Simplified stochastic gradient adaptive filters using partial updating

In some adaptive filtering applications, the least-mean-square (LMS) algorithm may be too computationally- and memory-intensive to implement. The authors present two adaptive algorithms that update only a portion of the coefficients of the adaptive system on average. These algorithms use a decimated version of the regressor vector signal and thus are particularly suited to filtered-regressor algorithms used in infinite-impulse-response (IIR) filtering and active noise control applications. The authors provide statistical analyses and simulations of these algorithms that indicate that their behavior with stationary random input signals is similar to that of a periodic update version of the LMS adaptive algorithm. The robustness of the proposed algorithms for periodic inputs is also discussed