Dario Farina, Luigi Fattorini, Francesco Felici and Giancarlo Filligoi

J Appl Physiol 93:1753-1763, 2002. First published Aug 9, 2002; doi:10.1152/japplphysiol.00314.2002

You might find this additional information useful...

This article cites 43 articles, 11 of which you can access free at:
http://jap.physiology.org/cgi/content/full/93/5/1753#BIBL

This article has been cited by 3 other HighWire hosted articles:

Motor Unit Synchronization Is Increased in Biceps Brachii After Exercise-Induced Damage
to Elbow Flexor Muscles
T. J. Dartnall, M. A. Nordstrom and J. G. Semmler
J Neurophysiol, February 1, 2008; 99 (2): 1008-1019.
[Abstract] [Full Text] [PDF]

The extraction of neural strategies from the surface EMG

D. Farina, R. Merletti and R. M. Enoka
J Appl Physiol, April 1, 2004; 96 (4): 1486-1495.
[Abstract] [Full Text] [PDF]

Gender differences in skeletal muscle fatigability are related to contraction type and EMG
spectral compression

Downloaded from jap.physiology.org on January 6, 2009

B. C. Clark, T. M. Manini, D. J. The, N. A. Doldo and L. L. Ploutz-Snyder
J Appl Physiol, June 1, 2003; 94 (6): 2263-2272.
[Abstract] [Full Text] [PDF]

Updated information and services including high-resolution figures, can be found at:
http://jap.physiology.org/cgi/content/full/93/5/1753

Additional material and information about Journal of Applied Physiology can be found at:
http://www.the-aps.org/publications/jappl

This information is current as of January 6, 2009 .

Journal of Applied Physiology publishes original papers that deal with diverse areas of research in applied physiology, especially
those papers emphasizing adaptive and integrative mechanisms. It is published 12 times a year (monthly) by the American
Physiological Society, 9650 Rockville Pike, Bethesda MD 20814-3991. Copyright © 2005 by the American Physiological Society.
ISSN: 8750-7587, ESSN: 1522-1601. Visit our website at http://www.the-aps.org/.
 J Appl Physiol 93: 1753–1763, 2002.
First published August 9, 2002; 10.1152/japplphysiol.00314.2002.

Nonlinear surface EMG analysis to detect changes of

motor unit conduction velocity and synchronization

DARIO FARINA,1 LUIGI FATTORINI,2,3 FRANCESCO FELICI,4 AND GIANCARLO FILLIGOI5,6

1
Centro di Bioingegneria, Dipartimento di Elettronica, Politecnico di Torino, Torino 10129;
2
Dipartimento Fisiologia Umana e Farmacologia, and 3Scuola di Medicina dello Sport, Università
di Roma, and 5Facoltà di Ingegneria, Dipartimento INFOCOM and 6Centro Interdipartimentale
Sistemi Biomedici, Università degli Studi, La Sapienza, Roma 00185; and 4Facoltà di Scienze
Motorie, Istituto Universitario di Scienze Motorie, Roma 00194, Italy
Received 9 April 2002; accepted in final form 8 August 2002

Farina, Dario, Luigi Fattorini, Francesco Felici, and EMG, because the tissues separating the muscle fibers
Giancarlo Filligoi. Nonlinear surface EMG analysis to de- and the recording electrodes act as low-pass filters (15,
tect changes of motor unit conduction velocity and synchro- 28). This determines low-spatial selectivity, which hin-
nization. J Appl Physiol 93: 1753–1763, 2002. First pub- ders the separation of the contributions of different
lished August 9, 2002; 10.1152/japplphysiol.00314.2002.—

Downloaded from jap.physiology.org on January 6, 2009

motor units (MUs). For this reason, past research ef-
Amplitude and frequency content of the surface
electromyographic (EMG) signal reflect central and periph- forts in the surface EMG field were mainly devoted to
eral modifications of the neuromuscular system. Classic sur- the development of processing techniques in time and
face EMG spectral variables applied to assess muscle func- frequency domain, which gave indications about the
tions are the centroid and median power spectral frequencies. global EMG activity, without aiming at an analysis at
More recently, nonlinear tools have been introduced to ana- the single MU level. Although, recently, techniques for
lyze the surface EMG; among them, the recurrence quantifi- noninvasive assessment of single MU properties have
cation analysis (RQA) was shown to be particularly promis- been proposed and refined (4, 6, 8), their applicability
ing for the detection of muscle status changes. The purpose of is still limited by the need of complex detection sys-
this work was to analyze the effect of motor unit short-term tems, to low-contraction levels, and controlled labora-
synchronization and conduction velocity (CV) on EMG spec-
tory conditions for signal recording. Thus the extrac-
tral variables and two variables extracted by RQA, the per-
centage of recurrence (%Rec) and determinism (%Det). The tion of global information from the surface EMG signal
study was performed on the basis of a simulation model, remains of paramount importance in many applica-
which allowed changing the degree of synchronization and tions, such as sport and occupational medicine, reha-
mean CV of a number of motor units, and of an experimental bilitation, and basic and applied physiology.
investigation of the surface EMG signal properties detected Spectral analysis and amplitude estimation are usu-
during high-force-level isometric fatiguing contractions of the ally performed from single differential signals (3) to
biceps brachii muscle. Simulations and experimental results obtain indications about the physiological processes
were largely in agreement and show that 1) spectral variables, occurring during sustained voluntary contractions. In
%Rec, and %Det are influenced by CV and degree of synchro- addition to the efforts devoted to the technical issue of
nization; 2) spectral variables are highly correlated with %Det
estimating amplitude and spectral variables in a reli-
(R ⫽ ⫺0.95 in the simulations and ⫺0.78 and ⫺0.75 for the
initial values and normalized slopes, respectively, in the exper- able way (12, 32, 41), many studies were focused, in the
imental signals), and thus the information they provide on past, on the clarification of the relationships between
muscle properties is basically the same; and 3) variations of these global variables and the underlying physical pro-
%Det and %Rec in response to changes in muscle properties are cesses, to extract information of physiological interest
significantly larger than the variations of spectral variables. from the global analysis of the surface EMG signal. It
This study validates RQA as a means for fatigue assessment has been clearly established that the rate of change of
with potential advantages (such as the higher sensitivity to spectral variables and conduction velocity (CV) during
changes of muscle status) with respect to the classic spectral a sustained contraction is indicative of muscle fatigue
analysis. (31) and may be correlated with MU type (39, 43). It
surface electromyography; motor unit short-term synchroni- has also been shown, both theoretically (28, 41) and
zation; surface electromyographic modeling; recurrence plot experimentally (2), that, during fatiguing contractions,
analysis CV and mean (MNF) or median spectral frequency
(MDF) of the surface EMG signal are highly correlated;
MNF and MDF reflect mainly the CV changes of the
THE SURFACE ELECTROMYOGRAPHIC (EMG) signal presents
active MUs. However, the comparison of the percent
smaller bandwidth with respect to the intramuscular
The costs of publication of this article were defrayed in part by the
Address for reprint requests and other correspondence: D. Farina, payment of page charges. The article must therefore be hereby
Dip. di Elettronica, Politecnico di Torino, Corso Duca degli Abruzzi marked ‘‘advertisement’’ in accordance with 18 U.S.C. Section 1734
24, Torino, 10129 Italy (E-mail: [email protected]). solely to indicate this fact.

