Content uploaded by Garry T Allison
Author content
All content in this area was uploaded by Garry T Allison on Nov 12, 2020
Content may be subject to copyright.
The relationship between EMG median frequency and
low frequency band amplitude changes at
different levels of muscle capacity
G.T. Allison
a,*
, T. Fujiwara
b
a
The Department of Surgery, The Centre of Musculoskeletal Studies, The University of Western Australia, Rear 50 Murray Street,
Level 2 Medical Research Foundation Building, Perth, WA 6000, Australia
b
School of Allied Medical Professionals, Shinshu University, Matsumoto, Nagano 390-8621, Japan
Received 18 January 2002; accepted 23 April 2002
Abstract
Objective. To test the validity of high and low frequency band amplitudes of the surface electromyography (EMG) profile as
representation of muscle fatigue.
Design. A within subjects (n¼10) repeated measures design was used to collect surface EMG signals from the biceps during an
isometric contraction under two levels of fatigue status.
Background. The use of the shift in the median frequency of the surface EMG power spectrum is a well known method of as-
sessing muscle fatigue. Fatigue also results in amplitude changes of the specific frequency bands. The use of frequency band analysis
may be an alternative option for the assessment of muscle fatigue in specific experimental settings.
Methods. Surface EMG profiles of the biceps were recorded at 1024 Hz during a sustained isometric hold at 60% of the indi-
viduals fresh and fatigued maximal voluntary isometric torque. The median frequency of the power spectrum was compared with
changes in the low frequency (15–45 Hz) and high frequency (>95 Hz) bands.
Results. There was a close association between median frequency shift and the amplitude of the 15–45 Hz bandwidth and the
high–low frequency amplitude ratio. The association was similar for performance under different muscle capacity states.
Conclusions. Frequency band amplitude analysis provides similar information to median frequency shift under isometric con-
ditions and may be suited to specific experimental protocols in workplace fatigue studies.
Relevance
The use of amplitude band analysis that closely approximates the standard median frequency changes allows greater possibility of
assessing muscle fatigue in different experimental settings and the use of lower sampling rates.
Ó2002 Elsevier Science Ltd. All rights reserved.
Keywords: Fatigue; EMG amplitude; Fast Fourier transformation; Frequency band analysis
1. Introduction
Demodulation of the muscle activity profile derived
from surface EMG electrodes is used in many ways to
assess muscle performance. Assessment of the frequency
domain (power density spectrum (PDS)) has been used
in determining muscle fatigue responses. As a muscle
fatigues there is a concomitant change in the power
spectrum derived from surface electrodes where there is
an increase in the amplitude of the low frequency band
and a relative decrease in the higher frequencies. Phys-
iologically, the frequency shift has been attributed to
changes in conduction velocity, changes in intra-mus-
cular pH, modification in the recruitment and synchro-
nisation of the motor units and the fibre type [1–6].
Derived variables from the power spectrum repre-
senting the concept of fatigue include central tendency
measures (mean, peak and median) and ratios of the
power of high and low frequency bands. Since the PDS
from the surface electrodes is not normally distributed
there is an argument for the use of the median frequency
(MF) as an assessment of central tendency. Changes in
MF are associated with the high frequency fatigue of a
*
Corresponding author.
E-mail address: gta@cms.uwa.edu.au (G.T. Allison).
0268-0033/02/$ - see front matter Ó2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 2 6 8 - 0 0 3 3 ( 0 2 )0 0 0 3 3 - 5
Clinical Biomechanics 17 (2002) 464–469
www.elsevier.com/locate/clinbiomech
muscle and is highly correlated with a decline in force
from the fresh state [7]. Power spectrum estimates of
fatigue however are less sensitive to decline in muscle
capacity associated with (low frequency) fatigue fol-
lowing repeated fatiguing trials. This inability to identify
changes in torque generating capacity is also noted when
using ratios of the power of high and low frequency
bands [8].
