A spike-based adaptive filter

Abstract

We propose a spike-based adaptive filter with supervised learning. Unlike standard adaptive filters, here the optimal MSE solution is not unique for the spike-based system identification problem. The simplex method is introduced to select one of the many possible optimal solutions. In simulations, an LMS-based learning procedure is designed and, for faster convergence, we introduce a credit assignment method which penalizes all the weights contributing to the current error signal. Finally, we discuss issues regarding the implementation of the spike-based adaptive filter in an analog VLSI circuit.