International Journal of Electrical and Computer Engineering (IJECE) Vol. 12, No. 1, February 2022, pp. 229~238 ISSN: 2088-8708, DOI: 10.11591/ijece.v12i1.pp229-238  229 Journal homepage: http://ijece.iaescore.com Effect of optimal filtering parameters for autoregressive model AR(p) on motor unit action potential signal Ayad Asaad Ibrahim, Mohammed Ehsan Safi, Eyad Ibrahim Abbas Department of Electrical Engineering, University of Technology Iraq, Baghdad, Iraq Article Info ABSTRACT Article history: Received Mar 22, 2021 Revised Jul 13, 2021 Accepted Jul 29, 2021 Error is one element of the autoregressive (AR) model, which is supposed to be white noise. Correspondingly assumption that white noise error is a normal distribution in electromyography (EMG) estimation is one of the common causes for error maximization. This paper presents the effect of a suitable choice of filtering function based on the non-invasive analysis properties of motor unit action potential signal, extracted from a non- invasive method-the high spatial resolution (HSR) electromyography (EMG), recorded during low-level isometric muscle contractions. The final prediction error procedure is used to find the number of parameters in the model. The error signal parameter, the simulated deviation from the actual signals, is suitably filtered to obtain optimally appropriate estimates of the parameters of the automatic regression model. It is filtered to acquire optimally appropriate estimates of the parameters of the automatic regression model. Then appropriate estimates of spectral power shapes are obtained with a high degree of efficiency compared with the robust method under investigation. Extensive experiment results for the proposed technique have shown that it provides a robust and reliable calculation of model parameters. Moreover, estimates of power spectral profiles were evaluated efficiently. Keywords: Autoregressive Final prediction error Power spectral Robust estimation Simulated EMG This is an open access article under the CC BY-SA license. Corresponding Author: Mohammed Ehsan Safi Department of Electrical Engineering, University of Technology Iraq Baghdad, Iraq Email: 30165@uotechnology.edu.iq 1. INTRODUCTION The surface electromyography (EMG) signal is increasingly used for tension state discrimination in a single muscle or a group of muscles and muscle fatigue measurements [1]. This noise-like signal is an interference pattern that is the temporal and the spatial sum of the action potential (AP) of all motor units (MU) in the region of the detecting electrodes [2]. This signal's statistical properties are related to muscle tension and spectral shape changes with muscle tension changes, but the changes are not significant [3]. The EMG signal is assumed to be a stationary zero-mean gaussian random process [4]. The muscles can change their contraction levels by changing the activated motor units [5]. These changes will cause some outlier samples to appear negatively, affecting the assumed distribution mentioned above [6]. Modeling the motor unit action potential (MUAP) has been particularly useful in developing a mathematical formulation to study, understand, and process techniques for this signal [7]. In this paper, a MUAP signals a model extracted from the IB2 spatial filter [8], based on the autoregressive model coefficients of this signal. Its parameters are derived from real recorded unipolar EMG signals. This model provides the means for generating MUAP signals whose amplitude and spectral characteristics depend on subjects and muscle type under normal contraction to be simulated. Such simulated signals can be used as test signals to evaluate the performance of