You are on page 1of 6

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 45, NO.

6, JUNE 1998 795

Communications
Influence of Smoothing Window Length on smoothing window (T , in seconds) applied to the data as1 [4]
Electromyogram Amplitude Estimates p
SNR = 2 1 2Bs 1 L 1 T : (1)
Yves St-Amant, Denis Rancourt, and Edward A. Clancy* (This formula assumes that each EMG channel has the same statistical
bandwidth.) Note that for sampled data, T = N=f , where N is the
number of samples in the smoothing window and f is the sampling
Abstract—A systematic, experimental study of the influence of smooth- frequency in Hz. The larger the SNR, the better the EMG amplitude
ing window length on the signal-to-noise ratio (SNR) of electromyogram
(EMG) amplitude estimates is described. Surface EMG waveforms were
estimate. Investigators have looked at experimental and theoretical
sampled during nonfatiguing, constant-force, constant-angle contractions methods to improve the SNR by increasing the signal bandwidth
of the biceps or triceps muscles, over the range of 10%–75% maximum (i.e., applying whitening filters to the EMG waveform) [3]–[9] and
voluntary contraction. EMG amplitude estimates were computed with the number of EMG channels [3], [4], [6], [10]–[12] contributing to
eight different EMG processor schemes using smoothing length durations an amplitude estimate. Limited attention, however, has been focused
spanning 2.45–500 ms. An SNR was computed from each amplitude
estimate (deviations about the mean value of the estimate were considered on determining experimentally the role of smoothing window length
6
as noise). Over these window lengths, average standard deviation SNR’s on the SNR. Inman et al. [1] examined three time constants ( =
6 6
ranged from 1.4 0.28 to 16.2 5.4 for unwhitened single-channel EMG 100, 200, 250 ms) with their original analog processor (a full-wave
processing and from 3.2 6
0.7 to 37.3 6
14.2 for whitened, multiple- rectifier followed by a resistor-capacitor low-pass filter). Qualitative
channel EMG processing (results pooled across contraction level). It was
found that SNR increased with window length in a square root fashion.
performance results were shown, but no quantitative SNR results
The shape of this relationship was consistent with classic theoretical were given. Kreifeldt [13] investigated the SNR for three settling
predictions, however none of the processors achieved the absolute perfor- times (250, 500, 1000 ms) from three analog smoothing circuits
mance level predicted by the theory. These results are useful in selecting (first-order Butterworth, third-order Butterworth, and third-order
the length of the smoothing window in traditional surface EMG studies. averaging filters) at four constant-force, constant-angle contraction
In addition, this study should contribute to the development of EMG
processors which dynamically tune the smoothing window length when levels [5%, 10%, 25%, and 50% maximum voluntary contraction
the EMG amplitude is time varying. (MVC)]. Because contraction durations were between 40 and 60 s,
it is likely that some of the contractions (particularly those at higher
Index Terms— Biological system modeling, electromyography, EMG
amplitude estimation, modeling, whitening. MVC levels) were fatiguing. Their results showed that increasing
the settling time increased the SNR. Thusneyapan and Zahalak [12]
examined four cutoff frequencies (5, 10, 20, and 30 Hz) with a
nine-channel analog processor which performed analog rectification,
I. INTRODUCTION combination and low-pass (second-order Butterworth) filtering.
For EMG amplitude estimates, the SNR ranged from 4.6 (with the
The amplitude of the surface electromyogram (EMG) waveform 30 Hz cutoff frequency) to 6.9 (with the 5-Hz cutoff frequency).2
has been used as a noninvasive tool for assessing the degree of Taken together, these prior investigations certainly demonstrate that
muscular exertion (e.g., as the command input of EMG-controlled the SNR increases with the smoothing window length for constant-
prosthetic limbs), and is being investigated as an indicator of the force, constant-angle, nonfatiguing contractions. However, processors
force developed by the muscles. Classical methods for processing which increase the statistical bandwidth of the EMG signal by
the EMG waveform (e.g., signal rectification followed by low-pass including whitening filters (single or multiple channel) have not been
filtering [1]) provide a processed EMG amplitude which is investigated. In addition, and most importantly, the data are too sparse
inherently noisy. For constant-force, constant-angle, nonfatiguing to validate the fit of an analytical function such as the theoretical
muscular contraction, the signal-to-noise ratio (SNR) of EMG square root relationship of Hogan and Mann [4]. Use of an analytical
amplitude estimates (defined as the sample mean of the estimate function to represent the variation of the SNR with window length is
divided by its sample standard deviation) using root-mean-square mathematically convenient for the optimization process of dynamic
(RMS) detection has been shown theoretically to be related to the EMG adaptive window length processors. In this case, higher fidelity
statistical bandwidth [2] of the EMG signal (Bs , in Hz), the number EMG amplitude estimates can be achieved if the window length
of EMG channels recorded on a muscle (L), and the length of the is tuned throughout the duration of a contraction [7], [14]–[18].
Basically, a tradeoff is maintained between error due to estimator
variance (which is diminished via a long smoothing window) and
Manuscript received July 7, 1997; revised December 10, 1997. This work error due to estimator bias (which is diminished in the dynamic case
was supported in part by National Institutes of Health under Grant AR40029,
1 Hogan and Mann [3], [4] actually present this square root formula as
in part by the United States Department of Education under NIDRR Grant
H133E80024, and in part by the Conseil de Recherche en Sciences Naturelles a simplification/approximation of the true SNR formula. We compared the
et en Génie du Canada. Asterisk indicates corresponding author. square root formula to the true formula using MATLAB (version 4.0). For
Y. St-Amant and D. Rancourt (e-mail: boloria@gmc.ulaval.ca) are with the values of 2Bs 1 L 1 T over the range of 12 to 10 000 the two formulas never
Mechanical Engineering Department, Laval University, Québec, P.Q., Canada differ by more than 0.05 and the percentage difference is always less than 1%.
G1K 7P4. The difference drops rapidly as 2Bs 1 L 1 T increases. Thus, for all intents and
*E. A. Clancy is with the Liberty Mutual Research Center for Safety purposes, the square root formula given here can be considered equivalent to
and Health, 71 Frankland Road, Hopkinton, MA 01748 USA (e-mail: ms- the more complex formula found in [3].
mail5.clancye@tsod.lmig.com). 2 These result values have been approximated from a graph in [12], since
Publisher Item Identifier S 0018-9294(98)03721-5. the actual values are not found in the text.

