Stochastic analysis of myoelectric temporal signatures for multifunctional single-site activation of prostheses and orthoses

This paper is concerned with a stochastic time-series analysis of the temporal signatures of myoelectric (ME) signals including the determination of model order and sampling rate. The paper considers the use of time-series parameters for the activation of artificial limbs for high-level amputees, of stimulation electrodes or of powered braces for paralysed persons, in several degrees of freedom, from a single or two surface-electrode pairs at locations where considerable ME cross-talk exists. The multifunctional capability from a single site is based on the differences between the time-series (TS) parameters for different muscle activation patterns at the same ME site, these differences being thus used for limb function discrimination via easily trainable muscle activation patterns at the vicinity of the electrode site. Specifically, the analysis is in terms of identifying the AR parameters of a time-domain autoregressive (AR) signature model both for the complete ME spectrum and for parts thereof, and in terms of the autocorrelation of the signal and of the models residual. Determination of sampling rate and of model orders is discussed in detail. It is shown that, using online real-time analysis, differences in the AR time-series parameters can be observed for different trainable patterns of muscle activation, at the same electrode location, even at the same ME power levels, as long as considerable cross-talk exists at the electrode site. These parameter differences can be accentuated if one considers the AR parameters for lower-frequency spectral windows. A case is made in this paper for employing TS analysis to squeeze out information in a distinct but low-level ripple of the low frequency spectrum of the signal. This information tends to be ignored in frequency domain, but is all that the AR parameters care for in TS analysis, since they are not concerned, with a flat-average low-frequency spectrum, i.e., its white-noise-like part, which is the residual term of the AR Model and not an AR parameter. Discrimination between different functions from a single electrode-site, at even the same power level, is thus shown to require considerable cross-talk at the given site, and to require the consideration of only the low-frequency part of the spectrum.