ORI GI NAL ARTI CLE

Changes in surface EMG assessed by discrete wavelet transform

during maximal isometric voluntary contractions following
supramaximal cycling
Luis Pen ailillo

Rony Silvestre

Kazunori Nosaka
Received: 18 April 2012 / Accepted: 11 September 2012 / Published online: 23 September 2012
Springer-Verlag 2012
Abstract To better understand characteristics of neuro-
muscular fatigue in supramaximal cycling exercise, this
study examined changes in surface electromyography
(sEMG) frequency during maximal voluntary isometric
contractions (MVC) following a 30-s Wingate anaerobic
test (WAnT) using discrete wavelet transform (DWT). The
changes in sEMG were also compared between DWT and
mean frequency (MNF) obtained by fast Fourier transform
(FFT). 17 healthy men performed a WAnT with a 7.5 % of
body mass load. Knee extensor MVC torque was measured
before and 1, 3, 6, 9, 12 and 15 min following WAnT, and
sEMG was recorded from vastus lateralis muscle during the
torque measures. sEMG was analysed for (RMS), MNF by
FFT and frequency domains of DWT (divided into six
domains). MVC torque decreased 2123 % at 315 min,
RMS increased 2634 % at 115 min, and MNF decreased
810 % from baseline (76.3 3.2 Hz) at 13 min post-
cycling (P\0.05). The DWT frequency domains showed
that the changes lasted longer than MNF such that the
intensity increased at 12 and 15 min for domain 2
(125250 Hz), all time points for domain 3 (62.5125 Hz),
and 16 min for domains 4 (31.262.5 Hz) and 5
(15.631.2 Hz). The magnitude of increase in the intensity
at 1 min post-exercise (4560 %) was largest for domains
3 and 5 (P\0.05). A signicant correlation was evident
only between the magnitude of changes in the domain 5
and MNF (r = -0.56). It is concluded that DWT provides
information on neuromuscular fatigue that is not detected
by MNF derived from FFT.
Keywords Wingate anaerobic test Neuromuscular
fatigue Frequency spectrum Maximal voluntary
isometric contraction Root mean square Fast Fourier
transform (FFT) Mean frequency
Introduction
Surface electromyography (sEMG) provides information of
global motor unit activation, and sEMG signals have been
analysed for their changes in amplitude and frequency in
the investigation of neuromuscular fatigue in various
exercises (Cifrek et al. 2009; Hunter et al. 2003). A Win-
gate anaerobic test (WAnT) is used to examine anaerobic
capacity and also as a model of supramaximal exercise to
investigate neuromuscular fatigue (Patton et al. 1985;
Bogdanis et al. 1995). Previous studies (Hunter et al. 2003;
Greer et al. 2006; Camata et al. 2010; Stewart et al. 2011)
examined changes in sEMG frequency obtained by a fast
Fourier transform (FFT) during WAnT, but controversy
exists regarding the changes in mean frequency (MNF)
derived from FFT.
Two studies reported a signicant decrease (1015 %)
in MNF over 30-s of WAnT (Hunter et al. 2003; Greer
et al. 2006), but other two studies (Camata et al. 2010;
Stewart et al. 2011) did not nd signicant changes in
MNF during WAnT. No previous study has examined
Communicated by Arnold de Haan.
L. Penailillo
School of Kinesiology, Ponticia Universidad Catolica de
Valpara so, Valpara so, Chile
L. Penailillo (&) K. Nosaka
School of Exercise and Health Sciences, Edith Cowan
University, 270 Joondalup Drive, Joondalup,
WA 6027, Australia
e-mail: l.penailillo@ecu.edu.au
R. Silvestre
School of Kinesiology, Universidad Mayor, Santiago, Chile
1 3
Eur J Appl Physiol (2013) 113:895904
DOI 10.1007/s00421-012-2499-1
changes in MNF after WAnT. To better understand neu-
romuscular fatigue and recovery, it is important to examine
post-WAnT frequency changes. It is also interesting to
know how long the change (e.g. decreases) in MNF lasts
after WAnT. It has been documented that further decreases
in pH and ATP are seen after than during WAnT (Bogdanis
et al. 1995). Thus, it is possible that further and greater
changes in sEMG frequency could be observed following
WAnT than during WAnT; however, this has not been
investigated.
