fbpx

Lattice Reduction Aided- Zero Forcing MIMO Detection using two Stage QR-Decomposition


Call for Papers Engineering Journal, May 2019

Download Full-Text PDF Cite this Publication

Text Only Version

Lattice Reduction Aided- Zero Forcing MIMO Detection using two Stage QR-Decomposition

Syed Moinuddin Bokhari. B1 and Bhagyaveni M. A2 1,2Department of Electronics and Communication Engineering College of Engineering, Guindy Campus, Anna University, Chennai, INDIA 600 025

Abstract Lattice Reduction Aided (LRA) MIMO Detection algorithms has been widely used in Multiple Input Multiple Output (MIMO) system for achieving better Bit Error Rate (BER) performance. QR- Decomposition (QRD) is a method applied to solve the performance-complexity trade- off issue of the channel matrix in mostly all the MIMO detection algorithms. In this paper, with the proposed two stage QRD method based on Recursive Block QRD (RBQRD), the performance enhancement of the LRA-Zero Forcing (ZF) MIMO detection algorithm in both the AWGN (Additive White Gaussian Noise) and Rician channels, has been investigated for 16- Quadrature Amplitude Modulation (QAM), comparing it with the traditional QRD algorithms.

Index Terms MIMO, Maximum likelihood detection , QAM, AWGN channels, Rician channels

  1. INTRODUCTION

    MIMO systems has been deployed in a variety of wireless communication fields such as 4G (LTE, LTE-A), Next Generation WLAN networks (IEEE 802.11n/ac), etc. because it has high spectral efficiency. Recently Large MIMO and Massive MIMO systems have gained importance in the existing MIMO technologies. Due to the high density of networks, detection of MIMO systems face serious issues. The trade-off between performance and complexity still exist in the present day networks. Lattice reduction is a method applied in MIMO systems which was originally introduced by Lenstra Lenstra Lovasz (LLL) in 1982. It is a method for factorizing polynomials with rational coefficients [1]. LLL algorithm is well explained in [2]. Complexity and diagonal variants of LLL algorithm was also proposed and the performance improvement was made [2, 3]. Various other reduction techniques such as Hermite-Korkine-Zolotareff (HKZ) and Minkowski reduction algorithms are also followed for Lattice reduction [4]. In MIMO the channel is usually faced with the least squares problem before Lattice reduction. Normal equations, QRD and Singular Value Decomposition (SVD) are the methods adopted to overcome least squares problem [5]. QRD methods has been followed to avoid least squares problem in a full-rank matrix. Most of the techniques used in the literature are based on Gram-Schmidt (GS), Modified GS (MGS), Householder QRD (House QRD) and Givens Rotation-QRD (GR-QRD). In this paper, LRA MIMO detection based on different two stage QRD is performed. The first stage considers MGS, House QRD, GR-QRD, RBQRD before lattice reduction and the second

    stage takes into account GR-QRD inside LLL algorithm of lattice reduction. Their performance are compared in AWGN and indoor Rician channel models, Task Group ac (TGac -channels B, C and D) using 16-QAM.

    The remainder of the paper is organized as follows. In Section

    II the system model for MIMO in AWGN and indoor channel is introduced. Section III gives a review of several QR decomposition algorithms. Simulation results are given in Section IV. Conclusion will be drawn in Section V.

  2. SYSTEM MODEL

    In this section system model for two separate channel under flat fading and indoor fading is considered for the performance evaluation of LRA MIMO detection.

    1. System with Channel Model under flat fading

      In this model AWGN channel is considered in which a symbol synchronized MIMO system with M transmit and N receive antennas is considered. This means M independent signals are multiplexed at the transmitter end and demultiplexed at N receivers, this, is known as spatial multiplexing. If we group all signals into vectors then the system can be viewed as transmitting an M×1 vectors through an N×M channel matrix H. Also, each receiver will have its own noise source (assumed Gaussian). Thus, the overall baseband system model can be mathematically represented as

      = + (1)

      where is N×1 received vector and is N×1 noise vector. The (i,j )th element, hij , of the matrix denotes the complex channel response from jth transmit antenna to ith receive antenna, is zero mean gaussian vector with a covariance matrix of = {} = 2 . The objective of the MIMO detection algorithm is to compute an estimate such that

      = arg min 2

      C

      (2)

      where is the set of complex entries from the QAM constellation and C is the size of QAM constellation.

