COMPARATIVE PERFORMANCE ANALYSIS OF SYMBOL TIMING RECOVERY FOR DVB-S RECEIVERS Panayot Savvopoulo, Unverty of Patra, Department of Electrcal and Computer Engneerng, 6500 Patra, Greece, Phone: 30-610-996483, Fax: 30-610-99734, e-mal: pavvop@upatra.gr Theodore Antonaopoulo, Unverty of Patra, Department of Electrcal and Computer Engneerng, 6500 Patra, Greece, Phone: 30-610-996487, Fax: 30-610-99734, e-mal: antonao@upatra.gr Abtract The need for relable broadband atellte communcaton ervce ha led to the development of the econd generaton DVB pec, DVB-S. Th paper conder the performance of a feedbac Symbol Tmng Recovery technque, whch baed on the Non-Data-Aded (NDA) Gardner Tmng Error Detector (TED), n an all-dgtal DVB-S IF recever archtecture. The Symbol Tmng Recovery performance meaured n term of loc-n tme and etmaton accuracy, wth the contrbuton of mulaton reult over all dfferent modulaton cheme ued by the DVB-S tandard (QPSK, 8PSK, 16APSK and 3 APSK) and under noe ource uch a non-lnearte, addtve whte Gauan noe (AWGN) and gnfcant redual carrer frequency offet. 1. Introducton New technologe have been preented n the feld of dgtal communcaton n the lat decade, epecally n modulaton and forward error correcton. The capablte ared by the ue of thee new cheme, have paved the way for more relable and more effectve communcaton, even under extremely hotle channel condton. The growng demand for more effcent dgtal atellte communcaton ervce and applcaton led to the development of the econd generaton DVB pecfcaton for broadband atellte communcaton, DVB-S [1]. The cope of th new pecfcaton to gradually replace the older and for many year globally deployed DVB-S, a t able to effectvely explot the avalable pace egment reource whle t provde the opportunty to older DVB-S recever to contnue ther operaton. DVB-S wa developed n order to provde hgh performance n broadcatng and uncatng applcaton, whch are totally baed on power and pectral adaptvty at reaonable recever complexty. Thee charactertc are achevable due to the veratlty of the DVB-S phycal layer that utlze a varety of modulaton and error codng cheme accordng to channel condton. A a conequence, new DVB-S termnal recever need to operate at low gnal-to-noe rato (SNR) and non-lnearte whch alo a major concern for the atellte communcaton ytem degner. Th wor deal wth a ymbol tmng recovery (STR) mechanm ued n DVB-S recever [], whch a econd order cloe-loop baed on the non-data-aded (NDA) Gardner tmng error detector (TED) [3]. Th cheme combne tructural mplcty along wth the capablty to operate at all dfferent DVB- S modulaton format, under even a gnfcant redual carrer frequency offet. Accordng to thee feature, STR ha to be able to converge wthout pror prece carrer ynchronzaton, thu comprng the frt ynchronzaton procedure performed n typcal DVB-S recever. Baed on thee apect, th paper preent the confguraton and etup approach for econd order cloed-loop STR technque baed on the NDA Gardner TED, a a part of an all dgtal IF DVB-S recever ncludng a dgtal down converter (DDC) module operatng at 70 MHz. Apart from degnng apect, a thorough performance analy, n term of loc-n tme and etmaton accuracy, alo gven under dfferent channel condton ncludng non-lnearte, AWG noe and redual carrer frequency offet. The functonal verfcaton a well a the performance evaluaton performed through extenve mulaton baed on dgtal ample of captured DVB-S IF gnal of all dfferent modulaton format (QPSK, 8PSK, 16APSK, 3 APSK) from a DVB-S complant modulator, a hgh level mparment model (non-lnearte or AWG noe) and the all-dgtal DVB-S IF recever archtecture contng of DDC and STR ubunt.
