Optimal extraction of EMG signal envelope and artifacts removal assuming colored measurement noise

Abstract

This paper develops the Kalman filter (KF) and unbiased finite impulse response (UFIR) filter to extract the electromyography (EMG) signal envelope and remove some artifacts with a maximum accuracy assuming colored measurement noise (CMN). The filters are developed in discrete-time state-space for Gauss-Markov CMN and termed as cKF and cUFIR. For EMG signals combined with low-density motor unit action potentials (MUAPs), the envelope is preliminary shaped using the Hilbert transform. To well pronounced envelopes shaped with high MUAP densities and having sharp edges, the filters are applied directly. Experimental verification is provided for different EMG signal databases and it is suggested that the color factor must be set optimally to approach the desired envelope in the best way. Testing of the algorithms is provided by the rectangular, triangular, trapezoidal, and Gaussian pulses in the mean square sense. It is shown experimentally that the cKF and cUFIR filter are generally most efficient when the MUAP density is low.