How to choose the "correct" low pass cut-off frequency for EMG signal?

[François Hug]

Dear Biomch-L-ers,

For studies focusing on muscle coordination, the choice of the low-pass filter for the EMG linear envelope is crucial for a proper neurophysiological interpretation of the results. A wide variety of cut-off frequencies has been used in the literature, from 3 Hz for gait analysis (Winter and Yack, 1987 - Electroencephalogr Clin Neurophysiol ) to 40 Hz (Guidetti et al., 1996 - JEK), that lead to very different EMG waveforms and thus different interpretations.
I guess that the that the low pass cut-off frequency has to be chosen that there is a neurophysiological basis for interpreting the EMG patterns. Thus, what is the “ideal” filter and, thus, the “ideal” waveform? The response is not obvious for me.

Does anyone know some studies about this issue ? What is your opinion?

Thanks in adavance for you help.

My best,

François

[Jesse W. Young]

Re: How to choose the "correct" low pass cut-off frequency for EMG signal?

Why not try the residual analysis procedure recommended by in Winter's "Biomechanics and Motor Control of Human Movement" text? In my 1990 edition, it's described on page 41 - Section 2.5.5.3 "Choice of Cutoff Frequency - Residual Analysis".

take care,
Jesse Young

Dear Biomch-L-ers,

For studies focusing on muscle coordination, the choice of the low-pass filter for the EMG linear envelope is crucial for a proper neurophysiological interpretation of the results. A wide variety of cut-off frequencies has been used in the literature, from 3 Hz for gait analysis (Winter and Yack, 1987 - Electroencephalogr Clin Neurophysiol ) to 40 Hz (Guidetti et al., 1996 - JEK), that lead to very different EMG waveforms and thus different interpretations.
I guess that the that the low pass cut-off frequency has to be chosen that there is a neurophysiological basis for interpreting the EMG patterns. Thus, what is the “ideal” filter and, thus, the “ideal” waveform? The response is not obvious for me.

Does anyone know some studies about this issue ? What is your opinion?

Thanks in adavance for you help.

My best,

François

[Zahra Safaeepour]

Re: How to choose the "correct" low pass cut-off frequency for EMG signal?

Hi
It is also possible to find out the main frequencies of a signal with power analysis( power spectrum) and select the cut off frequency based on this. By this method the main frequencies in which signal has power wont be missed.In Matlab there is a GUI (sptool) which is simple to use.
About filter, Moving average method(low frequncy filter) is useful.
Regards
Zahra

[David Gabriel]

Re: How to choose the "correct" low pass cut-off frequency for EMG signal?

Greetings.

Please forgive the shameless self-promotion but this topic is exactly why Gary Kamen and I wrote Essentials of Electromyography. The idea was to address all these intermediate level topics and questions that graduate students have. We cover this particular topic in great detail.

http://www.humankinetics.com/products/all-products/essentials-of-electromyograhy

Best Wishes,

David Gabriel

[Ton van den Bogert]

Re: How to choose the "correct" low pass cut-off frequency for EMG signal?

Hi David,

No problem with the self-promotion, by all means cite the book when these questions come up.

But.... Could you also post a brief answer to the question? Then you can cite the book as the source and for further reading on the topic.

Ton van den Bogert, Biomch-L co-moderator

[David Gabriel]

Re: How to choose the "correct" low pass cut-off frequency for EMG signal?

Thanks, Ton.

OK.

The correct choice for a low-pass cut-off frequency for linear envelop "depends on whom you ask" and what your data analysis goals are. If, for example, you are looking at the EMG-to-force relationship as part of a musculoskeletal modelling adventure then the recommendations by Winter (pick your edition, 1-4) are perfectly correct. However, if you are trying to determine EMG onset to examine muscle timing, then other factors must be considered. This includes balancing the Type I and Type II errors associated with a purely automated approach through the use of a computer algorithm approach. In this case, a 50 Hz low-pass cut-off is recommended by Walter (1984), in conjunction with two threshold criteria (duration and amplitude) for the EMG signal.

In the book, three different filtering options are compared to illustrate the various considerations: band-passed (10-500 Hz); a low-pass at 3.1 Hz; and 50 Hz. I will point out the Figure 4.8 on page 115 should show that the 3.1 Hz condition prematurely detects EMG onset before the other two filtering conditions. However, I do not think that it comes across well in the figure and will be revised in the future.

Best Wishes,

David

[François Hug]

Re: How to choose the "correct" low pass cut-off frequency for EMG signal?

Dear all,

Thanks a lot for your replies and help.
Regards,

François

[Garry T Allison]

Re: How to choose the "correct" low pass cut-off frequency for EMG signal?

Re: filtering and onsets. There are two stages in the filtering both has impact on Onset detection. The first - data conditioning- can alter the distribution of the data and impact on algorithm (& visual) detection. A second filtering process that also impacts on onsets is the creation of the linear envelop (LE). The generic class of onset detection protocols (Shewhart models that use the baseline Mean and SD distribution) are sometimes modified with additional requirments such as sustained above the threshold for x ms - this is the same as an additional LP filter. As a consequence - the ability to examine the literature to see which onset algorithm is optimal can rarely be compared since there are such a wide variety of combinations - 3SD above the mean depends on how the original data set was obtained and filtered. This is particularly true for the signal to noise ratio (between the baseline and onset). In the end the onset needs to be reliable (as are automatic algorithms) but also valid - (independent visual agreements). Both are often required to establish robust evidence - the latter has the greatest potential for bias within any specific laboratory. [Allison J Electromyogr Kinesiol. 2003 Jun;13(3):209-16. Trunk muscle onset detection technique for EMG signals with ECG artefact.]
regards
Garry.