Wavelet transform is a timefrequency method for sig-
nal analysis, which has been proposed to be more appro-
priate to analyse sEMG signals than FFT, because the
wavelet transform is better than FFT for non-stationary
sEMG signals (Karlsson et al. 2000), and wavelet function
captures spiky signals of sEMG more efciently than FFT
(Karlsson et al. 1999). The wavelet transform analysis
expresses a signal as a linear combination of a particular set
of functions obtained by shifting and dilating one single
function called mother wavelet, and the correspondence
between the resulting wavelet shape and the signal is cal-
culated (Subasi and Kiymik 2010). The intensity of this
coefcient is a measure of how much of the wavelet at a
particular scale and time point (i.e. frequency range) is
included in the signal (Subasi and Kiymik 2010). Since the
mother wavelet can be moved to various locations of the
signals and can be dilated (matching better with slow
component of the signal) or squeezed (matching better with
fast component of the signal), the wavelet transform pro-
vides a mapping of the signal and decomposes the signal
into several frequency domains that provide information
regarding a specic portion of the frequency spectrum at a
specic time (Subasi and Kiymik 2010; Letelier and Weber
2000). In the wavelet transform analyses, fast components
of sEMG signals (sharp edges and steep slopes) are rep-
resented by high-frequency domains and slower compo-
nents of the signals (repolarisation or baseline uctuations)
are represented by low-frequency domains (Letelier and
Weber 2000). Discrete wavelet transform (DWT) is a type
of wavelet transform and requires less computational pro-
cessing time than other wavelet transform techniques such
as continuous wavelet transform and wavelet packet
decomposition, and provides high-quality timefrequency
resolution (Constable and Thornhill 1993).
Several studies (Beck et al. 2005; Constable et al. 1994;
Kumar et al. 2003) used a wavelet transform for sEMG
analyses in relation to neuromuscular fatigue. For example,
Beck et al. (2005) compared between FFT and DWT for
sEMG frequency changes over 50 maximal isokinetic
concentric contractions of the elbow exors, and found that
changes in MNF of FFT and the wavelet centre frequency
were similar, both showing signicant decreases in fre-
quency over contractions, suggesting that both FFT and
wavelet analyses lead to the same physiological interpre-
tations. Using a different wavelet transform, So et al.
(2009) investigated changes in 11 wavelet domains during
50 maximal isokinetic concentric contractions of the knee
extensors, and reported that the intensity of low-frequency
domain with the centre frequency (CF) of 19.3 Hz
increased in vastus lateralis muscle over 50 contractions.
They speculated that the increases in the low-frequency
domain were attributed to decreases in conduction velocity,
which might indicate greater fatigue in fast than slow
twitch muscle bres. However, no previous studies have
assessed DWT changes following WAnT and the rela-
tionship between changes in individual DWT domains and
MNF after WAnT.
The present study therefore investigated frequency
changes in sEMG during maximal isometric contractions
of the knee extensors following a WAnT using FFT and
DWT, and assessed the relationship between the changes in
MNF and DWT intensity. We hypothesised that (1) MNF
would decrease and the intensity of low-frequency DWT
domains would increase following WAnT, and (2) the
changes in MNF would be signicantly correlated with the
changes in the intensity of low-frequency DWT domains.
Methods
Subjects
Seventeen healthy men (mean SD: age, 24.4 2.6 years;
height, 173.0 5.5 cm; body weight, 72.7 8.8 kg) par-
ticipated in this study. The Institutional Human Research
Ethics Committee reviewed and approved the study, and all
subjects gave their written informed consent before partici-
pation. All subjects had not been involved in any regular
training programme for at least 6 months prior to the study,
and did not have any musculoskeletal or neurological dis-
orders of lower limbs. They were instructed not to perform
any strenuous exercise at least 48 h before the testing day,
and to avoid caffeinated and alcoholic beverages for at least
8 h prior to the participation in the experiment.
Protocol
All subjects performed WAnT in the afternoon
(14:0017:00) at least 1.5 h after lunch, and several mea-
surements shown below were taken before and for 15 min
following WAnT. Maximal voluntary isometric contraction
(MVC) torque of the non-dominant knee extensors were
measured before and 1, 3, 6, 9, 12 and 15 min following
WAnT, and during which sEMG of vastus lateralis (VL)
muscle was recorded. In order for the subject to perform
896 Eur J Appl Physiol (2013) 113:895904
1 3
MVC immediately after the completion of WAnT, a cycle
ergometer was placed right next to a chair for the MVC
measurement (explained below), and each subject was
assisted by the investigator to have the MVC measures
within 1 min after the WAnT. sEMG electrodes were kept
on the subjects leg during WAnT, and were connected to a
PowerLab system (explained below) as soon as possible.
sEMG was analysed for root mean square (RMS) and
MNF, and by DWT. Blood lactate concentration was
assessed before each MVC torque measure.
Wingate anaerobic test (WAnT)
Ten minutes after the baseline measurements, the subjects
performed 5 min cycling warm-up with a 2 % of their body
weight load at a self-selected cadence, and two 5-s sprints
were included at the third and fourth minute. 2 min after
warm-up, subjects started the WAnT by cycling without a
load to reach the maximal pedalling rate for 3 s on a
mechanically braked ergometer (Enermax, SINEBI,
Argentina) and a load of 7.5 % of their body weight (range
4.57.1 kg) was suddenly added. They were encouraged to
pedal as hard and fast as possible for next 30 s. During the
test, the average power output of every 5-s segment (05,
510, 1015, 1520, 2025, 2530 s) was calculated, and
peak power output, mean power output over 30 s and
fatigue index [(highest 5 s peak power - lowest 5 s peak
power)/highest 5 s peak power] were obtained (Armstrong
et al. 1983; Reiser et al. 2002).
