deviation. If the value for the standard deviation is overestimated, it results in too pessimistic predictions and the radio network is more expensive than it needs to be. On the contrary, underestimating the standard deviation value produces too optimistic plans and thus too few sites which will result in more outages than estimated [Bee07]. The standard deviation is estimated for a single transmitter site cell. Nevertheless, in a realistic network, the transmitter site cells are overlapping. In SFN case, the reception of two or more signals from different transmitter site cells increases the area location probability both in the edge and over the whole site cell. The simulations can use large scale (slow) fading and fast (Rayleigh) fading individually or as a combination as shown in Figure 6-7. The simulator does not include though the AWGN profile which is basically representing the situation in pure LOS situation. Nevertheless, as the code is modular, it would be straightforward to implement additional fading profiles in a mathematical format. Equally, it would be possible to replace the formulas and include a real fading example

182 A 4 as a file that is based on the field measurement results. This would be a possible further development item in order to tune the simulator into real environments. In the simulator, the location probability at site cell edge is given in percentages, the realistic functional scale being %. The mapping of area location probability over the whole site cell and in the site cell edge appears after the simulation has been carried out, by selecting the filtered option of the simulation, i.e. limiting the results to the cases that have the mobile terminal only inside the pre-calculated site cell. Long-term fading Rayleigh fading Figure 6-7. Principle of the combination of fast and long-term fading in the receiving end. Depending on the simulator version, the investigated area is given either in x and y coordinates o the area corners (km), or in SFN reuse pattern size in K value (which refers to the number of the sites within SFN area). The simulator gives possibility to investigate the performance either inside the calculated site cell areas (applying the above mentioned values), or in the whole x/y map. In the former case, the terminal locations that occur outside of the site cells are rejected. The simulator uses Okumura-Hata path loss model for the coverage prediction. For this reason, the simulator requests the area type that can be large city, small and medium city, sub-urban or open area. The bandwidth is given in MHz (5, 6, 7 or 8 MHz). In the developed version of the simulator, the modulation can be QPSK or 16-QAM, code rate 1/2 or 2/3, MPE-FEC rate 1/2 or 2/3, guard interval 1/4, 1/8, 1/16 or 1/32, and FFT mode 2K, 4K or 8K. When all the parameter values have been defined, the initiation procedure begins. First, the probability tables are formed for the fading. L norm represents the fading loss caused by the long-term variations, and other losses may include the fast fading as well as antenna losses. For the long-term fading, a normal distribution is commonly used in order for modelling the variations of the signal level. The PDF of the long-term fading is the following: ( x x) 2 1 PDF( Llong _ term ) = exp (A-2) 2 2πσ 2σ The term x represents the loss value, and x is the average loss (0 in this case) in db. In the snap-shot based simulations, the L long_term is calculated for each arriving signal individually as

183 A 5 the different events do not correlate. The respective PDF and CDF are obtained by creating a probability table for normal distributions. Figure 6-8 shows an example of the PDF and CDF of normal distributed loss variations when the mean value is 0 and standard deviation σ is 5.5 db PDF and CDF of normal distribution, stdev=5.5 PDF CDF db Figure 6-8. PDF and CDF of the normal distribution representing the variations of long-term loss when the standard deviation is set to 5.5 db. Large scale fading is formed depending on the standard deviation value that was given as input. First, a PDF is formed in the following way presented in Pascal language: for count:=-30 to 30 do begin PDFnorm[count]:=(1/((sqrt(2* ))*stdev)) * exp(-(abs(count*count))/(2*stdev*stdev)); end; After this, the CDF table is created in the following way: (* Absolute values of CDF (which may not result exactly 1 in the tail): *) for count:=-30 to 30 do temp2[count]:=0.0; for count:=-30 to 30 do begin for count2:=-30 to count do begin temp2[count]:=temp2[count]+pdfnorm[count2]; end; (* for count2 *) end; (* for count *) (* Normalised values of CDF (the last value of CDF table is set to 1): *) for count:=-30 to 30 do CDFnorm[count]:=temp2[count]/temp2[30]; Now, the task is to convert the CDF table into an inversed version, i.e. instead of the function of the attenuation in scale of 30 to +30 db, the table should be presented as function of probability with scale of 0 to 1. The conversion can be done in the following way:

184 A 6 CDFnormprob[0]:=-30; for count:=1 to 100 do begin count2:=0; repeat count2:=count2+1; until ( (CDFnorm[count2] >= count/100) or (count2>=30) ); end; value_hi:=cdfnorm[count2]; value_lo:=cdfnorm[count2-1]; para:=value_hi-value_lo; parb:=(count/100.0)-value_lo; CDFnormprob[count]:=(count2-1)+(parB/parA); The resolution of the probability table is of 1 %. It is now straightforward to give a long-term snap-shot attenuation value in each simulation round by utilizing a random value of 0 to 100. Even the simulator does not give correlation for the location between the simulation rounds the overall channel type approaches to the mathematical model over the whole area as the number snap-shots grows. The next step is to assign the SFN limits based on the GI, i.e. the maximum radio signal propagation delay (us), the FFT mode which together defines the maximum diameter of the SFN area (the maximum distance of the extreme sites) in km. The simulator then assigns the noise floor depending on the band width. The simulator takes into account the noise figure of the typical terminal, in this case 5 db as stated in the DVB-H Implementation Guidelines [Dvb09]. The result is thus dbm for 5 MHz band, dbm for 6 MHz, for 7 MHz and dbm for 8 MHz band. The modulation scheme and code rate give requirement for the minimum functional C/N value. The simulator assigns 8.5 db for QPSK CR 1/2, 11.5 db for QPSK CR 2/3, 14.5 db for 16- QAM CR 1/2 and 17.5 db for 16-QAM CR 2/3. These values are derived from [Dvb09], and they are valid for TU3 channels with Rayleigh fading profile. For the other radio channel types, as presented in [Bou06, p. 27], the values can be added in the code and the rest of the parameter combinations can be coded directly to the source, but the respective simulations were not carried out in the studies of this thesis. Next, the theoretical maximum site cell size r for the useful signal (carrier C) as well as for the interference I site cell size r interference is calculated according to the respective Okumura-Hata model. The carrier distance is calculated based on the required carrier level compared to the noise floor (which in this case includes already the terminal s own noise figure). The calculation of the radius is done in iterative way, i.e. by growing the radius value from the initial one (0.01 km) with increments of 0.01 km, calculating the respective path loss L with the Okumura-Hata

185 A 7 using the respective area type correction, until the maximum allowed path loss L is achieved. The maximum path loss is achieved when the following formula is fulfilled: L = Ptx noisefloor Lant (cn) min (A-3) The term P tx is the transmitter s radiating power (EIRP) in dbm, noisefloor is the noise floor in dbm that takes into account the actual noise limit and receiver s noise figure of 5 db, L ant is the receiver s antenna gain (attenuation) in dbi and (cn) min is the minimum C/N requirement for the respective parameter set in db. The result of this calculation is the maximum allowed path loss for the carrier L max and interference L maxinterference. Next, there exist two variations of the area forming. The first variation is based on the filling the defined rectangular area with hexagonal site cells and the second is based on the SFN reuse patterns. In the first version, the simulator calculates the expected radius of single site cell in noninterfering case and fills the area with uniform site cells according to the hexagonal model. This provides partial overlapping of the site cells. Each simulation round provides information if that specific connection is useless, e.g. if the criteria set of 1) effective distance, i.e. the difference of the arriving signals D eff > D sfn (maximum allowed distance within SFN area) in any of the site cells, and 2) C/(N+I) < minimum C/N threshold. If both criteria are valid, and if the C/N would have been sufficiently high without the interference in that specific round, the SFN interference level is calculated. 100 TX locations km km Figure 6-9. Example of the transmitter site locations the simulator has generated. Figure 6-9 shows an example of the site locations. As can be seen, the simulator calculates the optimal site cell radius according to the parameter settings and locates the transmitters on map

186 A 8 according to the expected site cell radius, leaving ideal overlapping areas in the site cell border areas based on the hexagonal model. The size and thus the number of the site cells in the investigated area depend on the radio parameter settings without interferences in the whole investigated area. The same network setup is used throughout the complete simulation, and it is changed accordingly for the next simulations if the radio parameters of the following simulation require so. The second version of the presented SFN performance simulator is based on the hexagonal site cell layout as given in [Lee86]. Figure 6-10 presents the basic idea of the cell numbering. TX(1,y) TX(2,y) TX(x,y) TX(1,2) TX(2,2) TX(x,2) d y r TX(1,1) TX(2,1) TX(x,1) r d x Figure The active transmitter sites are selected from the 2-dimensional site cell matrix with the individual numbers of the sites. As can be seen from Figure 6-10, the site cells are located again in the map in such a way that they create ideal overlapping areas. The tightly located hexagonal site cells fill completely the circle-shape site cells. A uniform parameter set is used in each site cell, including the transmitter power level and antenna height, which results the same radius for each site cell per simulation case. For the calculation of the site cell locations, the simulator uses the principles shown in Figure 6-11 and Figure y r x Figure The x and y coordinates for the calculation of the site locations.

187 A 9 y cos (30 deg) = (x/2) / r x = 2r cos (30 deg) = 2*0.866r x x/2 30deg r 30deg 30deg x/2 tan (30 deg) = (x/2) / y y = x/2tan (30 deg) y = r cos(30 deg)/tan(30 deg) y=1.5r Figure The geometrical characteristics of the hexagonal model used in the simulator. As the relative location of the site cells is fixed, the coordinates of each site cell depends on the uniform site cell size, i.e. on the radius. Taking into account the characteristics of the hexagonal model, the x-coordinates can be obtained in the following way depending if the row for y coordinates is odd or even. The distance between two sites in x-axis is: ( 30 ) = 2 0. r x = 2r cos 866 (A-4) The common inter-site distance in y-axis is: ( 30 ) 3 ( 30 ) 2 r cos r y = = (A-5) tan For the odd rows the formula for the x-coordinate of the site m is thus the following: ( m 1) 1. r x( m) odd = r (A-6) In the formula, m represents the number of the site cell in x-axis. In the same manner, the formula for x-coordinates can be created in the following way: 1 x( m) even = r r + ( m 1) r (A-7) 2 For the y-coordinates, the formula is the following: y 3 = (A-8) 2 ( n) r + ( n 1) r

188 A 10 The simulations can be carried out for different site cell layouts. The symmetrical reuse pattern concept was selected for the simulations of N sites =K sites presented in Publication IV. The reuse pattern size refers in this case to the SFN size, i.e. the site cells utilize the same frequency within the reuse patter size (unlike the original use of the SFN reuse patterns, e.g. in GSM frequency planning). Different SFN areas can thus be repeated by utilizing this pattern and by applying different frequency for each SFN pattern, i.e. by utilising MFN for different SFN areas. The term SFN reuse pattern is utilized to distinguish the idea from the original reuse pattern concept. Figure 6-13 clarifies the SFN reuse pattern concept. Figure The idea of the SFN reuse pattern. In this case, K=9. The colours represents different frequencies, each forming a single SFN area with 9 sites. The most meaningful SFN reuse pattern size K can be obtained with the following formula by applying the TDMA reuse pattern concept [Lee86]: 2 K = ( k l) kl (A-9) The variables k and l are positive integers with minimum value of 0. In the simulations, the SFN reuse pattern sizes of 1, 3, 4, 7, 9, 12, 16, 19 and 21 was used for the C / (N + I) distribution in order to obtain the carrier and interference distribution in both non-interfering and interfering networks (i.e. the presence of SER depends on the size of the SFN area). In this way, the lower values of K provides with the non-interfering SFN network until a limit that depends on the GI and FFT size parameters. As a last step of the initial phase, the text files are named and opened for the storing of the results.

189 A 11 A.3 Simulation phase Figure 6-14 shows the block diagram of the simulator in the simulation phase. Once the initialization of the simulator is done, the simulation rounds will be started. The simulations are repeated a total of 60,001 times by placing the DVB-H receiver in the investigated geographical area, calculating the total level of carriers and interferences in that spot. The value 60,001 is a result of the applied loop which is repeated from integer value of 30,000 to +30,000. The first step is to initiate the random number generator. In this case, this is done by executing the only non-standard Pascal command of the simulator, i.e. Randomize. This provides the means to repeat the simulations with the same parameter values but with different locations of the receiver. As the round number of each complete simulation is sufficiently high, the final placement of the receiver is close to the uniform distribution over the whole area in x and y coordinates. Next, all the relevant tables are cleaned by giving zero-values for all the table indexes. This guarantees that the values are stored and cumulated correctly even in the case of possible residual values in the tables in the execution of the simulation. Then, the DVB-H receiver is placed in the field by applying the random function. As the exact location in x/y coordinates of the sites is known from the initial phase of the code execution, as well as the height of the transmitter antennas (which is uniform in the basic version of the simulator, and can be given separately for each site in the 3 rd version), the distance between the receiver and each one of the site antenna can be calculated in 3-dimensional space. These distance objects are stored in respective tables during each simulation round. Furthermore, the respective path loss L can be calculated by applying the Okumura-Hata models that are coded in respective functions as sub programs. Furthermore, the long term and/or fast fading is added to this path loss value by using the respective cumulative presentations of the probability density functions created in the initial phase. As the simulator is snap-shot type, this is done separately for each location, i.e. in each simulation round. It should be noted, that as the terminal can be located anywhere in the x/y-coordinates, the transmitter antenna height results the minimum distance with the terminal as shown in Figure If the terminal is relatively close to the site, the respective carrier could be higher than the upper scale limit of the C table (+50.0 db). In such cases, the value is added to the table index C[500] representing the db value, or if that already contains 30,000 values (which is highly theoretical and not probable), the index C[499] representing the db value is increased respectively. This is due to the fact that the utilised Pascal compiler s maximum integer value is limited to 32,000 unless possible longer integer values would be available and applied. This procedure assures that the cumulative presentation can be calculated correctly over

190 A 12 the whole area. Another solution would be to limit the minimum distance in x/y-coordinates between the terminal and site sufficiently in order to limit the C value, but in this investigation it is not necessary. It should also be noted that the transmitter and receiver antennas are considered isotropic. Initialization Initiate random generator. Initiate the log files: Write headers with the parameter values. Place the mobile randomly on the map. Calculate the 3-D distance and path loss (taking into account fast and long term fading and antenna gain) between the mobile and all the TXs. Find nearest TX to the terminal as a reference. Indicate if TXs are producing C or I (comparing each one with the reference TX). Calculate the db level of C and I for each TX. Convert each C and I value from db to absolute power levels (W) and sum the total C and total I. Convert the C and I back to db values. Increase round counter by 1. Write terminal s x/y coordinates, C and I to file number 1. Add C or I to cumulative tables. no Are 60,001 rounds complete? Save cumulative results to log file number 2. Finalization yes Figure The simulation phase.

191 A 13 z y z 1 D MS z 2 x 1 x 2 y 1 y2 x Figure The terminal is placed in the map varying the x and y coordinates, and the distance from the site antenna is calculated in 3D space. The idea of the simulator is based on snap-shot analysis, i.e. the terminal is located on the map during each round without continuum from the previous round. If the distance between terminal and investigated site is less than the theoretical SFN limit shows, the signal is considered as useful carrier C, otherwise it is considered as interfering signal I. The effective distance D eff is used to decide weather the SFN limit is exceeded per site or if the terminal is within the SFN limits, i.e. it expresses the difference between the physical 3D distances of the arriving signals of the sites that are compared. I 1 C 3 I 2 C 4 C 2 C 1 C 5 Effective SFN limit Figure The principle of the snap-shot of each simulation round. The nearest site is identified and handled as a reference carrier C 1. The comparison of the other sites, i.e. the effective distance that is the difference between the nearest site s signal and the other one, is calculated. When the distance is within the maximum allowed SFN distance, i.e. when D eff D sfn, the respective carrier signal is taken into account if its level is more than the noise level, i.e. the sum of the noise floor and terminal noise figure. Equally, for the D eff > D sfn, the interfering signal is taken into account only if it is greater than the sum of noise floor and terminal noise figure, in order to simplify the measurement procedure. The value for D sfn is based on the initial values of FFT size and GI which together dictates the maximum allowed

192 A 14 distance between the sites in non-interfering SFN. Figure 6-16 shows the basic principle of this idea. The distances between the terminal and each site are calculated separately as well as the respective signal level of the investigated site. If the effective distance with the reference site is inside SFN, the signal component is considered as carrier, and otherwise it is considered as interference. Now, all the relevant components of the carrier and interfering signals are summed in order to obtain the total level of C and I. This is done by converting the signal levels into absolute power levels and carrying out the summing, assuming that the signals are always non-coherent: n C tot = C i i= 1 n I tot = I i i= 1 (A-10) (A-11) In the formula, n represents the number of identified carriers, and m is the number of the interfering sources, per simulation round. After the summing, the value is again converted to the original presentation of received power level in dbm. The total C and I of the simulation round is now stored into a text file with the respective information about the coordinates. It is also added into a cumulative C and I table, which in this case is an indexed table with index values of 500 to If the C/N value is 12.5 db, the value of the table index C[125] is increased by one. In this format, the C scale can thus be obtained directly by observing the indexes, and the cumulative amount of the values per index is obtained by observing the respective value of the index. As can be noted, the raster of the indexed format is 0.1 db. In cases when the C or I value is outside of the indexed range, the respective value is added to the extreme indexes ( 500 or 500 presenting 50 db and +50 db and 499 or 499 presenting 49.9 db) in order to maintain the statistical accuracy of the cumulative presentation. When all the 60,001 simulation rounds have been executed, the text files are closed and the simulation ends. A.4 Data analysis The simulation model is based on the snap-shot principle, i.e. Monte-Carlo method. The receiver is thus located on the field randomly, and the received total carrier power, as well as the

193 A 15 possible received total interfering power, is calculated by taking into account the selected fading profile. The snap-shots are repeated until sufficient amount of samples has been collected. The main outcome is the cumulative function of the occurred carrier levels within the simulated area, as well as interfering power levels. In this format, the percentage of the useful coverage area can be investigated directly by observing the dimensioned area location values. The results are stored in a cumulative table which shows the CDF as such. Other option that is typically utilized in the coverage prediction of the planning tools like NetAct, would be to divide the area into relatively small sub-areas, e.g. 100 m 100 m sized squares. The simulations can be repeated within each square in order to observe if the level of the carrier is above the minimum functional carrier threshold for more or equal percentage of the snap-shots compared to de designed area location probability value. If the level is high enough, the square is selected as a part of the functional coverage area. This method is suitable for the visual presentation of the coverage area, but for the calculation of the CDF of the C/(N+I) distribution, the method presented in this thesis is obviously more accurate as it takes into account each simulated point in the area. In any case, the simulator stores all the locations in x, y coordinates with the respective C and I values, so the post-processing for the square format is possible. This was not done during the thesis though, as it would not give added value for the C/(N+I) analysis if the visual aspect is not taken into account. As a result of the simulation, there are two files produced in plain text format, with the values separated by semi-commas. One file contains the entire C, I and location information of each simulation round, and the other one contains the C/N and I/N distribution in 50.0 db db scale. These tables can be post-processed in order to obtain the geographical distribution of the C and I levels (as shown in Publication VII), and the CDF and PDF of C/N, I/N and C/(N+I) curves as shown in Publications III, IV and, X. The post-processing of the results was done by using MICROSOFT EXCEL. The respective text files can be imported to the EXCEL sheet by selecting the ; as a separator. The PDF of C and I distribution over the complete scale gives information about the presence and level of the carriers and interferences. If the whole area is within SFN, there are no I- components present whereas the extension of the SFN over its theoretical limits starts producing I components. The effect of parameter values like site antenna height and transmitter power levels on the interference levels can be thus investigated by repeating the complete simulations and varying the parameter values respectively. Publication III shows the antenna height and power level effects on the SFN interference level. In the second version of the simulator, the network layout can be designed by using SFN reuse patterns. Furthermore, the analysis can be limited into the coverage area of the used site cells. This gives possibility to compare the C/N and C/(N+I) distributions with different number of the used site cells. When multiple site cells

194 A 16 are compared with a single site cell, the effect of SFN gain can be investigated as shown in Publication IV. As an example about the interpretation of the results, the first log-file contains an individual simulation round s x, y, total C and total I values in a single row. It is thus possible to make an additional analysis that plots the x and y coordinates when certain C, I, C/N and C/(N+I) criteria is achieved. This gives means of producing coverage maps that takes into account the balance of SFN gain and SFN interference as is presented in Publication IV. The second log-file contains the C, I and C/(N+I) distribution in db scale of db. It is possible to present the C, I and C/(N+I) occurrences, e.g. in relative percentages over the whole scale as shown in Publication III. It is also possible to post-process the results in cumulative format, which gives the typical S-curves over the investigated area. In this way, it is possible to interpret the area outage or area location probabilities as function of the C/(N+I) as shown in Publication III. The idea is to identify from the simulation results the cumulative C/(N+I) value that complies with the minimum percentage of the occurred events. This gives directly the area location probability of the whole investigated area. If the area is not limited to the pre-calculated site cell areas, the percentage shows the area location probability over the whole x/y-area. Otherwise, the results reflect the realistic situation of only site cells (including the overlapping proportion) included. As for the interpretation of the results, Table 6-2 shows an example of the simulation that has been done by limiting the occurred events to the pre-calculated site cell area. This example refers to the second version of the simulator that uses the SFN reuse pattern sizes as a basis for creating the site cell areas. The related results are presented in deeper level in Publication IV by utilizing a more complete set of parameter values. The reference site cell (K=1) represents the situation without SFN gain as only one radio path is present. When the simulation is done for larger SFN reuse pattern sizes, more radio paths are present. The total amount of carrier and interference components are summed, which makes it possible to obtain the SFN gain by comparing the results with wanted area location criteria. In this specific case, the radio parameters were: FFT=2K, GI=1/4, 16-QAM modulation, CR 1/2, MPE-FEC 1/2. The standard deviation value was selected to 5.5 and the location probability in the cell edge to 70 % which corresponds to 90 % area location probability. The height of the transmitter antenna was 60 meters, and the height of the terminal was 1.5 meters. The transmitter power level (EIRP) was +60 dbm. If we seek for the location area probability of 90% (which corresponds to 70% of area location probability in the site cell edge), it is now found as the 10% outage point of the results shown in Table 6-2.

195 A 17 Table 6-2. An example of the simulation results, C/(N+I) as function of the cumulative area location probability. A total of 7 complete simulations are presented, for the SFN reuse patterns K of 1, 3, 4, 7, 9, 12 and 16. C/(I+N). db K=1 K=3 K=4 K=7 K=9 K=12 K= When investigating the 10% area outage probability point, a clear tendency of SFN gain can be observed up to K=7 both in Table 6-2 and graphical presentation of it in Figure Compared to the 1 site cell case which produces C/(N+I) of 14.5 db 90 % of the locations, the area with

196 A 18 seven site cells produces C/(N+I) of 17.4 db with the same 10% outage criteria. The SFN gain can thus be interpreted directly by comparing these values, resulting db = 2.9 db. Then, as the number of sites grows, the SFN reuse pattern K=9 includes already part of the sites outside of the SFN area as the GI is not long enough for all the farthest sites. It can be noted, though, that even the interference level produced via SFN reuse pattern size of K=16 of this example is still below the level that can be obtained via the SFN gain, so the balance of SFN interferences and SFN gain is still positive. In fact, as presented in Publication IV, the SFN reuse pattern size of K=21 is still producing around 0 db gain in this specific case, which indicates that the theoretical maximum SFN area, that would be clean of interferences, can be exceeded in a controlled way by utilizing the above mentioned parameter values. 0,2 SFN performance Outage probability 0,18 0,16 0,14 0,12 0,1 0,08 0,06 0,04 0,02 0 K=1 K=3 K=4 K=7 K=9 K=12 K= , , , ,5 db 15 15, , ,5 18 Figure An amplified view to the outage probability of 10% (area location probability of 90 %) with respective C/(N+I) values for SFN reuse pattern size of K=1 16. A.5 Functional analysis The simulator locates the mobile terminal in the map according to the uniform distribution. If the location is not inside of any of the site cells, there is an option to discard those results. Nevertheless, all the occurred locations are saved to the log file. In both cases, the uniformity of the locations is important in order to have good cumulative presentation of the whole map. When the amount of the snap-shots is increased, also the accuracy of the results gets better.

197 A 19 A.5.1 Selection and reliability of the models Propagation models The major part of the simulations was based on the Okumura-Hata path loss prediction model [Hat80]. The model was selected for this thesis because it is relatively straightforward to implement in the simulator code as it is based on the formulas for each area type. The model is based on the practical measurements and mathematical modelling of the typical results. It fits to the simulations as in these cases there is no need to calculate the path loss in more detailed level, and its typical cell range is within the studied examples. The model can be assumed to provide sufficiently accurate results as it has been utilized widely as a basis in commercial network planning programs. Also ITU-R P model was utilized in one of the cases to estimate the site cell coverage area in a large area due to the high mountain location. This model was selected because Okumura-Hata is not valid in the site area in question due to the much greater effective antenna height than Okumura-Hata supports. The ITU-R P model is more complicated to implement in the simulator code as it is based on the curve mapping and interpolation and extrapolation of the curves. Nevertheless, as the case utilized only one value for the frequency and effective antenna height, it was possible to create a mathematical formula by forming a regression curve. The accuracy of the produced formula is high enough for this case as the error margin analysis of A.5.2 shows. Also the model itself can be assumed to be enough accurate as indicated in [Öst06], which concluded that the model (its different variations) enhance the accuracy of the traditional models. Simulator model The simulations are based on the Monte-Carlo. It was selected as a base for the work because it suits well in the evaluation of the radio parameter effects over the whole investigated area. It is possible to take into account the channel types by including the effect of the path loss variations in a realistic yet relatively simple way. The model provides the possibility to investigate the inter-site effects on the power level contribution for both useful and interfering signals is possible to take into account in a large area that contains considerable amount of transmitters. A typical solution of investigating the level of the coverage would be to divide the investigated area into small sub-regions, e.g. 100 m 100 m to 500 m 500 m squares. The cumulative distribution of the signal level of each region can be investigated. If the area location probability criterion for the minimum C/N or C/(N+I) is fulfilled, the region can be marked as functional. Nevertheless, the simulator of this work was designed in such a way that the simulations were executed over the whole investigated area once by repeating 60,001 snap-shots per investigated parameter set. As Chapter A.5.3 and A.5.4 indicate, the accuracy of this method is sufficient for

198 A 20 the format of the results presented in this thesis. This is due to the fact that the values are shown typically as cumulative density functions in this thesis, and the visual coverage areas are not utilized in the numerical analysis. As the results are accurate enough for the CDF presentation of the performance, the benefit of this method is much faster simulation time compared to the individual region simulations. As an example, each one of Figure 5-1, Figure 5-2 and Figure 5-3 required complete set of 60,001 simulation rounds due to the altering of the parameters (antenna height or SFN reuse pattern size), i.e. each plot of these figures represent a complete simulation. The simulation time varies in this type of cases 5 60 seconds per simulation in 100 km 100 km area, depending on the amount of the sites that are required to fulfil the whole area perfectly overlapping hexagonal area. Nevertheless, for this type of repetitions, the separate sub-area simulations would take considerably longer time and would not increase the accuracy of the results, according to the error margin analysis presented in Chapter A.5.4. The comparison of the results obtained via the utilized simulation model was not possible during the work as the cases that had been investigated by utilizing the same method were not found. Nevertheless, the results for the SFN gain are compared in this work with the identified references even if the methods have been different. A.5.2 Accuracy of the path loss prediction models The Okumura-Hata prediction model gives the path loss as function of the site cell radius. In the reverse calculations, i.e. when investigating what is the maximum site cell radius as function of the path loss, the simulator uses recursive format for maximum of 20 km distances. Publication VII presents a set of sites with the coverage calculated with Okumura-Hata except for one mountain site with the respective propagation loss calculated with ITU-R P.1546 model. The original model consists of pre-defined curves. For this special case with respective parameter settings, the ITU-curve was calculated and plotted in MICROSOFT EXCEL by utilizing the tabulated format of the original curves and by interpolating the correct curve as advised in the documentation of the model. In order to have a single formula for the simulation purposes, a regression curve was formed with EXCEL. The curve for the distances of 1 20 km was created by using the EXCEL functionality and the resulting formula is: y = ln( x) (A-12) For the distances of km, the linear presentation is close to the interpolated curve: y = x (A-13)

199 A Path Loss 1-20 km y = 10,659Ln(x) + 113, L (db) Distance (km) Figure The formed curve for the distances of 1 20 km of the mountain site is close to the logarithmic form, as used in Publication VII Path Loss 1-20 km FSL 170 L (db) Distance (km) Figure The interpolated curve for the distance of km can be formed linearly. The regression curves can now be obtained from. Figure 6-18 and Figure 6-19 show the values obtained by interpolating the tabulated values of the report ITU-R P , and the values that can be obtained from the above mentioned regression curves.

200 A Difference of trend line and original values Diff (regression - original) Distance (km) Figure The error margin of the estimated path loss values obtained via the original tabulated values of the model ITU-R P and the respective regression curves is within db up to 60 km of distance. There is also a peak up to +0.8 db in close distance which does not have significance in this case. Table 6-3. Comparison of the values obtained by interpolating the tabulated form and the regression curves. d L_860m_regr L_860m_orig Diff reg_orig

201 A 23 Now, an analysis for the difference of the values shown in Table 6-3 can be obtained as shown in Figure As can be noted from Figure 6-20, the difference between the estimated path loss value of the original tabulated values and the values that can be calculated with the regression curves is not significant in this type of simulations. A.5.3 Uniformity of the locations In order to evaluate the level of uniformity when the terminal is located on the map, the given area can be divided into smaller pieces in x and y coordinates. In a uniform 2D area, the distribution of the MS locations should occur approximately equally in each slice of these subareas. Let s select an example of km area with sufficient amount of sites. The parameters for this test are: base station antenna height 100 m, MS height 1.5 m, MS antenna gain 7 dbi, frequency 700 MHz, standard deviation 5.5 db, TX power +70 dbm, Rayleigh fading not selected, large-scale fading on, location probability 90 %, small and medium city type, bandwidth 8 MHz, modulation QPSK, code rate 1/2, MPE-FEC rate 2/3, guard interval 1/4, FFT mode 8K. This parameter set creates a hexagonal site cell layout with maximum path loss of db, site cell radius of 10.0 km (i.e. where the carrier is still above the minimum C/N of 8.5 db), maximum interference path loss db and corresponding interfering site cell radius of 18.5 km (i.e. distance where the interfering signal is still above the noise floor). The total number of transmitter sites in that area is 33, with carrier to interfering site cells proportion of 3.5 (taking into account all the sites even if the interfering signal would be less than noise floor, i.e. there are total of 410 links from which interfering links are 118). Let s carry out a complete simulation round and consider the distribution of the MS location by post-processing the location information. We can divide the area into 10 km slices both in x and y axis as shown in Figure 6-21 and investigate, what is the percentage of the occurred locations in each one of the squares. In the ideal case, each slice should contain 10 % of the total occurred events in x-axis, and 10 % per slice in y-axis. The total amount of simulation rounds and thus location results is 60,001 in this revision.

202 A Y-coordinate, distance / km X-coordinate, distance / km Figure The uniformity of the coordinate distribution can be analyzed by slicing the x-axis and y-axis of the investigated area in 10 sub-areas and revise the percentage of the occurred locations in each slice. As can be seen from Table 6-4 and Table 6-5, the standard deviation between the 10 slices per simulation round is typically about %. Figure 6-22 shows also in graphical format the distribution of the occurred locations for x and y coordinates. Each slice of the figures represents 10 % of the geographical area. Table 6-4. The percentage of the samples in each sliced area of x-axis. The ideal distribution would result 10 % of samples in each slice. The table presents a total of 5 complete simulations with 60,001 rounds each. Simul x-axis, row number StDev # Table 6-5. The percentile values as function of y-axis slices. A total of five simulation rounds are presented. Simul y-axis, row number StDev #

203 A 25 10,40 X-axis 10,50 Y-axis 10,30 10,40 10,20 10,30 10,10 10,00 9,90 10,20 10,10 10,00 9,90 9,80 9,80 9,70 9,70 9, X-axis slice 9, y-axis slice Figure The graphical presentation of occurred x- and y-coordinates in the sub-slices of the area. As a conclusion, the random function combined with the initial randomizer functionality provides sufficiently variable values as a basis for the simulator. Furthermore, by applying 60,001 simulation rounds, the geographical distribution of the DVB-H terminal over the whole simulation area is sufficiently uniform in both x and y axis, the variation between extreme values (% of the occurred coordinate value per 1/10 th size slices) being less than ±0.4 percentage points in both x and y axis. A.5.4 Accuracy of the results Based on five consecutive simulations with the above mentioned parameter values and 60,001 simulation rounds, the comparison of the results shows small variations between these different curves when the area location probability values are observed per RSSI value. The cumulated C/I values per db-class behaves in constant way regardless of the different random schemes which is an indication that the used number of the simulation rounds per complete simulation is sufficiently high. As the resolution of the C/I scale is of 0.1 db and the whole scale of the simulator s table is db, there is a total of 1001 elements in the table. Assuming that all the table elements would be in active use for storing the C and I values, the average count of the events per table element is 60,001 / 1001 = 60. This is statistically sufficiently high value especially because the most populated and thus significant area of the S-curve occurs around 0 30 db range. The accuracy can be investigated by lowering the amount of simulation rounds until the cumulative C and I distribution starts varying from the full scale version. Table 6-6 shows the standard deviation of the C/I distribution s db values with different number of simulation round values when 5 consecutive simulation rounds are compared with each others. The starting case was 60,001 simulation rounds per simulation, which was divided to half for each of the consecutive simulations. For each case (simulation round number) a total of five simulations

204 A 26 was carried out. The respective maximum and minimum values was investigated in 8.5 db point in order to see the variations of the results in probability % scale. The respective results are shown in Table 6-6. Secondly, the C/(N+I) value variation in 5% point of probability scale was observed to see the effect in db values. The results are presented in Table 6-7. Table 6-6. The analysis of the probability variations when C/I = 8.5 db reference point is observed in each simulation. The Table shows the respective variations in probability values. Average Std Rounds per simulation Probability for C/I point (8.5 db), Min Probability for C/I point (8.5 db), Max Difference of probabilities, max-min, % unit 60, , , , , , Table 6-7. The analysis of the variations of the interpretation of C/I value when 5% probability point is observed. Average Std Rounds per simulation C/I for prob. point (5%), Min C/I for prob. point (5%), Max Difference of C/I, max-min, db 60, , , , , , Figure 6-23 shows the effect of the number of the simulation rounds on the accuracy of the results. When observing the cumulative distribution curve in the most typical C/I range, i.e. approximately in the minimum threshold area for the QPSK and 16-QAM cases, the effect can be seen clearly in visual format. It can be noted that 60,001 simulation rounds produces smooth and constant curves, i.e. their variance is sufficiently low in order to interpret the C/I and probability mapping in sufficiently accurate way.

