AN ADVANCED ALGORITHM FOR IMPROVING DVB-T COVERAGE IN SFN V. Mgnone, A. Morello, M. Vsntn RAI Research Centre, Italy ABSTRACT The paper analyses the coverage achevable n sngle frequency networks (SFN) by the European system for dgtal vdeo broadcastng on terrestral channels (DVB-T), usng two dfferent recever structures: the conventonal coherent demodulaton method usng the plot tones nserted n the OFDM (Orthogonal Frequency Dvson Multplexng) spectrum and the new scheme proposed by RAI, based on coded decson drected demodulaton () combned wth a new lock-n algorthm. The frst structure s charactersed by a rapd performance degradaton for echoes outsde the guard nterval, whle the second shows a more gradual degradaton, lke the DAB (Dgtal Audo Broadcastng) system. The performance evaluaton based on theoretcal analyss, software smulatons and hardware laboratory tests, shows the sgnfcant coverage mprovement offered by the gradual degradaton algorthm, especally n the cases of omndrectonal antennas, medum to hgh bt-rates and transmtter dstances exceedng 4 Km. INTRODUCTION The DVB system for terrestral vdeo broadcastng (DVB-T) was standardsed by ETSI (European Telecommuncaton Standard Insttute) n February 1997 [1]. The RAI Research Centre has drectly contrbuted to ts defnton through studes, computer smulatons, laboratory tests and hardware developments n the framework of the European projects DVB, RACE dttb and ACTS Valdate. Ths contrbuton s focused on the analyss of the coverage probablty achevable n the SFN servce area by dfferent types of DVB-T recevers, both n the cases of fxed recepton (roof-top antenna) and of portable statc recepton (omn-drectonal antenna). The nvestgaton takes nto account the network self-nterference produced by the SFN echoes fallng outsde the guard nterval, whose effect s strctly dependent on the equalsaton algorthms mplemented n the DVB-T recever. Frst the paper descrbes the prncples of conventonal coherent demodulaton methods usng the plot tones nserted n the OFDM spectrum of the DVB-T system and shows that they suffer from a steep performance degradaton for echoes fallng outsde the guard nterval. An alternatve method developed by the authors s then descrbed, based on [2, 3] combned wth a new lock-n algorthm, whch offers a more gradual performance degradaton for the echoes outsde the guard nterval, followng the behavour of the European system for dgtal audo broadcastng (DAB) [4]. The new recever performance has been verfed through computer smulatons. Fnally a new smple mathematcal DVB-T system model s proposed for SFN plannng, whch s presently adopted by the CEPT/FM PT24 plannng group. It takes nto account the sgnal dstortons due to natural and SFN echoes nsde and outsde the guard nterval, as well as the characterstcs of the recever algorthms. Ths model s used to evaluate, by means of Monte Carlo computer smulatons, the coverage probablty achevable by fxed and portable recevers n a regular SFN. THE DVB-T SYSTEM The European RACE dttb Project, n co-operaton wth Natonal projects as HD-DIVINE and HDTVT, has studed, desgned and developed the DVB-T system, takng nto account the servce requrements defned by DVB; among these, the most techncally demandng are: the possblty to obtan statonary portable recepton wth omn-drectonal antennas n addton to fxed recepton wth drectve roof-top antennas; the possblty to operate n sngle frequency networks (SFN), where the coverage of large areas (a Regon or even a Country) s acheved by synchronsed transmtters operatng on the same RF channel, wth sgnfcant advantages n terms of spectrum effcency.
