This action might not be possible to undo. Are you sure you want to continue?

**Computer Engineering Michigan Technological University Houghton, MI 49931, USA
**

ABSTRACT In this paper, it is demonstrated that decision feedback equalization (DFE) based on soft-decisions is able to outperform hard-decision DFE, with an attractive computational cost that is polynomial in the number of symbols to be recovered. A group of soft-decision feedback equalizers (SDFE) with probabilistic data association are developed for multiinput multi-output (MIMO) wireless communication systems operating in time dispersive channels. Block transmissions with zero-padded precoding are employed to guarantee channel identiﬁablity. Computer simulations indicate that SDFE offers close-to-optimal error probability performance without incurring exponential complexity. Further improvements on these SDFE schemes promise enhanced performance at a reduced computational cost. Keywords: soft decision feedback equalization, MIMO systems, probabilistic data association. 1. INTRODUCTION Enormous increase in bandwidth efﬁciency is promised by the use of MIMO systems in wireless radio frequency links [1]. In the case of frequency selective channels, orthogonal frequency division multiplexing (OFDM) can be deployed to transmit signals over frequency ﬂat subcarriers. Alternatively, channel-induced inter-symbol interference (ISI) can be mitigated using high-speed equalization techniques, among which decision feedback equalization (DFE) has been proved very effective at low computational cost. Equalization in MIMO systems is more challenging, however, because of the need for signal detection in the presence of both multiple access interference (MAI) as well as ISI. MIMO DFEs have been derived in [2], [3], [4]. In these schemes, tentative decisions on past symbols or symbols from MAI sources are made by quantizing properly derived decision statistics. Such a hard-decision based approach may suffer from catastrophic error propagation, and in most cases, incurs a nontrivial performance gap relative to an optimal maximum likelihood (ML) detector in terms of bit error rate (BER) performance. In this paper, we take on a soft decision approach to narrow the gap between DFE and ML detection for MIMO systems operating in frequency selective channels. We are inspired by the low-complexity soft-decision CDMA detector introduced in [5]. It uses a multistage detection structure, but replaces the intermediate ﬁnite alphabet symbol decisions by soft decisions, expressed via their associated probabilities. Such a soft-decision structure leads to signiﬁcant computation reduction when MAI is approximated to obey a single-mode Gaussian probability distribution, an idea originated from the probability data association (PDA) ﬁlter [6]. The PDA detector provides near-optimal performance, with overall complexity Ç´Ã ¿ µ, where Ã is the number of users. We adopt this soft-decision approach and develop soft decision feedback equalization (SDFE) schemes for FIR-MIMO channels in the quest for low-complexity equalizers with BER performance competitive to that of ML detection. The paper is organized as follows. In Section 2, soft decision feedback equalization using PDA ﬁltering is derived for a single-input single-output (SISO) frequency selective channel. The development of the SDFE scheme for a SISO system (SISO-SDFE) not only provides an essential component for building MIMO-SDFE, but also illustrates a key feature of SDFE: the importance of block transmissions with zero-padded (ZP) precoding. PDA ﬁltering requires a decorrelating preprocessor, which is infeasible for rank-deﬁcient channels. With ZP precoding, however, the equivalent channels are guaranteed to be invertible [7], making PDA-based SDFE applicable to all channel conditions. In Section 3, the soft-decision equalization approach is applied to frequency-selective MIMO channels. Depending on the desirable tradeoff between performance and complexity, three soft-decision based algorithms are developed for a general FIR MIMO model: (1) joint soft-decision equalization and multiuser detection; (2) soft-decision feedback equalization with hard-decision nulling-canceling multiuser detection; (3) hard-decision feedback equalization with softdecision PDA multiuser detection. Computer simulations are performed in Section 4 to compare the BER performance of the proposed MIMO-SDFE schemes with that of ML detection, and a summary is given in Section 5.

