Randy Schmitz

01-28-1997, 07:44 AM

Thanks to everyone who responded to my request. Below is the Original

inquiry and the list of responses.

Once again thanks to all to contributed.

-------------

>I'm currently designing a project that is to look at the decline of power

>production of the quadriceps following a fatigue protocol. The post test

>will be a dynamic knee extension performed on an isotonic dynamometer. My

>question to the forum is can I obtain spectral EMG data from the quads on a

>contraction that will generally take less that 1 second? I've seen several

>studies that have measured spectral alterations in a dynamic contraction of

>the muscle, although these contractions were of a longer duration. (Muro et

>al, Amer. J. Physical Med. 1982 & Moritani et al, Amer. J. Physical Med.

>1988) 2 and 5 sec dynamic contractions respectively.

>

>I know I can obtain the data, but is it a valid measurement?

>

-------------------

To: Randy Schmitz

From: Brian Bergemann

Subject: Re: EMG Spectral analysis

HI Randy: I did a power spectrum analysis in my dissertation, Brian

Bergemann, Penn State University, 1977. It should be on microfiche,

University of Oregon. In the Appendix is a fortran program that does

fourier analysis. If you need a program that will give you the power

spectrum of the frequencies of the EMG go to that dissertation and try it.

The duration of contractions were 3 seconds.

----------------------------

From: "A.Mannion"

Subject: Re: EMG Spectral analysis

To: rjs5s@curry.edschool.virginia.edu

Date: Mon, 20 Jan 1997 18:33:29 +0000 (GMT)

MIME-Version: 1.0

You usually need at least 1024 data points to do a decent FFT but if you

sample very rapidly then this should be possible in less than a second.

But why do you have to use an isotonic contraction? Could you not do a

little isometric contraction for about 10 s at the start and end of the

fatiguing protocol? This would then get round the problems of both the

short sampling time and the nature of the movement being dynamic

(non-stationary signal).

---------------------------

From: "V.BALTZOPOULOS"

Organization: ALSAGER, CREWE+ALSAGER FACULTY, MMU

To: rjs5s@curry.edschool.virginia.edu

Date: Wed, 22 Jan 1997 14:31:51 GMT

Subject: Re: EMG Spectral analysis

Priority: normal

Dear Randy

Assuming that the phenomenon you are recording is contained in your

time interval (say 0.5 s) then you can analyse your data, but the

sampling rate is crucial. If for example your sampling rate is 1000

Hz then you will end up with 500 points which means that the FFT will

calculate 250 harmonics in the frequency range 2-500 Hz. If your fs

is 100 Hz then you will collect 50 points and you will only be able

to see 25 harmonics in the range 2-50 Hz. If the signal contained

higher frequencies then these will appear distorted inside the above

range (leakage). Generally speaking with a high fs you should be ok.

What the time period does affect, however, is the frequency increment

(df). For example with a sampling frequency of 1000 Hz a period of

0.5 s will give a resolution of 2 Hz (i.e. 250 harmonics at

2,4,6..500 Hz) whereas a period of 1 s will give you a resolution of

1 Hz (i.e. 500 harmonics at 1,2,3..500 Hz). I don't think however

that you will want to observe say two distinct phenomena occuring at

frequencies that are less than 1 Hz apart?!

In view of the short record, and contrary to popular advice, you

should not pad your data with zeros if N is not a power of 2 and you

only have access to a radix 2 FFT. Although this does not alter the

SHAPE of the power spectrum if the frequency axis is normalised (i.e.

0-1, where 1 is the maximum frequency), the frequency of the

harmonics (and df) will be affected.

I hope that this is helpful and let me know if you require any

further assistance.

Best wishes

Dr V. (Bill) Baltzopoulos

Biomechanics Group

Manchester Metropolitan University

baltz@mmu.ac.uk

Tel: +44 161 2475659

----------------------------------

From: "Jim Potvin"

Organization: Animal & Poultry Science

To: rjs5s@curry.edschool.virginia.edu

Date: Sun, 26 Jan 1997 09:04:57 EDT

Subject: EMG Spectral analysis

Priority: normal

Hi Randy

With regards to using dynamic EMG to calculate the spectrum, the main

issue is stationarity. It is actually better to use shorter samples

during the dynamic contractions because the amplitude will likely

vary over time and this will adversely affect your stationarity.