The use of power ratios between different frequency
bands has a historical link to analogue processing prior
to digital processing age [8–10]. The use of high–low (H/
L) ratios remains in the research literature in circum-
stances where the data acquisition and the processing
using fast Fourier transformation (FFT) becomes
problematic compared to superficial surface EMG sig-
nal acquisition. These problems may include the non-
stationarity of the waveform seen in non-isometric
muscle activities or where the muscle moves relative to
the electrodes e.g. oesophageal electrodes for diaphrag-
matic muscle activity recordings [11–14].
The surface power spectrum is predominantly less
than 500 Hz and therefore according to the Nyquist
theorem the surface EMG signal should be recorded at
least 1 kHz. The resolution of the power spectrum is
frequency dependent and therefore to date there are
limited methods that provide a fatigue index from the
surface EMG profile without high sampling rates.
High frequency acquisition rates, simultaneous re-
cordings from multiple muscles, and the long duration
of fatigue tasks, all contribute to the need for re-
searchers to collect epochs of data during the fatigu-
ing task. Modern data acquisition systems with large
memory capacity however may make it possible to
overcome many of these problems. Yet as the systems
increase capacity the researchers tend to have a con-
comitant increase in the number of muscles or duration
of the task. Therefore low frequency band amplitude
methodologies may be developed to maximise experi-
mental opportunities in the assessment of tasks that
induce muscle fatigue. This may have specific applica-
tion to data logging assessments in specific experimental
settings. For a more recent example, low frequency band
technique was a more reliable index of fatigue in back
fatigue tasks when compared to changes in MF [15].
The purpose of this study was twofold, first, to deter-
mine if there is a clinical significant association between
H/L ratio and low frequency band amplitude change
and MF changes during an isometric fatigue tasks.
Secondly, to determine if these relationships are inde-
pendent of the torque producing capacity of the muscle.
2. Methods
Five males 26.4 (SD 4.8) years and 5 females 32.3 (SD
4.7) years volunteered to participate in the experiment
testing their muscle response to two fatiguing tasks. The
study received approval from the human research ethics
committee and all subjects signed an informed consent
document.
All subjects sat with their feet flat on the floor and
hips and knees at right angles. The subject positioned
their dominant arm by their side with the elbow at 90
flexion. With their wrist supinated they held a hand grip
that was attached to a force transducer. In this position
their maximal voluntary isometric contraction was per-
formed three times against an unmoveable load. The
highest value was recorded prior to each endurance task.
The endurance task was designed to fatigue the elbow
flexors and was performed twice with 2 min recovery
between trials. A 10 cm visual analogue scale was used
to determine the perceived ‘‘local muscle fatigue’’ before
the first trial and after each endurance task.
The endurance elbow flexion task was a sustained
isometric load set at 60% of the MVIC for each trial.
This value was entered into a visual and audio feedback
system on a Power Macintosh computer running Lab-
view 4.0. The custom built Labview program provided
visual feedback for the target force in the form of a
moving bar chart.
Subjects were asked to hold at 60% of their previous
recorded maximum for as long as possible or until they
reached 35% of their value.
The EMG signals from the biceps brachii were re-
corded using a surface electrode bipolar configuration
with an inter-electrode distance of 25 mm aligned par-
allel to the fibres of the biceps. An earth electrode was
placed on the medial epicondyle of the elbow. All elec-
trode sites were prepared with a razor to remove any
hair and then vigorously rubbed with an abrasive gel
(SkinPureâNihon Kohden, Tokyo, Japan). Inter-elec-
trode skin impedance was accepted if below 10 kX.
The data were collected using BIMUTAS software
(Nihon Kissei Comtech, Matsumoto, Japan) for the
duration of the test at 1024 Hz (16 bit A–D converter)
and stored to disc. The data were then filtered digitally
using 20–500 Hz bandpass filter. The force trace was
visually inspected and the initial and final end compo-
nents of less than 1 s were excluded from further anal-
ysis.
Ten single epochs (deciles) of the data were then
created to allow group comparisons of time normalised
data. For the frequency analysis the power spectrum
was calculated for each decile and the MF (the fre-
quency where the power of the FFT derived power
spectrum is halved) was calculated. For each decile the
integrated EMG (IEMG) was calculated for the fre-
quency bands 15–45 Hz (low frequency band) from 45–
95 Hz (medium frequency band) and above 95 Hz high
frequency band.