0-7803-7551-3/02/$17.00 ©2002 IEEE

594

and is associated ¡ È Ö´ with a probability ½ Ýµ. and noisy digital communication SISO system. let × ¡ ÁÌ ¼Ì Ì matrix È Æ ´Ã Æ µ¢Æ is constructed by choosing Ã Æ · Ä. dispersive. In a soft decision approach. This model considers each user at each transmitter as one input. we group the input time-serial ´Òµ into blocks of size Æ and deﬁne the th ¡ ´ Æ µ ¡ ¡ ¡ ´ Æ · Æ ½µ Ì . The input-output block Ý model of the block transmission system with ZP precoding is given by [7] Ý À ·Ú (1) where À ¾ ÊÃ ¢Æ is a tall Toeplitz matrix made from . To avoid this. update and Ê from (4) and (5). (iii) combination of both (i) and (ii). is the th column of the Æ ¢ Æ identity matrix Á Æ . The overall complexity of SISO-SDFE is on the order of Ç´Æ ¿ µ. we ﬁrst perform a zero-forcing (ZF) operation on the received block sequence Ý to obtain: number of elements in . transmitted block to be To eliminate the inter-block interference. as suggested in a PDA ﬁlter [5]. with Ä being the channel memory length. the ﬁnite alphabet symbol is treated as a binary random variable. Direct evaluation of from (3) incurs a computational cost exponential in the 595 . a binary transmission is assumed. which is guaranteed to be invertible due to the ZP precoding. (8) 3. ¾ ¦½ Æ . Using the conditional means ¡ Ý ½ ¡ · ´ ½µ ¡ ´½ µ ¾ ½. at the expense of transmission redundancy. For update based on (7) as follows: ½ ½ ¨ © ½ · ÜÔ ¾ Ì Ê ½ converge. The vector × is a block of Ã symbols to be transmitted over the channel. MIMO communications arise in three situations: (i) multiple transmit and receive antennas with a single user. We will adopt a general MIMO system model that describes all three cases. where a guard-inserting data blocks. ISI only exists inside each transmitted block. ½ Æ . ¡ ´¼µ ¡ ¡ ¡ ´Äµ as the symbol-spaced tap Denote coefﬁcients in the channel response. Make decisions on the block of symbols by Ý À· Ý · À· Ú ½ (2) where À· denotes the Moore-Penrose psuedo-inverse of the channel matrix À. For vector Ú simplicity. [6].2. For block processing. To do so. we have the following equalities ¡ Ê Ý ÓÚ ¨ Ý Ý ´¾ ½µ (4) ¡ ´½ µ Ì ¡ · ÊÚÚ (5) È Ö´ Ýµ ½ ÜÔ ½´ ¾ È Ö´ È Ö´ ´¾ µÆ ¾ Ê ¨ Ýµ ÜÔ ¾ Ýµ µÌ Ê ½ ´ ½¾ Ì © µ © (6) (7) Ê ½ accordingly. and 2. To construct a soft-decision feedback equalizer (SDFE). we write Ý as ½ ¼ ¼ ½ ¡¡¡ Æ (9) Reﬁnements discussed in [5] can be used to reduce the computational cost of the above procedure. This enables the use of block-by-block processing without loss of optimality. and does not propagate to adjacent blocks. We will henceforth drop the block index for notational simplicity. This apDecision on can be made from proach is summarized into the following procedure for softdecision DFE: SISO-SDFE: 1. MIMO SDFE We now consider a frequency-selective faded MIMO link with Æ inputs and ÆÓ outputs. we assume that conditioned on Ý is approximated as a uni-modal Gaussian random vector with matched mean and covariance. BLOCK SISO SDFE It has been argued that block equalization offers important advantages over serial equalization in terms of guaranteed channel identiﬁability and less performance degradation from noise enhancement or error propagation [7]. while each output refers to the signal received at each receive Ý Æ · ½ ·Ú (3) where is the th element of . Initialize the probabilities as ¼ ½ ¡¡¡ Æ. Repeat (2) if necessary until all 4. In this model. 3. and the covariance of the noise À· Ú becomes ÊÚÚ À· ÊÚÚ ´À· µÌ . We take the block equalization structure and start with a linear. (ii) multiple receive antennas and a single transmit antenna with multiple-access users. resulting in a received data ¡ Ý´ Ã µ ¡ ¡ ¡ Ý´ Ã·Ã ½µ Ì . Similar to [5]. a zero-padding precoder can be used to insert guarding bits between successive ¡ È . We assume that the noise vector Ú has zero mean and temporal covariance Ê ÚÚ .