Measurements
Maximal voluntary isometric contraction (MVC) torque
Subjects were seated in a chair adapted for this study to
keep the hip joint at 1208 exion and the knee joint at 908
exion, and knee extension torque of the non-dominant leg
was measured by a load cell (MLT003/D, AdInstruments,
Australia) connected to a PowerLab system ML818
(AdInstruments, Australia). The load cell was placed on the
lever arm attached to the chair, and positioned 8 cm above
the medial tibial malleolus. Subjects were asked to exert
force against the lever arm as hard and fast as possible
and verbal encouragement was given. For the measurement
taken before WAnT, three 3-s MVC torque measures with
1 min rest between measures were performed, and the
average of the three peak values was used for further
analysis. In the measurements after WAnT (1, 3, 6, 9, 12
and 15 min post), only one MVC torque measure (3 s) was
taken for each time point.
Blood lactate
A blood sample of 25 ll was collected from the earlobe by
a capillary tube prior to each MVC measure and was
analysed for blood lactate concentration using a YSI 1500
Sport Lactate Analyzer (YSI life Science, OH, USA).
sEMG recording and analyses
Two rectangular (20 9 30 mm) AgAgCl electrodes
(Skintact, Austria) were placed on the VL of the non-
dominant leg at two-third on the line from the anterior
superior iliac spine to the lateral side of the patella
(according to SENIAM) separated by 20 mm (Hermens
et al. 2000), and the ground electrode was placed on the
medial surface of tibia, after the skin was shaved and
cleaned with 95 % alcohol to reduce impedance to less
than 5 kX. During the 3-s MVC strength measures, sEMG
signals were recorded with the PowerLab system using a
Chart 5.5 software (AdInstruments, Australia) with a
sampling frequency of 1,000 Hz. The sEMG signals were
digitised on-line and stored in a computer for later off-line
analysis with an IGOR Pro 6.0 software (WaveMetrics,
Portland, USA). An epoch of the central 2 s of the raw
sEMG signal during which the MVC was plateau was
extracted for analyses (Merletti and Parker 2004). Using
the 2-s epochs, RMS of the amplitude was calculated, and
the MNF of the power spectrum density was obtained from
a FFT using 2,048 points (Cifrek et al. 2009; Hostens et al.
2004).
Wavelet transform processing
DWT was applied for the sEMG signals of the 2-s epoch
(Fig. 1) using a DWT toolbox of the IGOR Pro software.
The DWT toolbox with the Daubechies (db4) mother was
used to calculate the intensity of the wavelet coefcient
(Beck et al. 2005) for the eight domains (Table 1). How-
ever, since most of the sEMG frequency spectrum is in the
range of 10250 Hz (Wakeling et al. 2001), only the
middle six domains (domains 27) were focused in the
present study. As shown in Fig. 1, the timefrequency
matrix is shown by a grey scale for each domain (18) for
the 2-s epoch, and the intensity of the wavelet coefcient is
shown as white representing the highest wavelet coefcient
intensity and black indicating the lowest intensity of the
wavelet coefcient. The average intensity of the absolute
value of the wavelet coefcient for the 2-s epoch was
obtained for each domain, and percentage of change in the
intensity from baseline to 1, 3, 6, 9, 12 and 15 min post-
cycling was calculated.
Eur J Appl Physiol (2013) 113:895904 897
1 3
Statistical analysis
Changes in WAnT power output, MVC, blood lactate,
RMS, MNF and average of the wavelet coefcients over
time were assessed using a one-way repeated measures
analysis of variance (ANOVA). When a signicant time
effect was found, a Scheffes post hoc test was applied. To
compare the changes in the average intensity of the DWT
coefcient over time amongst six domains, a two-way
repeated measures ANOVA (domain 9 time) was used,
followed by post hoc pairwise comparisons using a Fishers
least signicant difference (LSD) test. Relationships
between the percentage change in MNF and the intensity of
each DWT domain from baseline were assessed using a
Pearson product moment correlation coefcient. The sig-
nicance level was set at P\0.05. All statistical analyses
were performed with PASW Statistics 18 software for Mac
(SPSS Inc, IBM Company, USA). Data are presented in
mean standard error of the means (SEM).
Results
WAnt
All subjects completed the exercise as instructed. The peak
power and the mean power output were 565.6 69.9 and
476.5 60.8 W, respectively, and the relative power per
body weight was 7.8 0.6 W kg
-1
for the peak and
6.6 0.4 W kg
-1
for the mean power. Figure 2 shows
changes in the power output normalised by body weight for
30 s. Power output decreased signicantly over time, and
the fatigue index was 33.2 8.8 %.
MVC torque
Pre-exercise MVC torque was 324.6 146.1 Nm (range
213464 Nm), and MVC torque decreased signicantly at
3 min (23.1 14.2 %), remaining low for 15 min post-
exercise without signicant changes between 3 and 15 min
(Fig. 3a).