    2. System with Channel Model under Indoor fading This model considers a MIMO indoor channel. For this the traditional kronecker MIMO model which deals with the

      indoor Rician fading is taken into account. The kronecker MIMO channel is given as [6],

      = 1/2 1/2 (3)

      where the elements of are independent and identically distributed as circular symmetric complex Gaussian with

      triangular matrix. Using Equation (7), the received signal can be modified as,

      = [1, 2, . . ] =

      = +

      = + (9)

      zero-mean and unit variance. and are the transmitter and receiver correlation matrices given by the kronecker matrix

      1,1 1,1

      = [

      0 0

      1,

      ,

      1

      ] [

      ] [

      1

      1

      1

      1

      ] + [ ]

      =

      1/2

      1/2 (4)

      (10)

      The main idea of QRD is not only to keep the channel

      For a 2 × 2 channel matrix the matrices , and is given by

      1 1 1 1

      1 2 2

      orthogonal but also can simplify the procedure of signal processing. The basic algorithms used in QRD are GS, MGS, Householder transformation and GR. We also use

      = [

      = [

      1

      1

      1

      2

      2

      1 2

      ] (5)

      some of the second variant of the existing algorithms,

      RBQRD and compare the performance of LRA-ZF MIMO

      1

      detection both in AWGN and indoor channel models

      1 2 2

      = [ 1 ], = [ 1 ] (6)

      (channel B, C and D).

      1

      1

      1. QRD algorithms

        where the elements r,t and s represents receive, transmit and cross correlation respectively. The estimate is now computed using the kronecker model of the Equation (3). So using Equation (2) the estimate is given as

        C

        C

        = arg min 2

        (7)

        Using the kronecker channel models, the indoor channel models for IEEE 802.11n/ac were obtained under the TGac amendment [7]. Channels (B, C and D) were defined [8] for different environment using Rician fading and are shown in Table I.

  3. QR DECOMPOSITION

    Nowadays most of the MIMO detection methods are combined with QRD [9]. In this paper also we apply various algorithms of QRD to SD.

    TABLE I. TGAC CHANNELS

    Parameters

    B

    C

    D

    Avg. 1st wall dist.(m)

    5

    5

    10

    RMS delay spread (ns)

    15

    30

    50

    Maximum delay (ns)

    80

    200

    390

    Number of taps

    9

    14

    18

    Number of clusters

    2

    2

    3

    K (dB) LOS/NLOS

    0/-

    0-

    3/-

    QRD is a technique applied to computationally solve the channel matrix rather than the traditional methods, which has huge errors if the matrix involved is the ill conditioned. The standard solution is to make the QR factorization or QRD.

    There are several QRD algorithms available for channel matrix factorization. Algorithms which are fast and numerically stable perform better compared to other algorithms. Table II. gives some comparison of QRD algorithms in terms of improvements, drawbacks and complexity.

    Taking into consideration the large deployment of MIMO networks and sacrificing the complexity aspects we would consider analyzing the performance of our proposed two stage QRD namely RBQRD with GR-QRD in a MIMO scenario (2×2) with 16-QAM constellation, comparing it with the other traditional QRD algorithms. Hence some insights of RBQRD are discussed in the next section.

    TABLE II. IMPROVEMENT AND DRAWBACKS OF VARIOUS QRD

    ALGORITHMS

    Algorithm

    Improvement and drawbacks

    Complexity

    Modified Gram Schmidt [10]

    Performs well but not numerically stable

    22

    House Holder transformation [10]

    Requires fewer operations than GS and more stable than GS

    3/3

    Numerical properties are closer

    Givens Rotation [11]

    to Householder but not efficient than Householder

    32 3

    transformation

    Numerically stable for large

    block size and more number of

    Recursive Block QRD [12]

    streams especially in dense networks. Additional complexity is introduced when

    2 + 23 + 19

    the number of antennas/streams

    increases

    = (8)

    where is the orthogonal matrix and is the upper

      1. Recursive Block QRD (RBQRD) algorithm

    RBQRD is the most flexible approach of QRD which uses recursion method of decomposing the block of

    channels 1 , 2 especially in heterogeneous networks [12]. According to the block QRD, if × the thin QR factorization of the channel matrix H can be given as

    = ( ( [ ]1 )) (15)

    [ 0

    [ 0

    [1| 2] = [1| 2] 11 12] (11)

    22

    From the Equation (10), 1 11 = 1, 12 =

    \

    \

    1

    1

    2, 222 = 2 112 the Algorithm 1 for the RBQRD is obtained.