Fgure 1: A DVB-S Recever Archtecture Secton gve an overvew of the all-dgtal IF DVB-S termnal recever archtecture, whle Secton 3 hghlght the degn and parameterzaton of the cloed-loop STR technque. Secton 4 preent the meaurement platform, the ued performance metrc and the repectve performance reult.. The DVB-S IF Recever A DVB-S recever cont of an out-door unt (parabolc antenna and LNB), an external L to IF down converter and an all-dgtal IF recever, a hown n Fgure 1. The all-dgtal IF recever ue an A/D converter, drven by a free runnng ocllator, for dgtzng the ncomng gnal and the dgtal ample feed the frt dgtal proceng unt, the dgtal down-converter (DDC). DDC convert the IF gnal nto baeband gnal, through two cacaded tage of frequency down-converon. The DDC output ample, n the form of n-phae and quadrature gnal component, are ued by the baeband dgtal proceng crcut for ynchronzaton, demodulaton and decodng. The frt ta performed nde the baeband proceng unt the ymbol tmng recovery (STR), whch mplemented a a econd order loop before the matched flter, generate ample ynchronzed to the ncomng ymbol, whch are ued by the ubequent module for frame, carrer and phae ynchronzaton and decodng purpoe. The ncomng RF gnal (at Ku/Ka band), ntally down-converted to the L-band, ung the LNB of the out-door unt, and then the gnal further down-converted to the IF-Band (F c = 70 MHz center frequency) by an analog down-converter module that nclude output power control and bandpa flterng operaton. Incomng gnal adjutment acheved by an automatc gan controller (AGC), mplemented n the dgtal down-converter (DDC) of the IF dgtal recever. The AGC drven by a power meaurement crcut mplemented n the DDC, n order to properly adjut the receved gnal to the dynamc range of the ued ADC. The AGC alo nclude a fxed bandwdth bandpa flter. If BW Pa the bandwdth of the bandpa flter (n MHz), the maxmum upported ymbol rate R (n MBaud) gven by: BWPa R (1) (1 α) where α the roll-off factor upported by the DVB-S tandard (0.0, 0.5 or 0.35). The IF gnal ampled by the ADC at a rate of F (n MSp), whch choen o a to atfy the Nyqut crteron and be tme the ymbol rate R of the receved gnal, that : (F BW c pa F = R /) F F where an even nteger and F ADC the maxmum amplng rate of the ued ADC. The electon of the even factor trongly related to the ample rate converon (SRC) tage at the DDC output. Gven F, the frt tep of the DDC proceng to mx the ampled gnal wth a quadrature ocllator of a fxed frequency F /4. Th typcally managed by multplyng the ADC ample wth a perodc equence reultng to two gnal component of half the ADC amplng rate (F /) [4]. A the actual amplng rate recovered, both gnal are low pa fltered n order to uppre the unwanted gnal mage at hgher frequence and at the flter output a down-ampler appled. ADC ()
The next proceng tage a programmable frequency down-converter that utlze a numercal control ocllator (NCO) and a complex multpler, wth both runnng at F /. The frequency downconverter controlled by a carrer recovery econd order loop, whch ntated after STR and frame ynchronzaton convergence. The functonalty of th tage baed on a phae accumulator regter that produce utable phae argument that are mapped to ample of a pre-tored nuod waveform n a loo-up table, accordng to a frequency control nput. The NCO frequency reoluton f (n Hz) determned by the runnng cloc frequency (F /) and the number of bt (B) of the phae accumulator regter, accordng to: F / F f = = (3) B B 1 Therefore, the output frequency of the NCO for a gven nput u (ungned value) to the phae accumulator, gven by: F u B 1 fnco =, u [0, 1] (4) B 1 Fnally, at the DDC output, ample rate converon (SRC) tae place, that provde the neceary ratonal rate-change factor o that the amplng rate of the DDC output adjuted to an nteger multple of the actual