0018–9294/98$10.00  1998 IEEE

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 1, 2009 at 06:05 from IEEE Xplore. Restrictions apply.
796 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 45, NO. 6, JUNE 1998

with a short smoothing window). Previous authors [7], [14], [17] the electrode-amplifiers, some electrode-amplifiers saturated during
have shown that this tradeoff can be written mathematically as: MSE some experiments. All data from any electrode-amplifier from a
2 2
= b +  , where MSE is the mean square error in the EMG amplitude given subject which saturated during any portion of any record
estimate, b is the bias error and  2 is the variance error. Since the were discarded from further analysis. A total of 100 multiple site
variance represents the purely stochastic portion of the error, it is recordings, comprised of 660 single site recordings, were available
related to the SNR in (1) as: SNR = s2 = 2 , where s is the EMG for analysis. Previous experimental studies based on the data from
amplitude estimate (e.g., from mean-absolute-value (MAV) or RMS these trials are described in [9] and [11].
processing) and  is the standard deviation of the estimate. Hence,
establishing the experimental relationship between SNR and window
length in the constant-force, constant-angle, nonfatiguing situation B. Methods of Analysis
has direct bearing on EMG processing for dynamic situations. This
report provides such evaluation. A preliminary report of this work The SNR versus smoothing window length relation was studied
appeared in [19]. with unwhitened/whitened single/multiple channel EMG processors
which used MAV/RMS detectors. Thus, eight EMG amplitude pro-
cessors were formed. As a preprocessing step, the sample-mean value
II. EXPERIMENTAL STUDY
of each 5-s EMG waveform was subtracted from the waveform. This
step provided a simple method for removal of individual channel
A. Experimental Apparatus and Methods
offsets due to the A/D converter and front-end electronics. The
The experimental apparatus and methods have been described in mean-adjusted waveform samples were then used to form the eight
detail elsewhere [9], [11], [20]. Briefly, a chair was instrumented processors as follows.
to measure the torque generated about the elbow. The subject was 1) A single channel unwhitened amplitude processor was
seated and secured into a straight-back chair via five quick-release formed as the simple RMS detector.
belts. The subject’s right arm was oriented so that the upper arm and 2) A single channel whitened processor was formed by tem-
forearm were in the plane parallel to the floor (shoulder abducted porally whitening each data record, followed by RMS
90 from the anatomic position), the forearm was oriented in the detection (see [9], [11], and [20] for details of the whitening
parasagittal plane, with the wrist in complete supination, and the technique). Briefly, a fourth-order moving average whiten-
elbow flexion angle was about 90 . The subject’s right wrist was ing filter was constructed for each electrode-amplifier for
mounted, via a wrist cuff, to a cantilever beam. The torque produced each subject, and applied to all trials recorded by that
by the elbow was measured by a complete strain gauge Wheatstone electrode-amplifier. For each subject, a whitening filter
bridge affixed on the beam. Eight commercial electrode-amplifiers was calibrated by averaging the autoregressive (AR) power
(Liberty Mutual MYO111 [21] with a second-order 100–500 Hz spectrum coefficients computed from each 50% MVC trial.
bandpass filter) were placed in-line, side-by-side, transversely across The whitening filter model order and choice of 50% MVC
the flexor (biceps brachii) or extensor (triceps) muscles of the elbow. as the calibration contraction were selected based on prior
