Resolving Emg Pulse Superpositions via Utility Maximization

A major task in the decomposition of multi-channel Electromyographic (EMG) signals into their uni-source pulse train (UPT) components is to resolve pulse superpositions. Even optimal segmental analysis that fits linear combinations of temporally aligned UPT pulses to short data segments gives rise to errors because of pulse evolution and the presence of non-stationary noise. To address such shortcomings, we use the results of segmental analysis to estimate the probability of occurrence of each UPT in every segment. These probabilities, in conjunction with validity constraints on UPT inter-pulse intervals, are then used in a utility maximization process to revise the initial hypotheses. Inclusion of such suprasegmental analysis in a second generation EMG decomposition system has increased the system's accuracy from under 75% to well over 95% on real EMG data.