MF (Hz) were recorded for each trial for each decile
and low frequency band amplitude data were recorded
G.T. Allison, T. Fujiwara / Clinical Biomechanics 17 (2002) 464–469 465
relative to total integrated EMG and expressed as a
percentage. The high/low frequency band ratio was de-
rived from the amplitude assessments of the high and
low frequency band amplitudes.
The MF and two different frequency band amplitude
assessments for both trials were compared using a sec-
ond order polynomial least squares best fit algorithm.
This resulted in second order polynomial equations be-
ing generated for changes in MF and changes in
low frequency band amplitude (two fatigue states) and
changes in MF and the high/low frequency band ratio
(two fatigue states). The root mean square (RMS) of the
residuals was recorded for each of these four equations.
3. Results
3.1. Performance and perception of effort
The duration of rest between the endurance trials was
set at 2 min. This clearly did not allow full recovery of
the elbow flexors. This is noted in a significant
(P<0:0001) 35.8% decreased MVIC performance (ar-
bitrary units) from the first trial 720 (316) to the second
trial 462 (178). The first decile (initial data set) for the
absolute power in the low frequency band between trials
did not significantly differ (P¼0:8193). Statistically
significant differences between initial MF and initial H/
LFB ratio between trials were noted. The MF decreased
(P¼0:0120) from a mean of 85.8–81.5 Hz. The high/
low frequency band ratio decreased (P¼0:0131) from
3.1 to 2.4.
Fig. 1 shows the change in torque production for all
subjects for both trials across the 10 deciles of perfor-
mance normalised to their maximal isometric contrac-
tion prior to each test occasion. The rate of torque
production fatigue for both trials was highly correlated
(r¼0:98). Fig. 2 also shows the mean (SD) of the per-
ceived fatigue of the elbow flexors before each trial and
after the second endurance task. The degree of perceived
fatigue was greater (mean 57, SD 18; mean 71, SD 17)
after the second test and each score suggests that all
subjects reported significant fatigue follow both endur-
ance tasks.
3.2. Amplitude changes
All subjects demonstrated the representative shift in
the high and low frequency band amplitudes during
each endurance task. The EMG amplitude of total sig-
nal for each decile for each frequency band of both
endurance tasks is illustrated in Fig. 2. The first trial
shows a greater variance between deciles when com-
pared to the second trial. There was a slight trend to-
wards an increased amplitude in both trials, however
this was true for about 70% of the endurance time. After
this point the amplitude tended to level off. This was
clearer in the second trial, where amplitudes were more
stable. The EMG amplitude of the first trial was greater
when compared to the second trial and can be explained
by a significantly lower absolute load (constant relative
load) during the second test. The decrease in total
EMG between trials, did not match the decrease in the
torque production between trials. This indicates a de-
crease change in the EMG amplitude/force ratio (more
integrated EMG amplitude per unit force) between trials
and reflects the cumulative affects of fatigue.
Fig. 2 also shows that in both trials there was a linear
increase in low frequency band amplitude with a con-
comitant decrease in the relative contribution of the high
frequency band. This increase in amplitude is shown in
Fig. 3 for each trial. Fig. 3 also shows the relative de-
crease in MF for each trial. Divergence between the two
trial data can be seen only in the latter 3 deciles of the
MF data and the last 2 deciles of the low frequency band
data.
3.3. Frequency changes
Figs. 4 and 5 show the superimposed data sets of both
trials where MF (Hz) was highly associated with the
Fig. 1. The lines show the mean (SD) of each decile of normalised
torque production for both trials. The bars show the mean (SD) of the
perceived local muscle fatigue (10 cm VAS) reported by each subject
before and after each trial.
Fig. 2. The amplitude of total integrated EMG for the signal for each
decile of both endurance task for each of the low (15–45 Hz), median
(45–95 Hz) and high frequency (>95 Hz) bands.
