DVB-S2 inner receiver design for broadcasting mode

Transcription

1 28 Yang et al. / J Zhejiang Univ Si A 27 8(1):28-35 Journal of Zhejiang University SCIENCE A ISSN (Print); ISSN (Online) jzus@zju.edu.n VB-S2 inner reeiver design for broadasting mode YANG Jian-xiao 1, WANG Kuang 1, ZOU Zhi-yong 2 ( 1 Institute of Information Siene and Eletroni Engineering, Zhejiang University, Hangzhou 3127, China) ( 2 Guoxin Co. Ltd, Hangzhou 3127, China) shawn_yangjx@hotmail.om Reeived Mar. 1, 26; revision aepted Sept., 26 Abstrat: This paper details on the design of VB-S2 reeivers whih is ompliant with the broadasting mode. Speial attention is paid to the speifi reeiver funtions neessary to demodulate the reeived signal. To show the system performane we onsider the design of a omplete reeiver onsisting of timing reovery unit, frame synhronization unit, frequeny reovery unit and phase reovery unit. The system is easier to hardware implementation omparing with that provided in (ETSI, 25; Sun et al., 2). After the performane of the algorithms is analyzed and a quantitative result is given, this allows us to draw onlusions onerning the ahievable system performane under realisti omplexity assumptions. Key words: VB-S2, Timing reovery, Frame synhronization, Frequeny reovery, Phase reovery doi:1.1631/jzus.27.a28 oument ode: A CLC number: TN93 INTROUCTION The seond generation satellite digital video broadasting standard (ETSI, 2) has ome out reently. Compared with the previous satellite standard VB-S (ETSI, 23), VB-S2 has more advantages, suh as the 3% inrease in hannel apaity, the more reliable performane, and the more effiient usage of transponders. The VB-S2 standard has been designed having in mind the peuliarities of the satellite hannel, partiularly the link fading impairments and the arrier phase noise (Nemer, 25) dominated by the user terminal RF front-end. The most hallenging issues that one has to deal with when designing a VB-S2 demodulator are the rapid frame synhronization and the arrier reovery unit. In broadband satellite transmission, the phase of the arrier is usually affeted by a number of distortions (Casini et al., 2). Many algorithms are available in the literature to address the issue of arrier phase reovery (Mengali and 'Andrea, 1997) for different modulation shemes. However, using low ost ommerial low noise bloks (LNBs) and tuners leads to a remarkable inrease of the arrier phase noise whih the system has to be able to tolerate making onventional arrier phase estimators suffer from an unaeptable yle slip rate. The rest of the paper is organized as follows. We introdue the struture of the physial layer frame in the seond setion. In the third setion, we propose a digital inner reeiver sheme and the aompanying problems of eah part. In the fourth setion, through researhing the algorithms in the key units of the inner reeiver, we give out the algorithms of the timing synhronization unit, the frame synhronization unit, and the arrier reovery unit. In the fifth setion, the performane of eah unit is analyzed by means of fixed point simulation. We make the onlusion in the final setion. Physial layer framing The VB-S2 physial layer frame (named PLFRAME) struture shown in Fig.1 is ritial in the design of the synhronization algorithms. The frame is well explained as follows. XFECFRAME (length: 68 bits for broadasting mode) is the data frame after inner and outer oding. After mapping, XFECFRAME is slied into an integer N of onstant