205 A 27 0, rounds 0, rounds 0,09 0,09 0,08 0,08 0,07 0,07 0,06 0,06 0,05 0,05 0,04 0,04 0,03 0,03 0,02 0,02 7,5 8 8,5 9 9, , , , ,5 14 7,5 8 8,5 9 9, , , , ,5 14 C per I C per I 0, rounds 0, rounds 0,09 0,09 0,08 0,08 0,07 0,07 0,06 0,06 0,05 0,05 0,04 0,04 0,03 0,03 0,02 0,02 7,5 8 8,5 9 9, , , , ,5 14 7,5 8 8,5 9 9, , , , ,5 14 C per I C per I 0, rounds 0, rounds 0,09 0,09 0,08 0,08 0,07 0,07 0,06 0,06 0,05 0,05 0,04 0,04 0,03 0,03 0,02 0,02 7,5 8 8,5 9 9, , , , ,5 14 7,5 8 8,5 9 9, , , , ,5 14 C per I C per I Figure Comparison of the effect of the number of the simulation rounds per complete simulation. A total of 5 simulations are presented in each case. A part of the S-curve is shown around the 5 % FER point.

206 A 28 By considering Figure 6-23, and more specifically the 5 % criteria, Table 6-8 can be created to indicate the variance of the results of each case. Table 6-8. The accuracy analysis of the simulation cases. Rounds P TX (1) P TX (2) P TX (3) P TX (4) P TX (5) x Diff (min - max) 1, , , , , , Std The analysis can be still done more detailed by observing the cumulative histogram of the standard deviation values as function of the number of simulation rounds. In this analysis, the standard deviation of five consecutive simulations is observed for each db category, and the standard deviation values is collected as a histogram with the resolution of (i.e. with 0.1% resolution of the probability). 100,00 Cumulative Standard Deviation (resolution 0.001) 90,00 80,00 70,00 60,00 50,00 40, rounds rounds rounds 7500 rounds 3750 rounds 1875 rounds Figure Cumulative presentation of the standard deviation classes per different lengths of the simulation. As can be observed from Figure 6-23, the five simulation results of 60,001-round case provides a good correlation and thus accuracy, 75 % of the standard deviation values being maximum 0.1

207 A 29 %, and all the results being within 0.4 %. Also 30,000 and 15,000 simulation rounds would provide an accuracy of 0.5 % in probability scale whilst the 7,500 and 3,750 rounds provides about 0.9 % accuracy, and 1875 rounds less than 1.5 %. As a conclusion, 60,001 simulation rounds provide sufficient accuracy for the simulation results. It can be assumed that higher number does not increase the accuracy considerably in practical interpretations of the performance; Table 6-6 shows that the maximum error in interpretation of the db values in typical probability range is about 0.3 db. With typical CPU power the complete simulation with 60,001 rounds takes only few minutes in the most demanding cases presented in this thesis. Furthermore, the amount of data is still manageable with typical post processing software that might limit the maximum row number to about 65,000. It can be estimated that the reliability of the results is sufficiently high for the presented examples in the radio propagation level simulations. This claim is based on the assumption that the higher protocol layers are functioning ideally.

208

209 Appendix B Publications B.1 Radio network planning studies The main contribution of this thesis for the general radio network planning area is presented in Publications I, V and VI. Publication I contains the investigations of the balancing of the site parameters and network costs. The objective of this part was to create a method that can be used as an iterative element in the cost-efficient network planning process. Publication I also presents selected case examples how the optimal cost can be obtained in SFN area. This publication clarifies the topic because the CAPEX and OPEX optimization of the DVB-H networks analysis as function of the radio network planning parameters was not found in the presented way by the publication of the study. Publication V clarifies the DVB-H radio network coverage planning principles in a dense urban area with selected case examples. It shows the useful coverage areas when the parameters are selected in such a way that they do not create SFN interference in the investigated area. Based on this study, the importance of the radio propagation model adjustment is shown. It is also proved that the basic theoretical prediction models can be applied in the initial phase of the network planning with a sufficiently good correlation with the more advanced coverage tools. These results can be used as a basis for the investigation of the effects of the parameter value variations, i.e. when the sites start to cause interferences. Publication VII shows the related interference analysis via simulations. As the service area is the same, Publications V and VII can be used as a pair for a coverage and interference analysis in a realistic urban network. The studies presented in this thesis show the estimated interference behaviour. The radiation of the DVB-H transmitter site can be limiting factor, which should be taken into account in all phases of the radio network planning and site selection process. Publication VI presents the analysis of EMC and human exposure limits of DVB-H transmitter sites. The method and related case results might move the optimal power level that is derived from the radio link budget, and might result a need for modifications of the initially planned transmitter and antenna types. The general safety zone calculations of mobile and broadcast networks can be found in various sources, but these case studies were done especially for DVB-H in order to complete the radio network process.

210 B.2 Field measurement methodology In order to assure the required quality level of the DVB-H radio network, it is essential to carry out field measurements as a basis for the network fine-tuning. The field measurement related studies with new analysis method are presented in Publications II, VIII, IX and XI. Publication II, and its extended version in Publication IX, shows a method that can be applied for the fast revisions of the network performance with respective data collection, postprocessing and analysis. The case examples and their results of the Publications show the importance of the parameter adjustment as function of the radio channel type in outdoor environment. The presented method for the post-processing of the measurement data as well as for the respective analysis and case results are based on the already published data collection software of the DVB-H terminals. The post-processing method and the analysis gives additional value, e.g. for the WingTV field measurement documentation [Bou06], [Apa06a] and [Apa06b] which clarifies the planning process of the DVB-H radio network. This thesis also analyses the accuracy of the presented measurement method of MPE-FEC gain in more detailed level compared to the principles presented in [Dvb06] and [Dvb09]. Publication VIII continues from the previous studies and shows the field measurement method, post-processing and analysis with respective results in indoor and outdoor environments. In addition to the methodology, the results indicate the logical set of values for the error correction related parameters which extend the results obtained in [Apa06a] and [Apa06b]. B.3 Simulation methodology A DVB-H radio performance simulator was developed for carrying out analysis of the Single Frequency Network (SFN) aspects. The summary of the principle and functionality of the simulator is presented in Appendix A. The simulation based studies are used in Publications III, IV, VII and X. Publication III presents a principle that can be applied for the simulations of the interference levels in over-sized SFN area. The results show the effect of the antenna height and transmitter power level on the error rate. The simulation method can be used for the estimation of the severity of the errors, and it provides information about the optimal setting of the antenna heights and transmitter power levels that still fulfils the final radio reception quality requirement even if the theoretical SFN limits are exceeded. The simulator is based on the geographical area that is filled with uniformly distributed cells.

211 Publication IV shows the further development of the above mentioned simulator. It is based on the SFN reuse patterns of the transmitters that define the number of the transmitters in the SFN area. This method gives means for estimating the SFN gain as function of the number of the transmitter sites and their radio parameters. The SFN area can also be extended, and the simulation results show the effect of the increased SFN interference levels. The results can be used in order to balance the SFN gain and interference level, which is useful in the selection of the related radio parameter values in the radio network planning process. Publication VII presents a further development of the above mentioned simulation principles. It shows that the basic idea can be applied also in the practical environment by setting the site specific location, antenna height and transmitter power level. It also combines two propagation prediction models, i.e. Okumura-Hata for the basic sites and ITU-R P for large coverage sites. The simulations show the distribution of the carrier and interferences as function of the radio parameter values in a real urban environment. The simulation results can be used for the estimation of the relation between the realistic coverage obtained in Publication VII and the SFN interference levels that are caused by the radio parameter adjustment. Publication X is an extended version of Publication IV, and it compiles the presentations of the simulator in the Publications III, IV and VII.

212

213 I Publication I CAPEX and OPEX optimisation as function of DVB-H transmitter power This paper presents a method which can be used for the analysis of an optimal balance between DVB-H transmitter power levels and related site costs of the DVB-H network. A case calculation is carried out as an example. The Third International Conference on Digital Telecommunications ICDT 2008 June 29 July 5, Bucharest, Romania 2008 IEEE Computer Society Press Reprinted with permission.

214

215 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 87 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA CAPEX and OPEX Optimisation in Function of DVB-H Transmitter Power Jyrki T.J. Penttinen Member, IEEE jyrki.penttinen@nsn.com Abstract DVB-H network planning and optimisation are essential tasks for the network operators in order to provide adequate service quality. As DVB-H can be deployed by using Single Frequency Network (SFN), the assumption for building maximum coverage within the SFN area is to use as powerful transmitters and as high antenna locations as possible. Nevertheless, in addition to the technical tuning, the complete network optimisation should take into account also the initial and operating costs of the system. This paper describes DVB-H optimisation methodology which is based on the analysis of the total cost of the network in function of the transmitter power levels. 1. Introduction DVB-H capacity and coverage can be achieved by many different combinations of the parameter values, including the variation of the transmitter power levels and antenna heights. Taking into account the limitations of SFN interferences, the maximum coverage can be achieved technically by locating the transmitter antennas as high as possible and by using maximum transmitter power levels. Nevertheless, in detailed optimisation, also the cost of different solutions, i.e. parameter settings and site specific installations should be taken into account. As an example, more output power the transmitter provides, higher the equipment complexity and power consumption are, affecting on the operating expenses (OPEX) of the network. In optimal deployment of the network, it is thus essential to identify the most relevant technical parameters and investigate their impacts on the initial, i.e. capital expenses (CAPEX), and operating expenses. Even in case of relatively small DVB-H network deployment, the proper election of the transmitter types (power levels) might reduce considerably the final costs of the network. Cable, connector, power splitter and filter loss Transmitter power level TRX Antenna gain and height EIRP Cell radius Fig. 1. The DVB-H coverage area depends on the antenna height and transmitter power level when the radio parameters are the same. 2. Identifying the key parameters The DVB-H planning is done for both core and radio sub-systems. In each case, the proper capacity is dimensioned taking into account the short-term operation and preferably the prediction for the midterm evolution. The capacity of the DVB-H network is calculated by taking into account the modulation scheme (QPSK, 16- QAM or 64-QAM) and error correction scheme (Code Rate of 1/2, 2/3, 3/4, 5/6 or 7/8). As a difference with the DVB-T, there is an enhanced correction method, MPE-FEC (Multi Protocol Encapsulation, Forward Error Correction), defined in DVB-H, which provides additional protection against the effects of multi-path propagation and impulse noise in mobile environment. This parameter can be set with the values of 1/2, 2/3, 3/4 and 5/6. Furthermore, the final design of the network depends on the value for the guard interval, interleaver mode (2k, 4k, 8k), area location probability (typically between 70-95%), shadowing margin and possibly SFN gain. Depending on these settings, the balance between capacity and coverage can be found.

216 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 87 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA The most important initial network planning input is normally the required capacity in radio interface, which dictates the adequate transmitter power level. The main limitations depends on the general regulations for the RF radiation (including the EMC and human exposure limits) as well as on the practical issues since the cost of different types of transmitters is not linear in function of the power levels. Other significant factor is the antenna height, which does have an impact on the DVB-H coverage. As the CAPEX and OPEX are considered, there are various other aspects that affects on the final cost of the network. Also the amount of leased or own items, like transmission lines, transmitter sites etc. do have their impact. The cost depends mainly on the transmitter equipment complexity, transmission lines, transmitter sites, towers or roof-tops, antenna feeders and antennas. The detailed cost list might include the material that is needed for the installation of the equipment. In addition, the in-depth cost optimisation should take into account the installation services and other immaterial items like the cost of the planning, preparation of the site drawings, site acquisition, license fees, maintenance costs etc. 3. Description of the methodology In order to identify the initial optimal parameter values when minimising the CAPEX and OPEX of the DVB-H networks, a systematic methodology can be applied. The process contains the following high-level revision as a basis for the investment decision: Identify the main items that affect on CAPEX and OPEX. Estimate the cost for each item. Calculate the total CAPEX and (yearly) OPEX for single transmitter site. Calculate the single cell radius for each case, averaging the investigated area as a uniform type, or selecting a set of separate uniform area types (that are calculated individually), using adequate RF propagation model. Select sufficiently large service area and estimate by using e.g. hexagonal model, how many sites there should be obtained in each case in order to cover the area with the uniform quality of service level. Calculate the total CAPEX and OPEX for each case for comparison purposes. For the accurate estimation, it is important to select all the major items that has cost impact on the network, and as many minor details as is seen reasonable. In this approach, the transmitter power level is selected as the variable whilst the core network with source signals, capacity, bit rate per channel etc. can be assumed to be the same in each case. 4. CAPEX and OPEX estimation For the realistic DVB-H CAPEX estimation, the following items can be taken into account: Transmitters. Antenna systems (with antenna feeders, power splitters, jumpers and connectors). Other material, like feeder brackets, tools etc. Transmitter site acquisition and preparation work. Transmitter, antenna system and other material installation and commissioning work. For the OPEX, the longer term items include at least the following: Transmission (leased lines, satellite transmission etc.). Maintenance of the transmitters, other equipment and site. Tower and/or site rent. Electricity consumption. The transmission has a key role in the OPEX per site. Depending on the needed capacity, the technical solution can be done by using e.g. sufficient amount of E1/T1 lines, or implementing fibre optics that provides sufficiently wide bit pipe. For the remote areas with relatively large proportion of sites difficult to access, a satellite transmission might provide with optimal solution. The basic cost of the satellite transmission is normally clearly higher than in terrestrial cable solution, but the single satellite link usually covers all the needed sites. For the electricity consumption, a rough estimation of 6 times the output power level can be used in this analysis unless the practical values are available. For the other items, the costs depends on the equipment and service provider list prices, taking into account the possible volume and other discounts. Furthermore, the prices can be estimated separately for the main transmitter sites and gap-fillers. In order to simplify the calculation, it is sufficient to take into account the number of main transmitter sites, and possibly estimating a lump cost for the gap-filler sites. The number and transmitting powers of gapfillers depends on the wanted level of the outdoor and indoor coverage areas. The CAPEX and OPEX include both common costs

217 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 87 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA as well as the costs that depend on the type of the transmitter site. The common CAPEX estimation includes the site acquisition and structural analysis whereas the transmitter, antenna system, power splitter, brackets and installation work depends on each site type. The common OPEX items includes the transmission lease (assuming the same bit pipe is delivered to each site), and average tower or roof-top site rent. The other OPEX items depend on the site type, and include e.g. electricity consumption and maintenance work which both depends on the transmitter type. For the CAPEX, the transmitter cost plays a key role. As the power level of the equipment gets higher, the relative price of the power (W) gets normally lower. This is logical as the equipment contains common parts, designing, racks etc. that generates equal type of costs regardless of the differences in the power amplifier block. On the other hand, the highest power level transmitters are more complicated with e.g. liquid cooling requirements which cause the rise of the cost as the power level is considered. The following Figure 2 summarises an example of the market prices of the DVB-H transmitters. Relative cost of transmitters compared to 500 W TX only an idea how the price of single watt produced by different transmitter types could possibly behave. The CAPEX estimation for this case study can be seen in the following Figure 3. An estimation of the absolute prices was used in this analysis. 3,50 3,00 2,50 2,00 1,50 1,00 0,50 0,00 Normalised CAPEX per site, compared to 500W TX Transmitter power level Fig. 3. An example of the CAPEX per site. The values are normalised using the 500 W transmitter type as a reference. The cost includes the transmitter and antenna system as well as related installation services. The following Figure 4 shows an example of the OPEX for the same cases as shown previously. 2,50 2,00 1,50 1,00 0,50 0, Transmitter power (W) Fig. 2. An example of the relative comparison of the DVB-H transmitter power level price. The curve presents the cost of the single watt produced by different transmitter types. 1,40 1,20 1,00 0,80 0,60 0,40 0,20 0,00 Normalised OPEX per site, ref 500 W TX Site rent Transmission Electricity Transmitter power level In Figure 2, the 500 W transmitter type represents the normalised reference (i.e. the price of single watt produced by 500 W transmitter type). According to this specific example, the cost for producing a single watt is half when using the 1500 W transmitter instead of 500 W type (i.e. in this specific example, W solution is two times more expensive compared to W). Note that the cost value depends totally on the vendor list prices, and that the Figure 2 gives Fig. 4. The relative behaviour of OPEX, with respective analysis of the OPEX items. As can be observed form the Figure 4, the OPEX items are basically constant except for the electricity consumption. For the highest power level cases, this might turn out to be an important cost and has thus negative impact if the network is planned with only few high-power sites. For the transmission, the

218 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 87 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA terrestrial cable solution with sufficiently high bit pipe was used commonly in each case of this analysis. 5. Coverage estimation In order to estimate the radius for single cell in each transmitter case, an Okumura-Hata propagation model can be used. In this analysis, a sub-urban area type was used with QPSK modulation, antenna height of 100 m, terminal height of 1.5 m, service level of 90 % (area location probability), shadowing margin of 5.5 db, frequency of 700 MHz and receiver antenna gain of -7 dbi. The code rate of ½ and MPE-FEC of ¾ was selected. The parameter selection yields the minimum received power level of about -87 dbm for the functional service. The cell radius was calculated for the transmitter output power levels of 100-9,000 W. Antenna gain of 13 dbi was used in each case, which is the result of directional antennas (or antenna arrays, depending on the power level) installed in 3 sectors without downtilting. Table 1. The calculated cell range for each case. TX power EIRP / W Range r / km The calculation takes into account the jumper, connector, power splitter and feeder losses. A feeder of 1 5/8 with the loss of 1.9 db/100m was selected for the power levels of W, and a 3 cable with the loss of 1.5 db/100m was used for the power levels of 4,700-9,000 W. An estimation of 10% for transmitter filter loss was used in each case. The assumption was to use the antenna in tower, which means that the same antenna feeder length (133 m) was used in each case. The cell range that was obtained by taking into account the above mentioned values can be seen in Table I. The next step of this methodology includes the selection of a physical service area. The cells are then placed in the planned area in order to estimate the total cost of the network for each power level case. The area is thus filled with cells as tightly as possible, using the hexagonal model. The cell coverage area is represented with a circle that touches the edges of the hexagonal element. The circles overlap partially in the cell edges resulting relatively realistic presentation of the coverage areas. In practice, this provides the service continuity as well as SFN gain due to the multiple path of the radio signal. The overlapping area presented in this analysis can be calculated geometrically by comparing the surface of the circle with the hexagonal area. r rcos(30) r/2 Overlapping area of single cell Fig. 5. The overlapping area of the individual cell can be calculated by the difference between the surface of the circle and hexagonal element. When the radius of the cell (circle) is r, the surface of the hexagonal inside of the circle is: r Ah = 6r cos(30) = 3r 2 2 cos(30) In this analysis, the overlapping area is taken into account as a reduction factor Rf when estimating the total cell number in given service area. It means that the overlapping area exists in the investigated area, but the reduction factor gives the possibility to calculate the single cell areas and the number of the cells by using the formula of the surface of the circles. The reduction factor can be obtained be the following formula: Rf = Ah Ac 3r = 2 cos(30) 3cos(30) = 82.7% 2 πr π With the reduction factor, it is thus possible to estimate, how many partially overlapping omni-cells (N cells ) with a form of the circle and the radius of r fits into the planned service area. The formula is the following: N A = A A Rf = πr tot tot cells 2 cell cell Rf

219 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 87 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA 6. Results When observing the results calculated for the total service area (in this analysis an area of 100 km 100 km was selected), the following CAPEX and OPEX relation can be obtained depending on the power level of the site. As can be seen from Figure 7, there exists an optimal point for both CAPEX and OPEX (when observing the 4 th year of operation costs) curves. In this specific case, the optimal power level is found in 3.4 kw category. 7,00 6,00 5,00 4,00 3,00 2,00 1,00 0,00 CAPEX and CAPEX+OPEX in 4 years, total area CAPEX/10000 km2 C+O/10000km2/Y Transmitter power level Fig. 7. The relative CAPEX and OPEX comparison of different power level cases. It is interesting to notice that the OPEX and CAPEX curves follows the general trend of the power unit price for different transmitter types, but nevertheless, the final relative CAPEX and combined CAPEX / OPEX grows faster for the highest power level cases that takes into account all the relevant cost items for each power level case. For the OPEX, a more specific analysis can be done. The Figure 8 shows the development of the cumulative operating costs during 4 years form the initial deployment of the network. The yearly OPEX is constant for each transmitter case, producing lines with certain angular coefficient. The following Figure 9 shows an amplified view of the OPEX development in order to observe the detailed behaviour of different power levels. It can be seen that e.g. for the power level of 750 W, the initial cost is lower than in average with the smallpower levels, but the operating cost of 750 W case is considerably higher than could perhaps be expected. In this very case, it can also be seen that the highest power level, i.e. 9 kw, is relatively expensive solution as the CAPEX is considered, and regardless of the considerably lower amount of the sites compared to the lower power level cases, the cumulative OPEX development (angular coefficient of the line) is only slightly lower than that of the mid-power transmitters, mainly because of the higher power consumption. 1,50 1,40 1,30 1,20 1,10 1,00 0,90 0,80 0,70 0,60 0,50 200W CAPEX + OPEX 500W 750W 3400W Y0 Y1 Y2 Y3 Y4 year Fig. 8. An amplified view of the OPEX development. 1500W 9000W 2800W 4700W When observing the angular coefficients of each case and taking into account the development of the network for 4 years of time period, the optimal power level is thus found in the mid-level power range, i.e. the respective transmitters provides with the lowest CAPEX and OPEX of the DVB-H network in this specific case. In generic situation, the coefficients can be calculated by the formula: y y0 k = ( x x ) 0 The term y 0 represents the CAPEX (the cost in initial year), and x 0 can be marked as 0 as it represents the beginning of the operation, i.e. the year 0. It is thus straightforward to calculate the total cost of the network after x years: y = kx + y 0 The coefficients of this specific analysis are shown in Table 2. It can be seen that the coefficient lowers when the transmitter power level is higher. The task would thus be to find the case that yields lowest total cost (CAPEX and cumulated OPEX) within x years.

220 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 87 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA Table 2. The angular coefficient of different transmitter power levels. TX power Y0 k In this specific analysis, the 3,400 W transmitter would provide the lowest total cost for the time scale of 0-6 years of operation. The 4,700 W turns out to be more attractive if the network would operate during 7-45 years, and theoretically, the 9,000 W transmitter would yield the lowest costs if the network would operate in the very same setup at least 46 years. 7. Conclusions The method presented in this paper shows the possibility to estimate the costs of the network, both the initial as well as the operating ones, by observing the angular coefficient of the CAPEX and OPEX as presented in Figures 8 and 9. It is obvious that in infinite scale, the solution with lowest angular coefficient does have the lowest final cost regardless of the initial investments. In practice, though, the DVB-H network does have a limited life time. This should be used as one of the parameters in the analysis of the final cost of the network in function of the power level of the transmitter. It is interesting to note that the behaviour of the OPEX depends strongly on the power level. This means that cost-wise, it is not same to build the coverage area with a big amount of low-power transmitters as with lower amount of mid-power transmitters. It is also worth noting that the optimal CAPEX and OPEX of the network is not necessarily achieved by using the highest power levels. In detailed optimisation of the DVB-H network, it is thus essential to obtain all the relevant CAPEX and OPEX related items and carry out the combined cost and technical analysis for different transmitter types, i.e. varying the power level, in the very same total coverage area for each case and observing the effect of the power levels on the cost of the network. The models presented in this paper are theoretical, and the final sites cannot be obtained from the ideal locations shown by the hexagonal cell distribution. In practice, there is thus need for cell-based adjustments of the power levels, and probably a combination of different power levels in different area types is needed with variable antenna heights, different antenna elements and down-tilting in selected locations. In addition, there might be need to limit the power levels in order to avoid the interferences outside a single SFN area, i.e. the theoretical limitations of SFN could possibly be avoided by using low power levels. Nevertheless, the method presented in this paper gives means to investigate the logical combinations as a basis for the initial network planning. It is thus probable that the use of the presented method yields savings in the DVB-H network deployment and operation. 8. References [1] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [2] Jukka Henriksson. DVB-H standard, principles and services. HUT seminar T Helsinki, Presentation material. 53 p. [3] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [4] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. IEEE p. [5] Editor: Davide Milanesio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Network issues. Project report, May p. [6] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D6 Wing TV Common field trials report. Project report, November p. [7] Myron D. Fanton. Analysis of Antenna Beam-tilt and Broadcast Coverage. ERI Technical Series, Vol 6, April p. [8] William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp

221 II Publication II Field measurement and data analysis method for DVB-H mobile devices This paper presents a methodology for the field test analysis by using a mobile DVB-H device. A set of field tests is carried out in the coverage area of a test network, varying the most important radio parameters. Also respective examples of the performance of the radio network are presented. The Third International Conference on Digital Telecommunications ICDT 2008 June 29 July 5, Bucharest, Romania 2008 IEEE Computer Society Press Reprinted with permission.

222

223 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 90 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA Field Measurement and Data Analysis Method for DVB-H Mobile Devices Jyrki T.J. Penttinen Member, IEEE jyrki.penttinen@nsn.com Abstract The field measurement equipment that provides reliable results is essential in the quality verification of DVB-H networks. In addition, sufficiently in-depth analysis of the post-processed data is important. This paper presents a method to collect and analyse the key performance indicators of the DVB-H radio interface, using a mobile device as a measurement and data collection equipment. 1. Introduction The verification of the DVB-H quality of service level can be done by carrying out field measurements within the coverage area. Correct measurement data, as well as the right interpretation of it, are fundamental for the detailed network planning and optimisation. During the normal operation of the DVB-H network, there are only few possibilities to carry out long-lasting, in-depth measurements. A simple and fast field measurement method based on mobile DVB-H receiver provides thus added value for the operator. The mobile equipment is easy to carry both in outdoor and indoor environment, and it stores sufficiently detailed performance data for post-processing. 2. Measurement equipment In some cases, DVB-H network element might fail in such way that the DVB-H operations and maintenance system is not able to interpret correctly the instance. As an example, the antenna element might turn around due to the loose mounting, resulting outages in the designed coverage area. The antenna feeder might still remain connected correctly, keeping the reflected power in acceptable level. As DVB-H is broadcast system, the only way to verify this kind of fails is to carry out field tests. This paper presents a method to post-process the basic field test data collected with mobile terminal. The method can be considered as an addition to the usual network performance tests, and is suitable for fast revisions of the quality and faults. The field measurement results presented in this paper are meant as examples and for clarifying the methodology. The data presented in the result chapter was collected with a commercial DVB-H hand-held terminal capable of measuring and storing the radio link related data. In this specific case, a Nokia N-92 terminal was used with a field test program. The program has been developed by Nokia for displaying and storing the most relevant DVB-H radio performance indicators 3. Test setup The methodology was verified by carrying out various field tests mostly in vehicle. There was also static and dynamic pedestrian type of measurements included to verify the usability of the equipment and to evaluate the usability of the method. The DVB-H test network consisted of one 200 W DVB-H transmitter and a complete DVB-H core network. The source data was delivered to radio interface by capturing real-time television program. The program was converted to DVB-H IP data stream with a standard DVB-H encoder. There was a set of 3 DVB-H channels defined in the same radio frequency, with audio/video bit rates of 128, 256 and 384 kb/s. The antenna system consisted of directional antenna panel array, each producing 65 degrees of horizontal beam width and 13.1 dbi gain. The vertical beam width of the single antenna element was 27 degrees, which was narrowed by locating two antennas on top of each others via a power splitter. Taking into account the loss of cabling, jumper, connector, power splitter and transmitter filter, the radiating power was estimated to be 62.0 dbm (EIRP). The Figure 1 shows the antenna setup. The transmitter antenna system was installed on a rooftop with 30 meters of height. The environment consisted of sub-urban and residential types with LOS

224 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 90 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA (line-of-sight) or nearly LOS in major part of the test route, except behind the site building which was non- LOS. Each test route consisted of two rounds in main lobe of the antenna. The maximum distance between the antenna and terminals was about 6.4 km. Figure 1. The antenna system setup. Nokia N-92 terminal was used during the test cases. Nevertheless, if the relevant data can be measured from the radio interface and stored in text format, the method presented in this paper is independent of the terminal type. It is important to notice, though, that the characteristics of the terminal affects on the analysis, i.e. the terminal noise factor and the antenna gain (which is normally negative in case of small DVB-H terminals) should be taken into account accordingly. On the other hand, unlike with the advanced field measurement equipment, the method gives a good idea about the quality that the DVB-H users observe in real life as the terminal type with its limitations is the same as used in commercial networks. There were a total of 3 terminals used in each test case, capturing the radio signal simultaneously. Multiple receptions provide respectively more data to be collected at the same time, which increases the statistical reliability of the measurements. It also makes possible the comparison of the differences between the terminal performances. The terminals were kept in the same position inside the vehicle without external antenna, and the results of each test case were saved in separate text files. 4. Terminal measurement principles The DVB-H parameter set was adjusted according to each test case. The cases included the variation of the code rate, MPE-FEC, guard interval and interleaving size (2k, 4k, 8k), in accordance with the Wing TV principles described in [2], [3] [4] and [6]. The parameter set was fixed for each case, and the audio/video stream was received with the terminals by driving the test route 2 consecutive times per each parameter setting. The needed input for the field test is the on and off time of the time sliced burst, PID (packet identifier) of the investigated burst, the number of FEC rows and the radio parameter values (frequency, modulation, code rate and bandwidth). The N-92 stores the measurement results to a log file after the end of each burst until the field test execution is terminated. According to the DVB-H implementation guidelines [1], the target quality of service is the following: For the bit error rate after Viterbi (BA), the DVB- H specific QEF (quasi error free) point should be better than The frame error rate should be less than 5%. The field test software of N-92 is capable of collecting the RSSI (received power in dbm), FER (frame error rate) and MFER (FER after MPE-FEC correction) values. In addition, there is possibility to collect information about the packet errors. The Figure 2 shows a high-level block diagram of the DVB-H receiver. [1] The reception of the Transport Stream (TS) is compatible with DVB-T system, and the demodulation is thus done with the same principles also in DVB-H. The additional DVB- H specific functionality consists of Time Slicing, MPE-FEC and the DVB-H de-encapsulation. RF reference DVB-T Demodulator TS reference DVB-H specific functionality DVB-H Time Slicing DVB-H MPE-FEC DVB-H De-encapsulation IP output FER reference MFER reference IP reference Fig. 2. A principle of the reference DVB-H terminal. As can be seen from the Figure 2, the FER information is obtained after the Time Slicing process, and the MFER is obtained after the MPE-FEC correction module. If the MFER is free of errors, the respective data frame is decoded correctly and the IP output stream can be observed without disturbances. The measurement point for the received power level is found after the antenna element and the optional

225 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 90 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA GSM interference filter. In addition, there might be optional external antenna connectors implemented before the RF reference point. The presence of the filter and antenna connectors has thus frequencydependent loss effect on the measured received power level in the RF point. The Figure 3 shows an example of the measurement data display of N-92. In this case, there was a frame error in the reception because the value of FER was 1. The FER value is either 0 for non-erroneous or 1 for erroneous frame. Furthermore, the MPE-FEC could still recover the error in this case, because the MFER parameter is showing a value of 0. FER 1 MFER 0 BB 1.10E-02 BA 8.00E-04 PE 111 RSSI -84 Fig. 3. Example of the measured objects. According to the Figure 3, the bit error level before Viterbi (BB) was above the QEF point, i.e The bit error level after the Viterbi (BA) was which is clearly better than the QEF point for the acceptable reception. The bit error rate had been thus low enough for the correct functioning of the MPE- FEC. In this example, the amount of packet errors (PE) was 111, and the averaged received power level, i.e. RSSI, was measured and averaged to -84 dbm. The RSSI resolution is 1 db for single measurement event in the used version of the field test software. The following Figure 4 shows the RSSI value during the complete test route. There were two rounds done during each test. The received power level was about -50 dbm close to the site, and about -90 dbm in the cell edge. The duration of the single test route was 25 minutes, and the total length was 22.4 km Fig. 4. The RSSI values measured during the test route. The maximum speed during the test route was about 90 km/h, and the average speed was measured to 50 km/h (excluding the full stop periods). The speed is sufficient for identifying the effect of the MPE-FEC. 5. Method for the analysis The collected data was processed accordingly in order to obtain the breaking points, i.e. the QEF of and FER / MFER of 5% in function of the RSSI values for each test case. The processing was carried out by arranging the occurred events per RSSI value. For the BB and BA, the values were averaged per RSSI resolution of 1 db. For the FER and MFER, the values represent the percentage of the erroneous frames per each RSSI value. The following Figure 5 shows the processed data for the bit error rate before and after the Viterbi. The results represent the situation over the whole test route in location-independent way, i.e. the results show the collected and averaged BB and BA values that have occurred related to each RSSI value. As can be noted in this specific example, the bit error rate before Viterbi does not comply with the QEF criteria of even in good radio conditions, whereas the Viterbi clearly enhances the performance. The resulting breaking point for the QEF with Viterbi can be found around -83 dbm of RSSI in this specific case. BER 1.00E E E E E E E E E-08 BB and BA, average for each Prx value QEF BB ave BA ave Fig. 5. Processed data for the bit error rate before and after the Viterbi. For the frame error rate, the similar analysis yields an example that can be observed in Figure 6. The Figure shows the occurred frame error counts (FER and MFER) as well as the amount of error-free events per each RSSI value. In this format, the Figure shows the amount of occurred samples per RSSI value (in 1 db raster) arranged to error free counts ( count FER0 MFER 0 ), to counts that had error but could be corrected with MPE-FEC ( FER 1 MFER 0 ), and to counts that were erroneous even after MPE-FEC ( MFER 1 ). It can be noted that the amount of the occurred events is low in the best field strength cases and does not provide with sufficient statistical reliability in that dbm

226 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 90 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA range of RSSI values. Nevertheless, as the idea was to observe the limits of the coverage area, it is important to collect sufficiently data especially around the critical RSSI value ranges. Counts Count FER0MFER0 MFER1 # FER1MFER0 # Count of samples per Prx level Fig. 6. Example of the analysed FER and MFER level of the signal. In this type of analysis, the data begins to be statistically sufficiently reliable when several tens of occasions per RSSI value are obtained, preferably around 100 samples. In practice, though, the problem arises from the available time for the measurements, i.e. in order to collect about 100 samples per RSSI value it might take more than one hour to complete a single test case. Next, the corresponding amount of total samples was normalized, i.e. scaled to 0-100% for each RSSI value. An example of this is shown in Figure 7. dbm By presenting the results in this way, the percentage of FER and MFER per RSSI and thus the breaking point of FER / MFER can be obtained graphically. The 5% FER and MFER level can be obtained graphically for each case observing the breaking point for the respective curves. The corresponding MPE- FEC gain is the difference between FER and MFER values (in db), which can be obtained by observing the 5% breaking point. 6. Other measurements For comparison purposes, there was also a set of test cases carried out in the pedestrian environment with the same measurement methodology. The following Figures 8 and 9 shows the results of a short snapshot type of measurement in about 800 m distance from the transmitter antenna, in the main lobe. The measurement consists of measurements inside and outside of a 1-floor building, with a slowly moving terminal. The slow movement is needed in order to average the received power level and to take into account correctly the Rayleigh fading Count of samples FER and MFER % criteria <5% criteria FER% MFER % Fig. 7. The samples that were collected in outdoor case. dbm Count of samples % MPE-FEC gain Bit error rate (5% breaking point) before MPE-FEC dbm Bit error rate (5% breaking point) after MPE-FEC Fig. 6. The post-processed data can be presented in graphical format which consists of the normalised percentage of FER and MFER for each RSSI value dbm Fig. 8. The samples that were collected in indoor case.