http://www.jap.org 8750-7587/02 $5.00 Copyright © 2002 the American Physiological Society 1753
1754 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS

rate of CV and MNF and MDF decrease showed that, eters, and 3) to analyze advantages and limitations of
although being the main contribution to spectral com- the linear and nonlinear approach for assessment of
pression, CV is not the only determinant of changes of changes in the surface EMG signals, as a consequence,
the characteristic spectral frequencies. A higher de- for example, of fatigue, pathological conditions, or
crease of MNF and MDF with respect to CV was indeed training. The study is performed on the basis of a
observed in a number of past studies (refer, for exam- simulation model, which allowed separate investiga-
ple, to Ref. 31 and, recently, to Ref. 30). To explain tion of different muscle parameters. The modeling re-
these results, changes in MU CV distribution spread, sults are used to interpret experimental findings ob-
increase of the depolarization zone length, or increase tained from the biceps brachii muscle during voluntary
in MU short-term synchronization has been suggested isometric contractions.
(31), although with no direct evidence. Among these
factors, recent results indicated a major role of MU METHODS
synchronization (25, 45). However, the ability of sur-
face EMG spectral analysis as an indicator of MU Linear and Nonlinear Surface EMG Analysis
short-term synchronization is still under debate.
Spectral surface EMG variables. Classic spectral variables
On the basis of the above considerations, nonlinear computed from the surface EMG signal are MNF and MDF.
analysis of surface EMG signal has been exploited (35, The estimates of these variables depend on the additive noise
42). Recurrence quantification analysis (RQA), de- and the estimators adopted (12); on anatomical, physical, and
scribed by Eckmann et al. (5), is a technique for the detection-system parameters (9); and on the number, sizes,
detection of state changes in drifting dynamic systems type, and firing rates of the active MUs (10). In the present

Downloaded from jap.physiology.org on January 6, 2009

that does not necessitate any a priori constraint on study, these variables have been computed from synthetic
data size, stationarity, and statistical distribution (18). and experimental surface EMG signals with well-known al-
RQA has been used recently in a number of experimen- gorithms (for a recent review, see Ref. 12). In particular, the
tal studies (17, 23, 42), showing its potential in detect- periodogram-based spectrum estimation was obtained from
adjacent nonoverlapping signal epochs of 1 s.
ing changes in surface EMG due to fatigue. Webber et
RQA. For the mathematical details about RQA, the reader
al. (42) tested the sensitivity of different indexes ex- can refer to a recent review by Filligoi and Felici (18), here
tracted from RQA. These authors indicated that subtle summarized in the APPENDIX. The procedure is based on
changes in surface EMG can be detected by the per- embedding EMG data in an N-dimensional Euclidean space.
centage of determinism (%Det), which reflects the Recurrence maps are built from signal epochs, and quantifi-
amount of rule-obeying structure in the signal dy- cation indexes are extracted from these maps (refer to the
namic, and the percentage of recurrence (%Rec), which APPENDIX). Among these indexes, %Rec and %Det will be
reflects the current state of the system. Furthermore, investigated in this study.
under particular experimental conditions, it was spec-
ulated (16, 18) that %Det should be more sensitive Simulation Model
than spectral analysis to MU short-term synchroniza-
Generation of MU action potentials. A model previously
tion, because it reveals embedded determinisms in an developed for the simulation of the surface EMG signal (13)
apparently stochastic signal. However, none of these was used in this study. The model simulates synthetic MU
studies has investigated in depth details which EMG action potentials (MUAPs) generated by finite-length fibers
signal parameters are indeed reflected by EMG vari- and detected by surface electrodes with physical dimensions.
ables extracted by RQA. For this reason, it is difficult The volume conductor is an anisotropic medium representing
to interpret experimental results, in particular when the muscle and two-layered isotropic media representing fat
they are compared with those obtained by spectral and skin tissues (15) (Fig. 1C). The transmembrane current
analysis. density was described as indicated by Rosenfalck (37). The
In this work, we focused on two phenomena, the MU fixed-model parameters (such as the conductivities of the
CV change and the degree of short-term synchroniza- subcutaneous layers) are the same as reported in a previous
study (10).
tion, which clearly have an impact on surface EMG The simulated signals were detected by a single differen-
spectral variables. The theoretical derivation of the tial system with electrodes 5 mm long and 1 mm thick. The
effects on RQA of mean MU CV and MU short-term interelectrode distance was 20 mm. The detection probe was
synchronization and its relationship with classic spec- located in the middle between the centers of the innervation
tral EMG analysis are difficult because of the complex- zone and the tendon region of a number of MUs with a mean
ity of the surface EMG generation and detection sys- semilength (in both directions) of 65 mm. The detection
tem. A modeling approach is, on the other hand, volume of the system was defined as suggested by Farina et
feasible and may be useful for the interpretation of al. (10) on the basis of a threshold in the energy of the surface
experimental results. MUAPs. The contributions to the surface EMG signal of the
The aims of this work were thus 1) to investigate the MUs out of the detection volume were neglected. The number
of MU fibers was uniformly distributed between 50 and 450,
effect of MU short-term synchronization and mean CV and the MU fiber density was 20 fibers/mm2 (19). The MU
on EMG variables extracted from linear and nonlinear territory was circular, and the fiber density in the muscle was
analysis, 2) to assess whether some variables of RQA 200 fibers/mm2. Thus the fibers of different MUs could be
and spectral analysis provide indications on the same intermingled. The surface-recorded MUAPs were obtained as
neuromuscular system parameters or whether they the sum of the action potentials generated by the muscle
have different sensitivity to variations of such param- fibers belonging to each specific MU. Sixty-five MUs were

J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org

 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS 1755

Fig. 1. A: the independently generated

firing patterns of all of the 65 active
motor units (MUs). B: the same pat-
terns to which synchronization has
been added (10% firings, and MUs syn-
chronized). C: the volume conductor
and an example of location and territo-
ries of the active MUs. D: represen-
tative single differential simulated
signal (10% firings and MUs synchro-
nized).