2 Yang et al. / J Zhejiang Univ Si A 27 8(1): PLHEAER SOF PLSCOE 9 symbols SLOT-1 N SLOTs XFECFRAME SLOT-2 SLOT-N from AGC Frame synhronization Timing synhronization Frame synhronization information Frequeny detetor loop Phase detetor loop to FEC 1 SLOT (π/2 BPSK) 16 SLOTs (seleted modulation) 36 symbols Fig.2 Inner reeiver arhiteture of VB-S2 PLHEAER SLOT-1 length SLOTs (length: 9 symbols eah). The physial layer frame header (named PLHEAER) is omposed of an SOF field (26 symbols, identifying the Start of Frame) and a srambled and oded PLSCOE field (6 symbols, identifying the physial layer transport information) whih will be modulated into π/2 BPSK symbols. The PLFRAME an be onfigured with or without pilot modes. In the former ase a PILOT BLOCK shall be omposed of 36 pilot symbols. The first PILOT BLOCK shall be inserted 16 SLOTs after the PLHEAER, the seond after 32 SLOTs and so on. PROBLEM STATEMENT as SLOT-16 PLFRAME After AGC, the reeived signal an be expressed rt () = ast ()exp(j2π ft) + nt ( ), (1) where a, s(t), f and n(t) denote the omplex amplitude fator, the base-band modulated signal, the residual arrier frequeny offset and the additive white Gaussian noise respetively. By sampling we get the disrete signal in the time t k =kt, rk ( ) = ask ( )exp(j2π fkt) + nk ( ), (2) PILOT BLOCK Fig.1 PLFRAME struture SLOT-N where T is the symbol duration, s(k)=aexp(jθ k )= I k +jq k, and A is the amplitude of the symbol. As shown in Fig.2, the VB-S2 demodulator we proposed onsists of a timing synhronization unit, a frame synhronization unit, a arrier frequeny detetor loop and a arrier phase detetor loop. Timing synhronization We adopt Gardner TE algorithm (Gardner, 1986) in timing error detetor. This algorithm is non-data aided and is insensitive to arrier phase error. The expression of the algorithm is et () = It ( T/2)[ It () It ( T)] + Qt ( T/2)[ Qt ( ) Qt ( T)], (3) where I(t) and Q(t) denote the real part and the imaginary part of the reeived signal, respetively. As we know, the performane of this algorithm is sensitive to the arrier frequeny error. The timing error with nonzero frequeny error should be modified as et () = {( It T/2)[ It () It ( T)] + Qt ( T/ 2)[ Qt ( ) Qt ( T)]}os(2π ft ) () {( It T/2)[ Qt () + Qt ( T)] Qt ( T/2)[ It ( ) + It ( T)]}sin(2π ft). The timing error spetrum before the loop filter is shown in Fig.3. Just as shown, the Gardner TE algorithm suffers from a slightly higher self-noise aused by the sin terms in Eq.() and also suffers from the systemati timing phase jitter aused by the fat that intermediate samples do not only depend on both adjaent data symbols but on all data samples. Timing error spetrum ( 1 ) Normalized frequene (π) Fig.3 Timing error spetrum without pre-filtering

3 3 Yang et al. / J Zhejiang Univ Si A 27 8(1):28-35 Frame synhronization Frame aquisition in VB-S2 suffers from two major impairments: the extremely low SNR, whih an indeed assume negative values in db, and the unknown arrier frequeny offset and phase. In addition, at terminal startup the unertainty region equals the entire frame length, TF, whih in the worst ase is as large as QPSK symbols. At first glane, one ould hope to improve performane by exploiting multi-dwell proedures (i.e., olleting information from multiple frames before making a final deision). Unfortunately, this approah annot be used in VB-S2, where the frame length depends on the seleted oding and modulation pair. In partiular, the frame format is signaled to the reeiver by a 6-symbol physial layer signaling (PLS) field and the PLS field annot be deoded aurately prior to frame synhronization. Frequeny reovery unit Frequeny reovery is also one of the most ritial units due to the fat that VB-S2 mass market terminals will typially inorporate low-ost osillators (the terminal LNB osillator instabilities) and oppler effets, whih introdue large initial frequeny offsets (e.g., 5 MHz at 27.5 Mbaud). Phase