Investigation of probability density functions in modeling sample distribution of surface electromyographic (sEMG) signals

The surface electromyography signal (sEMG) has been typically modeled as a Gaussian random process. However, some authors have reported that the probability density functions (pdfs) associated with the sample distribution of sEMG signal exhibits a more peaked shape than one could expected for a Gaussian pdf. This work aimed to reinvestigate the profile of the sEMG pdfs during five different load levels of isometric contractions of biceps brachii muscle, and compared the adequacy of four different pdfs (Gaussian, Logistic, Cauchy, and Laplacian) in describing the sample distribution of such signal. Experimental pdfs were estimated for each subject and load condition. The comparison between experimental pdfs obtained from sEMG data of forty volunteers and four theoretical pdfs was performed by fitting these functions to its experimental counterpart, and using a mean absolute errors in the assessment of the best fit. On average, the Logistic pdf seemed to be the best one to describe the sample distribution of sEMG signal, although the probabilistic results, considering binomial trials, were significant for both Gaussian and Logistic pdfs