The two electrode contacts of each electrode-amplifier were oriented investigation conducted on these data [9].
along the long axis of the upper arm (i.e., oriented along the assumed 3) A multiple channel (four channel) unwhitened amplitude
direction of action potential conduction). The electrode-amplifiers processor was formed by equalizing (normalizing to a value
were located approximately midway between the elbow and the of one) the variance of each channel (based on the average
midpoint of the upper arm, clustered about the muscle midline. Each variance of the five 50% MVC trials), and then performing
electrode-amplifier consisted of a pair of 4-mm diameter, stainless- spatial-temporal RMS detection. If the multiple channel
steel, hemispherical electrode contacts separated by a distance of 15 EMG waveforms (after variance equalizing) at sample k
mm (center to center). The center-to-center distance between adjacent are denoted ml (k) and the channel index l ranges from
electrode-amplifiers was approximately 1.75 cm. one to L = 4, then spatial-temporal RMS detection to
Five subjects (four male and one female, ranging in age from 23 to form this EMG amplitude estimate is written as s^3 (k) =
37 years), with no known neuromuscular deficits of the right shoulder, 1 L k 2 1=2
arm or hand, participated in the experiment. Informed consent was N 1L l=1 i=k0N +1 ml (i).
received from each subject. Two subjects were studied during an 4) A multiple channel (four channel) whitened amplitude pro-
elbow flexion task, three were studied during an elbow extension cessor was formed by temporally whitening each channel
task. During an experimental trial, the output voltage of the strain (which inherently normalizes the variance of each chan-
gauge circuit and a target torque level were presented as the two nel), and then performing spatial-temporal RMS detection.
displays of a dual-trace oscilloscope. The subject was instructed to Whitening filters were the same as in the single channel
begin at rest, then gradually increase flexion/extension torque until case. Spatial uncorrelation was not performed, since, based
the target torque level was achieved (typically over a period of 0.5–1 on prior work [11], [20] it provided little performance
s). The subject maintained the target torque level until a 5-s segment improvement with this electrode arrangement.
of data was recorded. 5–8) These processors parallel the above processors, except that
Two initial 3-s MVC trials were averaged to provide a rough RMS detection is replaced with MAV detection.
estimate of the strain gauge circuit output voltage corresponding For each processor, the smoothing window length was varied from
to MVC. A sequence of five sets of constant-force, constant-angle a minimum value of five samples (2.45 ms) to a maximum value
contractions was then conducted. Each set consisted of four trials, of 1024 samples (500 ms). SNR’s of each EMG amplitude were
one trial each at 10%, 25%, 50%, and 75% MVC. Trials within computed as the square root of the ratio of the squared amplitude
a set were randomized. A rest period of 2 min between trials (3 estimate sample mean divided by the amplitude estimate sample
min after 100% MVC trials) was provided to prevent fatigue [22]. variance. To avoid start-up transients, the first 1024 + 5= 1029
EMG and strain gauge circuit output data were sampled at 2048 Hz, samples of each amplitude estimate were discarded prior to computing
using a 12-bit A/D converter, and processed off-line. Each data record the SNR. All computation was performed using MATLAB (version
was plotted and visually inspected. Due to the high gain (3600) of 4.0) on an IBM-compatible PC.