reovery unit The arrier phase reovery algorithm has to ope with a residual arrier frequeny error from the arrier frequeny reovery unit as well as a strong phase noise aused by the terminal LNB RF osillator. INNER RECEIVER ISSUES Timing synhronization In this paper, we propose an IIR pre-filter sheme. Comparing with the pre-filter put before the TE in ('Andrea and Luise, 1993; 1996), we put the pre-filter between the TE and the loop filter. We do not propose higher order filter beause the 1st order loop filter is enough to suppress the noise and make both loop robustness and the loop filter parameters easily onfigurable. And the timing error spetrum after pre-filtering is shown in Fig.. Timing error spetrum ( 1 5 ) Normalized frequene (π) Fig. Timing error spetrum with pre-filtering Just as shown in Fig.3, the timing error fouses most information on the low frequeny range while the noise fouses on the high frequeny range. So the key parameter of the pre-filter is the bandwidth. After the bandwidth has been properly hosen, most information on the timing error has been reserved and most parts of the noise have been suppressed as shown in Fig.. Frame synhronization A differential orrelation based frame synhronization algorithm is given in (ETSI, 25), in whih the total 57 orrelation information is utilized (inluding 25 orrelations in SOF field and 32 orrelations in PLSCOE field). Under low SNR environment, the algorithm proposed in (ETSI, 25) generates many pseudo peaks whih are shown in Fig.9 and turns out to be too short to provide reliable onvergene at low SNRs by using the 57 orrelations. Sine the SOF is known, and only the information on adjaent orrelation is used, SOF information was not utilized suffiiently. In this paper, we strengthen the former algorithm by using the struture shown in Fig.5 with the detetion expression being shown in Eq.(5): loation = arg{max[ sum( n)]}, n L M 1 l sum( n) = r( n+ m+ l) r ( n+ m) ( m, l) (5) l= 1 m= P 1 + rn ( + M+ 2i+ 1) r( n+ M+ 2 il ) ( ), i=

4 Yang et al. / J Zhejiang Univ Si A 27 8(1): r(k) r * (k) To peak detetor Fig.5 Frame synhronization detetor where M is the SOF length whih is 26, P is the length of PLSCOE whih is 6, (m,l) denotes the taps of the differential orrelation, L is the maximum orrelation interval. Compared with the proposed algorithm in (ETSI, 25), the omputation omplexity inreases with L. The taps assoiated with shift register for omputing the orrelation should follow these steps: (1) Set all the registers to zero. (2) Shift the modulated SOF and a modulated and srambled ode word of PLSCOE into the iruit. (3) One the lowest registers beome nonzero, the tap assoiated with a register is just the omplex onjugate of the. Given that the modulated SOF and PLSCOE take only ±1, ±j, the taps only take these four possible values as well. Obviously, if L=1, the algorithm we propose redues to the one given in (ETSI, 25). The output is then further proessed by a peak searh algorithm. The onventional approah is to ompare the output orrelation sum with a predetermined threshold. If the value is larger than the threshold, it is a possible PLHEAER loation and an be alled the andidate. And the post verifiation is based on the detetion of the loation at a distane L s whih belongs to the set {329, 33282, 2169, 2219}. Comparing with the strategy proposed in (Sun et al., 2), we do not utilize the PLSC to verify that the PLSC an tolerate large noise but is very sensitive to the arrier frequeny offset (the range is less than f s ), espeially, while the frequeny arrier reovery unit is preeded by the frame synhronization, the residual frequeny offset is muh higher than f s. Frequeny reovery Many algorithms