What is the best way to choose 'ntw' using wavelet coherence?

I am comparing EEG signals. I don't know what the optimum value of ntw should be. It is the length of the smoothing window. The default is roughly half the length of the signal. Does anyone have any suggestions? Thanks Kim

Answer 1

[Wayne King]

NTW gives the order of a moving average filter on the wavelet coefficients so the optimum value depends really on what scales are most important for you and how noisy the signal is.

The default is the minimum of 20 and 1/20*length(signal) so how does that lead to 1/2 the signal length?

[Kim]

I guess I misread the default, that is clearly not half the signal length! If I am particularly interested in scales 30 to 150 using cmor1_3 and the signal is originally noisy but been wavelet denoised to level4 using bior 6.8 (I would prefer to denoise with cmor but I don't know how to do it as that is not one of the options in the WDEN function) would you happen to have a recommended NWT? If you do, would you mind giving a brief reason for your choice? Thanks

[Wayne King]

Is the default not working well for you? I would guess from your description above, you could use a smaller value because your signal has already been smoothed.

[Kim]

It's probably fine, but my signal processing experience is short (2 years) and I am not knowledgeable to know if it is optimum or not. You said it's already been smoothed- is that the DWT denoising that was done as part of the preprocessing? Kim

[Wayne King]

Yes, that's correct