Design and Development of a Practical Muscle Fatigue Monitor

Bahattin Karagzolu, Waleed H. Sindi, Ahmed A. Al-Omari

Department of Electrical and Computer Engineering
Faculty of Engineering, King Abdulaziz University, PO Box: 80204, Jeddah, 21589, Saudi Arabia

Abstract The purpose of the study is to develop a

method to monitor the patient during the exercise
and advice him to stop as he is approaching the
fatigue condition. The median frequency of the EMG
signal has been recognized as the best indicator of the
muscular fatigue during the voluntary contraction of
skeletal muscles.
The EMG signal is picked up from the body using a
pair of biopotential electrodes. A special preamplifier and a
band-pass filer are designed to process the raw signal. The
signal is applied into a personal computer (PC) via the
sound card. The power spectrum of the signal is calculated
using the Fast Fourier Transform (FFT) algorithm in
MATLAB. The median frequency is calculated and plotted
against time during the exercise. Experiments are designed
to test the procedure in the lab. The fluctuations in the
median frequency are eliminated using averaging over
time. The tests indicate a drop in the mean value of the
median frequency before the test subject feels the fatigue.
Index Terms Median frequency detection, Muscle
fatigue, MATLAB, FFT, Sound card

I. INTRODUCTION
The human body is a dynamic mechanism which is
under continuous motion. The extent of the activity
ranges from unintentional peristalic movements of
smooth muscles to severe physical exertions that require
generation of huge muscular forces. All these actions are
carried out by small motors inside the body distributed
around all organs and limbs. They produce mechanical
energy by burning the fuel carried to them through
special vessels. These motor organs of the body are the
muscles. For neuromuscular disorders, the action of the
muscle is partly replaced by others in the vicinity of it
until it recovers. If the motor nerve of a muscle is
interrupted somehow, then the muscle does not work
and decays very quickly if the interruption continues for
a while. This is called the paralysis of the muscle and
the recovery involves a programmed physical exercise.
The dose and duration of the exercise is set by a
physician or a physiotherapist. The exercise may yield a
temporary loss of power that is called the muscular
fatigue. At this stage, the patient may feel pain and he
might be reluctant to continue the exercise. Hence, the
dose and duration must be determined very carefully so
that the patient shall stop exercising before he feels the
fatigue. The current study concerns with development of
a method to monitor the patient during the exercise and
advice him to stop as he is approaching the fatigue
condition. The onset of muscle fatigue will be detected

and exercise will be stopped before the patient feels the

fatigue.
The paper starts with definition of the muscular
fatigue, its symptoms and causes, and realistic design
constraints for developing a method to study it. It then
presents the methodology of the project including design
requirements, feasibility discussion and implementation.
Finally, it presents test results in lab and filed tests, the
discussion of results and conclusions reached.
II. MUSCLE FATIGUE
The muscle fatigue is an exercise-induced reduction in
maximal voluntary muscle force. It has two main
components:
x The inability to produce force due to a decline
in the sensitivity of troponin to calcium; the
strength of the muscle has been exhausted and
the muscle no longer responses to signals from
the brain.
x Fatigue due to a buildup of lactic acid.
Fatigue from lactic acid usually occurs from aerobic
activities and the bodys need for oxygen. In this project,
the fatigue is taken as the first component from the
physiologists view that is the decline in the ability to
produce force due to the decrease in the troponins
sensitivity to calcium.
Brainstorming on muscle fatigue detection yields nontechnical and technical solutions.
Non-Technical Solutions
The intensity and duration of exercise can be
controlled by the patient himself, by a physiotherapist or
by a physician. This traditional technique does not need
any effort to achieve to the required results. The patient
needs to exercise to the maximal possible level to
improve his muscular strength. However, monitoring of
the activity and control of its duration based on personal
judgments may not yield objective assessments. At the
same time, patients freedom is limited and social
environment is interfered leading into the white-coat
syndrome. Hence, we eliminate this alternative from the
beginning and dive into the technical solutions.

