Material Information

Back-Man the foundation, development and performance of a portable, digital, real-time noninvasive EMG monitor
Standiford, Floyd Jay
Publication Date:
Physical Description:
viii, 59 leaves : illustrations ; 29 cm

Thesis/Dissertation Information

Master's ( Master of Science)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Electrical Engineering, CU Denver
Degree Disciplines:
Electrical engineering


Subjects / Keywords:
Electromyography ( lcsh )
Electromyography ( fast )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )


Includes bibliographical references.
General Note:
Submitted in partial fulfillment of the requirements for the degree, Master of Science, [Department of] Electrical Engineering.
General Note:
Department of Electrical Engineering
Statement of Responsibility:
by Floyd Jay Standiford.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
26922791 ( OCLC )
LD1190.E54 1992m .S72 ( lcc )

Full Text
j Floyd Jay Standiford
B.S., University of Colorado, 1990
i A thesis submitted to the
: Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
I Master of. Science
Electrical Engineering

1992 by Floyd Jay Standiford
All rights reserved.

This thesis for the Master of Science
degree by
Floyd Jay Standiford
has been approved for the
Department of
Electrical Engineering
Mike Radenhovic

Standiford, Floyd Jay (M.S., Electrical Engineering)
Back-Man: The Foundation, Development and Performance of a Portable,
Digital, Real-Time, Noninvasive EMG Monitor
Thesis directed by Professor Joe E. Thomas
When a muscle contraction is sustained, a EMG frequency spectrum is
produced which will compress as the muscle is fatigued. This compression
can be measured by a muscle fatigue monitor such as the one developed in
this thesis. Although the development of muscle fatigue monitors began prior
to 1982, until now, they have been based on analog techniques. This is the
first (to the author's knowledge) muscle fatigue monitor, which is
implemented as a portable, battery operated, microprocessor based, digital
design-utilizing real-time FFT frequency analysis of the EMG signal to find
the spectral compression associated with muscle fatigue.
This communication begins with the basic physiology of the EMG
signal and continues by describing the development of the EMG monitor,
ending with muscle fatigue test results. In addition, some of the applications
of a monitor of this design are described such as detection and treatment of
muscle disease, muscle pain diagnostics, sports therapy, job environment
analysis, muscle injury fraud and relaxation training.
This abstract accurately represents the contents of the candidate's thesis. I
recommend its publication.

The author wishes to express his gratitude to Dr. Dan Michaels and Dr.
Joe Thomas. Their valuable advice and support have been inmeasuasble toward
the evolution of Back-Man. I would also like to thank my loving wife, Linda, for
her understanding and support of my professional goals.

1. Introduction.........................................
2. Objectives for the Muscle Fatigue Monitor Prototype..
3. Explanation of EMG signals..........................
3.1 Physiological Basis of the EMG Signal....
3.2 Methods Used to Measure EMG Signal.......
3.2.1 Invasive Measurements...........
3.2.2 Noninvasive Measurements........
3.3 Electrical Characteristics of the EMG Signal.
4. Previous Instruments Used to Measure EMG............
5. The Muscle Fatigue Monitor System...................
5.1 Principal of Operations..................
5.2 Specifications Required for a Muscle Fatigue
, Monitor......................................
6. Hardware Resign.....................................
6.1 Hardware Design Problems.................
6.2 ^ Hardware Design Solutions..............
6.3 Hardware Design Results..................
7. Software Design.....................................
7. li; Problems...............................
7.2: Forth83.................................

7.3 Design Solutions.................................25
7.4 Design Results...................................27
8. Test Measurements of the System..............................28
9. Applications.................................................33
10. Recommendations and Improvements............................35
11. Conclusions.................................................37
Appendix A:; Pictorial View of the Muscle Fatigue Monitor........39
Appendix:: 'Hardware Schematics.................................40
B-l. Input Amplifier..................................40
B-2. Low Pass Filter..................................41
B-3. Final Amplifier Schematics..................... 42
B-4. Power Supply.....................................43
B-5. Display Logic....................................44
Appendix C: Test Data............................................45
C-l. Battery Life.....................................45
C-2. Shorted Input Noise..............................46
C-3. Input Sine Wave..................................47
C-4. Sine Wave After The Hanning Window...............48
C-5. The FFT and Median Frequency (without
Hanning Window)........................................49
( ________
C-6. The FFT and Median Frequency (with Hanning
Appendix D: Sample Procedure in Forth............................51

Appendix E: ' EMG Signal Test Data...........................52
E-l. Sample EMG Signal..............................52
E-j'2. Sample EMG Signal After the Hanning Window...53
i E-3. Magnitude of EMG Signal........................54

1. Introduction
In the past few years there has been increasing interest by many
different disciplines in the measurement of electromyography (EMG) signals.
Doctors are accepting the use of EMG signals as a diagnostic tool in their
clinical practice. Large corporations are using EMG to study the environment
. j
in which employees work and the employees' suitability to a position within
this environment! In addition, insurance companies are interested in EMG
signals to help determine the cause of muscle injury, in particular the cause of
back injuries. In 1989 Liberty Mutual Insurance Co., the largest worker's
compensation insurer, paid out $1 million every working day for cost related
to back pain.[l] EMG studies are also being used to aid physical therapy and
athletic training.
This thesis focuses on the development of an EMG monitor used to
measure the activity of a muscle, in particular the measurement of muscular
fatigue. When a muscle contracts a signal known as a myoelectric (ME)
signal is generated. Through the use of EMG, the ME signal can be detected
inside and/or outside the muscle. EMG is used to provide an objective
measure of muscle fatigue, muscle pain and muscle disorders. It has been
thought since 1912 that the frequency content of the EMG signal shifts to
lower frequencies as the muscle fatigues. [2] However, recently it has been
shown that as thej muscle fatigues, the EMG frequency spectrum is
compressed and it is the median frequency of the EMG signal that shifts to a

lower frequency;[3] The theory used for the muscle fatigue monitor is based
on this result and a prototype muscle fatigue monitor is implemented as a
portable, battery operated, microprocessor based design utilizing digital
methods to measure muscle fatigue.

