filtfilt vs self coded filtfilt giving different results.

Hello, I have a large iEEG dataset. I wanted to bandpass the iEEG signal between 0.5-125Hz. So, I used 2nd order butterworth filter. I used "filtfilt" to eleminate phase lage. Alternatively, I also applied the conventional "filter" in both directions to eliminate phase-lag. However, using "filtfilt" is changing my data: negative values to positive values and vice-versa, also phase-lagged. But, my conventional "filter" applied in both directions working perfectly fine. Can someone help me with this. I have attached the filtered signals and raw signal in plot and also the code. Happy to provide the data for the plot.

%filtfilt EEG is channles x datapoints
cut_off=[0.5 125];
sample_rate=512;
ny=sample_rate/2;
cut_off=cut_off/ny;
order=2;
[b,a] = butter(order, cut_off,'bandpass');
eeg_filtfilt = filtfilt(b,a,EEG);

%conventional filter applied in both directions (own code)
sample_rate=eeg_data.Fs;
ny=sample_rate/2;
cut_off=cut_off/ny;
order=2;
[b,a] = butter(order, cut_off,'bandpass');
EEG_f = filter(b,a,EEG);
EEG_f_reverse = EEG_f(end:-1:1, :);
EEG_f_reverse_f = filter(b,a,EEG_f_reverse);
eeg_own = EEG_f_reverse_f(end:-1:1, :);

[Micke Malmström]

I think that when you use filter forwards AND backwards the filter order is doubled. If you specify order 4 for filtfilt maybe they will be the same?

Answer 1

[Mathieu NOE]

hello

IMHO the most obvious error is that your second simulation runs on a completely different filter

see the second time you're doing :

cut_off=cut_off/ny;

your second filter has ny times lower cut off frequencies vs the first filter , so no surprise the home made filter (x2) method seems to fail;