You can do everything on the EMG signal with MatLab. The best techniques to be used depend on the protocol and the aim of the study.

Together with the important features reported by Priya Natarajan, you could also include muscle activation intervals in order to add quantitative information about "when" a muscle is active within the gait cycle, and precisely estimate the onset and offset time instants of the muscle activation bursts. In this way you can obtain a complete information about muscle coordination during gait.

Features in the frequency domain, obtained from the power spectrum of the EMG, may also help you if you deal with physiological changes (such as muscle fatigue).

I thought for pre processing we can use adaptive filters to remove noise

The DC offset should be removed following adaptive noise cancelling (for both power noise and sometimes ECG too if picked up). In case you don't have bandpass filter in your acquisition circuit design a digital one between 50 to 500 Hz, most usable range is in this... but your design would depend on which muscle you are acquiring the signal from.

For quantification there are various features you could calculate from the data...

Some include

1. IEMG

2. MAV

3. MMAV1

4. MMAV2

5. SIMPLE SQUARE INTEGRAL

6. VARIANCE

7. RMS

8. WAVEFORM LENGTH

9. WILISON AMPLITUDE

10. SSC

11. ZC

12. DIFFERENCE ABSOLUTE MEAN VALUE

13. HISTOGRAM

Good Luck

You can do everything on the EMG signal with MatLab. The best techniques to be used depend on the protocol and the aim of the study.

Together with the important features reported by Priya Natarajan, you could also include muscle activation intervals in order to add quantitative information about "when" a muscle is active within the gait cycle, and precisely estimate the onset and offset time instants of the muscle activation bursts. In this way you can obtain a complete information about muscle coordination during gait.

Features in the frequency domain, obtained from the power spectrum of the EMG, may also help you if you deal with physiological changes (such as muscle fatigue).

Parthiban,

Can you add a bit more information about the question that you are trying to answer? There are many ways to look at EMG, and each method is appropriate for a certain type of analysis.

You are doing analysing Gait, as in walking, right?

The following is an educated guess of what exactly you want and what COULD work.

First of all you need to have the data as voltage, that is, centred around 0 and varying from negative to positive.

Then, you can compute a measure of power using a sliding window. RMS is good, EMG analysts sometimes use "rectification", i.e., sum of absolute values.

Finally, do some analysis of periodicity of the resulting power signal (e.g., spectrogram) to identify the walking cycle.

YMMV

Gnu R is an alternative free Software with plenty analysis option - similar to MatLab. There are many packages, and calculation and plots are not to difficult, that - even I - did manage to write a script, analyzing my EMG data.

We are working on the use of the VRI (voluntary response index) for gait that essentially is a pattern matching to a template comprised of composite neurologically intact subject data (i.e., "normal" data). The deviation from this template can be quantified as the cosine of the angle between the subject data and the normal data at various intervals across the gait cycle, to identify when in the cycle the pattern deviates from normal, and by how much (see Lee DC, Lim HK, McKay WB, Priebe MM, Holmes SA, Sherwood AM: Toward an Objective Interpretation of Surface EMG Patterns: A Voluntary Response Index (VRI) J Electromyogr Kinesiol. 14(3): 379-388, 2004".,

Whatever filter u select: Always remember it should be linear and zero phase distortion.

Best of luck.

For pre-processing and noise cancellation, I propose Morphological Filters. As I did it for ECG signals. Maybe I can help you for this part!

If you are using a specified hardware for EMG acquisition then there is no need for further filtering........for further processing, wavelet analysis is quite good tool..

I would suggest to have a look to the following paper 'MYOELECTRIC CONTROL DEVELOPMENT TOOLBOX' and the Matlab codes in the website http://www.sce.carleton.ca/faculty/chan/matlab.html

They are very useful

All the best,

Hi Parthiban,

As you know, most biosignals are highly non- Gaussian and nonlinear. In addition to the suggested methods from Priya, you can also use nonlinear or chaotic analysis for your time series. But it should be noted that these methods are sensitive to phase changes due to the filtering process for noise cancellation. By the way, you can implement them simply in MATLAB environment.

Bests,

The very first step is to get a readC3D.m function or similar. You can find some tool on the web or ask for it here in this discussion. Maybe we all participant could compare our functions and improve them

Before any analysis, I would recommend to look at raw data, as filters could turn large motion artifacts into emg-like bursts. Tipically, at foot contact motion artifact can arise on the calf muscle signals and on the posterior thigh muscle data. Large artifacts can also appear on the vasti muscle if electrodes collide due to the large skin movements

psd is a useful tool to detect the presence of artifacts and of power line interference, until your not skilled to detect them by eyes in the raw signal

I've been using elliptic high-pass filters, as they are very steep, to remove artifacts. "Filtfilt" instead of "filter" allows for avoiding phase distorsion.

Further computation depends on your goals and on the subjects you are analyzing (healthy subjects or patients). The accuracy in computation depends on the minimum clinically significant difference.

Good Luck

Try Hilbert -Huang Transform ..........

With my group on EMG signal rocessing in Roma (Italy) we are working on EMG signal processing since almost 30 years. To give an answer in a few rows is completely impossible. There is a good book, edited by IEEE Press, from Prof. Parker & Prof Merletti that can help you very much.

dear Parthiban,

I would suggest you to get some important information on the matter from R. Merletti, P.A. Parker "Electromyography" IEEE Press 2004.

Hello Parthiban,

You already have some good answers to your question and I would only restate the last two in a slightly different way: Did you not think about your experiments and measurements before your started doing it? Part of planning an experiment is to decide how you treat the data!

With best wishes

Hello. I don't think I can't help a lot with this since it's been about 30 years since I did a project on EMG processing in my EE degree. At the time, and I don't know if the fundamentals have changed since then, the main problems were that the muscle stimulus is not directly observable, hence the use of EMG signals. The other problem was that the EMG is affected by multiplicative rather then additive noise. The system you are attempting to process signals from is also highly nonlinear except in the case of weak, isometric muscle contractions. Most of the system identification and filtering routines in Matlab are based on additive white noise and linear systems. This said, however, I would still use Matlab due to its flexibility.

The paper I used way back then, round about the time Matlab was developed (!) and one had to use things like Fortran 77, was from 1984 Proc. IEE by Lago and Jones:  "Parameter estimation of system dynamics with modulation-type noise: application to the modelling of the dynamic relationship between the EMG and force transients in muscle". Sorry if this is old hat, but it's nice to reminisce :-)

http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=4642266&filter%3DAND%28p_IS_Number%3A4642264%29

Hie Pathiban,

All answers given above are good contributions but the fundamental principle governing the use of EMG signals is its intended clinical use. Some computations are complex because they determine LIFE and DEATH applications while others are for simple control. If the final use is for disease screening then one has to be thorough with artifacts removal but if it is for controlling devices such as wheel chairs then additional robust algorithms for control can compensate. Try to read articles by D. Luca. they are helpful.