Wavelet packet analysis for angular data extraction from muscle afferent cuff electrode signals

Rehabilitation devices can greatly benefit from the use of natural sensors. Thus, we have extended on our efforts to extract angular information from muscle afferent nerves by means of cuff electrodes. Is this study we applied wavelet analysis to electroneurographic (ENG) data from rabbits. In order to estimate ankle flexion/extension angles, we recorded ENG signals from the left Tibial and Peroneal nerves, both during FES and under passive motion. Several processing methods were used for extraction of angular data and were compared with the wavelet analysis. An artificial neural network (ANN) was used with the analyzed features to improve on the accuracy of the angular predictions. The network has so far been tested for local generalization only. The ANN was found to work better with the wavelet features than with previously explored rectified and bin integrated (RBIN) signals. Best results were obtained by using ANN inputs that consisted of both the output from a single wavelet packet node and the RBIN signal: the mean angle prediction error was 1.2°. Exciting as this result is, we must keep in mind that due to the local generalization scope of this study, angle predictions have yet to be assessed regarding inter-rabbit variability.