Decomposition of Synthetic Surface Electromyograms Using Sequential Convolution Kernel Compensation

Abstract:- This paper reveals a sequential decomposition method based on Convolution Kernel Compensation (CKC). It operates in real time and can decompose linear mixtures of finite-length signals. Multiple-input multiple-output (MIMO) signal model is used with finite channel responses and trains of delta pulses as inputs. Our approach compensates the channel responses and reconstructs the input pulse trains in real time, when samples of the observed signals become available. Two versions of the sequential decomposition were tested: sample-by-sample and multiple sample input. The time complexity of the two approaches, is analytically evaluated and supported by the experimental measurements. Performance tests on the synthetic surface electromyograms (sEMG) provide estimations of average rate and standard deviation of the reconstructed pulse trains for single motor units (MU)