Downloaded from jap.physiology.org on January 6, 2009

simulated in each trial. This value was selected so that the total number of MUs (%M)]. Given %F and %M, for each
total number of simulated fibers was similar to that of a real train, the preset number of MUAPs was synchronized with
muscle in the detection volume (10). others by moving their firing instants. The MUAPs to be
Firing rates. The simulated signals corresponded to a full synchronized were moved to have their firing instants at a
recruitment of the MUs in the detection volume. To obtain distance described by a Gaussian random variable with 0
the distribution of firing rates, it was assumed that the mean and 2-ms standard deviation.
generated force was that corresponding to the recruitment In summary, after the generation of the independent firing
threshold of the highest threshold MU in the simulated set. patterns, the steps followed to induce MU synchronization
The MU recruitment threshold was computed with the expo- were as follows: 1) compute the number Nm of firings of
nential rule suggested by Fuglevand et al. (19) and recently different MUs synchronized in each synchronization event as
applied by Farina et al. (10). Recruitment of MUs was as- %M of the total number of MUs; 2) consider the first MUAP
sumed to take place at up to 80% of the maximal force, train generated and compute the number Nf of its firings,
simulating a muscle that recruits MUs almost until the which should be synchronized with others as %F of the total
maximum contraction level, as is the case of the biceps number of firings of the train; 3) select randomly the Nf
brachii (26). The firing rates were inversely related to the firings among those of the train; 4) for each of the Nf firings
recruitment threshold (34), with a minimum of 8 pulses/s at (reference firings), select randomly the (Nm ⫺ 1) MUs, of
the recruitment and a maximum of 35 pulses/s. The standard which one firing should be synchronized with the reference
deviation of the interpulse interval was fixed for all MUs to firing; 5) select, for each of the (Nm ⫺ 1) MUs, the firing
15% of the mean interpulse interval (Gaussian distribution). closest to the reference firing; 6) move the selected firings in
The 65 MUs had a Gaussian distribution of CV with a a position obtained as the realization of a Gaussian random
standard deviation of 0.3 m/s. The mean value of the distri- variable with mean value of the reference firing position and
bution was varied between 3 and 5 m/s to investigate its standard deviation of 2 ms; and 7) repeat the operations
effect on the EMG variables. The distribution of CV was described in steps 2–6 for all of the selected firings and for all
truncated at 2 and 7 m/s, which are the extremes of the of the MUAP trains. Once a firing has been shifted, it cannot
experimentally observed values. The smallest MUs had the be moved in subsequent operations. This procedure has been
lowest CV (1), and the recruitment order followed the size proposed by Yao et al. (45) and was validated by those
principle (21), with the small and low CV MUs recruited at authors by comparison of synchronization indexes proposed
the beginning of the contraction (and thus having higher in the past in simulated and experimental signals.
firing rate). Figure 1 shows the representative firing patterns gener-
Generation of the firing patterns. The firing patterns of the ated with and without MU synchronization and an example
active MUs were first generated independently (no synchro- of the generated signal. Figure 2 shows examples of simu-
nization) on the basis of the calculated mean firing rates and lated signals with three levels of MU synchronization. The
interpulse interval variability (Fig. 1A). After this step, some cross-histograms of two randomly selected firing patterns in
firing instants were moved to create the desired level of firing the set are also reported.
synchronization (Fig. 1B). In particular, the synchronization Simulation modalities. In each simulation, all of the 65
level was defined by two parameters, indicating the percent- MUs were active, with firing patterns computed as described
age of firings in each train synchronized with other firings above on the basis of their recruitment threshold, size, CV,
(%F) and the number of firings synchronized together for and degree of synchronization. In the simulations, the two
each synchronization event [expressed as a percentage of the parameters describing MU synchronization (%F and %M)

J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org

1756 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS

Fig. 2. Simulated signals with different de-

grees of synchronization among MU firing
patterns. The percentage of synchronization
is the percentage of potentials (with respect
to the total no. in the train) of each MU
synchronized with potentials of other MUs.
The no. of potentials involved in each syn-

Downloaded from jap.physiology.org on January 6, 2009

chronization event is a percentage of the total
no. of MUs (this percentage is equal to the
percentage of synchronization). The signals
are normalized with respect to the maximum
value of the signal with highest degree of
synchronization. The cross-histograms be-
tween 2 firing patterns (randomly selected)
are also reported. See text for details.

were always set equal to reduce the number of parameters nization levels, and 50 signals for each condition) were gen-
describing central changes. The values chosen were in the erated.
range between 0% (no synchronization) and 25% (very high
synchronization) at steps of 5%. For each of these synchroni- Experimental Protocol
zation levels, the distribution of CV had a mean from 3 to 5
m/s at steps of 0.5 m/s. Subjects. Ten healthy male volunteers (age, mean ⫾ SD:
To take into account the variability of the results, depend- 27.4 ⫾ 5.2 yr; body mass, 72.3 ⫾ 4.7 kg; stature, 174.3 ⫾ 6.1
ing on the locations of the MUs in the detection volume, the cm) participated in the study. No subject had known symp-
MUs were placed randomly in the muscle. For each physio- toms of neuromuscular disorders. The study was approved by
logical condition, 50 synthetic signals were generated. For the local ethics committee, and written, informed consent
each of the 50 signals in the same set, the positions of the was obtained from all participants before inclusion.
MUs in the muscle were randomly selected (with uniform Electrophysiological and mechanical recordings. The
distribution in the detection volume), whereas the firing torque of the elbow flexor muscles was measured with a
patterns (with synchronization), CV distribution, and num- modular brace, which incorporates two independent torque
ber of fibers for each MU were fixed for the entire set. Thus, meters (model TR11, CCT Transducers, Torino, Italy), on
in each physiological condition, 50 anatomic conditions were each side of the brace. The force signal was digitized by a
tested, describing a basic source of variability between sub- 12-bit analog-to-digital converter and sampled at 1,024 Hz.
jects. The simulated signals were all 5-s long, with a sam- A linear array of electrodes (silver bar electrodes, 5 mm
pling frequency of 1,024 Hz and noise free. In total, 1,500 long, 1 mm diameter, 10 mm interelectrode distance) was
synthetic signals (corresponding to 5 CV values, 6 synchro- applied between the two tendon regions of the long head of

