doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 1 A Novel Soft MIMO Detector for MIMO-OFDM (802.11n) Receivers Notice: This document has been prepared to assist IEEE It is offered as a basis for discussion and is not binding on the contributing individual(s) or organization(s). The material in this document is subject to change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or withdraw material contained herein. Release: The contributor grants a free, irrevocable license to the IEEE to incorporate material contained in this contribution, and any modifications thereof, in the creation of an IEEE Standards publication; to copyright in the IEEE’s name any IEEE Standards publication even though it may include portions of this contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the resulting IEEE Standards publication. The contributor also acknowledges and accepts that this contribution may be made public by IEEE Patent Policy and Procedures: The contributor is familiar with the IEEE 802 Patent Policy and Procedures, including the statement "IEEE standards may include the known use of patent(s), including patent applications, provided the IEEE receives assurance from the patent holder or applicant with respect to patents essential for compliance with both mandatory and optional portions of the standard." Early disclosure to the Working Group of patent information that might be relevant to the standard is essential to reduce the possibility for delays in the development process and increase the likelihood that the draft publication will be approved for publication. Please notify the Chair as early as possible, in written or electronic form, if patented technology (or technology under patent application) might be incorporated into a draft standard being developed within the IEEE Working Group. If you have questions, contact the IEEE Patent Committee Administrator at. Date: Author NameCompanyAddressPhone Behrouz Farhang-BoroujenyUniv of UtahSal Lake City, UT

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 2 Outline  Introduction  Channel model  Soft Information: Log-likelihood ratio, LLR values ■ what is the problem?  Zero-forcing / MMSE / VBLAST detectors ■ computation of LLR values  Our solution to LLR computation  Simulation results  Conclusions

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 3 Introduction  We answer the following question: ■ In a MIMO set-up how one can efficiently obtain soft information, e.g., log-likelihood ratio (LLR) values, of the data bits?

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 4 Introduction  We answer the following question: ■ In a MIMO set-up how one can efficiently obtain soft information, e.g., log- likelihood ratio (LLR) values, of the data bits? The material presented here are protected by a patent application owned by the university of Utah.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 5 Channel Model We consider:  A frequency selective channel.  OFDM is used to convert the frequency selective channel to a number of parallel flat fading channels.  Accordingly, each subcarrier channel has the following model: y = Hd+n where d is a vector of transmit symbols y is a vector of received signal H is the channel gain matrix n is an additive noise vector.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 6 Receiver Structure Channel Decoder MIMO Detector

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 7 Receiver Structure Channel Decoder MIMO Detector We address an implementation of this block

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 8 Receiver Structure Channel Decoder MIMO Detector Feedback from the output of the channel decoder to MIMO detector, allows near capacity performance.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 9 Soft Information: Log-likelihood ratio (LLR) values  Symbol vector d is obtained from a vector b=[b 1 b 2 … b N ] of information bits through some mapping.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 10 Soft Information: Log-likelihood ratio (LLR) values  Symbol vector d is obtained from a vector b=[b 1 b 2 … b N ] of information bits through some mapping.  We wish to calculate where is calculated in the same way.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 11 Soft Information: Log-likelihood ratio (LLR) values  Symbol vector d is obtained from a vector b=[b 1 b 2 … b N ] of information bits through some mapping.  We wish to calculate where is calculated in the same way.  Problem: the number of combinations that b -k takes is 2 N-1 !

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 12 Soft Information: Log-likelihood ratio (LLR) values  Symbol vector d is obtained from a vector b=[b 1 b 2 … b N ] of information bits through some mapping.  We wish to calculate where is calculated in the same way.  Problem: the number of combinations that b -k takes is 2 N-1 ! The key point here is that most of the terms in the numerator and denominator are insignificant. Thus, a handful of the significant terms may be sufficient for accurate estimation of k.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 13 Log-likelihood ratio values: max-log algorithm

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 14 Log-likelihood ratio values: max-log algorithm This incurs an insignificant loss (in the order of a a fraction of 1 dB) in performance.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 15 Zero-forcing / MMSE / VBLAST detectors  Zero-forcing detector: ■ Estimate of d = Q[(H*H) -1 H*y]

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 16 Zero-forcing / MMSE / VBLAST detectors  Zero-forcing detector: ■ Estimate of d = Q[(H*H) -1 H*y]  MMSE detector: ■ Estimate of d = Q[(H*H+  2 I) -1 H*y]

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 17 Zero-forcing / MMSE / VBLAST detectors  Zero-forcing detector: ■ Estimate of d = Q[(H*H) -1 H*y]  MMSE detector: ■ Estimate of d = Q[(H*H+  2 I) -1 H*y]  VBLAST/Successive Interference Canceller (SIC) detector: ■ Detects the strongest symbol first, subtract the detected symbol, and continue with the successive detection and cancellation of the rest of the symbols.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 18 Zero-forcing / MMSE / VBLAST detectors: computation of LLR values  Starting with the detected d, for a chosen bit b k, it is identified that b k belongs to which element of d, say d i.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 19 Zero-forcing / MMSE / VBLAST detectors: computation of LLR values  Starting with the detected d, for a chosen bit b k, it is identified that b k belongs to which element of d, say d i.  All the elements of d, except d i, are kept fixed. The symbol d i is then given all possible values from the symbol constellation, and from all these choices, the maximum values of P(b k =+1|y,d -i ) and P(b k =-1|y,d -i ) are found and substitute in the max-log LLR formula