    Algorithm 1: RBQRD algorithm

    Initialization:

    • Initialize the blocking parameter

    • Get the size of and obtain n Thin QR decomposition:

      if <

      • Perform thin factorization

        else

        Recursive Block QRD:

      • Split in half

        1 = /2, round off 1

      • Find of left half of the [1, 11] = ((: ,1: 1), )

      • Find of the modified right half of the

        1

        1

        12 = (: , 1 + 1: )

        (: , 1 + 1: ) = (: , 1 + 1: ) 1 12 [2], 22] = ((: , 1 + 1: ), )

    • Synthesize from the results of 1 2

    = [1 2]

    = [11 12; ( 1, 1) 22]

    end

    Using the Algorithm 1 we perform RBQRD in first stage before LRA-ZF MIMO detection and compare it with other QRD algorithms both in AWGN and TGac channels.

  4. PROPOSED LRA – ZF MIMO DETECTION

    A brief overview of lattice reduction and the algorithms used in LRA MIMO detectors are presented in this section. Consider a lattice as

    {: , 1 2} (12)

    In Equation (15), (. ) is the nearest neighbor quantization operation.

    Fig .1. LRA-ZF MIMO detector with two stage-QRD

  5. SIMULATION RESULTS

    A 16-QAM MIMO (2×2) is considered in our simulations. MATLAB was considered for the simulation purposes. The simulation parameters used are shown in Table III. The BER performance of LRA-ZF MIMO detection using MGS, GR and RBQRD in AWGN and TGac channels (B,C and D) are shown in Fig 2, 3 and 4. Clearly from the figures the BER performance for LRA-ZF MIMO detection is almost the same at low and medium SNR's. At high SNR (> 25dB), RBQRD has the best performance compared to other two algorithms. We can also

    TABLE III. SIMULATION PARAMETERS

    Parameters

    Value

    Number of symbols

    1000

    Number of antennas (TX ×RX)

    2×2

    Number of users

    1

    Bandwidth

    20 MHz

    Channel Type

    AWGN, TGac (B C and D)

    Mapper schemes

    16-QAM

    MIMO detection

    LRA-ZF

    Blocking parameter for RBQRD

    3

    and

    {: = , , 1 2} (13)

    In Equation (12) and (13) is the transmit lattice, is the received lattice with noise and is the number of transmit antennas. The channel matrix can be reduced in LRA MIMO detectors with the well known LLL algorithms [13]. Various LLL algorithms such as original LLL, complex LLL and MLLL had been used for LRA MIMO detection.

    We used an original LLL for the proposed LRA-ZF MIMO detection in this paper. In our proposed design shown in Fig 1. for a larger block size the channel is factorized using RBQRD and then fed to the lattice reduction phase which uses GR-QRD. From the algorithm, we obtain = , where is the uni-modular matrix. The system model of Equation (1) now changes to

    = + (14)

    where = 1. the estimate of obtained using LRA- ZF MIMO detection denoted as is given as

    Fig .2. Performance of LRA-ZF MIMO detection for a 16-QAM MIMO (2×2) using MGS in first stage for an AWGN and TGac channels

    Fig 3: Performance of LRA-ZF MIMO detection for a 16-QAM MIMO (2×2) using GR-QRD in first stage for an AWGN and TGac channels

    infer that LRA- ZF MIMO detection performs best in AWGN mode when compared to TGac channel. Among the TGac channels, performance of LRA-ZF MIMO detection using QRD is better in channel D rather than channel B and C. Hence using various other QRD algorithms the BER performance of LRA-ZF MIMO detection is compared for the same 16-QAM MIMO (2×2) scenario in AWGN and channel D (Fig. 5 and 6). Due to indoor fading of channel D, LRA-ZF MIMO detection performed well in AWGN channel. We can also view that the other QRD algorithms except the Householder QRD method (sub-optimal), performed even better than the optimal ML detection approach in medium and high SNR regime.

    Fig 6: Performance of LRA-ZF MIMO detection for a 16-QAM MIMO

    (2×2) in Tgac Channel D

  6. CONCLUSION

LRA-ZF MIMO detection with the variants of two-stage QRD algorithms were presented. Performance in terms of BER at low, medium and high SNR regime were considered with the AWGN and indoor channel variation. Results have shown that the first stage CGS, Householder transformation and RBQRD algorithms with a second stage of GR-QRD provide better results with the AWGN. In addition with the indoor channel variation we have shown that the performance of LRA-ZF MIMO detection using two-stage QRD with the first stage as RBQRD is better in channel D next to AWGN rather than B and C. As a future work the complexity of the QRD algorithms with the indoor channel variation in LRA-ZF MIMO detection will be studied.

Fig 4. Performance of LRA-ZF MIMO detection for a 16-QAM MIMO (2×2) using RBQRD in first stage for an AWGN and TGac

channels

REERENCES

  1. A. K. Lenstra , H. W. Lenstra, Jr. , and L. Lov z Factoring Polynomials with Rational Coefficients, Math. Annl., vol. 261, pp. 515-534,1982.

  2. Cong Ling, Wai Ho Mow and N. Howgrave-Graham, Reduced and Fixed-Complexity Variants of the LLL Algorithm for Communications, IEEE transactions on Communications, vol 61. no.3,2013, pp. 10401050

  3. Wen Zhang, Sanzheng Qiao, and Yimin Wei, A Diagonal Lattice Reduction Algorithm for MIMO Detection,

    IEEE Signal processing letters, vol 19. no.5,2012, pp. 311 314.

  4. Wen Zhang, Sanzheng Qiao, and Yimin Wei, HKZ and Minkowski Reduction Algorithms for Lattice-Reduction-

    Aided MIMO Detection, IEEE transactions on Signal processing, vol 60. no.11, 2012, pp. 5963 5976.

  5. James W. Demmel, Applied Numerical Algebra, 2nd Edition, SIAM publications, Prentice-Hall, Englewood Cliffs, New Jersey,1997.

  6. C. Oestges, Validity of the Kronecker Model for MIMO Correlated Channels, in Proc. IEEE Vehicular Technology Conference, Melbourne, Vic., May 2006, pp. 28182822.

  7. Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, Amendment 4: Enhancement for Very High Throughput for Operations in Bands below 6 GHz. IEEE P802.11ac / D3.0, June 2012.

  8. A. Chockalingam and B. Sundar rajan, Large MIMO systems, 1st Edition, Cambridge University press, New York,

    Fig 5. Performance of LRA-ZF MIMO detection for a 16-QAM MIMO

    (2×2) in AWGN channel

    2014.

  9. H.A. Mahmoud, and H. Arslan, A low-complexity high-speed QR decomposition implementation for MIMO receivers, in Proc. IEEE International symposium on Circuits and Systems, Taipei, 2009, pp.33- 36.

  10. S. Timothy, Numerical Analysis, Pearson Education Inc., George Mason University, 2006.

  11. Sven Hammarling and Craig Lucas,Updating the QR Factorization and the least squares problem, Manchester Institute for Mathematical Sciences,School of Mathematics, 2008.

  12. R. Thomas, R. Knopp B. T. Maharaj and L. Cottatellucci Detection using Block QR decomposition for MIMO HetNets, in Proc. Asilomar conference on signals, systems and computers, Pacific Groove, CA, USA, 2014,

  13. K.A Singhal, T.Datta and A.Chockalingam, Lattice reduction aided detection in large-MIMO systems, , in Proc. IEEE Signal Processing Advances in Wireless Communication (SPAWC) Darmstadt, June 2013, pp. 594 598.

B. Syed Moinuddin Bokhari completed his B.E from Government College of Engineering, Bargur, INDIA affiliated to University of Madras in 2001, received his M.E from Jayaram College of Engineering and Technology, Thuriayur, INDIA affiliated to Anna Univeristy, Chennai INDIA in 2005.Currently he is a research scholar at the department of Electronics and Communication Engineering, College of Engineering Guindy, Anna University ,Chennai INDIA. His field of interests include Wireless communication and networks, Signal processing. His current research interests are Multiuser detection, Multiuser scheduling applied to FPGA based SISO/MIMO test beds. He is a life member of ISTE.

M.A. Bhagyaveni received her B.E. degree in Electronics and Communication Engineering from GCT, Coimbatore, India in 1997, M.E. degree in Optical Communication from CEG, Gunidy, India in 1999 and Ph.D. degree from CEG, Guindy, India in 2006. She is currently working as Assistant Professor in the department of Electronics and Communication Engineering, CEG Campus, Anna University, Chennai, India. Her present research interests include Wireless communication, Digital communication, MIMO systems, Ad hoc networks, Sensor networks, Cloud computing, Cognitive radio technologies, Radio resource allocation in LTE and Next generation networks. She has published about 40 papers in Journals, National and International Conferences. She is a member of IEEE and several International association bodies.

Leave a Reply

Your email address will not be published. Required fields are marked *