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 1, 2009 at 06:05 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 45, NO. 6, JUNE 1998 797

Fig. 1. Effect of smoothing window length on the SNR (data pooled across subjects and force levels). Mean SNR’s are shown for the eight processors.
Lines are an aid to the eye only. One-sided error bars show the standard deviations for the MAV processors. (Standard deviations for the RMS processors
are similar.) Inset table tabulates the mean 6
standard deviation SNR’s for the MAV processors.

III. EXPERIMENTAL RESULTS on the logarithm of the SNR were initially performed because what
The aim of this study was to obtain a more complete picture of might be true at a given window length may not be so at another.
the SNR variation over a practical range of window lengths. The The tests were conducted at only two different window lengths: T =
SNR performance for the eight processors is shown in Fig. 1. Data 30 ms and T = 500 ms. Results being generally similar, we pursued
from all contraction levels and all subjects were pooled together. the statistical analysis using a parametric test where the SNR curves
Hence, results for single channel processors are derived from 660 (as a function of smoothing window length) were least-squares fitted
SNR’s, while those for multiple channel processors are derived from to the following model:
100 SNR’s. Using a 245-ms smoothing window, Clancy and Hogan
[9], [11] showed that the SNR curves perform differently from one
ln(SNR) = ln(2 1 2Bs L 1 T )b
1
processor to another. Our results show that this difference appears
to be true for any of the window lengths that were studied. Both = b1 ln(2 1 2Bs 1 L) + b1 ln T
whitening and spatio-temporal techniques seemed to improve the = b0 + b1 ln T:
SNR performance, and MAV processors consistently produced equal
or higher SNR values than RMS processors. A more detailed analysis
of the SNR variation was conducted by grouping data by subject or This model is a more generalized form of (1). The influence of
by contraction level. Grouping data by subject showed that only one different factors on the SNR was then analyzed using both b0 and
subject had generally higher SNR values than the others, and all five b1 variations. Values for b0 and b1 are given in Table I, for the four
subjects still exhibited a square root behavior. The SNR curves for contraction levels, for all eight processors. The analysis of variance
all contraction levels are shown in Fig. 2. The curves suggest that test on all the data showed that there were significant interactions
the SNR performance decreases with contraction level, but all curves between subjects, force levels and processors. Further analysis of
still exhibit a square root behavior. variations across subjects was not of interest and, thus, was not
In order to provide statistical power to the previous observations, a pursued.
multivariate analysis of variance (ANOVA) (with = 0:05) was per- Statistical tests showed that b1 was significantly different from
formed using SAS, version 6.09 (SAS Institute Inc., Cary, NC). The zero, hence supporting the observation that the SNR varies with
analysis was performed after applying a logarithmic transformation window length. In addition, all processors, at all contraction levels,
to the SNR data. This particular transformation was chosen to satisfy have a b1 close to 0.5, indicating a square root relationship between
the hypothesis of homogeneous variance required by the statistical SNR and window length. Both parameters, b0 and b1 , were ordered
analysis. Following this transformation, it was found that the residuals and statistical comparisons were performed using a Tukey (LSD)
of the fitted models were distributed uniformly. Nonparametric tests comparison test, with = 0:05. Statistical analysis of the data

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 1, 2009 at 06:05 from IEEE Xplore. Restrictions apply.
798 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 45, NO. 6, JUNE 1998

Fig. 2. Influence of contraction force level on SNR (data pooled across subjects). Mean SNR’s (and one-sided standard deviations) are shown. Lines are
an aid to the eye only. In all cases, EMG processing uses MAV detection and four whitened channels. Each entry is averaged across 25 SNR’s. Inset
table tabulates the mean 6standard deviation SNR’s shown in the plots.

