Journal ofScience & Technology 101 (2014) 118-121 A^ An Improved Blind Algorithm for Transmit - Reference UWB Receivers Tran Manh Hoang, Pham Van Binh* Hanoi University of Science and Technology, No 1, Dai Co Viet Str., HaiBa Trung, Ha Noi Viet Nam Received: March 03, 2014; accepted' April 22, 2014 Abstract This paper proposes an improved blind receiver algonthm for Transmit-Reference Ultra-Wideband (TRUWB) systems which uses rank-1 singular value decomposition (SVD) More specrficaliy, oversampling by integrate-and-dump to divide each frame into multiple samples associated with using only samples from 'useful" sub-frame which conveys data signal helps create a flexible receiver which meets various pertormance / complexity requirements Finally, the practical timing error issue is also investigated under vanous transceive schemes to evaluate and compare the BER performance of the proposed receiver algonthm with classic receiver algonthms Keywonds: TR-UWB, SVD Blind receiver algorithm I, Introduction Smce approved by FCC m 2002 [1] and by ECC in 2005 [2], Ultra-Wideband (UWB) has become a strong candidate for short range wireless communication systems due to many of its attractive characteristics, Transmit-Reference (TR) transceiver, a promismg non-coherent scheme introduced in [3, 4], can be implemented with reduced complexity TRUWB receiver detects data symbols without having to estimate individual multipath components of the channel at the expense of lower data rate and inferior noise performance (due to cross-conelation terms between signal and noise) Therefore, it is suitable to low power, lower rate UWB systems e.g WPAN-LR in IEEE 802.15.4a [5] So far, many research papers have been published on TR-UWB concept The original scheme can only support very low data rate and uses sliding window over multiple frames which is not practical in wideband circuits, a more accurate data model and transceiver scheme was proposed in [6] but the algorithm is too complex because of large matrix inversion An autoconelation receiver m [7] is based on Voltena model, of which the complexity grows exponentially with the channel length In [8], the authors propose oversampling by integrate and dump, but aim at higher rate systems with multiple delays A very simple TR-UWB scheme was presented in [9] in which each symbol has only one frame, with single delay However, this generic scheme suffers significant performance loss (even more than dB as stated in the same paper) • Corresponding author: Tel.: Tel: (+84) 912.629.062 Email Binh.phamvan@hust.edu.vn This paper will shortly present the TR concepi and discuss the receiver's performance m this generic scheme Subsequently, both blind algonthm (estimate the channel parameters, and detect the data symbols at the same time) and zero-forcing receiver (with known channel) are proposed These algorithms will be simulated for various scenanos and results will be discussed in depth, where the timmg issue will also be addressed A Generic TR-UWB Transceiver The Transmit-Reference concept is best introduced by a genenc transceiver in which BPSK symbols are mapped into frames Each symbol contains just one frame, where two pulses separated by D second are transmitted The first pulse is the reference g(t) (Gaussian monocycle) with fixed polarity, while the second pulse has the conesponding symbol value s.g(t: — D) The frame period Tf and the delay value D are implicitly chosen such that there is no interference between consecutive pulses or consecutive frames (i e, no IPI and no IFI) For simplicity, we choose r^ = 3DandD >rft, where Tj is the maximum channel delay spread The transmitted signal for a single symbol s is: y(t) = g{t) + s.g(t-D) The received signal (at antenna's output) is: r ( t ) = hit) -h s.ft{t- D) + n(t) Here n(t) is the white Gaussian noise and h(t) is the 'composite' channel: hit) = git)* k_p(t) Where /ipf^j is the physical channel impulse response Journal ofScience & Technology 101 (2014) 118-121 ••Jv_ Jr^ Fig A genenc TR-UWB receiver Fig One received signal frame of r(t) Fig One received frame of j:CrJ Since there is no IPI and no IFI, if we multiply the received signal r(£) in Fig.2 with a delay-by-D version of itself, the resulting signal %(t) at the input of the integrator will be: ar{t) = r(£)r(t - D) == sh^Zit) -I- noise More specifically, oversampling sub-frame II of a:,(C) conesponding to the transmitted symbol S[, we have : x,inT,) JD+J Here, all the cross-terms