The DVB-T system can transmt dgtal multprogramme, conventonal - defnton televson sgnals n MPEG-2 MP@ML format, but t s open to evolve towards HDTV by usng hgher MPEG-2 levels and profles. The transmsson scheme s based on a mult-carrer coded modulaton named C-OFDM [5, 6], charactersed by two operatonal modes, the frst one wth 2K carrers for conventonal mult-frequency networks (MFN), the second one wth 8K carrers for SFNs. Ths modulaton s partcularly sutable to operate on the terrestral multpath propagaton channel because of the narrow-band characterstc of each data carrer and of the presence of a "guard nterval" (wth duraton T g ) whch separates adjacent symbols and avods nter-symbol nterference (ISI) n the presence of echoes. The DVB-T system offers a btrate capacty rangng from 5 Mbt/s to 31.5 Mbt/s, dependng on the chosen level of m-qam modulaton (m = 4, 16 or 64), the nner code rate (1/2, 2/3, 3/4, 5/6 or 7/8) and the guard nterval duraton (T g / T u =1/4, 1/8, 1/16 or 1/32; T u = useful symbol duraton = 224 µs for 2K mode and 896 µs for 8K mode). The system s optmsed for 8 MHz channels (European UHF channellsaton), but t can be easly adapted to 7 MHz and 6 MHz channels by adjustng the recever samplng frequency. correct equalsaton T f, a part of the echo power s assocated to the same OFDM symbol as the - delay sgnal, and therefore contrbutes postvely to the global useful sgnal power; another part of the echo power s assocated to the prevous OFDM symbol and produces ISI whch has smlar effect as an uncorrelated Gaussan nose nterference. In addton to the ISI, a loss of orthogonalty n the OFDM carrers s nduced by the echoes outsde T g, producng Inter-Carrer Interference (ICI). Therefore, progressvely ncreasng the echo delay outsde T g, the useful contrbuton C decreases and the ISI-ICI nterference I ncreases wth a quadratc law. The echoes wth delay τ longer than the nterval of correct equalsaton T f are to be consdered as pure nterference (see fgure a). W C I THE DVB-T SYSTEM MODEL FOR SFN PLANNING To acheve coherent demodulaton, the DVB-T system equalses the ampltude and phase of the data carrers by dvdng the receved samples (after FFT) for the estmated channel frequency response. Conventonal recevers perform the channel estmaton through tme and frequency nterpolaton of plot carrers, regularly nserted n the transmtted OFDM symbol [2, 3]. The channel frequency response estmaton, before frequency nterpolaton, s sub-sampled of a factor 1/3 (one plot out of three carrers), therefore ts Fourer Transform (correspondng to the estmated channel mpulse response) s perodcal wth perod T u /3. To attenuate the spectral repettons (alasng), the frequency nterpolator flter must provde an attenuaton of at least 2-3 db at T u /3. Furthermore t should be flat n a band T f (ndcated n the followng as nterval of correct equalsaton) between and at least T g to allow a good estmaton of the echoes wthn the guard nterval. The DVB-T mode sutable for SFN applcaton has 8K carrers and T g = T u /4, therefore a conventonal recever must satsfy the nequalty T u /4 T g T u /3. If the above condtons are satsfed, the power of all the echoes receved wthn a wndow of duraton T g can be consdered as useful, and contrbutes postvely to the global avalable sgnal power. Outsde the guard nterval, but wthn the nterval of T g T u +T g t Fgure A. DVB-T model for SFN plannng: splttng of the echo power (W) nto useful (C) and nterferng (I) components. In formulae: C = w C w = T f I = ( 1 w ) C f t < 1 f t T g T u t + T 2 g f T T g < t T f u f t > T f Equaton A where C s the power contrbuton from the -th transmtter and w s the weghtng coeffcent for the -th component. A system wth an nterval of correct equalsaton T f equal to the total symbol duraton T g + T u wll be ndcated n the rest of the paper as gradual performance degradaton system. The DAB system for dgtal audo broadcastng [4] comples wth ths
