[ibdfe] matlab simulation of frequency domain equalization based on ibdfe

1. Software version

matlab2015b

2.IBDFE Frequency domain equalization scheme

What we have now IBDFE The structure is as follows ：

      From the structure ,IBDFE It is composed of feedforward filter and feedback filter , among C and B Represents the coefficients of feedforward filter and feedback filter . From the existing literature and data , At present, the structure is in the calculation process , Each iteration requires the estimation of coefficients , Thus, the complexity of system implementation is increased . Aiming at problems , At present, the main research results are as follows LC-IBDFE etc. , It separates the error in the decision signal from the expected signal , This reduces the complexity . But it's similar LC-IBDFE How to improve , It is based on the assumption that the bit error rate of each iteration is the same and very small , In reality, this situation is difficult to meet the conditions . The other is in IBDFE in , When there is serious channel fading , It will lead to overestimation of correlation factors , This leads to the spread of errors . In response to this question , The existing achievements mainly include the joint equalization algorithm of joint channel estimation and channel equalization . But the complexity of this algorithm is further increased .

3. Part of the source code

clc;
clear all;
close all;
warning off;
addpath 'func\'
rng('default');
rng(1);
Blk_size = 512;      % Block size 512
Chu_size = 64;       %chu Sequence , size 64
NFrame   = 1000;  
SNR      = [0:1:14]; % Signal-to-noise ratio dB
Modsel   = 2;        %QPSK
Fs       = 10*10^3;  % Sampling rate 
Ts       = 1/Fs;        
Fc       = 5*10^3;   % Carrier frequency 
Fd       = 10;      % Doppler shift 
tau = [0 0.5 0.5 1.2 1.2 2.1 2.1 3.3 3.3 4.8 4.8 6.5 6.5 8.5 10.8]*10^-4;    
pdb = [0 -0.967 -0.967 -0.967 -1.933 -1.933 -1.933 -1.933 -2.900 -2.900 -2.900 -2.900 -2.86 -3.86 -3.84];          
% Channel model 
Channel  = rayleighchan(Ts,Fd,tau,pdb);
%FFT Transformation 
H_channel0 = fft(Channel.PathGains./sqrt(sum((abs(Channel.PathGains)).^2)),Blk_size+Chu_size+Chu_size);

%CHU Sequence 
Chuseq = zeros(1,Chu_size);
for k = 0:Chu_size-1
    tmps(k+1) = pi*k^2./Chu_size;
end
I      = cos(tmps);
Q      = sin(tmps);
Chuseq = I+sqrt(-1)*Q;
% Bit error rate 
%turbo Parameters 
Mss    = 295;
for n = 1:length(SNR)
    ErrMMSE = 0;
    for k = 1:NFrame
        [n,k]
        rng(k);
        % Random 
        Tdin       = randint(1,Mss);
        % utilize turbo Interleaver for , structure TB-DEF, Three way output 
        output     = [func_turbo_code(Tdin)];
        output     = reshape(output, 1, []);
        seridata1  = [output,0,0];
        
        % modulation 
        Data       = modulation(seridata1,Modsel);
        Tx         = [Chuseq,Data,Chuseq];
        Channel0   = Channel.PathGains./sqrt(sum((abs(Channel.PathGains)).^2));
        Rx1        = filter(Channel0,1,Tx);           
        Rx2        = awgn(Rx1,SNR(n),'measured');
        Rx3        = Rx2;%(Chu_size+1:Chu_size+Blk_size);   
        H_channel  = H_channel0;     
        % Frequency domain equalization 
        Y          = fft(Rx3,Blk_size+Chu_size+Chu_size);       
        Wk         = conj(H_channel)./(H_channel.*conj(H_channel)+10^(-SNR(n)/10)); 
        Zk         = Y.*Wk;
        Qk         = zeros(size(Zk));
        Bk         = (Blk_size-Chu_size)*(abs(H_channel).^2+10^(-SNR(n)/10))./(sum(abs(H_channel).^2+10^(-SNR(n)/10)))-1;
        P          = 5;
        % call CNN Output weight of neural network  
        load CNNmodel.mat
        Iter       = 5;
        for iter = 1:Iter
            Wk = conj(H_channel)./(H_channel.*conj(H_channel)+10^(-SNR(n)/10)/P).*(1+Bk);  
            Zk         = Y.*Wk;
            Uk         = Zk-Qk;
            RxMMSE0    = ifft(Uk,Blk_size+Chu_size+Chu_size);    
            xn         = sign(real(RxMMSE0))+sqrt(-1)*sign(imag(RxMMSE0));
            % Go to UW
            RxMMSE1    = xn(Chu_size+1:Blk_size);
            % To adjudicate 
            RxMMSE     = demodulation(RxMMSE1,Modsel);   
            Tdecode    = round(func_turbo_decode(2*RxMMSE(1:end-2)-1));
            tmps       = Tdecode;
            
            XK         = fft([tmps,Chuseq],length(RxMMSE1));
            % call CNN Deep learning neural networks , Calculation Bk value 
            Bk0        =([H_channel.*conj(H_channel)]+10^(-SNR(n)/10)/P)/(mean(([H_channel.*conj(H_channel)]+10^(-SNR(n)/10)/P)))/(Blk_size)-1;
            Bk         = func_CNN(H_channel,Bk0,cnn);
            Qk         = [XK,ones(1,192)].*Bk;
        end
        CrrMMSE    = find((Tdin-Tdecode) == 0);
        ErrMMSE    = ErrMMSE+(Mss-length(CrrMMSE));
    end
    % Statistical bit error rate 
    errors(n) = ErrMMSE/(Mss*NFrame*Modsel);
end
 

figure
semilogy(SNR,errors,'b-o');
hold on;
grid on;
xlabel(' Signal-to-noise ratio SNR(dB)')
ylabel(' Bit error rate SBR')

save R1.mat SNR errors

4. Simulation results

 A1-169