Technical Solutions
There are many indicators that we can use for
detecting the muscular fatigue during an exercise. They
can be briefed as follows:
x Physiological tremor:
All external body organs have small vibrations whose
intensities increase with the muscular fatigue. They can
be detected and used as an indicator. The technique
requires external transducers and sophisticated signal
processors.
x Drop in the level of physical activity:
The speed of muscular contraction is expected to
decrease with the fatigue. The speed can be measured
using accelerometers and the fatigue condition can be
established via signal processing.
x The electromyogram (EMG):
During muscular fatigue, there are electrochemical
changes in the muscle that may alter some
characteristics of the electromyogram waveform. The
change occurs both in the amplitude and frequency
characteristics of the signal. EMG can be detected using
surface electrodes and processed to determine the onset
of fatigue. As will be illustrated later, there are good
indications that physiological fatigue can be detected
before the physical one via the EMG signal processing.
x Ultrasound imaging:
Detecting, ultrasound imaging can inherently provide
the morphological information of individual muscle,
thus the architectural changes of muscles during fatigue
can be obtained.
A decision analysis yields the detection of muscular
fatigue via EMG as the best solution.
III. MEASUREMENT AND PROCESSING OF EMG
The electromyogram (EMG) is an electrical signal
generated by the muscles during their contractions. It
shows however, an interference pattern of the action
potentials generated by muscle fibers underneath the
measuring site. As the surface electromyography
(SEMG) can be used to estimate the features of
neuromuscular activations associated with muscle
contractions, it has been widely employed as an
objective
tool
to
evaluate
muscle
fatigue.
Electromyogram (EMG) can be measured by applying
conductive elements or electrodes to the skin surface, or
invasively within the muscle. The surface EMG is the
more common method of measurement, since it is noninvasive and can be conducted by personnel other than
medics or paramedics, with minimal risk to the subject.
The amplitude of the signal in the surface EMG (sEMG)
ranges from micro volts to a few milli volts depending
upon the technique of measurement. The frequency
range covers the lower audio range that is from 20 Hz to
500 Hz. The amplitude, time, and frequency domain

properties of the sEMG signal depends upon factors

such as
x the timing and intensity of muscle contraction
x the distance of the electrode from the active
muscle area
x the properties of the overlying tissue (e.g.
thickness of overlying skin and adipose tissue)
x the electrode and amplifier properties
x the quality of contact between the electrode and
the skin
Measuring and accurately representing the sEMG
signal depends on the properties of the electrodes and
their interaction with the skin, amplifier design, and the
conversion and subsequent storage of the EMG signal
from analog to digital form (A/D conversion). The
quality of the measured EMG is often described by sinal
to noise ratio (SNR) which is the ratio between the
measured EMG signal and unwanted noise contributions
from the environment. The goal is to maximize the
amplitude of the signal while minimizing the noise,
assuming that the amplifier design and process of A/D
conversion exceed acceptable standards. Fig. 1 shows a
raw EMG signal taken from the biceps during
interrupted physical exercise.

Fig. 1. The raw EMG recording of 3 contractions bursts of

the biceps.

Sources of Noise in the EMG Signal

Before we can develop strategies to eliminate unwanted
noise we must understand what the sources of noise are.
There are three types of noise as the ambient noise,
transducer noise and electronic noise.
1- Ambient noise
The ambient noise is generated by electromagnetic
devices such as computers, force plates, power lines etc.
Essentially any device that is plugged into the wall A/C
(Alternating Current) outlet emits the ambient noise.
This noise has a wide range of frequency components,
however, the dominant frequency component is 50 Hz or
60 Hz, corresponding to the frequency of the A/C power
supply (i.e. wall outlet). It can be reduced by:
- Turning off the fluorescent light
- Unplugging unused instrument
- Using shielded cables or at least twisted wires
for input connections.
2- Transducer noise
Transducer noise is generated at the electrode skin
interface. Electrodes serve to convert the ionic currents