defnton (wth the extenson also to t < ) snce t s based on dfferental demodulaton and does not requre channel estmaton and equalsaton n the recever. Accordng to the hardware tests on the second dttb demonstrator [7], n the coverage evaluatons t s assumed that a conventonal recever s charactersed by T f = T g. Neglectng other nterference sources, the total avalable C/(N+I) [db] n a gven locaton of the servce area can be estmated wth conventonal methods (e.g. Monte Carlo smulatons wth lognormal feld-strength dstrbuton, medan level from ITU-R Rec. 37, varance 5.5 db). C s the sum of the power [Watt] of all the useful contrbutons C from the varous SFN transmtters, I s the sum of the power [Watt] of the nterferng contrbutons I from the varous transmtters (SFN self-nterference) and N the recever nose. The DBV-T system can operate on a Rcean or on a Raylegh multpath channel [8], dependng on the use of a drectve roof-top antenna or of a non drectve set-top antenna (assumng n a worst case the lack of the lne of sght). In these envronments the system requrements n terms of sgnal to nose rato C/N depend on the echo level, especally wth the hgher codng rates (even f the echo delay s nsde the guard nterval). Therefore n plannng exercses the C/N requred by the system must be predcted by a sutable law takng nto account the echo confguraton (type of recepton, MFN and SFN). A computer smulaton nvestgaton has been carred out by the authors on the DVB-T systems n SFN envronments and a smple emprcal law has been derved, as descrbed n Appendx. Ths system model has been used n the followng to predct the SFN coverage. AND THE NEW LOCK-IN ALGORITHM s a channel estmaton algorthm developed at RAI, whch allows to estmate drectly the channel transfer functon for every frequency poston. It s based on a feedback loop explotng the error correcton capablty of a forward error correcton (FEC) decoder and frequency and tme doman flterng to mtgate the effects of nose and resdual errors [2, 3]. A recever based on CD3- OFDM s not affected by frequency-doman subsamplng n the channel estmaton. Therefore a larger nterval of correct equalsaton T f can be adopted, allowng the recever to equalse also echoes exceedng the guard nterval duraton. For example, to obtan a gradual performance degradaton accordng to Equaton 1 for echoes up to 2 T g, the pass-band of the frequency doman flter must be extended up to 2 T g. In order to lock-n the orgnal algorthm n the presence of echoes outsde the guard nterval, a reference OFDM symbol would be requred [2], whch s not present n the DVB-T specfcaton. To solve ths problem, a new recever algorthm, as descrbed n Appendx, has been developed to lockn wthout reference symbols n the transmtted DVB-T sgnal. The new algorthm, as descrbed n Appendx, conssts of two phases. In the frst phase a "quck" synchronsaton (wthn 4 OFDM symbols) of CD3- OFDM s performed on the scattered plot carrers, ensurng lock-n n the presence of echoes nsde the guard nterval, whch s the typcal stuaton n the servce area. In the case the frst phase fals, the second phase s performed, where the channel estmaton s based both on the plot carrers and on the data carrers, to overcome the frequency doman subsamplng problem. The second phase allows a "slow" lock-n (wthn some tens of OFDM symbols) n the presence of echoes nsde and outsde the guard nterval, whch can happen n crtcal recevng locatons of large SFNs. Durng ths phase, lock-n can be assured provded that the channel remans quas-statc. Computer smulatons have demonstrated that, for 64-QAM 2/3 and 8K FFT, lock-n s possble wth a Doppler frequency shft up to.5 Hz for a db echo at T g. The lock-n algorthm s only relevant to the start of the recevng process, whle the trackng phase s based on channel estmaton, whch s charactersed by a good channel trackng speed. To mantan a hgh nose flterng effect, even wth T f = 2T g. a tmedoman averagng over two OFDM symbols has been mplemented n the smulated recever. The new lock-n algorthm has been smulated n the presence of combnatons of echoes nsde and outsde the guard nterval. The consdered transmsson modes were 8KFFT, T u /4 guard nterval, 64-QAM rate 2/3, 16-QAM rate 1/2 and QPSK rate 1/2 modulatons. fgure b compares the performance of a recever usng the new lock-n algorthm wth that of a conventonal recever, n the case of 64-QAM modulaton rate 2/3. The results refer to the case wth a db echo at T g. and a second echo wth varable delay τ. The curves gve the maxmum nterference level (.e., γ [db]=power of the useful sgnal / nterferng power) as a functon of the normalsed echo delay τ/t g, to acheve BER=2 1-4 after Vterb decodng wthout nose. As a reference, the performance of a theoretcal "gradual performance degradaton" recever s also drawn, accordng to the mathematcal model of Equaton 1. The performance of the algorthm s smlar to that of the gradual performance degradaton model for the echoes outsde the guard nterval, achevng a gan n terms of γ from 12 db to 7 db
