I have raw EEG dataset in .mat files ( Matlab format). I need to perform band pass filtering on the data in the certain bands between 3Hz and 30 Hz. All the EEG recordings of this dataset are sampled at 256 samples per second, at 16-bit
quantization. Most of the cases contain 23 bipolar EEG signals.
As I am relatively new in Matlab, how can I do that ?. please help!
Hi
filtering with zero-phase IIR filters as suggested by Ricardo is nice. However, it's better to apply two filters in cascade, one low-pass filter and, subsequently, a high-pass one. This is because the cut-off frequency for the high pass region; since it is so close to 0 Hz, the order that you assign to the band-pass filter could be enough for the "high-pass" region, but insufficient for the "low-pass" one, and if you attempt increasing the order, you take chances that your filter gets unstable.
Are you using EEGLAB toolbox? It is very recommended for EEG data analysis in Matlab.
If the only processing you need to do is frequency filtering, you can do it without EEGLAB. Try the following code for a Butterworth filter:
sampleRate = 256; % Hz
lowEnd = 3; % Hz
highEnd = 30; % Hz
filterOrder = 2; % Filter order (e.g., 2 for a second-order Butterworth filter). Try other values too
[b, a] = butter(filterOrder, [lowEnd highEnd]/(sampleRate/2)); % Generate filter coefficients
filteredData = filtfilt(b, a, inputData); % Apply filter to data using zero-phase filtering
Hi Huda,
You can do that using finite impulse response (FIR) filter.
One more thing, you may need to downsample your data to around 250 Hz.
Cheers,
Samer
Dear Huda,
This article contains a lot of information of ECG signal filtering. Hope that it's interested for you.
regards
Hi
filtering with zero-phase IIR filters as suggested by Ricardo is nice. However, it's better to apply two filters in cascade, one low-pass filter and, subsequently, a high-pass one. This is because the cut-off frequency for the high pass region; since it is so close to 0 Hz, the order that you assign to the band-pass filter could be enough for the "high-pass" region, but insufficient for the "low-pass" one, and if you attempt increasing the order, you take chances that your filter gets unstable.
If you use IIR filters, then yes, forward/backward (non-causal) filtering is the way to go. But in your application (EEG signal processing), I think you will ultimately be aiming for uninterrupted online processing, so you will only be able to do forward (causal) filtering. In that case, only FIR filters will give you a linear-phase response, if you don't mind a fairly long group-delay (latency). If you have plenty of processing power, and can use long filters, then I wouldn't bother down sampling. The weighted-integral-squared-error (WISE) procedure is my go-to FIR filter design method (low-pass, high-pass, band-pass, arbitrary group-delay, etc.).
There exists transient time-domain response for EEG filters. This is not a problem for real-time application, because you may wait several seconds to omit transient time-domain response caused by filters. However, you need to consider this problem for your raw EEG data set. It's better for you to get a big EEG data set. You may try different filters and cut the response with transient response.
Hello,
how can i select the:
filterOrder = 2; % Filter order (e.g., 2 for a second-order Butterworth filter). Try other values too
What is better for eeg signals?
Dear Huda da ,
Could you find any satisfactory filter for your work? May you guide me how can I perform a same filter on my data in matlab?
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Process