2. Objectives for the Muscle Fatigue Monitor Prototype
The system prototype design objectives for the muscle fatigue monitor
are many. The first system objective is, obviously, that it must accurately
measure the EMG signal. The muscle fatigue monitor must then analyze the
EMG signal and determine the muscle fatigue. It is desirable to have a
microprocessor based design, which implements digital techniques to find the
median frequency shift of the spectrum. The system is to be portable; thus, it
must be battery operated and should have a battery life of at least eight hours.
(This allows the muscle fatigue monitor to be used for an eight hour work day
testing situation.) The muscle fatigue monitor needs to be able to store
muscle fatigue information and to upload this information to a host computer.
Also, a display of the information is beneficial to enable feedback during
muscle fatigue testing. Finally, two channels of EMG are requested for the
prototype, with the capability to extent the number of channels to four or

3. Explanation of EMG signals
To provide a general understanding of the source of EMG signals, this
chapter will give a brief overview of the physiology that generates an EMG
signal. Special emphasis will be given to explain how the muscle reacts to
fatigue. Furthermore, a look at the different methods used to retrieve EMG
signals will also discussed, including the types of results that can be expected
from these different methods. The chapter is concluded with the electrical
characteristics of: the EMG signal
3.1 Physiological Basis of the EMG Signal
Although the physiological basis of EMG signals is not fully
understood, research has shown that the alpha motor system and its motor
units are the source of the EMG signal. A motor unit consists of a lower
motor neuron, its axion and the muscle fibers which it innervates. [3] The
muscle fibers are directed to contract when a nerve action potential is sent by
the central nerve system. Specifically, this nerve action potential travels
down the axion of the alpha motor system's motor unit to the neuromuscular
junction. When the nerve action potential reaches the neuromuscular
junction, it creates a release of the chemical acetylcholine (ACH). The
release of ACH initiates a chemical event which begins a process that
depolarizes the muscle fibers of the motor unit. Depolarization of the muscle
fiber releases energy which travels along the direction of the muscular fiber.

It is this energy that creates a muscle action potential (MAP). The MAP is the
source of the EMG signal.
The frequency content in the energy of the MAP signal has been
suggested by De Luca to emanate from two different types of MAP. One
developed from a group of slow twitch muscle fibers and another one from a
group of fast twitch muscle fibers. [3] The fast twitch fibers contain
frequencies higher than 100 Hz and the slow twitch fibers include the
frequencies less than 100 Hz. The fast twitch muscle fiber group is suggested
to be resporisible for phasic or fast ballistic movements, such as those used by
a sprinter. These also might be considered as the acute fight or flight muscles,
used when a person senses stress or danger. The slow twitch muscle fiber
group is consequently composed of the muscle fibers required for (tonic)
postural support. Furthermore, the slow twitch fibers provide muscular
endurance as would be needed by a marathon runner.
Using this physiology, as a muscle is fatigued the fast twitch muscle
fibers attenuate or stop firing. Thus, only the slow twitch muscle fibers
generate MAPs. This creates the compression of the frequency spectrum. The
most popular method of measuring spectral compression is to measure the
shift of the median frequency of the spectrum. Thus, as the spectrum
compresses the median frequency will shift to lower values. Through many
empirical studies, it has been found that most muscles are totally fatigued
when the median frequency shifts to 50% of its starting (non-fatigued)
value. [4]

3.2 Methods Used to Measure EMG Signal
Two common methods are used to measure EMG signals. The
invasive method utilizes needle probes inserted into the muscle and the non-
invasive method uses surface probes attached to the skin. It is important to
identify which type of EMG test is being performed as the measurements are
quite different. It has been suggested by the medical community that non-
invasive EMG studies should be called iEMG. This would signify the
integrating nature of the non-invasive measurement. [3] For simplicity the
author will assume that all references to EMG will mean iEMG except for the
section on invasive measurement which is presented next.
3.2.1 Invasive Measurements
Invasive rheasurements are performed using needle probes to record a
single MAP. Thus, allowing a single motor unit to be studied. Research of
this type, is normally conducted by physiatrists and neurologists, who are
interested in the integrity of the nerve that feeds the muscle. Note that only on
rare occasions will needle EMG studies measure the integrity of the muscle
fiber itself. [3]
3.2.2 Noninvasive Measurements
Surface type electrodes are used for noninvasive measurements of the
muscle activity. These electrodes actually measure the integrated signal from
many MAPs (or populations of motor units) in the localized area under the

electrodes. The genesis of this integrated signal is a collection of individual
MAPs, traveling; through the tissues of the body and collecting on the surface
of the skin, along the lines of energy from which the MAP(s) originated. A
number of factors determine the amplitude of the EMG signal produced by
MAPs. They are: 1) The number of MAPs; 2) The rate of the MAPs; 3) The
proximity of the MAPs to the electrodes; 4) The distance between the
electrodes; 5) The skin preparation, and 6) The obesity of the individual. [3]
Due to the nature of the non-invasive measure, the position chosen for
the electrodes is an important consideration. Either single muscles may be
monitored or a group of muscle may be monitored by a single set of
3.3 Electrical Characteristics of the EMG Signal
The electrical characteristics of the EMG signal are somewhat
ambiguous. Through researching the current literature and the author's own
tests it is suggested that the following electrical characteristics seem to apply.
The amplitude of the EMG signal appears to range from a few microvolts to
hundreds of millivolts, with the range of interest from 20 microvolts to 4
millivolts. The frequency content of the EMG signal is reported to extend
from 20 to 10,000 Hz. [5] However, little power seems to be found above 500
Hz and many systems measure only frequencies less than 300 Hz. Therefore,
the author has chosen a range of interest for the initial prototype as 8 to 512
Hz. A typical sample of the EMG spectrum follows in figure 1.

