SVD and higher-order statistical processing of human nerve signals

Abstract

Human, afferent, whole nerve signals recorded using an implanted nerve-cuff electrode were analyzed using two algorithms based on the statistical properties of the signals. The processing method typically described in the literature (RectiJication and BinIntegration -RBI) has serious shortcomings in processing these signals, which have very poor signal-to-noise ratios. Algorithms based on a Singular Value Decomposition @4l) of the signal's 2nd and Higher-Order Statistics @OS) have resulted in more robust signal detection. Reliable detection of afferent nerve signals is essential i f such signals are to be of use in artificial sensoy-based Functional Electn'cal Stimulation neuralprosthetics.