TABLE I
FITTING COEFFICIENTS b0 AND b1 FOR THE FOUR CONTRACTION LEVELS, FOR THE EIGHT PROCESSORS

b0 b1
10% 25% 50% 75% 10% 25% 50% 75%
Processor
MVC MVC MVC MVC MVC MVC MVC MVC
MAV, unwhite, single 3.20 3.14 3.01 2.95 0.494 0.484 0.469 0.456
MAV, unwhite, multiple 3.62 3.50 3.31 3.17 0.514 0.492 0.469 0.441
MAV, white, single 3.78 3.65 3.37 3.34 0.514 0.492 0.465 0.466
MAV, white, multiple 4.21 4.06 3.77 3.57 0.506 0.487 0.483 0.452
RMS, unwhite, single 3.12 3.05 2.92 2.84 0.473 0.463 0.447 0.431
RMS, unwhite, multiple 3.52 3.38 3.16 3.07 0.494 0.460 0.435 0.416
RMS, white, single 3.78 3.64 3.32 3.25 0.510 0.484 0.453 0.445
RMS, white, multiple 4.15 3.97 3.69 3.35 0.503 0.467 0.468 0.413
The coefficients fit the model: (SNR) , where SNR is the signal-to-noise ratio and is the
smoothing window length (in seconds).

confirmed that: 1) whitening improves the SNR performance (increase distinguished statistically. However, MVC levels 50% and 75% each
of b0 ), 2) multiple channel combination improves the performance in differed significantly from 10% MVC.
a similar manner, 3) both techniques together help even more, and
4) MAV processors, although always ranked higher than RMS, are IV. DISCUSSION
not statistically different than the RMS. This last result suggests that
MAV processors—which often have easier and faster realizations in A. SNR Square Root Relationship
both analog and digital systems—can be successfully substituted for As predicted by the theoretical work of Hogan and Mann [3],
the theoretically prescribed RMS processors. [4], the SNR performance generally exhibited a square root behavior
Results showed that both fitting parameters b0 and b1 decrease with respect to window length, since the b1 values were close
with increasing force level, and hence the SNR. In the global to 0.5. However, none of the processors achieved the level of
parametric analysis, the variation of b1 was ranked using a Tukey performance specified by the theory. Therefore, adaptive window
(LSD) comparison test ( = 0:05). Adjacent MVC levels—10% length EMG amplitude estimators which use this theoretical model
versus 25%, 25% versus 50%, and 50% versus 75%—could not be may underestimate the variance component of the error. For example,

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 1, 2009 at 06:05 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 45, NO. 6, JUNE 1998 799