Sample EMG Spectrum

4. Previous Instruments Used to Measure EMG
Many different designs and methods have been implemented to
measure the state of the EMG signals. Most real-time designs thus far have
been based on specialized analog circuits, which only measure the rms power
within a certain bandwidth. [2] [6] For example, the rms power is measured
using a narrow band of 100 to 200 Hz. Since the EMG spectrum has
frequency components in this bandwidth when the muscle is not fatigued, a
high rms level (amplitude) is measured. As the muscle fatigues, the median
power of the spectrum shifts to lower frequencies and the rms energy
decreases in the chosen bandwidth. Thus, the muscle fatigue is measured.
Obviously, this method of measure has many limitations. First, the frequency
content of the EMG signal varies with individuals and between male and
females. Second, the amplitude of the EMG signal varies depending on the
force of the contraction. Consequently, the signal measured in the bandpass
may change due to the force of the contraction as well as the fatigue of the
Of the designs which use a fast fourier transform (FFT) spectral
analysis, most have not been real-time designs. They require that the EMG
signals be recorded, for example on magnetic tape, and then be processed at a
later time. While this method has been good for the research of EMG signals,
it prevents direct feedback to the patient or researcher.

Recently; some new systems have allowed real-time processing in the
lab or clinic. They have been based on mini computers or fast AT class IBM
machines. The main limitation of these designs is that they are not truly
portable in nature or design, thus limiting the range of applications that they
are suited for. These new designs have shown the value of FFT analysis of
EMG signal and have furthered the need for a portable muscle fatigue

5. The Muscle Fatigue Monitor System
The fundamental system design of the muscle fatigue monitor is
schematically represented in the block diagram of figure 2. In the following
section, the basic operation of the muscle fatigue monitor is described.
Following the principals of operations, the general specifications that are
required of the muscle fatigue monitor are summarized in an outline format.
5.1 Principal of Operations
The EMG signal is obtained through surface electrodes. These
electrodes include a differential pair along with a reference ground which
helps maintain a high common mode rejection ratio (CMRR). The input
signal is amplified and filtered through a high pass filter to remove the low
frequencies which may occur from electrode contact potentials (DC) and
electrode movement artifacts. The signal is then filtered through a low pass
filter to prevent aliasing. After filtering, the signal is amplified again to the
overall gain values of 100 and 1,000 or 1,000 and 10,000. This allows some
flexibility, regarding the amplitude of signals which one may wish to
measure. At this .point the analog signal is digitized by a 10 bit analog to
digital (A/D) converter which is built into the Intel 80C196 microprocessor.
Through software or hardware interrupts, the signal is over-sampled at 2048
Hz for 0.25 seconds, resulting in 512 samples. The samples are then averaged

two at a time to give a final array of 256 samples, with an effective sampling
rate of 1024 Hz.
Figure 2. System Block Diagram of the Muscle Fatigue Monitor.

The EMG samples are then conditioned through a Hanning window,
reducing the high frequency leakage effects due to Gibb's phenomena and the
FFT window. The conditioned signal is then processed by a 256 point FFT,
resulting in a 128 point frequency spectrum which is used to End the total
energy of the spectrum from 8-512 Hz. The median frequency of this
spectrum is then obtained. At this time, the liquid crystal display (LCD) is
updated with the. new median frequency and the new total energy of the
spectrum. This cycle is repeated five times and a running average is
computed for the median frequency. The average median frequency is then
stored in RAM for future uploading to the host computer. Furthermore, a
battery test is performed to maintain the integrity of the data received.
Finally, the muscle fatigue monitor system will go into a sleep mode. The
sleep mode is entered after a predetermined number of results are obtained, or
when the memory is full. The sleep mode will also be entered if the battery
voltage drops below a preset threshold value.

5.2 Specifications Required for a Muscle Fatigue Monitor
The hardware specifications for the muscle fatigue monitor are:
1) Input Signal Amplification
. a) Frequency Range of at least 20 to 500 Hz
b) Signal Amplitude Range 20 pV to 4 mV
c) i Noise < 5 pV
d) CMRR> 110 dB
2) Data Acquisition
a) , 10 bit A/D Linear to 0.5 bits
b) ' Conversion time < 500 pSec.
3) Storage Requirements
The storage requirements will vary depending on the
particular application.
4) Display Requirements
LCD display of median frequency and total energy of the
spectrum. One might also provide a limited view of the
frequency spectrum.
5) Battery Life
. The battery life should exceed eight hours.

6. Hardware Design
A detailed discussion of the hardware design has been omitted here for
the sake of brevity. The major hardware problems are outlined and the
solutions are discussed. The important design specifications and results are
also given. For complete schematics, please refer to the appendices.
6.1 Hardware Design Problems
The hardware design problems of the muscle fatigue monitor system
are formidable. The input amplifiers must be able to detect a small EMG
signal that is generated through a large input source resistance, typically on
the order of 10^ ohms. The EMG signal is also usually accompanied by a
relatively large 60 Hz component due to the surrounding environment of our
modem world. In addition, the electrodes cause a substantial contact potential
and may also introduce extremely low frequency artifacts due to electrode
movement. Another consideration, regarding the inputs to the amplifiers is
electrostatic discharge (ESD). Since the muscle fatigue monitor is a portable
device, the possibility of ESD is quite high; therefore, the inputs to the
amplifier must be; protected from ESD. Once the signal is obtained, the task
of the hardware system is to condition the input through band pass filters.
The analog filters need to be designed such that their distortion effects, of
amplitude and phase, are minimal to the signal and yet still prevent low
frequency artifacts and high frequency aliasing from degrading the system.

