Summary -- Probability Density Function for Surface Emg

My original query some time ago – one computer crash ago was about the amplitude probability density function (pdf) of surface EMG data (from which we have subtracted the mean value). That is, if one plots the amplitude of the rectified EMG in an histogram, what does it look like? The common assumption is that we get a (one-sided) Gaussian pdf for stationary data. This is generally so. However, looking at a few data sets (static) I observed that the pdf:s tend to be *supergaussian*; that is, they have heavy tails. Indeed, these tails have a distribution more like the exponential density [5] which decays slower than the Gaussian distribution. Normally this is of little consequence. The supergaussian part seems to cover values greater than v_crit = 2.6 * average_rectified_EMG (the average calculated as { |EMG_1| + .... + |EMG_n| }/n ). The factor 2.6 is explained below. If the pdf were Gaussian then about 3.8 % of the points should be greater than v_crit; if the pdf is exponential then the percentage would be 7.5%. Thus, one may suggest a simple (instead of the more complicated kurtosis-parameters) statistical exceedance parameter TAIL defined as the percentage of points greater than v_crit. If TAIL is close to, or bigger than (3.8% + 7.5%)/2 := 5.7% then we might say that have a supergaussian candidate. One may also encounter subgaussian pdf:s with TAIL < 3.8%. Computing the TAIL-parameter is a simple way to screen data for *deviant* distributions. The bottom line is to find out whether the (frequency of) extreme high amplitude potentials are not just some artifacts but have a significant physiological correlate; that is, if the tail may provide some information about the generating process which has not yet been considered. The common practice is to subject EMG to heavy massaging which usually serve the intended purpose well. From nonlinear time series analysis [4] it is however known that processing data (filters etc) may destroy some important intrinsic features of the data. This is as yet just a speculation with regards to sEMG. Indeed, from the responses I got, and through the further surveys I have made myself, it seems that this avenue has not been explored much.