Effect of Smoothing Window Length on RMS EMG Amplitude Estimates - Bioengineering Conference, 1996., Proceedings of the 1996 IEEE Twenty-Second Annual Northeast

This paper describes the results of a study on the influence of smoothing window length on the signal-to-noise ratio (SNR) of electromyogram (EMG) amplitude estimates. In a previous experiment, surface EMG waveforms were recorded for non-fatiguing, constant-force, isometric contraction of the biceps and triceps muscles. Based on these data, four different EMG processors were studied using smoothing window length durations spanning 24-500 ms. Results show that the SNR increases in a square root fashion with smoothing window length, as predicted by classical theory. These results are useful in choosing the length of the smoothing window in traditional surface EMG studies, and to adaptively tune the smoothing window length when the EMG amplitude is time-varying. INTRODUCTION complete stain gauge Wheatstone bridge. Up to eight commercial electrode-amplifiers were placed on the flexor and extensor muscles of the elbow. A sequence of five sets of non-fatiguing, constant-force, isometric contractions was then conducted. Each set consisted of four randomized trials, one each at 10, 25, 50 and 75% of maximum voluntary contraction (MVC). A total of 660 singlei 100 multiple site recordings were sampled at 2048 Hz by an 12bit AID converter. The relation between SNR and smoothing window length was studied with unoptimizedioptimized singleimultiple channel EMG processors. The sample mean of each EMG waveform was substracted from the data. These adjusted waveforms were then used to form the four processors as follow: l )A single channel unwhitened amplitude processor was formed as the simple RMS processor. If the EMG waveform at sample t is denoted m(t ) and N is the smoothing window length, then the EMG amplitude estimate s^, ( t ) was formed as The surface EMG waveform is a pattern of the electrical activitv of a muscle. The amplitude of the surface EMG waveform has been used as a non-invasive tool for assessing the t degree of muscular exertion. The quality of amplitude estimation can be quantified using a signal-to-noise (SNR) ratio value. For isometric, isotonic, non-fatiguing muscular contraction, the SNR of EMG amplitude estimates has been shown theorically to be related to the statistical bandwith B, of the EMC signal, the single channel whitelled (optimized) processor was by temporally whitening each data record, and performing RMS detection. A fourth-order moving average whitening filter was "Inber EMG recorded On a and the length constmcted for each electrode amplifier for each subject, and Of the window to the data as applied to all trials recorded by that electrode-amplifier. For SNR4-F [ l ] . The B, and L factors have been each subject, a whitening filter was calibrated by averaging the experilnentally examined previously, however, the influence of autoregressive (AR) power spectruln coefficients computed smoothing window length has not been analysed experimentally. from each 50% MVC The window length becomes quite important when the EMG 3)A multiple channel (four channel) unwhitened amplitude amplitude is dynamically varied i.e. when muscular contraction processor was formed by the Of each varies with time. In this case, better results can be obtained if the (based On the average variance Of the five j0% MVC alnplitude is rapidly varying, a short window length should be multiple Chm~e l EMG (after variance equalizing) at sample used and when the EMG amplitude is slowly varying, the window is denoted m k ( t ) and ranges from to = 4, then this length should be made longer. This paper describes the variation EMG amplitude estimate was folmed as the appropriate window length for EMG amplitude estimation. window leilgth is tuned during the contraction, When the EMG trials), then performing spatial-temporal RMs detection. If the of the SNR over a pratical range of window lengths to help select L t k=li=t-N+1 4)A multiple channel (four channel) whitened amplitude METHODS This study is based on data acquired in a previous experilnent Subjects were seated and strapped into a straight-back processor was by whitening each [2.3], the variance Of each then performing chair. The subject's right arm was oriented so that the upper limb spatial-temporal detection. was in a plane parallel to the floor with the elbow at a 90" flexion angle, The subject's right wrist was mounted, via a wrist cuff, to The window length was varied from 24 ms to 500 SNRs Of each EMG amp1itude were ms for each processor. a cantilever beam. Torque about the elbow was measured by a computed as the 'quare root Of the ratio Of the 'quared 93 0-7803-3204-0/96$05.00