The low frequency nature of the EMG signal does not cause any
particularly special considerations regarding the analog to digital (A/D)
converter; however, it should be reasonable fast and linearly accurate. The
A/D should also provide good resolution, thus enabling it to accurately
measure the input signal over the range of interest
The microprocessor, on the other hand, is an important consideration.
Some of the most important features of the microprocessor are as follows.
First, the CPU needs to be able to compute an FFT and execute the supporting
program in a time frame that will allow the system to operate in real time.
Secondly, the microprocessor should also be a low power CMOS design,
allowing the CPU to be operated at lower power levels between
measurements. A sleep mode from a completely static design would also be
beneficial in extending battery life while the CPU is waiting to uploaded data
to a host computer. And lastly, the microprocessor should be upgradable in
the future, thus allowing more speed and function to be built into the muscle
fatigue monitor. :
User input and output considerations for the muscle fatigue monitor,
in the prototype sjtage, have been limited to a simple LCD and a link through
an RS232 port to ;a host computer. This allows some simple feedback through
the LCD and the capability to perform a more detailed observation and/or
analysis through a host computer.
One last hardware consideration that underlies all hardware design is
the portable, battery operated specification. Accordingly, the size and number

of components used must also be carefully considered and all circuits must be
designed to minimize the power required and to work with the available
power supplies.
6.2 Hardware Design Solutions
A block diagram schematically showing the major components of the
hardware design ,is given in figure 3. For additional hardware schematics
please refer to the appendices.
The first consideration in the hardware design was the choice of the
microprocessor. Through the suggestion of Dr. Michaels an SBC 196
embedded controller was chosen through Vesta Technology, Inc. The
SBC196 incorporates an Intel 80C196KB microprocessor operating at 12
MHz, with approximately 24K bytes of RAM and 32K bytes of ROM
programming space available to the user. The Vesta product also includes a
RS232 serial interface, and a real time clock with a supercap power backup.
Other features include an eight bit microprocessor data expansion bus, a 4K
bytes off-board address space, EEPROM, and additional serial busses. The
Intel 80096 is a!|16 bit microprocessor which includes an eight channel, 10
bit A/D converter and the capability to execute a 16x16 bit multiplication in
2.3 pS. Since an FFT requires 0(Nlog2N) operations, we find that the
execution time for a 256 point FFT is on the order of milliseconds. The
SBC 196 has an impressive Forth compiler available which will generate

ROMable code. |Thus, with an EPROM programmer, a complete stand alone
system can be developed.

Figure 3. Hardware Block Diagram of the Muscle Fatigue Monitor.
With the choice of the SBC 196 embedded controller, the rest of the
hardware was designed-starting with the input signal amplification and
conditioning. Due to the 60 Hz problem, the design of the input amplifier
was the most critical component. The typical approach is to include a 60 Hz

notch filter; however, since a large component of the EMG signals of interest
includes 60 Hz, a notch filter is unacceptable. The approach thus taken was to
improve the common mode rejection ratio (CMRR) at 60 Hz. Obtaining high
common mode rejection is a difficult task for this type of application since the
input source resistance is usually not balanced. The final solution was to use
an INA102, a special, low power, medical instrumentation amplifier, supplied
by Bur Brown. In addition, a compensation circuit is used to improve the
internal balance of the INA102, thus increasing the CMRR to values greater
than 120 dB. The inputs to the INA102 were DC coupled to the electrodes,
through a diode clamping circuit to prevent damage due to ESD. DC
coupling was chosen, even though surface electrodes may create large contact
potentials, to minimize the possibility of source imbalances. Remember that
source imbalances can seriously degrade the CMRR. The INA102 is
prevented from saturating through the diode clamping of the inputs, and
furthermore, the usual gain is limited to only ten. A passive high pass filter is
cascaded to INA102 to block DC and low frequency artifacts. Following the
high pass filter, ah active, four pole, butterworth low pass filter is used to
prevent aliasing. The signal is then gain compensated to give an overall gain
of 100 and 1,000. In addition, the signal is level shifted to give an output of
zero to five volts, as required by the A/D of the 80C196. Note that all
components and values were chosen to reduce the power requirements of the

A four line, 20 character per line LCD display was interfaced to the
80C196 through the expansion bus on the SBC196. Low power CMOS
buffers and decoders were used to interface the fast bus of the 80C196, to the
slow timing requirements of the LCD. Additional circuitry was also added to
allow the contrast of the display to be adjusted.
Power for the muscle fatigue monitor was generated through four
different power regulators, all supplied by six rechargeable, AA sized, nicad
batteries. Different power supplies were provided to allow for the
considerations of each subsystem. The first power supply powers the
SBC196, including the 80C196, RAM, and support chips. The second power
supply is a split power supply derived from the RS232 driver on board the
SBC196, and provides 9 volts for the INA102 instrumentation amps. The
third power supply is used to power the final op amps and the A/D voltage
reference, thus optimizing the linearity of the A/D conversion. Lastly, the
LCD and supporting logic chips are driven though a separate power regulator
to isolate the analog circuits from the digital circuitry. One final note, all the
power supplies, except for the first, have been designed to allow the 80C196
the ability to control their on and off state through software to enhance battery

.3 Hardware Design Results
The results of the hardware design is summarized in the following
ables. Additional results are also given in the appendices.
Power Supply Current
With Display 90.1 mA
Without Display 77.5 mA
Idle 55.2 mA
Sleep < 5 mA
Input Amplifiers
Gain xlOO, xlOOO, xlOOOO
Bandwidth 2.0- 1,000 Hz +0, -3 dB
CMRR > 120 dB @ 60 Hz
Noise ~5nV