J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org

 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS 1757

the biceps brachii muscle. A reference electrode was applied spectra are also shown. The difference between the two
at the wrist. Before electrode placement, the skin was recurrence maps in the two physiological conditions is
slightly abraded with abrasive paste and cleaned with water. evident (see also APPENDIX). In particular, a higher
The optimal orientation of the array was determined by number of rule-obeying structures in the signal is
visual inspection of the EMG signals in a few trials at the
beginning of the experimental session. The array was held in
clearly shown by a higher regularity of the recurrence
place by elastic straps, which did not obstruct blood flow. The map.
EMG signals were amplified and band-pass filtered (3 dB Figure 4 shows %Det and MNF as functions of mean
bandwidth ⫽ 10–500 Hz), sampled at 1,024 Hz, and con- CV and synchronization, normalized with respect to
verted in digital form by a 12-bit analog-to-digital converter. the highest mean CV simulated and to the lowest
General procedures. Subjects were seated comfortably and synchronization level. MDF led to results similar to
performed isometric contractions with the arm placed in the those for MNF. Assuming, for example, decreasing
isometric torque brace with the shoulder angle at 90° flexion mean CV (31) and increasing synchronization (25) dur-
and the elbow angle at 120°. Measurements were performed ing fatigue, these plots can be interpreted as indicative
on the right (always dominant) biceps. The hand was main- of the changes in linear and nonlinear variables during
tained halfway between pronation and supination. Maximal
voluntary elbow flexion contraction was measured during
fatiguing contractions with normalization with respect
three successive trials with 2 min of recovery in between. The to the initial values. Note that, for both CV and syn-
greatest value was assumed as the maximal voluntary con- chronization level changes, %Det shows greater rela-
traction (MVC) for the elbow flexion. After the determination tive changes with respect to MNF. The percent in-
of the MVC, 5-min rest was given to the subjects. After the crease of %Rec for CV decreasing from 5 to 3 m/s was
MVC assessments, the subjects performed an 80% MVC 130%. The percent increase of %Rec for the degree of
isometric contraction of the elbow flexors lasting 10 s. Sub-

Downloaded from jap.physiology.org on January 6, 2009

synchronization increasing from 0 to 25% was 73%.
jects held the contraction, matching a torque target by visual The average nonnormalized values of %Det for 0%
feedback. Each subject repeated the experimental session on synchronization level and 3 m/s CV were 28.09 ⫾ 12.31
2 different days separated by more than 1 day.
and 22.93 ⫾ 8.71%, respectively. For %Rec, they were
Experimental data processing and statistical analysis.
From the 15 single differential signals detected by the 16 3.10 ⫾ 0.88 and 3.52 ⫾ 1.25%, respectively. The aver-
electrodes of the array spaced by 10 mm, we computed the 14 age values for the two conditions were 97.72 ⫾ 16.27
single differential signals detected by electrodes spaced 20 and 101.34 ⫾ 10.90 Hz for MNF and 94.77 ⫾ 16.12 and
mm. This was achieved by summation of consecutive single 97.89 ⫾ 11.46 Hz for MDF, respectively.
differential signals, as described in Ref. 11. The location of Figure 5 shows the scatterplot between %Det and
the main innervation zone was assessed by visual inspection MNF normalized with respect to the mean values
of the multichannel EMG signals, and only the channels computed over all of the simulation set (n ⫽ 1,500). A
distal with respect to the innervation zone were considered high correlation was found between the two vari-
for further analysis. From these signals, we selected that ables in the simulated conditions (linear regression
corresponding to the minimal variability of spectral EMG
variables with respect to adjacent locations, as suggested in
analysis, R ⫽ ⫺0.95, P ⬍ 0.001). MDF showed very
Ref. 30. From this channel, EMG variables from linear and similar results, with a correlation with %Det result-
nonlinear analyses were computed. Thus the signals used for ing in R ⫽ ⫺0.93 (P ⬍ 0.001). Changes in MNF or
subsequent analysis were detected by a single differential MDF corresponded to opposite changes in %Det, as
system, with electrodes spaced 20 mm, placed between the already observed in Fig. 4. However, the normalized
innervation zone and the distal tendon region, and providing changes in %Det are almost twice those of the nor-
the minimal variability of spectral variables to shifts of the malized MNF and MDF.
recording system in that muscle zone. %Rec showed a decrease with CV increasing and an
%Det, %Rec, and spectral variables were estimated from increase with increasing degree of synchronization, as
the selected signals from adjacent nonoverlapping epochs of %Det. However, the correlation with spectral variables
1 s. For each EMG variable, the first-order regression lines
were computed from the values obtained during the 10-s-long
was poorer than for %Det (R ⫽ ⫺0.74 for MNF and R ⫽
contractions. The slope of EMG variables was defined as the ⫺0.72 for MDF), although statistically significant (P ⬍
slope of the regression line and the initial value as the 0.001).
intercept of the regression line. Normalized slopes, defined as
the slopes divided by the initial values, were expressed in Experimental Signals
percentages per second.
The experimental data were analyzed by using All subjects were able to maintain the selected force
two-way repeated-measures ANOVA, followed by post hoc level constant for the time interval of the contraction
Student-Newman-Keuls pairwise comparisons, when re- within ⫾5% MVC. For all of the subjects, a statistically
quired. Threshold for statistical significance was set to P ⫽ significant (linear regression analysis, P ⬍ 0.001) de-
0.05. Data are presented as means ⫾ SD. crease in MNF and increase (P ⬍ 0.001) in %Det and
%Rec was observed during the fatiguing task. As for
RESULTS the simulated signals, MNF and MDF showed very
Simulated Signals similar results; thus only those from MNF analysis are
reported in the following paragraph.
Figure 3 shows simulated signals with different The mean MNF, %Det, and %Rec slopes (absolute
mean values of CV distribution and degrees of synchro- values) were 2.03 ⫾ 0.55 Hz/s, 2.32 ⫾ 1.04%/s, and
nization level. The recurrence plots and the power 0.27 ⫾ 0.18%/s, respectively. The mean MNF, %Det,
J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org
1758 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS

Fig. 3. From top to bottom: raw simu-

lated signals, power spectra with mean
(MNF) and median spectral frequency
(MDF) values, and recurrence maps
and their percentage of determinism
(%Det) and recurrence (%Rec). The
electromyographic signals are gener-
ated with 5 m/s conduction velocity
(CV) and 0% MU synchronization level
(A) and 3 m/s CV and 25% MU syn-
chronization level (B). The signals in
the top row as well as the power spec-
tra in the middle row are in arbitrary

Downloaded from jap.physiology.org on January 6, 2009

units. The recurrence quantification
analysis parameters are D ⫽ 15, R ⫽
60, ␭ ⫽ 3, and L ⫽ 20 (see APPENDIX for
definitions), providing a plot of 982 ⫻
982 pixels.