Blood lactate
Blood lactate concentration increased signicantly from
pre (1.6 0.1 mmol l
-1
) to 1 min post-exercise (11.5
1.1 mmol l
-1
), further increased up to 6 min (15.3 0.5
mmol l
-1
) and remained the level until 15 min post-
exercise (Fig. 3b).
Root mean square
sEMG RMS increased signicantly following the exercise
by 34.4 8.3 % at 1 min, remaining the level for the rest
of recovery time (Fig. 4a).
MNF
As shown in Fig. 4b, MNF decreased signicantly from pre
(76.3 3.2 Hz) to 1 min (69.0 2.7 Hz) and 3 min
Fig. 1 sEMG signal processing for DWT domain analyses. From the
3-s sEMG signal during MVC, a central window of 2 s was selected.
DWT divides the signal into eight dyadic domains, and presents a
timefrequency matrix that shows time (2 s) at the X-axis and the
wavelet domains 18 as Y-axis. The intensity of the wavelet
coefcient is shown by a grey scale at the right side of the time
frequency matrix, where white indicates maximum intensity and black
shows the lowest intensity. For this example, the average intensities
for the domain 2, 3, 4, 5, 6 and 7 (1 and 8 domains were not analysed)
are 0.20, 0.43, 0.62, 0.49, 0.32 and 0.13, respectively
Table 1 Domains of the discrete wavelet transform and their
frequency range
Wavelet domain Frequency range (Hz)
Lower frequency Higher frequency
1 250 500
2 125 250
3 62.5 125
4 31.2 62.5
5 15.6 31.2
6 7.8 15.6
7 3.9 7.8
8 1.9 3.9
898 Eur J Appl Physiol (2013) 113:895904
1 3
(70.4 2.9 Hz) post-WAnT, then returned to the pre-
exercise level.
DWT
Figure 5 shows changes in the intensity of the wavelet
coefcient from pre-exercise level for each of the six fre-
quency domains following WAnT. The changes were sig-
nicantly different amongst the domains, and the domains
3 and 5 showed the largest increases in the intensity of the
wavelet coefcients amongst the domains. The one-way
repeated measures ANOVA showed a signicant increase
in the intensity of the wavelet coefcient for domains 2, 3,
4 and 5 from pre-exercise values. The intensity increased
signicantly at 12 and 15 min for the domain 2, at all time
points for the domain 3, and at 1, 3 and 6 min for the
domains 4 and 5.
Fig. 2 Changes (mean SEM) in cycling power output normalised
by bodyweight every 5 s over 30-s supramaximal cycling. Fatigue
index is also shown in the gure. *Signicantly (P\0.05) lower than
the peak 5-s power output
Fig. 3 Changes (mean SEM) in maximal voluntary isometric
contraction strength (a) and blood lactate concentration (b) before
(Pre) and 1, 3, 6, 9, 12 and 15 min following 30-s supramaximal
cycling. *Signicantly (P\0.05) greater than the pre-value.
#
Sig-
nicantly (P\0.05) greater than the 1 min value
Fig. 4 Changes (mean SEM) in sEMG root mean square (a) and
mean frequency (b) before (Pre) and 1, 3, 6, 9, 12 and 15 min
following 30-s supramaximal cycling. *Signicantly (P\0.05)
different than the pre-exercise value
Eur J Appl Physiol (2013) 113:895904 899
1 3
Correlation between changes in MNF and intensity
of DWT domains
Figure 6 shows the correlation between the magnitude of
changes in MNF and DWT for the domains that showed
signicant changes at 1 and 3 min post-WAnT (i.e.
domains 3, 4 and 5) using the two-time points of 1 and
3 min post-WAnT (n = 34: 17 subjects 9 2 time points).
A signicant correlation was found (r = -0.56) for the
domain 5 only (Fig. 6c).
Discussion
The main ndings of the present study were that (1) MNF
signicantly decreased at 1 and 3 min post-WAnT from the
baseline, and the DWT domains 25 signicantly increased
after WAnT, exhibiting longer lasting changes (e.g. 15 min
for domains 2 and 3) than MNF (13 min post-WAnT
only); and (2) the relative increase in intensity of the
wavelet domain 5 was signicantly correlated with the
relative decrease in the MNF at 1 and 3 min post-WAnT.
These results partially supported the hypotheses that MNF
would decrease and the intensity of low-frequency DWT
domains would increase following WAnT, and the changes
in MNF would be signicantly correlated with the changes
in the intensity of low-frequency DWT.
The peak and mean power output, and fatigue index
during WAnT (Fig. 2) and changes in blood lactate fol-
lowing WAnT (Fig. 3b) were similar to those reported in
previous studies (Hebestreit et al. 1996; Bogdanis et al.
1995; Beneke et al. 2007). Judging from the peak and mean
power outputs, the subjects in the present study were cat-
egorised as the lower 1020 % of physically active young
adults according to the normative values reported by Maud
and Shultz (1989). The present study was the rst to
measure MVC torque of the knee extensors following
WAnT. Since the fatigue index showed more than 30 %
Fig. 5 The magnitude of
changes (mean SEM) in
wavelet coefcient intensity for
the domains 27 from pre-
exercise (0 %) at 1, 3, 6, 9, 12
and 15 min following
supramaximal cycling.
