Unsupervised Bayesian EMG decomposition algorithm using tabu search

We introduce here the statistical model of the electromyographic (EMG) signal decomposition problem and then propose an algorithm with a Bayesian approach followed by its simulation data test results. We consider the classical problem of the multi-source impulse discharge separation from intramuscular EMG signals in the case where the impulse responses (motor unit action potentials, MUAPs) are supposed to be known to a certain degree. The main contribution of this work is the proposal of a fully unsupervised EMG decomposition algorithm that exploits both the signal model likelihood and the regularity of the motor unit discharge patterns in a Bayesian framework. The latter, though well-proven properties in the past, is essentially used as auxiliary information in an interactive procedure [1] involving human interventions. Another contribution consists of using the Tabu metaheuristics to solve the NP-hard problem over the complete search space of overlapped MUAP, unlike the existing methods that either performed on the restrained search spaces [2, 3] due to complexity or based on recursive algorithms [4, 1] with certain trial strategy and residual threshold estimations.