TABLE II increases, motor unit recruitment and firing rates increase, perhaps
FITTING FACTOR “c” AND AVERAGE ERRORS (BETWEEN THE introducing systematic differences in the EMG bandwidth (and, thus,
FIT AND EXPERIMENTAL DATA) FOR ALL EIGHT PROCESSORS the SNR).
Fitting
Mean
Processor factor
error C. Whitening at Low Contraction Levels
"c"
MAV, unwhite, single 526 0.12
As expected from (1) and prior experimental work [3]–[9], in-
MAV, unwhite, multiple 1093 0.22 creasing the signal bandwidth (Bs ) via whitening generally improved
MAV, white, single 1278 0.29 SNR performance. From Fig. 1 it would appear that whitening always
MAV, white, multiple 2854 0.47 provides a processing benefit. This is not necessarily the case. Initial
RMS, unwhite, single 469 0.13 evaluation of whitening filters calibrated with a 50% MVC trial,
RMS, unwhite, multiple 927 0.23 then applied to data recorded during contractions at less than 10%
RMS, white, single 1235 0.30 MVC (a situation not investigated in this report) provided poor results
RMS, white, multiple 2524 0.43 [9], [20]. An additive background noise was found superimposed on
The factor fits the model: SNR , where SNR the EMG signal, e.g., a nonzero EMG amplitude was noted when
is the experimentally adjusted signal-to-noise ratio and is the the subjects were asked to fully relax their muscles. At low EMG
smoothing window length (in seconds).
amplitude, this additive background noise (not included in classical
EMG models) seemed to dominate the output of the whitening filters.
a multiple channel, whitened, RMS processor (i.e., L = 4; Bs = The relative contribution of background noise to the EMG amplitude
1024 Hz) has a theoretical SNR versus T relationship of estimates seemed to increase progressively as the EMG amplitude
p decreased. Similar difficulties have been noted by Kaiser and Peterson
SNRL=4;B =1024 Hz = 16 384 1 T : [5]. The use of adaptive whitening filters, which tune the shape of
the whitening filter as a function of the EMG amplitude, offers a
Thus, for T = 200 ms, the theoretically predicted SNR value is
promising solution to this problem [5], [20].
57.2. Fig. 1 shows that this performance level is not achieved—the
actual average SNR is 24.5 (with the MAV detector). Further,
experimentally achieved performance seems to be consistently below V. SUMMARY AND CONCLUSIONS
the theoretical prediction. Not all of the statistical degrees of freedom This study has demonstrated that a simple square root formula
modeled in the theoretical analysis have been recovered from the can be used to represent the experimental SNR performance of
experimental data. This observation was explored by least-squares eight different EMG estimators. An exhaustive mapping of the
fitting the SNR versus T relationship to anp“experimentally adjusted SNR performance versus window length was obtained for the EMG
(EA)” square root model; SNREA = c 1 T , where “c” is a estimators. Both whitening and spatio-temporal techniques improved
multiplicative factor representing the theoretical factor 2 1 2Bs 1L. The SNR performance, and MAV processors consistently produced equal
fitting factors (and fit errors) for all eight processors are tabulated in or higher SNR values than RMS processors. It was also found that
Table II. In general, the fit is quite good. These fitting factors can be SNR performance decreased with increasing contraction level. The
used to more accurately represent the experimentally achieved SNR results of this study can be used to select an appropriate smoothing
versus T relationship for the eight EMG processors studied. window length in traditional surface EMG studies. In addition, the
At least two reasons for the poorer experimental performance results of this study can be used to develop adaptive window length
(compared to the theoretical model) can be postulated. First, the EMG processors in which the length of the smoothing window is
theory does not model measurement noise in the electrode recording dynamically tuned when the EMG amplitude is time varying.
when, in fact, such noise clearly exists. This noise would likely
decrease the performance of any EMG amplitude processor. Second, ACKNOWLEDGMENT
the theoretical model is based on the assumption that the EMG
waveform samples are distributed as a Gaussian random process. Experimental data for this study were collected at the Eric P. and
However, a recent analysis of the data used in this study suggests Evelyn E. Newman Laboratory for Biomechanics and Human Reha-
that these recordings are more closely represented as a Laplacian bilitation at the Massachusetts Institute of Technology, Cambridge.
random process. Further, theoretical and simulation analysis of MAV The authors would like to thank N. Plante of the Statistics Consulting
and RMS processing of Laplacian data showed that the predicted Service, Department of Mathematics and Statistics, Laval University,
SNR of the amplitude estimate is considerably reduced (by 30%) Québec, Canada, for serving as a statistical consultant to this project.
[23]. Hence, minor variations in the density function of the sampled
REFERENCES
data can have a large influence on the achieved SNR.
[1] V. T. Inman, H. J. Ralston, J. B. de C. M. Saunders, B. Feinstein, and E.
B. Influence of Contraction Level W. Wright, “Relation of human electromyogram to muscular tension,”
EEG Clin. Neurophysiol., vol. 4, pp. 187–194, 1952.
The results indicated that experimental SNR decreased as the [2] J. S. Bendat and A. G. Piersol, Random Data: Analysis and Measurement
contraction level increased. Although no definitive reason for this Procedures. New York: Wiley, 1971.
observation can be provided, two possibilities will be discussed. [3] N. Hogan and R. W. Mann, “Myoelectric signal processing: Optimal
estimation applied to electromyography—Part I: Derivation of the
First, note that high-frequency (14 Hz) periodic oscillations were
optimal myoprocessor,” IEEE Trans. Biomed. Eng., vol. BME-27, pp.
found superimposed on the force recordings during the 50% and 75% 382–395, 1980.
MVC’s. These oscillations seemed to be due to normal tremor activity [4] , “Myoelectric signal processing: Optimal estimation applied to
associated with the high force output of the muscles. With tremor electromyography—Part II: Experimental demonstration of optimal my-
activity during high contractions, muscle activation deviated, in part, oprocessor performance,” IEEE Trans. Biomed. Eng., vol. BME-27, pp.
396–410, 1980.
from a constant-force contraction, and likely hindered the achieved [5] E. Kaiser and I. Petersen, “Adaptive filter for EMG control signals,” in
SNR. Second, actual changes in the physiologic EMG spectrum The Control of Upper-Extremity Prostheses and Orthoses. Springfield,
may occur as a function of contraction level. As contraction force IL: Thomas, 1974, pp. 54–57.