. . . Initialization for NC-MUD: ½ À· ´¼µ ½ 596 . . . The non-diagonal matrix inside each block. and treats both using soft-decision feedback. À´¼µ. . ¾ where À´Ðµ ¡ ´½ ½µ ´Ðµ ¡ ¡ ¡ . where Ä is the maximum length of all the ÆÓ Æ channel impulse responses. While the soft decision approach promises near-optimal BER performance without incurring intractable complexity. Ð ¼ ½ ¡ ¡ ¡ Ä. We propose to use the nullingcanceling technique for multiuser detection over equalized ﬂat fading channels. we ﬁrst diagonalize the MIMO channel matrix À into a block-diagonal form. The number of elements in the symbol block is now Æ Æ . MIMO systems present greater challenges because of the existence of capacity limiting multiple access interference and the signiﬁcant increase in computational burden. This can be achieved by premultiplying À with a row trun¡ Á cating matrix Ì¼ Æ ÆÓ ¼Æ ÆÓ ¢´Ã Æ µÆÓ . À´Äµ ¼ ¿ (10) ¡¡¡ . Ì ¡ Á ÆÓ ¼ . namely nulling-canceling. . . . and Æ is enlarged to Æ Æ . we make no distinction among the three terms inputs. .2. The transformed channel matrix À is block diagonal.. with each element denoted by ´ µ ´Ðµ. ¿ Á´Æ ½µÆÓ ª À´¼µ ßÞ ¡À · ÌÚ (12) The rest of the algorithm is described as follows: NC-MUD with SDFE for FIR-MIMO channels: 1. À´Äµ À´¼µ (11) ´ ½µ ´Ðµ Æ ¿ ´½ µ ´Ðµ ¡ ¡ ¡ ¡¡¡ ´ . . which is typically on the order of Ç´Æ ¾ µ in the SISO case. To balance the tradeoffs between performance and complexity for various MIMO system setups. Such an approach considers multiuser detection and channel equalization simultaneously.. Accordingly. we develop three symbol detection and equalization schemes that blend the soft-decision and hard-decision methods in different ways. the hard-decision approach is attractive for its computational simplicity. . ÆÓ Æ µ ´Ð µ Efﬁcient soft-decision feedback equalization has been developed for a dispersive SISO system.e. Ý ´ µ and ¡ Ý´½µ ´ µ ¡ ¡ ¡ Ý´ÆÓµ ´ µ Ì ¾ ´ µ are augmented into Ý´ µ ÊÆÓ ¢½ and ´ µ ¡ ´½µ´ µ ¡ ¡ ¡ ´Æ µ ´ µ Ì ¾ ÊÆ ¢½ . transmit antennas. Similar to Section 2. and¡the computational complexity is on the order of Ç ´Æ Æ µ¿ accordingly. In such applications. Joint soft-decision equalization and detection The SISO-SDFE algorithm in Section 2 can be extended to the MIMO model in (10) in a straightforward manner. a MIMO system with a wireless spectral efﬁciency higher than 20 bps/Hz is demonstrated and ÆÓ ½¾. the number of transmit and receive antennas Æ and ÆÓ could be large. In analogy to (1). Ý ¡ ÌÆ ½ ¡ ¡ ¡ Ì½ Ì¼ Ý ßÞ ¡Ì ¼ . where the operator ª denotes the Kronecker product. Such a block diagonalization process is equivalent to the decorrelating preprocessor in SISO-SDFE. . that is. the number of inputs is determined by the number of transmit antennas and the number of multiple access users as well. ÆÓ ¼ . 3. For convenience. 3. .1. To make the MIMO system model amenable to singletransmitter soft-decision equalization. In [8].antenna. . i. SDFE with nulling-canceling MUD When multiple antennas are employed for high data-rate transmissions.. complexusing Æ ity is heavily dependent on the multiuser detection schemes used. coupled with SDFE within each user. respectively. The channel response matrix between the symbol block at the th input and the th output takes the same form as the À matrix in (1). and users. except that (1) is replaced by (10). À´ÄµÀ·´¼µ À´½µÀ·´¼µ ÁÆ ÁÆÓ ¼ . The overall receiver complexity of such a MIMO system is ¡ ¾¡ reduced to the maximum of Ç Æ ¿ Æ ¾ and Ç Æ ¿ ÆÓ . Each element in (1) is then augmented into a block element. the baseband equivalent MIMO signal model for block processing can be expressed in a matrix-vector form as follows Joint SDFE/SDFD for FIR-MIMO channels: The same as SISO-SDFE. a zero-padding precoder is employed at each input. indicates the existence of nonorthogonal multi-access interference to be suppressed by a MUD technique. . The nulling-canceling (NC) algorithm is an attractive low-complexity method that has been widely deployed for the BLAST systems [8]. . followed by a series of Æ ÆÓ ¢ Æ ÆÓ matrices Ì in the form of ¾ Ý À ¡ À ·Ú ¾ À´¼µ .