compared to a recever based on the plots, for delays n the range [1.3 T g, 2 T g ]. The average convergence delay of the new lock-n algorthm s of the order of L=1-2 OFDM symbols, wth a worst case of 4 symbols. Addtonal smulatons on 16-QAM 1/2 and QPSK 1/2 have demonstrated that the gradual performance degradaton model s well approxmated by for all the cases n whch the theoretcal value of γ s larger than about 3-4 db. Ths s n fact the maxmum echo level outsde the guard nterval allowng convergence of the lock-n algorthm (.e., recovery of the π/2 ambguty, see Appendx). In the case of QPSK 1/2, the convergence of the lock-n algorthm was already acheved n phase 1 of the procedure. and fxed roof-top recepton at 1 m a.g.l. wth a drectonal antenna. Ths antenna has a gan of 1 db and a front-to-back rato of 16 db, the gan fallng lnearly from 1 db at 2 to -6 db at ±6. Transmtter Area under test 3 25 2 15 1 5 γ [db] τ = k*t g Plots CD3..2.4.6.8 1. 1.2 1.4 1.6 1.8 2. Normalsed echo delay, k gradual degradaton Fgure B. 64-QAM 2/3. Maxmum echo level versus normalsed echo delay to acheve BER = 2 1-4 wthout nose. Fgure C. Transmtter confguraton n the regular SFN The followng table shows the crtera adopted to evaluate the qualty of the coverage n the worst small area wthn the servce area: % of the locatons of the small area Qualty of the coverage 95 Good 7 Acceptable The guard nterval wndow poston depends on the recever synchronsaton strategy: the best algorthm should place the wndow so that the global useful power s maxmsed wth respect to the global nterferng power: max{ ΣC / ΣI }. COVERAGE RESULTS Coverage probablty evaluaton Network model The SFN structure consdered n the coverage evaluatons s the hexagonal transmtter lattce of fgure c and conssts of 37 transmtters at a dstance D, charactersed by omn-drectonal antennas at a heght of 3 m. The mnmum coverage probablty has been evaluated only on a small porton (see fgure c) of the total area, snce the network s hghly symmetrcal. The ITU-R Rec. 37 propagaton model over land for UHF band has been employed at 5 % of the tme for contrbutons wthn the guard nterval and 1% of the tme for the other contrbutons, both wth a standard devaton for the feld strength log-normal dstrbuton of 5.5 db. The coverage calculatons have been performed for portable outdoor recepton wth omn-drectonal antenna at a 1.5 m heght (antenna heght converson factor from 1 m to 2 m set to 12 db) Two recever models have been analysed (see Appendx), all based on 8K FFT and guard nterval duraton T g = 224 µs: a conventonal recever based on the plot tones wth pass-band T f = T g ; a recever based on channel estmaton, usng a frequency doman flter wth a bandwdth T f = 2 T g and a tme doman flter averagng on two successve symbols. The 15 non herarchcal confguratons of the DVB-T specfcaton have been tested n absence of nose, to evaluate the effects of the SFN self-nterference. fgure d compares the coverage achevable n the worst locaton of the coverage area by the conventonal recever and the one for a transmtter dstance of 6 Km wthout nose. Doublng the nterval of correct equalsaton T f