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 1, 2009 at 06:05 from IEEE Xplore. Restrictions apply.
800 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 45, NO. 6, JUNE 1998

[6] M. I. A. Harba and P. A. Lynn, “Optimizing the acquisition and Optimal Detection of Visual Evoked Potentials
processing of surface electromyographic signals,” J. Biomed. Eng.,
vol. 3, pp. 100–106, 1981. Carlos E. Davila,* Richard Srebro, and Ibrahim A. Ghaleb
[7] T. D’Alessio, “Some results on the optimization of a digital processor
for surface EMG signals,” Electromyogr. Clin. Neurophysiol., vol. 24,
pp. 625–643, 1984.
[8] G. C. Filligoi and P. Mandarini, “Some theoretic results on a digital Abstract—We consider the problem of detecting visual evoked potentials
EMG signal processor,” IEEE Trans. Biomed. Eng., vol. BME-31, pp. (VEP’s). A matched subspace filter is applied to the detection of the
333–341, 1984. VEP and is demonstrated to perform better than a number of other
[9] E. A. Clancy and N. Hogan, “Single site electromyograph amplitude evoked potential detectors. Unlike single-harmonic detectors, the matched
estimation,” IEEE Trans. Biomed. Eng., vol. 41, pp. 159–167, 1994. subspace filter (MSF) detector is suitable for detecting multiharmonic
[10] W. R. Murray and W. D. Rolph, “An optimal real-time digital processor VEP’s. Moreover, the MSF is optimal in the uniformly most powerful
for the electric activity of muscle,” Med. Instrum., vol. 19, pp. 77–82, sense for multiharmonic signals with unknown noise variance.
1985.
Index Terms—Detection, evoked potentials, prewhitening, visual acuity.
[11] E. A. Clancy and N. Hogan, “Multiple site electromyograph amplitude
estimation,” IEEE Trans. Biomed. Eng., vol. 42, pp. 203–211, 1995.
[12] S. Thusneyapan and G. I. Zahalak, “A practical electrode-array myopro-
cessor for surface electromyography,” IEEE Trans. Biomed. Eng., vol. I. BACKGROUND
36, pp. 295–299, 1989. Visual grating acuity (GA) is useful in the clinical evaluation
[13] J. G. Kreifeldt, “Signal versus noise characteristics of filtered EMG
of patients with eye and neurologic disease. GA is obtained by
used as a control source,” IEEE Trans. Biomed. Eng., vol. BME-18, pp.
16–22, 1971. having the subject view a contrast grating at a fixed contrast (usually
[14] T. D’Alessio, “Analysis of a digital EMG signal processor in dynamic 100%) while the spatial frequency is increased until the subject
conditions,” IEEE Trans. Biomed. Eng., vol. BME-32, pp. 78–82, 1985. can no longer detect the contrast grating. In adults, measurements
[15] R. R. Fullmer, S. G. Meek, and S. C. Jacobsen, “Optimization of an can be accomplished psychophysically; however, infants, young chil-
adaptive myoelectric filter,” in Proc. Ann. Int. Conf. IEEE Eng. Med.
Biol. Soc., 1984, pp. 586–591. dren, and nonverbal patients cannot be studied with psychophysical
[16] S. C. Jacobsen, S. G. Meek, and R. R. Fullmer, “An adaptive myoelectric methods. Several researchers have proposed using the steady-state
filter,” in Proc. Ann. Int. Conf. IEEE Eng. Med. Biol. Soc., 1984, pp. visual evoked potential (VEP) as an objective method of determining
592–596. GA [1]–[3]. Most of these methods utilize the second harmonic of
[17] E. Park and S. G. Meek, “Adaptive filtering of the electromyographic
stimulus contrast reversal frequency to detect the presence of a VEP.