ymbol data rate. The rate converon mechanm combne a polyphae FIR nterpolaton flter wth the approprate decmator at t output, whch eep the cloc of each polyphae ubflter to F /. The complexty of the SRC tage play a gnfcant role n the electon of the even nteger of the equaton () for a gven ADC amplng rate, eeng for the mallet nteger factor for nterpolaton and decmaton procedure. The NDA cloed-loop STR technque preented n the next ecton, requre that the ample rate at the DDC output or 4 tme the ymbol rate. 3. Cloed-Loop STR technque baed on the NDA Gardner TED Th ecton preent the archtecture of the dgtal STR feedbac loop, whch hown n Fgure, and perform ntal acquton and tracng of the actual ymbol rate by ung an etmate of the tmng error gven by the Non-Data-Aded (NDA) Tmng Error Detector (TED) propoed by Gardner [3]. The element comprng the STR loop are the Gardner TED, the frt order loop flter and the Farrow tructured dgtal nterpolator along wth t control logc (Tmng-NCO). Matched flterng and the repectve down-amplng performed on the ample generated by the STR loop. All element wthn the feedbac loop contrbute to the ymbol ynchronzaton proce. Tmng error etmated by the Gardner TED n an average rate equal to the actual ymbol rate and then fltered by the loop flter, whoe output drve the control logc of T-NCO. Fnally, the nterpolator controlled by the T- NCO, and that reult to the mnmzaton of the detected tmng error. Fgure : Cloed-Loop STR ung the NDA Gardner TED
3.1 Interpolaton Flter The functon of the nterpolaton flter to calculate a ample y(t ) ung a et of adjacent nput ample x(mt ), and a fractonal nterval µ (0 µ 1) whch both are obtaned from the T-NCO control logc, explaned n a followng ubecton. T compre the amplng perod of the nput tream, derved from the DDC output and T the perod of the nterpolated ample. Baed on the fact that after the nterpolator the NDA Gardner TED ued, whch need two ample per ymbol for t tmng error detecton, T equal to T/M, where T ndcate the actual ymbol perod and M an nteger number or 4. A well nown archtecture ued n tmng recovery applcaton utlze a pecewe polynomal nterpolator [5] tructure whch provde a great balanced trade-off between effcent hardware mplementaton and performance. The fundamental equaton for dgtal nterpolaton gven by the followng equaton: y(t ) = y[(m µ )T ] = x[(m I = I1 )T ] h [( µ )T ] where x(m) are the nput ample and h I (t) a fnte duraton mpule repone of a tme contnuou nterpolaton flter. Farrow mplementaton aume that the mpule repone a pecewe polynomal n each T nterval, wth ndex beng n the nterval [I 1, I ], (I 1, I nteger number). Practcally, baepont ndex m dentfe the I=I -I 1 1 gnal ample to be ued for the th output y and fractonal nterval µ κ dentfe the I coeffcent employed n equaton (5) for the generaton of th output y. Equaton (5) can be wrtten a a neted and more effcent repreentaton n term of mplementaton, whch for cubc Farrow tructure, ued n our cae, : where: I y() = [{u(3)µ u()}µ u(1)]µ u(0) (6) u(n) = = I I1 b n ()x(m Coeffcent b n are contant, ndependent of µ κ, olely determned by the flter mpule repone h I (t). There are total N I coeffcent f all mpule-repone egment are decrbed by polynomal of the ame degree N. In the cae of the cubc nterpolator, N=3 whch repreent the polynomal degree. Cubc Farrow tructure nclude 4 column of FIR tranveral flter of fxed coeffcent. Each FIR column ha I=4 tap. The coeffcent of cubc nterpolator column are gven n [5]. ) (5) (7) 3. NDA Gardner TED The bac element of the STR loop the tmng error detector (TED). Gardner TED compre an NDA algorthm that propoed for ue n DVB-S ymbol ynchronzer due to t ndependence from the receved data ymbol and the carrer frequency error. The Gardner TED requre two ample for every ncomng ymbol. The extracton of the tmng error for