Reasearch On Off-line And On-line Decomposition Methods Of Surface EMG Signal

EMG signals contain important information such as the coding characteristics of human muscle nerve.The analysis of EMG signals can provide theoretical basis for the diagnosis of clinical diseases of neuromuscular system,and promote the development of clinical rehabilitation medicine and health care electronics.Surface electromyography(s EMG)signal is the s EMG signal on the surface of the skin,which can be decomposed to obtain the action potential and time series of the motor unit.These information can be used in the monitoring and diagnosis of neuromuscular diseases,the control of medical assisted prosthesis and other fields.However,most of the current decomposition methods of surface EMG signals are too inefficient and only suitable for off-line analysis scenarios.In order to solve this problem,a fast off-line decomposition method is proposed,and on this basis,two real-time on-line decomposition methods are further improved and proposed.Aiming at the inefficiency of existing off-line decomposition methods,an off-line decomposition method based on fast gradient convolution kernel compensation is proposed.The core idea is to improve the calculation process of cross-correlation vector:in the iterative calculation of cross-correlation vector,not only the gradient of the current time is used,but also the influence of the gradient of all the past time on the gradient of the current time is considered,so as to speed up the convergence speed and improve the decomposition efficiency.Simulation results show that,compared with the traditional methods,the proposed algorithm can significantly improve the decomposition speed while ensuring the accuracy.For example,when the signal-to-noise ratio of EMG signal is 20 d B,the decomposition speed can be increased by 5.2 times,and the accuracy is only reduced by 0.2%.For the application scenarios that need real-time response,based on the above offline decomposition methods,two online decomposition methods are proposed: the decomposition method based on sliding window and the decomposition method based on separation matrix.The core idea of the first on-line decomposition method is to add window to segment the EMG signal sequence: select a window with appropriate length to slide on the EMG signal sequence,and improve the calculation rules of cross-correlation matrix to ensure that the decomposition time of the signal in the window is less than the forward time of the sliding window,so as to realize on-line decomposition.Due to the limitation of hardware computing power,the window length of this method usually needs to be set to a relatively large value(such as 3 seconds),and once it is determined,it cannot be changed arbitrarily.In order to improve the decomposition flexibility,a second on-line decomposition method is proposed,which can make the length of each decomposition signal small enough(such as 200 ms),and change the length in real time during the decomposition process.The core idea is to divide the whole decomposition system into off-line training module and on-line decomposition module to achieve different functions respectively: the off-line training module uses K-means clustering algorithm to cluster the time series of motor units,so that the training outputs a separation matrix;the on-line decomposition module directly uses the separation matrix to decompose the EMG signal.Simulation results show that,compared with the fast off-line decomposition method,the proposed two on-line decomposition methods can reduce the number of motion units by 3 and the accuracy by 10% and 4% respectively when the signal-to-noise ratio is 20 d B.In the two online decomposition methods,the decomposition method based on separation matrix has higher accuracy,and the signal length of each decomposition can reach millisecond level,which is more in line with the requirements of practical application scenarios.