Performance Analysis of Iterative Receiver in 3GPP/LTE DL MIMO OFDMA System

Transcription

1 Performance Analysis of Iterative Receiver in 3GPP/LTE DL A System Laurent Boher, Rodolphe Legouable and Rodrigue Rabineau Orange Labs, 4 rue du Clos Courtel, Cesson-Sévigné Cedex, France {laurent.boher;rodolphe.legouable;rodrigue.rabineau}@orange-ftgroup.com Abstract This paper provides a link level study of a downlink B3G system using A techniques. A 3GPP/LTElike scenario, in terms of frame structure and main system parameters is considered. We especially focus on the impact of iterative receiver implementation when transmitting via double Alamouti scheme. A performance analysis is carried out regarding to the channel coding scheme implementing into the iterative loop. Several scenarios have been considered (different modulation and coding schemes (MCS), real or perfect channel estimation...) to quantify the gain provided by such iterative receivers. Keywords -A, iterative receiver, B3G. I. INTRODUCTION The Third Generation Partnership Project (3GPP) is currently normalizing the Long Term Evolution (LTE) of the third generation (3G) cellular system [1]. This paper is focusing on LTE downlink (DL). DL LTE targets are challenging: a Mb/s peak data rate in 2 MHz spectrum allocation, an average user throughput per MHz multiplied by 3-4, a cell-edge user throughput per MHz multiplied by 2-3 and a spectrum efficiency multiplied by 3-4 compared to 3GPP Release 6 High Speed Data Packet Access. The DL LTE air interface is based on orthogonal frequency division multiple access (A), which offers good flexibility and performance for a reasonable complexity. The users of a same cell are multiplexed in frequency, each user s data being transmitted on a subset of the sub-carriers of an orthogonal frequency division multiplexing () symbol. Depending on its velocity and the reliability of its channel quality indicator, a user may be allocated either to subset of sub-carriers scattered over the whole bandwidth (distributed user) or to subset of adjacent sub-carriers (localized user). In order to achieve the challenging spectral efficiency and user throughput targets, multiple transmit and receive antennas schemes (), turbo coding and link adaptation are also included in the physical layer specifications. In [2], it has been shown that double Alamouti scheme is a good trade-off between robustness and throughput. However, this scheme causes co-antenna interference that can be advantageously cancelled with iterative receiver. This paper aims at quantifying the link performance gains, being as much cell coverage gain, provided by such receiver compared to a classical one, i.e. without iterative loop processed between the equalization and the decoding. We have considered several scenarios in order to assess the impact of iterative receivers implementation according to the channel coding scheme used in the iterative loop, the coding block size and the channel estimates errors. This paper is organized as follows. Section II describes system model and notations. In section III, the iterative receiver is detailed. Simulation parameters are listed in section IV. Finally, simulation results are given in section V and conclusion is drawn in section VI. II. SYSTEM MODEL A. Transmitter The block diagram of the complete LTE-like transmitter is depicted in Figure 1. Firstly, information bits of each user are scrambled with a pseudo-random sequence. Channel coding is then performed on the scrambled bits. Two different codes are considered: Duo-binary convolutional turbo coding (DBTC) based on recursive systematic 8-state convolutional codes and classical 128-state convolutional coding (CC). Coded bit are then interleaved and modulated into QAM complex symbols. encoding is performed afterwards in order to exploit the spatial diversity of the channel. Space Time block coding allows transmission of groups of symbols on the different antennas for each subcarrier. mapping affects coded symbols of each user to a specific subcarriers subset for the 4 transmit antennas. f raming added pilots and control subcarriers and modulation is performed on each antenna before transmission. B. Channel model The studied system is made of transmit and N r receive antennas. For a discrete time representation of the time and frequency selective channel, the frequency response of subchannel between the m-th transmit antenna and the n-th receive antenna can be expressed by: user x s Bit Scrambling Coding Interleaving Coding user 1 s1 Bit Scrambling Coding Interleaving Coding user K sk Bit Scrambling Coding Interleaving Coding Fig. 1. A transmission scheme x1 xk /8/$ IEEE

