RLS自适应滤波与Matlab仿真

@RLS自适应滤波与Matlab仿真

一、RLS自适应滤波模型

  假设d(n)为参考信号，k(n)为增益向量，e(n)为先验估计误差，w(n)为滤波器抽头系数，y(n)为RLS滤波后输出信号。

RLS采用指数加权的最小二乘法，其代价函数为指数加权的误差平方$$J(n)=\sum _{i=0}^n{\delta }^{n-i}|e(n){|}^2$$ 误差计算公式：$$e(n)=d(n)-y(n)$$
权值更新公式:$$w(n+1)=w(n)+k(n)e(n)$$
RLS滤波模型如下图所示：

算法步骤：
a.估计当前值：$y(n)=w\ast X{N}^{\prime };$
b.估计误差：$e(n)=s(n)-y(n);$
c.计算增益：$k=p\ast XN/(step+X{N}^{\prime }\ast p\ast XN);$
d.更新权值：$w=w+k\ast conj(e(n));$
e.更新协方差矩阵：$p=(1/step)\ast (1-k(n)\ast u(n))\ast p;$

二、matlab程序

1.代码

%--------------Function:Adaptive filter--------------------------
%--------------Remark: RLS Method -------------------------------
%--------------Data：2024.11.13----------------------------------
%--------------Author：Clemence----------------------------------
%--------------Reference：---------------------------------------



clc;
clear all;
close all;

g = 100;% Simulation Count
N=1024;% Sample Points
k=128;% Filter Coefficient Number
pp = zeros(g,N-k);% Error Memory

%-------------------------------Signal Paras--------------------------
f = 24e6;       % Signal Frequency
phase = 0;      % Initial Phase
a = 1;          % Amplitude

%------------------------------Sample Paras----------------------------
Fs = 100e6;     % Sample Frequency
ts = 1/Fs;      % Sample Interval
t = 0:ts:(N-1)*ts;      % Sample Time

s= a*cos(2*pi*f*t + phase); % Signal
 
%------------------------------RLS Paras-------------------------------
delta = 0.001;
step = 0.98;

for q = 1:g
   
    xn = awgn(s,10);
    %---Initial-----------------
    y=zeros(1,N);  % Output Signal y
    y(1:k)=xn(1:k); % k
    w=zeros(1,k);% Filter Coefficient
    e=zeros(1,N);% Error
    p = (1/delta)*eye(k); %Covariance Matrix

    %-----RLS---------------------
    for i=k+1:N
%         XN=xn((i-k+1):(i));
        XN=xn(i:-1:i-k+1);
        y(i)=w*XN';% Estimate y
        e(i)=s(i)-y(i);% Estimate Error
        v = p*XN';
        kk = (1/step)*v/(1+(1/step)*XN*v);% Calculate Gain
        w=w+kk'*conj(e(i));% Update Weight
        p = (1/step)*(eye(k)-kk*XN)*p;% Update Covariance Matrix
    end
    pp(q,:) = e(k+1:N).^2;
    
end

for b = 1:N-k
    bi(b) = sum(pp(:,b))/g;% Calculate Average Error
end
figure;% Output Signal -y
plot(y);

t=1:N-k;
figure(2);% Error Curve
plot(t,bi,'r');
title("Mean Square Error Curve");
hold on    

2.仿真结果

三、结果分析

  从仿真结果可以看出RLS的收敛速度较快，滤波器系数更新时计算量较大，收敛速度的提升是以计算速度为代价的。因此，实际应用中，可根据收敛后误差允许范围，合理地选择RLS或者LMS自适应滤波器

四、参考资料

https://zhuanlan.zhihu.com/p/256998244