7. Software Design
A summary of the software design is given, which includes some of
the software problems that had to be resolved. The solutions to these issues
are discussed in a general fashion-stating only the methods used to approach
the problems. A brief discussion of the high-level Forth computer language is
also given here to explain its advantages in this type of system. Furthermore,
a sample procedure written in Forth code may be found in the appendices.
7.1 Problems
The execution of the software is critical to the real-time operation of
the muscle fatigue monitor. The tasks that must be performed are
computation intensive and until recently microprocessors with adequate
processing power were unavailable. To operate in a true real-time sense the
software must be able to calculate the median frequency within the time frame
required to obtain the samples. For example, using a 256 point array,
sampled at 1024 Hz, we find that it only takes 0.25 seconds to obtain an input
array. If sampled continuously we would only have 0.25 seconds to calculate
the median frequency of the data. Fortunately, it turns out that the rate of
muscle fatigue is slow enough that one doesn't need to always sample
continuously to achieve a real-time measurement of muscle fatigue. From
empirical studies and research of the literature, the author has found that rapid
muscle fatigue occurs over a range of 30 to 45 seconds. Thus, to provide at

least ten median frequency measurements, the software must complete a
sample and a result within 3.0 to 4.5 seconds. Note that a previous real-time
digital system used 6 seconds to perform a single median frequency
measurement. [5] For slower fatigue test this should not present a problem as
the data will need to be averaged or compressed for storage and trend
analysis. Even with our relaxed understanding of real-time measurements it
can be seen that software speed is critical.
Other software design problems include the limitations of integer
' |
numbers to represent floating point numbers, 16 bit overflows and the speed
penalties of transcendental functions. Additionally, software timing is critical
for the input sampling routines. In order to sample the input properly, the
software must initiate an A/D conversion at correct intervals.
7.2 Forth '83
The Vesta SBC 196 has the option of including a Forth interpreter /
compiler. This system was used extensively for the software development
portion of the muscle fatigue monitor, therefore, a brief discussion of the high
level language Forth will be presented. Forth was developed in the early '70s
for critical scientific and industrial applications. It is an efficient, compact,
high level language that features fast execution and small executables. Forth
consist of many things. It is a high level language, an assembly language, an
operating system, a set of development tools, and a unique design philosophy.

The fundamental approach to the Forth language is to build on itself.
The Forth language starts with a set of basic definitions, which are called
words. The user can then create new definitions which are built on these
words. New definitions can also be formed directly from assembly language,
further extending the language. This building style allows application specific
words to be generated. Thus, the resulting program can then be an English
style reading of the functions to be performed. Forth also has the capability to
compile words interactively. Therefore, a user can test definitions
immediately; this helps shorten the development time required to complete a
Forth has] been growing in popularity, finding applications in many
systems. Examples include business and personal computing software, data
acquisition analysis, expert systems, graphics, medical, and robotics systems.
Fourth has also been developed into a hardware engine through the Rockwell
International R65F11 and R65F12. This implementation is said to execute
faster that code written directly in assembly language. [7]
The Forth system that is developed by Vesta has the capability to
generate ROMable code for the SBC 196. This gives the user the ability to
write a program and then, with the use of an EPROM programmer, bum the
program into hardware for autostart applications.

7.3 Design Solutions
A block diagram of the software is schematically represented in figure
4. The software written at this time is an alpha version, written completely in
Forth. Future versions will include some inline assembly code to further
enhance the execution time of the software.
A detailed discussion of how the software is constructed will be
omitted. Instead, a general overview of the basic functions performed and the
basic algorithms used will be discussed. From the block diagram of the
software, the general flow can be observed. After basic initializations, the
software starts by obtaining a sample of the data. A simple looping algorithm
has been implemented at this time with an interrupt driven method planned
for the future. E)ata from channels A and B are read into memory
simultaneously. Once the data is gathered, it is processed through a Hanning
window to reduce the leakage effects of the Gibbs phenomena. The data then
undergo a transformation to the frequency domain through a digital FFT. A
Cooley-Turkey, radix two, decimation in frequency algorithm, as described
by Burrus and Pajrks is implemented to perform the 256 point FFT. [8] A
technique described by Brigham for combining two real FFTs into one
complex FFT is also utilized to cut the processing time.[9] This allows both
channels A and B to be calculated at the same time. Once in the frequency
domain, the total energy is calculated by integrating the area of the spectrum.

The median frequency is then determined, with the results displayed to the
I 11
LCD and the host computerif connected.
Figure 4. Block Diagram of the Software for the Muscle Fatigue.

The problems of integer arithmetic are overcome through proper
scaling. The transcendentals for the twiddle factors and the Hanning window
are stored in a look up table, thus speeding the execution times of a
transcendental to that of a memory read. An interesting approximation for the
square of the sum of squares was found in an article written by Williams.[10]
The paper concluded that the square of the sum of squares could be
approximated to !an accuracy greater than 94% by:
! yjx2 + y2 = Ax + By
Where A== 1.00 and B = 0.375, provide that x > y. This
approximation was used to enhance the speed of the magnitude function.
7.4 Design Results
The results of the software design produced code that is capable of
finding the median frequency of two sampled EMG inputs, channels A and B,
in an average time of 3.4 seconds. This is the total time required to execute a
loop of the software as shown in figure 4. While this is slower than the author
would like, it will be possible to improve this execution time through the use
of inline assembly code. The code, as written now, does provide an adequate
starting point'for the test measurements of the EMG signal and the analysis of
muscle fatigue. A sample of the LCD Display and the host display is given in
chapter 8.

