Hallo Eric, You probably had an answer for your question but I will give you my suggestion anyway. It might help someone else in the future :).
First, taking the window size equal to the length of your data is wrong. In fact, the window size is a function of your sample frequency, and the frequency you want to resolve while using the function mscohere. Let's say the data is collected at 50 Hz and you want to resolve the coherence between two time series at 0.5 Hz, then you will need a window length = 50/0.5 = 100 samples. In practice, it is ALWAYS better to multiplied this value by 4 or 5 so to be sure that the desired frequency falls within the frequency bin. Therefore, your window length = 400 or 500.
Now, concerning the noverlap question. Depending on how noisy is your signal, increasing the noverlap size will decrease the noise variance. If your signal is a clean one, perhaps an overlap = 0 or 10% would not change anything. I suggest, you play with the noverlap to see which value corresponds best in your case.
Taking your example,
for noverlap = [0 0.1 0.2 0.3 0.4 0.5 0.6 0.9]*window_size
mscohere(xdata,ydata,[],noverlap,window_size,samp_rate);
end
I hope this help.
Best regards, Axel