mproves sensbly the mnmum coverage probablty: for example good portable coverage, possble wth conventonal recevers only wth QPSK 2/3 (6.6 Mbt/s), becomes feasble also wth 16-QAM 3/4 and wth 64-QAM 1/2 (15 Mbt/s), whle acceptable portable coverage s possble even wth 64-QAM 3/4 (22 Mbt/s). allows good fxed coverage also for the hgher bt-rate confguratons, whle a conventonal recever s lmted to about 17 Mbt/s. Mnmum coverage probablty % 1 8 Fxed 1 8 6 4 2 transmtters for 16-QAM 3/4 (R u = 15 Mbt/s). Mnmum coverage probablty [%] Conventonal Portable Fxed 6 4 5 6 7 8 9 Dstance [Km] 4 2 1/2 2/3 3/4 5/6 7/8 1/2 2/3 3/4 5/6 7/8 1/2 2/3 3/4 5/6 7/8 4.98 Transmtter Dstance D = 6 Km 6.64 QPSK 16QAM 64QAM 7.46 Conventonal 8.29 8.71 9.95 13.27 14.93 16.59 17.42 14.93 19.91 Code rate, modulaton scheme and bt rate Portable Fgure D. Mnmum coverage probablty for CD3- OFDM compared to that obtaned for a conventonal recever. fgure e and fgure f present the behavour of 16- QAM 3/4 (15 Mbt/s), and 64-QAM 2/3 (2 Mbt/s) respectvely, n absence of nose, as a functon of the dstance between the transmtters n the network both for the conventonal and CD3 recevers. 1 8 6 4 2 Mnmum coverage probablty [%] Conventonal 4 5 6 7 8 9 Dstance [Km] 22.39 24.88 Fxed Portable Fgure E. Mnmum coverage probablty for a conventonal and a CD3 system as a functon of the dstance between 26.13 Fgure F. Mnmum coverage probablty for a conventonal and a CD3 system as a functon of the dstance between transmtters for 64-QAM 2/3 (R u = 2 Mbt/s). The results show that CD3 allows to ncrease the dstance between the transmtters n the network of about 2 5 % for the same percentage of coverage. For example n fgure e to acheve a good portable coverage at 15 Mbt/s (16-QAM 3/4), the CD3 recevers allow a transmtter dstance of about D=6 Km, to be compared wth the 35 Km relevant to conventonal recevers. Smlarly n fgure f good fxed coverage at 2 Mbt/s s acheved wth transmtter dstances of 5 and 72 Km for conventonal and CD3 recevers, respectvely. Snce D 2 s nversely proportonal to the number of transmtters n the network (assumng a very large coverage area), ths mples that regardng the network self-nterference the CD3 system potentally supports a large SFN wth a number of transmtters halved or dvded by three. Especally for portable ndoor recepton, ths noseless condton s true only wth hgh power SFNs, whch are forecast more n the medum term perspectve than n the ntroductory phase of DVB-T. Addtonal smulatons have been performed n presence both of nose and network selfnterference, verfyng that the gans of the new recevng algorthm are mantaned, unless the power constrants n the network become the lmtng factors of the coverage. CONCLUSIONS Computer smulatons have been performed by RAI on a modfed CD3 channel estmaton algorthm, allowng to acheve gradual performance degradaton for echoes outsde the guard nterval, wthout performance losses wthn the guard
nterval. A new lock-n algorthm appled to CD3- OFDM allows full compatblty wth the current DVB- T specfcaton, wth plot carrers but wthout reference OFDM symbols. The SFN coverage evaluaton demonstrates the sgnfcant mprovement offered by the CD3 algorthm, especally n the cases of portable recepton, medum to hgh bt-rates, hgh transmtter power and transmtter dstances exceedng 4 Km. The lower senstvty to the SFN network selfnterference allows the re-use of the current transmttng stes for analogue televson, and therefore a reducton of the nvestments requred to nstall a DVB-T SFN. It should be noted that the hgh bt-rate modes of the DVB-T specfcaton (e.g. 64QAM 1/2 or 2/3) are the most attractve from a servce pont of vew, and the possblty to serve both fxed and portable recevers wth the full capacty (although wth dfferent coverages) s of prmary mportance for some broadcasters. A hardware prototype has been mplemented by the RAI to perform laboratory tests and feld trals on the new CD3 algorthms. APPENDIX C/N requrements evaluaton The system can operate satsfactorly (quas error free bt stream) when the total avalable C/N+I s larger or equal to the system Equvalent Protecton Target (EPT). Ths parameter has a physcal nterpretaton smlar to that of the C/N requred by the modulaton/codng system, but ncludes also the dstorton produced by the echoes nsde the equalsaton nterval T f, whch determne the crtcalty (frequency selectvty) of the channel. The followng emprcal expresson s proposed by RAI, based on computer smulatons, to estmate EPT, where all the tems are expressed n db: C N F EPT = + C C 5. ( N P N F ) ( C C ) N P N F K A 1 + where EPT s the requred Equvalent Protecton Rato for the system n