´Àµ denotes the th column of À. SDFD-MUD with ZF-DFE for FIR-MIMO channels: For Ò ¾ ½ ¡¡¡ ½ ÑÒ ´ µ ¾ (13) Û Ï Ö ÁÆ ª Û Ï Ý ¾ ÊÆ ¢½ ´ µ (14) (15) (16) (c) SISO-SDFE: note that Ö contains signal components from the th transmitter only. and the optimal ML detector implemented using the sphere decoding algorithm [9]. ´ µ is the th row of . and À´ µ denotes the matrix obtained by zeroing columns ½ ¡ ¡ ¡ of À. Some observations on the NC-MUD/SDFE algorithm are in order: Apply Steps 1-4 of SISO-SDFE to (21) to obtain the ´½µ´Òµ ¡ ¡ ¡ ´Æ µ ´Òµ Ì . For such cases. The over ¡ all complexity is now Ç Æ ¾ Æ ¿ . we may combine soft-decision feedback detection (SDFD) with the conventional hard-decision feedback equalizers. It is assumed to ½¾ symbols. 597 . NC-MUD blends with SISO-SDFE nicely. The BER performance versus signalto-noise ratio (SNR) of these receivers is depicted in Figure 1. The hard-decision based DFE and ZF receivers are computationally simple (Ç´Æ ¾ µ). the vector Ö happens to possess the “decorrelated structure” (17) required by the ensuing SISO-SDFE processor. but suffer from approximately 2 to 4 dB performance losses. It takes advantage of the block Toeplitz structure of the MIMO channel matrix in (11). Due to the preprocessing step (12). 1. SISO-SDFE is compared against the conventional zero-forcing hard-decision feedback equalizer (DFE). but becomes more pronounced for high SNRs. the computational task on multiuser detection becomes less stringent than block equalization. The following gives a hybrid solution using hard-decision ZF-DFE and the soft-decision PDA [5] method for MUD. 4.3. we consider a single-user FIR fading channel with a symbol-spaced memory length of Ä . the structure of the noise covariance matrix is changed and the ordering algorithm in Step 2(a) is no longer optimal. Recursion: (a) User ordering: arg (b) Nulling: 3. After the zero-forcing nulling in Step 2(b). (17) to ´ µ. It is shown that the performance of SISO-SDFE is close to that of sphere decoded ML. starting with a preprocessor: Ý´Òµ ¡ À· ´¼µÝ´Òµ ´Òµ · Ú´Òµ (21) As in [8]. Soft-decision PDA multiuser detection of symbols transmitted at the Òth time. Each FIR tap coefﬁcient is Rayleigh faded. ½ ¡ ¡ ¡ Æ. but is remain constant over each block of Æ allowed to change from block to block.2. Ý´Òµ À´¼µ ´Òµ·Ú´Òµ. SDFD with hard-decision DFE In MIMO system setups where the block size Æ is signiﬁcantly greater than the number of transmit and receive antennas. zero-forcing linear ﬁlter (ZF-only). Ö where ÊÚÚ Ð ´ µ· Ð Æ ½ ´ µ·Ú Ð (17) ½ Æ Ï ÌÊÚÚ ÌÀ ÏÌ . Ý ·½ Ý Æ ·½ À· µ ´¼µ ´ ·½ ½ ´ µ ´Àµ ´ ½µ · Æ (18) (19) 2. Hard-decision feedback equalization of previously detected symbols of all users: Apply Steps 1-4 of SISO-SDFE to Eqn. obtain (d) Cancellation: Ý´Òµ ¡ Ý´Òµ Ò ½ Ñ ½ À´Ñ Òµ ´Ñµ (20) With correct past decisions. Reﬁnements through noise whitening or other optimized user ordering criteria must be devised in order to improve the detection performance. Æ ¢½ symbol vector ´Òµ End Ò. The performance gap is negligible in the low SNR region. Both zero-forcing and MMSE based hard-decision feedback equalization can be used. SIMULATIONS In the ﬁrst example. ½ ¡ ¡ ¡ Æ denotes a permutation of the integers ½ ¡ ¡ ¡ Æ specifying the order in which components of the transmitted symbol vector ´ µ are to be processed. ¯ ¯ Structurally.

