Abstract—This work presents the classify of four channel of myoelectric signal employing higher order statistics, principal components analysis and Cero Crossing. The results let us to obtain six distinctive activities to control a parallel mechanism for a prosthetic transhumeral elbow. The results are commands of movements for the prosthesis in order to emulate a biological elbow. This work use 2 of 3 feature extraction methods, the best method is higher order statistics because is easy to implement and have fast processing and can be used in FPGA or DSP. I. INTRODUCTION The electromyographic signal observer at the skin surface (sMES) is the sum of many small potential generated in the muscles fibers [1]. EMG signal are non-stationary and have highly complex time-frequency characteristics. Consequently, these signals cannot be analyzed using classical methods such as Fourier Transform. Although the short time Fourier Transform can be used to satisfy the stationary condition for such no stationary signal, it suffers from the fact that the performance depends on choosing an appropriate length of the desired segment of the signal. To solve such problem, High Order Statistics was used as a feature extraction method and has been widely used in signal analysis [2]. The process to obtain the signal myoelectric is showing in Figure 1 In which the signal is acquired from patient through skin sensor and then proceeds to the amplification, processing and classification of the signals. This process is the start point to obtain myolecric signal in order to control the prosthesis of 3DOF. Fig. 1. SMES Classification Bloc Diagram. II. FEATURE EXTRACTION Historically, in stochastic process has been used a second order analysis. Nevertheless, the second order statistics is deficient in the analysis of certain type signals. Therefore, in some applications it is necessary to extend the analysis to statistics of high order. A Gaussian stochastic process is totally defined by first and second order statistics; and therefore it is possible to be analyzed using only the autocorrelation and the power spectral density (PSD), nevertheless, the signals in the real world are of non-Gaussian nature like the biomedical signals, voice, [3], [15], [16], [17] among others. A. High Order Statistics The moments provide information of the pdf, nevertheless, the mathematical treatment to calculate moments is quite complicated. A statistics set exists that has mathematical characteristics that can be been useful in the signal analysis [4]. This set of statistics is the cumulants. The cumulants and the statistics moments are widely related. The cumulants of a certain process can be calculated by the relations between moments and statistics of the process [5]. Let x(t) a process with mean 0, the k-order cumulant, C kx can be calculated using the relation between cumulants and moments. { } ) ( ) ( 1 , 1 t x E C x = τ (1) { } ) ( ) ( ) ( 1 1 , 2 τ τ + = t x t x E C x (2) { } ) ( ) ( ) ( ) , ( 2 1 2 1 , 3 τ τ τ τ + + = t x t x t x E C x (3) { } ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) , , ( 2 1 , 2 3 , 2 1 3 , 2 2 , 2 3 2 , 2 1 , 2 3 2 1 3 2 1 , 4 τ τ τ τ τ τ τ τ τ τ τ τ τ τ τ − − − − − − + + + = x x x x x x x C C C C C C t x t x t x t x E C (4) Where E{x(t)} is the expected value for each x (t) ( x(t) is the sMES sample), and τ are the times of each sample. Is possible consider the values of τ 1 and τ 2 constants to calculate the third and quarter order cumulants [6].The equation 1 Myoelectric signal feature extraction based on Higher Order Statistics for Parallel Elbow Control Salvador A. Arroyo Díaz* National Institute for Astrophysics, Optics and Electronics, Electronics Department. Tonantzintla, Puebla, MEXICO sarroyo@inaoep.mx Alejandro Díaz Sánchez National Institute for Astrophysics, Optics and Electronics, Electronics Department. Puebla Institute of Technology, MEXICO adiazsan@inaoep.mx