generated in muscles into an electronic current that can

be manipulated in electronic circuits and stored in either
analog or digital form as a voltage potential. There are
two types of noise sources that result from this
transduction from an ionic to an electronic form:
D/C (Direct Current) Voltage Potential: caused by
differences in the impedance between the skin and the
electrode sensor, and from oxidative and reductive
chemical reactions taking place in the contact region
between the electrode and the conductive gel .
A/C (Alternating Current) Voltage Potential:
generated by factors such as fluctuations in impedance
between the conductive transducer and the skin. One
effective method to decrease impedance effects is to use
Ag-AgCl electrodes. The effect of the transducer noise
can be minimized by reducing the electrode impedance
via careful skin preparation and cleaning.
3- Electronic noise
The electronic noise is inherent in all electronic devices.
It can be minimized by
- Using low noise amplifiers at the preamplifier
level
- Avoiding the use of high-value (over a few
M)
- Match the noise impedances at amplifier inputs
- Matching the impedance in all electrodes
Typical EMG Processing Techniques
It is reported by many research workers that during the
muscular fatigue the frequency spectrum of the EMG is
shifted toward lower frequencies accompanied by an
increase in its amplitude. Eventually there are two
distinct techniques available to analyze the EMG signal
to detect the muscular fatigue.
1- Time domain analysis:
The root mean square (RMS) value of a signal is the
indicator of its power. It is a somewhat easy technique.
However, it is affected by various environmental
conditions and it is not recommended in studying the
behavior of the signal as a function of force or time
under sustained contractions. There is a linear
relationship between the number of zero crossings and
the number of motor unit action potential turns
(MUAPT) at low level contractions. Additionally
recruited motor units contribute to the EMG signal as
the contraction level increases and the linear relationship
does not hold any more. Hence, it is difficult to use it as
an indicator of the onset of muscle fatigue.
2- Frequency domain analysis:
The analysis of the EMG signal in the frequency domain
involves determination of parameters that describe
specific aspects of the frequency spectrum of the signal.
Fast Fourier transform (FFT) techniques are used to
obtain the power density spectrum usually through
digital computers on off-line basis. There are several
parameters that characterize the spectrum and can be

detected by electronic means like mean and

frequencies are the most reliable ones. The
frequency is less sensitive to noise. Electronic
are developed to study the changes in the
frequency during localized muscular fatigue.

median
median
systems
median

IV. MUSCLE FATIGUE MONITORING AS A

DESIGN PROJECT
It is a requirement in the Faculty of Engineering to do a
capstone design project before graduation. The intention
is to give students training on an in-depth design work
on a practical engineering problem at a technical level
similar to that might be encountered in the industry. In
due course, students will gain experience in planning
and managing projects, as well as documenting and
communicating engineering works. We feel that design
of a method for the monitoring of the muscular fatigue is
feasible and will provide us with the necessary design
exposure. The project is feasible in the sense that:
1- The project is sufficiently open-ended with
many technical and non-technical solutions
available.
2- We are a team of two students. Hence, in two
semesters that is partially allocated for the
project, and at the beginning of the third
semester after finishing courses we can study
the problem in detail and hopefully come up
with a useful solution.
3- Projects were carried out in the BME lab
around 20 years ago on this subject. Hence, we
have academic staff with necessary expertise
and qualifications to supervise the project.
4- We feel that we have necessary technical
background needed as pre-requisite for the
design work on this topic from biological and
physiology courses. We also took all technical
courses in electronics, computer and
biomedical engineering fields necessary for
implementing such a project.
5- Experiments can be performed in the
biomedical engineering lab, hence there is no
political and social problem related to them. No
license is required for production and/or
utilization of the developed devices.
6- Possible cost for a system like this, including
the materials, should not exceed 500 SR.
Necessary electronic components can be
procured from the local market. Our
expectation for the marketing cost is about 150
SR.
V. PROJECT METHODOLOGY
The prime objective of this project is to develop a
technique to confirm that, under voluntary contraction,

the frequency spectrum of an EMG signal shifts to lower

frequencies as a muscle fatigue occurs. A second
objective is to gain engineering experience through the
design and construction of the EMG apparatus. The
developed system shall be practical in using commonly
available facilities in the patient's environment.
The system involves the design of hardware and
software for data acquisition and signal analysis,
measurement and signal processing as illustrated in Fig.
2. It is realized in four distinct stages:

Band Pass
Filter

Fig. 2. EMG measurement and acquisition system

x
x
x

detect the EMG and resolving the relatively

small EMG signal from background noise by
the design and construction of the EMG
apparatus.
use the sound card of a personal computer and
the data acquisition functions in MATLAB to
collect data.
design of a program to sample an EMG signal
in real time, to transform them to the frequency
domain, and to calculate the median frequency.
experiment to try the developed system under
real life conditions.

The System Hardware

There are several considerations in building a working
device. First, the device must be constructed so that
internal noise does not obscure the signal. Second, the
electrodes must be designed to minimize ambient noise,
including signals from other parts of the body. Lastly,
frequencies outside the usable range of the EMG signal,
20-500 Hz, need to be filtered out.
Three surface electrodes are required; two exploring
(signal) and one reference. of the electrodes are for the
differential signal and the third is for the reference. The
signal electrodes are placed on the muscle with spacing
of about 2 cm and the reference electrode is placed into
a silent region away from the muscle.
The system hardware is composed of an EMG
amplifier and a band-pass filter, and a PC with its

interface. The EMG-amplifier acts as a differential

amplifier to eliminate the power line artifacts and
unwanted physiological signals such as the ECG. The
"Common Mode Rejection Ratio" (CMRR) represents
the relationship between differential and common mode
gains and is therefore a criteria for the quality of the
chosen amplification technique. The CMRR should be
as high as possible because the elimination of interfering
signals plays a major role in the signal quality. A value
over 95dB is regarded as acceptable. An amplification
level between 500 and 1000 is generally acceptable for
the sEMG processing. The band-pass filter that follows
the amplifier limits the frequency range of the signal that
goes to the sound card between 10 Hz and 500 Hz.
Personal computers are available everywhere with
integrated sound cards. Hence, utilization of a readily
available PC reduces the cost of the project and
contributes a lot to its practicality.
Software Requirements
Since the frequency range of the EMG signal is within
the audio band, it is possible to use the sound card as the
sampling and digitization tool. The MATLAB is a
scientific program that has the necessary functions to
control the sound card and to use it as a data acquisition
device. After acquiring the EMG signal with suitable
resolution and sampling rate, the signal is saved in a data
file for further processing. Then, the Fast Fourier
Transform (FFT) routine available in the signal
processing toolbox of the MATLAB is used for
computing the median frequencies of these sampled
data. MATLAB program developed for data acquisition
by sound card and median frequency computations are
given in Appendix A and B respectively.
The internal circuits of each recording channel of the
sound card include a sampler and an analog to digital
converter (ADC). The sampler samples the voltage
signal presented on the microphone input or the line-In
port of the sound card at a specified sampling rate. Each
sample is then digitized using a binary code word ( 8
bits or 16 bits ). These code words are available to any
software application interfaced to the sound card. The
frequency range of the sound card is limited to the
audible frequencies, 10 Hz to 15000 Hz. These limits
may vary slightly from one sound card to another. The
sampling rate at which the sound card functions is
specified by the application, and 8000 Hz, 16000 Hz,
and 44100 Hz are frequently used standard sampling
frequencies.
VI. SIGNAL DISPLAYS
The sound card has a dynamic range 0.5 V and it
offers a resolution of 16 bits per sample. The sampling
frequency must be at least 1000 Hz to avoid aliasing.
However the lowest sampling frequency is 8000

samples per second which is much more than sufficient

for the EMG signal works. The only concern with higher
sampling rate is the occupation of more memory space
due to extra data points collected. This is not an issue for
the computers used today.

Fig. 3. EMG signal from biceps captured during a

repetitive exercise for 20 seconds
Signal Recording and Display
A MATLAB program script is written to capture the
EMG signal coming through the sound card (AppendixA). The resulting data is plotted as indicated in Fig. 3.
The signal is recorded from the biceps during an arm
flexion exercise. The exercise continued for 20 seconds
with 2 seconds intervals.
Fig. 4 shows an expended display of the first spell. It
is clear that the signal amplitude increases during the
contraction and only noise appears at the rest time.