and %Rec initial values were 73.15 ⫾ 8.34 Hz, 35.15 ⫾ DISCUSSION
9.31%, and 3.63 ⫾ 1.04%, respectively. The normalized
Although RQA is being used in many studies on
slopes (absolute values) were 2.66 ⫾ 0.66%/s, 7.50 ⫾ muscle assessment, surprisingly, it was not clearly
5.05%/s, and 8.41 ⫾ 7.32%/s, for MNF, %Det, and known what information it provides. Mean CV and
%Rec, respectively. short-term synchronization were the two phenomena
A two-way (three EMG variables, MNF, %Det, and investigated in this work, because they have been
%Rec, and 2 days) ANOVA of the normalized slope was proven to affect surface EMG spectral features during
significant for the EMG variable (F ⬍ 9.38, P ⬍ 0.01) fatigue assessment. In this study, we were interested
and not for the day. The post hoc Student-Newman- in analyzing the information that is extracted by RQA
Keuls test disclosed pairwise differences (P ⬍ 0.01) from the surface EMG signals and to compare this with
between MNF normalized slopes and the normalized that obtained by classic spectral analysis.
slopes of both %Det and %Rec. We followed a simulation approach proposed in a
The initial values and normalized slopes of MNF and previous study for addressing the issue of MU recruit-
%Det were significantly correlated. For the initial val- ment strategies (10). The main advantage of this ap-
ues, the correlation coefficient was R ⫽ ⫺0.78 (linear proach is to take into account a number of anatomic
regression analysis, P ⬍ 0.001), whereas for the nor- conditions, thus avoiding results biased by a particular
malized slopes it was R ⫽ ⫺0.75 (P ⬍ 0.001). The location of the MUs in the muscle.
correlation between MNF and %Rec was significant
Simulation of the MUAPs
(P ⬍ 0.05), although, as for the simulations, much
lower than that between MNF and %Det (R ⫽ ⫺0.55 The simulation of the MUAPs was performed with a
for the initial values and R ⫽ ⫺0.46 for the normalized number of simplifications, which have been discussed
slopes). Figure 6 reports the scatterplots between MNF already in previous works (9, 10, 13) and which are
and %Det initial values and normalized slopes. related to the choices of the simulation parameters
J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org
 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS 1759

modeling was accurate enough to clearly interpret the

experimental findings.
Simulation of the MU Firing Patterns
The assumed correlation among MU size, firing rate,
and recruitment threshold is a consequence of the size
principle (21), which has been validated in a number of
previous studies. The association among CV, MU size,
and firing rate is based on the results reported by
Andreassen and Arendt-Nielsen (1), who indicated a
relationship between CV and twitch MU properties
and thus proposed CV as an additional size principle
parameter. The correlation between firing rate and CV
was also directly shown by Farina (6) using intramus-
cular and surface EMG techniques for assessing con-
trol and conduction properties of single MUs (7).
Some previous studies aimed at the simulation of
MU short-term synchronization. Hermens et al. (22)
shifted the firings of independently generated firing
patterns to create synchronization of all of the firings of

Downloaded from jap.physiology.org on January 6, 2009

MU pairs. The complete synchronization of MU firings
was also assumed by Weytjens and van Steenberghe
(44) for their analytic derivation of the power spectrum
of the surface EMG signal. This simplification implies
that the two synchronized MUs have the same mean
firing rate. Yao et al. (45) proposed a method to intro-
duce synchronization in a pool of MUs, again based on
the shift of MU firings. This approach was validated by
the analysis of classic indexes of synchronization (36)
applied to the simulated signals and was adopted in
the present study, although with a different model of
MUAP generation and different simulation modalities.
Recently, Kleine et al. (25) proposed a different ap-
proach based on the model suggested by Matthews
Fig. 4. %Det and MNF as functions of the mean of CV distribution
(A) and the degree of MU synchronization (B). The data are normal-
ized with respect to (wrt) the mean values obtained for the maximum
CV value (A) and the minimum degree of synchronization (B). The
second-order polynomials interpolating the data are also shown. The
mean CV values in A are normalized wrt the maximum value (5 m/s)
to compare the relative rates of change of CV, %Det, and MNF. In
each case, mean and SD are computed from the 50 anatomic condi-
tions and by varying the nonfixed parameter over all of the values
simulated.

(such as the conductivities of the subcutaneous layers),

most of them not exactly known in practice. Different
choices for these parameters probably would not affect
the general conclusions drawn in this study, which are
mostly related to the statistical properties of the MU
firing patterns and to the relationships between CV
and surface MUAPs. Another limitation of the physical
model used for the generation of the surface MUAPs is
the description of the subcutaneous layers and of the
muscle as infinite parallel layers (13, 15). Other models
may be better in this respect, as, for example, the
approach followed by Kleine et al. (25), who assumed a
cylindrical volume conductor, which probably simu-
lates the arm better (40). Again, we do not think our
Fig. 5. Scatterplot of %Det as function of MNF for all of the simu-
main conclusions would be different by assuming a lated signals (n ⫽ 1,500). The values are normalized wrt mean values
different volume conductor shape. Moreover, the agree- (on all of the set of signals) to compare the relative changes of the 2
ment with the experimental results indicated that the variables.

J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org

1760 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS

how the distribution of CV changes independently on

the mean value, that is, how the relative weight of
large and small MUs changes with fatigue.
The dependence of %Det and %Rec on CV was not
previously known, because the only results in this
respect came from experimental fatigue studies in
which a number of phenomena occurred and could not
be separated from each other. The finding that the two
variables extracted from the recurrence maps are
changing with CV, as the spectral variables, clarifies
that CV is a parameter strongly affecting nonlinear
surface EMG analysis. The result can be explained by
observing that the number of recurrent structures
identified depends on the duration of the segments in
the EMG signal, which repeat in time with almost the
same shape, i.e., on the duration of the MUAPs. CV is
directly related to the duration of the surface MUAPs if
the length of the depolarized zone is constant during
time, as assumed in the simulations. Thus, the higher
CV is, the smaller is the duration of the surface

Downloaded from jap.physiology.org on January 6, 2009

MUAPs. From this point of view, %Det and %Rec
detect changes in the frequency content of the signal, a
conclusion important for the interpretation of experi-
mental findings. From our results, it clearly appears
that there is no possibility of separating the changes
due to CV and to the degree of synchronization by
using %Det or %Rec.

Effect of MU Synchronization Level on Linear and

Nonlinear EMG Analysis
Fig. 6. Scatterplot of MNF initial values (A) and normalized slopes
(B) as function of %Det initial values and normalized slopes for all of The effect of the degree of synchronization on the
the subjects investigated and the 2 days (n ⫽ 20). spectral frequencies of the surface EMG signal was
investigated in some previous works. Weytjens and
van Steenberghe (44) extended the derivations of Lago
(29), which described the afterhyperpolarization after and Jones (27) on the effect of MU firing statistics on
a firing as an exponential function on which membrane the surface EMG power spectrum to the case of depen-
noise is superimposed. The noise was then divided into dent firings and showed that cross-terms arise that
a common and an individual part to simulate different increase the energy of the signal mostly at low frequen-
degrees of synchronization. However, when comparing cies. Hermens et al. (22) validated this analytic obser-
the synchronization level introduced by their method vation by a simulation model, and similar results were
with that introduced by Yao et al. (45), these authors shown, more recently, by Yao et al. (45) and, with the
concluded that similar results were obtained (25); thus inclusion of the electrode location variability, by Kleine
the choice of one of the two models is probably not et al. (25). Our results are in line with these previous
critical in the context of this study. studies and confirm the role of synchronization in de-
Effect of CV on Linear and Nonlinear EMG Analysis termining MNF and MDF patterns during fatigue.
%Det and %Rec also depended on MU synchroniza-
The effect of muscle fiber CV on spectral features of tion, exhibiting an increase for increasing degree of
the EMG signal is well known from past studies (2, 31). synchronization. This study is the first that clearly
Theoretically, neglecting the effects of the firing statis- shows that these EMG variables increase with an in-
tics on spectral features, the relative change of CV and creasing level of MU short-term synchronization; the
MNF or MDF are equal if 1) no MUAP shape change result is in agreement with the expectation that a
occurs (only scaling), 2) the MU pool is stable (10), and higher number of deterministic structures in the signal
3) all of the MUs have the same CV. In our simulations, should be detected by RQA. In previous studies, this
condition 3 was not met, to better approximate exper- dependence was suggested, but, in experimental inves-
imental signals. However, it was shown (Fig. 4) that tigations on muscle fatigue, it was not possible to
the relative changes of MNF (MDF provided the same separate changes of MU synchronization and CV.
results) and CV are similar also in this condition. Felici et al. (16), for example, reported RQA analysis on
Direct interpretation of the plots of Fig. 4 in terms of a group of weight lifters, assuming that a higher de-
changes occurring in CV during fatigue is, however, gree of synchronization should occur with fatigue in
rather critical, because in real cases it is not known this group with respect to a control group of sedentary
J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org
 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS 1761