have been designed for frequeny detetor (F), e.g. the maximum-likelihood based Gardner frequeny detetor (GF) proposed in (Gardner, 199) and the redued version redued omplexity frequeny detetor (RCF) (Karam and Sari, 1995). These two algorithms need modifiation of the mathed filter, whih will inrease the omplexity and hardware expense. The algorithm proposed in (Mengali and 'Andrea, 1997) an only utilize 26 information symbols in SOF field and is onstrained by the number of the information symbols per-frame; it will spend more time to reah the frequeny synhronization. In this paper we propose a novel F algorithm whih does not require the PLHEAER information. The only required information is the position of the PLHEAER with the derivation of the algorithm being shown below. Suppose tr(k)=a s(k)exp(j2πf kt), where tr(k) is the transmission signal, the differene between the adjaent symbols s(k) and s(k 1) is ±π/2 aused by the π/2 BPSK modulation, we have 2 2 Im{ tr ( k)[ tr ( k 1)] } = Im{ A exp(π ft )} = A sin(π ft), (6) where Im{} and [] * denote imaginary part and onjugate, respetively. Modified tr(k)=a s(k)exp(j2πf kt)+n(k), and denote n(k)=n I (k)+jn Q (k), the term tr 2 (k) an be extended as follows: tr ( k) = [ as( k)exp(j2π fkt )] + n ( k) + 2 nkask ( ) ( )exp(j2π fkt ) (7) = [ ask ( )exp(j2π fkt )] + {[ ni ( k) nq( k)] + 2j n ( k) n ( k) + 2 n( k) a s( k)exp(j2π f kt)}. I Q Useful output of the algorithm is the loop filter output whih equals the average over many samples, not the value of an isolated sample, then the noise generated by the marked terms will be filtered out, whose remaining term is of the form Avg tr k tr k Avg tr k Avg tr k { ( )[ ( 1)] } = { ( )} {[ ( 1)] } = A exp(π ft). (8)

5 32 Yang et al. / J Zhejiang Univ Si A 27 8(1):28-35 Based on Eq.(6), the frequeny detetion error yields Eek E r k r k = A sin(π f T), 2 2 {( )} = {Im[ ( )[ ( 1)]]} (9) where r(k) is the reeived signal after AGC, and A is the amplitude of the symbol. The frequeny aquisition range is equal to the range of the algorithm given in (Mengali and 'Andrea, 1997), but the frequeny error information auray of the new algorithm an be greatly improved and does not require any information symbol. With PILOT and PLHEAER information being utilized, the onvergene speed an be greatly improved. And the number of training symbols inreases up to 9, whih inreases the auray and the robustness of the estimation. Phase reovery The MPSK modulated signal an be demodulated using M-power Costas loop, whih an inrease the hardware expense greatly when M is large. A feed-forward maximum-likelihood phase ompensation based algorithm is proposed by Sun et al.(2). But the algorithm an work only if the residual frequeny offset is lower than f s after F whih is diffiult for F unit to reah suh a riterion at low SNR. In this paper, the polarity-type Costas loop has been adopted by van der Wal and Montreuil (1995) and Huang et al.(2), whose expression is et () =sgn[ It ()] Qt () sgn[ Qt ()] It (). (1) QPSK But it is only suitable for QPSK. The modifiations must be done for 8PSK as follows et () 8PSK sgn( SI) SQ, if SI > SQ, = sgn( SQ) SI, else. (13) We do not adopt high power phase reovery algorithm, whih an not only greatly inrease the hardware expense, but also be a high energy onsumption solution. Sine the low energy onsumption design is one of the most ritial parts in VB-S2 sheme, lower power less than 2 is an optimization solution. SIMULATION RESULTS Timing synhronization The timing errors after loop filtering with and without pre-filtering are shown in Fig.6 and Fig.7 respetively. The jitter in Fig.6 has been greatly dereased and the pre-filtering has effetively suppressed the noise. In