8. Test Measurements of the System
The author has performed several muscle fatigue tests on various
muscles to prove, the operation and results of the system. Figures 5,6,7, 8,
and 9 show the result of one of these tests. The test muscle was the right
biceps, using, an isometric contraction. Figure 5 shows the results as
displayed on a host computer. Note that only channel B was used for this test,
The inputs for channel A were shorted to disable the output of random data.
At the start of the test, the muscle fatigue measured 56 Hz (via median
frequency) and at the end of the tests, the muscle fatigue measured 28 Hz.
The time increment between steps is approximately 3.4 seconds and the total
time of the test was 51.0 seconds. The author will verify that his right biceps
was fatigued at the end of the test!
0 , 8 , 56 , 0 , 1636
1 , 8 ' 52 , 0 , 2044
2 , 8:, , 52 , 0 , 2112
3 , 8 ; 52 , 0 , 1978
4 , 8 , 44 , 0 , 1904
5 , 8 , 36 , 0 , 1746
6 , 8 l 48 , 0 , 1540
7 , 8 , 44 , 0 , 1334
8 , 8 , 44 , 0 , 990
9 , 8 , 36 , 0 , 1066
10 , 8, : 32 , 0 , 920
11 , 8 ' 36 , 0 , 1046
12 , 8 , 32 , 0 , 606
13 , 8 , 32 , 0 , 620
14 , 8 i, 28 , 0 , 682
END FFT R2 Program ok
, Figure 5. Host display of Muscle Fatigue Test.

In Figurej 6, a sample of the LCD display is given-showing the final
muscle fatigue measurement. Figure 7 is the starting frequency spectrum as
found, and stored in memory, by the muscle fatigue monitor. Note the high
frequency components and the median frequency of the spectrum. Figure 8
shows the ending frequency spectrum. It is very obvious that the spectrum
has compressed, with most of the high frequencies becoming attenuated.
: !>
Figure 9 shows a three dimensional view of the two spectrums, overlaid to
give a direct coniparison. This test case of muscle fatigue shows that the
t1 I
muscle fatigue monitor does indeed measure the fatigue of a muscle. Further
testing to compare the results with existing muscle fatigue monitors would
also be desirable!,in the future.
*** BACK-MAN k
Bat 7.32 Ch A Ch B
Med Freq 08 28
Power 0000 0682
Figure 6. Sample LCD Display.


U> ^
o o
TI 228
| 240
a 252
| 264
I 276
N 208
in m *>j
a o o
Back-Man Muscle Fatigue Test #2 Startkig Spectrun

Back-Man Muscle Fatigue Test #2 Begbibtg & Ending
Spec trim
Figure 9. Combination of the Beginning and Ending Spectrum

9. Applications
The abilities of a portable, digital, real-time EMG monitor provide the
power required to meet the demands of many applications. Some of these
applications include:
Detection of muscle disease. Muscle disease may be identified
through spectral .analysis of the EMG signal. Thus, popular use of an EMG
monitor could aid in early detection of muscle disease and improve the odds
of recovery.
Physical therapy. With a portable device, physical therapists may use
an EMG monitor more frequently, quantifying the results of a therapy
session. Furthermore, there are muscle diseases where therapy can maintain
muscle strength. However, if the muscle is fatigued beyond a specific
threshold, additional muscle damage will result. A muscle fatigue monitor
can be used to prevent excessive fatigue from occurring.
Muscle pain diagnostics. Pain within a muscle may be derived from
different causes. Using a muscle fatigue monitor, it can be determined if the
muscle is contracted, or if it has shortened. Contraction is evident when the
muscle is hard and there is an EMG signal present. This indicates that a
treatment for muscle relaxation is required to remedy the situation. The
alternative scenario is that the muscle is hard and there is no EMG signal
present. This is usually the result of some form of posture guarding. This
happens when the body receives an injury and the muscle reacts to prevent

pain. Posture guarding results in a shortening of the muscle fibers and
requires that the!patient undergo physical therapy to re-stretch the muscle to
its original length.
Sports therapy. A portable EMG monitor would aid sports therapy
since it can be used in real activity, such as running, rowing, weight training,
bicycling, even swimming-provided the unit and electrodes were made water
Job environment analysis. Large corporations and insurance
companies are interested in providing a job environment that does not cause
muscle injury or excessive muscle fatigue. A portable EMG monitor will
allow more job positions to be evaluated to help improve the working
environment^ reducing muscle injury and workers' compensation claims.
Muscle injury fraud. Unfortunately, it has been insinuated that many
claims for workers' compensation have been deceitful. Imitation of muscle
injury is very easy for some people. The use of an EMG monitor can help
determine if the injuries are real and help everyone by reducing insurance
Relaxation training. EMG monitors are beneficial in helping a person
learn to relax. For example, stress may be the cause of many migraine
headaches. An EMG monitor can be used to determine stress by measuring
the facial muscles. Using the EMG signals, a person can be trained how to
relax, reducing stress and migraine headaches.

10. Recommendations and Improvements
The success of the EMG monitor developed in this thesis has left only
a few improvements and recommendations. The first, and main,
improvement would be to increase the execution time of the software. This
could be done through the use of some inline assembly code and increasing
the radix size of the FFT from two to four. The execution time could also be
improved by upgrading the processor to the new 16 MHz release of the Intel
The second major improvement would be to make the device even
smaller. The size of the EMG monitor could be reduced by using a
production style, surface mounted printed circuit board for the amplifier
sections. Also, a completely integrated design could be made that would
eliminate the Vesta board. The six AA nicad batteries could also be
rethought, possibly using only one or two sub-C cells.
The third major recommendation would be to add the capability for
user input-through a small key pad for example. This would allow the
custom use of the EMG monitor for many applications. Obviously, this
would also require additional software routines to be added to the basic
programs already written.
Finally, for special purpose applications a small speaker could be
added to provide audio feedback in addition to the visual feedback already
present. Audio sounds could be made to represent the median frequency

and/or the energy of the EMG signal. Sounds could be made to indicate
certain muscle fatigue thresholds. For example, these thresholds could
indicate over or under fatigue, which would be helpful for many muscle
therapy goals.

