A neural network method for adaptive noise cancellation

In this paper, we present an adaptive FIR filter for noise cancellation whose coefficients are adjusted by an analog neural network instead of numerical adaptive algorithms. Due to its real time processing capabilities, the neural network can optimize the coefficients of the adaptive FIR filter at each new received sample, which is especially useful in non-stationary environments. Due to the parallel and analog nature of the processing, the time needed by the neural network for computation of those coefficients is short. Compared to the traditional LMS and RLS adaptive algorithms, the proposed adaptive method is characterized by inherent stability and fast convergence, while eliminating the need to choose learning rates. Simulation results are given which demonstrate satisfactory performance