466 G.T. Allison, T. Fujiwara / Clinical Biomechanics 17 (2002) 464–469
percentage low frequency band and the high/low fre-
quency band ratio. A second order polynomial describes
the association between the percentage low frequency
band and the MF change for trial 1 (R2¼0:894 RMS
error 5.1 Hz) and trial 2 (R2¼0:928 RMS error 4.1 Hz).
Fig. 5 shows the association between MF and high/
low frequency band ratio. A second order polynomial
described the association for trial 1 (R2¼0:957 residual
RMS error is 3.22 Hz) and trial 2 (R2¼0:964 residual
RMS error 2.9 Hz). The greatest source of error for both
sets of data were obtained early in the testing on each
trial.
4. Discussion
4.1. Cumulative effects of fatigue
The decrease in maximal voluntary isometric capac-
ity and increase of perception of local muscle fatigue
demonstrated that the subjects had a cumulative local
muscle fatigue affect from one trial to the other. Rest
periods of greater than 5 min have been suggested for
complete recovery for the elbow flexor high frequency
fatigue [2] therefore the short rest period in this study
was expected to have a large degree of change between
trials in both torque production and the EMG para-
meters.
Although there was a clear decrease of maximal
muscle capacity between trials there was little evidence
that the gross pattern of torque production fatigue was
fundamentally different between the two tests. This is a
result of using a sub-maximal criteria related to their
current muscle capacity (i.e. lower absolute load for the
second testing), and the use of time normalisation for all
data sets. The second test consistently lies below the
initial trial (Fig. 2). This is a result of the initial decile
(used for normalisation) being relatively greater in the
second trial compared to the first trial. This may re-
flect both central and peripheral factors associated with
muscle performance of repeated fatigue tasks with a
short 2 min recovery period. A greater initial decrease in
performance in the second trial may have been associ-
ated with a decreased recovery of the high frequency
fatigue component and remaining intra-muscular me-
tabolites. The latter has been highly associated with
muscle performance during fatigue tasks. Although real-
time feedback was identical in both trials this study did
not differentiated between the psychological (central)
and peripheral factors. Psychological factors may con-
tribute to as much as 20% of the variance in maximal
performance tasks [16]. Clearly the fact that the subjects
reported high perceptions of local muscle fatigue, the
initial high burst of muscle torque during the second
trial may be motivational and this was not explicitly
controlled in this study.
The low frequency band amplitude changes demon-
strated similar pattern of increase for both trials. There
was no statistical difference in the initial values between
Fig. 3. The mean (SD) of the change in MF of the power spectrum for
all subjects for both trials (top). The mean (SD) of the amplitude
change of the low frequency band for both trials (bottom). Both data
sets are expressed as a relative change normalised to the first decile.
Note the last decile was not included since there was large variance in
the data as the individual reach exhaustion.
Fig. 5. The association between changes in MF of the power spectrum
and the percentage of the total integrated EMG derived from the low
frequency band during two trials of isometric elbow flexion fatigue.
Note: increasing low frequency band percentage represents fatigue
response.
Fig. 4. The association between changes in MF of the power spectrum
and the high–low frequency amplitude ratio during two trials of iso-
metric elbow flexion fatigue. Note decreasing ratio corresponds to
fatigue.
G.T. Allison, T. Fujiwara / Clinical Biomechanics 17 (2002) 464–469 467
trials. This infers that the parameter is not related to
decrease in the initial torque production capacity of the
muscle between trials. It is acknowledged, however, that
this may represent a type II statistical error due to the
large variance in the data. In comparison, there was a
significant systematic decrease in both initial MF and
high/low ratio between trials.
The magnitude of the change in the initial MF values
was relatively small (5%) compared the decrease in the
maximal isometric torque production (36%). This dem-
onstrates that the initial MF is largely independent of
the torque generating status of the muscle although
there is a suggestion that some residual high frequency
fatigue in the elbow flexors following the rest period.
