View Full Version : Summary of EMG averaging replies

Pamela Joy Wise
08-24-1998, 11:44 AM
Thanks to everyone who replied so quickly and thoroughly to my question.
To summarize, most suggested I normalize to percent gait cycle and
after the trials after they have been processed,thereby elimating any
problems with sampling rates altogether.


Hello. I am a first year graduate student currently working on software
for analyzing on and off times for EMG signals. I see that this has been
a recent topic on the biomech list, but I have a question that has yet to
be answered. I am collecting signals from 10 different muscles over the
gait cycle, for 5 trials per subject. I would like to present my
resulting on/off signal as an average for the 5 trials. To do this, I
think it is necessary to normalize the lengths of the trials (they are
all different lengths, due to natural variation in the gait cycle) so
that I can average on a point by point basis. However, if I do this,
won't I be changing the sampling frequency and therefore invalidating any
kind of averaging? I have seen kinematic data presented in the
literature as an average of x number of trials
+/- a certain standard deviation and assume the same sort of problem has
been encountered and solved in order to present the data this way. I
would very much appreciate any insight or references to literature that
might help me solve this problem. I will post a summary of replies in a
few days. Thank you.

Hi Pam,

Here are three published papers you might find applicable to your

Computer Algorithms to Characterize Individual Subject EMG Profiles During
Gait - Ross A. Bogey, DO, Lee A. Barnes, MA, Jacquelin Perry, MD Arch Phys
Med Rehabil Vol 73, September 1992, Pages 835-841

A Computer Algorithm for Defining the Group Electromyographic Profile From
Individual Gait Profiles - Ross A. Bogey, Do, Lee A. Barnes, MA, Jacquelin
Perry, MD Arch Phys Med Rehabil Vol 74, March 1993, Pages 286-291

The Rancho EMG analyzer: a computerized system for gait analysis - J.
Perry, E.L. Bontrager, R.A. Bogey, J.K. Gronley, L.A. Barnes J. Biomed.
Eng. 1993, Vol. 15, November 1993, Pages 487-496

-Lee Barnes

Most either standardize to % gait cycle, or break it up as % stance phase
% swing phase.

Gregory Rash
Director, Gait & Biomechanics Lab
Frazier Rehab Center


Isn't it possible to eliminate the problem by presenting your results in
terms of per cent of gait cycle, rather than as specific time
In other words, soleus turned on at blank % of gait and off at blank %.
Calculate results for each trial independently, then average across
Are you using heel strike of the first foot to toe off of the next as one

Good luck

Lou Rosenfeld


Very quickly (off the top of my head)... Could you not just normalize the
time to pecentage of gait cycle? That would probably be sufficient for on
and off times.

I do not think that you would have to do anything to sampling rates if you
approached it from this angle.

Good luck, feel free to drop me a line to discuss this further.


Stephen J. Kinzey, Ph.D.
Assistant Professor / Director of Biomechanics Laboratory
The University of Mississippi
Department of ESLM
University, MS 38677
e-mail: skinzey@olemiss.edu
office: (601) 232 - 5540
fax: (601) 232 - 5525

Dear Pam:

I did almost the same thing, excpet for the upper limb during reaching
motions. The EMG signals were linear envelope at detected 10 Hz with
Butterworth Filter (Winter 1990, page 280). I then determined the onset
end of the EMG relative to the hand kinematics for each trial. To look at
average patterns, I then normalized to each waveform to the duration of
movement, which was slightly different from trial to trial. This is
possible based on the work of Shavi and Green (1983) who suggest that
interpolation retains the frequency content of the signal.

1. Gabriel, D.A. (1997). Shoulder and elbow muscle activity in
goal-directed arm movements. Experimental Brain Research, 116, 359-366.

2. Shavi, R. & Green, N. (1983) Ensemble averaging of locomotor
electromyographic patterns using interpolation. Medical & Biological
Engineering & Computing, 21, 573-578.

Best Wishes,


Since you are interested in activity relative to phase rather than latency
from some zero point, and assuming the conditions of each trial are
sufficiently similar to each other (and your description suggests that
are), then it is appropriate to normalize cycle duration for the purposes
of averaging, at least in principle. Another consideration is to ensure
that you obtain an accurate measure of signal amplitude at each point in
time. In the software package that our company produces, one can rectify
and smooth the EMG signals, using either a linear or RMS algorithm prior
averaging. The averaging operation employs a linear interpolation
to estimate amplitude at percentage intervals across each defined trial.
The number of percentage intervals is user-selectable. Standard deviation
at each point is also calculated. Averaged results can be displayed
graphically, subjected to a burst analysis to determine on and off times
for each muscle, and the results of the average and/or burst analysis can
be exported in the form of ASCII files for further statistical analysis,
desired. Let me know if I can be of further assistance.

RUN Technologies

Dear Pamela Joy Wise,
You are correct that time normalization will be required.
Typically one normalizes to either 100% of stance or 100% of stride
(stance and swing). A simple linear interpolation algorithm is often
adequate. You must decide how many data points you desire in your
output curves which is dependent upon the time resolution you need in
reporting your on/off EMG data.
Best of luck,

Howard J. Hillstrom, Ph.D.
Director, Gait Study Center
Temple University School of Podiatric Medicine
Eighth and Race St.
Philadelphia, PA 19107
phone: (215) 625-5366
fax: (215) 629-1622
email: hhillstrom@pcpm.edu


Have a look at Banato et al " A statistical method for the measurement of
muscle activation intervals...", IEEE Trans in Biomed Engg March 1998 vol
45 (3), pp 287-299.


Ed Biden

You could either resample, choosing poitnats pre-defined intervals of your
normalized time frame and interpolating between raw data, using a fitting
technique like Woltring's, or a far better solution in my opinion is to
transform your data into the frequency domain via FFT and average the
harmonic coefficients.

__________________________________________________ _______________

Pamela Joy Wise _/ _/ _/ _/
Biomedical Engineering Graduate Student _/_/ _/_/ _/ _/
Marquette University _/ _/_/ _/ _/ _/
1504 W. Kilbourn Ave. Apt. D _/ _/ _/ _/ _/
Milwaukee, WI 53233 _/ _/ _/ _/
(414)344-6907 _/ _/ _/_/ _/
__________________________________________________ ______________

To unsubscribe send UNSUBSCRIBE BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://www.bme.ccf.org/isb/biomch-l