Skip to content
Advertisement

Define correct scipy.signal.spectrogram input parameters

I have the following code:

sampling_rate=128
N = sampling_rate
_f, t, Sxx = signal.spectrogram(_signal, sampling_rate, nperseg=N, nfft=N, noverlap=N-1, mode="complex")
cm = plt.pcolormesh(t, _f, np.log(np.abs(Sxx)), cmap="viridis")
plt.savefig('Spectogram.png', dpi=300, frameon='false') 

which is giving me the following plot: enter image description here

What is the correct way to define the correct parameters, namely: nperseg, nfft and noverlap to obtain a correct and smooth plot?

Thank you!

Plot of the input signal: enter image description here

Advertisement

Answer

I was able to solve this issue by normalising the input signal and doing:

t, f, Sxx = signal.spectrogram(filt_signal[sampleIDX], fs=128, 
                        nperseg=200, noverlap=180, return_onesided=True)    
Sxx = np.log(Sxx)
Sxx = (Sxx - np.min(Sxx)) / (np.max(Sxx) - np.min(Sxx))
image_Sxx = Image.fromarray(np.uint8(cm.viridis(Sxx)*255))
User contributions licensed under: CC BY-SA
8 People found this is helpful
Advertisement