Re: Remez Octave Signal vs Scipi.signal

[Thomas D. Dean]

On 09/03/2017 09:28 PM, Dmitri A. Sergatskov wrote:
> On Sun, Sep 3, 2017 at 10:49 PM, Thomas D. Dean <address@hidden
> <mailto:address@hidden>> wrote:
>
>
>     figure()
>     hold on
>     subplot(3,1,1)
>     pwelch(x2,4,0,2048,Fs)
>     title('pwelch x2')
>     subplot(3,1,2)
>     pwelch(x4_lpf,4,0,2048,Fs_y)
>     title('pwelch x4_lpf')
>     subplot(3,1,3)
>     pwelch(x4_filt,4,0,2048,Fs_y)
>     title('pwelch x4_filt')
>     hold off
>
>
>
> ​I am not a digital filter guy, but
> ​
> pwelch(x4_lpf,4,0,2048,Fs_y)
>
> ​does not make any sense to me.
>
>
> ​Try
> ​pwelch(x4_lpf,[],[],2048,Fs_y, "loglog", "linear")
>
> (you pick up 4 data-point window, pad it with 0 to 2048, do fft on that,
> make a zero overlap -- what is the point of pwelch then?)
>
> Also there is a decimate() function in signal pkg.

Thanks,  I know less about DSP than almost everyone.  I did the math for
DSP 40 years ago at Cal.

The difficulty I had was in the values returned by the
Octave:signal:remez function as compared to the scipy.signal.remez function.

I don't remember why one cannot do an fft, zero all the data that does
not fit the filter criteria, and do an ifft.  The math must back this
up, but, I don't remember it...  I am looking for a book that says, "if
you want that result, do this."

Tom Dean