An Interleave-Division Multiplexing (IDM) based Modulation
for 3GPP LIE Downlink
Dageng Chen, Yi Wang, and Jiayin Zhang Huawei Technologies Co., Ltd. No.450, Jinyu road, Shanghai, China, 201206 Tel: +86-21-50993168, Fax: +86-21-50993619 E-mail: {dagengchen.yLwang.zhangjiayin}@huawei.com Abstract-Interleave-Division Multiple-Access (IDMA) is an efficient multiple access technique to improve high spectrum efficiency for cellular communications. In this paper IDM-based modulation (called layered modulation (LM)) is proposed for OFDM systems. The idea of LM is to multiplex multiple parallel layers of data streams by performing repetition and layer-specific interleaver on each layer. Since LM code bits are fully distributed in time-frequency plane, by the aid of low-complexity chip-by-chip interference cancellation time-frequency diversity can be well exploited in LM. Applying LM to 3GPP LTE downlink, we compare LM to M-PSKlQAM modulation. Numerical results show that the LM can achieve superior performance over M-PSK/QAM. The performance gain over QPSK, 8PSK and 16QAM is 1.1dB, 2.2dB and O.6dB, respectively. Index Terms-Orthogonal Frequency Division Multiplexing (OFDM), Interleave-Division Multiple-Access (IDMA), Layered-Modulation (LM), Code Divison Multiple Access (CDMA) I. INTRODUCTION Combined orthogonal frequency-division multiple-access (OFDMA) and code-division multiple-access (CDMA) is a promising multiple access solution for future broadband wireless communications. OFDMA has the advantages of low-complexity, robustness to multi-path channel, suitable for link adaptation and multiuser diversity etc. But frequency diversity is not fully exploited in OFDMA since only part of frequency band is assigned for one user. On the other hand, in CDMA every user occupies the whole frequency band and frequency diversity can be well exploited at the cost of increased complexity. IDMA can be regarded as a kind of CDMA technique. It spreads code bits by repetition sequence, as performed in CDMA. The difference is that IDMA uses a user-specific interleaver to distinguish users whereas CDMA utilizes user-specific signature sequence for each user. Many efforts have been made in IDMA researches [1-6]. The primary idea ofIDMA is presented in [1] as an improved CDMA scheme, where interleaver in CDMA is put after spreading unit, and with a chip-by-chip multiuser detection the modified CDMA scheme can achieve significant performance gain over CDMA. In [2], Li Ping firstly proposed to use only interleaver (without user-specific signature) to distinguish users, and receive algorithm is derived. Sequential researches focus on the performance of IDMA, MIMO-IDMA, link adaptation based on reliability information, receive algorithm convergence, channel coding etc. See [3] for a survey of IDMA researches. Recently, IDMA is applied to OFDM system and OFDM-IDMA structure is studied [4, 5]. OFDM-IDMA inherits most of the merits of OFDM and IDMA. The key advantage of OFDM-IDMA is that MUD can be realized efficiently with complexity per user independent of the channel length and the number of users, which is significantly lower than that of other alternatives. Meanwhile, it can achieve better performance or high throughput than OFDMA systems. In this paper, we propose to use IDM as a modulation method for OFDM-based systems, called layered-modulation. That is, users are multiplexed in OFDMA method, and data flows from one user are modulated in IDM form. Compared with OFDMA system with M-PSK/QAM, we will show that such IDM-OFDMA scheme may well exploit time-frequency diversity and achieve better performance. Moreover, since IDM is performed within user data instead of multiple users, receiver complexity is lower than OFDM-IDMA system. The remainder ofthis paper is organized as follows: In Section II, the system model including transmitter and receiver is presented. In Section III, an enhanced scheme for Layered Modulation is proposed. Numerical results are shown in Section IV, and conclusions are drawn in Section V. II. LAYERED-MODULATION (LM) A. System model The transmitter