between signal and noise are incorporated into "noise" The data symbol can then be detected by ^sign li/'''*l (1) While this is a very simple receiver, i e data symbol is detected by taking the sign of the ADC's output, its BER vs SNR performance is poor The loss (when compared to coherent RAKE receiver) is much greater than dB This is because of following problems: (i) one frame is now three times longer and the required signal energy per frame is twice higher, (ii) the integration is over tiie whole frame while the "usefiil signal" s.h^it) concentrates only in one subframe (see Fig,3), (lii) this signal is not evenly distributed (but usually exponentially decaying proportional to the channel delay profile) The first problem is the system design constraint, but the last two problems can be mitigated by following receiver algorithms J Receiver algorithms In our proposed receiver, instead of having just one sample per frame, we now use oversampling (also by "mtegrate and dump") to get multiple samples per frame Each frame has SW samples, but for this blind algorithm we only need N samples from the sub-frame II (illusfrated in Fig,3) which contains "usefril signal" s.h^t) -• s,.K Where Tj = — is the sampling rate and h^ is the "new" channel coefficient By oversampling with integrate and dump, the charmel is divided mto several segments Each segment is represented by a coefficient h^, which is actually the energy of the conesponding segment (therefore always positive) h„ = h^it)dt •'nTs Collecting all samples within the i-th frame, Vn = 0,1, ,(W — 1), into a vector, we have the data model for a single symbol s, Where I , - [^,o,:*:,j^,.,,, j:[ft,_i]'' and h=[ho,h^ 'h^_'^Y 3.1 ZF receiver algorithm Under certain cfrcumstances when channel is known (e.g by measurement or by training), we can easily denve Zero Forcing (ZF) receiver algorithm from the data model in (2) when noise and other crossterms are assumed zero Each data symbol is detected separately by •• sign{h'^Xi} (3) Journal ofScience & Technology 101 (2014) 118-121 Fig BER vs SNR plots for vanous receivei algorithms Fig The algorithm's robusteess E enor 3.2 Blind receiver algorithm Consider the transmission of a data packet of M consecutive symbols where the channel is assumed constant Stackuig all vector;!', from (2) into a matrix, we have the data model for multiple symbols •'^M-lJ = [so/t,Sih, ,SM-i/i] X = hs^ (4) (5) Given the received matrix X, we can estimate both vectors h and s m a typical rank-1 approximation problem by using SVD and taking the first left and right singular vectors The data symbols can be subsequently detected by hard decision X=UlV.s = sign[vo) (6) The known polarity ambiguity caused by SVD can be easily solved by applymg differential coding to the transmitted bits 3.3 Remarks It can be seen that the symbol detection operator in (I) is actually equivalent to (3) if h^ is replaced with an all-one vector [1,1, ,1], The implication is that the generic receiver sums all x, with equal weights while the ZF receiver chooses h, as weighting factors when summing all x, together The blind receiver also indirectly use weighting factors in its approximation problem Note that omitting subframe I and sub-frame 111 also means usmg zeros as the weighting factors for the samples in these sub-frames Therefore, our maximum-ratio-combining (MRC) approach (i,e stronger signals have proportionally greater weights) does have superior received SNR than the generic receiver This will be confirmed shortly in the simulation results presented next 4.1 Settings We simulate a TR-UWB system, where lOOQ BPSK symbols are transmitted through IEEE CMl channels [10] The chosen UWB pulse is the secood derivative of the Gaussian monocycle of duration [ns] The delay between the reference pulse and the information pulse is D = 64 [ns], which is longer than most CMl channels The frame penod is Tf = 3D = 192 [ns], 1000MonteCarlorunsareexecuted to obtain BER vs SNR plots for different receiver algorithms under different scenarios Here SNR is defined as the pulse energy over the noise power spectral density in dB BER performance of the blind receiver algorithm is compared with the generic algonthm The reference curve is the ZF receiver when channel is known Additionally, we also investigating the effect of sampling rate on BER performances by varying the number of samples per frame, N e [1,2,4,16] As timing is an important issue in every communication systems, we will