2 Channel Estim. User i Demod. Demod. Deframing Deframing Demap. Demap. Channel Estim. x i Decoding s i Soft Bit Channel Demap. Deinter. Decoding Soft Bit Interleaving channel decod. loop Bit Descramb. interference cancellation loop Fig. 2. iterative receiver scheme L H nm,j = h nm,l e j2πml N FFT (1) l= where h ij,l is a time varying channel tap, L the number of taps and N FFT the Fast Fourier Transform (FFT) size. demodulation is applied at each receive antenna n and signal r n,j is obtained for each subcarrier j. Therefore symbol x m,j sent on the j-th subcarrier is only affected by a complex coefficient and noise: ρ r n,j = h nm,j x m,j + n n,j (2) m=1 with n n,j the FFT processed Additive White Gaussian Noise with variance σn/2 2 per dimension and ρ the Signal Noise Ratio measured at the reception side. For each subcarrier affected to user i, the signal block r i C TNr 1 collecting received symbols on the N r antenna over T time slot is so: ρ r i = H i x + n i (3) with H C TNr TNt containing the channel matrices of the T time slot and N C Nr T the zero-mean complex Gaussian noise. With associating channel matrix and coding, the equivalent received vector r i is given by: ρ r i = H i s i + ñ i (4) where H i and ñ i stand respectively for equivalent channel matrix and noise vector. III. ITERATIVE RECEIVER After demodulation on each antenna, each user recovers his data symbols on his affected subcarriers. After, the reception process consists in detecting information bits thanks to a equalizer and a channel decoder. The corresponding receiver is depicted on Figure 2. We consider two types of process. First, equalization and channel decoding are performed separately. In the second case, equalization and channel decoding exchange soft information on coded bits according to turbo equalization principle [3]. A. Low complex equalizer Iterative equalization consists in cancelling interference on received signals thanks to estimated symbols [4]. The output of the MMSE Interference Canceller (IC) can be expressed as: s i = p H i r i q H i ŝ i (5) where ŝ i is an estimate of s i given by previous iteration. The two vectors p i and q i are respectively N R 1 and Q 1 complex vectors optimized under the MMSE criterion: (p opt i, q opt i ) = arg min s i s i 2 (6) p i,q i Since no prior information on transmitted symbols is available at the first iteration, the equalization process is thus reduced to a classical linear MMSE solution: s (1) i = [ H H i H i + σ2 n I] H H σs 2 i r i (7) For next iterations, coefficients of filters depend on estimated symbols ŝ i. Exact solution of the calculation of these filters can be found in [5], [6]. However, as each filter has to be determined by matrix inversion, the optimum receiver would induce a high complexity as new filters have to be computed for each symbol. A classical approximation consists in selecting a solution invariant per Space Time (ST) block. In that case, filters are computed from the variance σŝ 2 i of estimated signals (MMSE- IC): p MMSE IC σs 2 p i i = σs 2 + σŝ 2eT H (8) i H p i and q MMSE IC i = H H p MMSE IC i (9) with p i = σ 2 s[ H H H (σ 2 s σ 2 ŝ )+σ2 ni] H H e i and e i is a Q 1 vector with all elements equal to except the i-th element that is equal to 1. B. Demapping/ After equalization, symbols are sent to the soft demapper that produces Logarithm Likelihood Ratio (LLR)

3 of coded bits Λ eq [i, k] according to the formula: ) s X Λ eq [j, i] =ln j exp ( si βis 2 1 α i ( ) () s X j exp si βis 2 α i where X j b denotes the subset of X for which j-th bit is equal to b whereas β i is the bias introduced by the equalizer and α i represents the total variance of the interference terms plus the residual noise, such that: β i = p H i He i (11) α i = σsβ 2 i (1 β i ) (12) The soft estimated symbols ŝ i are computed from the knowledge of the a posteriori LLR Λ dec [j, i] provided by the turbo decoder. Owing to the presence of the interleaver, independence of modulated bits is assumed and we get: ŝ i = s Pj,i cj (13) s X where P cj j,i = Pr ( ĉ j i = cj) = s:[c 1,..,c m ] exp(cj Λ dec [j, i]) 1 + exp(c j Λ dec [j, i]) is the probability to have the value c j for the bit ĉ j i. (14) C. Channel decoder LLR values