based on layered-modulation and OFDMA system is shown in Fig.l. For user k, the serial data stream after channel coding is divided into M parallel data streams by SIP unit. Each data stream (or each layer of code bits) is repeated with a common repetition sequence (e.g. + - + - ...). Then a layer-specific interleaver is performed on the repeated code bits (or chips). The interleaver length is the code bit length multiple with repetition factor. Next the interleaved chips are modulated by modulation unit. Finally, each layer of modulated symbols are summed together as the output ofLM, and sent to share OFDM modulation with other users. The iterative receiver structure is similar to that of IDMA receiver, as shown in Figure 2. After performing OFDM demodulation on received signals, a chip-by-chip multi-layer detection (MLD) is used to separate each layer of data. Then the detected sequences are sent to layer-specific de-interleaver followed by de-repetition operation. Before channel decoding, an iterative computation of soft information (called ipternal iteration) can be performed 2 where (8) (9) (7) (6) M = H k LE(x m ,) = E(Yk) - HkE(x m ) m'=l,m';tm Ext(x m ) =IOg(P(X m =+1)J p(x m = -1) M = IH k 1 2 L Var(x m ,)+ (j2 m'=l,m';tm A. Comparison with PSK/QAM In this section, we will simulate the LM-based system performance. Simulation assumptions are listed in Table I.As the baseline, performance of OFDMA with M-PSK/QAM is also given. To achieve the same spectrum efficiency, LM needs to have the number of layers (QPSK is adopted for each layer in LM-based system): b if 1 spectrumefficiency x repetiton factor num er 0 ayers = 2 III. PHASE-SHIFTED LAYERED-MODULATION (PSLM) The LM is the non-orthogonal superposed scheme, the multilayer interference is assumed to be suppressed by the low complexity iterative receiver. If some modifications are used to alleviate the multilayer interference at the transmitter, the performance will be improved at least while the iterative receiver does not work effectively. Assigning unique angle to each layer is one of the ways to alleviate interference. E.g. assume we have two BPSK signals {+1, -I}. The superposition oftwo BPSK signals will produce a three level output signal {+2, 0, -2}. The "+2" level indicates that both the two BPSK signals are "+1" and the "-2" level indicates that both the two BPSK signals are "-I". The "0" level indicates that one BPSK signal is "+1" and the other is "-1" but we cannot distinguish which is which. The distance between these two output signals "(+1) + (-1)" and "(-1) + (+1)" are zero. This will case a performance loss even under a noiseless channel. If we introduce an angle rotation before adding these two signals, then the ambiguous level "0" will not occur (the distance can be larger than zero) and hence improve the performance. IV. NUMERICALRESULTS E(x m ) = (+I)p(x m = +1)+ (-l)p(x m = -1) =exp(Ext(xm ))-1 = tanh(Ext(x )/2) exp(Ext(x m )) +1 m Var(x m ) =1- (E(X m ))2 (10) Detailed descriptions of receive algorithm of other modules such as De-Rep, Re-Rep, DEC etc, were given in [2]. From (5)-(10), we can see that the MLD detection algorithms inherit the low complexity of IDMA, the normalized computational cost is only about 6 additions, 6 multiplications and a tanh(x) function per chip per layer per iteration which is the same to IDMA iteration detection [2]. so = Var(Yk) -IHk1 2 Var(x m ) E(x m ) and Var(x m ) both are determined by the input extrinsic information Ext(x m ), defined by (8), given by re-interleaver module. (5) (4) (3) M c;m =H k LXm' +nk m'=l,m':#:m MLD where is the interference in yk with respect to layer m. From the central limit theorem, can be approximated as a Gaussian variable, and yk can be characterized by a conditional Gaussian probability density function, e.g. BPSK is adopted, xm E {+I,-I}. P(Yk IX m = 1) = 1 ex p [ (Yk - (H k + E(;m )))2] J21rVar(;m) 2Var(;m) where E(*) and Var(*) are the mean and variance functions, respectively. The output Log Likelihood Rate (LLR) of x can be expressed as llr(x ) = logP(Yk IXm= +1) = 2H k Yk -E(c;m) m