What is the algorithm used by freqz?

I have taken a look at the help pages of fft and frezq but I am still unclear.

The function fft computes a DFT fast; if I fft a sinusoid whose frequency is an multiple of 1/N (where N is the signal length) I get back a N-length signal that contains two bins equal to N/2 while all other bins contain quantization noise. Now, if I freqz the same signal with N points, I get spectral leakage all around the place. For example,

cos_signal = cos(2*pi*[0:1:1023]/8); h1=fft(cos_signal); plot(20*log10(abs(h1)/max(abs(h1))));

Gives a spectrum with two discrete deltas, and zero (ie quantization error), everywhere else.

If I call freqz(cos_signal), which supposedly uses 512 frequency points, I get the same thing as with fft (even though the fft was a 1024 point fft). However, if I call freqz(cos_signal,1,1024), I get spectral leakage.

I would appreciate an explanation with the implementation details of freqz.

Thanks,

Fernando.

Answer 1

[Paulo Silva]

edit freqz

and see how it works