227 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 90 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA The results show an average of dbm for the outdoor RSSI, with a standard deviation of 6.1 db. For the indoor, the average of RSSI was dbm with the standard deviation of 2.9 db. It is thus straightforward to estimate the average building loss to be about 14.1 db in this specific case. As the terminal speed was low, the MPE-FEC is not able to correct the possible frame errors, and the respective analysis that was described previously for MPE-FEC gain is thus not needed for this measurement type. The terminal measures the received power level after the possible (optional) GSM interference suppression filter. There might also be external antenna connectors in either side of the filter. The terminal characteristics thus affects on the received power level interpretation. In order to obtain information about the possible differences of the terminal displays, separate comparison measurements were carried out. There were a total of three N-92 terminals used during the testing. As the terminals were still prototypes, the calibration of the RSSI displays was not verified. This adds uncertainty factor to the test results. The following Figure 9 shows a test case that was carried out in laboratory by keeping all the terminals in the same position and making slow-moving rounds within relatively good coverage area. 7. Results As a result of the vehicle based field tests performed in this study, the following Tables 1-3 summarises the RSSI thresholds for the QEF point of and FER / MFER of 5% criteria with different parameter values. In addition, the effect of MPE-FEC was obtained graphically for each parameter setting. The analysis was made for the post-processed data by observing the breaking points of BB, BA, FER and MFER of the averaged values of 2 terminals. The guard interval (GI) was set to ¼ in each case. The MPE-FEC gain was obtained for each studied case. The effect seem to be lowest in 64-QAM modulation, which might mean that the receiver has been optimised for the modes that are most probable in mobile environment, i.e. the QPSK and 16-QAM are more like to be used outdoors whereas 64-QAM could be most logical in indoor environment with slow moving terminals. It should be noted, though, that the terminals were not calibrated especially for this study. The RSSI display might thus differ from the real received power levels with some decibels. The calibration should be done e.g. by examining first the level of the noise floor of the terminal and secondly examining the QEF point, i.e. investigating the signal level which is just sufficient to be received correctly Fig. 9. An example of the laboratory test case for the comparison of the RSSI displays of the terminals. The systematic difference in RSSI displays can be noted, being about 2 db between the extreme values. The same 2 db difference between the terminals was noted in the field test analysis. The values obtained from the radio network tests cannot thus be considered accurate. Nevertheless, the idea of the testing was to investigate rather the methodology of the measurements than to obtain accurate values of the defined parameter settings. TABLE I THE RESULTS FOR QPSK CASES. THE VALUES REPRESENTS THE RSSI IN DBM, EXCEPT FOR THE MPE-FEC GAIN, WHICH IS SHOWN IN DESIBELS. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/2-88,1-83,8-78,4-86,6 5%, FEC 1/2-84,0-77,3-72,7-81,4 MPE-FEC 1/2 gain 4,1 6,5 5,7 5,2 5%, MPE-FEC 2/3-87,3-83,0-76,7-84,4 5%, FEC 2/3-83,3-72,9-73,5-81,0 MPE-FEC 2/3 gain 4,0 10,1 3,2 3,4 BA QEF average: -85,8-81,7-78,4-84,9 TABLE II THE RESULTS FOR 16-QAM CASES. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/2-77,6-61,8-77,4-77,7 5%, FEC 1/2-69,8-61,6-74,2-73,4 MPE-FEC 1/2 gain 7,8 0,3 3,7 4,3 5%, MPE-FEC 2/3-77,0-63,5-75,5-77,0 5%, FEC 2/3-72,1-59,0-71,0-74,7 MPE-FEC 2/3 gain 4,9 4,5 4,5 2,3 BA QEF average: -78,3-73,7-77,2-77,4

228 THE THIRD INTERNATIONAL CONFERENCE ON DIGITAL TELECOMMUNICATIONS PAPER NUMBER 90 ICDT 2008, JUNE 29 - JULY 5, BUCHAREST, ROMANIA TABLE III THE RESULTS FOR 64-QAM CASES. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/2-59,9-51,6-65,0-67,3 5%, FEC 1/2-59,7-51,6-57,5-65,2 MPE-FEC 1/2 gain 0,2 0,1 7,5 2,1 5%, MPE-FEC 2/3-61,0-51,3-59,5-68,1 5%, FEC 2/3-60,3-50,6-54,5-65,8 MPE-FEC 2/3 gain 0,7 0,7 5,0 2,4 BA QEF average: -60,7-53,0-61,9-68,3 It can be assumed that the most reliable results are obtained by observing the FER / MFER of the data. The frame rate error reflects the practical situation as the user interpretation of the quality depends on the amount of correctly received frames. For the bit error rate before and after Viterbi, it is not necessarily clear how the terminal calculates the value shown in the displays especially in the cell edge with high error rates. Nevertheless, the results correlate with the theory of different parameter settings, as well as with the MPE- FEC gain. As the test route contained different radio channel types (different vehicle speeds, LOS, near- LOS and non-los behind the building), the mix of the propagation types causes uncertainty to the results. In order to obtain the values nearer to the theoretical ones, it would be important to carry out the test cases in separate, uniform areas as the radio channel type is considered, but on the other hand, these results represent the real situation in the investigated area with a practical mix of radio channel types. 8. Conclusion The benefit of the hand-held receiver is obvious in the measurements presented in this paper as the equipment is easy to carry to different environments, including indoors. The data collection with hand-held terminal is fast, and the collected radio interface performance indicators provide sufficiently data for the post-processing. The tests presented in this paper shows that the realistic DVB-H measurement data can be collected with the terminals. The analysis showed correlation between the post-processed data and estimated coverage that was calculated and plotted separately with a network planning tool. The results correlate mostly with the theoretical DVB-H performance, although there was a set of uncertainty factors identified that affects on the accuracy of the results. This study was merely meant to develop and verify the functionality of the analysis methodology instead of the verification of accurate data. The test environment consisted of multiple radio channel types, and the terminal displays were not calibrated specifically for these tests. An error of few decibels is thus expected. Nevertheless, the results show that the terminals can be used as an additional tool for fast revision of the overall functioning of the network. With the collected data and respective post-processing, it is possible to observe the DVB-H audio/video quality in detailed level compared to the subjective studies. The field test results show clearly the effect of the parameter values on radio performance in a typical sub-urban environment. Even if the hand-held terminal is not the most accurate device for the scientific purposes, it gives an overview about the general functioning and quality level of the network and the estimation of the effects of different network parameter settings. 9. References [1] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [2] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [3] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D6: Common field trials report. November p. [4] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. Wing TV Country field report. November p. [5] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. Proceedings of the Vol. 94, No. 1. January p. [6] Editor: Davide Milanesio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Network issues. Project report, May p. [7] William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp

229 III Publication III The simulation of the interference levels in extended DVB-H SFN areas This paper describes a method for the investigation of the interference level caused by the overdimensioning of the Single Frequency Network of DVB-H radio network. A simulator has been programmed as a basis for the investigation and relevant cases are studied by varying the radio parameters. The SFN error levels are identified as function of the radiating power levels and antenna heights. The Fourth International Conference on Wireless and Mobile Communications ICWMC 2008 July 27 August 1, Athens, Greece 2008 IEEE Computer Society Press Reprinted with permission.

230

231 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 224 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE The Simulation of the Interference Levels in Extended DVB-H SFN Areas Jyrki T.J. Penttinen Member, IEEE jyrki.penttinen@nsn.com Abstract The maximum size of the Single Frequency Network (SFN) of DVB-H depends on the guard interval (GI) and FFT mode. The distance limitation between the extreme transmitter sites is thus straightforward to calculate. Nevertheless, there might be need to extend the theoretical SFN areas e.g. due to the lack of frequencies. This paper describes a simulation method for obtaining information about the effect of parameter settings on the error levels caused by the over-sized DVB-H SFN area. The functionality of the method was tested by programming a simulator and analysing the variations of carrier per interference distribution. The results show that the limits can be exceeded e.g. by selecting the antenna height in optimal way and accepting certain increase of the error level that is called SFN error rate (SER) in this analysis. 1. Introduction The DVB-H network can be deployed with Single (SFN) or Multi Frequency Network (MFN) modes. In SFN, the transmitters can be added inside the allowed area without inter-symbol interferences. Furthermore, the overlapping signals increase the performance of the network due to the SFN gain. Sites located within the SFN area minimises the risk of the interferences as the guard interval (GI) protects the OFDM signals of DVB-H. If certain degradation in the quality level of the received signal is accepted, it could be justified to even extend the SFN limits. This paper presents a method to simulate the respective quality level of DVB-H by observing the variations of carrier-to-interference levels in function of radio parameters in over-sized SFN. The additional error rate caused by the exceeding of SFN is called SFN error rate, or SER, in this paper. The impact can also be investigated by calculating a ratio of SER events over the total simulation area. In the latter case, the term can be called SFN area error rate, SAER. 2. Theory of SFN limits According to [1], the GI and the FFT mode of DVB-H network determinate the maximum delay that the mobile can handle in order to receive correctly the multi-path propagated components of the signals. The Table 1 summarises the SFN distances. Table 1. The guard interval lengths and respective SFN distances. GI FFT 2K FFT 4K FFT 8K 1/4 56 µs/16.8 km 112 µs/33.6 km 224 µs/67 km 1/8 28 µs/8.4 km 56 µs/16.8 km 112 µs/33.6 km 1/16 14 µs/4.2 km 28 µs/8.4 km 56 µs/16.8 km 1/32 7 µs / 2.1 km 14 µs/4.2 km 28 µs/8.4 km As long as the distance between the transmitter sites is less than the maximum allowed, the difference of the delays between the signals originated from these sites never exceeds the allowed margin excluding strong multipath propagated signal. If the site distance is greater than the theoretical limitation allow, the situation changes. As an example, the GI of ¼ and 8K mode provides 224 µs margin for the safe propagation delay. Assuming the signal propagates with the speed of light, the SFN size limit is 300,000 km/s 224 µs yielding about 67 km SFN distance. If any distance combination of the sites using the same frequency exceeds this value, they start producing interference in spots where the difference of the signal propagation delays is higher than 224 µs. If the level of interference is greater than the noise floor, and the minimum C/N value that the respective mode requires is not any more complied, the signal in that specific spot is interfered. The interference thus increases the required received power level of the carrier up to C/(N+I). Even if the C/(N+I) level gets lower when moving from one site to another, the situation is not necessarily critical as the effective distance D eff of the signals (difference of the delays) might be within the SFN

232 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 224 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE limits e.g. in the middle of two sites, although their distance from each others would be greater than the maximum allowed. The otherwise interfering site might thus not be considered as interference in the respective spot, but instead, it produces SFN gain. The interference increases especially when the terminal moves away from the centre of the SFN network. The setup for the simulation consists of area type and geometrical area where the cells are located. The cell radius is dimensioned according to the minimum C/N requirement. When estimating the total carrier per interference levels, both total level of the carriers and interferences are calculated separately by the following formulas, using the respective absolute power levels (W) of C and I components: I/TX1 C/TX2 C tot = C1 + C C (1) n I tot = I1 + I I (2) n Noise floor SFN limit Extended SFN area Carrier from TX2 Interference from TX1 Noise Fig. 1. The principle of interference when the site is out of the SFN limit. The sum of thermal noise floor and noise figure of the terminal is the reference for C and I, and it depends on the bandwidth. As an example, the combined noise floor and noise figure value is dbm for the 8 MHz case. The original minimum received power level C/N of the signal raises up to C/(N+I) when the interference is present. For the multi-site network, the distance between the terminal and the sites determines if the signal of individual site is carrier or interference. The total C/(N+I) consists of the sum of separate components of C and I. The required C/N for some of the most commonly used parameter setting is seen in Table 2. [1] Table 2. The minimum C/N for the selected parameter settings. The terminal antenna gain (loss) is taken into account in the values. Parameters C/N (db) QPSK, CR 1/2, MPE-FEC 1/2 8.5 QPSK, CR 1/2, MPE-FEC 2/ QAM, CR 1/2, MPE-FEC 1/ QAM, CR 1/2, MPE-FEC 2/ Methodology for the SFN simulations In order to estimate the error level of multi-sites that is caused by extending the theoretical geometrical limits of the SFN network, a simulation can be carried out. The investigated variables can be e.g. the antenna heights, radiating power levels of the site and the GI and mode that define the SFN limits. In each simulation round, the site with the highest field strength is identified. In case of uniform network and equal site configurations, the site with the lowest propagation loss corresponds to the nearest cell TX1 which is selected as a reference in each round. Once the nearest cell is identified, the task is to investigate the propagation delays of signals form the nearest and each one of the other sites, and calculate if the difference of arriving signals D eff is greater or lower than the SFN limit D sfn. In general, if the difference of the signal arrival times of TX1 and TXn is greater than GI defines, the TXn is producing interfering signal (if the signal is above the noise floor), and otherwise it is adding the level of total carrier energy (if the signal level is above the minimum requirement for carrier). In order to obtain the level of C and I, the path loss can be estimated e.g. with Okumura-Hata. The total path loss can be calculated by the following formula: L = L + L (3) tot pathloss norm L norm is the fading caused by the long-term variations in the path loss. For the long-term fading, a normal distribution is commonly used in order to model the variations of the signal level. The formula is the following [5]: 2 ( x x) 1 PDF( L ) = exp 2 (4) norm 2πσ 2σ The term x represents the loss value, and x is the average loss (0 in this case). In the snap-shot based simulations, the L norm is calculated for each arriving signal individually as the different events does not have correlation. The respective PDF and CDF are obtained by creating a probability table for normal distributions. Figure 3 shows an example of the PDF and CDF of normal distributed loss variations.

233 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 224 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE PDF and CDF of normal distribution, stdev=5.5 PDF CDF db Fig. 3. PDF and CDF of normal distribution representing the variations of long-term loss. 4. Simulator The block diagram of the SFN simulator is shown in the Figure 4. The simulator was programmed with standard Pascal code. It produces the results to text files, containing the C/N, I/N and C/(N+I) values showing the distribution in scale of db and with 0.1 db resolution. Also the MS coordinates and respective C/N, I/N and C/(N+I) is produced. A total of 60,000 simulation rounds per each case were carried out. It corresponds to an average of 60,000 / (50 db 10) = 120 samples per C/I resolution. The terminal was placed in 100 km 100 km area according to the uniform distribution in function of the coordinates (x, y) during each simulation round. The raster of the area was set to 10 m. Small and medium city area type was selected for the simulations. The total C/I value is calculated per simulation round by observing the individual signals of the sites. Inputs: Geographical and simulation area, radio parameters. difference t 2 t 1 is less than D SFN defines, the respective signal is marked as a carrier C, otherwise it is marked as an interference I. In generic format, the total C/(N+I) can be obtained in db units from the simulation results by comparing the level of the carrier (C / db) with the level of interference (I / db), having the noise floor (i.e. the sum of thermal noise floor and terminal noise figure) as a reference for both values. The final C/(N+I) is the difference between C and I. The noise figure depends on the terminal, but in these simulations, it is estimated to 5 db as defined in [1]. Each simulation round provides information if that specific connection is useless, e.g. if the criteria set of 1) effective distance D eff >D SFN in any of the cells, and 2) C/(N+I) < min C/N threshold. If both criteria are valid, and if the C/N would have been sufficient without the interference in that specific round, the SFN interference level is calculated. The Figure 5 shows an example of the site locations. The simulator calculates the optimal cell radius according to the parameter setting and locates the transmitters on map according to the hexagonal model, leaving the overlapping areas in the cell border areas. The size and thus the number of the cells depends on the radio parameter settings without interferences, and in each case, a result is a uniform service level in the whole investigated area. The behaviour of C and I can be observed by the probability density functions, i.e. PDF of the results. The Figures 6 and 7 gives indication about the overall quality of the network with two different parameter settings. The first one does have only moderate interference level whilst the interference of the latter case is considerably higher. 100 TX locations Tables: Create CDF of lognormal distribution for long-term fading and optional CDF for Rayleigh fading. 80 Initialisation: Calculate the single cell radius with Okumura-Hata and place the sites on map according to the hexagonal principles. Simulation: Place the mobile station on map. Calculate the respective C/I levels. Repeat the simulation rounds until the statistical accuracy has been reached. Data storing: Save the C/I PDF and CDF distribution ( db, 0.1 db resolution), coordinates of MS in each simulation round and respective C/I value. Fig. 4. The block diagram of the simulator. The nearest site is selected as a reference during the respective simulation round. If the arrival time delay km km Fig. 5. Example of the transmitter site locations the simulator has generated.

234 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 224 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE # of counts # of counts PDF, C and I, QPSK, FFT=8K, GI=1/4 C distribution I distribution C and I (db) Fig. 6. Example of the PDF for C and I with antenna height of 200 m PDF, C and I, QPSK, FFT=4K, GI=1/4 C distribution I distribution C and I (db) Fig. 7. The C and I distribution for 4K. The interference level is considerably higher with 4K than with 8K. Area location probability (%) CDF, QPSK, GI=1/4 C/(I+N)_QPSK_8K 0.1 C/(I+N)_QPSK_4K C/(N+I) (db) Fig. 8. Example of the CDF of C/(N+I) for QPSK 4K and 8K modes with antenna height of 200 m and transmitter power P tx of 60 dbm. The specific values of the interference levels can be obtained by producing a CDF from the simulation results as shown in the Figure Results By applying the principles of the DVB-H simulator, the C/I distribution was obtained. The variables were the modulation scheme (QPSK and 16-QAM), antenna height ( m) and FFT mode (4K and 8K). The following Figures 9 and 10 shows the resulting networks that were used as a basis for the simulations. # of sites Site number and cell radius for QPSK # sites r_c r_i Ant h (m) Fig. 9. The network dimensions for the QPSK simulations. # of sites Site number and cell radius for 16-QAM # sites r_c r_i Ant h (m) Fig. 10. The network dimensions for the 16- QAM simulations. The simulator locates randomly the mobile terminal on the map and calculates the C/I by taking into account the long-term fading. This is repeated 60,000 times. One of the results is the estimation of the errors due to the interferences from the sites exceeding the SFN distance (i.e. if the arrival times of the signals exceed the maximum allowed delay difference). This event can be called SFN error rate, or SER Cell radius (m) Cell radius (m)

235 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 224 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE C/I (db) C/I w ith outage of 2 10%, 16-QAM, 8K, 100x100km2 Out_2% Out_5% Out_10% y/km x/km Fig. 11. An example of the SER for QPSK in geographical format. The plots indicates the locations where the events results C/(I+N) level corresponding less than 8.5 db. It can be assumed that when the SER level is sufficiently low, the end-users will not experience remarkable reduction in the quality due to the extended SFN limits. The respective interferences tend to cumulate to locations outside of the centre of the network as can be observed form the figure 11. According to the simulations, the SFN interference level varies clearly when the radio parameters are tuned. The following Figures summarises the obtained cumulative C/(N+I) for outage probabilities of 2, 5 and 10 %. In order to have a uniform reference for the performance comparisons, the outage level of 10 % was selected. It is logical point, corresponding to the location probability of 70% in the cell edge in the single cell case. C/I (db) C/I w ith outage of 2 10%, QPSK, 8K,100x100km2 Out_2% Out_5% Out_10% Antenna height (m) Fig. 13. The summary of the case 2 (16-QAM, 8K). C/I (db) C/I w ith outage of 2 10%, QPSK, 4K, 100x100km Antenna height (m) Out_2% Out_5% Out_10% Fig. 14. The summary of the case 3 (QPSK, 4K). C/I (db) C/I w ith outage of 2 10%, 16-QAM, 4K, 100x100km2 Out_2% Out_5% Out_10% Antenna height (m) Fig. 12. The summary of the case 1 (QPSK, 8K). The results show the C/(I+N) with 2%, 5% and 10 % SER criteria Antenna height (m) Fig. 15. The summary of the case 4 (16-QAM, 4K). The Figures shows that with the uniform radio parameters and by varying the antenna height, modulation and FFT mode, the functional settings can be found exceeding the theoretical SFN limits.

236 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 224 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE The analysis shows that the 8K mode provides with sufficiently low SFN interferences for both QPSK and 16-QAM when GI of ¼ is used. For the QPSK, the minimum required C/(N+I) requirement of 8.5 db with the outage probability of 10 % is achieved with all the antenna heights of m. If the mode is changed to 4K, the respective antenna height should be lowered down to 35 m. 16-QAM provides with smaller cell sizes and higher capacity compared to the more robust QPSK cases. The effect of FFT mode can be seen clearly also in this case. For the 16-QAM and 8K, and with the minimum C/(N+I) requirement of 14.5 db for this specific case, there seem to be no limits for the antenna height as the SFN interference limits are considered. If the mode of this case is switched to 4K, the antenna should be lowered down to 30 m in order to still fulfil the maximum of 10 % outage criteria. The following Figure 16 summarises the previously presented cases presenting the outage percentage of different modes in function of transmitter antenna height. As previously, 8.5 db C/(N+I) limit was used for QPSK and 14.5 % for 16-QAM modulation. Outage (%) Outage-% for antenna heights m Outage%_QPSK,8K Outage%_16-QAM,8K Outage%_QPSK,4K Outage%_16-QAM,4K Antenna height (m) Fig. 16. The summary of the simulations, showing the outage-% in function of the transmitter antenna height. The 60 dbm EIRP that was used in the simulations represents relatively low power class for DVB-H. The higher power level raises the SER level accordingly. For the mid and high power sites the optimal setting depends thus even more on the combination of the power level and antenna height. According to these results, it is clear that the FFT mode 8K is the only reasonable option when the SER should be kept in acceptable level in over-sized SFN areas. According to the results, the over-sized SFN network could be planned using initially QPSK, FFT 8K and GI of ¼ providing large coverage areas but lowest capacity. On the other hand, when the capacity requirement increases, the switching to the 16-QAM modulation is the most logical solution. In practice, the SER level can be further decreased by minimising the propagation of the interfering components. This can be done e.g. by adjusting the transmitter antenna down-tilting and using narrow vertical beam widths, producing thus the coverage area of the carrier and interference as close to each others as possible. Also the natural obstacles of the environment can be used efficiently for limiting the interferences far away outside the cell range. 6. Conclusions The controlled extension of the SFN limit might be interesting option for the DVB-H operator. The simulation method and results presented in this paper shows logical behaviour of the SFN error rate when varying the essential radio parameters. The results also show that the optimal setting can be obtained using the respective simulation method. As expected, the 8K mode is the most robust when extending the SFN whilst 4K limits the site antenna height considerably. 16-QAM provides suitable performance for the extension, but according to the results, even QPSK is not useless in SFN extension when selecting the parameters correctly. 7. References [1] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [2] Jukka Henriksson. DVB-H standard, principles and services. HUT seminar T Helsinki, 24 Feb Presentation material. 53 p. [3] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [4] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. IEEE p. [5] William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp

237 IV Publication IV The SFN gain in non-interfered and interfered DVB-H networks This paper shows a method for the dimensioning of the DVB-H network s SFN gain in noninterfering case (inside the allowed SFN limits) as well as in over-sized SFN network that causes interferences. A simulator has been designed for the investigation. A respective simulation is carried out for the most relevant parameter settings and the optimal point of the balance between the SFN gain and SFN error levels has been identified for these case examples. The Fourth International Conference on Wireless and Mobile Communications ICWMC 2008 July 27 August 1, Athens, Greece 2008 IEEE Computer Society Press Reprinted with permission.

238

239 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 446 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE The SFN gain in non-interfered and interfered DVB-H networks Jyrki T.J. Penttinen Member, IEEE jyrki.penttinen@nsn.com Abstract The Single Frequency Network mode (SFN) is one of the benefits of DVB-H network. In addition to the possibility to use only one frequency in given area, the reception of the same contents via two or more radio paths provides with SFN gain which enhances the quality of service whenever the respective sites are located inside the SFN limits. If the same frequency is used outside the theoretical SFN area, the sites that exceed the relative guard distance converts as interfering sources. This paper describes a simulation method for the investigation of the SFN gain in noninterfered network as well as in the environment with SFN interferences. Also simulation results and analysis with the most logical network parameter set is presented. 1. Introduction The benefit of Single Frequency Network (SFN) compared to the Multi Frequency Network (MFN) is the possibility to achieve additional signal levels in the coverage area of DVB-H. The level of SFN gain depends on the number of the useful carriers. The situation changes when the SFN limit is exceeded, i.e. if the maximum diameter of some of the sites is longer than the maximum guard distance of the SFN area. The respective leg of these sites causes interference whenever the difference between the arriving signals exceeds the SFN guard interval. The paper presents a method for obtaining the value of SFN gain and SER via simulations, by varying the most essential radio parameters. In addition, a set of results for selected cases are shown in order to obtain an optimal balance between SFN gain and the SER 2. Theory of the SFN limits According to [1], the combination of GI (Guard Interval) and FFT mode gives the maximum delay that the DVB-H terminal can handle in order to receive correctly the multi-path components of the signals. The Table 1 summarises the values and the respective SFN distances. Table 1. The guard interval lengths and respective SFN distances. GI FFT = 2K FFT = 4K FFT = 8K µs km µs km µs km 1/ / / / The FFT size has impact on the maximum velocity of the terminal, and the GI affects on both the maximum velocity of the terminal as well as on the capacity of the radio interface. In these simulations, if the velocity and capacity are not considered, the following parameter combinations results the same C/N and C/(N+I) performance due to their same requirement for the safety distances: FFT 8K, GI 1/4: unique case FFT 8K, GI 1/8: same as FFT 4K, GI 1/4 FFT 8K, GI 1/16: same as FFT 4K, GI 1/8 and FFT 2K, GI 1/4 FFT 8K, GI 1/32: same as FFT 4K, GI 1/16 and FFT 2K, GI 1/8 FFT 4K, GI 1/32: same as FFT 2K, GI 1/16 FFT 2K, GI 1/32: unique case When the distance between the extreme transmitter sites is less than the SFN determines, the difference of the delays between the signals from different sites is always within the allowed margin. On the other hand, if the distance is greater, the sites interferes, depending if the relative arriving distance of the respective sites is greater than the safety margin shown in Table 1. The received signal is still functional if the total C/(N+I) is at least in the same level as the original minimum required C/N for the respective parameter set.

240 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 446 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE The required C/N depends on the values of code rate (CR), MPE-FEC rate (multi protocol encapsulator, forward error correction) and modulation scheme. The minimum C/N requirement for some of the most common parameter setting can be seen in Table 2. [1] Table 2. The minimum C/N (including antenna loss) for the selected parameter settings. For the carrier and interfering signal, the total level of C/(N+I) is the squared sum of the C/N and I/N components in absolute powers (watt), thermal noise floor and terminal noise factor being the reference. 3. Methodology for SFN simulations The SFN performance simulator is based on the hexagonal cells. The Figure 1 presents the idea of the cell distribution. TX(1,y) Parameters TX(2,y) TX(x-1,y) TX(1,3) TX(2,3) TX(3,3) TX(x,3) C/N (db) QPSK, CR 1/2, MPE-FEC 1/2 8.5 QPSK, CR 1/2, MPE-FEC 2/ QAM, CR 1/2, MPE-FEC 1/ QAM, CR 1/2, MPE-FEC 2/ TX(x,y) y x/2 30deg r 30deg 30deg x/2 Fig. 2. The geometrical characteristics of the hexagonal model used in the simulator. The distance between two sites in x-axis is: ( 30 ) = 2 0. r x = 2r cos 866 (1) The common inter-site distance in y-axis is: ( 30 ) 2 ( 30 ) 3 r cos r y = = (2) tan In odd row case, the formula for the x-coordinates of the site m is thus the following: x ( m 1) 1. r x( m) odd = r (3) In the formula, m represents the number of the cell in x-axis. In the same manner, the formula for x- coordinates can be created in the following way: 1 x( m) even = r r + ( m 1) r (4) 2 TX(1,2) TX(2,2) TX(x-1,2) TX(x,2) For the y-coordinates, the formula is the following: TX(1,1) TX(2,1) TX(3,1) TX(x,1) Fig. 1. The active transmitter sites are selected from the 2-dimensional cell matrix. A uniform parameter set is used for each cell site including the transmitter power level and antenna height. This yields the same radius for each cell per simulation case. The coordinates of each site depends on the uniformly calculated cell radius. Investigating the Figures 1-2 and the hexagonal model behaviour, the x-coordinates can be obtained in the following way depending if the row number for y coordinates is odd or even. y 2 = (5) 3 ( n) r + ( n 1) r The simulations can be carried out for different cell layouts. Symmetrical reuse pattern concept was selected for the simulations presented in this paper. The most meaningful reuse pattern size K can be obtained with the following Formula [5]: 2 K = ( k l) kl (6) The variables k and l are positive integers with minimum value of 0. In the simulations, the reuse pattern sizes of 1, 3, 4, 7, 9, 12, 16, 19 and 21 was used for the carrier and interference distribution in both non-interfering and interfering networks (i.e. SER either exists or not depending on the size of the SFN

241 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 446 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE area). In this way, the lower values of K provides with the non-interfering SFN network until a limit that depends on the GI and FFT size parameters. The single cell (K=1) is considered as a reference in all of the cases. The fixed parameter set was the following: Transmitter power: 60 dbm Transmitter antenna height: 60 m Receiver antenna height: 1.5 m Long-term fading with normal distribution and standard deviation of 5.5 db Area coverage probability in the cell edge: 70% Receiver noise figure: 5 db Bandwidth: 8 MHz Frequency: 700 MHz For the used bandwidth, the combined noise floor and noise figure yields dbm as a reference for calculating the level of C and I. The path loss was calculated with Okumura-Hata prediction model for small and medium sized city. The 70 % area coverage probability corresponds with 10% outage probability in the single cell area. These settings result a reference C/N of 8.5 db for QPSK and 14.5 db for 16-QAM. The value is the minimum acceptable C/N, or in case of interferences, C/(N+I) value that is needed for the successful reception of the signal. The Figure 3 presents the symmetrical reuse pattern sizes that were selected for the simulations. The grey hexagonal cells means that the coordinates has been taken into account calculating the order number of the sites (formulas 3-5), but the respective transmitter has been switched off in order to form the correct reuse pattern. The Figure 4 shows an example of the network setup and locations for the QPSK and K=7. y (km) Network layout for K= x (km) Fig. 4. An example of the QPSK network, K=7. The Figure 5 shows an example of the C/N distribution with the parameter values of K=7, GI=1/4, and FFT=8K. According to the C/I link analysis, the case presented in Figure 5 is free of SFN interferences. Fig. 5. An example of the simulated case with QPSK and K=7. Fig. 3. The reuse patterns, K=1, 3, 4, 7, 9, 12, 16, 19 and 21. The actual simulation results for C/N, or in case of the interferences, for C(/N+I), is done in such way that only the terminal locations inside the calculated cell areas is taken into account. If the MS is found outside of the network area (the circles) in some simulation round, the result is simply rejected. The Figure 6 shows the principle of the filtered simulation. As the terminal is always inside the coverage area of at least one cell, it gives the most accurate estimation of the SFN gain with different parameter values. Furthermore, the method provides a reliable means to locate the MS inside the network area according to the uniform distribution.