This suggestion however can not be quantified for this
study since assessment of high frequency fatigue using
electrical stimulation was not used. Although the abso-
lute magnitude of the inter-trial change in the initial
high /low frequency band ratio was larger (22.5%), the
probability of determining statistical differences between
trials were the same for each derived parameter (P
0:013). The findings of this study suggest that the pa-
rameters are measuring essentially the same domain of
fatigue––an element of high frequency fatigue. Both are
therefore limited in their assessment of cumulative ef-
fects of repeated trials with short rest periods and their
ability to quantify low frequency fatigue [8].
4.2. Methodological issues
Research in the 1970s examined the relationship be-
tween the high/low frequency band ratios and fatigue.
The basis of these studies was the poor digital capacity
of data acquisition systems. Therefore the high/low ra-
tio and low frequency band amplitude technique could
be performed using analogue techniques (filtering fol-
lowed by integration) prior to the data acquisition or
chart recording [8–10]. Since the improvements in digital
technology the focus of post-processing has moved away
from analogue techniques. The use of FFT and the MF
shift is now the choice of assessment of muscle fatigue
derived from the surface EMG profile. This requires
high sampling rates and stationarity of the surface
EMG signal or alternatively newer dynamic or non-
linear processing protocols [11,12].
The findings of this study demonstrate that the
changes in High/Low frequency band ratio and ampli-
tude of the percentage low frequency band correlate
strongly with the changes in the MF during fatiguing
isometric tasks. The former variables can be derived in
an analogue setting and therefore recorded at low fre-
quencies (e.g. 20 Hz) in concert with other transducer
information such as angle and force.
It would seem that these amplitude data (like MF
analysis) reflect the selective high frequency fatigue
components and are independent of the cumulative (low
frequency fatigue) effects of repeated trials with mod-
erate rest periods. The RMS error in predicting the
change in MF for high/low frequency band ratio was
less than the low frequency band method. Both tech-
niques however generated residual errors of less than 5
Hz with a discernible improvement on the second trial.
The 5 Hz predictive error compares favourably with
reliability data of initial MF values 95% tile CI of 10
Hz [17]. The clinical and experimental impact of this
needs further consideration.
This study found that a second order polynomial
fitted both curves. Other types of curves were also ap-
plicable to describing the data set. Future studies may
find different relationships when post-processing ampli-
tude assessments are based on electrical power (RMS) as
opposed to average voltage (IEMG).
The low frequency band in this study was between 15
and 45 Hz. The high pass frequency of 15 Hz was se-
lected to avoid movement artefact even in the most
dynamic tasks. The low frequency cut-off was selected
to limit any contribution from mains (50 or 60 Hz) or
electrical motor noise. The high/low cut-off ranges vary
from older studies within the literature with the main
difference in the selection of the high frequency band.
Bai et al. [18] and Esau et al. [19] used a H/LFB ratio of
130–238/20–40 Hz. Other studies utilised different ran-
ges [14] 150–350/20–46.7 Hz and have modified these
ranges for different muscles and adapted to finewire
electrodes. For surface EMG the power of the signal
above 500 Hz is limited and therefore the upper limit is
unlikely to impact on the selection of the range as much
as the low cut-off frequency value. Further research to
optimise the selection of the lower cut-off for the high
frequency band should be considered in the context of
other muscles and non-isometric actions, however the
selection is likely to be limited by the configuration of
(analogue) hardware.
5. Conclusion
The decrease in MF from surface EMG profiles is
a recognised method of determining (high frequency)
fatigue in an isometric muscle action. The methods
employed to generate MF data however often rely on
stationary waveforms and high sampling rates. This
study was able to demonstrate that relative changes in
the amplitude of specific frequency bands are highly
correlated to changes in MF. Both frequency banding
techniques, like MF changes, are sensitive to elements of
fatigue that have a shorter recovery time than those
associated with muscle performance or perception of
local muscle fatigue. Therefore the frequency banding
technique may be an alternative methodological con-
sideration where experimental circumstance are indi-
cated. The results of this study warrant examination of
468 G.T. Allison, T. Fujiwara / Clinical Biomechanics 17 (2002) 464–469
similar EMG processing techniques in more dynamic
and repetitive muscle activities.
