Robert Newton

09-09-1994, 09:33 PM

Thankyou to those who responded to my request for information on

frequency spectrum analysis of EMG signals. Following is a list of

references and a summary of the replies.

Regards

Robert Newton

================================================== =====================

De Luca, C.J. Physiology and mahmatics of myoelectric signals. IEEE

Transactions on Biomedical Engineering, Vol. 26(6):313-325, 1979.

Desmedt, J.E. New concepts of the motor unit, Neuromuscular Disorders

Electromyographical kinesiology. New Developments in

Electromyography and Clinical Neurophysiology. Vol. 1.

S. Karger. 1973

Kwatny, E., Thomas, D.H. and H.G. Kwatney. An application of signal

processing techniques to the study of myoelectric signals.

IEEE Transactions on Biomedical Engineering,

Vol. 17(4):303-313, 1970.

Stulen, F.B. and De Luca, C.J. Frequency parameters of the myoelectric

signal as a measure of muscle conduction velocity. IEEE

Transactions on Biomedical Engineering, Vol. 28(7):515-523, 1981.

================================================== =======================

From: Oyvind Stavdahl

Message-Id:

To: run1@psu.edu

Subject: Re: Frequency Analysis of EMG using FFT

Dear Robert,

The questions asked in your BIOMCH-L posting are interesting to me,

because I am just plowing into the world of EMG myself. I am studying

EMG as a means of controlling powered hand prostheses. I have just

realized that in order to obtain really good (smooth) estmates of

contraction force I have to utilize the information "hidden" in the

frequency spectrum of the signals (in contrast to only using amplitude).

If your work is somehow related to mine, or if you have any good

litterature references on the relationship between contraction force

and the correspopnding EMG spectrum I would greatly appreciate hearing

fron you. (I already have _some_ references.)

I will try to answer your questions:

1. The Fourier transform decomposes the time series into a number of

sine components that are equidistant with respect to frequency

(linearly distributed along the frequency axis). I therefore believe that

division by frequency is unnecessary. (I don't KNOW this - it just seems

to me that this is the case. By the way you will probably receive other

replies that will give you facts on this issue.) I think the division -by-

frequency issue might have something to do with conversion between linear

and logarithmic frequencies.

2. If you use your raw time series as input data to your FFT sofware, it

will produce the magnitude - in the _amplitude_ sense - of the 512 sine

components. To obtain the power spectrum you have to square each of these

values. The power spectrum (power spectral density) can also be obtained

by Fourier transforming the autocorrelation function of your time series.

3. I think you should always use a Window before FFT processing, as this

reduces the expected error in the FFT output - regardless of the signal's

frequency content. (This error is a consequence

of your time series always being a finite part of an ideally "everlasting"

signal - you sample this signal only in a limited time "window", and the

FFT acts as if all samples outside the window were equal to zero. The Rectangular

window does _nothing_ to your time series, while the Hamming, Hanning etc.

force the sampled values close to the ends of the series to approach zero,

reducing the "step" from nonzero samples to the (nonexisting) zero samples

outside the window.)

I won't go into more details about this, but I think most textbooks on Numerical

Signal Processing would give you the answers you request. The none-rectangular

windows have very similar effects, and I find it diffycult to recommend one

more than the others.

I hope you can use this.

Good luck in your further work!

Best regards,

Oyvind Stavdahl

From: "Claudia Ranniger"

Subject: Re: Frequency Analysis of EM

To: run1@psu.edu

Reply to: RE>Frequency Analysis of EMG u

Robert-

A discussion of mean vs median frequency, and how to calculate them, is

found in Stulen and DeLuca, "Frequency parameters of the Myoelectric Signal as

a Measure of Muscle Conduction Velocity," IEE Transaction of Biomed

Engineering, Vol 28, #7, July 1981.

good luck

-Claudia

================================================== ===================

Robert Newton Email: run1@psu.edu

Center for Sports Medicine Telephone: Int+ 1 814 865 7107

The Pennsylvania State University Facsimile: Int+ 1 814 865 7077

117 Ann Building, University Park, PA 16802

================================================== ===================

frequency spectrum analysis of EMG signals. Following is a list of

references and a summary of the replies.