*Signicantly (P\0.05)
different from pre-exercise
value
Fig. 6 Correlation between the
magnitude of changes in mean
frequency and coefcient of
DWT for the domains 3 (a), 4
(b) and 5 (c) from the baseline
values at 1 and 3 min following
WAnT. The two-time point data
(1 and 3 min post-cycling,
n = 34) were used for the
analyses
900 Eur J Appl Physiol (2013) 113:895904
1 3
decrease in power output, a decrease in knee extensor
MVC torque immediately after WAnT was expected.
However, the MVC torque did not show a signicant
decrease at 1 min post-WAnT, but showed a signicant
decrease from the baseline (23 %) after 3 min post-WAnT
(Fig. 3b). It could be argued that repeated MVC mea-
surements following WAnT would have induced additional
neuromuscular fatigue and/or affected the recovery from
WAnT. However, only a 3-s MVC contraction was per-
formed for each time point with 2 min rest before the next
contraction, thus it seems likely that the effect of the MVC
measures on recovery was minimum, and they did not
affect the changes in MVC and sEMG.
Sahlin and Seger (1995) reported 34 % decrease in
MVC immediately after cycling to exhaustion at 75 % of
the VO
2max
, and Giacomoni et al. (2006) reported 1315 %
MVC decrease immediately after ten sets of 6-s cycling
sprints. Nicolas et al. (2008) also showed 17 % MVC
decrease immediately after cycling to exhaustion at 95 %
of maximal power output, and Theurel and Lepers (2008)
reported 712 % decrease in MVC at 1 min after 33 min
cycling at 70 % of maximal aerobic power. It should be
noted that the exercise duration of these studies was much
longer (485 min) than that of the present study (30 s). It
might be that the short duration of the WAnT delayed the
decrease in MVC torque. In the present study, the blood
lactate further increased from 1 to 3 min post-WAnT
(Fig. 2). Bogdanis et al. (1995) showed that the intramus-
cular pH was lowest at 1.5 min after WAnT, and Street
et al. (2001) showed a delayed acidosis (range 0.52 min)
after knee extensor fatiguing contractions. Thus, it is pos-
sible that delayed peripheral metabolic changes are asso-
ciated with the delayed decrease in MVC torque. It should
be noted that sEMG RMS was elevated after WAnT
(Fig. 4a). It might be that central fatigue was limited after
WAnT, and increased cortical excitability minimised the
decrease in MVC at 1 min post-WAnT. However, it can be
speculated that this effect was no longer potent at the
subsequent time points. Further studies are necessary to
elucidate why a decrease in MVC torque was delayed after
WAnT.
No previous studies have reported changes in sEMG
following WAnT, but some studies examined sEMG
amplitude changes during WAnT and found no signicant
changes (Hunter et al. 2003; Rana 2006; Camata et al.
2010; Stewart et al. 2011). For example, Stewart et al.
(2011) have recently reported that the average rectied
values of EMG amplitude did not change during WAnT. In
contrast, the present study found 2634 % increase in RMS
of sEMG during MVC measures that were performed after
WAnT (Fig. 3a). Sarre and Lepers (2005) reported that
RMS of VL muscle increased during cycling (from 5 to
60 min) at 65 % of maximal power output. Furthermore,
Giacomoni et al. (2006) showed that RMS of VL muscle
during MVC increased by 810 % from baseline at
15 min after a repeated sprint cycling exercise (ten sets of
6 s, 30 s rest between sets). Arabadzhiev et al. (2010)
stated that increases in sEMG RMS indicated increases in
motor unit action potential amplitude that could be related
to increases in the activity of the Na
?
K
?
pump and
intracellular action potential (IAP) length. IAP length is
increased with hyperpolarisation of the muscle membrane,
which is likely due to an accumulation of extracellular K
?
and H
?
that impairs the repolarisation of the muscle
membrane (Allen et al. 2008). It can be speculated that the
increases in RMS amplitude following WAnT were asso-
ciated with increases of extracellular K
?
and decreases in
pH produced by WAnT.
The sampling frequency in the present study was
1,000 Hz in agreement with the Nyquist rate stating that
sampling frequency must be at least double the highest
frequency components in the signal (Merletti and Parker
2004). The highest frequency components of the sEMG
signals have been reported to be around 400500 Hz for
isometric contractions (Ives and Wigglesworth 2003), thus
the sampling frequency (1,000 Hz) seems appropriate.
Although it has been reported that MNF depends on the
sampling frequency such that the higher the sampling fre-
quency, the greater the MNF (Sadhukhan et al. 1994),
changes in the MNF shown in Fig. 4b should not be
affected by the sampling frequency. Moreover, 1,000 Hz
sampling frequency has been used for MNF analyses dur-
ing isometric contractions in previous studies (e.g. Willems
and Ponte 2012; Coorevits et al. 2008; Beck et al. 2005).