the th data ymbol gven by: T T e = y(t ){yi[( 1)T] yi(t)]} yq(t ){yq[( 1)T] y Q(T)]} (8) The S-curve of the TED computed by cloely followng the dervaton of [6], and ung the equaton: 4C contk πδ S( δ ) = n( ) (9) T T where δ= τ-τˆ ( τ : actual tmng error, τ ˆ : etmated value) and aumng ndependent random data ymbol {c} from an alphabet of ze N wth zero-mean: * C cont, = E{c c } = (10) 0, Parameter K gven by the followng ntegral, n whch H(f) equal to the root-raed cone pule functon (n frequency doman): K = H( - 1 T 1 f)h( T f)co( πft)df a TED appled drectly before the matched flterng operaton, and equal to: α πα K = co( ) (1) π(α - 1) (11)
Equaton (1) and (9) ndcate that the S-curve a nuodal functon of perod T (ymbol perod) and pae from the orgn when δ=0. The curve ampltude proportonal K and alo a functon of the roll-off factor α. From the ampltude of the S-curve we can determne the gan K D of the Gardner TED whch equal to 4C cont K π. 3.3 Proportonal-Integral Frt Order Loop Flter The loop flter play a gnfcant role n the STR loop nce t determne t loop bandwdth, B L T, whch trongly related to the requred loc-n tme and the etmaton accuracy durng the acquton and tracng mode, repectvely. The loop flter alo characterze the order of the STR loop. In order to degn a econd order feedbac loop, a well nown frt order PI (Proportonal-Integral) loop flter ued, whoe tranfer functon equal to: 1 g0 g1z F(z) = (13) -1 1- z where the flter coeffcent g 0 and g 1 are gven by: 1 α1 1 α g0 = ( ) and g1 = ( 1) (14) KDK 0 α0 KDK 0 α0 K D and K 0 are the gan of TED and T-NCO, repectvely. Parameter α 0, α 1, α are obtaned from the dcrete tme econd-order PLL model, whch reult to: α = 4(ω T) ζω T 4, α = (ω T) 8, α = 4(ω T) ζω T 4 (15) 0 n n 1 n n n ω n tand for the natural frequency (rad/ec) and ζ for the dampng factor (0<ζ<1) and uually et to 1/. The confguraton of the loop flter baed on the normalzed to the ymbol rate loop bandwdth parameter, whch for econd order feedbac loop the repectve natural frequency gven by: 1 1 BL, = ωn(ζ ) (16) 4ζ The loop flter bandwdth determne the STR accuracy and acquton tme. The hgher the bandwdth, the fater the acquton proce but wth le accurate reult. In order to acheve a fater crcut wth ncreaed accuracy, the mplemented STR ue two loop flter wth dfferent loop bandwdth, one wde and one narrow, and a control crcut that determne whch flter ued a the tme progree. Both flter are alway fed wth new ample, but ther output are wthced n order to drve the T-NCO. The above approach acheve to accelerate the acquton proce n order for the feedbac loop to reult to an error gnal n the vcnty of the real error and at th pont the output of the flter wth the narrower loop bandwdth ued for achevng uffcent etmaton accuracy. The STR loop tart t operaton wth the loop flter aocated wth the wder loop bandwdth and f the normalzed tmng error, whch feed T-NCO, mantan the ame mean value (over a number of ample determned by the narrow loop bandwdth 1/( B L T LOW )) n a pecfc range for a perod of tme, the narrower flter get the control of the loop. 3.4 Interpolator Control Logc (T-NCO) TED produce an error gnal at ymbol rate 1/T ung fractonally paced ample T =T/M (M nteger). In our cae M can be ether or 4. Snce the amplng rate 1/T not an exact multple of the ymbol rate, ample of perod T need to be mapped onto the tme cale of STR loop nput ample of perod T. Th expreed a: T ε Τ = L [T ε Τ ]T µ T = m T µ T (17) INT L INT (x): larget nteger x For every nput ample, control logc compute the correpondng baepont m and the fractonal delay µ n order to obtan the repectve nterpolated value. A new ample nerted nto the Farrow tructure everyt, whle a new output computed at each baepont m T. Th procedure explan the decmator appearng at the output of the nterpolator n Fgure. Due to the fact that the error gnal at the TED produced at ymbol rate, a econd decmaton need to be performed whch elect every Mth baepont m (=nm) to output one error gnal. The prevou decmator alo