242 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 446 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE The interference link proportion can be obtained for each case by calculating the potential interference links over the total links. It gives a rough idea about the severity of the exceeding of the SFN limit, with a value range of 0-100% (from non-interfering network up to interfered network where all the transmitters are a potential source of interference). 4. Results The Tables 3 and 4 summarises the C/I link analysis for the different reuse pattern sizes and for FFT and GI with the selected radio parameter values. Fig. 6. The results in filtered area. This principle is used in the simulations. The network was dimensioned in such way that the area location probability is 70% in the cell edge. The dimensioning can be made by taking into account the characteristics of long-term fading. The Figure 7 shows a snap-shot type example in filtered area with the C/N values resulting less than 8.5 db, i.e. the limit for the respective parameter settings of QPSK cases. Table 3. The C/I link analysis for QPSK cases. Reuse pattern size (K) FFT,GI K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ Fig. 7. An example of the distribution of the simulation results that yields C/N < 8.5 db. As a verification of the geographical interference class, a study that can be called a C/I-link analysis was carried out. It is a method to revise all the combinations (hash) of the distances between each pair of sites (TX1-TX2, TX1-TX3, TX2-TX1, TX2-TX3 etc.) marking the link useful (C) if the guard distance between the respective sites is less than the maximum allowed SFN diameter (D SFN ). If the link is longer, it is marked as a potential source of interference (I). Table 4. The C/I link analysis for 16-QAM. Reuse pattern size (K) FFT,GI K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ The C/I link analysis shows that in case of large network (21 cells in the SFN area), the only reasonable parameter set for the QPSK modulation seems to be FFT=8K and GI=1/4. This is due to the fact that QPSK provides with the largest cell sizes (with the investigated parameter set the r=7.5 km). The cell size of the investigated 16-QAM case is smaller (r=5.0 km)

243 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 446 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE which provides the use of the parameter set of (FFT = 8K, GI = 1/4), (FFT = 4K, GI = 1/4) and (FFT = 8K, GI = 1/8). The interference distance r interference =13.5 km is the same in all the cases as the interference affects until it reaches the reference level (the sum of noise floor and terminal noise figure). The C/I link investigation gives thus a rough idea about the most feasible parameter settings. In order to obtain the information about the complete performance of DVB-H, the combination of the SFN gain and SER level should be investigated. The Figures 8 and 9 shows examples of two extreme cases of the simulations, i.e. PDF of noninterfered and completely interfered situation. As for all of the simulation cases, a total of 60,000 simulation rounds were performed. The figures shows the PDF, i.e. occurred amount of samples per C/N and C/I in scale of 0-50 db, with 0.1 db resolution. # of samples PDF, C/N, QPSK, K=7, FFT=8K, GI=1/4 C distribution I distribution The PDF gives a visual indication about the general quality of the network. Nevertheless, in order to obtain the exact values of the performance indicators, a cumulative presentation is needed. The following Figure 10 shows an example of the CDF in the noninterfering QPSK network with the reuse pattern size as a variable CDF, C/N, QPSK, K=1...21, FFT=8k, GI=1/ C/N (db) Fig. 10. The CDF of C/N in non-interfered network for reuse pattern sizes of The Figure 11 shows an amplified view to the breaking point, i.e. the 10% outage probability criteria C/N (db) Fig. 8. An example of the C/N distribution in non-interfered SFN network. # of samples PDF, C/(N+I), 16-QAM, K=7, FFT=2K, GI=1/32 C distribution I distribution C/(N+I) (db) Fig. 9. An example of C/N and I/N in interfered case CDF K= K= K= K= K= K=12 K=16 K=19 K= C/(N+I) (db) Fig. 11. An amplified view of the example of the simulation results for QPSK network As can be seen form the Figure 11, the single cell (K=1) results a minimum of 8.5 db for the 10 % outage probability, i.e. for the area location probability of 90% in the whole cell area which corresponds to the 70% area location probability in the cell edge. The cell is thus correctly dimensioned for the simulations.

244 THE FOURTH INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMMUNICATIONS PAPER NUMBER 446 ICWMC 2008, JULY 27 - AUGUST 1, ATHENS, GREECE In order to find the respective SFN gain level, the comparison with single cell and other reuse pattern sizes was made in the 10% outage point. The Figures shows the simulation results for the reuse pattern sizes SFN gain (db) SFN gain, QPSK GI1/4,8K GI1/4,4K GI1/4,2K GI1/8,2K GI1/ 16,2 K GI1/32,2K Fig. 12. The SFN gain levels for QPSK cases, with the reuse pattern sizes of SFN gain (db) K SFN gain, 16-QAM GI1/4,8K GI1/4,4K GI1/4,2K GI1/8,2K GI1/ 16,2 K GI1/32,2K Fig. 13. The SFN gain levels for 16-QAM cases, with the reuse pattern sizes of The simulation results show the level of SFN gain. The reference case (FFT 8K and GI ¼) results the maximum gain for the reuse pattern sizes of K=1-21 for both QPSK and 16-QAM and provides a noninterfering network. In addition, the parameter set of (FFT = 4K, GI = 1/4) and (FFT = 8K, GI = 1/8) results a network where SFN errors can be compensated with the SFN gain. According to the results shown in Figure 12, the QPSK case could provide SFN gain of 3-4 db in noninterfering network. It is interesting to note that in the interfering cases, also the parameter set of (GI = ¼, FFT = 4K corresponding FFT = 8K, GI = 1/8) results K positive SFN gain even with the interference present for all the reuse pattern cases up to 21. Also the parameter set of (GI = ¼, FFT = 2K, corresponding FFT = 8K, GI = 1/16 and FFT = 4K, GI = 1/8) provides an adequate quality level until reuse pattern of 16 although the error level (SER) increases. According to the Figure 13, the 16-QAM gives equal SFN gain, resulting about 3-4 db in noninterfering network. For the (FFT = 4K, GI = ¼) and the corresponding parameter set of (FFT = 8K, GI = 1/8), the SFN gain is higher than the SER even with higher reuse pattern sizes compared to QPSK, because the 16-QAM cell size is smaller. As can be seen from the Figures and from the C/I link analysis of Tables 3-4, the rest of the cases are useless with the selected parameter set. 5. Conclusions The simulation method presented in this paper can be used in the in-depth optimisation of the DVB-H. The results show the SFN gain in non-interfered as well as in the over-sized and interfered SFN network. Especially in the interfered SFN network, the correct selection of the radio parameters is essential. The results show that the interference level can be compensated with the SFN gain, and sufficiently good quality of service (e.g. the level that would be achieved by using only one site without SFN gain) can thus be obtained by selecting correctly the radio parameters. 6. References [1] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [2] Jukka Henriksson. DVB-H standard, principles and services. HUT seminar T Helsinki, 24 Feb Presentation material. 53 p. [3] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [4] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. IEEE p. [5] William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp

245 V Publication V DVB-H coverage estimation in highly populated urban area This paper compares the DVB-H coverage area estimations via the Okumura-Hata model and a typical radio network planning program that is based on the Okumura-Hata propagation model with practical correction factors. The investigation shows the coverage estimations in the urban area of Mexico City. 58th Annual IEEE Broadcast Symposium October Alexandria, VA, USA 2008 IEEE Computer Society Press Reprinted with permission.

246

247 DVB-H Coverage Estimation in Highly Populated Urban Area Jyrki T.J. Penttinen Nokia Siemens Networks, Spain Abstract The methodology for the sufficiently accurate DVB-H coverage estimation is one of the most important topics in the network planning of the system. The final coverage area depends on the planned capacity, i.e. the bit stream per channel, as well as the number of the channels. Also the surrounding geographical area has a big impact on the coverage, as well as the antenna height. This paper shows a case analysis about the coverage estimation as a basis for the initial radio network planning in large urban environment. The study was carried out for Mexico City, which is one of the most populated areas in the world. Keywords DVB-H, radio network planning. INTRODUCTION The coverage area of a single DVB-H transmitter site depends on the provided capacity and thus on the radio parameters that provides the required bit pipe. Logically, higher total capacity demand in the given band width results smaller cell sizes. The environment has a big impact on the radio wave propagation. According to the Okumura-Hata [8] prediction model, the dense urban area attenuates the signal considerably compared to the other environment types. This paper shows an example of the coverage planning in Mexico City, which is considered as one of the largest urban and dense urban environments. Despite of the challenges of the presented environment type, the definite advantage of DVB-H can be seen most clearly in this case as big amount of potential customers can be served in reduced area. In Mexico City alone, the estimated total population is about million from which the number of the potential customers can be estimated to be considerable. The study presented in this paper compares the accuracy and usability of the basic Okumura-Hata model and more indepth coverage planning tool with respective terrain height and area type information included. THE PLANNED ENVIRONMENT Mexico City is situated in 1.4 miles (2.2 km) of height from the sea level, and it is surrounded by the mountains with about 1.9 miles (3 km) of height compared to the sea level. Figure 1 shows an overview of the city. As can be observed, the area consists of urban buildings in large area. Figure 1. An overview of Mexico City, near the center area. The city is in general tightly built and large in size. The following Figure 2 shows the area cluster type of Mexico City. As can be seen, the dense urban and urban type is very large with the respective cluster type proportion of roughly 420 square miles (1000 km 2 ). residential forest dense urban urban open km Figure 2. The area type, i.e. cluster map of Mexico City. This map shows the area of about miles (50 50 km) from which half is urban and dense urban area.

248 THEORETICAL COVERAGE ESTIMATION The assumption for the analysis was to use 16-QAM modulation, which provides about two times more channel capacity compared to QPSK. On the other hand, the link budget of 16-QAM produces about 6-7 db smaller coverage than QPSK. For the sufficiently robust channel coding and error recovery, the Code Rate (CR) of ½ and MPE-FEC rate of ¾ was selected. With 16-QAM modulation and 17.5 db C/N requirement, this results a total channel capacity of 6.2 Mb/s. Using the basic SFN network, this combination would be possible to use e.g. for 1 ESG (electrical service guide, about kb/s) and for high quality audio-video channels of about 450 kb/s each, or for about 20 good quality A/V channels of about 250 kb/s each. In the initial phase of the planning, an Okumura-Hata based rough estimation about the needed amount of sites can be carried out using the respective area correction. For the estimation of the cell size (radius), a link budget is a proper tool. In this specific case, the link budget shown in Table 1 can be created. Table 1. An example of the DVB-H link budget. General parameters Frequency f MHz Noise floor for 6 MHz bandwidth P n dbm RX noise figure F 5.2 db TX Transmitter output power P TX W Transmitter output power P TX 63.8 dbm Cable and connector loss L cc 3.0 db Power splitter loss L ps 3.0 db Antenna gain G TX 13.1 dbi Antenna gain G TX 11.0 dbd Eff. Isotropic radiating power EIRP 70.9 dbm EIRP W Eff. Radiating power ERP 68.8 dbm ERP W RX Min C/N for the used mode (C/N) min 17.5 db Sensitivity P RXmin dbm Antenna gain, isotropic ref G RX -7.3 dbi Antenna gain, 1/2 wavelength dipole G RX -5.2 dbd Isotropic power P i dbm Location variation for 95% area prob L lv 5.3 db Building loss L b 14.0 db GSM filter loss L GSM 0.0 db Min required received power outdoors P min(out) dbm Min required received power indoors P min(in) dbm Min required field strength outdoors E min(out) 62.8 dbuv/m Min required field strength indoors E min(in) 76.8 dbuv/m Maximum path loss, outdoors L pl(out) db Maximum path loss, indoors L pl(in) db According to the link budget, the outdoor reception of this specific case with 2,400 W transmitter yields a successful reception when the radio path loss is equal or less than db. The Okumura-Hata model [8] can be applied in order to obtain the cell radius (unit in kilometers) in large city type. L( db) = lg( f ) 13.82lg( h + [ lg( h )] lg( d) a( h a( h ) MS LC1 ) MS LC 2 d = 10 L( db) BS [ ( h )] = 8.29 lg 1.54 MS BS ) a( h 2 [ ( h )] 4.97, f 400MHz = 3.2 lg MS [ lg( f ) lg( h ) a( h ) ] lg( hbs ) BS MS i, f MS ) 200MHz The following Figure 3 presents the estimated cell range calculated with the large city model and by varying the transmitter antenna height and power level. d (km) Cell radius Antenna height (m) i P=1500W P=2400W P=3400W P=4700W Figure 3. The cell range calculated with the Okumura- Hata model for the large city, varying the transmitter power levels. According to the Figure 3, it is clear that higher the antenna is located and higher the transmitter power is, lower is the transmitter site number. In practice, though, it is not always possible to obtain the site locations and antenna heights in the technically best locations. The access for already existing towers might be limited as well as the available heights in the towers, and rooftops might be challenging to obtain. In many cases, the antenna height is limited to feet (20 30 m). In some cases, it could be possible to obtain a higher antenna location in broadcast tower near the city center, or even better, in skyscraper s rooftop in the downtown area. On the other hand, the transmitter type is important to select correctly. In case of high-power transmitter type, the respective power consumption increases. It can be estimated 200

249 that the power consumption might be about 6 times the produced power level fed to the antenna cable. Also the complexity rises among the higher power levels, and e.g. liquid cooling is needed instead of air cooling, affecting on the maintenance. These might be limitations for the highest power classes when selecting the optimal power levels. In this analysis, a 2,400 W transmitter type was selected due to the above mentioned reasons. According to the Figure 3, it provides a cell radius of miles (3 4 km) with the transmitter antenna installed to feet (20 40 m) of height. As a comparison, the antenna height of 330 feet (100 m) provides about 4.4 miles (7 km) cell radius, and 660 feet (200 m) antenna yields about miles (10 11 km) radius. If possible to install, one or two high antenna locations would provide a good basic coverage in the city area whilst the sites with lower antenna heights fills the rest of the area. If only the urban and dense urban areas are to be covered, the following theoretical coverage map can be created by selecting 2 sites with antenna height of 660 feet (200 m) and cell radius of 6.8 miles (11 km), and the rest of the sites could use antenna heights of 100 feet (30 m) which provides a cell radius of about 2.2 miles (3.5 km). The map represents the Okumura-Hata estimated coverage for the outdoor assuming the sites can be selected without restrictions. km Figure 4. Theoretical coverage plan when using Okumura-Hata model for the cell radius estimation. The drawback of the above presented coverage plan is that the prediction is inaccurate depending on the actual terrain type. Nevertheless, it gives an idea about the rough cell number. In any case, this analysis shows the importance of the antenna height as with only two high antenna locations, it looks possible to cover about half of the given area whilst the low antenna locations results a need of about 15 sites. PLANNING TOOL ANALYSIS In order to compare the theoretical Okumura-Hata approach with the more realistic methods, Nokia NetAct Planner was used as a basis for more in-depth analysis of the coverage planning. The tool consists of the digital maps of Mexico City, with respective clutter data. A total of 7 sites were selected for the analysis based on the practical site considerations, i.e. the selected sites could possibly be real candidates with realistic antenna heights. The NetAct consists of several propagation models. Extended Okumura-Hata prediction model with respective digital cluster maps was used in the analysis. The initial parameter tuning for the DVB-H plan was made based on the estimated local clutter attenuation factors. The final clutter values and other propagation model parameterization for the estimated coverage area should be adjusted by carrying out field tests as each area type differs from the others. The coverage map was plotted for outdoor and indoor environments based on the previously used DVB-H link budget, taking into account the relevant parameters (bandwidth, modulation scheme, code rate, MPE-FEC rate and receiver antenna gain). The same 2,400 W transmitter type was used in all the sites like in previous analysis, with EIRP of about dbm, depending on the site configuration, i.e. cable lengths and losses. In this analysis, 2 or 3 directional antennas with the horizontal beam width of 65 degrees and vertical beam of 27 degrees was used, with respective antenna gain of 13.1 dbi. In some cases of high antenna installations, a slight antenna element downtilting was used in order to optimize the coverage. Other essential global parameters for the link budget were: Code rate (CR) ½, MPE-FEC rate ¾, radio channel TU6 (with 30 Hz Doppler), channel bandwidth 6 MHz and 680 MHz operating frequency. The average building penetration loss was estimated to be 14 db. With the area location probability of 95 %, the link budget yields a minimum requirement of dbm for the received power level in this specific case. The following Table 2 presents the selected sites with the antenna height h ant, direction (degrees), down-tilt (DT) and the final EIRP (dbm) values. The EIRP shown in the Table includes the transmitter filter, cable, connector and power splitter loss. Table 2. The DVB-H site configuration. Site TX P h ant deg DT EIRP Tres Padres 2400 W 60 m 140/220 2/ WTC 2400 W 190 m 0/150/240 2/2/ Iztapalapa 2400 W 30 m 0/120/240 0/0/ Santa Fe 2400 W 20 m 0/120/240 0/0/ Tlalpan 2400 W 20 m 330/90 0/ Vallejo 2400 W 30 m 0/120/240 0/0/ Azteca 2400 W 30 m 120/240 0/0 71.1

250 The coverage plots indicates the functional areas for 16- QAM in outdoor and indoor environments. As a reference, also QPSK outdoor coverage is presented, with the colors shown in the following Figure 5. In the following coverage maps, the raster size is miles (5 5 km). Outdoor QPSK coverage Okumura-Hata is not valid in this case as the radius is more than 20 km, but e.g. ITU-R P.1546 model yields 9.4 miles / 15 km with the parameter values of this example which is in align with the estimation shown in the Figure 6. EIRP: 70.5 dbm Ant: 140/220 deg, ~3km above sea level Downtilt: 2/2/2 deg Out door 16-QAM coverage Indoor 16-QAM coverage Figure 5. The meaning of the plotted colors. Two possibilities were identified for the high antenna installation; the WTC (skyscraper) and Tres Padres (mountain). The WTC site (with 623 feet / 190 m antenna height) provides a good basic coverage with more than 6.2 miles / 10 km radius (NetAct) in main beam, or 5.5 miles / 8.7 km (Okumura-Hata). It is worth noting, though, that the used prediction model does not take into account the variations of the obstacles like detailed building heights, so there might be holes in the presented map especially in the street canyons. EIRP: 69.3 dbm Ant: 0/150/240 deg, 190 m Downtilt: 2/2/2 deg Figure 6. Tres padres site. Tlalpan was planned with 2 sectors in a tower (antenna in 65 feet / 20 m), which provides only local coverage. The outdoor coverage is about 1.9 miles / 3 km (1.8 miles / 2.8 km via Okumura-Hata) of radius in the main beam of the antennas. EIRP: 71.3 dbm Ant: 330/90 deg, 20 m Downtilt: 0/0 deg Figure 5. WTC site. The other high antenna site, Tres Padres, was planned with 2 sectors pointing south-east and south-west in high mountain tower (about 1.9 miles / 3 km from sea level, whilst the average value of the city area is about 1.4 miles / 2.2 km). It provides a large basic coverage. Depending on the obstacles in LOS and observed direction, the outdoor coverage varies within miles / 5 25 km of radius. Figure 7. Tlalpan site Iztalapa was planned with 3 sectors in a tower (antenna height 100 feet / 30 m). It provides only a local coverage. Depending on the obstacles in LOS, the outdoor coverage is

251 about 2.5 miles / 4 km (1.9 miles / 3.1 km via Okumura- Hata) of radius in main beams. EIRP: 69.3 dbm Ant: 0/120/240 deg, 30 m Downtilt: 0/0/0 deg 1.6 miles / 2.5 km (1.6 miles / 2.5 km via Okumura-Hata) of radius. The challenge of this site is the variations of the mountain heights nearby. EIRP: 69.5 dbm Ant: 0/120/240 deg, 20 m Downtilt: 0/0/0 deg Figure 8. Iztalapa site. Vallejo contains 3 sectors (antenna height 100 feet / 30 m). It provides a local coverage due to the low antenna height, but nevertheless, it increases the indoor coverage in Tres padres sector. Depending on the obstacles in main beams, the outdoor coverage is about 2.1 miles / 4 km (1.9 miles / 3.1 km via Okumura-Hata) of radius. Figure 10. Santa Fe site. Azteca with 2 sectors (antenna height 65 feet / 20 m) provides relatively good local coverage in the directions without obstacles. Depending on the obstacles, the outdoor coverage varies in range of miles / 3 10 km (2.2 miles / 3.6 km via Okumura-Hata) of radius. EIRP: 69.3 dbm Ant: 0/120/240 deg, 30 m Downtilt: 0/0/0 deg EIRP: 71.1 dbm Ant: 120/240 deg, 30 m Downtilt: 0/0/0 deg Figure 9. Vallejo site. Santa Fe with 3 sectors in a tower (antenna height 65 feet / 20 m) provides relatively small local coverage for the low antenna height and non-uniform terrain. Depending on the obstacles in man beams, the outdoor coverage is about Figure 11. Azteca site. The following Figure 12 presents the complete network coverage with the 7 sites. The plot shows the network coverage with 16-QAM and 95 % area location probability, code rate of ½ and MPE-FEC of 2/3.

252 Figure 12. The complete network coverage prediction. RESULTS The cell size estimation obtained by calculating with pure Okumura-Hata prediction model has relatively good average correlation with the results that can be obtained with NetAct Planner when the general limits of the models are taken into account. The results of the NetAct Planner shows that the used prediction takes well into account the terrain heights and cluster types, which would be challenging to do with using only theoretical Okumura-Hata approach. When sufficiently good line of sight is found in the planned sector, the useful cell size may be considerably better that obtained with the use of Okumura-Hata. The number of identified site number was relatively low in the NetAct analysis. More sites are obviously needed if the same area should be covered as shown in Okumura-Hata analysis in Figure 4. By observing the Figure 12, about 4 6 additional sites might be necessary for the full coverage. Okumura-Hata estimated well the relatively low antenna installation sites, but the model estimated the coverage area of the high WTC site in pessimistic way which affects on the final estimation of the sites. It is worth noting that especially the indoor coverage in selected areas requires the use of repeater type of solution, e.g. in shopping centers and other centralized locations, where the potential customers are typically using the service. CONCLUSIONS The results of the case analysis shows that the theoretical Okumura-Hata prediction model with DVB-H link budget gives a good first-hand estimate about the cell sizes and thus about the needed amount of the sites in the planned area. Taking into account the characteristics of the model, this method can be applied especially in the initial phase of the network planning. Due to the restrictions of the Okumura-Hata ranges as the antenna height and maximum estimated cell radius are considered, the methodology applies for the relatively decent radiating power levels. When the cell radius exceeds the maximum predictable value of 12.4 miles / 20 km, as the case is for the Tres Padres site, the model is not feasible and adjusted models should thus be used. Especially for the high antenna locations, one of the most logical models at the moment is the ITU recommendation P.1546, which is based on the curve mapping and is valid practically for all the environments where DVB-H can be constructed. On the other side, the power levels are limited due to the EMC and human exposure regulation resulting sufficiently small cell ranges in order to be estimated with Okumura-Hata in mayor part of the cases in urban areas. The advanced planning tool with respective digital maps including the terrain height and correct cluster attenuation information is essential in the detailed network planning. It is also worth noting that the predictions presented in this paper gives indication only about the coverage areas. Especially in the case of Single Frequency Network, the correct balancing of the FFT size and Guard Interval values is important in order to avoid too high level of the possible inter-symbol interferences in large single frequency network areas. The results shows that the coverage estimation presented in this paper can be used in the first phase of the DVB-H radio network planning for the initial estimation of the transmitter sites. As the clutter types vary in practice, the more detailed prediction estimations with respective model tuning via the field tests are thus needed in the following phases. REFERENCES [1] Limits of Human Exposure to Radiofrequency Electromagnetic Fields in the Frequency Range from 3 khz to 399 GHz. Safety Code 6. Environmental Health Directorate, Health Protection Branch. Publication 99-EHD-237. Minister of Public Works and Government Services, Canada ISBN p. [2] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [3] Jukka Henriksson. DVB-H standard, principles and services. HUT seminar T Helsinki, Presentation material. 53 p [4] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [5] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB- H: Digital Broadcast Services to Handheld Devices. IEEE p. [6] William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp [7] Myron D. Fanton. Analysis of Antenna Beam-tilt and Broadcast Coverage. ERI Technical Series, Vol 6, April p. [8] Masaharu Hata. Empirical Formula for Propagation Loss in Land Mobile Radio Services. IEEE Transactions on Vehicular Technology, Vol. VT-29, No. 3, August p. [9] Recommendation ITU-R P Method for point-to-area predictions for terrestrial services in the frequency range 30 MHz to 3000 MHz p.

253 VI Publication VI DVB-H Radiation Aspects This paper shows a method to estimate the safety distances for both human exposure and EMC in typical installation environments of the DVB-H antennas. IEEE International Symposium on Wireless Communication Systems 2008 (ISWCS 08) October Reykjavik, Iceland 2008 IEEE Computer Society Press Reprinted with permission.

254

255 DVB-H Radiation Aspects Jyrki T.J. Penttinen Nokia Siemens Networks Camino del Cerro de los Gamos 1, Pozuelo de Alarcón, Spain Abstract The typical radiating power levels of the DVB-H sites are normally between the values used in mobile and television broadcast networks. It is thus important to take into account accordingly the EMC (Electro Magnetic Compatibility) as well as the human exposure limits in the DVB-H transmission facilities. This paper presents a practical method for estimating the safety limits of human exposure and EMC in different DVB- H installation environments. Furthermore, case studies are presented for the most typical environments. Index Terms DVB-H, planning, optimisation, radiation, safety I. INTRODUCTION HE radiation of the DVB-H transmitter sites can be T assumed to be higher than in mobile networks. As there are normally mobile or broadcast systems installed in the same site as DVB-H, it is essential to dimension the safety distances in order to minimise the risk of inter-system interferences. On the other hand, the human exposure limits must be calculated for both technical installation personnel and for the public. The DVB-H transmitter site antennas can be located to the telecom or broadcast tower as well as on the rooftop or indoors of the buildings. The power level should be adjusted accordingly in order to avoid any interferences or safety zone extensions. This paper presents the methodology to estimate the safety distances in typical environments. II. SAFETY ASPECTS The DVB-H radio transmission is defined to the frequency range of MHz (UHF IV and V) with a channel bandwidth of 5, 6, 7 or 8 MHz. The channels can be divided into several sub-channels as the used bit stream is relatively low. DVB-H can also be deployed to VHF III and L bands. The power level of DVB-H depends on the transmitter manufacturer solutions. Typically, the power amplifiers can produce around W (output power to the feeder), although the maximum power level might be limited to few thousands of watts in practical solutions. DVB-H radiation is non-ionising like in case of mobile network technologies. The radiation does not thus alter the human cell structure as can happen in ionising systems like x- ray equipment. Nevertheless, a sufficiently high non-ionising radio transmission power can increase the cell temperature. The radiation in given distance can be estimated with various propagation models. The simplest one giving the maximum values is the far-field attenuation in free space: ( d ) 20log( f ) L = log + (1) The safety distance of the DVB-H antennas must be assured in the deployment of the system. A simple but practical method can be created based on the fact that the antenna is located to the tower or rooftop, and that the most meaningful area to be investigated is found below the antenna. The DVB-H antenna solution can be based on omniradiating poles or directional antennas. The Figure 1 shows an example of the far-field horizontal and vertical radiation patterns of the directional antenna used in DVB-H. As can be observed from the Figure, in the practical installation environments, the vertical radiation pattern is the most meaningful when estimating the field strength and safety zones Fig. 1. Example of the horizontal plane of the antenna radiation pattern. The 3 db attenuation determinates the beam width. In this specific case, the beam width is about 60 degrees. The second pattern shows an example of the vertical plane of the directional antenna radiation. In this case, the beam width in 3 db attenuation points is about +/- 14 degrees, i.e. 28 degrees. In order to obtain the safety limits below the DVB-H antenna, a loss analysis with different angles from the antenna can be done. Having the vertical antenna pattern, respective coordinate system and linear scale for the antenna radiation attenuation as shown in Figure 2, an antenna attenuation table can be created with the scale from 270 to 180 degrees (back lobe) and from 180 to 90 degrees (main beam). 180 degree vertical angle means the point below the antenna. In this specific example, the following Table I can be obtained from the respective antenna data. The 90 degree angle represents the main beam of the antenna with 0 db loss. Please note the difference with the conventional marking of the angles, as normally 0 degree elevation represents the horizon and -90 degrees the point below the antenna.

256 deg 90 deg db res deg Fig. 2. An example of the vertical pattern in linear scale with 3 db resolution. TABLE I EXAMPLE OF THE ANTENNA ATTENUATION VALUES (IN DB). Deg Av Deg Av Deg Av The values of the table can now be analysed in two phases. The first task is to obtain the minimum attenuation value for the back and side lobes, i.e. in beam angle of degrees. In this example it is 20.9 db. As also the minimum attenuation of the horizontal pattern falls into this value, it can be used as a reference for the complete back side (hemisphere) of the antenna. The vertical angle between degrees is used for the analysis of the main lobe. The angle is the independent variable used to obtain the respective safety distances. The next step is to calculate the safety distance for the radiation power, which is the result of the DVB-H transmitter power, transmitter filter, cable, connector and jumper losses, possible power splitter loss for multiple antenna arrays and the radiation pattern loss in function of the elevation angle. The safety distance depends on the regulatory decisions. As a common type of regulations, the reference [1] can be used in order to obtain the respective formulas. The immediate area close to the antenna is the near-field region. Most of the electromagnetic energy in this region is stored instead of radiating. The field has considerable variations within this zone, making the field estimation extremely challenging. Further away from the antenna, the reactive near-field decreases and the radiating field becomes predominating as a function of the distance until the far-field zone finally stabilises the characteristics of the radiation making the calculation of the field strength reliable. The dimensions of the radiating antenna have impact on the minimum distance where the far-field starts dominating. Assuming the variable D indicates the largest dimension of the antenna and if λ is the wave length of the observed signal, the following formula can be used for the calculation of the minimum far-field limits in case of large antennas (i.e. if the greatest dimension of the antenna is much more than the wave length), according to [1]: d far field = 0.5D λ 2 (2) For the small antennas, the following formula can be used: λ d = (3) far field 2π The next calculations are valid for the far-field, in frequency range of MHz, which falls into the operating frequency range of DVB-H. In this specific case, the maximum power density for the general public in the DVB-H frequency range can be obtained by the following formula: f (MHz) W = (4) 150 Now, the power density in the far-field region can be obtained by the following formula: EIRP PG W = = (5) 2 2 4πr 4πr In the formula, EIRP is the effective isotropic radiated power (W), r is the distance from the radiating antenna (m), P is the power fed to the antenna, and G is the antenna gain compared to the isotropic antenna (10log 10 [G] in dbi). It is now possible to estimate the safety distance in front of the DVB-H antenna. Investigating Table I, it is possible to estimate the safety distances also in different sides of the antenna. Let s select the 700 MHz frequency and transmitter power of 1500 W after the filter loss. The additional cable, connector and jumper loss (L) can be estimated to total of 3 db. Using the antenna pattern presented in Figure 2 assuming its gain (G) is dbi and the physical dimensions are mm, the power density and EIRP values are: 700MHz W = = 4.67W / m (6) EIRP( W ) = P kW G L 10 = 1500W The minimum safety distance in main beam is thus: EIRP r = 4 W π 17.2m kW = W / m π 0.5 (7) (8)

257 The wavelength of the used frequency is, when c is the speed of light: c 300,000km / s λ = = = 0.43m (9) 6 f Hz Let s revise the minimum distance for the far-field in order to make sure the calculation is valid. As the extreme dimension of the antenna is more than wave length (1m), the antenna is considered large, and the far-field distance limit is thus: α = deg: Area to be avoided in back beam, r=1.6 m for all the values of α α = deg: Area to be avoided in main beam, r=f(α) according to Table II d 0.5D = λ 0.5 (1m ) = 0.43m 2 2 far field 1. 2 m (10) The calculation is valid as the value is in far-field zone. For the radiation angles of degrees, the following Table can be created. TABLE II EXAMPLE OF THE SAFETY DISTANCE VALUES FOR THE ANGLES DEGREES WITH THE RESPECTIVE VERTICAL PATTERN ATTENUATION. Deg Av r min Deg Av r min Fig. 4. An experimental antenna installation. With the values presented in the example, the vertical safety distance behind the antenna is 1.6 m. In case of the Figure 4, using a 2 meter high antenna pole, the safety distance below the antenna is thus achieved for the personnel inside the building. As the antenna is in the edge of the roof top, the main beam is also secured. In addition, the roof material provides with additional attenuation for the indoor. The Figure 5 shows and extended analysis with the same radio parameters as previously but varying the transmitter power level between 100 and 9000 W. For the vertical back and side lobes, i.e. for the vertical radiation angles of α = degrees, the common value of 20.9 db was obtained from Table I. This value corresponds the minimum safety distance of 1.6 meters for the whole range of the above mentioned angles. As can be observed from the Tables I and II, the relatively narrow vertical beam width provides reduced safety distances outside of the main lobe. The Figure 3 clarifies the idea of the safety distances in graphical format. Safety distance r (m) Safety distance per TX pow er r_9000w r_100w Distance (m) r_safety Angle (deg) Fig. 3. An example of the safety distances in function of the angle of the vertical antenna radiation pattern Vertical radiation pattern angle (deg) Fig. 5. The safety distances for a set of transmitter power levels presented in logarithmic scale, with the parameter setting of the presented example. As can be observed, lower the power level is and closer to the back side of the antenna, many of the calculated points falls below the minimum far-field distance of 1.2 m. As the calculation is not accurate in those near-field spots, the 1.2 m limit can be considered as a minimum limit for all the cases in near-field. In case of the Figure 4, the safety limits on the rooftop should be marked accordingly in case of e.g. the maintenance personnel closing the antenna. The back side of the antenna can be marked as round with the minimum of the calculated safety distance, with preferably extra margin as the antenna