Shankar et al (1989) Electro. Clin. Neurophys 73:142-150 is the only

paper I have found that has tested the stationarity during dynamic

contractions. They studied the biceps brachii during flexion

rotations between 40 and 160 deg/s. Their sample durations ranged

from 0.25 to 1.0 seconds and they found that 96% of their data

records were at least weakly stationary. We have a paper coming out in

J. Electromyography and Kinesiology showing that the changes in MnPF

from 250 ms segments recorded continuously throughout a fatiguing

contraction are similar to those recorder isometrically before and

after the trial. I also have a paper coming out in the Journal of

Applied Physiology (this month I think) that looks at the effects of

muscle length and velocity (concentric vs. eccentric) on the EMG

spectrum and amplitude at rest and under fatigued conditions. These

articles and the references cited may be of further help to you.

__________________________________________________ _________

Jim Potvin, PhD

Biomechanics Laboratory

Department of Human Biology and Nutritional Sciences

University of Guelph

Guelph, Ontario, Canada

N1G 2W1

Tel 519-824-4120 ext 4589

Fax 519-763-5902

---------------------------------------

Dear Randy,

In my opinion this can be safely done. Just use a FFT over

the complete EMG burst. The median frequency has of course some

random error (1), but you have to cope with that anyay. With some

others I have done similar experiments on the triceps surae (2,3) and

the vastus lateralis (4). If your reviewer happens to be a signal

processing 'fundamentalist' , he will ask you difficult questions on

stationarity etc. For that case you may try some more complicated

time-frequency transforms (5). Our main problem was that we did

find only very small spectral effects, even in very hard exercise

(2,3,4). As a consequence, I personally am hesitant to couple fatigue

with any EMG frequency effect.

Good Luck anyway,

At Hof

Refs.

1) Hof,A.L. IEEE Trans BME 38: 1077-1088 (1991)

2) Ament, W, et al. J. Electromyogr. Kinesiol. 3: 214-220 (1993)

3) Ament, W. et al. Eur J Appl Physiol 74: 180-186 (1996)

4) Jansen, R. et al Eur J. Appl Physiol (in press)

5) Bonato, P. et al IEEE EMBS Magazine 15/6:102-111 (1996).

At Hof

Department of Medical Physiology

University of Groningen

Bloemsingel 10

NL-9712 KZ GRONINGEN

The Netherlands

Phone: (31) 50 3632645

Fax: (31) 50 3632751

e-mail: a.l.hof@med.rug.nl

---------------------------------------------------

To: Randy Schmitz

From: iallison@info.curtin.edu.au (Garry Allison)

Subject: Re: EMG Spectral analysis

You can always argue the case of validity the trade off will be the

*reliability* of the derived variable.

Good luck.

GTA.

--------------------------------------------

From: "John P. Peach"

To: "'Randy Schmitz'"

Subject: RE: EMG Spectral analysis

Date: Sat, 18 Jan 1997 13:17:06 -0500

MIME-Version: 1.0

Hello;

In order to use spectral parameters of EMG you have to take your

EMG and perform a fast fourier transformation (FFT) on it. The harmonics

of the FFT (or if you like, its resolution) is a function of the sample

rate (SR) and the number of samples collected (N). There is restrictions

on the value of N. N must be a value of 2^n, so you are looking at numbers

like, 2,4,8,16,32,64,128,256,512,1024,2048 etc... so you will have to

choose your sampling rate (for a given collection time) to optimize the

value of N.

Now for the real problem. The harmonics are determined by:

SR / N = harmonic

this can also be expressed as

SR/ (Time * SR) = harmonic

this simplifies to;

1/Time = harmonic

This means that the FFT harmonic (i.e. the resolution of the FFT)

is a function of the collection time. So if you are looking at the fatigue

rate of the quads the smallest change you can detect is 1 Hz if the

movement takes a full second. With out knowing your protocol I cannot

really comment on how good this is. My gut feeling is that it is going to

be problematic.

The use of MPF as a measure of fatigue is only valid if the

contraction is isometric. If the contraction is dynamic the motor pool

changes and this violates one of the basic assumptions of the MPF

calculation.

I just submitted a paper to the Journal of Electromyography and

Kinesiology looking at the reliability of the MedianPF decay rate

calculation. There is some problems with it in that it is quite variable.

On the back extensor muscles the intercept values are reliable if you have

5 or more samples and the slope (decay rate) is not reliable unless you

have several hundred measures.

John

----------------------------------------------------------

Randy J. Schmitz

Ph.D. student in Sports Medicine

University of Virginia

rjs5s@virginia.edu

(804) 984-5243

inquiry and the list of responses.