Fig. 5. A sinusoidal test signal at 150 HZ and it's spectrum

A real time EMG recording is carried out and it's

spectrum is computed as shown in Fig. 6. The signal
continued for 5 seconds with two spells of exercise.

Fig. 6. Part of the EMG signal and it's power spectrum

VII. THE MEDIAN FREQUENCY

Fig. 4. The EMG signal from a single contraction
The EMG Spectrum
The FFT subroutine from the MATLAB's signal
processing toolbox is used to compute the spectral
components of the signal. A sinusoidal test signal is
applied at various frequencies to test the system. Fig. 5
shows the 150 Hz signal and it's spectrum.

The median frequency is defined as the particular

frequency that would divide the power spectrum into
two parts of equal areas. It is calculated in two steps as:
x The total area of the power spectrum of the
signal by numerically integrating the power
spectrum curve,
x Adding up the area from frequency 10 Hz
onwards until the accumulated area is equal to
half the total area of the power spectrum.
The frequency when this half signal power is reached
is identified as the median frequency and recorded into a
file. In order to get a closer look at the results of median
frequency analysis, the following steps are performed:
1- The arm muscle is exercised for several
minutes continuously with a load and EMG
signal is captured for analysis.

2- This signal is packed into six 5-second

sections. The power spectrum and median
frequency are computed for each section.
3- The average and standard deviation of median
frequencies of each package are computed such
that each group has 6 sections and all
information is graphed at every 30 seconds.
Fig. 7 shows the progress of the median frequency
during the exercise. The test subject is 25 year old male
with height 165 cm and weight 73 kg. The recording is
done from the right arm as the subject lifts a load of 5
kg. Sudden drop in the median frequency two minutes
after the onset of the exercise is a clear indication of
muscular fatigue.

operational one by more than a standard deviation and

stop the exercise as this value is reached.
Conclusion
The median frequency is an indicator for the muscle
weakness (Fatigue) as illustrated in the previous
chapters. We studied the frequency domain of the picked
up signal by FFT (Fast Fourier transform) to detect the
median frequency from that spectrum. The system can
be improved by adding a wireless link between the
system hardware and the PC. Then, the patient can
exercise freely without limitations of the harness and it
can be applied on any muscle in locomotion. The system
can give early warnings to the patient to stop the
exercise before he reaches into a state of exhaustion.

ACKNOWLEDGEMENT
The work was carried out as a graduation project by the
2nd and 3rd authors and supervised by the 1st author. The
authors express their gratitude to the staff in the
Biomedical Engineering Group for their continual
supports throughout the project work.

REFERENCES

Fig. 7. Error bar graph for median frequency for test -1

VIII. DISCUSSION AND CONCLUSION

Discussion
From the results of the median frequency analysis, we
can see that the median frequency does indeed decrease
as the muscle exercise continues and this is due to the
fact that the muscle is getting closer to the fatigue state.
However, we see also that the median frequency
fluctuates due to errors from several sources such as
signals from the heart, signals from other muscles and
outside sources such as the 60 Hz power line noise.
Another reason for error is the limited accuracy of the
instrument used, namely the computer sound card. The
error magnitude introduced in the median frequency is
+/- the standard deviation. The mean value of the MF
indicates a statistically stable score as illustrated in Fig.
6. The decay in the MF as the muscle gets tired is very
clear.
The standard deviation is used as an indicator of the
error in the MF. It also changes during the exercise.
Nevertheless, the change is around a central value that
can be detected from the early parts of the records. We
can set a threshold value that is safely below the normal

[1] B. Karagzolu M. N. Abdulqadir, A System for

Analyzing Performance of Muscles Revealed by
Electromyogram During Exercise", Project 410/051,
Faculty of Engineering, KAU, 1991.
[2] B. Karagzolu, Development of an electronic

system to study muscular activity, J. of King

Abdulaziz University, Engineering Sciences, special
issue, pp. 135-141, 1999.
[3] W.H. Sindi, A.A. Al-Omari, Development of a Method
to Detect Onset of Muscle fatigue Senior Project Report,
Faculty of Engineering, KAU, March, 2008.