our simulation, pre-filtering bandwidth being 5~6 times that of the loop filter bandwidth is optimal. Fig.8 shows the performane of the timing synhronization in terms of the RMS of the residual timing frequeny error Number of sampling ( 1 symbols) Fig.6 LPF output without pre-filtering sgn[ Qt ( )] It ( ), if Qt ( ) > 2.1 It ( ), et () 8PSK = sgn[ It ()] Qt (), if Qt () <.1 It (), sgn[ It ( )] Qt ( ) sgn[ Qt ( )] It ( ), else. (11) We also present further modifiations to the above algorithm as shown below. enote S I =I 2 (t) Q 2 (t) and S Q =2I(t)Q(t), we yield that e(t) QPSK =sgn(s Q )S I, (12) Number of sampling ( 1 symbols) Fig.7 LPF output with pre-filtering

6 Yang et al. / J Zhejiang Univ Si A 27 8(1): Normalized frequeny RMS fs Fig.8 Performane of timing synhronization Frame synhronization In Eq.(5), the parameter L should be properly seleted. With the inrease of L, performane of the andidate frame synhronization algorithm an be greatly inreased. But onsidering the hardware expense, there should be tradeoff between the performane and the expense. In our simulation, we hoose 7. Performane of the andidate is shown in Fig.9. The simulation is performed with 1. db SNR and.2f s (the maximum tolerable frequeny offset of the Gardner timing algorithm) residual frequeny offset. Extensive omputer simulations have been performed. Table 1 summarizes the mean time and the time with 99.9% onfidene to aquire frame synhronization under the environment with 1 M symbol rate,.5 db E s /N and.2f s arrier frequeny offset. Modulation Table 1 Time to frame synhronization Method Mean time (ms) Time with 99.9% onfidene (ms) QPSK SOF QPSK SOF+PLSC PSK SOF PSK SOF+PLSC Table 2 gives the mean time to aquire frame synhronization under environment with 2 db E s /N and different symbol rates. It shows learly that by suffiiently utilizing the PLHEAER, the aquiring time an be greatly shortened. Comparing with the performane proposed by Sun et al.(2), signifiant improvement has been obtained Number of sampling ( 1 symbols) (a) Number of sampling ( 1 symbols) () Number of sampling ( 1 symbols) (b) Number of sampling ( 1 symbols) (d) Fig.9 Candidate algorithm performane with different L s. (a) L=1; (b) L=3; () L=5; (d) L=7 Middle subfigures and top subfigures denote the orret and the andidate positions, respetively. Bottom subfigures are the ones ombined with the orret position (amplitude is 1.) and andidate position (amplitude is.5)

7 3 Yang et al. / J Zhejiang Univ Si A 27 8(1):28-35 Table 2 Time to frame synhronization with different symbol rates Modulation Method Mean time (ms) Time with 99.9% onfidene (ms) QPSK SOF QPSK SOF+PLSC PSK SOF PSK SOF+PLSC Frequeny reovery Computer simulations were onduted to test the arrier reovery algorithm for pilotless modes in the presene of strong AWGN and the phase noise speified by VB-S2. The reeived signal has a 25 M symbol rate and a 5 M frequeny offset. Fig.1 shows the RMS frequeny error of the F unit operating on QPSK frames in the pilotless mode. Autoorrelation is aumulated only on 57-symbol PLHEAER. At 2 db, the RMS error is f s over 2 frames. performane of the phase reovery algorithm with phase noise and without pilots. Table 3 Phase reovery performane Mode RMS ( ) Cyle-slip rate QPSK <1 8 8PSK <1 8 System performane The ultimate goal of the reeiver design is not to degrade the performane of diret deision when ompared with ideal AWGN hannel. The urves of the QPSK and 8PSK SER vs SNR are shown in Fig.11 and Fig.12 respetively. It an be easily seen that the SER urve of the simulation results approahes the theoretial one, espeially at low SNR. But with the inrease of the SNR, the simulation SER urve departs from the theoretial one beause of the quantization preision RMS frequeny error 1 SER Simulation P s Theoretial P s Fig.1 Performane of frequeny reovery unit Fig.11 QPSK SER vs SNR Phase reovery The phase error and the residual frequeny offset an be anelled by PF loop after the F loop loks. Under the ondition of AWGN, we should onsider the tradeoff between the onvergene speed and the stability error, inluding the apability of ombating the white noise and the phase noise. We adopt Eqs.