Analysis and Smoothing of EMG Signal Envelope Using Kalman and UFIR Filtering under Colored Measurement Noise

Abstract

This article describes some filtering methods to remove artifacts from the EMG signal envelope. Diverse EMG waveforms are studied using the Kalman filter (KF) and unbiased finite impulse response (UFIR) filter. The filters are developed in discrete-time state-space for Gauss-Markov colored measurement noise (CMN) and termed as cKF and cUFIR. It is shown that a choice of a proper CMN factor allows extracting the EMG waveform envelope with a high robustness. Extensive investigation have shown that the cKF and cUFIR filter are most efficient when the density is low of the motor unit action potential (MUAP) of the EMG and the Hilbert transform is required. Otherwise, when the envelope is well-pronounced and well-shaped with sharp edges due to a high MUAP density, the filters can be applied directly without using the Hilbert transform.

Nonparametric Extrema Analysis in Time Series for Envelope Extraction, Peak Detection and Clustering

“…Many algorithms have been developed in literature to extract the EMG envelope, which primarily exploit the signal's moving average activity together with a noise reduction [39]. However, such methods may suffer large bias errors and be unable to avoid measurement spikes [40]. The Savitsky-Golay smoother combined with a low-pass filter was used to address such problems [41], where the smoothing improves envelope shaping but introduces time delays, which may be intolerable in certain applications (such as robotics).…”

“…In addition, Kalman and Wiener filter based algorithms have been developed for EMG by many works [44]- [48]. While the traditional Kalman setting has effective tracking when noise is white Gaussian; it provides no meaningful advantage against other methods, since EMG noise is strictly non-white [40]. Another approach by [49] completely forgoes the unbiased noise assumption and is considered more robust [50].…”