
How can we make this code shorter?
1 view (last 30 days)
Show older comments
function APSM
figure
L=200;%Dimension of the unknown vector
N=3500; %Number of Data
theta=randn(L,1);%Unknown parameter
IterNo=100;
MSE1=zeros(N,IterNo);
MSE2=zeros(N,IterNo);
MSE3=zeros(N,IterNo);
MSE4=zeros(N,IterNo);
noisevar=0.3;%For the high noise scenario set this value equal to 0.3
epsilon=sqrt(2)*sqrt(noisevar); %Epsilon parameter used in the APSM
X = randn(L,N);
inputvec = @(n) X(:,n);
noise = randn(N,1)*sqrt(noisevar);
y = zeros(N,1);
y(1:N) = X(:,1:N)'*theta;
y = y + noise;
for It=1:IterNo
It
xcorrel = randn(N+L-1,1);
xcorrel = xcorrel/std(xcorrel);
X = convmtx(xcorrel,L)';%Generate noise process
X(:,1:L-1) = [];
inputvec = @(n) X(:,n);
noise = randn(N,1)*sqrt(noisevar);%Generate noise process
y = zeros(N,1);
y(1:N) = X(:,1:N)'*theta;
y = y + noise;
w=zeros(L,1);%Estimate vector
mu=0.2;
delta=0.001;
q=30;% Number of window used for the APA and the APSM.
for i=1:N
if i>q
qq=i:-1:i-q+1;
yvec=y(qq);
Xq=inputvec(qq);
Xq=Xq';
e=yvec-Xq*w;
eins=y(i)-w'*inputvec(i);
w=w+mu*Xq'*inv(delta*eye(q)+Xq*Xq')*e;%APA update
MSE1(i,It)=eins^2;
end
end
w=zeros(L,1);
for i=1:N
if i>q
qq=i:-1:i-q+1;
yvec=y(qq);
Xq=inputvec(qq);
Xq=Xq';
sum1=zeros(L,1);
sum2=0;
for jj=1:q
p=projection(yvec(jj),Xq(jj,:),w,epsilon);%Compute projection onto hyperslab
sum1=sum1+(1/q)*p;
sum2=sum2+(1/q)*(((p-w)'*(p-w)));
end
if sum1==w%Extrapolation Parameter Computation
Mn=1;
else
Mn=sum2/((sum1-w)'*(sum1-w));
end
eins=y(i)-w'*inputvec(i);
w=w+Mn*0.5*(sum1-w);%APSM update
MSE2(i,It)=eins^2;
end
end
w=zeros(L,1);%RLS recursion
delta=0.001;
Pi=delta*eye(L);
P=(1/delta)*eye(L);
for i=1:N
gamma=1/(1+inputvec(i)'*P*inputvec(i));
gi=P*inputvec(i)*gamma;
e=y(i)-w'*inputvec(i);
w=w+gi*(e);%RLS update
P=P-(gi*gi')/gamma;
MSE3(i,It)=e^2;
end
w=zeros(L,1);%NLMS Recursion
delta=0.001;
mu=1.2;
for i=1:N
e=y(i)-w'*inputvec(i);
mun=mu/(delta+inputvec(i)'*inputvec(i));
w=w+mun*e*inputvec(i);%NLMS update
MSE4(i,It)=e^2;
end
end
MSEav1=sum(MSE1')/IterNo;
MSEav2=sum(MSE2')/IterNo;
MSEav3=sum(MSE3')/IterNo;
MSEav4=sum(MSE4')/IterNo;
plot(10*log10(MSEav1),'r');hold on
plot(10*log10(MSEav2),'g');hold on
plot(10*log10(MSEav3),'b');hold on
plot(10*log10(MSEav4),'k')
function [fnew]=projection(y,x,f,epsilon)%Projection onto the hyperslab function
x=x';
inner=f'*x;
if((inner-y<-epsilon))
fnew=f+((y-inner-epsilon)/(x'*x))*x;
else if (inner-y>epsilon)
fnew=f+((y-inner+epsilon)/(x'*x))*x;
else
fnew=f;
end
end
Answers (1)
KALYAN ACHARJYA
on 13 Apr 2020
One Suggestion use array or cell array, where it can be, for example-
#Instead of this
MSE1=zeros(N,IterNo);
MSE2=zeros(N,IterNo);
MSE3=zeros(N,IterNo);
MSE4=zeros(N,IterNo);
Use
MSE=cells(1,4)
Instead of this
MSEav1=sum(MSE1')/IterNo;
MSEav2=sum(MSE2')/IterNo;
MSEav3=sum(MSE3')/IterNo;
MSEav4=sum(MSE4')/IterNo;
#DO Array indexing
MSEav(i)=sum(MSE{1}')/IterNo;
See more about vectorization.
See Also
Categories
Find more on Descriptive Statistics in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!