11. Conclusions
Frequency spectrum analysis of the EMG signal provides many
: :l
applications. Injthe past real-time EMG studies were limited to power level
diagnostics over a wide band of 25-1000 Hz and a narrow band of 100-200
Hz. The development of a portable, digital, real-time EMG monitor opens the
door of frequency spectrum analysis to many applications. The primary
purpose of this thesis was to develop such a unit and use it to measure muscle
fatigue. The author is very happy to report that this has been successfully
accomplished. The measurement of muscle fatigue is just a start of what a
device like this is capable of doing and many applications will become
possible for a portable, digital, real-time EMG monitor.


Appendix A: Pictorial View of the Muscle Fatigue Monitor
The physical dimensions of the proto-type, Back-Man, are 4.75" x
6.75" x 2.0". The complete system weight of the unit is less than 2 pounds.


B-2. Low Pass Filter

B-3. Final Amplifier Schematics

i-5 (V0)

i|= C2<
'A/O Inputs
10Q / *1.000
-*-5 JVB)
Jay Slondlford
Mle BackMon Plnol Goln Section
bn* L>ecwm*n\ numoi'
A 0M_3.clr 1.0
Ttprll t. iy?!J Ibheel 1 ol

B-4. Power Supply
_ 0300
. 6 AA

' ji c joo ---r~
5 o.i ^
i § r
1 0301 1 L
rv § t
I C3Q2* I 1 1-
ztz C302 I
T 0.1 uF
JP300 R300
^ 100 uF
__ C305
100 uP
__ csoe
T 100 uF
C3Q7 I
C307 ^ C309
1000 0.1
J; C30S jcsio
~p 10OO uf|~ 0.1
4.3 Veil* (for SBC 196)
'+5 Veits (for Amps)
4-5 Veils (for Dlaploy)
9 Veits (for Input Amps)
9 VeMs (for Input Amps)
Joy Stondlferd
i me Bock-Won Power Supply Section
Size" UGcumeni iNwmoer
A BM-4.CIR 1.0
Uoie April /. i^neet i oi

B-5. Display Logic

wmf m
Time (h:mm)
Appendix C: Test Data

Back-Man Shorted hput Noise
C-2. Shorted Input Noise
Note that;;this is +1 bit of the A/D at a gain of 1000. Thus, the noise
is less than +5 |iV referenced to the input.

BACK-MAN 50 Hz Input Wave
C-3. Input Sine Wave

BACK-MAN 50 Hz tfter flamntf VndOw
C-4. Sine Wave After The Hanning Window

C-5. The FFT and Median Frequency (without Hanning Window)

Back-Man JO Hz (tlamkig) ITT
C-6. The FFT and Median Frequency (with Hanning Window)

Appendix D: \ Sample Procedure in Forth
\ Procedure to calculate the Mag of the FFT output
\ Note this is approximate (see Williams paper)
\ JS ;i
\ '
N+l/2 );129 1 do
\ MAG_A(I) = IX(I)I + 375 IY(I) I
\ j = |X(I)I + I(Y(I)I right shift 3 times)*3
I AR @ abs R1 !
I AI @ alas R2 !
R1 @ R2 3(3 > if
R1 @ 3 -7 shift
else ;
R2 @ -7 shift
then '
R2 @
Rl @
-10 shift 3 *
-10 shift 3 *


Appendix E:
EMG Signal Test Data
E-l. Sample EMG Signal


E-2. Sample EMG Signal After the Hanning Window

Back-Man Frequency Spec trim cf OiG Signal
30 -r
E-3. Magnitude of EMG Signal

[1] N. Freunjdlich, "A High-Tech Probe of Your Aching Back," Business
Week, April 24, 1989.
[2] L. D. Gilmore and C. J. De Luca, "Muscle Fatigue Monitor (MFM):
Second Generation," IEEE Transactions on Biomedical Engineering,
vol. BME-32, no. 1, pp. 75-78, January 1985.
[3] J. R. Cram, Clinical EMGfor Surface Recordings: Volume 2, Clinical
Resources, Nevada City, California, 1990.
[4] C. J. De Luca, "Myoelectrical Manifestations of Localized Muscular
Fatigue in Humans," CRC Critical Reviews in Biomedical
Engineering, vol. 11, no. 4, pp. 251-279.
[5] R. Seroussi, M. H. Krag, P. Wilder, and M. H. Pope, "The Design and
Use Of a Microcomputerized Real-Time Muscle Fatigue Monitor
Based on the Medial Frequency Shift in the Electromyographic
Signal," IEEE Transactions on Biomedical Engineering, vol. BME-
36, n6. 2j pp. 284-286, February 1989.
[6] F. B. Stolen and C. J. De Luca, "Muscle Fatigue Monitor: A
Noninvasive Device for Observing Localized Muscular Fatigue,"
IEEE Transactions on Biomedical Engineering, vol. BME-29, no. 12,
pp. 760-7|]68, December 1982.
[7] L. Brodiejl Starting Forth, Second Edition, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 1987.
[8] C. S; Burrus and T. W. Parks, DFT/FFT and Convolution Algorithms,
Theory and Implementation, John Wiley & Sons, New York, 1985.
[9] E. O. Brigham, The Fast Fourier Transform, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 1974.