Once again thanks to all to contributed.

-------------

>I'm currently designing a project that is to look at the decline of power

>production of the quadriceps following a fatigue protocol. The post test

>will be a dynamic knee extension performed on an isotonic dynamometer. My

>question to the forum is can I obtain spectral EMG data from the quads on a

>contraction that will generally take less that 1 second? I've seen several

>studies that have measured spectral alterations in a dynamic contraction of

>the muscle, although these contractions were of a longer duration. (Muro et

>al, Amer. J. Physical Med. 1982 & Moritani et al, Amer. J. Physical Med.

>1988) 2 and 5 sec dynamic contractions respectively.

>

>I know I can obtain the data, but is it a valid measurement?

>

-------------------

To: Randy Schmitz

From: Brian Bergemann

Subject: Re: EMG Spectral analysis

HI Randy: I did a power spectrum analysis in my dissertation, Brian

Bergemann, Penn State University, 1977. It should be on microfiche,

University of Oregon. In the Appendix is a fortran program that does

fourier analysis. If you need a program that will give you the power

spectrum of the frequencies of the EMG go to that dissertation and try it.

The duration of contractions were 3 seconds.

----------------------------

From: "A.Mannion"

Subject: Re: EMG Spectral analysis

To: rjs5s@curry.edschool.virginia.edu

Date: Mon, 20 Jan 1997 18:33:29 +0000 (GMT)

MIME-Version: 1.0

You usually need at least 1024 data points to do a decent FFT but if you

sample very rapidly then this should be possible in less than a second.

But why do you have to use an isotonic contraction? Could you not do a

little isometric contraction for about 10 s at the start and end of the

fatiguing protocol? This would then get round the problems of both the

short sampling time and the nature of the movement being dynamic

(non-stationary signal).

---------------------------

From: "V.BALTZOPOULOS"

Organization: ALSAGER, CREWE+ALSAGER FACULTY, MMU

To: rjs5s@curry.edschool.virginia.edu

Date: Wed, 22 Jan 1997 14:31:51 GMT

Subject: Re: EMG Spectral analysis

Priority: normal

Dear Randy

Assuming that the phenomenon you are recording is contained in your

time interval (say 0.5 s) then you can analyse your data, but the

sampling rate is crucial. If for example your sampling rate is 1000

Hz then you will end up with 500 points which means that the FFT will

calculate 250 harmonics in the frequency range 2-500 Hz. If your fs

is 100 Hz then you will collect 50 points and you will only be able

to see 25 harmonics in the range 2-50 Hz. If the signal contained

higher frequencies then these will appear distorted inside the above

range (leakage). Generally speaking with a high fs you should be ok.

What the time period does affect, however, is the frequency increment

(df). For example with a sampling frequency of 1000 Hz a period of

0.5 s will give a resolution of 2 Hz (i.e. 250 harmonics at

2,4,6..500 Hz) whereas a period of 1 s will give you a resolution of

1 Hz (i.e. 500 harmonics at 1,2,3..500 Hz). I don't think however

that you will want to observe say two distinct phenomena occuring at

frequencies that are less than 1 Hz apart?!

In view of the short record, and contrary to popular advice, you

should not pad your data with zeros if N is not a power of 2 and you

only have access to a radix 2 FFT. Although this does not alter the

SHAPE of the power spectrum if the frequency axis is normalised (i.e.

0-1, where 1 is the maximum frequency), the frequency of the

harmonics (and df) will be affected.

I hope that this is helpful and let me know if you require any

further assistance.

Best wishes

Dr V. (Bill) Baltzopoulos

Biomechanics Group

Manchester Metropolitan University

baltz@mmu.ac.uk

Tel: +44 161 2475659

----------------------------------

From: "Jim Potvin"

Organization: Animal & Poultry Science

To: rjs5s@curry.edschool.virginia.edu

Date: Sun, 26 Jan 1997 09:04:57 EDT

Subject: EMG Spectral analysis

Priority: normal

Hi Randy

With regards to using dynamic EMG to calculate the spectrum, the main

issue is stationarity. It is actually better to use shorter samples

during the dynamic contractions because the amplitude will likely

vary over time and this will adversely affect your stationarity.