258 pole can affect on the final radiation pattern of the antenna (i.e. the pole might act as a part of the antenna elements). For the main beam, the horizontal radiation pattern must be taken into account by calculating the safety limits on the sides of the antenna. A mask with sufficient additional margin can be used e.g. by observing the reference attenuation points of 20, 10 and 3 db and the respective angles. In practice, it is important to assure the personnel can not cross the main beam accidentally as the full EIRP might mean considerable safety distances in front of the beam. the same vertical line of the tower. As can be observed form the Figure 2, the vertical pattern in this case example is symmetrical above and below the antenna element. The attenuation value around 180 degrees (below the antenna) as well as 0 degrees (above) is thus 20.9 db as shown in Table I. In order to calculate the interfering electrical field, the following formula can be used: 1 E i = 2r ηp π (11) Rooft op Fig. 6. In case of the rooftop installation, the respective safety distance on sides of the antenna should be calculated according to the horizontal radiation pattern of the antenna. III. EMC LIMITS When the DVB-H antenna is installed in the tower, there might be antennas of the other telecommunications systems on top and below of the DVB-H antenna. The EMC calculations should thus take into account mainly the upper and lower side lobes of the DVB-H antenna. The analysis methodology presented in the previous chapter can be used also for the EMC calculations. The following Figure 7 shows a principle of the EMC in the towers. The DVB-H antennas create a certain electro magnetic field strength around the antenna, which might be considered as an interference field for the antennas of the other systems nearby. Fig. 7. When using a directional antenna, the relative field strength above and below the antenna depends on the attenuation of the vertical radiation pattern. Like in previous case, the vertical radiation pattern can be used as a basis for the EMC field estimations. In fact, the vertical pattern right above and below the antenna is the most important as the antennas of the other systems are normally in The variable η is the wave impedance in air with the value of 377 Ω, P is the transmitter power (EIRP) including the gains G and attenuations A. Let s revise the electrical field for the previously presented case examples with the transmitter power levels of 1500 W in 2 meter distance of the main beam. As shown previously, the example yields a total EIRP of kw, which produces the following field in 2 meters distance in the main beam. 1 ηgp 1 E i = = 2r π 4 366V / m W π (12) The field strength reduces according to the radio propagation environment. Assuming the worst case, the free propagation loss can be assumed as shown in Formula 1. The antennas of the other systems might be relatively near of the DVB-H antennas due to the lack of the space in the tower. The essential question is thus, what is the minimum allowed distance between the antennas. The allowed field strength depends on the immunity level of each system. As an example, the commercial electrical equipment are able to handle interference field strengths of about 2-10 V/m. For the professional electrical equipment as well as e.g. mobile network elements, the requirement is considerably higher. In addition to the pure field strength, also the frequency is essential. As an example, the GSM 850 and GSM 900 might suffer from the distortion caused by the harmonic components of the DVB-H transmission in the radio spectrum whilst GSM 1800/1900 is safe due to the frequency difference and filtering of the equipment. For the situation with the antennas located on top of each others, the attenuation value of the vertical radiation pattern of the DVB-H should be taken into account. The value for the EMC calculations can be estimated by observing the 180 degree angle of the vertical radiation pattern. In practice, the value might need additional margin in order to take into account the possible variations e.g. for the pattern inequality around 180 degrees, and the possible side-effect of the tower itself which can interfere the radiation pattern. The effect of the radiation pattern can now be extended in the following

259 way. P ( dbm) = P( dbm) + G( db) A ( db) (13) tot Ptot (dbm) P tot ( W ) = (14) η Ptot ( W ) Ei = 2r π v (15) The A v represents the attenuation (db) in the observed vertical direction; in this case, the value is 20.9 db both right above as well as below the antenna as can be seen in Table I ,0 E (V/m) 1000,0 100,0 10,0 EMC, main beam ,5 5,0 7,5 10,0 12,5 15,0 17,5 20,0 r (m) Fig. 8. The electrical interference field in the main beam of DVB-H antenna with the transmitter power levels of W (cooresponding EIRP of kw). 1000,0 E (V/m) 100,0 10,0 1,0 EMC, vertical 180 deg and 0 deg ,5 5,0 7,5 10,0 12,5 15,0 17,5 20,0 r (m) Fig. 9. The electrical interference field above and below the typical directional DVB-H antenna with the transmitter power levels of W (corresponding EIRP of W). The severity of the interfering field depends on the frequency, i.e. how considerable the frequency separation is. The effect of the main beam should be observed especially when the DVB-H antenna faces the antenna of another system. The EMC safety distance is sufficient in most of the installation cases as there should be sufficiently open area in front of the antenna, but there might be situations e.g. with antennas located on rooftops in both sides of a street. It is thus important to make sure that the facing antennas (e.g. DVB-H antenna on rooftop and GSM 850 antenna on the other side) are producing sufficiently low field within the respective distance. The field in the main beam might be quite high as can be observed from the Figure 8, meaning that the mid level and highest power transmitters should be avoided in rooftops. IV. CONCLUSIONS The results of the case study shows that the installation of the directional antenna in telecom or broadcast tower, as well as on the rooftop of the building, can be done in controlled and safe way. The logical way of obtaining the safety distance limits is to observe the vertical radiation pattern of the antenna as the antenna is above the population in the tower installations. In rooftop cases, the vertical beam dictates the safety distance which should be analysed more detailed in all the angles, from the point below the antenna up to the main beam direction. The corresponding height of the antenna is relatively straightforward to design based on this information. As for the EMC, the vertical beam analysis can be used in the designing of the location of the DVB-H antenna in the telecom or broadcast tower. In multiple directional antenna installation, the final vertical radiation pattern depends on the antenna array and should be calculated or measured taking into account the power splitter loss. The EMC safety distance above and below the antenna depends on the frequency separation and on the maximum field strength that the other systems can support taking into account the respective vertical beam attenuation of the other systems. REFERENCES [1] Limits of Human Exposure to Radiofrequency Electromagnetic Fields in the Frequency Range from 3 khz to 399 GHz. Safety Code 6. Environmental Health Directorate, Health Protection Branch. Publication 99-EHD-237. Minister of Public Works and Government Services, Canada ISBN p. [2] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [3] Jukka Henriksson. DVB-H standard, principles and services. HUT seminar T Helsinki, Presentation material. 53 p [4] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [5] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. IEEE p. [6] William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp [7] Myron D. Fanton. Analysis of Antenna Beam-tilt and Broadcast Coverage. ERI Technical Series, Vol 6, April p. [8] Myron D. Fanton. Analysis of Antenna Beam-tilt and Broadcast Coverage. Electronics Research, Inc. ERI Technical Series, Vol. 6, April p.

260

261 VII Publication VII DVB-H Performance Simulations in Dense Urban Area This paper shows a method to simulate inter-symbol interference levels in dense urban areas by utilizing realistic DVB-H site definitions, including site specific antenna heights and power levels. Okumura-Hata and ITU-R P models were used in the simulations. Case examples were simulated by varying the relevant DVB-H radio parameter values. Mexico City was used as a basis for the simulations. The Third International Conference on Digital Society (ICDS 2009) February 1 7, Cancun, Mexico 2009 IEEE Computer Society Press Reprinted with permission.

262

263 DVB-H Performance Simulations in Dense Urban Area Jyrki T.J. Penttinen Member, IEEE Abstract The correct estimation of the DVB-H (Digital Video Broadcast, Handheld) coverage areas is essential part in the pre-planning of the system. In addition to the coverage and respective capacity planning, the indepth work also requires the estimation of the quality of service levels, which depends mainly on the radio related parameters. This paper presents a simulation method to estimate the interference caused by the Single Frequency Network (SFN) mode of DVB-H in dense urban area type. Mexico City area type was used as a basis for the investigations and case results. 1. Introduction The link budget and theoretical radio propagation models can be used in the pre-planning and initial estimation of the DVB-H radio network size with the given capacity requirement. If the assumptions are tuned accordingly with the special characteristics of the planned area, this type of coverage estimation can be accurate. Nevertheless, these methods provide only a limited view to the practical network performance especially when the Single Frequency Network mode is used as the interference level may increase. This paper describes a method to simulate the SFN interference levels in a realistic network environment. A simulation tool was developed for the investigation of the interferences both geographically as well as by cumulative distribution of the power levels. 2. SFN limits As defined by the ODFM principles, the GI (Guard Interval) and FFT mode defines the maximum delay the DVB-H mobile terminal is able to handle in SFN mode [1]. Table 1 summarizes the maximum time delays with the respective maximum functional distances of the site when the single frequency approach is applied. Table 1. The guard interval lengths and respective safety distances. GI FFT = 2K FFT = 4K FFT = 8K µs km µs km µs km 1/ / / / The FFT size has impact on the maximum velocity of the terminal, and the GI affects on the maximum velocity of the terminal and on the radio capacity. As for the interferences only, the following parameter combinations results the same C/(N+I) performance due to their same requirement for the safety distances: {FFT 8k, GI 1/4}: only one set {FFT 8k, GI 1/8}, {FFT 4k, GI 1/4} {FFT 8k, GI 1/16}, {FFT 4k, GI 1/8}, {FFT 2k, GI 1/4} {FFT 8k, GI 1/32}, {FFT 4k, GI 1/16}, {FFT 2k,GI 1/8} {FFT 4k, GI 1/32},{FFT 2k, GI 1/16} {FFT 2k, GI 1/32}: only one set The required C/N depends on the code rate (CR), MPE-FEC rate (multi protocol encapsulator, forward error correction) and modulation. The minimum carrier per interference and noise C/(N+I) level requirement for e.g. {QPSK, CR ½, MPE-FEC ½} is 8.5 db and for {16-QAM, CR ½, MPE-FEC ½} 14.5 db [1]. 3. SFN simulator The dense and urban area of Mexico City was used as a basis for the simulations by applying suitable propagation prediction models. The city is located on relatively flat ground level with high mountains surrounding the center area which was taken into account in the radio interface modeling. The Figure 1 presents the location of the selected sites, and the Table 2 shows the site parameters.

264 40 30 km Geographical site location km Fig. 1. The site locations and informative site sizes of the simulator. Table 1. The site parameters. Site Coord, km EIRP Radius, km nr x y dbm W QPSK 16- QAM The height of the antenna was 60, 190, 30, 20, 20, 30 and 30 meters from the tower base, respectively for the sites 1-7. The site number 7 represents the mountain installation with the tower base located 800 meters above the average ground level, resulting the effective antenna height of 860 meters compared to the city center level. Site number 4 is also situated in relatively high level, but in this case, the surrounding area of the site limits its coverage area. The rest of the sites are in base ground level of Mexico City center. As the cell radius of the investigated sites is clearly smaller than 20 km, the Okumura-Hata [3] is suitable for the path loss prediction for all the other sites except for the mountain site number 7. For the latter case, ITU-R P.1546 (version 3) [2] model was applied for the simulations by interpolating the path loss for each simulation round, using antenna height of 860 meters. The frequency was set to 680 MHz. The link budget of the simulator takes into account separately the radiating power levels and antenna heights of each site as seen in Table 1. During the simulations, the receiver was placed randomly in the investigated area (45km 45km = 2025km 2 ) according to the snap-shot principle and uniform geographical distribution. In each simulation round, the separate sum of the carrier per noise and the interference per noise was calculated by converting the received power levels into absolute powers. The result gives thus information about the balance of SFN gain and SFN interference levels. Tables of geographical coordinates with the respective sum of carriers and interferences were created by repeating the simulations 60,000 times. Also carrier and interference level distribution tables were created with a scale of db. The long-term as well as Rayleigh fading was taken into account in the simulations by using respective distribution tables independently for each simulation round. A value of 5.5 db was used for the standard deviation. The area location probability in the cell edge of 90% was selected for the quality criteria, producing about 7 db shadowing margin for the long-term fading. Terminal antenna gain of -7.3 dbi was used in the calculations according to the principles indicated in [1]. Terminal noise figure of 5 db was taken into account. Both Code Rate and MPE-FEC Rate were set to ½. 4. Simulation results The usable coverage area was investigated by postprocessing the simulation results. The simulations were carried out by using QPSK and 16-QAM modulations and all the possible variations of FFT and GI. As expected, the parameter set of {QPSK, FFT 8k, GI 1/4} produces the largest coverage area practically without interferences (Figure 2). The results of this case can be considered thus as a reference for the interference point of view. When the GI and FFT values are altered, the interference level varies respectively as can be observed from the figures 2-7 (QPSK) and 8-13 (16- QAM). It can be seen that in addition to the parameter set of {FFT 8k, GI1/4}, also {FFT 8k, GI 1/8}, {FFT 4k, GI 1/4} produces useful coverage areas, i.e. the balance of the SFN gain and SFN interferences seem to be in acceptable levels, whilst the other parameter settings produces highly interfered network. The 16-QAM produces smaller coverage areas compared to the QPSK as the basic requirement for the C/(N+I) of 16-QAM is 14.5 db instead of the 8.5 of QPSK. The SFN interferences tend to cumulate to the outer boundaries of the planned coverage area. Nevertheless, in case of high interferences, the investigated parameter settings would obviously still function in the Multi Frequency Network because the adjacent site of this mode uses different frequencies.

265 Coverage, C/(N+I) Coverage, C/(N+I) 40 (C/I)<=8.5dB 8.5dB<=(C/I)<35dB (C/I)>=35dB 40 (C/I)<=8.5dB 8.5dB<=(C/I)<35dB (C/I)>=35dB km 20 km km Fig. 2. The geographical distribution of the C/(N+I) values for QPSK, FFT 8k and GI ¼, with the respective minimum limit of 8.5 db km Fig. 5. Parameter setting of QPSK, FFT 8k and GI 1/32 produces highly reduced useful coverage. Coverage, C/(N+I) Coverage, C/(N+I) 40 (C/I)<=8.5dB 8.5dB<=(C/I)<35dB (C/I)>=35dB 40 (C/I)<=8.5dB 8.5dB<=(C/I)<35dB (C/I)>=35dB km 20 km km Fig 3. The parameter setting of QPSK, FFT 8k and GI 1/8 results small outages compared to the previous case km Fig. 6. Parameter setting of QPSK, FFT 4k and GI 1/32 reduces further the useful coverage area. Coverage, C/(N+I) Coverage, C/(N+I) 40 (C/I)<=8.5dB 8.5dB<=(C/I)<35dB (C/I)>=35dB 40 (C/I)<=8.5dB 8.5dB<=(C/I)<35dB (C/I)>=35dB km 20 km km Fig. 4. The parameter setting of QPSK, FFT 8k and GI 1/16 affects on the coverage clearly due to the increased interference levels km Fig. 7. The simulation results show that the parameter setting QPSK, FFT 2k and GI 1/32 is practically useless in the planned area.

266 Coverage, C/(N+I) Coverage, C/(N+I) 40 (C/I)<=1.5dB 14.5dB<=(C/I)<35dB (C/I)>=35dB 40 (C/I)<=14.5dB 14.5dB<=(C/I)<35dB (C/I)>=35dB km 20 km km Fig. 8. The Coverage estimation for 16-QAM, FFT 8k and GI ¼. This case shows a clean noise limited coverage area km Fig QAM, FFT 8k and GI 1/32 is practically useless for the coverage area due to the high interference levels. Coverage, C/(N+I) Coverage, C/(N+I) 40 (C/I)<=14.5dB 14.5dB<=(C/I)<35dB (C/I)>=35dB 40 (C/I)<=14.5dB 14.5dB<=(C/I)<35dB (C/I)>=35dB km 20 km km Fig QAM, FFT 8k and GI 1/8 reduces the useful coverage area, but the case is still feasible km Fig QAM, FFT 4k and GI 1/32 provides a minimum coverage due to the very high interference levels. Coverage, C/(N+I) Coverage, C/(N+I) 40 (C/I)<=14.5dB 14.5dB<=(C/I)<35dB (C/I)>=35dB 40 (C/I)<=14.5dB 14.5dB<=(C/I)<35dB (C/I)>=35dB km 20 km km Fig QAM, FFT 8k and GI 1/16 provides clearly reduced coverage area due to the increased interference levels km Fig QAM, FFT 2k and GI 1/32 is useless for the coverage in the planned area due to the extremely high interference levels.

267 The following Figures summarizes the results of the interference level simulations, showing the noninterfered areas (when I is 0), lightly interfered areas (when I is between 0 and 5 db), and highly interfered areas (when I is greater than 5 db), depending on the selection of the FFT and GI values. It can be seen that the parameter setting of {FFT 8k, GI 1/4} as well as {FFT 8k, GI 1/8} and {FFT 4k, GI 1/4} produces a network without or with only few interferences. The difference between the other parameters is clear as the Figures shows a very high rise of the interference levels. The results indicate that the parameter sets of {FFT 8k, GI 1/16}, {FFT 4k, GI 1/8} and {FFT 2k, GI 1/4} increases considerably the interference level reducing the useful coverage, and the rest of the parameters are practically useless. km I levels I=0dB 0dB<I<=5dB I>5dB km Fig. 16. Parameter setting of FFT 8k and GI 1/16 interference level makes large part of the original coverage area useless. I levels I levels 40 I=0dB 0dB<I<=5dB I>5dB 40 I=0dB 0dB<I<=5dB I>5dB km 20 km km Fig. 14. The interference level simulation shows that the FFT 8k and GI ¼ provides a non-interfered network in the planned area km Fig. 17. Parameter setting of FFT 8k and GI 1/32 interferences level is considerable in major part of the network. I levels I levels 40 I=0dB 0dB<I<=5dB I>5dB 40 I=0dB 0dB<I<=5dB I>5dB km 20 km km Fig 15. the FFT 8k and GI 1/8 still provides with a clean network as the interference level is considered km Fig. 18. Parameter setting FFT 4k and GI 1/32 produces a heavy interference level almost everywhere in the original coverage area.

268 40 30 I levels I=0dB 0dB<I<=5dB I>5dB terminal (FFT 4k), or give more capacity (GI 1/8), but with the cost of useful coverage area due to the increased interference levels. As for the rest of the parameter settings, the optimal balance can not be achieved due to the very high interference levels. km Conclusions km Fig 19. The simulation shows massive interference levels in the whole planned area with FFT 2k and GI 1/ CDF, C/(N+I) QPSK, FFT 8k, GI 1/4 QPSK, FFT 8k, GI 1/8 QPSK, FFT 8k, GI 1/16 QPSK, FFT 8k, GI 1/32 QPSK, FFT 4k, GI 1/32 QPSK, FFT 2k, GI 1/ C/(N+I), db Fig. 20. The cumulative C/(N+I) distribution of different modes. The Figure 20 shows a résumé of the cumulative C/(N+I) distribution of different modes. By observing the 90% probability in the cell edge (about 95% in the cell area), i.e. 5 % outage probability of the Figure, the mode {FFT 8k, GI1/4} provides a minimum of about 9 db and the set of {FFT 8k, GI1/8} and {FFT 4k, GI1/4}gives about 5 db in the whole investigated area. It is worth noting that these values are calculated over the whole map of 45km 45km. It seems that the QPSK mode would provide a good performance in the investigated area when using the non-interfering {FFT 8k, GI 1/4} parameters, whilst 16-QAM gives smaller yet non-interfered coverage area. The advantage of the latter case is the double radio channel capacity compared to the QPSK with greater coverage area. The parameter set of {FFT 8k, GI 1/8} and {FFT 4k, GI 1/4} looks to be still useful, providing the possibility to either rise the maximum velocity of the The presented simulation method provides both geographical and cumulative distribution of the SFN gain and interference levels. The method can thus be used in the detailed optimization of the DVB-H networks. The principle of the simulator is relatively straightforward and the method can be applied by using various different programming languages. In these investigations, a standard Pascal was used for programming the core simulator. The results show that the radio parameter selection is essential in the detailed planning of the DVB-H network. As the graphical presentation of the results indicate, the effect of the parameter value selection on the interference level and thus on the quality of service can be drastic, which should be taken into account in the detailed planning of DVB-H. 6. References [1] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). EBU. 108 p. [2] Recommendation ITU-R P Method for point-to-area predictions for terrestrial services in the frequency range 30 MHz to 3000 MHz p. [3] Masaharu Hata. Empirical Formula for Propagation Loss in Land Mobile Radio Services. IEEE Transactions on Vehicular Technology, Vol. VT-29, No. 3, August pp [4] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D6 Wing TV Common field trials report. Project report, November p. [5] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Country field trial report. Project report, November p. [6] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [7] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. IEEE p.

269 VIII Publication VIII MPE-FEC Performance as function of the Terminal Speed in Typical DVB-H Radio Channels This paper shows case examples of the MPE-FEC performance in the laboratory as well as in the live network in urban and sub-urban area types. Some typical radio parameter values were varied in the laboratory environment in order to identify the indoor performance. In the live network, a set of typical urban and sub-urban indoor cases were studied. Also vehicular tests were carried out in order to identify the relationship between the MPE-FEC performance and vehicular speed. IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB) May, Bilbao Spain 2009 IEEE Computer Society Press Reprinted with permission.

270

271 mm MPE-FEC Performance in Function of the Terminal Speed in Typical DVB-H Radio Channels Jyrki T.J. Penttinen, Member, IEEE, Eric Kroon, Member, IEEE Abstract The detailed network planning of DVB-H requires understanding about the regional differences and their effects on the functionality of the system performance. A deep city center contains various radio components whereas rural area normally consists of a single line-of-sight signal resulting differences in the error behavior in the DVB-H reception. In the city area, the terminal velocity is normally limited to pedestrian and relatively slowly moving vehicular speeds. On the other hand, the Doppler shift might be the limit of DVB-H performance in the motorways across rural areas due to the higher velocities. This paper presents a radio interface quality analysis methodology based on the erroneous frames and bit error rates. A selected set of field and laboratory test cases and respective results of the performance analysis is presented. The results can be used as a basis for balancing the MPE-FEC rate and capacity of DVB-H during the related in-depth radio network optimization. Index Terms Field trials and test results, mobile broadcast, mobile TV, DVB-H, radio network planning, performance evaluation, propagation and coverage, radio network planning. I. INTRODUCTION VB-H (Digital Video Broadcasting, Handheld) is based D on the terrestrial version of DVB, i.e. DVB-T, and it has been designed especially for the moving environment. The performance of the system takes into account the variations of the received power levels and bursty interferences of the radio path. DVB-H is adequate especially for the small handsets as the typical terminal screen size requires only a fraction of the bit stream capacity compared to DVB-T which is meant for the fixed reception with the non-movable antennas located in rooftop. Unlike in case of DVB-T, the handheld version is typically used in street level which brings challenges for the radio reception due to the varying radio conditions. One of the methods of DVB-H to cope with the Doppler effects and multi-path radio signal propagation is the MPE-FEC (Multi Manuscript received April 9, This work was supported in part by Nokia Siemens Networks which provided the field measurement equipment and test laboratory for the study. Jyrki T.J. Penttinen is with Nokia Siemens Networks, Av Ronda de Europa 5, Tres Cantos, Madrid, Spain ( jyrki.penttinen@nsn.com). Eric Kroon is with Nokia Siemens Networks, Karaportti 2, Espoo, Finland ( eric.kroon@nsn.com). Protocol Encapsulator, Forward Error Correction). It is an optional feature of DVB-H and has been defined as an additional functionality compared to the original DVB-T specifications, which uses the basic FEC for the error recovery. Depending on the radio channel type, MPE-FEC enhances the performance of DVB-H by applying an additional error coding to the radio link. As a result, the frame error rate can be lowered and thus the usable coverage area and/or the radio channel capacity can be extended. This paper presents a field test methodology, results and respective analysis in order to estimate the usability of different MPE-FEC code rates in urban, sub-urban and rural environments, and gives recommendations for the related parameter values taking into account typical terminal velocities per area type. In order to verify the behavior and level of the link budget enhancement via the MPE-FEC functionality in urban environment, case studies were carried out in laboratory environment as well as in functional urban DVB-H network coverage areas of Helsinki, Finland. The work contains the verification of the behavior of the error correction capabilities of MPE-FEC via field tests and by studying the postprocessed data. Comparative field test measurements were carried out in variable radio channel types, altering the terminal speed around the threshold values for the MPE-FEC. The test cases were repeated in urban, sub-urban and rural areas in order to investigate the effect of the radio channel type on the error correction performance. II. THE EFFECT OF MPE-FEC ON RADIO NETWORK PLANNING The MPE-FEC that DVB-H uses is based on Reed-Solomon (RS) code. The DVB-H Implementation Guidelines [1] explains the method, with the possibility to use puncturing. On the other hand, the decoding is left open and the receiver solutions might thus differ from each others. There are various possibilities to decode the signal as indicated in [15]. The IP datagrams of DVB-H are encapsulated into MPE- FEC frame. The time slicing burst size effects on the MPE- FEC in such way that the amount of MPE-FEC rows can be set to 256, 512, 768 or The datagrams are encapsulated column-wise to the frame, and the encoding is done row-wise with Reed-Solomon variant RS(255,191), indicating that the

272 mm total size of the frame is 255 columns, which consists of 191 columns of application data (IP datagrams) and 64 columns of Reed-Solomon data (parity bytes sections and possibly punctured RS fields). The Figure 1 shows the principle of the frame. The final MPE-FEC rate depends on the filling of the data and RS columns, and can be selected from the values of 1/2, 2/3, 3/4, 5/6 or 1/1 (i.e. when no MPE-FEC is used). IP datagram 1 IP datagram 1 IP datagram 2 IP datagram 2... Columns for the application data (MPE section) Fig. 1. DVB-H frame structure. IP datagram n Padding Padding 1st section of parity bytes 2nd section of parity bytes Last section of parity bytes 1st column of punctured RS 2nd column of punctured RS Columns 1-64 for the RS data table (FEC section) The IP datagram consists of IP header (20 bytes) and IP payload (maximum of 1480 bytes). The IP datagram then consists of MPE header (12 bytes) and CRC-32 check field (4 bytes). On the other hand, the FEC sections (columns) consists of FEC header (12 bytes) and CRC-32 tail (4 bytes). These MPE and FEC packets (with respective MPE or FEC headers) are then fragmented to a transport stream (TS) in such way that TS header (4-5 bytes; 5 if the TS packet contains the first byte of a section) is sent first, then MPE payload (183 bytes) followed by FEC TS header (4 bytes) and FEC payload (184 bytes). The frame error quality criterion of 5 % has been used in DVB-H radio network quality and limit estimations. The 5 % criteria can be studied when MPE-FEC is not used (frame error rate, FER), or with MPE-FEC involved (MFER, frame error rate after MPE-FEC). The 5 % criterion was originally selected in subjective way in the early phase of DVB-H evaluations, as it provides with practical means of estimating the quality. It is based on the intuitive idea of having maximum of 1 erroneous frame with 1 second of length during 20 seconds of measuring period. III. METHODOLOGY This study concentrates on the MPE-FEC performance via the post-processing and analysis of the radio measurement results. The field tests were carried out in mainly urban and rural areas in Helsinki inside of partially overlapping coverage areas of a commercial DVB-H network. Also selected laboratory measurements were carried out in indoor environment, and vehicular tests in the metropolitan area of Beijing, China. The field test equipment consisted of the commercial DVB- H terminal and a data collection unit which is based on the available radio interface measurement equipment. All the Last column of punctured RS relevant performance indicators were collected from the radio interface by storing the results to text files for the further postprocessing. The performance identifiers included the received power and packet error levels, carrier to noise level, frame errors before and after the MPE-FEC correction, as well as the bit error rates before and after Viterbi. Also the geographical position was stored to the log files in order to identify the specific location of occurred errors. In the quality revision of the network, the frame error rate indicates sufficiently well the real performance of the radio network as it is comparable with the end-user s interpretation of the service quality. More specifically, the DVB-H planning guideline [1] proposes that frame error rate of 5% or less indicates that the reception is sufficiently good as for the user experiences. Furthermore, the [1] shows that the 10% frame error level is already annoying in practical reception. After the first phase DVB-H evaluation tasks, more indepth criteria have been searched for. As an example, [16] has concluded that the limit for the acceptable quality based on the subjective interpretation is roughly between 6.9 % and 13.8 % as for the MFER when the channel type is close to vehicular urban (VU). This indicates that the MFER5 might be slightly pessimistic, but it is used in this study for the comparison purposes for the results obtained in other studies. The level of the MPE-FEC correction can be obtained by investigating what is the level of frame errors (FER, Frame error Rate) compared to the frame errors after the MPE-FEC functionality (MFER, MPE-FEC frame error rate). In this study, both laboratory and live network tests were carried out. The post-processing of the data was done by exporting the measurement raw data to Excel spreadsheet. The post-processing included the organization of the data to PDF format, received power level representing the x-axis with 1 db resolution, and MPE-FEC results being the variable as seen in Figure 2. When the values per RSSI level are normalized to 100 %, the format shows the proportion of the occurred frame errors that could be corrected, arranged in function of the RSSI levels as seen in figure 3. The format gives indication about the general probability distribution of the MPE-FEC functionality in investigated area type in function of RSSI. For the comparison of the performance with the terminal speed, also a cumulative presentation was calculated for frame error occasions showing the expected success rates of MPE- FEC with different parameter values. This gives indication of the effectiveness of the MPE-FEC per case in different area types, i.e. with what RSSI value the MPE-FEC starts functioning, and up to what RSSI limit it is still able to correct sufficiently effectively the frame errors. An example of this format can be seen in Figure 4. In the Figures 2-4, the term FER1, MFER0 means that there has been frame error which could be recovered by MPE- FEC. FER1, MFER1 means that the occurred error could not be recovered any more, and FER0, MFER0 means that there were no frame errors present in the first place.

273 mm Count of s am ple s per Prx leve l FER0, MFER0 # FER1, MFER0 # FER1, MFER1 # IV. THE CASE SETUP The investigation of the frame error correction capability of DVB-H was divided into three parts: Laboratory tests, indoor tests in live network and vehicular tests in live network. The setup consisted of commercial DVB-H terminals with additional software developed by Nokia for the logging of the relevant measurement values in the radio interface as described in [13] with the respective analysis method. Counts dbm Fig. 2. An example of the collected and post-processed data. The RSSI raster is 1 db, and each RSSI value corresponds to the error-free samples (FER 0), as well as occurred frame errors (FER 1) that could be corrected by MPE-FEC (0) or not (1). % FER% MFER % 10% criteria 5% criteria FER and M FER, % per RSSI value A. Laboratory tests In the laboratory tests, the 16-QAM was selected for the modulation method. In each test case, code rate of 1/2 was utilized, as well as FFT 8k and GI 1/4. The MPE-FEC rate was varied between 15% (5/6), 25% (3/4) and 35% (2/3). The test setup consisted of one radiating DVB-H indoor antenna (omni-pole with about +2 dbi gain in horizontal plane). There was a complete DVB-H core and radio network in use. The radio transmitter was adjusted to 250 mw input power level, which was fed to the antenna cable of about 15 meters of length and about 7 db loss per 100 m. The core network of the setup captured live audio/video content and delivered it in DVB-H format to the radio interface. The Table 1 shows the channel configuration for each test case. TABLE 1. THE DVB-H LABORATORY CHANNEL CONFIGURATION. Case CR FEC PID FEC rows A 50% 15% B 50% 25% C 50% 35% The measurements took place in indoor environment as presented in the Figure 5. There was a single corridor where the cell limits as well as the MPE-FEC performance areas were identified. dbm Fig. 3. An example of the MPE-FEC analysis FER/MFER%, cumulative over the route FER0, MFER0 cum% FER1, MFER1 cum% FER1, MFER0 cum% dbm Fig. 4. An example of the cumulative analysis of the measurement data. This format gives idea about the RSSI scale where the MPE-FEC was taken place and where it was effective. In this specific example, the MPE-FEC took place for 80 % of the samples (from 10% to 90% of cumulated values) in about dbm scale creating a 5 db window for the functional MPE-FEC. TX RX 50 m Fig. 5. The measurement area for the laboratory cases. The terminal speed was kept constantly about 1-2 m/s, i.e. keeping the terminal in constant slow motion with walking speed. The aim was to collect sufficiently data especially around the breaking point of the RSSI scale, i.e. where MPE- FEC takes place. B. Indoor tests in live network The indoor tests were carried out in the downtown area of Helsinki by observing the performance of the live DVB-H network. In this specific case, the radio network covered the city area and sub-urban of Helsinki, with partially overlapping

274 mm areas in SFN mode. The aim was to identify cell edge areas inside the typical public buildings like shopping centers. For the comparison purposes, also the respective outdoor measurement was done. This merely shows the received power level compared to the indoor cases, as there was practically no frame errors observed in the measurement points of the outdoor environment. During the measurements, the relevant radio interface parameters were the following: 16-QAM modulation, FFT 8k, GI 1/8, code rate 1/2, MPE-FEC rate 5/6 (15%). C. Vehicular tests As a third part of the study, there were outdoor measurements carried out by using a vehicle mounted calibrated field tester. The aim was to collect the performance indicators in variable vehicle speeds and different environments from the city center, sub-urban and rural areas. During the drive test an outside mounted antenna was used for the reception of the radio signal. The testing was conducted in three stages. The first stage was conducted within the city of Helsinki and Espoo (urban and sub-urban environment) where the same route was drives with different speeds to find the speed where the FER are starting to happen. The speeds driven were 20 km/h, 40 km/h and 60 km/h. The environment chosen was excellent (received power level RSSI limit minimum of -75 dbm and C/(N+I) > 15 db). This environment was chosen to eliminate all possibilities of degradation of the signal other then the speed. The second stage was conducted on the highway in urban areas in the greater Helsinki area and the speed tested were 80 km/h, 100 km/h and 120 km/h, again with excellent conditions. Both cases were conducted with the UdCast Navigator. In the third and final stage a drive test was conducted in the metropolitan area of Beijing in different speeds in an urban environment. Measurement equipment used was the Rohde&Schwarz TSM- DVB-H scanner. Finland network settings were: 16-QAM, CR 1/2, MPE- FEC 1/2, GI 1/8. China network settings were: QPSK, CR 1/2, MPE-FEC= 1/2, GI = 1/8. V. RESULTS AND ANALYSIS A. Laboratory tests The following Table 2 summarizes the laboratory test cases and respective results for the MPE-FEC breaking points (i.e. showing the RSSI values with the 5 % criteria) in received power levels (dbm) when using 16-QAM, code rate of 1/2, FFT of 8k and GI of 1/8. In this case, there were 2 terminals used in the data collection, Nokia models N-77 (terminal 1) and N-96 (terminal 2). As the results show, the terminals were not commonly calibrated as for the RSSI display. Nevertheless, the relative frame error information with and without MPE-FEC per terminal can be obtained even without the information about the exact RSSI breaking point. TABLE 2. THE DVB-H LABORATORY MEASUREMENT RESULTS. * THE GAIN COULD NOT BE SEEN EXPLICITLY DUE TO THE INSTABILITY IN MFER5 POINT. Case / FEC Terminal FER5 MFER5 Diff A / 15% B / 25% C / 35% A / 15% B / 25 % / /2.1 * C / 35% As the Table 2 shows, the difference between FER5 and MFER5 grows as the MPE-FEC rate grows, i.e. there is a logical behavior noted for the frame error capability in function of the MPE-FEC rate in all of the cases. In the least protected mode (MPE-FEC 15%), the difference, i.e. the MPE-FEC gain, is close to zero. This can be explained by the building layout, which did not allow too much multipath components in the receiving end. In the most protected mode that was applied in these measurements (MPE-FEC 35%), the MPE-FEC gain is in order of 2 db. The MPE-FEC rate of 25% seems to result only about 0.5 db gain. This case study shows that the MPE-FEC gain is not relevant in indoor environment with low amount of reflected multipath components in radio interface and with slowly moving terminal, if the MPE-FEC rate is defined to 15% or 25%. The effect starts to be notable when using MPE-FEC rate of 35%, giving about 2 db gains compared to the situation without MPE-FEC in the investigated channel types (which are close to typical urban 3, or TU3 type of models). On the other hand, this 35% MPE-FEC rate reduces accordingly the capacity that can be offered in the radio interface. B. Indoor tests in live network The following Table 3 summarizes the post-processed indoor measurement results for the city area of Helsinki, i.e. the breaking points of FER 5% and MFER 5% in function of the received power level (dbm). The counts column summarizes the amount of collected measurements (frames) per case. The cases are the following. A represents a ground floor of a shopping center with relatively large open areas. It can be estimated that some multi-path propagated components arrived to the measurement routes. B represents another relatively large shopping center in the very center of the city. There are probably plenty of multipath component present in this environment. C represents the ground floor of a relatively large 2-floor bookstore with open center area. D is the upper floor of the case C. E is the ground floor of a very large shopping center in the city center. F is the upper floor of the case E. G, H and I represents the ground, second and third floor of university campus building in sub-urban area.

