When the fft length is even, fftshift flips the input so that the Nyquist point is to the left. But pwelch with freqrange = ‘centered’ keeps the Nyquist point on the right.
The behavior in pwelch is documented, so can’t complain.
Just wondering why the developers might have made that choice for pwelch (and others, like periodogram), rather than maintaining consistency.
rng(100);
x = rand(1,10);
X = fftshift(fft(x));
% this usage of pwelch is undocumented because noverlap is supposed to be
% a positive integer
[P,f] = pwelch(x,ones(size(x)),0,10,1,’centered’);
figure
plot(f,P*10,’-o’,(-5:4)/10,X.*conj(X),’-x’)
axis paddedWhen the fft length is even, fftshift flips the input so that the Nyquist point is to the left. But pwelch with freqrange = ‘centered’ keeps the Nyquist point on the right.
The behavior in pwelch is documented, so can’t complain.
Just wondering why the developers might have made that choice for pwelch (and others, like periodogram), rather than maintaining consistency.
rng(100);
x = rand(1,10);
X = fftshift(fft(x));
% this usage of pwelch is undocumented because noverlap is supposed to be
% a positive integer
[P,f] = pwelch(x,ones(size(x)),0,10,1,’centered’);
figure
plot(f,P*10,’-o’,(-5:4)/10,X.*conj(X),’-x’)
axis padded When the fft length is even, fftshift flips the input so that the Nyquist point is to the left. But pwelch with freqrange = ‘centered’ keeps the Nyquist point on the right.
The behavior in pwelch is documented, so can’t complain.
Just wondering why the developers might have made that choice for pwelch (and others, like periodogram), rather than maintaining consistency.
rng(100);
x = rand(1,10);
X = fftshift(fft(x));
% this usage of pwelch is undocumented because noverlap is supposed to be
% a positive integer
[P,f] = pwelch(x,ones(size(x)),0,10,1,’centered’);
figure
plot(f,P*10,’-o’,(-5:4)/10,X.*conj(X),’-x’)
axis padded pwelch, fftshift, centered MATLAB Answers — New Questions

​