Shankar et al (1989) Electro. Clin. Neurophys 73:142-150 is the only

paper I have found that has tested the stationarity during dynamic

contractions. They studied the biceps brachii during flexion

rotations between 40 and 160 deg/s. Their sample durations ranged

from 0.25 to 1.0 seconds and they found that 96% of their data

records were at least weakly stationary. We have a paper coming out in

J. Electromyography and Kinesiology showing that the changes in MnPF

from 250 ms segments recorded continuously throughout a fatiguing

contraction are similar to those recorder isometrically before and

after the trial. I also have a paper coming out in the Journal of

Applied Physiology (this month I think) that looks at the effects of

muscle length and velocity (concentric vs. eccentric) on the EMG

spectrum and amplitude at rest and under fatigued conditions. These

articles and the references cited may be of further help to you.

__________________________________________________ _________

Jim Potvin, PhD

Biomechanics Laboratory

Department of Human Biology and Nutritional Sciences

University of Guelph

Guelph, Ontario, Canada

N1G 2W1

Tel 519-824-4120 ext 4589

Fax 519-763-5902

---------------------------------------

Dear Randy,

In my opinion this can be safely done. Just use a FFT over

the complete EMG burst. The median frequency has of course some

random error (1), but you have to cope with that anyay. With some

others I have done similar experiments on the triceps surae (2,3) and

the vastus lateralis (4). If your reviewer happens to be a signal

processing 'fundamentalist' , he will ask you difficult questions on

stationarity etc. For that case you may try some more complicated

time-frequency transforms (5). Our main problem was that we did

find only very small spectral effects, even in very hard exercise

(2,3,4). As a consequence, I personally am hesitant to couple fatigue

with any EMG frequency effect.

Good Luck anyway,

At Hof

Refs.

1) Hof,A.L. IEEE Trans BME 38: 1077-1088 (1991)

2) Ament, W, et al. J. Electromyogr. Kinesiol. 3: 214-220 (1993)

3) Ament, W. et al. Eur J Appl Physiol 74: 180-186 (1996)

4) Jansen, R. et al Eur J. Appl Physiol (in press)

5) Bonato, P. et al IEEE EMBS Magazine 15/6:102-111 (1996).

At Hof

Department of Medical Physiology

University of Groningen

Bloemsingel 10

NL-9712 KZ GRONINGEN

The Netherlands

Phone: (31) 50 3632645

Fax: (31) 50 3632751

e-mail: a.l.hof@med.rug.nl

---------------------------------------------------

To: Randy Schmitz

From: iallison@info.curtin.edu.au (Garry Allison)

Subject: Re: EMG Spectral analysis

You can always argue the case of validity the trade off will be the

*reliability* of the derived variable.

Good luck.

GTA.

--------------------------------------------

From: "John P. Peach"

To: "'Randy Schmitz'"

Subject: RE: EMG Spectral analysis

Date: Sat, 18 Jan 1997 13:17:06 -0500

MIME-Version: 1.0

Hello;

In order to use spectral parameters of EMG you have to take your

EMG and perform a fast fourier transformation (FFT) on it. The harmonics

of the FFT (or if you like, its resolution) is a function of the sample

rate (SR) and the number of samples collected (N). There is restrictions

on the value of N. N must be a value of 2^n, so you are looking at numbers

like, 2,4,8,16,32,64,128,256,512,1024,2048 etc... so you will have to

choose your sampling rate (for a given collection time) to optimize the

value of N.

Now for the real problem. The harmonics are determined by:

SR / N = harmonic

this can also be expressed as

SR/ (Time * SR) = harmonic

this simplifies to;

1/Time = harmonic

This means that the FFT harmonic (i.e. the resolution of the FFT)

is a function of the collection time. So if you are looking at the fatigue

rate of the quads the smallest change you can detect is 1 Hz if the

movement takes a full second. With out knowing your protocol I cannot

really comment on how good this is. My gut feeling is that it is going to

be problematic.

The use of MPF as a measure of fatigue is only valid if the

contraction is isometric. If the contraction is dynamic the motor pool

changes and this violates one of the basic assumptions of the MPF

calculation.

I just submitted a paper to the Journal of Electromyography and

Kinesiology looking at the reliability of the MedianPF decay rate

calculation. There is some problems with it in that it is quite variable.

On the back extensor muscles the intercept values are reliable if you have

5 or more samples and the slope (decay rate) is not reliable unless you

have several hundred measures.

John

----------------------------------------------------------

Randy J. Schmitz

Ph.D. student in Sports Medicine

University of Virginia

rjs5s@virginia.edu

(804) 984-5243