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 20 Our solution to LLR computation: Markov chain Monte Carlo (MCMC) simulation technique y=Ad+n, Stated3d3 d2d2 d1d1 S0S0 S 1` +1 S2S2 +1 S3S3 +1 S4S4 S5S S6S6 S7S7 +1

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 21 Our solution to LLR computation: Markov chain Monte Carlo (MCMC) simulation technique y=Ad+n, Stated3d3 d2d2 d1d1 S0S0 S 1` +1 S2S2 +1 S3S3 +1 S4S4 S5S S6S6 S7S7 +1 This procedure gives us a set of selections of d that result in small distances |y-Ad (n) |. These may be viewed as important samples of d that correspond to significant terms in the LLR equation or its max-log version.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 22 Our solution to LLR computation: Markov chain Monte Carlo (MCMC) simulation technique y=Ad+n, Stated3d3 d2d2 d1d1 S0S0 S 1` +1 S2S2 +1 S3S3 +1 S4S4 S5S S6S6 S7S7 +1 This procedure gives us a set of selections of d that result in small distances |y- Ad (n) |. These may be viewed as important samples of d that correspond to significant terms in the LLR equation or its max-log version. If implemented in some clever way, the number of samples that is required for estimation of each k is in the order of 10 to 30, even though the size of the state space can be in the order of billions.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 23 How Complex is MCMC?  It turns out that MCMC can be implemented VERY efficiently.  MCMC simulator for a MIMO channel with 4 transmit antenna and 16 QAM symbols has a complexity that is comparable or lower than that of a 16 bit-by-16 bit multiplier.  An implementation of this MCMC simulator on FPGA requires 600 slices.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 24 Simulation Results  MIMO-OFDM ■ FFT size: 64 ■ Cyclic prefix length: 16 ■ Channel is estimated through pilot symbols transmitted at the beginning of each frame ■ Channel: convolutional code with polynomials [133, 171], R = 3/4

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 25 Simulation Results

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 26 Simulation Results

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 27 Simulation Results

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 28 Simulation Results

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 29 Simulation Results

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 30 Simulation Results

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 31 Wireless Communications Lab at ECE Dept of Univ of Utah  We are actively involved in development of MIMO detection techniques  In collaboration with L-3 Communication West in Salt Lake City, we have developed a MIMO testbed with 4 transmit and 4 receive antennae  A new version of our testbed that facilitates our research on MIMO detectors is under development.  We are open and seeking collaboration with industry. In particular, we are looking forward to any collaboration with IEEE n consortia.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 32 Conclusions  The problem of soft estimation of information bits in a MIMO setup was addressed.  Using Markov chain Monte Carlo simulation technique, in the Wireless Communications lab of UofU, we have developed a very efficient detector for this task.  The proposed method could be used along with any conventional detector (ZF/MMSE/VBLAST- SIC) to improve its performance.  Gains in the order of 6 dB or more have been observed.  The proposed method is an excellent choice in systems that employ advanced channel coding, i.e., turbo and LDPC codes.  The proposed technology is extremely hardware friendly. The complexity of the MCMC simulator is not greater than a 16 bit-by-16 bit multiplier. Therefore, in a MIMO-OFDM where many subcarrier channels have to be examined in parallel, a number of MCMC simulators can be run in parallel at a minimum cost.

doc.: IEEE /0790r1 Submission July 2005 Behrouz Farhang-Boroujeny, Univ of UtahSlide 33 Publications [1] B. Farhang-Boroujeny, H. Zhu, and Z. Shi, “Markov chain Monte Carlo algorithms for CDMA and MIMO communication systems,” IEEE Trans. Signal Processing, Accepted for publication. [2] H. Zhu, B. Farhang-Boroujeny, and R-R. Chen, “On performance of sphere decoding and Markov chain Monte Carlo detection methods,” IEEE Signal Processing Letters, Accepted for publication. [3] R-R. Chen, B. Farhang-Boroujeny and A. Ashikhmin, “Capacity-approaching LDPC codes based on Markov chain Monte Carlo MIMO detection,” Submitted to IEEE Communications Letters, March [4] H. Zhu, Z. Shi, and B. Farhang-Boroujeny, “MIMO detection using Markov chain Monte Carlo techniques for near- capacity performance,” Int. Conf. Acoustics, Speech and Signal Processing, ICASSP’05, Philadelphia, March 18 – 23, [5] Z. Shi, Haidong Zhu, and B. Farhang-Boroujeny, Markov chain Monte Carlo techniques in iterative detectors: a novel approach based on Monte Carlo integration, IEEE Global Telecommunications Conference, GLOBECOM'04., vol. 2, 29 Nov.-3 Dec., 2004, pp. 325 – 329. [6] H. Zhu, B. Farhang-Boroujeny, and R-R. Chen, “On performance of sphere decoding and Markov chain Monte Carlo detection methods,” SPAWC 2005, the sixth IEEE International Workshop on Signal Processing Advances for Wireless Communications, June 5-8, 2005, Invited. [7] R-R. Chen, B. Farhang-Boroujeny and A. Ashikhmin, “Capacity-approaching LDPC codes based on Markov chain Monte Carlo MIMO detection,” SPAWC 2005, the sixth IEEE International Workshop on Signal Processing Advances for Wireless Communications, June 5-8, 2005.