Digital signal processing algorithms for the detection of afferent nerve activity recorded from cuff electrodes.

Due to the very poor signal-to-noise ratios (SNR's) usually encountered with whole nerve-cuff signals, the processing method typically applied, rectification and windowed (bin)-integration (RBI), can have serious shortcomings in extracting reliable information. In order to improve detection accuracy, these signals were further analyzed using statistical signal detection algorithms based on their second and higher order spectra (HOS). A comparison with both analog and digital RBI processing suggests that the statistical methods, due to their ability to separate the signal and noise subspaces, are superior. It was determined that the noise typically encountered with nerve-cuff electrode signals is normally (Gaussian) distributed. Therefore, third-order statistics can be applied to, ideally, completely reject the noise component. When cutaneous nerve recordings from the calcaneal nerve (innervating the heel area) were used in a drop-foot correction neural prosthesis, the detection percentage and the insensitivity to algorithm parameters were increased through the use of these statistical methods as to warrant their real-time implementation, and the inherent additional processing hardware that entails.