ENH: signal: remezord implementation

[nils-werner]

Rebased remezord implementation as mentioned in #1002

[zaazbb]

I think, remezord() should have a Stopband Attenuation param, to input a Stopband Attenuation (dB) list.

firpmord() in mathlab, can input Passband Ripple (dB) and Stopband Attenuation (dB) both.
[n,fo,ao,w] = firpmord(f,a,dev,fs)
dev is a vector the same size as a that specifies the maximum allowable deviation or ripples between the frequency response and the desired amplitude of the output filter for each band.

is the rips param in remezord() equal to the dev param in firpmord()?

[zaazbb]

Can I rewrite this matlab code to python? as below:

[n,fo,ao,w] = firpmord(f,a,dev,fs);
b = firpm(n,fo,ao,w);

change to:

[n,fo,ao,w] = remezord(f,a,dev,fs);
b = remez(n,fo,ao,w, fs);

[rgommers]

There is still an unresolved discussion about whether ripples are defined correctly. Other things that need to be addressed:

• code needs unit tests.
• all functions except remezord need to start with an underscore.
• the docstrings for remlplen_* repeat the same text 3 times. The 2nd and 3rd docstring can refer to the first one. The references to papers should all be moved to the remezord docstring.
• the code needs some PEP8 cleanup (spaces around operators etc.)
• the Hz parameter is missing from the docstring.
• docstring format: See should be See Also, typo in Paramters, remove blank lines right below section headings.
• An example in the docstring would be very useful.
• remezord should be added to the function listing in signal/__init__.py.

[zaazbb]

yes, it seems have some problems.
If i input frequency in Hz directly, this code can not work correctly, the plot output is nothing.

    fs = 1200000 # 1200 KHz.
    # bandpass, 269 KHz - 392 KHz.
    fstop1 = 269000 - 6780
    fpass1 = 269000
    fpass2 = 392000
    fstop2 = 392000 + 6780
    dstop1 = 0.002 # First Stopband Attenuation (dB)
    dpass = 0.019 # Passband Ripple (dB)
    dstop2 = 0.002 # Density Factor
    dens = 20
    N, Fo, Ao, W = remezord([fstop1, fpass1, fpass2, fstop2],
                            [0, 1, 0], [dstop1, dpass, dstop2], fs)
    bpf = remez(N, Fo, Ao, W, fs, grid_density = 20)
    w, h = freqz(bpf)
    plt.plot(w * fs / 2 / np.pi, 20 * np.log10(abs(h)))
    plt.ylabel('Amplitude [dB]')
    plt.xlabel('Frequency [Hz]')
    plt.grid()
    plt.show()

and if i input param as firpmord() in matlab, the output frequency is half of except. as below:

    fs = 1200000 # 1200 KHz.
    # bandpass, 269 KHz - 392 KHz.
    fstop1 = 269000 - 6780
    fpass1 = 269000
    fpass2 = 392000
    fstop2 = 392000 + 6780
    dstop1 = 0.002 # First Stopband Attenuation (dB)
    dpass = 0.019 # Passband Ripple (dB)
    dstop2 = 0.002 # Density Factor
    dens = 20
    N, Fo, Ao, W = remezord([i/fs/2 for i in [fstop1, fpass1, fpass2, fstop2]],
                            [0, 1, 0], [dstop1, dpass, dstop2])
    bpf = remez(N, Fo, Ao, W, grid_density = 20)
    w, h = freqz(bpf)
    # the plot result is half of (269 KHz - 392 KHz).
    plt.plot(w * fs / 2 / np.pi, 20 * np.log10(abs(h)))
    # use this line, the plot result is correct.
    #  plt.plot(w * fs / np.pi, 20 * np.log10(abs(h)))
    plt.ylabel('Amplitude [dB]')
    plt.xlabel('Frequency [Hz]')
    plt.grid()
    plt.show()

[zaazbb]

and i need this funciton now, wich function can replace it?

[rgommers]

@zaazbb I'm not sure, I'm not aware of a remezord implementation in another Python package.

[rgommers]

@charris IIRC you know this stuff quite well and still have some remez improvements lying around. Apart from the things I pointed out above, what do you think about this remezord implementation?

[charris]

I'll take a look at it. Know quite well means I studied it a bit some years ago )

[larsoner]

@nils-werner are you interested in coming back to this? It looks like you force pushed an update four years ago that nobody saw (GitHub does not notify for force push updates). If the comments above are addressed and you're still interested in working on this we can review. If not perhaps we should close?

[nils-werner]

Yes, I would be interested to give this another go.

[luxedo]

Any updates on this? I can give it a try since it looks like mostly bureaucracy.

• code needs unit tests.
• all functions except remezord need to start with an underscore.
• the docstrings for remlplen_* repeat the same text 3 times. The 2nd and 3rd docstring can refer to the first one. The references to papers should all be moved to the remezord docstring.
• the code needs some PEP8 cleanup (spaces around operators etc.)
• the Hz parameter is missing from the docstring.
• docstring format: See should be See Also, typo in Paramters, remove blank lines right below section headings.
• An example in the docstring would be very useful.
• remezord should be added to the function listing in signal/init.py.

[rgommers]

I think that would indeed be nice, thanks @luxedo.

[j-bowhay]

continued in #20983