275 mm TABLE 3. THE DVB-H INDOOR MEASUREMENT RESULTS. Case Counts Max speed FER5 MFER5 Diff A m/s B m/s N/A N/A N/A C m/s D m/s N/A N/A N/A E m/s F m/s G m/s H m/s I m/s Note 1. The environment contained impulse noise sources probably due to the power systems of the mechanical escalators. The Figure 6 shows the behavior of the effect. Note 2. There were relatively low amount of data collected in this case which affects on the statistical accuracy. There was also probably an impulse noise peak in FER figure which adds uncertainty to the interpretation of the corresponding MPE-FEC gain. Nevertheless, the MPE-FEC could correct the occurred interference peak. Note 3. The received power level was too high in this area in order to interpret the MPE-FEC effect FER and MFER (re sults in scale 0-20 % pe r Prx value ) FER% MFER % 10% criteria 5% criteria % BER samples 100% 80% 60% 40% 20% 0% 0 7 BER cummulative over the route speed (km/h) Fig. 7. BER analysis as a function of the terminal speed measured in Finland. The following graph shows the average PER (non correctable RS packets over the total received packets [17]) and the BER (Bit Error Rate before RS) in relation to the speed from the China measurements. From the graph can be seen that around 65 km/h the PER (> 5000 packets/sec) and BER (> ) are getting high. This is due to the challenging environment, which consists of densely packed flat buildings which create many multipath fading signals. PER (avg packet loss/ sec) BER and PER in relation to speed 0,008 0,007 0,006 0,005 0,004 0,003 0,002 0,001 0 BER (%) CDF PER BER Speed (km/h) % 10 5 Fig. 8. PER and BER as a function of the speed measured in China It can be seen from the Figure 8 that the 50% BER point is at 63 km/h dbm Fig. 6. An example of the impulse type of noise that affected on the measurements in case B. The results obtained from the laboratory environment are a bit more pessimistic compared to the dense urban indoor results. With 15% of MPE-FEC rate and common parameter set, the laboratory cases showed a maximum of 0.2 db gains whereas city center s indoor environment provided up to 1 db gains. This is due to the varying multipath components in radio interface, laboratory being more limited in this sense. In sub-urban environment, though, the indoor gain was around in the same low level as was obtained in laboratory. C. Vehicular tests From the first drive test results no errors were found until 60 km/h. The errors started to appear with higher speeds. The following figure represents the CDF of the BER with different speeds. The 50% point is at 80 km/h. cdf 100% 80% 60% 40% 20% % Fig. 9. BER CDF measured in China BER Cummulative over the route Speed CDF BER VI. MPE-FEC OPTIMIZATION The analysis shows that the MPE-FEC is normally useful only when the received power level is low enough. The Figure 4 shows the basic principle of the functional scale of MPE- FEC, indicating that the RSSI window where MPE-FEC wakes up and is able to correct the erroneous frames is relatively small and near the performance limits of the cell. This means, that in decent and good radio field, MPE-FEC does not give added value even if there are multi-propagated

276 mm radio components present. This applies for the indoor and outdoor pedestrian and vehicular environments. With the used parameter set, the Doppler shift was not the limiting factor. The drive test results show that the speed for the BER breakpoint is around 80 km/h in perfect conditions and -65 dbm in challenging environments. The lowest MPE-FEC rates result only minor enhancement in indoor environment based on the laboratory measurements and the FER/MFER analysis. MPE-FEC gain was about 0.5 db or less for 25% MPE-FEC rate, and near zero for 15% MPE-FEC rate. For 35% MPE-FEC rate, values of about 2 db were observed for the MPE-FEC gain. The Table 4 summarizes the averaged MPE-FEC gain results and the respective standard deviation values for laboratory, city indoor, suburban indoor and sub-urban motorway cases when analyzed via FER and MFER differences and MPE-FEC 15% was used. TABLE 4. THE SUMMARY OF THE MPE-FEC GAINS IN DIFFERENT ENVIRONMENTS. Environment Average Stdev Laboratory Urban indoor Sub-urban indoor Sub-urban motorway 1.0 N/A As the MPE-FEC reserves a proportion of capacity from the total radio channel bandwidth, the balancing of the capacity and the MPE-FEC gain should be planned according to the operational environment. Based on this analysis, it would be recommendable to limit the MPE-FEC rate to lower values of the scale as it reserves thus only small proportion of the capacity but still corrects at some extend the occurred frame errors. This would be justified in indoor environment even with the multi-propagated radio paths as the field tends to drop fast in the cell edge according to the laboratory and indoor field measurements carried out in this study. On the other hand, in the typical urban and dense urban environment, the DVB-H radio coverage is normally good enough in order to provide decent indoor coverage. This means that the field in outdoor is so high that MPE-FEC is not normally needed, but recommendable due to multipath fading. Also sub-urban area within the DVB-H coverage seems to benefit only little from MPE-FEC. In the buildings, the gain seems to be even less than in urban areas due to the lower amount of multipath components. Nevertheless, approaching the service area edge, the field strength is sufficiently low in order to activate the MPE-FEC, which brings about 1 db MPE-FEC gain in normal motorway environment with the lowest MPE-FEC rate. VII. CONCLUSIONS In the in-depth DVB-H network planning, the fine-tuning of the relevant radio parameters is essential for different geographical clutter types. This study presents the MPE-FEC test methodology which is based on the collection of the relevant radio performance indicators via the field test equipment. Furthermore, an analysis of the MPE-FEC behavior was done in order to seek the optimal functional radio parameter values in the investigated area types. This paper describes the methodology that can be applied for DVB-H radio reception error level measurements and analysis in order to identify the optimal MPE-FEC rate parameter values in different network environments. The outcome of the study is the optimal MPE-FEC parameter value set for the typical DVB-H environments in function of the terminal velocity. The paper also shows that MPE-FEC can be altered in different DVB-H areas, although the tuning of the parameter should be done based on the field measurements of individual cases as the carrier per noise level varies in each network setup. REFERENCES [1] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [2] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [3] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D6: Common field trials report. November p. [4] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. Wing TV Country field report. November p. [5] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. IEEE p. [6] Joan Mazenc. Multi protocol Encapsulation Forward Error Correction (MPE-FEC). ETUD/Insa. Toulouse, France, January 9, p. [7] G. Gardikis, H. Kokkinis, G. Kormentzas. Evaluation of the DVB-H data link layer. European Wireless April 2007, Paris, France. 6 p. [8] Heidi Joki, Jarkko Paavola. A Novel Algorithm for Decapsulation and Decoding of DVB-H Link Layer Forward Error Correction. Department of Information Technology, University of Turku. 6 p. [9] Masaharu Hata. Empirical Formula for Propagation Loss in Land Mobile Radio Services. IEEE Transactions on Vehicular Technology, Vol. VT- 29, No. 3, August p. [10] Recommendation ITU-R P Method for point-to-area predictions for terrestrial services in the frequency range 30 MHz to 3000 MHz p. [11] Tero Jokela, Eero Lehtonen. Reed-Solomon Decoding Algorithms and Their Complexities at the DVB-H Link-Layer. Turku Centre for Computer Science, Department of Information Technology, University of Turku. 5 p. [12] ETSI EN , V1.4.1 ( ). Digital Video Broadcasting (DVB); DVB specification for data broadcasting. European Telecommunications Standards Institute p. [13] Penttinen, J. Field Measurement and Analysis Method for DVB-H Terminals. The Third International Conference on Digital Telecommunication. ICDT 2008, Bucharest, Romania. 6 p. [14] D. Plets, W. Joseph, L. Verloock, E. Tanghe, L. Martens, E. Deventer, H. Gauderis. Influence of Reception Condition, MPE-FEC Rate and Modulation Scheme on Performance of DVB-H. IEEE Transactions on Broadcasting, Vol. 54, No. 3, September Pp [15] J. Paavola, H. Himmanen, T. Jokela, J. Poikonen, V. Ipatov. The Performance Analysis of MPE-FEC Decoding Methods at the DVB-H Link Layer for Efficient IP Packet Retrieval. IEEE Transactions on Broadcasting, Vol. 53, No. 1, March Pp [16] H. Himmanen. Studies on Channel Models and Channel Characteristics for Mobile Broadcasting. Department of Information Technology, University of Turku, Turcu Centre for Computer Science, Finland. 9 p. [17] IP datacast over DVB-H : Implementation guidelines for mobility, DVB document A117, DVB organization, July 2007, p 18-19

277 IX Publication IX Field measurement and data analysis method for DVB-H mobile devices This paper is an extended version of the Publication II. The original conference paper received a Best Paper Award of IARIA. This extended version shows more thoroughly the measurement methodology, and shows in a deeper level the analysis of the field test results by dividing the investigated area into separate radio propagation channel types. International Journal on Advances in Systems and Measurements, ISSN X, Vol. 2, No. 1, 2009, pages IARIA Reprinted with permission.

278

279 Field Measurement and Data Analysis Method for DVB-H Mobile Devices Jyrki T.J. Penttinen Member, IEEE Abstract The field measurement equipment that provides reliable results is essential in the quality verification of DVB-H networks. In addition, sufficiently in-depth analysis of the post-processed data is important. This paper is based on [12] and presents a method to collect and analyze the key performance indicators of the DVB-H radio interface, using a mobile device as a measurement and data collection unit. 1. Introduction The verification of the DVB-H quality of service level can be done by carrying out field measurements within the coverage area. Correct way to obtain the most relevant measurement data, as well as the right interpretation of it, are fundamental for the detailed network planning and optimization. During the normal operation of the DVB-H network, there are only few possibilities to carry out long-lasting and in-depth measurements. A simple and fast field measurement method based on mobile DVB- H receiver provides thus added value for the operator. The mobile equipment is easy to carry both in outdoor and indoor environment, and it stores sufficiently detailed performance data for the post-processing. The measurements are required for the network performance revisions and for the indication of potential problems. As an example, the transmitter site antenna element might move due to the loose mounting, which results outages in the designed coverage area. The antenna feeder might still remain connected correctly, keeping the reflected power in acceptable level. As there are no alarms triggered in this type of instance, and as the basic DVB-H is a broadcast system without uplink and its related monitoring / alarming system, the most efficient way to verify this kind of fails is to carry out field tests. This paper presents a method to post-process the basic field test data collected with DVB-H mobile terminal. The method can be considered as an addition to the usual network performance testing carried out by the operator and is suitable for the fast revisions of the radio quality levels and possible network faults. The results presented in this paper are meant as examples and for clarifying the analysis methodology. The data was collected with a commercial DVB-H hand-held terminal capable of measuring and storing the radio link related data. In this specific case, a Nokia N-92 terminal was used with a field test program. The program has been developed by Nokia for displaying and storing the most relevant DVB-H radio performance indicators. 2. Coding of DVB-H In order to carry the DVB-H IP datagrams of the MPEG-2 Transport Stream (TS), a Multi Protocol Encapsulator (MPE) is defined for DVB. Each IP datagram is encapsulated into single MPE section. Elementary Stream (ES) takes care of the transporting of these MPE sections. Elementary Stream is thus a stream of packets belonging to the MPEG-2 Transport Stream and with a respective program identifier (PID). The MPE section consists of 12 byte header, 4 byte CRC-32 (Cyclic Redundancy Check), as well as a tail and payload length. [4] The main idea of MPE-FEC is to protect the IP datagrams of the time sliced burst with an additional link layer Reed-Solomon (RS) parity data. The RS data is encapsulated into the same MPE-FEC sections of the burst with the actual data. The RS part of the burst belongs into the same elementary stream (MPE-FEC section), but they have different table identifications. The benefit of this solution is that the receiver can distinguish between these sections, and if the terminal does not have the capability to use the DVB-H specific MPE-FEC, it can anyway decode the bursts although with lower quality when it experiences difficult radio conditions. [13, 14, 15, 16] The part of the MPE-FEC frame that includes the IP datagrams is called application data table (ADT). ADT

280 has a total of 191 columns. In case the IP datagrams does not fill completely the ADT field, the remaining part is padded with zeros. The division between ADT and RS table is shown in the Figure 1. In DVB-H system, the number of the RS rows can be selected from the values of 256, 512, 768 and The amount of the rows is indicated in the signaling via the Service Information (SI). RS data has a total of 64 columns. For each row, the 191 IP datagram bytes are used for calculating the 64 parity bytes of RS rows. Also in this case, if the row is not filled completely, padding is used. The result is a relative deep interleaving as the application data is distributed for the whole bust. [7] 256, 512, 768 or 1024 rows... First byte (element) IP datagrams data columns... Padded space Padded space Padded space RS parity bytes 64 RS columns Punctured columns Figure 1. The MPE-FEC frame consists of application data table for IP datagrams and Reed- Solomon data table for RS parity bytes. The error correction of DVB-H is carried out with the widely used Reed-Solomon coding. The Reed- Solomon code is based on the polynomial correction method. The polynomial is encoded for the transmission over the air interface. If the data is corrupted during the transmission, the receiving end can calculate the expected values of the data within the certain limits that depends on the settings. The RS data of DVB-H is sent in encoded blocks with the total number of m-bit symbols in the encoded block is n=2 m -1. With 8-bit symbols the amount of symbols per block is n= = 255. This is thus the total size of the DVB-H frame. The actual user data inside of the frame is defined as a parameter with a value k, which is the number of data symbols per block. Normal value of k is 223 and the parity symbol is 32 (with 8 bits per symbol). The universal format of presenting these values is (n,k) = (255,223). In this case, the code is capable of correcting up to 16 symbol errors per block. RS can correct the errors depending on the redundancy of the block. For the erroneous symbols which location is not known in advance, the RS code is capable of correcting up to (n-k)/2 symbols that contains errors. This means that RS can correct half as many errors as the amount of redundancy symbols is added in the block. If the location of the errors is known (i.e., in case of erasures), then RS can correct twice as many erasures as errors. If N err is the number of errors and N ers the number of erasures in the block, the combination of error correction capability is according to the formula 2N err + N ers < n. The characteristic of RS error correction is thus well suited to the environment with high probability of errors occurring in bursts, like in DVB-H radio interface. This is because it does not matter how many bits in the symbol are erroneous if multiple errors occur in byte, it is considered as a single error. It is possible to use also other block sizes. The shortening can be done by padding the remaining (empty) part of the block (bytes). These padded bytes are not transmitted, but the receiving end fills in automatically the empty space. FEC (Forward Error Correction) is widely used in telecommunication systems in order to control the errors. In this method, the transmitting party adds redundant data to the message. On the other side, the receiving end can detect and correct the errors accordingly without the need to acknowledge the data correction. The method is thus suitable for especially uni-directional broadcast networks. FEC consists of block coding and convolutional coding. RS is an example of the block coding, where the blocks or packets of bits (symbols) are of fixed size whereas convolutional coding is based on bit or symbol lengths. In practice, the block and convolutional codes are combined in concatenated coding schemes; the convolutional coding has the major role whereas block code like RS cleans the errors after the convolutional coding has taken place. In mobile communications, the convolutional codes are mostly decoded with the Viterbi algorithm. It is an error-correction scheme especially suitable for noisy digital communication links. It is used e.g. in GSM, dial-up modems, satellite communications and in LANs. It is also included in DVB-H radio transmission. The idea of Viterbi is to find the most probable sequence, Viterbi path in the information flow, i.e. in the sequence of observed events.

281 The MPE-FEC has been designed taking into account the backwards compatibility. DVB-T demodulation procedure contains error correction so both DVB-T and DVB-H utilizes it for the basic coding with Viterbi and Reed-Solomon decoding, whereas DVB-H can use also a combination of Viterbi, Reed- Solomon and additional MPE-FEC to improve the C/N and Doppler performance. [13] The detection of the presence of MPE-FEC is done based on the single demodulated TS packet, as its header contains the error flag. MPE-FEC adds the performance in moving environment as it uses so called virtual interleaving over several basic FEC sections. In this study, the bit rates before and after the Viterbi as well as the frame error rates of DVB-H MPE-FEC and DVB-T specific FEC were observed by collecting and analyzing data in radio interface. The measurement principles of WingTV guidelines [3] were taken into consideration in the selection (limiting) of the parameter values. The guideline defines the MFER (MPE Frame Error Ratio) as a ratio of the number of residual erroneous frames that can not be recovered to the total number of the received frames: residual _ erroneous _ frames MFER(%) = 100 received _ frames There are possibilities to obtain the MFER value by storing the frames during certain time (e.g. 20 seconds), or as proposed by DVB, at least 100 frames should be collected in order to calculate the MFER with sufficient statistical reliability. The measurement principles of WingTV was used as a basis in the study by collecting 25 minutes of field data during a drive test with varying speeds, and the data was later post-processed in order to obtain the more specific relation between the FER, MFER and received power level categories. Although WingTV does not recommend the use of QEF as a criterion, also the bit errors were analyzed by utilizing the same principles for the comparison purposes. investigated area with nearly-los in major part of the investigated area. Figure 2. The environment in main lobe of the transmitter antenna. The methodology was verified by carrying out various field tests mostly in vehicle. There was also static and dynamic pedestrian type of measurements included in the test cases in order to verify the usability of the equipment and methodology for the analysis. The DVB-H test network consisted of a single 200 watt DVB-H transmitter and a basic DVB-H core network. The source data was delivered to the radio interface by capturing real-time television program. The program was converted to DVB-H IP data stream with standard DVB-H encoders. There was a set of 3 DVB-H channels defined in the same radio frequency, with audio / video bit rates of 128, 256 and 384 kb/s. The number of FEC rows was selected as 256 for MPE-FEC rate of ½, and 512 for MPE-FEC rate of 2/3. The audio part of the channel was coded with AAC using a total of 64 kb/s for stereophonic sound. The bandwidth was 6 MHz in 701 MHz frequency. The antenna system consisted of directional antenna panel array with 2 elements as shown in Figure 3. Each element produces 65 degrees of horizontal beam width and provides a gain of dbi. 3. Test setup A test setup with a functional DVB-H transmitter site was utilized in order to investigate the presented field test methodology with respective post-processing and analysis. The site antenna safety distance zones were estimated by applying [9]. The investigated area represents relatively open sub-urban environment as shown in the Figure 2. There are mainly relatively small trees, residential areas and highways in the Figure 3. The antenna system setup. The vertical beam width of the single antenna element was 27 degrees, which was narrowed by locating two antennas on top of each others via a

282 power splitter. Taking into account the loss of cabling, jumpers, connectors, power splitter and transmitter filter, the radiating power was estimated to be dbm (EIRP) in the main lobe. The transmitter antenna system was installed on a rooftop with 30 meters of height from the ground. The environment consisted of sub-urban and residential types with LOS (line-of-sight) or nearly LOS in major part of the test route, excluding the back lobe direction of the site which was non-los due to the shadowing of the site building. Each test route consisted of two rounds in the main lobe of the antenna with a minimum received power level of about -90 dbm. The maximum distance between the antenna system and terminals was about 6.4 km during the drive tests. If the relevant data can be measured from the radio interface and stored in text format, the method presented in this paper is independent of the terminal type. It is important to notice, though, that the characteristics of the terminal affects on the analysis, i.e. the terminal noise factor and the antenna gain (which is normally negative in case of small DVB-H terminals) should be taken into account accordingly. On the other hand, unlike with the advanced field measurement equipment, the method gives a good idea about the quality that the DVB-H users observe in real life as the terminal type with its limitations is the same as used in commercial networks. Figure 4. The terminal measurement setup. There were a total of 3 terminals used in each test case for capturing the radio signal simultaneously. Multiple receptions provide respectively more data to be collected at the same time, which increases the statistical reliability of the measurements. It also makes possible the comparison of the differences between the terminal performances. The terminals were kept in the same position inside the vehicle without external antenna, and the results of each test case were saved in separate text files. The terminal setup is shown in the Figure 4. The external antenna was not used because the aim was to revise the quality that the end-user experiences in normal conditions inside the moving vehicle. On the other hand, the test cases were not designed for certain coverage area, but the aim was to classify the performance indicators in function of the received power levels. It does not matter thus if the received power level is interpreted via external or internal antenna. 4. Terminal measurement principles The DVB-H parameter set was adjusted according to each test case. The cases included the variation of the code rate (CR) with the values of ½ and 2/3, MPE- FEC rate with the values of ½ and 2/3 and interleaving size FFT with the values of 2k, 4k, 8k, in accordance with the Wing TV principles described in [3], [4], [5] and [6]. The guard interval (GI) was fixed to ¼ in each case. The parameter set was tuned for each case, and the audio / video stream was received with all the terminals by driving the test route two consecutive times per each parameter setting. The needed input for the field test is the on and off time of the time sliced burst, PID (Packet Identifier) of the investigated burst, the number of FEC rows and the radio parameter values (frequency, modulation, code rate and bandwidth). The terminal stores the measurement results to a log file after the end of each burst until the field test execution is terminated. According to the DVB-H implementation guidelines [1], the target quality of service is the following: For the bit error rate after Viterbi (BA), the reception should comply at least DVB-H specific QEF (quasi error free) point The frame error rate should be less than 5%. In case of FER, i.e. DVB-T, this criterion is called FER5, and for the DVB-H specific MPE-FEC, its name is MFER5. The field test software of N-92 is capable of collecting the RSSI (received power levels in dbm), FER (Frame Error Rate) and MFER (MPE Frame Error Rate, i.e. FER after MPE-FEC correction) values. In addition, there is possibility to collect information about the packet errors. The Figure 5 shows a high-level block diagram of the DVB-H receiver. [1] The reception of the Transport Stream (TS) is compatible with DVB-T

283 system, and the demodulation is thus done with the same principles also in DVB-H. The additional DVB- H specific functionality consists of Time Sliced burst handling, MPE-FEC module and the DVB-H deencapsulation. be corrected any more with MPE-FEC. In the latter case, the bit error information could not be calculated either. It seems that in this specific case, the RSSI value of about -87 dbm to -89 dbm has been the limit for the correct reception of the frames with MPE-FEC. DVB-H specific functionality RF reference DVB-T Demodulator TS reference DVB-H Time Slicing DVB-H MPE-FEC DVB-H De-encapsulation IP output FER reference MFER reference IP reference Figure 5. A principle of the reference DVB-H terminal. As can be seen from the Figure 5, the FER information, i.e. frame errors before MPE-FEC specific analysis, is obtained after the Time Slicing process, and the MFER is obtained after the MPE-FEC module. If the data after MPE-FEC is free of errors, the respective data frame is de-encapsulated correctly and the IP output stream can be observed without disturbances. The measurement point for the received power level is found after the antenna element and the optional GSM interference filter. In addition, there might be optional external antenna connector implemented in the terminal before the RF reference point. The presence of the filter and antenna connector has thus frequency-dependent loss effect on the measured received power level in the RF point. The Figure 6 shows an example of the measurement data file. The example shows 4 consecutive test results. Each measurement field contains information about the occurred frame error (FER), frame error after MPE- FEC correction (MFER), bit error rate before Viterbi (BB) and after Viterbi (BA), packet errors (PA) and received power level (RSSI) in dbm. In this case, there was a frame error in the reception of the first measurement sample because the value of FER was 1. The FER value is either 0 for non-erroneous or 1 for erroneous frame. The MPE-FEC procedure could still recover the error in this case, because the MFER parameter is showing a value of 0. The second sample shows that there were no frame errors before or after the MPE-FEC. The third sample shows again frame error that could be corrected with MPE- FEC. The fourth sample shows an error that could not Figure 6. Example of the measured objects with four consecutive results for FER, MFER, BB, BA, PE and RSSI. The plain measurement data has to be postprocessed in order to analyze the breaking points for the edge of the performance. Microsoft Excel functionality was utilized in order to arrange the data in function of the received power levels. According to the first sample of the Figure 6, the bit error level before Viterbi (BB) was close to the QEF point, i.e The bit error level after the Viterbi (BA) was which is already better than the QEF point for the acceptable reception. The bit error rate had been thus low enough for the correct reception of the signal. In this example, the amount of packet errors (PE) was between 7 and 75, and the averaged received power level was measured and averaged to dbm. It is worth noting that the RSSI resolution is 1 db for single measurement event in the used version of the field test software. The Figure 7 shows the measured RSSI values during the complete test route. There were two rounds done during each test. The back lobe area of the test route can be seen in the middle of the Figure, with fast momentarily drop of received power level. The received power level was about -50 dbm close to the site, and about -90 dbm in the cell edge. The duration of the single test route was approximately 25 minutes, and the total length of the route was 22.4 km. The maximum speed during the test route was about 90 km/h, and the average speed was measured to 50 km/h (excluding the full stop periods). The speed is sufficient for identifying the effect of the MPE-FEC.

284 RSSI (db) Received power level Sample # Fig. 7. The RSSI values measured during the test route. The following Figures 8 and 9 present the estimated coverage area for QPSK and 16-QAM cases with the code rate of ½ and MPE-FEC rate of ¾. The Okumura- Hata based propagation model [10] was used with a digital map that contains the elevation data and cluster type information. The minimum received outdoor power level limit for QPSK was estimated as -84 dbm and for 16-QAM as -78 dbm. The 80 % area location probability criterion was used in these coverage plots. The grid size is 1.6 km. (Background map source: Google Map). Figure 8. The predicted coverage area for 16- QAM. The raster shown in the map is 1.6 km. The main lobe and the test route is to north-west direction form the site. Figure 9. The predicted coverage area for QPSK, with the raster size of 1.6 km. Based on the coverage plots, the test route was selected accordingly. The coverage plots correlated roughly with the drive tests, although the in-depth location-dependent signal level measurements were not carried out during this specific study. 5. Method for the analysis The collected data was processed accordingly in order to obtain the breaking points, i.e. the QEF of and FER / MFER of 5% in function of the RSSI values for each test case. The processing was carried out by arranging the occurred events per RSSI value. For the BB and BA, the values were averaged per RSSI resolution of 1 db. For the FER and MFER, the values represent the percentage of the erroneous frames compared to the total frame count per each individual RSSI value (with the resolution of 1 db). The following Figure 10 shows an example of the processed data for the bit error rate before and after the Viterbi for 16-QAM, CR 2/3, MPE-FEC 2/3 and FFT 2k. The results represent the situation over the whole test route in location-independent way, i.e. the results show the collected and averaged BB and BA values that have occurred related to each RSSI value in varying radio conditions. As can be noted in this specific example, the bit error rate before Viterbi does not comply with the QEF criteria of even in relatively good radio conditions, whereas the Viterbi clearly enhances the performance. The resulting breaking point for the QEF

285 with Viterbi can be found around -78 dbm of RSSI in this specific case. 1.00E E BB and BA, average for each Prx value and does not necessarily provide with sufficient statistical reliability in that range of RSSI values. Nevertheless, as the idea was to observe the performance especially in the limits of the coverage area, it is sufficient to collect reliable data around the critical RSSI value ranges. BER 1.00E E E Count of samples per Prx level FER0, MFER0 # FER1, MFER0 # FER1, MFER1 # 1.00E E E E-08 dbm QEF BB ave BA ave Figure 10. Post-processed data for the bit error rate before and after the Viterbi presented in logarithmic scale. 5.0E E E E E E+00 QEF BA ave BB ave BB and BA averaged for each Prx value dbm Figure 11. Processed data for the bit error rate before and after the Viterbi with an amplified view around the QEF point in linear scale. For the frame error rate, the similar analysis yields an example that can be observed in Figure 12. The Figure shows the occurred frame error counts (FER and MFER) as well as the amount of error-free events analyzed separately for each RSSI value. In this format, the Figure shows the amount of occurred samples in function of RSSI in 1 db raster arranged to error free counts ( FER0, MFER0 ), to counts that had error but could be corrected with MPE-FEC ( FER1, MFER0 ), and to counts that were erroneous even after MPE-FEC ( FER1, MFER1 ). It can be noted that the amount of the occurred events is relatively low in the best field strength cases Counts dbm Fig. 12. Example of the analyzed FER and MFER levels of the signal. In this type of analysis, the data begins to be statistically sufficiently reliable when several tens of occasions per RSSI value are obtained, preferably around 100 samples as stated in [3]. In practice, though, the problem arises from the available time for the measurements, i.e. in order to collect about 100 samples per RSSI value in large scale it might take more than one hour to complete a single test case. There were a total of 32 test drive rounds carried out, 25 minutes each. In this case, the post-processing and analysis was limited to 2 terminals though due to the extensive amount of data. The corresponding amount of total samples was normalized, i.e. scaled to 0-100% separately for each RSSI value. An example of this is shown in Figures 13 and 14. By presenting the results in this way, the percentage of FER and MFER per RSSI and thus the breaking point of FER / MFER can be obtained graphically. The 5% FER and MFER levels, i.e. FER5 and MFER5, can be obtained graphically for each case observing the breaking point for the respective curves. The corresponding MPE-FEC gain can be interpreted by investigating the difference between FER5 and MFER5 values (in db). The graphics shows the observation point directly along the 5% error line. The parameter values of the following examples are still 16-QAM, CR 2/3, MPE-FEC 2/3 and FFT 2k.

286 Counts 100% 80% 60% 40% 20% Count of samples per Prx level, normalized FER0, MFER0 # FER1, MFER0 # FER1, MFER1 # As additional information about the FEC and MPE- FEC performance in the whole scale of 0-100%, the cumulative presentation can be observed as shown in the Figure 15. This format gives indication about the RSSI range where the MPE-FEC starts correcting. The results presented in this case can be postprocessed further in order to fragment the test routes into the more specific area and radio channel types. The following Figure 16 shows the segments during the test drive % dbm Figure 13. The post-processed data can be presented in graphical format with FER and MFER percentages for each RSSI value FER and MFER (results in scale 0-20 % per Prx value) FER% MFER % 10% criteria 5% criteria A B C E D F % MPE-FEC gain dbm Figure 14. An amplified view to FER5 and MFER5 criteria shows the respective RSSI breaking points FER/MFER%, cumulative over the route FER0, MFER0 cum% FER1, MFER1 cum% FER1, MFER0 cum% dbm Fig. 15. The processed data can also be presented in cumulative format over the whole route. This presentation gives a rough estimate about the RSSI range of the correction. Figure 16. The segments of the test route. Segments A and E represents the terminal moving away from the site in LOS within the main lobe of the antenna beam, with a maximum speed of about 70 km/hour and with a good field strength. Segments B and F represents the LOS or nearly LOS with the terminal moving with a maximum speed of km/h, and the area represents the cell edge or near the cell edge, depending on the selected modes. Segment C represent the terminal moving towards the site with a maximum speed of 90 km/h, with a good field strength and LOS. Segment D represents the situation in back lobe of the antenna with N-LOS situation due to the shadowing of the site building, with the terminal speed varying between 0 and 80 km/h. Having the segmentation done according to the figure 16, it can be seen that in the good field, as expected, the occurred FER and MFER instances are minimal and they do not affect on the reception of the DVB-H audio / video streams. Furthermore, the MPE- FEC does not bring enhancements for the performance within such a good field. The most interesting parts of the segments are thus the ones that represents the situation nearer to the cell edge, both in main lobe (B and F) and in back lobe (D).

287 As an example, the analysis for the case QPSK, CR 2/3, MPE-FEC ½, and FFT 4k, shows quite controlled behaviour of the FER and MFER until the breaking point, when the results from the segments B and F are combined and analyzed as a one complete block. % FER and MFER (results in scale 0-20 % per Prx value) FER% MFER % 10% criteria 5% criteria MPE-FEC gain It is worth noting, though, that this format gives only an indication about the RSSI range where the different modes of error correction tends to occur, i.e. the clean samples (FER0, MFER0), the successful MPE-FEC corrections (FER1, MFER0), and when the MPE-FEC is not able to correct the data (FER1, MFER1). As a comparison, the N-LOS segment D yields the following figures 19 and 20. The behaviour of the curves is not as clear as it is in main lobe. In addition to the attenuated N-LOS, the multi-path propagated signals are not strong in this area. It is though worth noting that the amount of the collected data is quite low, in order of samples, which reduces considerably the reliability of the back lobe analysis dbm Figure 17. An example of the main lobe analysis with LOS. The curves are in general more controlled compared to the analysis of the whole route. % FER and MFER (results in scale 0-20 % per Prx value) FER% MFER % 10% criteria 5% criteria It is also interesting to investigate where in the RSSI scale the FER and MFER occasions have been occurred during these segments. The following Figure 18 shows the cumulative presentation of different FER and MFER occasions with above mentioned parameter values FER/MFER%, cumulative over the route FER0, MFER0 cum% FER1, MFER1 cum% FER1, MFER0 cum% dbm Figure 18. The cumulative presentation of FER and MFER gives a rough indication about the performance within the investigated area. Compared to the Figure 15, the main lobe analysis produces clearer picture about single radio channel type dbm Figure 19. The back lobe analysis with N-LOS shows that the MPE-FEC is not able to correct the occurred frame error as efficiently as in main lobe with LOS. The respective segment has been selected from the same data file as shown in Figure 17 for the main lobe FER/MFER%, cumulative over the route FER0, MFER0 cum% FER1, MFER1 cum% FER1, MFER0 cum% dbm Figure 20. The cumulative presentation of back lobe analysis. It can be seen that the low amount of data affects clearly on the smoothness of the curves due to lack of data.