provided by soft demapping are deinterleaved and sent to the channel decoder. When iterative equalization is considered, a posteriori LLR of coded bits obtained in output of the decoder are sent back to the decoder. If convolutional code is used, there is as many iterations of convolutional decoding than iteration of detection (as convolutional decoding is not iterative). Thus, complexity increase due to the iterative process is relatively important. On the other hand, when turbo coding is considered, it is possible to adjust the number of iterations of channel decoding. In order to propose a receiver with a minimum increase in complexity, we decided to perform the same number of turbo decoding iterations as of equalization iterations. Thus complexity of turbo-coded iterative equalizer is only slightly increased compared to a non-iterative solution (such complexity analysis is detailed in [7]). We will notice that in that case turbo-coded iterative receiver is less complex than convolutional coded iterative receiver due to less complex trellis decoding. IV. SIMULATION PARAMETERS The main purpose of the simulation campaign is to quantify the gain obtained when implementing iterative receivers in various scenarios. To assess the performance of the LTE downlink, link level simulations are conducted in different scenarios at 3Km/h mobile speed. The channel used in the simulations is the SCME (Space Channel Model Extension) channel model described in [8] and [9] and which corresponds to the SCM-B channel typical of the Urban Macro with low spread and close to the well known Typical urban environment. The main parameters are summarized in Table I and are closed to those already taken into account in [2] and implemented inside the French RNRT OPUS project. We consider the parameters and the frame structure devoted to the MHz bandwidth as defined in the future 3GPP/LTE standard [1]. However for channel diversity consideration, we have only implemented the distributed Resource Block (RB) allocation. The user data symbol information are transmitted either by encoded packets of 1RB or RBs. The bit interleaving is carried out over the packet length, which is dependent on the channel coding rate (1/2 or 3/4) and on the modulation level (QPSK, 16QAM or 64QAM). We have adopted the 4x2 double Alamouti coding scheme, implemented in the time domain that is a good trade-off between robustness and spectral efficiency. Double-Alamouti space-time coding consists in simultaneously transmitting two Alamouti codes on two blocks of two antennas []. Double-Alamouti matrix is the following: [ ] T x1 x X DA = 2 x 3 x 4 x 2 x 1 x 4 x (15) 3 For equalization process, either perfect or real channel estimation is performed. In the latter case, RF front-ends are modelled and the channel coefficients are estimated from the symbol pilots via Wiener filtering in the frequency domain applied for each Tx/Rx branch. Then linear interpolation in the time domain is implemented in order to estimate the channel coefficients on the whole frame. Channel coefficients are then transmitted to the equalizer. In real channel estimation configuration, channel estimates are not updated at each iteration. Three iterative systems have been compared. First, a system with only iterations in channel decoding. In that case MMSE equalization is followed with 8 iterations of duo-binary turbo decoding (curve labelled (TC) in section V). Then, a system with only iterations in interference cancellation (IC). Here FFT size 24 Sampling frequency (in MHz) Number of modulated carriers 6 Guard interval (in samples) 72 scheme Double Alamouti 4x2 Number of users 5/5 RB size (in carriers) 12/12 RB allocation Distributed Number of complex data 12/12 schemes QPSK,16-QAM and 64QAM Channel coding rates 1/2 and 3/4 Bit interleaver size 24/72 Channel coding type Channel estimation Propagation channel type Convolutional K=7 for MMSE-IC DB TC with 8 it. for MMSE DB TC inside MMSE-IC TABLE I SYSTEM PARAMETERS Perfect/real Typical urban SCME at 3Km/h

