A Novel and Efficient Surface Electromyography Decomposition Algorithm Using Local Spatial Information

Decoding Multi-DoF Movements Using a CST-Based Force Generation Model With Single-DoF Training

“…3) MUST-Based Method: All training data was decomposed to MUSTs by an improved CKC method [18]. MUs with pulseto-noise ratio (PNR) higher than 25 and the coefficient of variation (CoV) for the interspike interval less than 50% were discarded [30].…”

“…1) Feature Extraction: A CST estimation method was employed to extract the CST of MUs under each channel [28]. The firing spikes of activated MUs were allocated to the closest regions of channels through the SSD method.…”

“…T comprises consecutive samples of the DRs of all N MUs. Since identifying all recruited MUs is challenging, we employed the SSD method to directly estimate the DRs of the CSToCs [28]. We can then estimate the force function of MU pools on the tth sample as follows:…”

“…A MUST is the firing time sequence of a recruited motor unit (MU), representing the fundamental functional unit of neuromuscular system [15]. The MUSTs can be partially identified by blind source separation (BSS) algorithms [16]- [18] and classified into distinct MU pools by cross correlation, based on the motor neuron synergy perspective [19]- [21]. This perspective considers that the CNS modulates muscle contractions through common input which can be estimated through the discharge rate (DR) of the cumulative spike train (CST) of the recruited MUs in an MU pool [22]- [24].…”

“…The method directly estimate the CST of MUs under the coverage of a single electrode, deemed as the CST of channel (CSToC). This approximation can obtain more MU firing information than MU decomposition method with acceptable potential estimation errors [28]. Inspired by the NMF, we propose a nonnegative matrix optimization method to cluster the CSToCs into MU pools, differing from the typical clustering method of MUST-based approach.…”