(1), (11) and Eqs.(12), (13) into aquisition proess and traking proess respetively. Furthermore, in eah modulation mode, there are two sets of filter parameters for the aquisition proess and the traking proess respetively. Table 3 summarizes the SER Simulation P s Theoretial P s Fig.12 8PSK SER vs SNR

8 Yang et al. / J Zhejiang Univ Si A 27 8(1): The inner reeiver onneted with the LPC deoder was extensively simulated to verify the final performane. The performane of MPEG PER (paket error rate) in the presene of the phase noise in AWGN hannel is shown in Fig.13 and Fig.1. PER PER CONCLUSION 1/2 3/5 2/3 5/6 8/9 9/1 3/ / Fig.13 Performane of QPSK 3/5 2/3 3/ Fig.1 Performane of 8PSK In this paper we have onsidered the systemati design of inner reeiver algorithms for VB-S2 based transmission systems. Taking into aount the physial layer frame struture of the VB-S2 senario, we have derived algorithms with lose-tooptimum performane while providing robust and fast aquisition. Simulation results showed that the reeiver does not degrade the performane when ompared with ideal AWGN hannel. 5/6 8/9 9/1 Referenes Casini, E., de Gaudenzi, R., Ginesi, A., 2. VB-S2 modem algorithms design and performane over typial satellite hannels. Int. J. Satell. Commun. Network., 22(3): [doi:1.12/sat.791] 'Andrea, N.A., Luise, M., esign and analysis of a jittter-free lok reovery sheme for QAM systems. IEEE Trans. Commun., 1(9): [doi:1.119/ ] 'Andrea, N.A., Luise, M., Optimization of symbol timing reovery for QAM data demodulators. IEEE Trans. Commun., (3): [doi:1.119/ ] ETSI, 23. EN3 21 v igital Video Broadasting (VB): Framing Struture, Channel Coding and Modulation for 11/12 GHz Satellite Servies. ETSI, 2. EN32 37 v1.1.1 igital Video Broadasting (VB): Seond Generation Framing Struture, Channel Coding and Modulation Systems for Broadasting, Interative Servies, News Gathering and Other Broadband Satellite Appliation. ETSI, 25. TR v1.1.1 igital Video Broadasting (VB): User Guidelines for the Seond Generation System for Broadasting, Interative Servies, News Gathering and Other Broadband Satellite Appliation. Gardner, F.M., A BPSK/QPSK timing-error detetor for sampled reeivers. IEEE Trans. Commun., 3(5): [doi:1.119/tcom ] Gardner, F.M., 199. Frequeny etetors for igital emodulators via Maximum-likelihood erivation ESA Final Rep: Part 2, ESTEC Contrat 822/88/NL/G. Huang, Z.J., Yi, Z.Q., Zhang, M., Wang, K., 2. 8PSK emodulation for New Generation VB-S2 Communiations, Ciruits and Systems. ICCCAS 2, 2(27-29): Hughes Network Systems, 23. LPC Frame Synhronization. VB-S2-1. Karam, G., Sari, H., A redued-omplexity frequeny detetor derived from the maximum-likelihood priniple. IEEE Trans. Commun., 3(1): [doi:1.119/ ] Mengali, U., 'Andrea, N.A., Synhronization Tehniques for igital Reeivers. Plenum Press, New York, USA. Nemer, E., 25. Physial Layer Impairments in VB-S2 Reeivers. Consumer Communiations and Networking Conferene, p Sun, F.W., Jiang, Y., Lee, L.N., 2. Frame synhronization and pilot struture for VB-S2. Int. J. Satell. Commun. Network., 22(3): [doi:1.12/sat.793] van der Wal, R., Montreuil, L., QPSK and BPSK demodulator hip-set for satellite appliations. IEEE Transations on Consumer Eletronis, 1(1):3-1. [doi:1.119/3.3737]