persons. They reported a higher rate of change of %Det to the different experimental setup used (for example,
in the weight lifters with respect to the control group interelectrode distances were different).
during sustained isometric contractions, but it was not
possible to attribute this difference exclusively to the Implications for Fatigue Assessment
different change in MU synchronization degree, be- The present results show that the two linear and
cause CV was also changing. nonlinear tools tested provide similar indications
about CV and MU synchronization changes, two mech-
Correlation Between Variables Extracted from Linear anisms that affect surface EMG signal features during
and Nonlinear Analysis fatigue. The claimed higher sensitivity of %Det to MU
synchronization than to other fatigue-related changes
One of the main results of this study is the observed
was not confirmed, and thus RQA cannot be used to
high correlation between %Det and MNF (or MDF) in
separate MU synchronization changes from other fac-
all of the simulated conditions. The regression analysis
tors. During fatigue, in nonpathological conditions,
reported in Fig. 5 shows a correlation coefficient be-
%Det is highly correlated to spectral variables. This is
tween the two variables close to one. This result indi- in agreement with previous experimental studies in
cates that, as spectral variables, %Det also reflects which RQA and spectral analysis always provided the
physiological phenomena of interest in muscle assess- same indications of muscle fatigue (16, 17, 42), al-
ment. Indeed the %Det changes found in previous though a correlation between variables obtained with
studies could not be explained on a physiological basis the two tools was reported for experimental data only
before the present study. by Webber et al. (42). The results suggest that no new

Downloaded from jap.physiology.org on January 6, 2009

As it always happens in simulation studies, the cru- information is extracted by RQA with respect to that
cial point is the validation and the extension of the provided by classic spectral analysis. However, another
results to experimental paradigms. We designed a sim- important finding of the present study is that the
ple fatigue protocol, and we assumed that, during the sensitivity of RQA to muscle changes is higher than
muscle contractions performed by the subjects, phe- that of spectral variables. This was already suggested
nomena similar to those simulated occurred. The con- by Webber et al. and has been demonstrated in the
traction level was selected to exclude recruitment of present study on a large set of simulation conditions
MUs during fatigue, because it corresponded to almost (including a number of anatomic configurations). From
the maximal level at which MUs are recruited in the the above considerations, RQA seems promising for
biceps brachii muscle. The simulation results were detection of muscle changes due to fatigue or other
largely in agreement with the experimental findings. factors and could allow a better separation among
We found, indeed, a highly significant correlation be- groups of subjects and/or muscles.
tween MNF and %Det in the experimental signals also,
in agreement with the simulation predictions. Webber Conclusions
et al. (42) already showed a correlation between the
The main conclusions of this work are as follows. 1)
rates of %Det and MNF change on 12 subjects who
Spectral variables, %Det, and %Rec are influenced by
performed fatiguing contractions. We also observed
CV and the degree of synchronization; thus none of
high correlation between initial values. Thus, although
them can be used to separate the two effects during
many factors of variability among subjects were not fatiguing contractions. 2) Spectral variables and %Det
included in the simulations for limiting the computa- are highly correlated, and thus the information they
tional effort (differences of subcutaneous layer thick- provide on muscle properties is similar. The correlation
nesses, variability in the orientation of the detection between %Rec and spectral variables is much poorer
system with respect to the fibers, length of the fibers, than that for %Det, although %Rec, like %Det, is sim-
etc.) and within the limitations of the model used, as ilarly sensitive to changes in muscle properties. These
mentioned above (see Simulation of the MUAPs), the results are confirmed by both simulation and experi-
model results were still a valid tool for interpreting the mental findings and are particularly important for the
experimental signals. correct interpretation of linear and nonlinear variables
For both the simulations and the experimental sig- when comparing muscles or subjects. 3) Relative vari-
nals, %Rec resulted in poorer correlation with spectral ations of %Det (and %Rec) in response to changes in
variables with respect to %Det. Thus interpretation of muscle properties are significantly larger than relative
%Rec changes from the surface EMG signals may be variations of spectral variables. Thus nonlinear analy-
less direct than that of %Det changes. sis of the surface EMG signal is a technique more
The observed larger changes in %Det and %Rec in sensitive than spectral analysis for the assessment of
response to fatigue than those in MNF were predicted muscle fatigue and can be used for more detailed non-
by the simulation analysis. Webber et al. (42), in their invasive muscle assessment.
experimental study, indicated an even higher differ- The application of RQA to surface EMG still needs,
ence between the normalized rate of change in MNF however, research efforts in determining its sensitivity
and %Det (changes in %Det were approximately five to other factors of variability between subjects and
times larger than those in MNF). The discrepancy of muscles, such as the thickness of the subcutaneous
the latter result with those of our study is probably due layers, the orientation of the detection system, the
J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org
1762 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS

interelectrode distance, the electrode location, the re- Recurrence map RM. The recurrences between the states
cruitment of MUs, the additive noise, and so on. are evaluated from the matrix DM. A recurrence occurs if the
Whereas the influence of all of these parameters on distance d(i,j) is smaller than a threshold (referred to as
spectral variables has been analyzed in many past radius R), i.e., if v(i) and v(j) fall within a sphere of radius R
centered at v(i). The comparison of the distances in DM with
publications (see, for examples, Refs. 9–12, 14, 20, 24, R provides the recurrence map RM as the collection of the
33, 38, 47), there is a complete lack of this information pixels b(i,j), defined as follows
for RQA.
YES 3 pixel b共i,j兲 ⫽ ON
APPENDIX d共i,j兲 ⱕ R (5)
RQA NO 3 pixel b共i,j兲 ⫽ OFF
Although RQA has been described in detail in some past The value of R was always set, in accordance with Ref. 46,
works, it is necessary in the context of this study to recall the smaller than 10% of the normalized mean distance (its value
basic principles of this technique. Given the surface EMG was adjusted depending on the data set). Because RM is
signal samples s(m) (m ⫽ 0, . . ., M), the following steps have symmetric, for the computation of the quantitative indexes of
been followed to compute the recurrence maps and the rela- the recurrence map, only the upward or downward portion of
tive indexes. it is considered.
Embedding procedure. From the surface EMG, vectors Quantification of the recurrence map. Often nonlinear plots
representing the states of the system are extracted. To do contain subtle patterns that are difficult to detect by simple
this, data are lagged by an integer number ␭ of samples visual inspection. Thus quantitative descriptors that empha-
(“time delay”), which defines the time distance between two size different features of the map have been proposed (42).

