Stochastic approximation with smoothing for optimization of an adaptive recursive filter

A major concern of adaptive IIR filter is that with the objective function being non-convex, currently used gradient methods have a tendency to converge to the local minimum. The stochastic approximation with convolution smoothing represents a simple approach for deriving a global optimization algorithm for adaptive filtering. This stochastic approximation method has been derived for adaptive system identification. Optimization is based on minimizing the mean square error objective function. The mean square error is a function of time series data that is statistically varying. An experimental result demonstrates the viability of using stochastic approximation for adaptive filtering.