4 Eb/N (db) Eb/N (db) Fig. 3. Performance results in QPSK 1/2, 1RB, perfect channel estimation. Fig. 5. Performance results in 16QAM 3/4, RB, perfect channel estimation Eb/N (db) Eb/N (db) Fig. 4. Performance results in QPSK 1/2, RB, perfect channel estimation. Fig. 6. Performance results in 64QAM 3/4, RB, perfect channel estimation. five IC iterations are performed between the convolutional code and MMSE equalization (curve labelled (IC+CC)#i in section V, where i denotes the i th iteration). Finally a system with iterations in both channel decoding and interference cancellation. Eight iterations of turbo decoding and equalization are so performed (curve label (IC+TC)#i in section V). V. SIMULATION RESULTS As usual, the performance results of the different detectors are evaluated by a comparison of the Bit Error Rate () as a function of the E b /N ratio, with E b the useful bit energy and N the noise power. Figures 3, 4, 5, 6, 7 and 8 are a representative panel of numerous performance results. This set of curves allows to draw the following conclusions: System with iterative interference cancellation with convolutional or turbo codes outperform turbo-coded MMSE receiver. The gains are in order to 2dB to 5dBs at = regarding to the MCS. Besides, turbo-coded iterative receiver obtain equivalent or better performance then a convolutional system with less complexity. In most cases, only two iterations of turbo-coded iterative receiver are necessary to outperform the MMSE turbo coded receiver with 8 iterations. Thus, iterative equalization can be viewed as a way to increase performance but also as a way to decrease reception complexity and latency. Globally, the obtained performance are better when using RBs instead of 1RB showing a better exploitation of the channel diversities (mainly in the frequency domain) and highlight the role played by the bit interleaving. We will notice that the increase in interleaving size is particularly beneficial to the iterative turbo-coded system (gain of more than 1dB in QSPK 1/2 for (IC+TC) instead of only.5db for (IC+CC)). The impact of a bad channel estimation is more detrimental to a iterative receiver compared to a classical one. There is a degradation of 3dB for iterative receiver instead of 2dB for others. However, when modulation order increases, the effect of real channel estimation becomes equivalent between iterative and non-iterative receivers.

5 Fig. 7. Eb/N (db) Performance results in QPSK 1/2, 1RB, real channel estimation Eb/N (db) Fig. 8. Performance results in 64QAM 3/4, RB, real channel estimation. VI. CONCLUSION In this paper, we have studied B3G A link level performance by considering 3GPP/LTE-like scenario. We have focused our attention on the use of advanced receiver in order to improve the link level performance that can be translated into cell coverage increase. Indeed, we have proposed to implement iterative receiver to deal with co-antenna interference, unavoidable when wishing for data cell throughput increase. The study has showed significant gain with low complexity increase especially when implementing a DB TC into the iterative loop. An hardware implementation has been carried out in the Orange labs confirming the feasibility and the performance of such receiver. REFERENCES [1] 3GPP TSG-RAN, 36 series: Evolved UTRA aspects, [2] D. P. Huy, R. Legouable, D. Ktnas, L. Brunel, and M. Assaad, Downlink B3G A link and system level performance, IEEE Vehicular Technology Conference (Spring), May 28. [3] C. Douillard, A. Picart, P. Didier, M. Jezequel, C. Berrou, and A. Glavieux, Iterative correction of intersymbol interference: Turboequalization, European Transactions on Telecommunications, vol. 6, no. 5, Sept [4] A. Glavieux, C. Laot, and J. Labat, Turbo equalization over a frequency selective channel, International Symposium on Turbo Codes and related topics, pp. 96 2, [5] M. Witzke, S. Baro, F. Schreckenbach, and J. Hagenauer, Iterative detection of signals with linear detectors, Asilomar Conference on Signals, Systems and Computer, vol. 1, pp , 22. [6] D. Reynolds and X. Wang, Low complexity turbo-equalization for diversity channels, Signal Processing, vol. 85, no. 5, pp , May 21. [7] L. Boher, R. Rabineau, and M. Hélard, FPGA implementation of an iterative receiver for - systems, To appear in IEEE Journal on Selected Areas in Communications, 28. [8] 3GPP TSG-RAN, 3GPP tr , physical layer aspects for evolved UTRA (release 7), v1.3.1 (26-5). [9] D. S. Baum, J. Hansen, G. D. Galdo, M. Milojevic, J. Salo, and P. Kysti, An interim channel model for beyond-3g systems - extending the 3GPP spatial channel model (SCM), IEEE Vehicular Technology Conference (Spring), May 25. [] S. Baro, G. Bauch, A. Pavlic, and A. Semmler, Improving BLAST performance using space-time block codes and turbo decoding, IEEE Global Telecommunications Conference, pp , 2.