Two previous studies reported a signicant decrease
(1015 %) in MNF over 30 s of WAnT (Hunter et al. 2003;
Greer et al. 2006). Stewart et al. (2011) found a tendency of
decreases in MNF that was in parallel with decreases in
power output and muscle bre conduction velocity during
WAnT. Hunter et al. (2003) showed that MNF decreased
by 15 % from the rst 5 s to the last 5 s during WAnT. It is
important to note that in the present study, MNF was
assessed during MVC measures after WAnT, not during
WAnT. However, as shown in Fig. 4b, MNF decreased
from pre to 1 min post-WAnT by 10 %, which was com-
parable to that reported by Hunter et al. (2003). Hunter
et al. (2003) stated that the decrease in MNF was probably
due to fatigue of fast twitch bres and decreases in pH and
phosphocreatine in those muscle bres. It is interesting that
MNF was returned to the baseline value by 6 min post-
WAnT in our study (Fig. 4b), when the blood lactate and
presumably intracellular H
?
concentration were still high
(Fig. 3b). Tenan et al. (2011) have recently reported that
MNF during MVC assessed intermittently over an incre-
mental cycling exercise to exhaustion did not change, but
blood lactate and plasma K
?
increased signicantly over
Eur J Appl Physiol (2013) 113:895904 901
1 3
time. They have concluded that changes in MNF are not
associated with changes in blood lactate and plasma K
?
.
The present study results also showed the dissociation
between MNF changes and blood lactate. Another possible
cause of the decrease in MNF is a decrease in muscle bre
conduction velocity (Kupa et al. 1995), but dissociation
between the changes in MNF and muscle conduction
velocity has been reported (Stewart et al. 2011; Zwarts
et al. 1987). It is speculated that central factors such as
changes in motor unit discharge rate and/or de-recruitment
of fatigued motor units also inuence MNF changes
(Hermens et al. 1992; Stewart et al. 2011).
The DWT analysis showed increases in the domains
25, and the magnitude of increase in the intensity of these
domains (2160 %) was similar to that of sEMG RMS
(2634 %). This would suggest that the increase in the
RMS was associated with the increases in the intensity
of wavelet coefcients. It should be noted that the range of
the frequency is not the same amongst domains, and the
greater the range, the less the precision and resolution. The
domains 3 and 5 (frequency range of 62.5125 and
15.631.2 Hz, respectively) showed the largest increases
amongst the domains after WAnT (Fig. 5). So et al. (2009)
reported that the intensity of a low-frequency domain
(centre frequency = 19.3 Hz) increased largest during 50
maximal fatiguing contractions, and speculated that the
large increase in the low-frequency domain was due to
decreases in the ring rate and conduction velocity of fast
twitch muscle bres. In the present study, domains 2 and 3
showed increases at 915 min post-WAnT, which may
reect some recovery of fast twitch muscle bres. Although
it is not known whether changes in the high-frequency
domains 2 and 3 represent physiological changes, the results
of the present study show that sEMG frequency was still
affected at 15 min post-WAnT according to the DWT
analysis.
The DWT domain 5 showed the largest relative increase
(60 % at 1 min) following WAnT amongst the six domains
(Fig. 5). Furthermore, the magnitude of increase in the
intensity of the domain 5 (15.631.2 Hz) was signicantly
correlated with the magnitude of decrease in MNF at
13 min post-WAnT. We speculate that the decrease in
MNF and the increase of domain 5, which contains a range
of low frequencies, could indicate a relative decrease in
muscle bre conduction velocity and/or a decrease in fast
twitch muscle bre activation. It might be that the larger
increases of the domain 5 compared with other domains
after WAnT suggest that the compression of the frequency
spectrum was shifted towards low frequencies. The chan-
ges in domain 5 lasted longer (i.e. 9 min) than those of
MNF (i.e. 6 min). It is possible that domain 5 could detect
smaller changes in the frequency spectrum, due to its fre-
quency range decomposition. Because DWT matches
better the spiky shape of the sEMG signals and manages
the non-stationarity in muscle contractions better than FFT
(Karlsson et al. 2000), it may provide better accuracy in
detecting changes in the sEMG frequency spectrum
(Letelier and Weber 2000). It is plausible that since MNF is
a measure of central tendency (i.e. mean, median, mode) of
the whole frequency spectrum, the sensitivity of MNF
analysis to detect changes in frequency spectrum is less
than that of DWT. DWT appears to provide information
that MNF does not provide such as the decomposition in
ranges of the frequencies that allows the examination of
changes at different portions of the frequency spectrum,
which could increase the understanding of muscle recovery
from fatiguing tasks (von Tscharner 2002; Karlsson et al.
1999).
It has been documented, however, that changes in fre-
quency domains do not necessarily reect a change in
muscle bre recruitment or conduction velocity (Farina
2008; Farina et al. 2004). The present study assessed the
sEMG during MVC measures that are often used to assess
neuromuscular fatigue, and compared the relative changes
in sEMG during this standardised task over 15 min fol-
lowing WAnT. We speculate that the relatively larger
increases in the DWT low-frequency domains most likely
indicate a decrease in muscle bres conduction velocity,
due to the decrease in the ring rate of fast twitch muscle
bres. More studies are necessary using multi-electrode
arrays and motor unit decomposition technique to investi-
gate more directly changes in muscle bre recruitment and
conduction velocity after WAnT, and their relationship to
DWT frequency changes.
