Offset after Butterworth bandpass filtering

Question

guess I'm stuck with this filtering problem and I'm thinking it has something to do with my raw data but I'm not so sure. So the problem is the following, please refer to the figures below, basically, all I'm trying to do is implement a bandpass Butterworth filter to filter the raw data trace (green). The raw data consists of 2501 points and was sampled at 20 kHz (100-ms). I'd like to bandpass filter this trace between 3 Hz and 170 Hz using the Matlab function 'butter' followed by either 'filter' (top figure) or 'filtfilt' (bottom figure), preferably the latter.

My Matlab code is as follows:

A = Average(1,:); % raw data
fs_Hz = 20000; % sampling rate (Hz)
order = 3;
fcutlow = 3; % Hz
fcuthigh = 170; % Hz
[b,a]  = butter(order,[fcutlow,fcuthigh]/(fs_Hz/2), 'bandpass');
x = filtfilt(b,a,A);
plot(x);
hold on
plot(A,'color','g');
legend('filtered data','raw data'); 

As you might guess the problem is the offset that I observe after both filtering functions. These 2501 point traces actually stem from a very long data-trace, should I possibly filter this long trace prior to extracting these smaller traces? Is the reason for the offset the fluctuating pattern of the trace? I can't detrend the data as I need the y-axis information for the subsequent analysis on this data. Should I be using a completely different type of filtering method; FIR instead of IIR?? Any advice on my problem would be greatly appreciated.

IljaPilja

Answer 1

Your data have a non-zero DC component which is filtered out by the bandpass filter. So the filter output has approximately zero DC, which causes the corresponding vertical shift of the signal. If you want to retain the DC information in the signal you must use a lowpass filter instead of a bandpass filter.