a partcular echo envronment, C/N F s the sgnal to nose rato requred by the system on the Rcean channel (see [1]), C/N P s the sgnal to nose rato requred by the system on the Raylegh channel (see [1]), K A the channel crtcalty due to artfcal echoes, that s the rato (converted n db) between the power receved from the man sgnal and the total power of the echoes nsde the nterval of correct equalsaton T f (f K A < db, K A s set to db, the most crtcal case), the mplementaton margn (respectvely set to 2.5 and 3 db for and conventonal channel estmaton). In the fgures of ths contrbuton relevant to the deal model, the above expresson and Equaton 1 are solved numercally to obtan the rato γ between the power of the echoes nsde the guard nterval and of the echoes outsde the guard nterval gvng BER= 2. 1-4. The new lock-n algorthm The lock-n algorthm (fgure g) operates nto two consecutve phases: 1. Channel estmaton based on the scattered plots (wth a frequency nterpolaton flter bandwdth T' f = T g ). In the case the lock-n fals (.e. echoes outsde T g ), the second phase s performed. 2. Channel estmaton based both on the plot carrers and on the data carrers. The ampltude and phase of the channel frequency response correspondng to the data carrers are estmated by applyng a non-lnearty (e.g. elevaton to the fourth power) and a tme-doman flter (e.g. average) over L consecutve OFDM symbols. The phase ambguty of multples of π/2, due to the symmetry of the m-qam constellatons, s solved by adoptng carrer-by-carrer the phase whch s closer to that of the channel estmaton obtaned n phase 1. A frequency doman flter wth bandwdth T f equal, for example, to 2 T g, s then appled to the estmated channel frequency response (ncludng the plot and the data carrers). In phase 2, the π/2 phase ambguty resoluton s based on the hypothess that the echoes outsde the guard nterval are constraned n ampltude. Therefore they produce only small addtonal phase rotatons (.e. < π/4) compared to those produced by the echoes nsde the guard nterval (as estmated n phase 1). As a frst approxmaton, confrmed by the smulaton results, t s possble to say that echoes attenuated by more than 3-4 db produce phase rotatons smaller than π/4 and therefore compatble wth the proposed algorthms. When the lock-n s acheved, the channel trackng speed of the recever becomes that typcal of, allowng to mantan a relable recepton also n the presence of movng reflectors around the portable recever.
scattered plots H(n,k) estmaton tme & frequency doman nterpolaton Bw = T f H p (n,k) A y(n,k) FFT nonlnearty tme doman flter phase ambguty resoluton H(n,k) reconstructon frequency doman flter Bw = T f' H d (n,k) B C H(n,k) data carrers swtch channel estmaton loop Bw = T f' H CD3 (n,k) equalser y (n,k) z(n,k)= H(n-1,k) H(n-1,k) decoded data FEC decoder mux symbol & bt denterleaver M-QAM demapper Fgure G. tranng algorthm REFERENCES [1] European Telecommuncaton Standard: "Dgtal broadcastng systems for televson, sound and data servces; Framng structure, channel codng and modulaton for dgtal terrestral televson", ETS 3 744 February 1997 [2] V. Mgnone, A. Morello, ": a novel demodulaton scheme for fxed and moble recevers", IEEE Transactons on Communcatons, COM-44(9), pp. 1144-1151, September 1996 [3] V. Mgnone, A. Morello, M. Vsntn, "CD3- OFDM: A new channel estmaton method to mprove the spectrum effcency n dgtal terrestral televson systems", IBC'95, Amsterdam, 14-18 September, 1995 [4] European Telecommuncaton Standard: "Rado broadcast systems; Dgtal Audo Broadcastng (DAB) to moble, portable and fxed recevers", pr ETS 3 41, November 1994 [5] L. Cmn, "Analyss and smulaton of a dgtal moble channel usng Orthogonal Frequency Dvson Multplexng", IEEE Transacton on Communcatons, COM-33, p.665-675, July 1985 [6] M. Alard, R. Lassalle: "Prncples of modulaton and channel codng for dgtal broadcastng for moble recevers", EBU Techncal Revew, n.224, p.168-19, August 1987 [7] A.Morello, G.Blanchett, C.Benz, B.Sacco, M.Tabone: Performance assessment of a DVB-T televson system, Proceedngs of the 2th Internatonal Televson Symposum, Montreaux 97 [8] J. G. Proaks, Dgtal Communcatons, Mc Graw - Hll, 1983