P(Yk IX m =-1) Var(c;m) b) Receiver for user k Fig. 1. Transmitter and receiver structure based on layered-modulation (LM) where M is the total layer number for k-th user; Hk is the channel coefficient for user-k, xm is the modulated symbol for m-th layer, and {nk} are samples ofa zeros-mean AWGN with variance a2=NO/2.Due to the use ofrandom interleavers, the MLD operation can be carried out in a chip-by-chip manner, (1) can be rewritten as Yk = Hkx m +;m (2) between MLD and de-repetition unit. After internal iteration, the output of de-repetition is sent to decoder for decoding. Decoded information is fed back to MLD for further iteration (called external iteration). Since de-repetition unit has much lower complexity than channel decoding, using the internal iteration can significantly reduced the complexity as in traditional IDMA receiver where only external iteration is utilized. OFDM is adopted to transmit the LM symbols, so the LM symbols can be thought as passing through a single path flat fading channel. For user k, the received symbol after FFT (Fast Fourier Transform) can be written as M (1) Yk =HkLxm +nk m=l TABLE 1SIMULATION PARAMETERS FOR LAYERED-OFDMA SYSTEM 3 TABLE 2PERFORMANCE GAIN OF LM-BASED SYSTEM WITH PSK/QAM The repetition factor (RF) in LM-based system is selected to 4, so with the same spectrum efficiency to QPSK, 8PSK and 16QAM, the corresponding number of layers (nLayer) is 4, 6 and 8, respectively. Fig. 2 shows the LM-based system performance in different block sizes. A RB is defined as a block with 12 subcarriers by 14 OFDM symbols. Ideal channel estimation is applied. The solid lines are OFDMA with PSKlQAM as the baseline while the dash lines present Layered-OFDMA. The marks on the lines are the same while the spectrum efficiency of LM and PSKlQAM is the same. Block-error-rate (BLER) is used as the performance measurement merit. Given BLER threshold as 10%, we investigate the performance gain of LM-based system over OFDMA system with M-PSK/QAM. The gain is defined as: Gain = (Et/NO)OFDMA-(Et/NO)LM-OFDMA @ BLER=10% The performance comparison between LM and PSKlQAM based OFDM in Fig. 2 is summarized in Table 2. As shown in Table 2, with the RB number increasing, the gain grows. In the case of QPSK and I6QAM, Layered-OFDMA shows better performance selectively. For 8PSK, LOFDMA always shows better performance. At 8RB, the gain ofQPSK, 8PSK, I6QAM is I.IdB, 2.2dB and 0.6dB respectively. With the RB size increasing, the interleaver length of each layer increases, that means the data correlation between layers is weaker, the iteration procedure works more effectively which leads to eliminate Multi-Layer interference effectively. That is why the performance gain grows with data block size increasing. B. Angle Rotation influence A special case of LM (S-LM) is studied by setting RF=I, nLayer=2 and phase-shifted adopted. Fig. 3 gives the minimize Euclidean distance by simulation. As shown in Fig.3, the phase-shifted factor is 30 or 60 differential which has the maximum of the minimum Euclidian distance for the I6-point superposed constellations. As shown in Fig. 4, the BLER performance of S-LM in TU30 is studied. Considering the 16QAM as the baseline, the gain of LM at BLER=O.1 is only -O.1dB, while the gain of S-LM is 2 dB. When data block is only 4RB, the interleaver Parameters Transmission Bandwidth Sample frequency Timeslot Duration Subcarrier spacing FFT size Number of occupied subcarriers Short CP Primary resource block (RB) Number of Antennas Channel model UE speed (km/h) Channel Code Rate Channel Code Polynomials Decoder algorithm Modulation Iteration number Values 5 MHz 7.68 MHz 0.5 ms 15kHz 512 301 (4.69us 136 samples) X6, (5.21us 140 samples) XI 12 subcarriersx 14 symbols lxl Ray-6 TU 30 1/3 013015017 Max-log-MAP QPSK 4(inner) 16 (outer) Gain QPSK 8PSK I6QAM IRB -0.1dB 0.2dB -1.2dB 2RB 0.4dB 0.8dB -0.7dB 4RB 0.9dB 1.9dB -O.OdB 8RB l.IdB 2.2dB 0.6dB length is not enough for iterative receiver to eliminate the MLI effectively. S-LM adopts the phase-shifted module to alleviate the MLI at the transmitter, which helps