1.软件版本

MATLAB2013b

2.本算法理论知识

算法源于文献：

3.部分核心代码

clc;
clear;
close all;
warning off;

Num_Sensor   = 16;
Num_signal   = 2;
Num_babe     = 2;
b            = zeros(Num_Sensor,1);
H            = zeros(Num_Sensor,Num_signal);
DOA          = zeros(Num_signal,1);
wavelength   = 0.06; 
Iteration    = 1:1:20;
SNR          = 12;
d            = 0.5*wavelength*ones(Num_Sensor,1);%阵元间距
f            = zeros(Num_signal,1);
f(1)         = 1;
Len_Signal   = 100000;
Len_train    = round(0.01*Len_Signal);
%设置延迟长度
Len_delays   = round(0.05*Len_Signal);
Train_Signal = zeros(Num_signal,Len_train); 
Train_Noise  = zeros(Num_Sensor,Len_train); 
%信号和噪声以及反射信号和反射噪声
Signal       = zeros(Num_signal,Len_Signal);
Noise        = zeros(Num_Sensor,Len_Signal);
%全值，LCMV不需要进行迭代更新
W            = zeros(Num_Sensor,1);
%根据所提供的场景，设置各个入射信号的方位角
DOA          = [0,pi-atan(2)];
%根据场景设置H（tableI算法为H已知，所以根据场景来设置）
for jj=1:Num_signal
    for ii=1:Num_Sensor
        b(ii)=exp(-j*2*pi*(ii-1)*d(ii)*sin(DOA(jj))/wavelength) ;
    end 
H(:,jj)=b/(b'*b)^0.5;
end 

%产生两路随机信号
Signalo        = randint(Num_signal,Len_Signal);
Signalo        = 2*(Signalo-0.5);
%反射信号
for i = 1:Num_signal
    Signal_reflect(i,:) = 0.2*[zeros(1,Len_delays),Signal(i,1:Len_Signal-Len_delays)];
end
Signal         = Signalo;

%产生babe噪声源
babble        = randn(Num_signal,Len_Signal)+j*randn(Num_signal,Len_Signal);
babble        = 2*(babble-0.5);
for i = 1:Num_signal
    babble_reflect(i,:) = 0.2*[zeros(1,Len_delays),babble(i,1:Len_Signal-Len_delays)];
end
%实际输入到麦克风的声源
Signal        = (Signalo + Signal_reflect + babble + babble_reflect)/4;

%传感器之间的噪声干扰
Noise         = randn(Num_Sensor,Len_Signal)+j*randn(Num_Sensor,Len_Signal);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
SNRs   = zeros(1,length(Iteration));
%在几种不同的信噪比下计算误码率
diBi   = 1/(2*10^(SNR/10))^0.5;
Noise  = Noise*diBi;




for Iter = Iteration    
    Iter
    Train_Signal = Signal(:,(1:Len_train));
    Train_Noise  = Noise(:,(1:Len_train));
    x            = H*Train_Signal + Train_Noise; %加噪声
    x1           = H*Signal       + Noise;       %加噪声
    %LCMV
    R            = x*x';
    W            = inv(R)*H*inv((H'*inv(R)*H))*f;
    y            = W'*x1;
    %以第一个信号源为准计算，如果是第二个信号源，也是同样处理
    %计算SNR值
    SNRs(Iter)   = 10*log10(mean(abs(y(1,:)).^2)/mean(abs(Noise(1,:)).^2));
    %计算SDR值
    SDRs(Iter)   = 10*log10(mean(abs(Signal(1,:)).^2)/mean(abs(Signal(1,:)-y(1,:)).^2));
end

figure;
subplot(211);
plot(Iteration,SNRs,'b--','LineWidth',2);
xlabel('Iteration');
ylabel('SNR[dB]')
axis([0,20,0,15]);
subplot(212);
plot(Iteration,SDRs,'b--','LineWidth',2);
xlabel('Iteration');
ylabel('SDR[dB]')
axis([0,20,0,15]);

save fig3.mat SNRs SDRs Iteration 
%计算后面的MSE值
save w_mat.mat W

4.操作步骤与仿真结论

      我们对tableI的算法进行仿真，同时给出了用于对比的LCMV算法，通过对比，我们得到如下的仿真结果：

图一：LCMV和LC-DANSE的SNR对比仿真图；

图二：LC-DANSE的MSE图（论文中MSE只是DANSE系列的对比，TableI对应其中一种）

图三：LC-DANSE中不同P之的SNR图。

图四：LCMV和LC-DANSE的随着p变化的SNR图。A27-03