It is of interest to characterize a dynamic function of muscle during movement, which is a non-stationary case, from surface EMG signals. This paper presents a new approach. This is based on a block algorithm in which a signal is divided in time into blocks and time-varying parameters are estimated in each block. A locally quasi-stationary processing, which is proposed here, is the method to estimate the parameters of an AR (autoregressive) model more precisely. It is assumed that an AR model represents a surface EMG generation system.
There have been many methods for the non-stationary analysis, for example, the synchronous averaging in stochastic approaches and the estimation methods of short-time power spectrum and system function by a locally stationary processing. However, the synchronous averaging method is restricted to the evoked responses of EEG or EMG and the locally stationary processing is not sufficient for essentially non-stationary signals.
In this paper, AR parameters were estimated from the surface EMG of masseter muscle by the locally quasi-stationary processing and the results were compared with the parameters by the conventional locally stationary processing. There are non-stationary intervals around evoked response or onset of masticatory EMG. AR parameters are linear prediction coefficients, reflection coefficients and poles.
As the results, the estimated characteristics of the time-varying parameters were reasonable in relation to the locus of lower jaw movement and the significant differences were showed in the non-stationary intervals.
抄録全体を表示