References
[1] Basmajian JV, De Luca CJ. Muscles alive: their functions
revealed by electromyographyed. Baltimore: Williams & Wilkins;
1985.
[2] Rainoldi A, Galardi G, Maderna L, Comi G, Lo Conte L,
Merletti R. Repeatability of surface EMG variables during volun-
tary isometric contractions of the biceps brachii muscle. J Elec-
tromyogr Kinesiol 1999;9:105–19.
[3] Gerdle B, Karlsson S, Crenshaw AG, Elert J, Friden J. The
influences of muscle fibre proportions and areas upon EMG
during maximal dynamic knee extensions. Eur J Appl Physiol
2000;81:2–10.
[4] Solomonow M, Baten C, Smit J, Baratta R, Hermens H,
D’Ambrosia R, et al. Electromyogram power spectra frequencies
associated with motor unit recruitment strategies. J Appl Physiol
1990;68:1177–85.
[5] Roy SH, De Luca CJ, Emley M, Oddsson LI, Buijs RJ, Levins
JA, et al. Classification of back muscle impairment based on the
surface electromyographic signal. J Rehab Res Dev 1997;34:405–
14.
[6] Wretling ML, Henriksson-Larsen K, Gerdle B. Inter-relationship
between muscle morphology, mechanical output and electromy-
ographic activity during fatiguing dynamic knee-extensions in
untrained females. Eur J Appl Physiol Occup Physiol 1997;
76:483–90.
[7] Mannion AF, Dolan P. Relationship between myoelectric and
mechanical manifestations of fatigue in the quadriceps femoris
muscle group. Eur J Appl Physiol 1996;74:411–9.
[8] Moxham J, Edwards RH, Aubier M, De Troyer A, Farkas G,
Macklem PT, et al. Changes in EMG power spectrum (high-
to-low ratio) with force fatigue in humans. J Appl Physiol
1982;53:1094–9.
[9] Gross D, Grassino A, Ross WR, Macklem PT. Electromyogram
pattern of diaphragmatic fatigue. J Appl Physiol 1979;46:1–7.
[10] Bellemare F, Grassino A. Evaluation of human diaphragm
fatigue. J Appl Physiol 1982;53:1196–206.
[11] Filligoi G, Felici F. Detection of hidden rhythms in surface EMG
signals with a non-linear time-series tool. Med Eng Phys 1999;21:
439–48.
[12] Knaflitz M, Bonato P. Time–frequency methods applied to muscle
fatigue assessment during dynamic contractions. J Electromyogr
Kinesiol 1999;9:337–50.
[13] Webber CL, Schmidt MA, Walsh JM. Influence of isometric
loading on biceps EMG dynamics as assessed by linear and
nonlinear tools. J Appl Physiol 1995;78:814–22.
[14] Gallagher C, Hof V, Younes M. Effect of inspiratory muscle
fatigue on breathing pattern. J Appl Physiol 1985;59:1152–8.
[15] Dolan P, Mannion AF, Adams MA. Fatigue of the erector spinae
muscles. A quantitative assessment using ‘‘frequency banding’’ of
the surface electromyography signal. Spine 1995;20:149–59.
[16] Kent-Braun JA. Central and peripheral contributions to muscle
fatigue in humans during sustained maximal effort. Eur J Appl
Physiol 1999;80:57–63.
[17] Elfving B, Nemeth G, Arvidsson I, Lamontagne M. Reliability of
EMG spectral parameters in repeated measurements of back
muscle fatigue. J Electromyogr Kinesiol 1999;9:235–43.
[18] Bai TR, Rabinovitch BJ, Pardy RL. Near-maximal voluntary
hyperpnea and ventilatory muscle function. J Appl Physiol 1984;
57:1742–8.
[19] Esau SA, Bellemare F, Grassino A, Permutt S, Roussos C, Pardy
RL. Changes in relaxation rate with diaphragmatic fatigue in
humans. J Appl Physiol 1983;54:1353–60.
G.T. Allison, T. Fujiwara / Clinical Biomechanics 17 (2002) 464–469 469