288 6. Terminal comparison The terminal measures the received power level after the possible GSM interference suppression filter. There might also be external antenna connectors in either side of the filter. The terminal characteristics thus affects on the received power level interpretation. In order to obtain information about the possible differences of the terminal displays, separate comparison measurements were carried out. There were a total of three terminals used during the testing. As the terminals were still prototypes, the calibration of the RSSI displays was not verified. This adds uncertainty factor to the test results. The following Figure 21 shows a test case that was carried out in laboratory by keeping all the terminals in the same position and making slow-moving rounds within relatively good coverage area. 1.8E E E E E E E E E E BA ave 2BA ave 3BA ave Figure 22. A comparison of 3 terminal models via BER measurement. It can be seen that two of the phones behave similarly whilst one is showing smaller bit error values Figure 21. An example of the laboratory test case for the comparison of the RSSI displays of the terminals. The systematic difference in RSSI displays can be noted, being about 2 db between the extreme values. The same 2 db difference between the terminals was noted in the field test analysis. The values obtained from the radio network tests cannot thus be considered accurate. Nevertheless, the idea of the testing was to investigate rather the methodology of the measurements than to obtain accurate values of the defined parameter settings. In the in-depth analysis, in addition to the RSSI, also more specific differences between the terminals can be investigated. As the following Figures 22, 23 and 24 shows, there is a systematic difference margin between the three utilized terminals as for the QEF, FER5 and MFER5. As a conclusion of the Figures, in order to minimize the error margin that arises from the differences of the BER, FER and MFER interpretation of different terminals, it is important to calibrate the models accordingly FER% 2FER% 3FER% Figure 23. The systematic difference of the performance measurement values between the three terminals can be observed also via the FER curves MFER % 2MFER % 3MFER % Figure 24. The differences of the terminals via the MFER analysis

289 7. Building penetration loss For comparison purposes, there was also a set of test cases carried out in the pedestrian environment with the same measurement methodology as described previously. When designing the indoor coverage, the respective building loss should be taken into account. The loss depends on the building type and material. As an example, the following Figure 25 shows a relatively short measurement carried out in outdoor and indoor of a 10-floor hotel building within the coverage area of the investigated DVB-H network. height etc. in each environment type. In case of new areas, the best way for the estimation is to carry out sufficient amount of sample measurements for the most typical building types, although for the initial link budget estimations, the average of db could be a good starting point according to these tests. 8. Effects on the coverage estimation Based on the measurement results, a tuned link budget can now be build up by using the essential radio parameters according to the Figure 26. Prx (dbm) RX level Sample # RSSI (in) RSSI (out) Figure 25. An example of short snap-shot type indoor and outdoor measurements carried out in the DVB-H trial network coverage area. The building loss of the Figure 25 can be obtained by calculating the difference of the received power values. The indoor average received power level is dbm with a standard deviation of 5.7 db. The outdoor values are dbm and 3.9 db, respectively. The building loss in this specific case is thus 16.1 db. The value is logical as the building represents relatively heavy construction type, although the roof top was partially covered by large areas of class resulting relatively good signal propagation to interior of the building via the diffraction. The test was repeated in selected spots inside and outside of the buildings, in variable field strengths both in main lobe and back lobe of the site antenna radiation pattern. As a general note, the static or low-speed pedestrian cases did not initiate the MPE-FEC functionality. Obviously more multi-path propagated signals and/or higher terminal speed would have been needed in order to wake up the MPE-FEC. In the general case, the building loss should be estimated depending on the overall building type, Carrier Noise figure Thermal noise floor Minimum received power level outside Location variation margin C/N min Building loss RX antenna gain RX sensitivity Distance r Fig. 26. The main principle of the DVB-H link budget. In order to estimate the useful cell radius of the DVB-H radio transmission, propagation prediction models can be used. One of the most used one in broadcast environment is the Okumura-Hata model, which is suitable as such for the macro cell type of environments [10]. The model functions sufficiently well in practice when the height of the transmitter site antenna is below 200 meters, and the frequency range is of MHz. The model is reliable for the cell ranges up to 20 km. The corrected Okumura-Hata prediction model for the distances over 20 km is defined in CCIR report It is sufficient to estimate the cell radius for the high-power DVB-H transmitter sites with the radius up to 50 km. The most recent model that is especially suitable for the large variety of distances and transmitter antenna heights is ITU-R P.1546, which is based on the mapping of the pre-calculated curves. [11] Based on the field test examples shown previously, the MPE-FEC gain can be included for the sensitivity figure, compared to the DVB-T case which contains only FEC. The gain varies depending on the terminal speed and environment type, so the fine-tuning of the link budget should be considered depending on the local conditions.

290 9. Field test results 9.1. Complete route As a result of the vehicle based field tests performed in this study, the following Tables 1-3 summarizes the RSSI thresholds for the QEF point of and FER / MFER of 5% criteria with different parameter values for the whole test route. In addition, the effect of MPE-FEC was obtained graphically for each parameter setting. The analysis was made for the post-processed data by observing the breaking points of BA, FER and MFER of the averaged values of 2 terminals (the other one resulting 2 db lower RSSI). The guard interval (GI) was set to ¼ in each case. The MPE-FEC gain was obtained for each studied case. The effect seems to be lowest for the 64-QAM modulation, which might be an indication of too small amount of collected data in the relatively good field that this mode requires, although 64-QAM seems to be in general very sensible for errors. It should be noted, though, that the terminals were not calibrated especially for this study. The RSSI display might thus differ from the real received power levels with roughly 1-2 decibels. The calibration should be done e.g. by examining first the level of the noise floor of the terminal and secondly examining the QEF point, i.e. investigating the signal level which is just sufficient to provide the correct receiving. Table 1. The results for QPSK cases. The values represent the RSSI in dbm, except for the MPE- FEC gain, which is shown in db. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/2-88,1-83,8-78,4-86,6 5%, FEC 1/2-84,0-77,3-72,7-81,4 MPE-FEC 1/2 gain 4,1 6,5 5,7 5,2 5%, MPE-FEC 2/3-87,3-83,0-76,7-84,4 5%, FEC 2/3-83,3-72,9-73,5-81,0 MPE-FEC 2/3 gain 4,0 10,1 3,2 3,4 BA QEF average: -85,8-81,7-78,4-84,9 Table 2. The results for 16-QAM cases. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/2-77,6-61,8-77,4-77,7 5%, FEC 1/2-69,8-61,6-74,2-73,4 MPE-FEC 1/2 gain 7,8 0,3 3,7 4,3 5%, MPE-FEC 2/3-77,0-63,5-75,5-77,0 5%, FEC 2/3-72,1-59,0-71,0-74,7 MPE-FEC 2/3 gain 4,9 4,5 4,5 2,3 BA QEF average: -78,3-73,7-77,2-77,4 Table 3. The results for 64-QAM cases. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/2-59,9-51,6-65,0-67,3 5%, FEC 1/2-59,7-51,6-57,5-65,2 MPE-FEC 1/2 gain 0,2 0,1 7,5 2,1 5%, MPE-FEC 2/3-61,0-51,3-59,5-68,1 5%, FEC 2/3-60,3-50,6-54,5-65,8 MPE-FEC 2/3 gain 0,7 0,7 5,0 2,4 BA QEF average: -60,7-53,0-61,9-68,3 As stated in the WingTV Measurement Guidelines [3], the moving channel produces fast variations already in TU6 channel type (typical urban 6 km/h) in the QEF criterion making the correct interpretation of the bit error rate before Viterbi very challenging. This also leads to the uncertainty of the correct calculation of the bit error rate after Viterbi. For the bit error rate before and after Viterbi, it is not thus necessarily clear how the terminal calculates the BB and BA values especially in the cell edge with high error rates. This phenomena can be noted in the field test results as the breaking points of the QEF does not necessarily map to the corresponding FER / MFER criteria of 5% or near of it. It can thus be assumed that the most reliable results are obtained by observing the FER / MFER of the data, because their error detection is carried out after the whole demodulation and decoding process. Furthermore, especially the frame error rate reflects the practical situation as the user interpretation of the quality of the audio / video contents depends directly on the amount of correctly received frames. Nevertheless, the results correlate with the theory of different parameter settings, as well as with the MPE- FEC gain although it varies largely in the obtained results depending on the mode. As the test route contained different radio channel types (different vehicle speeds, LOS, near-los and non-los behind the building), the mix of the propagation types causes this effect to the results. In order to obtain the values nearer to the theoretical ones, it is important to carry out the test cases in separate, uniform areas as the radio channel type is considered, but on the other hand, these results represent the real situation in the investigated area with a practical mix of radio channel types Segmented route in main lobe The following Tables 4 and 5 shows the analysis for selected parameter settings of the terminal 1 (with 2 db lower RSSI display compared to others) in order to present the principle of the segmented analysis in the main lobe with around dbm of RSSI and the

291 terminal speed of about 80 km/s. In case the breaking points could not be interpreted explicitly from the graphics, N/A was marked to Tables. Table 4. The results for the QPSK cases in the main lobe. The channel type is nearly LOS and the terminal speed is 80 km/h. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/ %, FEC 1/ MPE-FEC 1/2 gain %, MPE-FEC 2/3 N/A %, FEC 2/3 N/A MPE-FEC 2/3 gain N/A BA QEF average: Table 5. The results of the setup for the 16-QAM cases. FFT 8k 8k 4k 2k CR 1/2 2/3 2/3 2/3 5%, MPE-FEC1/2 N/A N/A %, FEC 1/2 N/A N/A MPE-FEC 1/2 gain N/A N/A %, MPE-FEC 2/ N/A %, FEC 2/ N/A MPE-FEC 2/3 gain 6.7 N/A BA QEF average: It is now interesting to observe the behaviour of the FEC and MFEC values in this single radio channel type. The following Figures summarises the performance of the investigated parameter set in the main lobe in the graphical format. -65 FER 5% and MFER 5%, RSSI per parameter set, terminal 1 QPSK, CR 1/2, MPE 1/2, FFT 8k 16-QAM, CR 2/3, MPE 2/3, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 2k 16-QAM, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 2/3, FFT 4k QPSK, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 1/2, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 1/2, FFT 8k QPSK, CR 2/3, MPE 1/2, FFT 2k dbm Figure 28. The summary of FER5 and MFER5 analysis. It can be noted that the MPE-FEC functions more efficiently in this specific fragmented case, i.e. in the uniform radio channel of 80 km/h in cell edge area compared to the results obtained by analysing the whole test area with fixed radio channel types. As a comparison, the difference of the QEF point and FER / MFER can now be obtained as shown in Figures 29 and 30. In this case, it seems that the breaking point of QEF occurs somewhere between the FER5 and MFER5 values. As stated before, the QEF calculation is not necessarily as reliable as FER and MFER can show, but nevertheless, it is interesting to note that in this specific setup the QEF breaking point is within ±1.6 db margin if we take simply the average of the FER and MFER values. QEF indicates thus the RSSI limits roughly in the same range as the FER and MFER does QAM, CR 2/3, MPE 2/3, FFT 8k 16-QAM, CR 2/3, MPE 1/2, FFT 8k 16-QAM, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 2/3, MPE 2/3, FFT 4k QPSK, CR 2/3, MPE 2/3, FFT 4k 16-QAM, CR 1/2, MPE 1/2, FFT 8k 16-QAM, CR 2/3, MPE 1/2, FFT 2k 16-QAM, CR 2/3, MPE 2/3, FFT 2k 16-QAM, CR 1/2, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 1/2, FFT 4k QPSK, CR 2/3, MPE 1/2, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 1/2, FFT 2k QPSK, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 1/2, MPE 1/2, FFT 8k QPSK, CR 1/2, MPE 2/3, FFT 8k -60 QEF, RSSI per parameter set, terminal dbm Figure 27. The summary of the QEF breaking points for the investigated parameter sets QAM, CR 2/3, MPE 2/3, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 2k 16-QAM, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 2/3, FFT 4k QPSK, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 1/2, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 1/2, FFT 8k QPSK, CR 2/3, MPE 1/2, FFT 2k Diff(QEF/MFER5%) Figure 29. The difference of the QEF breaking points compared to the MFER5 results.

292 Diff(QEF/FER5%) 16-QAM, CR 2/3, MPE 2/3, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 2/3, MPE 1/2, FFT 2k 16-QAM, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 2/3, FFT 4k QPSK, CR 2/3, MPE 1/2, FFT 4k 16-QAM, CR 1/2, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 8k QPSK, CR 2/3, MPE 2/3, FFT 2k QPSK, CR 2/3, MPE 1/2, FFT 8k QPSK, CR 2/3, MPE 1/2, FFT 2k Figure 30. The difference of the QEF breaking points compared to the FER5 results. 10. Conclusions The benefit of the hand-held receiver is obvious for the measurements presented in this paper as the equipment is easy to carry to different environments, including indoors. The data collection with hand-held terminal is fast, and the collected radio interface performance indicators provide sufficiently data for the post-processing. The tests presented in this paper shows that the realistic DVB-H measurement data can be collected with the terminals. The analysis showed correlation between the post-processed data and estimated coverage that was calculated and plotted separately with a network planning tool. The results correlate mostly with the theoretical DVB-H performance, although there was a set of uncertainty factors identified that affects on the accuracy of the results. This study was meant to develop and verify the functionality of the methodology for measurements, post-processing and analysis, and as a secondary result, it also gave performance values for selected parameter set. The test environment consisted of multiple radio channel types, and the terminal displays were not calibrated specifically for these tests. An error of 1-2 decibels in RSSI values is thus expected. Nevertheless, the results show that the terminals can be used as an additional tool for fast revision of the overall functioning of the network. With the collected data and respective post-processing, it is possible to observe the DVB-H audio / video quality in detailed level compared to the subjective studies. The field test results show clearly the effect of the parameter values on radio performance in a typical sub-urban environment. Even if the hand-held terminal is not the most accurate device for the scientific purposes, it gives an overview about the general functioning and quality level of the network and the estimation of the effects of different network parameter settings. The vehicle related test cases showed a logical functioning of MPE-FEC for different parameter settings. The results are in align with the theories especially when the analyzed data is limited to a single radio channel type in coverage edge of the main lobe with nearly LOS situation and with a vehicle speed of about 80 km/h. This case showed relatively significant effect of the MPE-FEC functionality giving a gain of up to 7 db, which thus enhances the link budget and extends the coverage area compared to the basic FEC. On the other side, in the good field, the MPE-FEC is not needed as there are no frame errors present. The whole test route can be analyzed also as one complete case, giving the overall information about the network performance in variable radio channels. It should be noted though that this case does not give comparable values for the performance measurements like the case is for the single radio channel. The pedestrian test cases showed that the building loss can be obtained in easy way with the test setup. The analysis also showed that the MPE-FEC does not take place with such a low speeds even in low field strengths, if there is lack of reflected multi-path propagated components. As the city center areas contains normally relatively good field and their main usage can be estimated to be pedestrian cases, the results indicates that the MPE-FEC setting could thus be relatively light in city areas, in order of 3/4 or 5/6, which gives still some MPE-FEC gain for sufficiently fast moving terminals, yet saving capacity. Regardless of the Excel sheet functionality that was developed for post-processing the data, there was a considerable amount of manual procedures in order to obtain the final results. As a possible future work item, basically all of the manual work can be automated by creating respective macro functionality for transferring the data from the terminal to the processing unit, to organize the data in function of the RSSI values, and to present the analysis in graphical and numerical format. Based on the methodology presented in this paper, the processing of the data is independent of the terminal type, but the special characteristics should be taken into account in the result tuning, as well as the proper calibration of the equipment. Furthermore, if the interface between the terminal and data processing unit supports proper protocols for fetching the data during the measurements, the post-processing and display can be performed in real time whilst carrying out the drive tests.

293 11. References [1] DVB-H Implementation Guidelines. Draft TR V1.2.2 ( ). European Broadcasting Union. 108 p. [2] Jukka Henriksson. DVB-H standard, principles and services. HUT seminar T Helsinki, Presentation material. 53 p. [3] Editor: Thibault Bouttevin. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Measurement Guidelines & Criteria. Project report. 45 p. [4] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D6 Wing TV Common field trials report. Project report, November p. [5] Editor: Maite Aparicio. Wing TV. Services to Wireless, Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D8 Wing TV Country field trial report. Project report, November p. [6] Editor: Davide Milanesio. Wing TV. Services to Wireless Integrated, Nomadic, GPRS-UMTS&TV handheld terminals. D11 WingTV Network Issues. Project report, May p. [7] Gerard Faria, Jukka A. Henriksson, Erik Stare, Pekka Talmola. DVB-H: Digital Broadcast Services to Handheld Devices. IEEE p. [8] William C.Y. Lee. Elements of Cellular Mobile Radio System. IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 2, May pp [9] Limits of Human Exposure to Radiofrequency Electromagnetic Fields in the Frequency Range from 3 khz to 399 GHz. Safety Code 6. Environmental Health Directorate, Health Protection Branch. Publication 99- EHD-237. Minister of Public Works and Government Services, Canada ISBN p. [10] Masaharu Hata. Empirical Formula for Propagation Loss in Land Mobile Radio Services. IEEE Transactions on Vehicular Technology, Vol. VT- 29, No. 3, August p. [11] Recommendation ITU-R P Method for point-to-area predictions for terrestrial services in the frequency range 30 MHz to 3000 MHz p. [12] Jyrki T.J. Penttinen. Field Measurement and Data Analysis Method for DVB-H Mobile Devices. The Third International Conference on Digital Telecommunications, p. [13] Transmission System for Handheld Terminals (DVB-H). ETSI EN V1.1.1 ( ). 14 p. [14] Tero Jokela, Eero Lehtonen. Reed-Solomon Decoding Algorithms and Their Complexities at the DVB-H Link-Layer. IEEE p. [15] Teodor Iliev et al. Framing Structure, Channel Coding and Modulation for Digital Terrestrial Television. ETSI EN V1.5.1 ( ). IEEE p. [16] Heidi Joki, Jarkko Paavola. A Novel Algorithm for Decapsulation and Decoding of DVB-H Link Layer Forward Error Correction. Department of Information Technology, University of Turku, p. Biography Mr. Jyrki T.J. Penttinen has worked in telecommunications area since 1994, for Telecom Finland and it s successors until 2004, and after that, for Nokia and Nokia Siemens Networks. He has carried out various international tasks, e.g. as a System Expert and Senior Network Architect in Finland, R&D Manager in Spain and Technical Manager in Mexico and USA. He currently holds a Senior Solutions Architect position in Madrid, Spain. His main activities have been related to mobile and DVB-H network design and optimization. Mr. Penttinen obtained M.Sc. (E.E.) and Licentiate of Technology (E.E.) degrees from Helsinki University of Technology (TKK) in 1994 and 1999, respectively. He has organized actively telecom courses and lectures. In addition, he has published various technical books and articles since His main books are GSM-tekniikka ( GSM Technology, published in Finnish, Helsinki, Finland, WSOY, 1999), Wireless Data in GPRS (published in Finnish and English, Helsinki, Finland, WSOY, 2002), and Tietoliikennetekniikka ( Telecommunications technology, published in Finnish, Helsinki, Finland, WSOY, 2006).

294

295 X Publication X The SFN gain in non-interfered and interfered DVB-H networks This paper is an extended version of the Publication IV. The original conference paper received a Best Paper Award of IARIA. This extended version presents more thoroughly the simulations and respective results by explaining the methodology applied in Publications III, IV and VII. International Journal on Advances in Internet Technology, ISSN , Vol. 2, No. 1, 2009, pages IARIA Reprinted with permission.

296

297 The SFN Gain in Non-Interfered and Interfered DVB-H Networks Jyrki T.J. Penttinen Member, IEEE Abstract The DVB-H (Digital Video Broadcasting, Handheld) coverage area depends mainly on the area type, i.e. on the radio path attenuation, as well as on the transmitter power level, antenna height and radio parameters. The latter set has effect also on the audio / video capacity. In the detailed network planning, not only the coverage itself is important but the quality of service level should be dimensioned accordingly. This paper is based on [14] and describes the SFN gain related items as a part of the detailed radio DVB- H network planning. The emphasis is put to the effect of DVB-H parameter settings on the error levels caused by the over-sized Single Frequency Network (SFN) area. In this case, part of the transmitting sites converts to interfering sources if the safety distance margin of the radio path is exceeded. A respective method is presented for the estimation of the SFN interference levels. The functionality of the method was tested by programming a simulator and analyzing the variations of carrier per interference distribution. The results show that the theoretical SFN limits can be exceeded e.g. by selecting the antenna height in optimal way and accepting certain increase of the error level that is called SFN error rate (SER) in this paper. Furthermore, by selecting the relevant parameters in correct way, the balance between SFN gain and SER can be planned in controlled way. 1. Introduction The DVB-H is an extended version of the terrestrial television system, DVB-T. Both are defined in the ETSI standards along with the satellite and cable versions of the DVB. The mobile version of DVB suits especially for the moving environment as it has been optimized for the fast variations of the field strength and different terminal speeds. Furthermore, DVB-H is suitable for the delivery of various audio / video channels in a single bandwidth, and the small terminal screen shows adequately the lower resolution streams compared to the full scale DVB-T. As DVB-H is meant for the mobile environment, the respective terminals are often used on a street level for the reception. This creates a significant difference in the received power level compared to DVB-T which uses fixed and directional rooftop antenna types. Furthermore, the DVB-H terminal has normally only small, in-built panel antenna, which is challenging for the reception of the radio signals. In the case of DVB-H, the factors affecting on the quality are mainly related to the radio interface due to the variation of the received signal levels as well as the Doppler shift, i.e. the speed of the terminal. The DVB-H service can be designed using either Single Frequency Network (SFN) or Multi Frequency Network (MFN) modes. In the former case, the transmitters can be added within the SFN area without co-channel interferences even if the cells of the same frequency overlap. In fact, the multi-propagated SFN signals increases the performance of the network by producing SFN gain. In MFN mode, the frequency hand-over is performed each time the terminal moves from the coverage area of one site to another and no SFN gain is achieved in this mode. The most logical way to build up the DVB-H network is to use separate SFN isles covering e.g. single cities, so the practical network consists of both SFN and MFN solutions. Especially in the Single Frequency Network, the coverage planning is straightforward as long as the maximum distance of the sites does not exceed the allowed value defined by the guard interval (GI). The guard interval takes care of the safe reception of the multi-path propagated signals originated from various sites or due to the reflected radio waves. If the GI and FFT dependent geographical SFN boundary is exceeded, part of the sites starts to act as interferers instead of providing useful carrier. The maximum size of the non-interfered Single Frequency Network of DVB-H depends on the guard interval and FFT mode. The distance limitation

298 between the extreme transmitter sites is thus possible to calculate in ideal conditions. Nevertheless, there might be need to extend the theoretical SFN areas e.g. due to the lack of frequencies. Sites that are located within the SFN area minimises the effect of the inter-symbol-interferences as the guard interval protects the OFDM signals of DVB-H, although in some cases, sufficiently strong multipath signals reflected from distant objects might cause interferences in tightly dimensioned network. On the other side, if certain degradation in the quality level of the received signal is accepted, it could be justified to even extend the SFN limits. This paper presents a simulation method that was developed for estimating the SFN interference levels as well as the SFN gain. Case studies were carried out by utilizing a set of DVB-H radio parameters. The simulator shows the variations in the carrier per noise and interference levels, C / (N + I), in function of related radio parameters in over-sized SFN. The additional error rate caused by the exceeding of SFN is called SFN error rate, or SER, in this paper. 2. DVB-H Dimensioning 2.1. Capacity Planning In the initial phase of the DVB-H network planning, the offered capacity of the system is dimensioned. The total capacity in certain DVB-H band defined as 5, 6, 7, or 8 MHz does have effect also on the size of the coverage area. The dimensioning process is thus iterative, with the aim to find a balance between the capacity, coverage and the cost of the network. The capacity can be varied by tuning the modulation, guard interval, code rate and channel bandwidth. As an example, the parameter set of QPSK, GI ¼, code rate ½ and channel bandwidth 8 MHz provides a total capacity of 4.98 Mb/s, which can be divided between one or more electronic service guides (ESG) and various audio / video sub-channels with typically around kb/s bit stream dedicated for each. The capacity does not depend on the number of carriers (FFT mode) but the selected FFT affects though on the Doppler shift tolerance. As a comparison, the parameter set of 16-QAM, GI 1/32, code rate 7/8 and channel bandwidth of 8 MHz provides a total capacity of 21.1 Mb/s. It should be noted, though, that the latter parameter set is not practical due to the clearly increased C/N requirement. The relation between the radio parameter values, Doppler shift tolerance and capacity can be investigated more thoroughly in [1] Coverage Planning When the coverage criteria are known, the cell radius can be estimated by applying the link budget calculation. The generic principle of the DVB-H link budget can be seen in the following Table 1. The calculation shows an example of the transmitter output power level of 2,400 W, with the quality value of 90% for the area location, but assuming the SFN gain does not exist. According to the link budget, the outdoor reception of this specific case yields a successful reception when the radio path loss is equal or less than db. Table 1. An example of DVB-H link budget. The Okumura-Hata model [10] can be applied in order to obtain the estimation for the cell radius (unit in kilometres) e.g. in large city type: L( db) = lg( f ) lg( h a( h Parameters Symbol Value General parameters Frequency f MHz Noise floor for 6 MHz BW P n dbm RX noise figure F 5.2 db Transmitter (TX) Transmitter output power P TX 2,400.0 W Transmitter output power P TX 63.8 dbm Cable and connector loss L cc 3.0 db Power splitter loss L ps 3.0 db Antenna gain G TX 13.1 dbi Antenna gain G TX 11.0 dbd Eff. Isotropic Radiating Power EIRP 70.9 dbm Eff. Isotropic Radiating Power EIRP 12,308.7 W Eff. Radiating Power ERP 68.8 dbm Eff. Radiating Power ERP 7,502.6 W Receiver (RX) Min. C/N for the used mode (C/N) min 17.5 db Sensitivity P RXmin dbm Antenna gain, isotropic ref. G RX -7.3 dbi Antenna gain, ½ wave dipole G RX -5.2 dbd Isotropic power P i dbm Loc. variation. L iv 7.0 db Building loss L b 14.0 db GSM filter loss L GSM 0.0 db Min. req. received power outd. P min(out) dbm Min. req. received power ind. P min(in) dbm Min. req. field strength outd. E min(out) 64.5 dbµv/m Min. req. field strength ind. E min(in) 78.5 dbµv/m Maximum path loss, outdoors L pl(out) db Maximum path loss, indoors L pl(in) db MS ) type + [ lg( h )] lg( d) BS BS ) (1)

299 For f 400 MHz, the area type factor is: [ ( )] a ( h MS ) LC = 3.2 lg 11.75h (2) MS [ lg( f ) lg( hbs ) a( hms ) ] L( db) i lg( hbs ) d = 10 (3) The following Figure 1 presents the estimated cell range of the example that is calculated with the large city model and by varying the transmitter antenna height and power level. As can be noted, the antenna height has major impact on the cell radius compared to the transmitter power level. d (km) Cell radius Antenna height (m) P=1500W P=2400W P=3400W P=4700W Figure 1. The cell range calculated with the Okumura-Hata model for the large city, varying the transmitter power levels. In the SFN related reference material, different values for the SFN gain is proposed to be added to the DVB-H link budget, typically from 0 to 3 db. As an example, [5] mentions that due to the large standard variation of the combined signal reception in environment with multi-path propagation, no SFN gain is recommended for the planning criteria. In the same document, a SFN gain of around 1.5 db was obtained for the 2 transmitter case by carrying out field measurements. A 3-transmitter field measurement case can be found in [9], which shows that about 2 db SFN gain was achieved. For the large amount of sites, no practical field results can be found due to the complexity of the test setup. Nevertheless, the possible SFN gain of e.g. 2 db would have around same effect on the coverage area growth as changing the 1500 W transmitter to 2400 W power level category, i.e. there can be an important cost effect in selected areas of the DVB-H network depending on the functionality of the SFN gain Theory of the SFN Limits The DVB-H radio transmission is based on the OFDM (Orthogonal Frequency Division Multiplexing). The idea of the technique is to create various separate data streams that are delivered in sub-bands. The error correction is thus efficient as the sufficiently high-quality sub-bands are used for the processing of the received data, depending on the level of the correction schemes used in the transmission. In order to work, the system needs to minimize the interference levels between the sub-bands. The subband signals should thus be orthogonal. In order to comply with this requirement, the carrier frequencies are selected in such way that the spacing between the adjacent channels is the inverse of symbol duration. According to [1], the GI and the FFT mode determinate the maximum delay that the mobile can handle for receiving correctly the multi-path components of the signals. The Table 2 summarises the maximum allowed delays and respective distances. The maximum allowed distance per parameter setting has been calculated assuming the radio signal propagates with the speed of light. Table 2. The guard interval lengths and respective safety distances. GI FFT = 2K FFT = 4K FFT = 8K 1/4 56 µs / 16.8 km 112 µs / 33.6 km 224 µs / 67 km 1/8 28 µs / 8.4 km 56 µs / 16.8 km 112 µs / 33.6 km 1/16 14 µs / 4.2 km 28 µs / 8.4 km 56 µs / 16.8 km 1/32 7 µs / 2.1 km 14 µs / 4.2 km 28 µs / 8.4 km As long as the distance between the extreme transmitter sites is less than the safety margin dictates, the difference of the delays between the signals originated from different sites never exceeds the allowed value unless there is a strong multipath propagated signal present. TX1 D 1 <D sfn SFN area D 2 <D sfn TX2 Figure 2. If all the sites of certain frequency band are located inside the SFN area with the distance between the extreme sites less than D sfn, no intersymbol interferences are produced.