” IEEE Transactions on Communications. 9. Sept. 361-363. Hasegawa. [3] N. 50. 10. “On the limits of wireless communications in a fading environment when using multiple antennas. G. The second example studies multiuser detection and channel equalization in a frequency selective MIMO channel with Æ ¾ transmitters and ÆÓ receivers. Further performance improvement can be expected with prudent reﬁnements over the current work. W. pp. R. P. J. [2] A. 1. Signal Processing. FIR-MIMO fading channel: Æ Æ ½¾. Golden. pp. Finkce and M. 1992. Pattipati. “The ﬁnite-length multi-input multi-output MMSE-DFE. [6] Y.” IEEE J. “Equalizers for multiple-input multipleoutput channels and PAM systems with cyclostationary input sequences. X. Papadias. [7] Z.” Electronics Letters. Artech House. Luo. C. R. no. G. Bar-Shalom. D. 2001. “Layered space-time receivers for frequency-selective wireless channels. Their BER performance is close to but still away from that of the optimal ML algorithm. The BER curves of the three softdecision related algorithms developed in Section 3 are plotted in Figure 2 with reference to that of the sphere decoded ML. [4] A. [8] G. 1985. Estimation and Tracking: Principles.” Mathematics of Computation. 10.” IEEE Signal Processing Magazine. [1] G. F. and Software. “Nearoptimal multiuser detection in synchronous CDMA using probabilistic data association. 65-73.Ä . H. Li. Wang. but they are much more tractable in terms of the computational load. Jan. 2002. Wolniansky. 1993. pp. [9] U. M. J. Selected Area in Communications. no. Sayed. the joint SDFE/SDFD algorithm offers closer to optimal BER performance. The mem. pp. Lozano. 29-48. pp. the tap coefﬁcients of each link are generated independently from Rayleigh distributions. Foschini.” Wireless Personal Comm. R. 598 . vol. ory lengths of these 8 channels are assumed to be Ä Over every block of Æ ½¾ transmitted symbols. A.. “Wireless multicarrier communications .10 0 10 0 10 −1 10 −1 BER 10 −2 BER 10 −2 10 −3 soft−decision DFE hard−decision DFE ZF−only Sphere decoding 10 −3 Joint SDFD / SDFE NC−MUD / SDFE SDFD / ZF−DFE Sphere decoded ML 10 −4 0 5 10 SNR (dB) 15 20 10 −4 −8 −6 −4 −2 0 2 SNR (dB) 4 6 8 10 Fig.vol. 2921-2936. pp. 463–471. vol. pp. 44. vol. 630-639. Gans. 5. A. 1. Vol. 5. 48.” IEEE Trans. Duel-Hallen. These receivers balance between the performance advantage offered by the soft-decision approach and the low computational cost of the conventional hard-decision feedback approach. Techniques. 2. vol. Foschini. Apr. Giannakis. K. at the expense of higher complexity. REFERENCES ¾. no. “Improved methods for calculating vectors of short length in lattice. 2000. high-performance receivers for frequency selective MIMO channels. The suboptimal NC-MUD/SDFE and SDFD/ZF-DFE algorithms exhibit performance gaps of about 2 dB from the ML solution. Al-Dhahir. May 2000. including a complexity analysis. FIR-SISO fading channel: Ä Æ ½¾. SUMMARY The soft-decision approach with probabilistic data association is utilized to construct three types of low-complexity. 6. Fig. Pohst. pp. with overall ¡ complexity no greater than Ç ´Æ ¢ Ñ Ü´Æ ÆÓ µµ¿ . 1998. 1999. 1. Valenzuela. “Detection Algorithm and Initial Laboratory Results using the V-BLAST Space-Time Communication Architecture. J. B. 315-335. Comparably. 35. P. 6.” IEEE Communications Letters. 14-15. Æ Ó .where Fourier meets Shannon. K. No. Willett. no. [5] J.

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd