Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

download Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

of 21

Transcript of Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    1/21

    IMPROVED EQUALIZATION

    FOR

    CODED, ZERO-PADDED OFDM(ZP-OFDM) SYSTEMS

    SUBMITTED BY:

    ANKIT SANCHETI

    2011PEC5324

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    2/21

    INTRODUCTION

    In order to combat the effect of multipath, OFDM systems is

    mainly used.

    Most common OFDM systems use cyclic-prefix so referred as

    CP-OFDM.

    a N sample OFDM symbol is cyclically extended by NG

    samples where NG > L, and L is the length of the channel

    impulse response.

    OFDM symbol is cyclic convolution in the case of the CP-

    OFDM system.2

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    3/21

    INTRODUCTION CONTINUED.

    Another way to deal with multipath is to pad the N sample

    OFDM symbol with NG zeroes, referred as Zero-padded

    OFDM (ZP-OFDM) system.

    One of the main advantages of ZP-OFDM as compared to

    CP-OFDM is that the total transmitted energy can be reduced.

    Equalization at the receiver needs to be performed differently

    because the convolution of the channel with the OFDM

    symbol is now a linear convolution than cyclic convolution.

    3

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    4/21

    SIGNAL MODEL

    For any L, let us first define a (L x L) Fourier matrix as follows:

    Data vector of length N be a = [a0 a1 aN1]

    The OFDM symbol derived by taking the IFFT of a be

    = A = [A0 A1AN1]

    The vector A is extended by NG zeroes, where NG is thelength of the guard interval.

    A is transmitted over a channel with impulse response

    denoted by

    we have explicitly forced the impulse response of the channelto be at most NG +1 samples long.

    4

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    5/21

    SIGNAL MODEL CONT.

    The received vector r can be expressed in terms of the channel

    h and transmitted vector A as follows:

    2

    The eigen-vector matrix is the Fourier matrix, Hc can be written

    as: 5

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    6/21

    is a N +NGdimensional diagonal matrix of eigen values i ofHc.

    These eigen values are also the FFT of the channel i.e.

    The received vector is

    ..3

    We will now describe a number of estimators for a based onequations (2) and (3).

    6

    SIGNAL MODEL CONT.

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    7/21

    A. OVERLAP-ADD(OLA) EQUALIZER

    Adding last NG samples of the received vector r to the first NGsamples, we get the signal model:

    We get overlap-add (OLA) equalizer by taking N-point FFTs ofboth sides.

    Exhibit similar performance as the CP-OFDM system, sincespectral-nulls will be inverted.

    No performance improvement as compared to a CP-OFDM

    system.

    7

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    8/21

    B. LEAST-SQUARES(LS) EQUALIZER

    For the transmitted signal vector a

    LS equalizer is

    The estimation noise is colored, which may have an effect oncoded systems.

    For large N, the complexity can be high since the equalizerinvolves the inverse of a channel-dependent matrix that is not adiagonal matrix..

    Guarantees symbol recovery even in the presence of spectralnulls.

    Better performance than the OLA equalizer

    8

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    9/21

    C. SUBOPTIMAL LEAST-SQUARES(SLS) EQUALIZER

    First multiply R by 1 , and then do a LSestimator for a.

    Much easier to implement than the LS equalizer.

    If there is a spectral null, the estimate error will be

    high. Performance is similar to the OLA equalizer.

    9

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    10/21

    D. TRUNCATEDSUBOPTIMAL (TS) EQUALIZER

    For the truncated, suboptimal (TS) equalizer:

    Matrix T has been truncated to remove the small eigen

    values. Hence the inverse will not cause excessive noise

    enhancement.

    For large N even this implementation would become

    extremely complex. 10

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    11/21

    E. SIMPLIFIEDTRUNCATEDSUBOPTIMAL (STS)

    EQUALIZER

    It can be described as

    It is very simple to implement, since there are no non-diagonal

    channel-dependent matrix inversions required.

    since the small eigen values have been removed from T ,

    noise enhancement is suppressed.

    11

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    12/21

    SOFT-METRIC CALCULATION FOR

    CODED ZP-OFDM

    Let us now consider a bit-interleaved-coded-modulation (BICM)

    system.

    One way to compute the soft-metrics of the bits is to first

    equalize, using one of the above equalizers to form an estimate

    a and then evaluate the soft-metric for the ith bit in the kth

    symbol as follows:

    where 2k is the variance of the kth component of the estimation

    noise, which in general is colored.12

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    13/21

    SOFT-METRIC CALCULATION FOR

    CODED ZP-OFDM CONT.

    An alternative way would be to start with eqn (3)

    These soft-metrics are then used in a Viterbi decoder

    to decode the transmitted sequence.

    The decoder using the above metric is the maximum

    likelihood (ML) receiver and in general this will give us

    a lower bound on achievable performance. 13

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    14/21

    SIMULATIONS

    We consider two ZP-OFDM systems.

    Un-coded 16-QAM ZP-OFDM

    Coded 16-QAM ZP-OFDM

    OFDM symbol length N = 64

    The number of zero-padded samples NG = 16.

    Channels :-

    The first is defined by h = [1/2 0 0 0 0 1/2]T .

    The second channel we consider is an exponentially Rayleigh fadedchannel with rms delay spread of 80ns.

    Sample rate = 20 MHz.

    Length of impulse response = 17 = (NG +1)14

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    15/21

    MSE performance of different equalizers in

    h = [1/2 0 0 0 0 1/2] channel.

    15

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    16/21

    MSE performance of different equalizers in exponentiallyfaded Rayleigh channel with 80ns rms delay spread.

    16

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    17/21

    PER performance of non-iterative equalizers in

    h = [1/2 0 0 0 0 1/2] channel.

    17

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    18/21

    PER performance of iterative equalizers in

    h = [1/2 0 0 0 0 1/2] channel.

    18

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    19/21

    PER performance of non-iterative equalizers inexponentially faded Rayleigh channel with 80 ns rms delay

    spread channel.

    19

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    20/21

    PER performance of iterative equalizers in exponentiallyRayleigh faded channel with 80 ns rms delay spread

    channel.

    20

  • 8/2/2019 Improved Equalization for Coded, Zero-Padded OFDM (ZP-OfDM) Systems

    21/21

    CONCLUSIONS

    We developed two new equalizers for ZP-OFDM: the TS and STSequalizers and showed their performance advantage over the more

    commonly used OLA equalizer in channels with and without spectral

    nulls.

    STS equalizer can improve performance, especially in channels with

    spectral nulls, using a larger FFT size.

    STS initialization provides a faster convergence than the OLA

    initialization especially for channels with spectral nulls.

    It is feasible to obtain the performance of a LS equalizer in a coded

    ZP-OFDM system without requiring matrix inverses of channel-

    dependent non-diagonal matrices, by using the iterative method of

    soft-metric calculation

    21