Downloaded from jap.physiology.org on January 6, 2009

uncorrelated samples. Selection of ␭ is performed on individ- Pixels ON are defined as recurrent (Re) if they share local
ual time series; in the present study (simulated and experi- neighborhoods in higher dimensional space (42). The per-
mental data), ␭ values ranged between 3 and 6. centage of plot occupied by Re points is defined as the %Rec.
The space dimension (phase space), referred to as embed- The percentage of Re forming lines parallel to diagonal is
ding dimension, has been selected in the present study as 15 defined as %Det. Lines are constituted by two or more points
(according to Ref. 18). The following set of D-dimensional that are diagonally adjacent with no intervening white
vectors v(n) is extracted from the original myoelectric time spaces. Thus %Det is defined as %Det ⫽ L/Re ⫻ 100, where
series L is the number of points forming lines.
Figure 3 shows examples of recurrence maps computed
v共1兲 ⫽ 兵s共1兲 s共␭ ⫹ 1兲 · · · s关共D ⫺ 1兲 ␭兴其 from simulated signals. The presence of diagonal structures
v共2兲 ⫽ 兵s共2兲 s共␭ ⫹ 2兲 · · · s关共D ⫺ 1兲 ␭ ⫹ 1兴其 can be clearly observed for the two simulated signals shown.
(1)
··· ··· ··· ··· ··· %Det and %Rec quantify this difference, as it appears from
v共N兲 ⫽ 兵s共N兲 s共N ⫹ ␭兲 · · · s关N ⫹ 共D ⫺ 1兲␭兴其 Fig. 3.

where N ⫽ M ⫺ (D ⫺ 1)␭. The authors are sincerely grateful to Prof. Marco Marchetti of the
Distance matrix DM. The states of the dynamic system Dipartimento di Fisiologia Umana, Università La Sapienza, Roma,
under consideration are represented by the vectors v(n) and to Prof. Roberto Merletti of the Dipartimento di Elettronica,
defined in Eq. 1. To define the recurrences between the Politecnico di Torino, Torino, for useful discussions and comments on
the manuscript.
states, their closeness should be quantified. To do so, the This work was partially supported by the European Shared Cost
most commonly used metric is the Euclidean distance (d) Project Neuromuscular Assessment of Elderly Workers (QLRT-
2000–00139), Fondazione CRT, and Compagnia di San Paolo di
d共i, j兲 ⫽ 兵 ⬍ 关v共i兲 ⫺ v共 j兲兴2 ⬎ 其 ⁄2
1
(2) Torino.

where ⬍ ⬎ indicates the summation of the elements of the REFERENCES

argument vector.
To make the results of the analysis independent of the 1. Andreassen S and Arendt-Nielsen L. Muscle fibre conduction
energy of the observed signal, the values adopted in RQA are velocity in motor units of the human anterior tibial muscle: a
new size principle parameter. J Physiol 391: 561–571, 1987.
normalized with respect to the average distance dav between
2. Arendt-Nielsen L and Mills KR. The relationship between
vectors mean power frequency of the EMG spectrum and muscle fibre
conduction velocity. Electroencephalogr Clin Neurophysiol 60:
N N
130–134, 1985.
兺 兺 d共i, j兲 3. Broman H, Bilotto G, and De Luca CJ. A note on the non-
invasive estimation of muscle fiber conduction velocity. IEEE
i⫽1 i⫽j
d av ⫽ (3) Trans Biomed Eng 32: 341–344, 1985.
N共N ⫺ 1兲/2 4. Disselhorst C, Silny J, and Rau G. Improvement of spatial
resolution in surface-EMG: a theoretical and experimental com-
where the denominator represents the number of possible parison of different spatial filters. IEEE Trans Biomed Eng 44:
distances d(i,j). 567–574, 1997.
The collection of the normalized distances provides a sym- 5. Eckmann JP, Kamphorst SO, and Ruelle D. Recurrence
metric matrix, indicated as distance matrix DM(N ⫻ N) plots of dynamical systems. Europhys Lett 4: 973–977, 1987.
6. Farina D. Advances in Surface EMG Signal Detection, Process-

冨 冨
d共1,1兲 d共1,2兲 · · · d共1,N兲 ing and Interpretation in Motor Control Studies (PhD Thesis).
Turin, Italy: Dept. of Electronics, Politecnico di Torino and Ecole
d共2,1兲 d共2,2兲 · · · d共2,N兲
DM ⫽ (4) Centrale de Nantes, 2001.
··· ··· ··· ··· 7. Farina D, Arendt-Nielsen L, Merletti R, and Graven-
d共N,1兲 d共N,2兲 · · · d共N,N兲 Nielsen T. Assessment of single motor unit conduction velocity

