Hi,

According to the literature, Butterworth filters of order 4 are suitable. Here is a couple of codes for high-pass, low-pass, and notch filter design using 4th-order Butterworth filter:

%% Filtering using butterworth filter of order 4

% High-pass filter

filt1 = fdesign.highpass('n,f3db',4,2*10*(1/1000)); %high-pass filter, cut off frequency at 10Hz, sampling frequency of 1kHz

H1 = design(filt1,'butter');

highpass_EMG = filter(H1, here insert the file name of your EMG data being imported in workspace); % sampling frequency of 1kHz

% Low-pass filter

filt2 = fdesign.lowpass('n,f3db',4,2*500*(1/1000)); %low-pass filter, cut off frequency at 500Hz

H2 = design(filt2,'butter');

lowpass_EMG = filter(H2,highpass_EMG);

% Notch Filter (50Hz)

filt3 = fdesign.notch(4,0.05,10); %notch filter (50Hz)

H3 = design(filt3);

cleaned_EMG = filter(H3,lowpass_EMG);

figure(1);plot(cleaned_EMG);

Regards,

It is the purpose of any filter to "affect" the signal to be filtered. If you do not want any effect do not use any filtering.

(Sorry for this simple answer).

Hi,

I am not familiar with this EMG setup but keep in mind that most built-in filters are designed to display clean signals in the interface but they still give you raw unfiltered data. You can check this simply by Fourier transforming your data, as you said. If the frequencies above 40hz (in your case) is attenuated then it is the effect of buit-in filter. Otherwise check for other adjustments/recording-tricks.

This filter has several effects on your signal :

* it removes noise (and signal if any) above 40Hz

* if digital, your new data is affected by the data immediately before or after (at least 3 samples for third order).

EMG signals are sometimes presented rectified (absolute value) with no filtering except analogic antialiasing acquisition filter. There is a lot of EMG signal above 40 Hz but also electric power lines noise. Notch filters may be used to remove power line noise at 60Hz or 50Hz (depending on your country) and harmonics.

Remove digital filtering and rectification if you want to look at your frequency spectrum.

As Anne mentioned, if you automatically applied a 40Hz low-pass filter, then you have discarded a large portion of the useful EMG signal. Many commercial systems have bandpass filters built in that preserve the signal between about 10 or 20 Hz (high-pass) and 450 or 500 Hz (low-pass) --- although, the range varies somewhat depending on whether you are interested in collecting surface or intramuscular EMG. The latter has a higher frequency content.

Hi,

Almost all the frequency content of sEMG signals lies between 10 and 500Hz. In addition, the majority of recording systems have low- and high-pass filters with adjustable cut-off frequencies. Specifically, since your system is designed for both EKG and EMG recordings, I am pretty sure that there is a facility to adjust filters in a desirable manner.

Anyway, as an alternative, you may record raw EMG signals without any filtering and, afterwards, filter them off-line with digital filters designed in for example MATLAB software (Butterworth filter of order 4) to compare and contrast the outcomes.

Regards,

Hi all.

Thank you for the responds. I already record the raw EMG signals without any filtering and then I want to filter them in MATLAB.

How can I determine order of the filter?

Thank you

Hi,

According to the literature, Butterworth filters of order 4 are suitable. Here is a couple of codes for high-pass, low-pass, and notch filter design using 4th-order Butterworth filter:

%% Filtering using butterworth filter of order 4

% High-pass filter

filt1 = fdesign.highpass('n,f3db',4,2*10*(1/1000)); %high-pass filter, cut off frequency at 10Hz, sampling frequency of 1kHz

H1 = design(filt1,'butter');

highpass_EMG = filter(H1, here insert the file name of your EMG data being imported in workspace); % sampling frequency of 1kHz

% Low-pass filter

filt2 = fdesign.lowpass('n,f3db',4,2*500*(1/1000)); %low-pass filter, cut off frequency at 500Hz

H2 = design(filt2,'butter');

lowpass_EMG = filter(H2,highpass_EMG);

% Notch Filter (50Hz)

filt3 = fdesign.notch(4,0.05,10); %notch filter (50Hz)

H3 = design(filt3);

cleaned_EMG = filter(H3,lowpass_EMG);

figure(1);plot(cleaned_EMG);

Regards,

The filter design somewhat depends on your protocol (e.g., if dynamic contractions, there might be movement artifacts that you want to reduce/remove; or depending on the muscle location, you might have picked up heart beat) and what information you are hoping to glean from the signals. Providing some details about the experimental protocol and the question you are trying to answer would be helpful. What are you hoping to use the signals for (e.g., predict muscle force, muscle onset/offset relative to force, fatigue)?

Dear Abu Bakar Yahya,

signal processing matlab toolbox includes clever butterworth filter design. Pay attention to the fact that, with higher order butterworth filters (i.e, greater than 6th-8th orders) you can be faced up with highly non-linear phase response in the pass band, which means great deformations of signals morphology.

In EMG signal, usually people want to eliminate the movement artifact, which normally in the low range between 10 to 20 Hz. Most biosignal amplifier is provided with the 50 Hz power line filtering. So, to get a clean raw data that really shows the muscle contraction, eliminating the low artifact (and ECG for chest wall muscles) would be enough regardless of whatever order or filter type it is.

But then, if you have other specific criteria for your EMG, perhaps you might need to further narrowing your filter option. Normally, people are interested in the linear envelope where low pass filter or moving average filter would come into the picture.

If you are working with programming, especially the MATLAB, it is not difficult to try various types of filters with different type of orders. You just need to compare the results and pick the one that suits your application.

Hello Mr. Yahya,

You have some very good responses to your question, or should I say questions, because the original one was about the spectrum analysis of your acquired EMG and then you moved on to filter design.

I would like to add a little more to your collection of answers and hope this will complement what others have said and in particular the two very excellent answers from Ms. Biyouki

I suggest you try and devise an experiment to identify and document the characteristics of your recording setup (by set up I do not just mean the amplifiers. Setup includes everything that is used to acquire and record the EMG minus the human subject). I suggest you start by using a good quality signal generator and set it up to produce clean sinusoidal signal of a couple of mV amplitude. Put this signal into your recording electrodes and acquire this signal. Use your spectrum analysis (be it a program or a piece of hardware) to see if the power and frequency are really what they should be. You can also look at the functioning of the analogue filters, including the notch (mains) filter and see how they affect your signal, after all you you know the signal should be a clean sinusoid of a certain amplitude and frequency.

You can then begin to vary some of the parameters of the setup, for example the gain of the system and see how that affects your signal. Similarly you can change the filter settings in steps and at each see how much of the frequency you introduced into the setup is actually removed.

Using this technique you can also investigate other parameters for example your electrodes. You may find different electrodes can affect the signal.

Finally, please ensure that your system is safe for use on human subjects and if you have a computer connected to it then power the entire set up from an isolation transformer.

With best wishes

Hamid

Hi All,

I'm working on developing a protocol for EMG analysis of Shoulder muscles during some activities of daily living besides wheelchair propulsion. During my research, I found different opinions regarding which type of filtering is more applicable for such signal processing task. Could you please advice me about which is better filter to use in such analysis between the fourth order Butterworth or Daubechies wavelet.

Kindest Regards,

Hass.  

I think that our new filter designer + signal analyser tool may help you here. Please have a look at: http://www.advsolned.com/asn_filter_designer.html

download the demo version, and have a play with the biomedical data example.

Bandpass filter is applicable for filtering EMG signal. However, it's traditionally believed that a butterworth filter of higher order is most suitable. This can be achieved by cascading highpass and lowpass filters, isolating each stage with buffer circuit.