In summary, both DWT and MNF frequency analyses
showed a swift towards a low-frequency spectrum; how-
ever, DWT domain changes lasted longer and the magni-
tude of the changes in low-frequency domains were greater
than those observed in MNF. Thus, it appears that DWT
provides information of the frequency changes in sEMG
that is not detected by FFT. It is possible that DWT,
especially domain 5, could detect small changes in sEMG
frequency following exercise. It seems possible that DWT
has better timefrequency resolution, which could detect
smaller changes in sEMG signals than MNF. It is con-
cluded that DWT analysis is useful to understand neuro-
muscular fatigue and recovery.
References
Allen DG, Lamb GD, Westerblad H (2008) Impaired calcium release
during fatigue. J Appl Physiol 104:296305
Arabadzhiev TI, Dimitrov VG, Dimitrova NA, Dimitrov GV (2010)
Interpretation of EMG integral or RMS and estimates of
neuromuscular efciency can be misleading in fatiguing
contraction. J Electromyogr Kinesiol 20:223232
902 Eur J Appl Physiol (2013) 113:895904
1 3
Armstrong N, Davies B, Heal M (1983) The specicity of energy
utilisation by trained and untrained adolescent boys. Br J Sports
Med 17:193199
Beck TW, Housh TJ, Johnson GO, Weir JP, Cramer JT, Coburn JW,
Malek MH (2005) Comparison of Fourier and wavelet transform
procedures for examining the mechanomyographic and electro-
myographic frequency domain responses during fatiguing isoki-
netic muscle actions of the biceps brachii. J Electromyogr
Kinesiol 15:190199
Beneke R, Jumah MD, Leithauser RM (2007) Modelling the lactate
response to short-term all out exercise. Dyn Med 6:10
Bogdanis GC, Nevill ME, Boobis LH, Lakomy HK, Nevill AM
(1995) Recovery of power output and muscle metabolites
following 30 s of maximal sprint cycling in man. J Physiol 482:
467480
Camata TV, Dantas JL, Abrao T, Brunetto MA, Moraes AC, Altimari
LR (2010) Fourier and wavelet spectral analysis of EMG signals
in supramaximal constant load dynamic exercise. Conf Proc
IEEE Eng Med Biol Soc 2010:13641367
Cifrek M, Medved V, Tonkovic S, Ostojic S (2009) Surface EMG
based muscle fatigue evaluation in biomechanics. Clin Biomech
24:327340
Constable R, Thornhill RJ (1993) Using the discrete wavelet
transform for time-frequency analysis of the surface EMG
signal. Biomed Sci Instrum 29:121127
Constable R, Thornhill RJ, Carpenter DR (1994) Time-frequency
analysis of the surface EMG during maximum height jumps
under altered-G conditions. Biomed Sci Instrum 30:6974
Coorevits P, Danneels L, Cambier D, Ramon H, Druyts H, Karlsson
JS, De Moor G, Vanderstraeten G (2008) Testretest reliability
of wavelet- and Fourier based EMG (instantaneous) median
frequencies in the evaluation of back and hip muscle fatigue
during isometric back extensions. J Electromyogr Kinesiol
18:798806
Farina D (2008) Counterpoint: spectral properties of the surface EMG
do not provide information about motor unit recruitment and
muscle ber type. J Appl Physiol 105:16731674
Farina D, Merletti R, Enoka RM (2004) The extraction of neural
strategies from the surface EMG. J Appl Physiol 96:14861495
Giacomoni M, Billaut F, Falgairette G (2006) Effects of the time of
day on repeated all-out cycle performance and short-term
recovery patterns. Int J Sports Med 27:468474
Greer F, Morales J, Coles M (2006) Wingate performance and surface
EMG frequency variables are not affected by caffeine ingestion.
Appl Physiol Nutr Metab 31:597603
Hebestreit H, Meyer F, Htay H, Heigenhauser GJ, Bar-Or O (1996)
Plasma metabolites, volume and electrolytes following 30 s
high-intensity exercise in boys and men. Eur J Appl Physiol
Occup Physiol 72:563569
Hermens HJ, Bruggen TA, Baten CT, Rutten WL, Boom HB (1992)
The median frequency of the surface EMG power spectrum in
relation to motor unit ring and action potential properties.
J Electromyogr Kinesiol 2:1525
Hermens HJ, Freriks B, Disselhorst-Klug C, Rau G (2000) Develop-
ment of recommendations for SEMG sensors and sensor
placement procedures. J Electromyogr Kinesiol 10:361374
Hostens I, Seghers J, Spaepen A, Ramon H (2004) Validation of the
wavelet spectral estimation technique in biceps brachii and
brachioradialis fatigue assessment during prolonged low-level
static and dynamic contractions. J Electromyogr Kinesiol 14:
205215
Hunter AM, St Clair Gibson A, Lambert MI, Nobbs L, Noakes TD
(2003) Effects of supramaximal exercise on the electromyo-
graphic signal. Br J Sports Med 37:296299
Ives JC, Wigglesworth JK (2003) Sampling rate effects on surface
EMG timing and amplitude measures. Clin Biomech 18:543552
Karlsson S, Yu J, Akay M (1999) Enhancement of spectral analysis of
myoelectric signals during static contractions using wavelet
methods. IEEE Trans Biomed Eng 46:670684
Karlsson S, Yu J, Akay M (2000) Time-frequency analysis of
myoelectric signals during dynamic contractions: a comparative
study. IEEE Trans Biomed Eng 47:228238
Kumar DK, Pah ND, Bradley A (2003) Wavelet analysis of surface
electromyography to determine muscle fatigue. IEEE Trans
Neural Syst Rehabil Eng 11:400406
Kupa EJ, Roy SH, Kandarian SC, De Luca CJ (1995) Effects of
muscle ber type and size on EMG median frequency and
conduction velocity. J Appl Physiol 79:2332
Letelier JC, Weber PP (2000) Spike sorting based on discrete wavelet
transform coefcients. J Neurosci Methods 101:93106
Maud PJ, Shultz BB (1989) Norms for the Wingate anaerobic test
with comparison to another similar test. Res Q Exerc Sport
60:144151
Merletti R, Parker P (2004) Electromyography : physiology,
engineering, and noninvasive applications. In: IEEE Press series
in biomedical engineering. Wiley, New Jersey
Nicolas A, Gauthier A, Bessot N, Moussay S, Thibault G, Sesboue B,
Davenne D (2008) Effect of time-of-day on neuromuscular
properties of knee extensors after a short exhaustive cycling
exercise. Isokinet Exerc Sci 16:3340
Patton JF, Murphy MM, Frederick FA (1985) Maximal power outputs
during the Wingate anaerobic test. Int J Sports Med 6:8285
Rana SR (2006) Effect of the Wingate test on mechanomyography
and electromyography. J Strength Cond Res 20:292297
Reiser RF 2nd, Peterson ML, Broker JP (2002) Inuence of hip
orientation on Wingate power output and cycling technique.
J Strength Cond Res 16:556560
Sadhukhan AK, Goswami A, Kumar A, Gupta S (1994) Effect of
sampling frequency on EMG power spectral characteristics.
Electromyogr Clin Neurophysiol 34:159163
Sahlin K, Seger JY (1995) Effects of prolonged exercise on the
contractile properties of human quadriceps muscle. Eur J Appl
Physiol Occup Physiol 71:180186
Sarre G, Lepers R (2005) Neuromuscular function during prolonged
pedalling exercise at different cadences. Acta Physiol Scand
185:321328
So RC, Ng JK, Lam RW, Lo CK, Ng GY (2009) EMG wavelet
analysis of quadriceps muscle during repeated knee extension
movement. Med Sci Sports Exerc 41:788796
Stewart D, Farina D, Shen C, Macaluso A (2011) Muscle bre
conduction velocity during a 30-s Wingate anaerobic test.
J Electromyogr Kinesiol 21:418422
Street D, Bangsbo J, Juel C (2001) Interstitial pH in human skeletal
muscle during and after dynamic graded exercise. J Physiol
537:993998
Subasi A, Kiymik MK (2010) Muscle fatigue detection in EMG using
time-frequency methods, ICA and neural networks. J Med Syst
34:777785
Tenan MS, McMurray RG, Blackburn BT, McGrath M, Leppert K
(2011) The relationship between blood potassium, blood
lactate, and electromyography signals related to fatigue in a
progressive cycling exercise test. J Electromyogr Kinesiol
21:2532
Theurel J, Lepers R (2008) Neuromuscular fatigue is greater
following highly variable versus constant intensity endurance
cycling. Eur J Appl Physiol 103:461468
von Tscharner V (2002) Time-frequency and principal-component
methods for the analysis of EMGs recorded during a mildly
fatiguing exercise on a cycle ergometer. J Electromyogr Kinesiol
12:479492
Wakeling JM, Pascual SA, Nigg BM, von Tscharner V (2001) Surface
EMG shows distinct populations of muscle activity when
Eur J Appl Physiol (2013) 113:895904 903
1 3
measured during sustained sub-maximal exercise. Eur J Appl
Physiol 86:4047
Willems ME, Ponte JP (2012) Divergent muscle fatigue during
unilateral isometric contractions of dominant and non-dominant
quadriceps. J Sci Med Sport [Epub ahead of print]
Zwarts MJ, Van Weerden TW, Haenen HT (1987) Relationship
between average muscle bre conduction velocity and EMG
power spectra during isometric contraction, recovery and applied
ischemia. Eur J Appl Physiol Occup Physiol 56:212216
904 Eur J Appl Physiol (2013) 113:895904
1 3