J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org

 LINEAR AND NONLINEAR SURFACE EMG ANALYSIS 1763

during sustained contractions of the tibialis anterior muscle with biceps and adductor pollicis during isometric contractions. Brain
advanced spike triggered averaging. J Neurosci Methods 115: Res 219: 45–55, 1981.
1–12, 2002. 27. Lago P and Jones NB. Effect of motor unit firing time statistics
8. Farina D and Cescon C. Concentric ring electrode systems for on EMG spectra. Med Biol Eng Comput 15: 648–655, 1977.
non-invasive detection of single motor unit activity. IEEE Trans 28. Lindstrom L and Magnusson R. Interpretation of myoelectric
Biomed Eng 48: 1326–1334, 2001. power spectra: a model and its applications. Proc IEEE 65:
9. Farina D, Cescon C, and Merletti R. Influence of anatomical, 653–662, 1977.
physical and detection system parameters on surface EMG. Biol 29. Matthews PB. Relationship of firing intervals of human motor
Cybern 86: 445–456, 2002. units to the trajectory of post-spike after-hyperpolarization and
10. Farina D, Fosci M, and Merletti R. Motor unit recruitment synaptic noise. J Physiol 492: 597–628, 1996.
strategies investigated by surface EMG variables. An experi- 30. Merletti R, Farina D, Gazzoni M, and Schieroni MP. Effect
mental and model based feasibility study. J Appl Physiol 92: of age on muscle functions investigated with surface electromyo-
235–247, 2002. graphy. Muscle Nerve 25: 65–76, 2002.
11. Farina D, Madeleine P, Graven-Nielsen T, Merletti R, and 31. Merletti R, Knaflitz M, and De Luca CJ. Myoelectric mani-
Arendt-Nielsen L. Stardardising surface electromyogram re- festations of fatigue in voluntary and electrically elicited con-
cordings for assessment of activity and fatigue in the human tractions. J Appl Physiol 69: 1810–1820, 1990.
upper trapezius muscle. Eur J Appl Physiol 86: 469–478, 2002. 32. Merletti R and Lo Conte L. Advances in processing of surface
12. Farina D and Merletti R. Comparison of algorithms for esti- myoelectric signals. Part 1. Med Biol Eng Comput 33: 362–372,
mation of EMG variables during voluntary isometric contrac- 1995.
tions. J Electromyogr Kinesiol 10: 337–349, 2000. 33. Merletti R and Roy S. Myoelectric and mechanical manifesta-
13. Farina D and Merletti R. A novel approach for precise simu- tions of muscle fatigue in voluntary contractions. J Orthop
lation of the EMG signal detected by surface electrodes. IEEE Sports Phys Ther 24: 342–353, 1996.
Trans Biomed Eng 48: 637–646, 2001. 34. Milner-Brown HS, Stein RB, and Yemm R. Changes in firing
14. Farina D, Merletti R, Nazzaro M, and Caruso I. Effect of rate of human motor units during linearly changing voluntary
joint angle on EMG variables in leg and thigh muscles. IEEE contractions. J Physiol 230: 371–390, 1973.

Downloaded from jap.physiology.org on January 6, 2009

Eng Med Biol Mag 20: 62–71, 2001. 35. Nieminen H and Takala EP. Evidence of deterministic chaos
15. Farina D and Rainoldi A. Compensation of the effect of sub- in the myoelectric signal. Electromyogr Clin Neurophysiol 36:
cutaneous tissue layers on surface EMG: a simulation study. 49–58, 1996.
Med Eng Phys 21: 487–496, 1999. 36. Nordstrom MA, Fuglevand AJ, and Enoka RM. Estimating
16. Felici F, Rosponi A, Sbriccoli P, Filligoi GC, Fattorini L, the strength of common input to human motoneurons from the
and Marchetti M. Linear and non-linear analysis of surface cross-correlogram. J Physiol 453: 547–574, 1992.
electromyograms in weightlifters. Eur J Appl Physiol 84: 337– 37. Rosenfalck P. Intra and extracellular fields of active nerve and
342, 2001. muscle fibers. A physico-mathematical analysis of different mod-
17. Felici F, Rosponi A, Sbriccoli P, Scarcia M, Bazzucchi I, els. Acta Physiol Scand 321: 1–49, 1969.
and Iannattone M. Effect of human exposure to altitude on 38. Roy SH, De Luca CJ, and Schneider J. Effects of electrode
muscle endurance during isometric contractions. Eur J Appl location on myoelectric conduction velocity and median fre-
Physiol 85: 507–512, 2001. quency estimates. J Appl Physiol 61: 1510–1517, 1986.
18. Filligoi G and Felici F. Detection of hidden rhythms in surface 39. Sadoyama T, Masuda T, Miyata H, and Katsuta S. Fiber
EMG signals with a non-linear time-series tool. Med Eng Phys conduction velocity and fibre composition in human vastus late-
21: 439–448, 1999. ralis. Eur J Appl Physiol 57: 767–771, 1988.
19. Fuglevand AJ, Winter DA, and Patla AE. Models of recruit- 40. Stegeman DF, Blok JH, Hermens HJ, and Roeleveld K.
ment and rate coding organization in motor unit pools. J Neu- Surface EMG models: properties and applications. J Electro-
rophysiol 70: 2470–2488, 1993. myogr Kinesiol 10: 313–326, 2000.
20. Gydikov A and Kosarov D. Volume conduction of the poten- 41. Stulen FB and De Luca CJ. Frequency parameters of the
tials from separate motor units in human muscle. Electromyogr myoelectric signals as a measure of muscle conduction velocity.
Clin Neurophysiol 12: 127–147, 1972. IEEE Trans Biomed Eng 28: 515–523, 1981.
21. Henneman E. Skeletal muscle: the servant of the nervous 42. Webber CL Jr, Schmidt MA, and Walsh JM. Influence of
system. In: Medical Physiology (14th ed.), edited by Mountcastle isometric loading on biceps EMG dynamics as assessed by linear
VB. St. Louis, MO: Mosby, 1980, p. 674–702. and nonlinear tools. J Appl Physiol 78: 814–822, 1995.
22. Hermens HJ, Bruggen TAM, Batem CTM, Rutten WLC, 43. Westbury JR and Shaughnessy TG. Associations between
and Boom HBK. The median frequency of the surface EMG spectral representation of the surface electromyogram and fiber
power spectrum in relation to motor unit firing and action type distribution and size in human masseter muscle. Electro-
potential properties. J Electromyogr Kinesiol 2: 15–25, 1992. myogr Clin Neurophysiol 27: 427–435, 1987.
23. Ikegawa S, Shinohara M, Fukunaga T, Zbilut JP, and 44. Weytjens JL and van Steenberghe D. The effects of motor
Webber CL Jr. Nonlinear time-course of lumbar muscle fatigue unit synchronization on the power spectrum of the electromyo-
using recurrence quantifications. Biol Cybern 82: 373–382, 2000. gram. Biol Cybern 51: 71–77, 1984.
24. Jensen C, Vasseljen O, and Westgaard RH. The influence of 45. Yao W, Fuglevand RJ, and Enoka RM. Motor-unit synchro-
electrode position on bipolar surface electromyogram recordings nization increases EMG amplitude and decreases force steadi-
of the upper trapezius muscle. Eur J Appl Physiol 67: 266–273, ness of simulated contractions. J Neurophysiol 83: 441–452,
1993. 2000.
25. Kleine BU, Stegeman DF, Mund D, and Anders C. Influence 46. Zbilut J and Webber C. Embeddings and delays as derived
of motorneuron firing synchronization on SEMG characteristics from quantification of recurrence plots. Phys Letters 171: 199–
in dependence of electrode position. J Appl Physiol 91: 1588– 203, 1992.
1599, 2001. 47. Zuniga N, Truong XT, and Simons DG. Effects of skin elec-
26. Kukulka CG and Clamann HP. Comparison of the recruit- trode position on average electromyographic potentials. Arch
ment and discharge properties of motor units in human brachial Phys Med Rehabil 51: 264–272, 1970.

J Appl Physiol • VOL 93 • NOVEMBER 2002 • www.jap.org