Regards

Robert Newton

================================================== =====================

De Luca, C.J. Physiology and mahmatics of myoelectric signals. IEEE

Transactions on Biomedical Engineering, Vol. 26(6):313-325, 1979.

Desmedt, J.E. New concepts of the motor unit, Neuromuscular Disorders

Electromyographical kinesiology. New Developments in

Electromyography and Clinical Neurophysiology. Vol. 1.

S. Karger. 1973

Kwatny, E., Thomas, D.H. and H.G. Kwatney. An application of signal

processing techniques to the study of myoelectric signals.

IEEE Transactions on Biomedical Engineering,

Vol. 17(4):303-313, 1970.

Stulen, F.B. and De Luca, C.J. Frequency parameters of the myoelectric

signal as a measure of muscle conduction velocity. IEEE

Transactions on Biomedical Engineering, Vol. 28(7):515-523, 1981.

================================================== =======================

From: Oyvind Stavdahl

Message-Id:

To: run1@psu.edu

Subject: Re: Frequency Analysis of EMG using FFT

Dear Robert,

The questions asked in your BIOMCH-L posting are interesting to me,

because I am just plowing into the world of EMG myself. I am studying

EMG as a means of controlling powered hand prostheses. I have just

realized that in order to obtain really good (smooth) estmates of

contraction force I have to utilize the information "hidden" in the

frequency spectrum of the signals (in contrast to only using amplitude).

If your work is somehow related to mine, or if you have any good

litterature references on the relationship between contraction force

and the correspopnding EMG spectrum I would greatly appreciate hearing

fron you. (I already have _some_ references.)

I will try to answer your questions:

1. The Fourier transform decomposes the time series into a number of

sine components that are equidistant with respect to frequency

(linearly distributed along the frequency axis). I therefore believe that

division by frequency is unnecessary. (I don't KNOW this - it just seems

to me that this is the case. By the way you will probably receive other

replies that will give you facts on this issue.) I think the division -by-

frequency issue might have something to do with conversion between linear

and logarithmic frequencies.

2. If you use your raw time series as input data to your FFT sofware, it

will produce the magnitude - in the _amplitude_ sense - of the 512 sine

components. To obtain the power spectrum you have to square each of these

values. The power spectrum (power spectral density) can also be obtained

by Fourier transforming the autocorrelation function of your time series.

3. I think you should always use a Window before FFT processing, as this

reduces the expected error in the FFT output - regardless of the signal's

frequency content. (This error is a consequence

of your time series always being a finite part of an ideally "everlasting"

signal - you sample this signal only in a limited time "window", and the

FFT acts as if all samples outside the window were equal to zero. The Rectangular

window does _nothing_ to your time series, while the Hamming, Hanning etc.

force the sampled values close to the ends of the series to approach zero,

reducing the "step" from nonzero samples to the (nonexisting) zero samples

outside the window.)

I won't go into more details about this, but I think most textbooks on Numerical

Signal Processing would give you the answers you request. The none-rectangular

windows have very similar effects, and I find it diffycult to recommend one

more than the others.

I hope you can use this.

Good luck in your further work!

Best regards,

Oyvind Stavdahl

From: "Claudia Ranniger"

Subject: Re: Frequency Analysis of EM

To: run1@psu.edu

Reply to: RE>Frequency Analysis of EMG u

Robert-

A discussion of mean vs median frequency, and how to calculate them, is

found in Stulen and DeLuca, "Frequency parameters of the Myoelectric Signal as

a Measure of Muscle Conduction Velocity," IEE Transaction of Biomed

Engineering, Vol 28, #7, July 1981.

good luck

-Claudia

================================================== ===================

Robert Newton Email: run1@psu.edu

Center for Sports Medicine Telephone: Int+ 1 814 865 7107

The Pennsylvania State University Facsimile: Int+ 1 814 865 7077

117 Ann Building, University Park, PA 16802

================================================== ===================