C. Williams, "Approximating the Square Root of the Sum of the
Squares," Micro, vol. 75, pp. 36-37, September 1984.
': j
M. Waite, "Build a Muscle Feedback Monitor," Popular Electronics,
pp. 39-42, May 1975.
A. K. Gcjdden, "Amplify biological signal with ICs," Electronic
Design, vol. 17, pp. 218-224, August 16,1969.
G. G. Naples, J. T. Mortimer, A. Scheiner, and J. D. Sweeney, "A
Spiral Nerve Cuff Electrode for Peripheral Nerve Stimulation," IEEE
Transactions on Biomedical Engineering, vol. 35, no. 11, pp. 905-
915, November 1988.
S. C. Jacobson, D. F. Knutti, R. T. Johnson, and H. H. Sears,
"Development of the Utah Artificial Arm," IEEE Transactions on
Biomedical Engineering, vol. BME-29, no. 4, pp. 249-269, April
1982. 'j
R. PallasjAreny, "Interference-Rejection Characteristics of
Biopotential Amplifiers: A Comparative Analysis," IEEE
Transactions on Biomedical Engineering, vol. 35, no. 11, pp. 953-
959, November 1988.
K. C. Mc Decomposition of the Clinical Electromyogram," IEEE Transactions
on Biomedical Engineering, vol. BME-32, no. 7, pp. 470-477, July
1985. j
G. N. Saridis, and T. P. Gootee, "EMG Pattern Analysis and
Classifica!tion for a Prosthetic Arm," IEEE Transactions on
Biomedical Engineering, vol. BME-29, no. 6, pp. 403-412, June 1982.
S. Losee,
"Pocket Ergometer," Fortune Magazine, pp. 138, November
P. Cassidy, "Back Track Fitness: Using the new Muscle Fatigue
Monitor, liioctors will tailor physical therapy to the individual," Omni,
pp. 24, October 1989.

S. Woolley, "Timing Your Ticker's Workout," Business Week, pp.
162, October 1989.
A. Van Boxtel and L. Schomaker, "Motor Unit Firing Rate During
Static Contraction Indicated by the Surface EMG Power Spectrum,"
IEEE Transactions on Biomedical Engineering, vol. BME-30, no. 9,
pp. 601-609, September 1983.
D. Hary.iM. Belman, J. Propst, and S. Lewis, "A Statistical Analysis
of the Spectral Moments Used in EMG Tests of Endurance," American
Physiological Society, pp. 779-783, April 1982.
G. Kondraske, and T. Carmichael, "Myoelectric Spectral Analysis and
Strategies for Quantifying Trunk Muscular Fatigue," Arch Phys
Medical Rehabilitation, vol. 68, pp. 103-110, February 1987.
L. Lindstrom, R. Kadefors, and I. Petersen, "An Electromyographic
Index fori
Localized Muscle Fatigue," IEEE Transactions on
Biomedical Engineering, vol. BME-24, pp. 750-754, April 1977.
R. Merletjti, D. Biey, M. Biey, G. Prato, and A. Orusa, "On-Line
Monitoring of the Median Frequency of the Surface EMG Power
Spectrum!," IEEE Transactions on Biomedical Engineering, vol.
BME-32,ijpp. 1-7, January 1985.
S. H. Roy, C. J. De Luca, and D. A. Casavant, "Lumbar Muscle
Fatigue and Chronic Lower Back Pain," Spine, vol. 14, no. 9, pp. 992-
1001, September 1989.
C. J. De Luca, NeuroMuscular Research Center, Activities Report
1989, Boston University, Massachusetts, 1989, pp. 1-22
J. S. Petrofsky, "Frequency and Amplitude Analysis of the EMG
During Exercise on the Bicycle Ergometer," Eur. J. Appl. Physiol,
vol. 41, pp. 1-15,1979.
C. Zetterberg, R. Bjork, R. Ortengren, and G. B. J. Anderson,
"Electromyography of the Paravertebral Muscles in Idiopathic
Scoliosis," Acta Orthop, vol. 55, pp. 304-309,1984.

[30] I. Yaar, A. R. Mitz, and E. W. Pottala, "Fatigue Trends IN and The
Diagnosis of Myasthenia Gravis by Frequency Analysis of EMG
Interferehce Patterns," Muscle & Nerve, vol. 8, pp. 328-335,1985.
[31] M. Hagb'erg, "Electromyographic Signs of Shoulder Muscular Fatigue
in Two Elevated Arm Positions," American Journal of Physical
Medicinej, vol. 80, no. 3 pp. 111-121,1981.
[32] R. E! Seroussi, and M. H. Pope, "The Relationship Between Trunk
Muscle Electromyography and Lifting Moments in the Sagittal and
Frontal Planes," The Journal of Biomechanics, vol. 20, no. 2, pp. 135-
146, 1987.
[33] K. R. Boff, and J. E. Lincoln, Engineering Data Compendium: Human
Perception and Performance, AAMRL, Wright-Patterson AFB, OH,
pp. 2022-2027,1988.
[34] P. H. Velltink, J. A. Van Alste, and H. B. K. Boom, "Influences of
Stimulation Conditions on Recruitment of Myelinated Nerve Fibers: A
Model Stjbdy," IEEE Transactions on Biomedical Engineering, vol.
35, no. 11, pp. 917-603, November 1988.
[35] R. Ortengren, G. Andersson, H. Broman, R. Magnusson, and I.
Petersen," Ergonomics, vol. 18, no. 2, pp. 157-174,1975.
[36] Intel, 16-Bit Embedded Controllers, Intel Corporation, 1991.
[37] R. W. Hamming, Digital Filters, Second Edition, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 1983.
[38] R. W. Ramirez, The FFT Fundamentals and Concepts, Prentice-Hall,
Inc., Englewood Cliffs, New Jersey, 1985.
[39] Vesta, SBC196 Hardware Manual, Vesta Technology, Inc., Wheat
Ridge, CO, 1991.
[40] Vesta!, Vesta Forth-83i96 Revision 1.68, Vesta Technology, Inc.,
Wheat Ridge, CO, 1991.
[41] C. H. Chen, Digital Waveform Processing and Recognition, CRC
Press, Inc|., Boca Raton, Florida, 1982.

[42] G. D. Bergland, "A Guided tour of the Fast Fourier Transform," IEEE
Spectrum, vol. 6, pp. 41-52, July 1969.
[43] C. L. Levkov, "Fast Integer Coefficient FIR Filters to Remove the AC
Interference and the High-Frequency Noise Components in Biological
Signal," Medical & Biological Engineering & Computing, pp. 330-
332, Ma} 1989.