depcted n Fgure. Afterward the TED gnal proceed by the loop flter, whoe output e(m nm ) ued to adjut the control word w(m nm ) of the T-NCO. w(m nm )= w(m (n-1)m )K 0 e(m (n-1)m ) (18) where n our cae K 0 =1.In the abence of noe the control word aume t correct value: T 1 T w = = ( ) (19) T M T Baepont and fractonal delay are computed recurvely a follow: Ung equaton (17), the next ample (1)T ε T gven by: T ( 1)T ε Τ = mt µ T Τ = mt (µ )T (0) T If the unnown rato replaced wth the etmate w(m ) (equaton (19)) then obtaned: ( ε [µˆ w(m )]T [µˆ w(m )] T (1) 1)T Τ = mt LINT mod1 and fnally, the etmate of m and µ are gven by: m µˆ 1 INT n 1 = m = [µˆ n L w(m [µˆ n )] w(m mod1 )] () 4. Performance Reult In th ecton, varou performance reult are preented whch depct the behavour of the prevouly decrbed STR feedbac loop, over everal mparment encountered n forward atellte channel. The mot gnfcant mparment found n uch a channel related to the non-lnearte ntroduced by the Travellng Wave Tube Amplfer (TWTA) of the atellte tranponder. Addtonally, nce the STR loop the frt ynchronzaton mechanm actvated n a DVB-S recever after power up, redual carrer frequency offet alo an mportant pecularty that regularly added to the prevou condton and need to be tuded how t affect the performance of the STR mechanm. Apart from non-lnearte and frequency offet, addtve whte Gauan noe alo condered. The performance metrc ued for tudyng the STR behavour under the above channel condton are the ntal acquton tme (loc-n tme) and the accuracy of the ymbol rate etmaton error durng tracng. A loc-n tme we conder the tme requred to acheve the teady-tate for the frt tme. The econd metrc the RMS value of the normalzed etmated tmng error. All performance reult have been derved from a meaurement etup, whch baed on off-lne proceng ung real ample from a DVB-S complant tranmon gnal. The etup cont of a DVB-S complant IF (70 MHz) modulator, a hardware platform wth a hgh-peed ADC and a large torage memory and a hgh level model decrbng the channel mparment (AWGN or non-lnearte) along wth the all-dgtal IF DVB-S recever decrbed n Secton. The DVB-S IF modulator wa ued n order to generate the repectve gnal for all dfferent modulaton format (dfferent MODCOD) ncludng alo the mparment of the redual carrer frequency error. Concernng the hgh level model of non-lnearte, we have utlzed the AM/AM and AM/PM charactertc of the Hella- SAT [7]. All experment have been performed wth the followng parameter: Nomnal Symbol Rate: 10 MBaud, roll-off 0.35, M= (at the cubc nterpolator), wde normalzed loop bandwdth 10-4, narrow loop bandwdth 5 10-5. Fgure 3(a) and 3(b) depct the RMS etmaton accuracy and the requred acquton tme n the preence of AWG noe repectvely, for all modulaton cheme and for zero frequency offet error. A the frt fgure demontrate, the RMS etmaton accuracy mprove a the noe level decreae whle t lghtly affected by the dfferent modulaton cheme. The acquton tme lghtly affected by the ncreae of noe power for all modulaton cheme, but t ncreaed a the ze of the ued contellaton ncreae. Fgure 3(c) preent the RMS accuracy n relaton to carrer frequency offet error at a gven AWG noe level (E /N 0 =0dB). A t hown, the RMS accuracy not affected gnfcantly ether by the frequency offet error or the ued contellaton. On the other hand, Fgure 3(d) how how the frequency offet affect the acquton tme for all dfferent modulaton cheme. More populated contellaton ntroduce longer acquton tme n contrat to le populated due to the fact that the Gardner TED generate an ncreaed error gnal a the average dtance between the receved ymbol become larger.