300 On the other hand, when the terminal drifts outside of the original SFN area and receives sufficiently strong signals from the original SFN, no problems arises either in this case as it can be shown that the difference of the signal delays from the respective SFN sites are always within the safety limits. SFN area including the extreme radio coverage edges TX1 Sites are within this physical (geographical) SFN area Delay 1 Terminal inside of SFN area Distance TX1-TX2 TX2 Terminal outside of physical SFN area but within the radio coverage of SFN Figure 3. The GI applies also outside the physical SFN cell area where the signal level originated form the SFN is sufficiently high. The situation changes if the inter-site distance exceeds the allowed theoretical value. As an example, GI of ¼ and 8K mode provide 224 µs margin for the safe propagation delay. Assuming the signal propagates with the speed of light, the SFN size limit is 300,000 km/s 224 µs yielding about 67 km of maximum distance between the sites. If any geographical combination of the site locations using the same frequency exceeds this maximum allowed distance, they start producing interference in those spots where the difference of the arriving signals is higher than 224 µs. If the level of interference is greater than the noise floor, and the minimum C/N value that the respective mode required in non-interfered situation is not any more obtained, the signal in that specific spot is interfered and the reception suffers from the frame errors that disturb the fluent following of the contents. In order to achieve correct reception, the additional interference increases the required received power level of the carrier to C/N C / (N + I). The Figure 4 shows that if the D eff, i.e. the difference between the signals arriving from the sites, is more than the allowed safety distance in over-sized SFN, the site acts as an interferer. Whilst the carrier per interference and noise level from the TX 2 complies with the minimum requirement for the C/N, the transmission is still useful. Delay 2 MS D eff =Delay1-Delay2 Delay 1 Delay 2 Figure 4. When the mobile station is receiving signals from the over-sized SFN-network, the sum of the interfering signals might destroy the reception if their level is sufficiently high compared to the sum of the useful carrier levels. Even if the C / (N + I) level gets lower when the terminal moves from one site to another, the situation is not necessarily critical as the effective distance D eff of the signals might be within the SFN limits e.g. in the middle of two sites, although their distance from each others would be greater than the maximum allowed. In other words, the otherwise interfering site might not be considered as interference in the respective spot but it might give SFN gain by producing additional carrier C 2. This phenomenon can be observed in practice as the SFN interferences tends to accumulate primarily in the outer boundaries of the network. The Figure 5 shows the principle of the relative interference which increases especially when the terminal moves away from the centre of the SFN network. I/TX1 Noise floor SFN limit C/TX2 Carrier from TX2 C/(N+I) MS Interference (I) from TX1 Noise (N) Useful coverage area of TX2 defined by C/(N+I) D/km Figure 5. The principle of interference when the location of transmitter TX 1 is out of the SFN limit. When moving outside of the network, the relative difference between the carrier and interfering signal gets smaller and it is thus inevitable that the C / (N + I) will not be sufficient any more at some point for the correct reception of the carrier, although the C/N level without the presence of interfering signal would still be sufficiently high. The essential question is thus, where the critical points are found with lower C / (N + I)

301 value than the original requirement for C/N is, and where the interference thus converts active, i.e. when D eff is longer than the safety margin. As an example, the distance of two sites could be 70 km, which is more than D sfn with any of the radio parameter combination of DVB-H. For the parameter set of FFT 8k and GI 1/8, the safety distance for D sfn is about 34 km, which is clearly less than the distance of these sites. Let s define the radiating power (EIRP) for each site to +60 dbm. We can now observe the received power level of the sites in the theoretical open area by applying the free space loss, f representing the frequency (MHz) and d the distance (km): L = 20 log f + 20logd (4) The Figure 6 shows the carrier (or interference) from TX 1 located in 0 km and carrier (or interference) from TX 2 located in 70 km, when the parameter set allows D sfn of 34 km. The interference is included in those spots where the D eff is higher than D sfn. If the D eff is shorter than D sfn, the respective received useful power level is shown taking into account the SFN gain of these two sites by summing the absolute values of the power levels: Carrier level (db) C tot = + (5) C 1 C2 C/N (TX1) C/N (TX2) C/(N+I) combined 10.0 D (km) Figure 6. The combined C/(N+I) along the route from 0 to 100 km, taking into account the SFNgain when the interference is not present. As the Figure 6 shows, the TX 1 is acting as a carrier and TX 2 as interferer from 0 km (TX 1 location) to 18 km, because the D eff > D sfn. Nevertheless, the carrier of TX 1 is dominating within this area in order to provide sufficiently high C / (N + I) for the successful reception for the QPSK, CR ½ and MPE-FEC ½ as it requires 8.5 db. The segment of 18 km to 52 km is clear from the interferences as all the D eff < D sfn, and in addition, the receiver gets SFN gain from the combined carriers of TX 1 and TX 2. The TX 1 starts to act as an interferer from 52 km to 100 km (or, until the C/N limit of the used mode). Nevertheless, the interference of TX 1 is already so attenuated such a far away from its origin that the C / (N + I) is high enough for the successful reception from TX 2 of above mentioned QPSK still within the area of km. With any other parameter settings, the SFN interference level is high enough to affect on the successful reception in these breaking points where the D eff makes the signal act as interferer instead of carrier. As can be seen from this example, the interference level takes place when the terminal moves towards the boundaries or boundary sites of the network. As a result, the boundary site s coverage area gets smaller, and depending on the parameter setting, there will be interferences between the sites. The required C/N for some of the most commonly used parameter setting can be seen in Table 3. [1] The terminal antenna gain (loss) is taken into account in the presented values. The Table present the expected C/N values in Mobile TU-channel (typical urban) for the "possible" reference receiver. Table 3. The minimum C/N (db) for the selected parameter settings. Parameters C/N QPSK, CR 1/2, MPE-FEC 1/2 8.5 QPSK, CR 1/2, MPE-FEC 2/ QAM, CR 1/2, MPE-FEC 1/ QAM, CR 1/2, MPE-FEC 2/ The FFT size has impact on the maximum velocity of the terminal, and the GI affects on both the maximum velocity as well as on the capacity of the radio interface. In fact, in these simulations, if only the requirement for the level of carrier is considered without the need to take into account the maximum functional velocity of the terminal or the radio channel capacity, the following parameter combinations results the same C/N and C / (N + I) performance due to their same requirement for the safety distances: FFT 8K, GI 1/4: only one set FFT 8K, GI 1/8: same as FFT 4K, GI 1/4 FFT 8K, GI 1/16: same as FFT 4K, GI 1/8 and FFT 2K, GI 1/4 FFT 8K, GI 1/32: same as FFT 4K, GI 1/16 and FFT 2K, GI 1/8 FFT 4K, GI 1/32: same as FFT 2K, GI 1/16 FFT 2K, GI 1/32: only one set

302 4. Methodology for the SFN simulations: first variation (unlimited SFN network) 4.1. General In order to estimate the error level of various sites that is caused by extending the theoretical geometrical limits of SFN network, a simulation can be carried out as presented in [13]. For the simulation, the investigated variables can be e.g. the antenna height and power level of the transmitter, in addition to the GI and FFT mode that defines the SFN limits. The setup for the simulation consists of radio propagation type and geometrical area where the cells are located. The most logical way is to dimension the network according to the radio interface parameters, i.e. the cell radius should be dimensioned according to the minimum C/N requirement. For this, a link budget calculator is included to the initial part of the simulator. It estimates the radius for both useful carriers as well as for the interfering signals, noise level being the reference. Depending on the site definitions, there might be need to apply other propagation models as the basic Okumura-Hata [10] is valid for the maximum cell radius of 20 km and antenna heights up to 200 m. One of the suitable models for the large cells is ITU-R P.1546 [11], which is based on the interpolation of the pre-calculated curves. When estimating the total carrier per interference levels, both total level of the carriers and interferences can be calculated separately by the following formulas, using the respective absolute power levels (W) for the C and I components: C tot = C1 + C C (6) n I tot = I1 + I I (7) n In each simulation round, the site with the highest field strength is identified. In case of uniform network and equal site configurations, the site with lowest propagation loss corresponds to the nearest cell TX 1 which is selected as a reference. Once the nearest cell is identified, the task is to investigate the propagation delays of signals between the nearest and each one of the other sites, and calculate if the difference of arriving signals D eff is greater or lower than the SFN limit D sfn. In general, if the difference of the signal arrival times of TX 1 and TX n is greater than GI defines, the TX n is producing interfering signal (if the signal is above the noise floor), and otherwise it is adding the level of total carrier energy (if the signal level is above the minimum requirement for carrier). In order to obtain the level of C and I in certain area type, the path loss can be estimated e.g. with Okumura-Hata radio propagation model or ITU-R P The total path loss can be calculated by applying the following formula: L = L + L + L (8) tot pathloss norm other L norm represents the fading loss caused by the longterm variations, and other losses may include e.g. the fast fading as well as antenna losses. For the long-term fading, a normal distribution is commonly used in order for modelling the variations of the signal level. The PDF of the long-term fading is the following [5]: 2 ( x x) 1 PDF( L ) = exp 2 (9) norm 2πσ 2σ The term x represents the loss value, and x is the average loss (0 in this case). In the snap-shot based simulations, the L norm is calculated for each arriving signal individually as the different events does not have correlation. The respective PDF and CDF are obtained by creating a probability table for normal distributions. Figure 7 shows an example of the PDF and CDF of normal distributed loss variations when the mean value is 0 and standard deviation is 5.5 db PDF and CDF of normal distribution, stdev=5.5 PDF CDF db Figure 7. PDF and CDF of the normal distribution representing the variations of long-term loss when the standard deviation is set to 5.5 db. The fast (Rayleigh) fading is present in those environments where multi-path radio signals occurs, e.g. on the street level of cities. It can be presented with the following PDF:

303 L log norm 2 x ( ) 2 2δ x = e (10) 2 δ The Figure 8 shows the PDF and CDF of the fast fading representing the variations of short-term loss when the standard deviation is set to 5.5 db. Inputs: Geographical and simulation area, radio parameters. Tables: Create CDF of lognormal distribution for long-term fading and optional CDF for Rayleigh fading. Initialisation: Calculate the single cell radius with Okumura-Hata and place the sites on map according to the hexagonal principles. PDF PDF and CDF of log-normal distribution, stdev=5.5 PDF CDF db CDF Figure 8. PDF and CDF of the log-normal distribution for fast fading Simulator A block diagram of the presented SFN interference simulator is shown in the following Figure 9. The simulator was programmed with a standard Pascal code. It produces the results to text files, containing the C/N, I/N and C / (N + I) values showing the distribution in scale of db and with 0.1 db resolution, using integer type table indexes of -500 to +500 that represents the occurred cumulative values. Also the terminal coordinates and respective C/N, I/N and C / (N + I) for all the simulation rounds is produced. If the value occurs outside the scale, it is added to the extreme db categories in order to form the CDF correctly. A total of 60,000 simulation rounds per each case were carried out. It corresponds to an average of 60,000 / (50 db 10) = 120 samples per C/I resolution, which fulfils the accuracy of the binomial distribution. Each text file was post-processed and analysed with Microsoft Excel. The terminal was placed in 100 km 100 km area according to the uniform distribution in function of the coordinates (x, y) during each simulation round. The raster of the area was set to 10 m. Small and medium city area type was selected for the simulations. The total C/I value is calculated per simulation round by observing the individual signals of the sites. Simulation: Place the mobile station on map. Calculate the respective C/I levels. Repeat the simulation rounds until the statistical accuracy has been reached. Data storing: Save the C/I PDF and CDF distribution ( db, 0.1 db resolution), coordinates of MS in each simulation round and respective C/I value. Figure 9. The simulator s block diagram. The nearest site is selected as a reference during the respective simulation round. If the arrival time delays difference t 2 t 1 is less than D sfn defines, the respective signal is marked as useful carrier C, or otherwise it is marked as interference I. In the generic format, the total C / (N + I) can be obtained from the simulation results in the following way: C N = ( Ctot[ db] noisefloor[ db] ) + I ( I [ db] noisefloor[ db] ) tot (11) The term N represents the reference which is the sum of noise floor and terminal noise figure. The noise figure depends on the terminal characteristics. In the simulations, it was estimated to 5 db as defined in [1]. The simulator calculates the expected radius of single cell in non-interfering case and fills the area with uniform cells according to the hexagonal model. This provides partial overlapping of the cells. Each simulation round provides information if that specific connection is useless, e.g. if the criteria set of 1) effective distance D eff > D sfn in any of the cells, and 2) C / (N + I) < minimum C/N threshold. If both criteria are valid, and if the C/N would have been sufficiently high without the interference in that specific round, the SFN interference level is calculated. The Figure 10 shows an example of the site locations. As can be seen, the simulator calculates the optimal cell radius according to the parameter setting and locates the transmitters on map according to the hexagonal model, leaving ideal overlapping areas in the cell border areas. The size and thus the number of the cells depends on the radio parameter settings without interferences, and in each case, a result is a uniform service level in the whole investigated area. The same network setup is used throughout the

304 complete simulation, and changed if the radio parameters of the following simulation require so. km TX locations km Figure 10. Example of the transmitter site locations the simulator has generated. The behaviour of C and I can be investigated by observing the probability density functions, i.e. PDF of the results. Nevertheless, the specific values of the interference levels can be obtained by producing a CDF from the simulation results. The following Figure 11 shows two examples of the simulation results in CDF format. In this specific case, the outage probability of 10% (i.e. area location probability of 90 %) yields the minimum C / (N + I) of 10 db for 8K, which complies with the original C/N requirement (8.5 db) of this case. On the other hand, the 4K mode results about 7 db with 10% outage, which means that the minimum quality targets can not quite be achieved any more with these settings Results By applying the principles of the DVB-H simulator, the C/I distribution was obtained according to the selected radio parameters. The variables were the modulation scheme (QPSK and 16-QAM), antenna height ( m) and FFT mode (4K and 8K). The following Figures show the resulting networks that were used as a basis for the simulations. The simulator selects randomly the mobile terminal location on the map and calculates the C/I that the network produces at that specific location and moment. This procedure is repeated during 60,000 simulation rounds. One of the results after the complete simulation is the estimation for the occurred errors due to the interfering signals from the sites exceeding the safety distance (i.e. if the arrival times of the signals exceed the maximum allowed delay difference). This event can be called SFN error rate, or SER. # of sites Site number and cell radius for QPSK # sites r_c r_i Ant h (m) Figure 12. The network dimensions for the QPSK simulations Cell radius (m) Area location probability (%) CDF, QPSK, GI=1/4 C/(I+N)_QPSK_8K 0.1 C/(I+N)_QPSK_4K C/(N+I) (db) Figure 11. Example of the cumulative distribution of C/(N+I) for QPSK 4K and 8K modes with antenna height of 200 m and Ptx +60 dbm. # of sites Site number and cell radius for 16-QAM # sites r_c r_i Ant h (m) Figure 13. The network dimensions for the 16- QAM simulations Cell radius (m)

305 In the Figure 14, the plots indicates the locations where the results of C / (N + I) corresponds 8.5 db or less for QPSK. In this case, the interfering plots represent the relative SFN area error rate (SAER) of 0.83%, i.e. the erroneous (SAR) cases over the number of total simulation rounds as for the simulated plots C/I (db) C/I w ith outage of 2 10%, QPSK, 8K,100x100km2 Out_2% Out_5% Out_10% y/km Antenna height (m) Figure 15. The summary of the case 1 (QPSK, 8K). The results show the C/(N+I) with 2%, 5% and 10 % SER criteria x/km Figure 14. An example of the results in geographical format with C/(N+I)<8.5 db. C/I (db) C/I w ith outage of 2 10%, 16-QAM, 8K, 100x100km2 Out_2% Out_5% Out_10% It can be assumed that when the SER level is sufficiently low, the end-users will not experience remarkable reduction in the DVB-H reception due to the extended SFN limits. In this analysis, a SER level of 5% is assumed to still provide with sufficient performance as it is in align with the limits defined in [1] for frame error rate before the MPE-FEC (FER) and frame error rate after the MPE-FEC (MFER). The nature of the SER is slightly different, though, as the interferences tend to cumulate to certain locations as can be observed form the Figure 14 obtained from the simulator. According to the simulations, the SFN interference level varies clearly when the radio parameters are tuned. The following Figures summarises the respective SFN area error analysis, the variable being the transmitter antenna height. The Figures shows that with the uniform radio parameters and varying the antenna height, modulation and FFT mode, the functional settings can be found regardless of the exceeding of the theoretical SFN limits. If a 5% limit for SER is accepted, the analysis show that antenna height of about 80 m or lower produces SER of 5% or less for QPSK, 8K, with minimum C / (N + I) requirement of 8.5 db. If the mode is changed to 4K, the antenna height should be lowered to m from ground level in order to comply with 5% SER criteria in this very case. 16-PSK produces higher capacity and smaller coverage Antenna height (m) Figure 16. The summary of the case 2 (16-QAM, 8K). C/I (db) C/I w ith outage of 2 10%, QPSK, 4K, 100x100km Antenna height (m) Out_2% Out_5% Out_10% Figure 17. The summary of the case 3 (QPSK, 4K).

306 C/I (db) C/I w ith outage of 2 10%, 16-QAM, 4K, 100x100km2 Out_2% Out_5% Out_10% possible. Also the natural obstacles of the environment can be used efficiently for limiting the interferences far away outside the cell range. The following Figure 19 shows the previously presented results presenting the outage percentage for the different modes having 8.5 db C / (N + I) limit for QPSK and 14.5 % for 16-QAM cases in function of transmitter antenna height Antenna height (m) Figure 18. The summary of the case 4 (16-QAM, 4K). The results are showing this clearly as the respective SER of 5% (16-QAM, 8K and minimum C / (N + I) requirement of 14.5 db for this modulation) allows the use of antenna height of about 120 m. If the mode of this case is switched to 4K, the antenna should be lowered down to 50 m in order to still fulfil the SER 5% criteria. The +60 dbm EIRP represents relatively low power. The higher power level raises the SER level accordingly. For the mid and high power sites the optimal setting depends thus even more on the combination of the power level and antenna height. According to these results, it is clear that the FFT mode 8K is the only reasonable option when the SER should be kept in acceptable level. Especially the QPSK modulation might not allow easily extension of SFN as the modulation provides largest coverage areas. On the other hand, when providing more capacity, 16-QAM is the most logical solution as it gives normally sufficient capacity with reasonable coverage areas. The stronger CR and MPE-FEC error correction rate decreases the coverage area but it is worth noting that the interference propagates equally also in those cases. The general problem of the SER arises from the different loss behaviour of the useful carrier and interfering signal. Depending on the case, the interfering signal might propagate 2-3 times further away from the originating site compared to the useful carrier as can be seen from Figures 12 and 13. In practice, the SER level can be further decreased by minimising the propagation of the interfering components. This can be done e.g. by adjusting the transmitter antenna down-tilting and using narrow vertical beam widths, producing thus the coverage area of the carrier and interference as close to each others as Outage (%) Outage-% for antenna heights m Outage%_QPSK,8K Outage%_16-QAM,8K Outage%_QPSK,4K Outage%_16-QAM,4K Antenna height (m) Figure 19. The summary of the cases 1-4 presenting the outage percentage in function of the transmitter antenna height. This version of the simulator gives indication about the behaviour of the C / (N + I) in geographical area. In order to estimate the SFN gain, the individual cells could be switched on and off for the comparison of the differences in overall C / (N + I) distribution. Nevertheless, when the investigated area is filled with the cells, it normally leaves outages in the northern and eastern sides as the area cannot be filled completely as shown in Figure 10. It also produces partial cell areas, if the centre of the site fits into the area but the edge is outside. An enhanced version of the simulator was thus developed in order to investigate the SFN gain in more controlled way, i.e. instead of the fixed area size the method uses the variable reuse pattern sizes. The following Chapter 6 describes the method. 5. Methodology for the simulations: second variation (SFN network with fixed reuse patterns) 5.1. Simulator The second version of the presented SFN performance simulator is based on the hexagonal cell layout [14]. The following Figure 20 presents the basic idea of the cell distribution.

307 TX(1,y) TX(2,y) TX(x-1,y) TX(x,y) teristics of the hexagonal model, the x coordinates can be obtained in the following way depending if the row for y coordinates is odd or even. The distance between two sites in x-axis is: TX(1,3) TX(2,3) TX(3,3) TX(x,3) ( 30 ) = 2 0. r x = 2r cos 866 (12) TX(1,2) TX(2,2) TX(x-1,2) TX(1,1) TX(2,1) TX(3,1) TX(x,1) TX(x,2) Figure 20. The active transmitter sites are selected from the 2-dimensional cell matrix with the individual numbers of the sites. As can be seen from the Figure 20, the cells are located in such way that they create ideal overlapping areas. The tightly located hexagonal cells fill completely the circle-shape cells. A uniform parameter set is used in each cell, including the transmitter power level and antenna height, yielding the same radius for each cell per simulation case. y r Figure 21. The x and y coordinates for the calculation of the site locations. y x/2 30deg r 30deg 30deg x/2 Figure 22. The geometrical characteristics of the hexagonal model used in the simulator. As the relative location of the cells is fixed, the coordinates of each cell depends on the uniform cell size, i.e. on the radius. Taking into account the charac- x x The common inter-site distance in y-axis is: ( 30 ) 2 ( 30 ) 3 r cos r y = = (13) tan For the odd rows the formula for the x-coordinate of the site m is thus the following: ( m 1) 1. r x( m) odd = r (14) In the formula, m represents the number of the cell in x-axis. In the same manner, the formula for x- coordinates can be created in the following way: 1 x( m) even = r r + ( m 1) r (15) 2 For the y coordinates, the formula is the following: y 2 = (16) 3 ( n) r + ( n 1) r The simulations can be carried out for different cell layouts. Symmetrical reuse pattern concept was selected for the simulations presented in this paper. The most meaningful reuse pattern size K can be obtained with the following formula [8]: 2 K = ( k l) kl (17) The variables k and l are positive integers with minimum value of 0. In the simulations, the reuse pattern sizes of 1, 3, 4, 7, 9, 12, 16, 19 and 21 was used for the C / (N + I) distribution in order to obtain the carrier and interference distribution in both noninterfering and interfering networks (i.e. SER either exists or not depending on the size of the SFN area). In this way, the lower values of K provides with the noninterfering SFN network until a limit that depends on the GI and FFT size parameters. The single cell (K=1) is considered as a reference in all of the cases. The fixed parameter set was the following: Transmitter power: 60 dbm Transmitter antenna height: 60 m

308 Receiver antenna height: 1.5 m Long-term fading with normal distribution and standard deviation of 5.5 db Area coverage probability in the cell edge: 70% Receiver noise figure: 5 db Bandwidth: 8 MHz Frequency: 700 MHz For the used bandwidth, the combined noise floor and noise figure yields dbm as a reference for calculating the level of C and I. The path loss was calculated with Okumura-Hata prediction model for small and medium sized city. The 70 % area coverage probability corresponds with 10% outage probability in the single cell area. These settings result a reference C/N of 8.5 db for QPSK and 14.5 db for 16-QAM. The value is the minimum acceptable C/N, or in case of interferences, C / (N + I) value that is needed for the successful reception of the signal. The Figures 23 and 24 presents the symmetrical reuse patterns that were selected for the simulations. The grey hexagonal means that the coordinates has been taken into account calculating the order number of the sites according to the formulas 14-16, but the respective transmitter has been switched off in order to form the correct reuse pattern. y (km) Network layout for K= x (km) Figure 25. An example showing the layout of the QPSK network with K=7. Figure 26. An example of the simulated case with QPSK and K=7. Figure 23. The reuse patterns with K of 1, 3, 4, 7, 9 and 12. Figure 24. The reuse patterns with K of 16, 19 and 21. The Figure 25 shows the site locations for the QPSK and K=7, and the Figure 26 shows an example of the C/N distribution with the parameter values of K=7, GI=1/4, and FFT=8K. According to the C/I link analysis, the case presented in Figure 26 is free of SFN interferences. The actual simulation results for C/N, or in case of the interferences, for C / (N + I), is done in such way that only the terminal locations inside the calculated cell areas are taken into account. If the terminal is found outside of the network area (the circles) in some simulation round, the result is simply rejected. The Figure 27 shows the principle of the filtered simulation. As the terminal is always inside the coverage area of at least one cell, it gives the most accurate estimation of the SFN gain with different parameter values. Furthermore, the method provides a reliable means to locate the MS inside the network area according to the uniform distribution. The network is dimensioned in such way that the area location probability is 70% in the cell edge. The

309 dimensioning can be made according to the characteristics of long-term fading. allowed SFN diameter (D sfn ). If the link is longer, it is marked as a potential source of interference (I). The interference link proportion can be obtained for each case by calculating the interference links over the total links. It gives a rough idea about the severity of the exceeding of the SFN limit, with a value range of 0-100% (from non-interfering network up to interfered network where all the transmitters are a potential source of interference) Results Figure 27. The filtered simulation area. This principle is used in the simulations in order to keep the network borders always constant. If the mobile station is inside the planned network area, it provides a reliable estimation of the SFN gain. The Figure 28 shows snap-shot type example of the C/N values with less than 8.5 db, which is the limit for the respective parameter settings of QPSK cases. The following Tables 4 and 5 summarises the C/I link analysis for the different reuse pattern sizes and for FFT and GI parameter values. The values presents the percentage of the over-sized legs of distances between the cell sites compared to the amount of all the legs. Table 4. The C/I link analysis for QPSK cases. Reuse pattern size (K) FFT,GI K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ Figure 28. An example of the distribution of the simulation results that yields less than 8.5 db for the C/N. As a verification of the geographical interference class, a study that can be called a C/I-link analysis can be carried out. It is a method to revise all the combinations (hash) of the distances between each pair of sites (TX 1 -TX 2, TX 1 -TX 3, TX 2 -TX 1, TX 2 -TX 3 etc.) marking the link as useful (C) if the guard distance between the respective sites is less than the maximum The C/I link analysis shows that in case of large network (21 cells in the SFN area), the only reasonable parameter set for the QPSK modulation seems to be FFT=8K and GI=1/4. This is due to the fact that QPSK provides with the largest cell sizes (with the investigated parameter set the r is 7.5 km). The cell size of the investigated 16-QAM case is smaller (r=5.0 km) which provides the use of the parameter set of (FFT = 8K, GI = 1/4), (FFT = 4K, GI = 1/4) and (FFT = 8K, GI = 1/8). The interference distance r interference = 13.5 km is the same in all the cases as the interference affects until it reaches the reference level (the sum of noise floor and terminal noise figure). The C/I link investigation gives thus a rough idea about the most feasible parameter settings. In order to obtain the information about the complete performance of DVB-H, the combination of the SFN gain and SER level should be investigated as shown next.

310 Table 5. The C/I link analysis for 16-QAM cases. Reuse pattern size (K) FFT,GI K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ K, 1/ The following Figures 29 and 30 shows examples of two extreme cases of the simulations, i.e. PDF of non-interfered and completely interfered situation. # of samples PDF, C/N, QPSK, K=7, FFT=8K, GI=1/4 C distribution I distribution C/N (db) The figures 29 and 30 shows two examples of the PDF, i.e. occurred amount of samples per C/N and C/I in scale of 0-50 db, with 0.1 db resolution. The PDF gives a visual indication about the general quality of the network. Nevertheless, in order to obtain the exact values of the performance indicators, a cumulative presentation is needed. The following Figure 31 shows an example of the CDF in the noninterfering QPSK network with the reuse pattern size as a variable. The case shows the C/N for the parameter set of QPSK, GI 1/4 and FFT 8K. This mode is the most robust against the interferences as it provides with the longest guard distance CDF, C/N, QPSK, K=1...21, FFT=8k, GI=1/ C/N (db) Figure 31. The CDF of C/N in non-interfered network for reuse pattern sizes of The Figure 32 shows an amplified view to the critical point, i.e. to the 10% outage probability point. Figure 29. An example of the C/N distribution in non-interfered SFN network. 0.4 CDF # of samples PDF, C/(N+I), 16-QAM, K=7, FFT=2K, GI=1/32 C distribution I distribution K=1 K=3 K=4 K=7 K=9 K=12 K=16 K=19 K= C/(N+I) (db) Figure 30. An example of C/N and I/N in interfered case. This parameter combination does not provide functional service in simulated area C/(N+I) (db) Figure 32. An amplified view of the example of the processed simulation results for QPSK.

311 As can be seen form the Figure 32, the single cell (K=1) results a minimum of 8.5 db for the 10 % outage probability, i.e. for the area location probability of 90% in the whole cell area which corresponds to the 70% area location probability in the cell edge. The cell is thus correctly dimensioned for the simulations. In order to find the respective SFN gain level, the comparison with single cell and other reuse pattern sizes can be made in this 10% outage point. The following Figures 33 and 34 shows the respective simulation results for all the symmetrical reuse pattern sizes 1-21 for QPSK and 16-QAM. SFN gain (db) SFN gain, QPSK GI1/4,8K GI1/4,4K GI1/4,2K GI1/8,2K GI1/ 16,2 K GI1/32,2K Figure 33. The SFN gain levels for QPSK cases, with the reuse pattern sizes of SFN gain (db) K SFN gain, 16-QAM GI1/4,8K GI1/4,4K GI1/4,2K GI1/8,2K GI1/ 16,2 K GI1/32,2K Figure 34. The SFN gain levels for 16-QAM cases, with the reuse pattern sizes of The simulation results show the level of SFN gain. The reference case {FFT 8K and GI ¼} results the maximum gain for the reuse pattern sizes of K=1-21 for both QPSK and 16-QAM and provides a noninterfering network. In addition, the parameter set of {FFT = 4K, GI = 1/4} and {FFT = 8K, GI = 1/8} K results a network where SFN errors can be compensated with the SFN gain. According to the results shown in Figure 33, the QPSK case could provide SFN gain of 3-4 db in noninterfering network. It is interesting to note that in the interfering cases, also the parameter set of {GI = ¼, FFT = 4K corresponding FFT = 8K, GI = 1/8} results positive SFN gain even with the interference present for all the reuse pattern cases up to 21. Also the parameter set of {GI = ¼, FFT = 2K, corresponding FFT = 8K, GI = 1/16 and FFT = 4K, GI = 1/8} provides an adequate quality level until reuse pattern of 16 although the error level (SER) increases. According to the Figure 34, the 16-QAM gives equal SFN gain, resulting about 3-4 db in noninterfering network. For the {FFT = 4K, GI = ¼} and the corresponding parameter set of {FFT = 8K, GI = 1/8}, the SFN gain is higher than the SER even with higher reuse pattern sizes compared to QPSK, because the 16-QAM cell size is smaller. As can be seen from the Figures and from the C/I link analysis of Tables 4-5, the rest of the cases are practically useless with the selected parameter set. 6. Methodology for the simulations: third variation (urban SFN network) 6.1. Simulation environment The dense and urban area of Mexico City was used as a basis for the next simulations by applying suitable propagation prediction models (Okumura-Hata and ITU-R P ). The city is located on relatively flat ground level with high mountains surrounding the centre area which was taken into account in the radio interface modelling. The height of the planned DVB-H site antennas was 60, 190, 30, 20, 20, 30 and 30 meters from the tower base, respectively for the sites 1-7. The site number 7 represents the mountain installation with the tower base located 800 meters above the average ground level which results the effective antenna height of 860 meters compared to the city centre level. Site number 4 is also situated in relatively high level, but in this case, the surrounding area of the site limits its coverage area. The rest of the sites are located in the base level of Mexico City centre. As the cell radius of the investigated sites is clearly smaller than 20 km, the Okumura-Hata [3] is suitable for the path loss prediction for all the other sites except for the mountain site number 7. The Figure 38 shows the principle of the geographical profile of this site, varying the horizontal angle from the site to the centre

312 by 10 degree steps. As the profile shows, there are smaller mountains found in front of the site. frequency (via 600 and 2000 MHz). The Figure 39 shows the resulting curve after the iterations. Geographical site location open km 6 7 residential forest dense urban urban km Figure 35. The clutter type of the investigated area. Terrain height (m from sea level) km Figure 37. The site locations and informative relative site sizes of the simulator Terrain height profile Tres Padres - Center Height 1 (TP-Zocalo) Height 2 (TP-Chapultepec) Height 3 (TP-middle) Distance (km) Figure 38. The profile of the mountain site 7. Figure 36. The predicted coverage area of the investigated network as analyzed with a separate radio network planning tool FSL Path loss The Figure 37 presents the location of the selected sites, and the Table 6 shows the site parameters. For the site number 7, ITU-R P.1546 (version 3) [2] model was applied by using the antenna height of 860 meters and frequency of 680 MHz. The calculation of the path loss for the site number 7 was done in practice by interpolating the correct ITU-R P.1546 curve for 860 meter antenna height (via 600 and 1200 meter heights) and for 680 MHz L (db) Distance (km) Figure 39. The estimated path loss L for the mountain site. FSL is free space loss reference.

313 Next, a trend line was created in order to present the tabulated values with a closed formula and to ease the simulations. For this specific case, there was one formula created for the path loss in distances of 1-20 km (L 20 ) and another one for the distances of km (L 100 ), d being the distance (km): L = ln( d) (18) 20 + L = d (19) In order to estimate the error between the trend lines and the original ITU-model, the following Figure 40 was produced. In the functional area of the mountain site, the maximum error is < 0.5 db. Diff (regression - original) Difference of trend line and original values Distance (km) Figure 40. The estimated error margin for the trend lines used for the mountain site path loss calculation. round, the separate sum of the carrier per noise and the interference per noise was calculated by converting the received power levels into absolute powers. The result gives thus information about the balance of SFN gain and SFN interference levels. Tables of geographical coordinates with the respective sum of carriers and interferences were created by repeating the simulations 60,000 times. Also carrier and interference level distribution tables were created with a scale of db. The long-term as well as Rayleigh fading was taken into account in the simulations by using respective distribution tables independently for each simulation round. A value of 5.5 db was used for the standard deviation. The area location probability in the cell edge of 90% was selected for the quality criteria, producing about 7 db shadowing margin for the long-term fading. Terminal antenna gain of -7.3 dbi was used in the calculations according to the principles indicated in [1]. Terminal noise figure of 5 db was taken into account. Both Code Rate and MPE-FEC Rate were set to ½ Simulation results The usable coverage area was investigated by postprocessing the simulation results. The simulations were carried out by using QPSK and 16-QAM modulations, CR ½, MPE-FEC ½ and all the possible variations of FFT and GI. The link budget of the simulator takes into account separately the radiating power levels and antenna heights of each site as seen in Table 6. Table 6. The site parameters. Site Coord., km EIRP Radius, km nr x y dbm W QPSK 16- QAM During the simulations, the receiver was placed randomly in the investigated area (45km 45km = 2025 km 2 ) according to the snap-shot principle and uniform geographical distribution. In each simulation Figure 41. Example of the non-interfered network with the parameter setting of QPSK, GI=1/4 and FFT = 8k. As expected, the parameter set of {QPSK, FFT 8k, GI 1/4} produces the largest coverage area practically without interferences (Figure 41). The results of this

314 case can be considered thus as a reference for the interference point of view. When the GI and FFT values are altered, the level of interference varies respectively. The results show that in addition to the parameter set of {FFT 8k, GI1/4}, also {FFT 8k, GI 1/8}, {FFT 4k, GI 1/4} produces useful coverage areas, i.e. the balance of the SFN gain and SFN interferences seem to be in acceptable levels, whilst the other parameter settings produces highly interfered network. The 16-QAM produces smaller coverage areas compared to the QPSK as the basic requirement for the C / (N + I) of 16-QAM is 14.5 db instead of the 8.5 of QPSK CDF, C/(N+I) QPSK, FFT 8k, GI 1/4 QPSK, FFT 8k, GI 1/8 QPSK, FFT 8k, GI 1/16 QPSK, FFT 8k, GI 1/32 QPSK, FFT 4k, GI 1/32 QPSK, FFT 2k, GI 1/ C/(N+I), db Figure 43. The cumulative C/(N+I) distribution of different modes for QPSK. Figure 42. Comparative example of the simulation results for the parameter set of 16-QAM, FFT 8k and GI 1/16. As can be observed from the previous analysis and Figure 42, the SFN interferences tend to cumulate to the outer boundaries of the planned coverage area. For the QPSK with FFT 8k and GI ¼, the area is practically free of interferences, i.e. the C / (N + I) is > 8.5 db in every simulated location. The dark colour in the middle shows the site locations with the C / (N + I) greater than 35 db. By observing the 90% probability in the cell edge (about 95% in the cell area), i.e. 5 % outage probability of the Figures 43-44, the mode {FFT 8k, GI1/4} provides a minimum of about 7 db and the set of {FFT 8k, GI1/8} and {FFT 4k, GI1/4} provides the same performance in the whole investigated area. This similarity is due to the D sfn limit, which is complied totally in both of the cases as the sites are grouped inside about 30 km diameter. It is worth noting that the values are calculated over the whole map of 45km 45km CDF, C/(N+I) 16-QAM, FFT 8k, GI 1/4 16-QAM, FFT 8k, GI 1/8 16-QAM, FFT 8k, GI 1/16 16-QAM, FFT 8k, GI 1/32 16-QAM, FFT 4k, GI 1/32 16-QAM, FFT 2k, GI 1/ C/(N+I), db Figure 44. The cumulative C/(N+I) distribution of different modes for 16-QAM QPSK,8k,1/4 QPSK,8k,1/8 Area-% for C/(N+I) >= 8.5 db QPSK,8k,1/16 QPSK,8k,1/32 QPSK,4k,1/32 QPSK,2k,1/32 Figure 45. The functional area percentage for QPSK modes compared to the total area.