investigate the system's robustness against tuning enor by modifying the algorithm to include the whole frame (all subframe I, II, and III) when integrate and dump at the cost of tripling the number of samples per frame In this case, the BER performances are compared between an assumed perfect timing receiver and the modified receiver when there are timing enors at both framelevel and sample-level as illustrated in Fig.4, In this simulation, the timing enor T IS generated randomly within [—D,D] (positive value of T means frame shift to the right, negative value of T means frame shift to the left) Journal of Science & Technology 101 (2014) 118-121 4.2 Results Fig.5 shows that our proposed algonthms have superior performance to the genenc receiver The gain can be from dB to dB, where the former is of the case N = and the latter is of the ZF receiver Note that N — is actually quite similar to the generic receiver except that it only mtegrate over subframe II, not the whole frame, which helps avoid noise accumulation in the region with no "useful signal" As N increases, the BER performance improves but it can he seen that when N reaches a certain value, the gam will be negligible In this case that value is N = Since the algorithm's computationcomplexity IS proportional to N, keeping N small enough can help reduces receiver's complexity significantly The modified algorithm's robustness agamst timing enor is shown in Fig 6, where the performance loss IS very small (less than dB) This can be explained by looking at the shifted frame in Fig.4, As long as the whole shifted frame contains all of the "usefiil signal", i.e —D < T < D, we can shll collect signal energy and detect symbols correctly Conclusions In this paper, we have developed a blind algorithm for TR-UWB receivers usmg integrate and dump for oversampling and a simple rank-l SVD which have superior BER performance than the generic receiver The proposed algonthm can adapt the sub-Nyquist sampling rate to available hardware, and to achieve a suitable tradeoff between performance and receiver's complexity It is shown by simulation that a very small value of N can have good enough performance (less than dB loss) while significantly reduces complexity The modified algonthm is very robust against timing enor, the working margin is[—D.D], which is very favorable when compared to some coherent receivers when even a sub-nanosecond timing enor can break down the receiver's algonthms Therefore, it is quite straightforward to derive the synchronization algonthm as well as FPGA implementation of this system [ ] "FCC notice of proposed mle making, revision of part 15 of the commission's rules regarding ultrawideband transmission systems," Washington, D C , 2002, [2,] "Harmonise radio spectrum use for UlfraWideband Systems in the European Union," Copenhagen, Denmark, 2005 [3 ] N v, S, a A D, a, K W, a R G a R H, a H Tomlinson, "Delay Hopped Transmitted Reference Expenmental Results," 2002 [4] R Hoctor and H Tomlinson, "Delay-Hopped Transrmtted-Reference {RF} Communications" [5 ] "Homepage of IEEE 802 15.4a Task Group," [Online] Available hnp://www.ieee802 org/15/pub/TG4a,html [6 ] Q,H Dang and A.J van der Veen and A, Tnndade and G Leus, "Signal model and receiver algorithms for a transmit-reference ultrawideband communication system," IEEE Journal on Selected Areas in Communications, 2006 [7 ] K Witrisal and G Leus and M Pausini and C Krall, "Equivalent system model and equalization of differential impulse radio UWB systems," JSAC, vol 23, pp 1851-1862,2005 [8.] Q H- D, a A.-J, v d Veen, "A Deconelatmg Multiuser Receiver for Transmit-Reference UWB Systems," JSTSP (2007) 431-442,2007 [9.] L Yang and G.B, Giaimakis, "Ultra-wideband communications, an idea whose time has come," SPm, voL 21, 2004 [10.] Andreas F, Molisch and Karman Balaknshnan and Chiachm Chong, "IEEE 802,15.4a channel model - final report," 2004 ... The implication is that the generic receiver sums all x, with equal weights while the ZF receiver chooses h, as weighting factors when summing all x, together The blind receiver also indirectly... simulate a TR-UWB system, where lOOQ BPSK symbols are transmitted through IEEE CMl channels [10] The chosen UWB pulse is the secood derivative of the Gaussian monocycle of duration [ns] The delay between... whose time has come," SPm, voL 21, 2004 [10.] Andreas F, Molisch and Karman Balaknshnan and Chiachm Chong, "IEEE 802,15.4a channel model - final report," 2004