signal for prosthetic control and force estimation,” IEEE Trans. Biomed.
Eng., vol. 42, pp. 1048–1052, 1995. The generalized T 2 statistic [4], the Tcirc
2
statistic [5], [6], and the
[18] S. G. Meek and S. J. Fetherston, “Comparison of signal-to-noise ratio Rayleigh phase criterion (RPC) [7], are representative of these types
of myoelectric filters for prosthesis control,” J. Rehab. Res. Dev., vol. of detection algorithms. All of the above statistics are parametric
29 pp. 9–20, 1992. in the sense that they assume that under the null hypothesis, the
[19] Y. St-Amant, D. Rancourt, and E. A. Clancy, “Effect of smoothing
window length on RMS EMG amplitude estimates,” in Proc. IEEE 22nd noise [electroencephalogram (EEG)] has a Gaussian density. This
Northeast Bioeng. Conf., 1996, pp. 93–94. assumption has been found to be reasonable by several investigators
[20] E. A. Clancy, “Stochastic modeling of the relationship between the [8], [9]. The record orthogonality test by permutation (ROTP) detector
surface electromyogram and muscle torque,” Ph.D. dissertation, Dept. looks at the power in ensemble averages derived via all possible
Elect. Eng. Comput. Sci., Massachusetts Inst. Technol., Cambridge, MA,
sign permutations of the data frames. If the average corresponding
to all + signs (i.e., no sign changes) is in the top 5% of all possible
Jan. 11, 1991.
[21] S. J. Greelish, “Five new amplifiers for detecting myo- and bioelectric
signals,” in Proc. 10th Annu. Conf. Rehab. Tech.—RESNA’87, San Jose, ensemble average powers, a detection is made, hence this detector is
CA, June 19–23, 1987, pp. 190–192. nonparametric [10].
[22] E. A. Muller, “Physiological methods of increasing human physical Victor and Mast compared the RPC, T 2 , and Tcirc 2
statistics and
work capacity,” Ergonomics, vol. 8, pp. 409–424, 1965. 2
[23] E. A. Clancy and N. Hogan, “Theoretic and experimental comparison of found that their Tcirc statistic outperformed the others [5]. One
root-mean-square and mean-absolute-value electromyogram amplitude drawback of all of these statistics is that they are all based on the
detectors,” in Proc. Ann. Int. Conf. IEEE Eng. Med. Biol. Soc., 1997, second harmonic of the contrast reversal frequency; there is no reason
pp. 1267–1270. to expect near-threshold evoked potentials (EP’s) to contain only the
second harmonic, and if other signal harmonics are present, then
current methodology does not appear to have exploited them.

Manuscript received November 20, 1996; revised January 13, 1998. This
work was supported in part by a grant from the Whitaker Foundation, in part
by the National Science Foundation under Grant BCS-9308028, and in part
by an unrestricted grant from Research to Prevent Blindness, Inc. Asterisk
indicates corresponding author.
*C. E. Davila is with the Electrical Engineering Department, Southern
Methodist University, P. O. Box 750338, Dallas, TX 75275–0338 USA (e-
mail: cd@seas.smu.edu).
R. Srebro is with the Departments of Ophthalmology and Biomedical
Engineering, University of Texas Southwestern, Medical Center, Dallas, TX
75275 USA.
I. A. Ghaleb is with the Electrical Engineering Department, Southern
Methodist University, Dallas, TX 75275-0338 USA.
Publisher Item Identifier S 0018-9294(98)03722-7.

0018–9294/98$10.00  1998 IEEE

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 1, 2009 at 06:05 from IEEE Xplore. Restrictions apply.

You might also like