Fgure 3(e) and 3(f) preent the RMS accuracy and acquton tme n the preence of non-lnearte for zero frequency offet, repectvely. Under th mparment, the RMS accuracy trongly affected by mall change of the Output Bac Off (OBO) parameter n the non-lnear area, for all modulaton cheme. Addtonally, there a gnfcant dfference between the RMS value reached by two rng modulaton (16APSK and 3APSK) compared to ngle rng modulaton (QPSK and 8PSK), due to maller dtance between adjacent ymbol. The ame concluon alo derved from the acquton tme fgure that how a more gnfcant ncreae n the acquton tme for 16APSK and 3APSK, n the ame wndow of OBO value. From both fgure t evdent that for OBO hgher than -db, 16APSK and 3APSK fal to loc n contrat to the other contellaton. Fgure 3(g) and 3(h) how that for -6dB OBO level, the RMS accuracy not affected by the ncreae of frequency offet error n contrat to the acquton tme whch affected gnfcantly. Furthermore, QPSK and 8PSK have better acquton tme and RMS accuracy a they are more robut n non-lnear condton. 5. Concluon In th paper, an STR feedbac loop utlzng the well nown NDA Gardner TED wa thoroughly preented and analyed a a part of an all-dgtal DVB-S IF recever. A detaled analy for the STR loop confguraton wa preented along wth nformaton for each loop component, whle extenve meaurement were ued for performance evaluaton n term of loc-n tme and etmaton accuracy n ntal acquton and tracng repectvely. The mparment condered durng the experment were non-lnearte, AWG noe and redual carrer frequency offet. 6. Acnowledgment Th wor wa upported by the European Regonal Development Fund (ERDF) and the Gree Mntry of Development General Secretarat for Reearch and Technology (GSRT). 7. Reference [1] ETSI EN 30 307 V1.1., Dgtal Vdeo Broadcatng (DVB); econd generaton framng tructure, channel codng and modulaton ytem for broadcatng, nteractve ervce, new gatherng and other broadband atellte applcaton, June 006. [] E. Can, R. De Gaudenz, and A. Gne, DVB-S modem algorthm degn and performance over typcal atellte channel, Internatonal Journal of Satellte Communcaton and Networng, vol., no. 3, pp. 81 318, 004. [3] F.M. Gardner, A BPSK/QPSK tmng-error detector for ampled recever, IEEE Tran. Commun., Vol. 5, pp. 43-685, May 1986. [4] Peter B. Kenngton, RF and Baeband Technque for Software Defned Rado, Artech Houe, 005. [5] L. Erup, F. M. Gardner, and R. A. Harr, Interpolaton n Dgtal Modem Part II: Implementaton and Performance, IEEE Tran. Commun., vol 41, no. 6, pp 998-1008, June 1993 [6] U. Mengal and A. D Andrea, Synchronzaton Technque for Dgtal Recever, Plenum Pre, 1997. [7] HELLAS-SAT SATELLITE HANDBOOK, March 004.
RMS of Normalzed Etmated Error 10-3 10-4 10-5 Etmaton Accuracy, Freq. Offet= 0 MHz 3APSK, MODCOD:4 Tme (Sec) 0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 Aquton Tme, Freq. Offet= 0 MHz 3APSK, MODCOD:4 RMS of Normalzed Etmated Error 10-6 4 6 8 10 1 14 16 18 0 E /N o (db) 10-3 10-4 10-5 Etmaton Accuracy, E /N o =0dB 3APSK, MODCOD:4 10-6 0 0.5 1 1.5.5 3 Frequency Offet (MHz) (a) (c) Tme (Sec) 0.0 0.01 4 6 8 10 1 14 16 18 0 E /N o (db) 0.16 0.14 0.1 0.1 0.08 0.06 0.04 0.0 3APSK, MODCOD:4 Aquton Tme, E /N o =0dB (b) (d) 0 0.5 1 1.5.5 3 Frequency Offet (MHz) RMS of Normalzed Etmated Error 10-3 10-4 10-5 Etmaton Accuracy, Freq. Offet= 0 MHz 3APSK, MODCOD:4 Tme (Sec) 0.55 0.5 0.45 0.4 0.35 0.3 0.5 0. 0.15 0.1 Aquton Tme, Freq. Offet= 0 MHz 3APSK, MODCOD:4 10-6 -9-8 -7-6 -5-4 -3 - Output Bac Off (db) (e) 0.05-9 -8-7 -6-5 -4-3 - Output Bac Off (db) (f) RMS of Normalzed Etmated Error 10-3 10-4 10-5 Etmaton Accuracy, OBO=-6 db 3APSK, MODCOD:4 Tme (Sec) 0.3 0.5 0. 0.15 0.1 3APSK, MODCOD:4 Aquton Tme, OBO=-6 db 10-6 0 0. 0.4 0.6 0.8 1 1. 1.4 1.6 1.8 Frequency Offet (MHz) (g) 0.05 0 0. 0.4 0.6 0.8 1 1. 1.4 1.6 1.8 Frequency Offet (MHz) Fgure 3: STR Performance Metrc under dfferent mparment (h)