the iterative receiver work effectively at the spectrum efficiency corresponding to I6QAM, if the iterative receiver works effectively, not only the multilayer interference can be eliminated, but also the channel diversity can be achieved. v. CONCLUSION Layered-Modulation based on IDM technique is proposed in this paper. Since code bits are well distributed over time-frequency plane, LM can well exploit time-frequency diversity. Since only one channel coding is applied for the LM, receiver complexity is lower than OFDM-IDMA. Simulation results show that LM may achieve performance gain up to I.IdB over QPSK, 2.2dB over 8PSK, 0.6dB over I6QAM. A trend is that performance gain increases with data block size. Furthermore, a special case of Layered-modulation is studied. With the aid of angle-rotation, LM can improve the gain over I6QAM from 0.6dB to 2dB. At the view of modulation, the symbol superposed by 2 independent QPSK symbols with angle rotation could perhaps be regarded as a special mapping constellation corresponding to I6QAM. REFERENCES [1] R. H. Mahadevappa and J. G. Proakis, "Mitigating multiple access Interference and intersymbol interference in uncoded CDMA systems with chip-level interleaving," Trans. Wireless Commun.,vol. 1, no. 4, pp. 781-792, Oct. 2002. [2] P.Li, "Interleave-division multiple access and chip by chip iterative multi-user detection," IEEE Commun. Magazine, vol. 43, no. 6, pp. S19-523, June 2005. [3] H. Schoeneich and P. A. Hoeher, "Adaptive interleave division multiple access - A potential air interface for 4G bearer services and wireless LANs," in Proc. WOCN 2004, Muscat, Oman, pp. 179-182, June 2004. [4] P.Li, Q. Guo, and J. Tong, "The OFDM-IDMA approach to wireless communication systems," IEEE Wireless Commun. vol. 14, pp. 18-24, June 2007. [5] I. Mahafeno, C. Langlais, and C. Jego, "OFDM-IDMA versus IDMA with ISIC for quasi-static Rayleigh fading multipath channels," in Proc. 4th International Symposium on Turbo Codes & Related Topics, Munich, Germany, April 2006. [6] K. Kusume and G. Bauch, "A Simple Complexity Reduction Strategy for Interleave Division Multiple Access," IEEE Vehicular Technology Conference (VTC2006-fall), Montreal, Canada, September 2006. [8] R. H. Mahadevappa, "Interference Cancellation Techniques for Multiuser Communication Systems," PhDthesis supervised by J. Proakis, Northeeastem University, USA, Nov. 2002. [9] P. Li, L. Liu, K. Y. Wu, and W. K. Leung, "A unified approach to multiuser detection and space-time coding with low complexity and nearly optimal performance," 40th Allerton Conference, pp. 170-179, Oct. 2002. [10] H. Schoeneich and P. Hoeher, "A Hybrid Multiple Access Scheme Approaching Single User Performance," in Proc. Sixth Baiona Workshop on Signal Processing in Communications,Baiona, Spain, September 2003. [11] J. C. Fricke, P. A. Hoeher, and H. Schoeneich, "An uplink proposal based on interleave - division multiple access," in Proc. Wireless World Research Forum 14th Meeting (WWRFI4), San Diego, California, July 2005. 4 LH vs PSK/QAM in OFDMA data block=2RB 10- 3 '-===::::r:====:::::r::==_.L-_---l..__--L__..L-_----.J o 6 8 W EblNO (dB) -1 10 14 12 10 6 8 Eb/NO (dB) --&-QPSl<
-B-16QAH - e - LM - - LM RF=4 -3 - 8 - LM RF=4 10 o o LH vs PSl</QAH in OFDHA data block=lRB 10 , .
: ...70' . :::::: ... :!.: . :.... ::.: . : : ............ ..:- -: -. < .. .""" ". 10- 1 ... ..... ........ r C : :J : :0'(' :. ::: ::TN 0 ...................................... " . W W a) data block = 1 RB b) data block = 2 RBs LM vs PSK/QAM in OFDHA data block=4RB -1 10 -2 10 ........................-.-.................................................... -e-QPSK --+--8PSK --a-16QAM - e - LM - - LM RF=4 -3 - B - LM RF=4 10 o 6 8 Eb/NO (dB) 10 12 LM vs PSK/QAM in OFDMA data block=SRB -1 10 -2 10 _ -e-QPSK --+--SPSK --a-16QAM - e - LM - - LM RF=4 -3 - B - LM RF=4 10 6 8 ro u Eb/NO (dB) c) data block = 4 RBs d) data block = 8 RBs Fig. 2 BLER performance ofLM and PSK/QAM based OFDMA with different number ofRBs TU30 S-LH Perforlllance data block=4RB -2 10 -1 10 o 10 ::::::1:::::::::. . : : . . : .. "" : . :