APPENDIX-A:THE SIGNAL CAPTURE PROGRAM

% Capture a signal for the given duration T
% Plot the signal in time
clear all
close all
clc
T=20; %Duration
%-------------------------------------------AI = analoginput('winsound');
chan = addchannel(AI,1);
duration = T;
set(AI,'SampleRate',8000)
ActualRate = get(AI,'SampleRate');
set(AI,'SamplesPerTrigger',duration*ActualRate)
set(AI,'TriggerType','Manual')
BB = get(AI,'BitsPerSample')
blocksize = get(AI,'SamplesPerTrigger');
Fs = ActualRate;
start(AI)
trigger(AI)

data = getdata(AI);
delete(AI)
clear AI
%----------------------------------------------------t=1/8000:1/8000:T;
plot(t,data)
grid on
ylabel('Magnitude')
xlabel('Time (Sec)')
title('Captured Signal')
p = [0 0.5];
figure,
plot(t,data)
xlim([p(1) p(2)])
grid on
ylabel('Magnitude')
xlabel('Time (Sec)')
title('Captured Signal')
%----------------------------------------------------save('emg_1','data','t')

APPENDIX B: MEDIAN FREQUENCY

CALCULATION AND DISPLAY PROGRAM
% Read the captured EMG signals, cut each signal and find
spectrum
clear all
close all
clc
% load('Ahmed_3','data','t')
% load('Waaleed_1','data','t')
% load('Hessain_L1','data','t')
% load('sami_2','data','t')
load('L_5','data','t')
% load('ex_360_sec','data','t')
% figure,
% plot(t,data)
% grid on
% ylabel('Magnitude')
% xlabel('Time (Sec)')
% title('Captured Signal')
%--------------------------------------------T = length(data)/8000;
k=1;
for n = 1:5:T
s = data((n-1)*8000+1:(n+4)*8000);
tt = t((n-1)*8000+1:(n+4)*8000);
%
% figure,
% subplot(2,1,1);plot(tt,s)
% ylabel('Magnitude')
% xlabel('Time (Sec)')
% title('Captured Signal')
%----------------------------------------------------Fs = 8000;
z1 = fft(s)/Fs;
z1 = abs(z1(1:length(z1)/2+1));
z1 = z1.^2;
% Power spectrum
f1 = [0:length(z1)-1]*Fs/length(z1)/2;
subplot(2,1,2);plot(f1,z1);
ylabel('Magnitude (abs)')
xlabel('Frequency (Hz)')

title('Frequency Spectrum')
xlim([10 500]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% compute median frequency
% step1 compute area of spectrum
n1 = length(z1);
df = f1(20)-f1(19);
area = 0;
for m = 1:n1-1
area = area + df*(z1(m)+z1(m+1))/2;
end
% step2 start from zero and compute
% 1/2 area of spectrum
%The median frequency defined as the particular
%frequency that would divide the
%power spectrum into two parts of equal area.
%http://www.mathworks.com/support/solutions/data/12UVGQ4.html?product=SG&solution=1-2UVGQ4

ar = 0;
for m = 1:n1-1
ar = ar + df*(z1(m)+z1(m+1))/2;
if ar >= area/2
break
end
end
median_freq(k) = m*df;
k = k+1;
end
median_freq';
figure,
plot(median_freq)
length(median_freq);
h = 6; % number of median frequencies to be averaged
med =[];
for k = 1:h:length(median_freq)
s = median_freq(k:k+h-1)';
med = [med s];
end
med;
tt = 5*h:5*h:T;
av = mean(med);
st = std(med);
figure,
errorbar(tt,av,st)
ylabel('Median Frequency (Hz)')
xlabel('Time (Sec)')
title('Median Frequency Error Bar Graph')
grid on