A low-complexity AFF-RLS algorithm using a normalization technique

In this paper, we propose a modified RLS algorithm with Adaptive Forgetting Factor using a low-complexity forgetting factor update equation. The conventional AFF-RLS method has a high-complexity update equation to update forgetting factor. In order to reduce complexity, an approximated version of the AFF-RLS method had been derived by Song. But this modified AFF-RLS method shows degraded performance because it suffers from 'gradient error amplification' problem. In order to obtain the same performance as AFF-RLS with relatively low computational cost, we noted that AFF-RLS had been derived by 'method of steepest descent', and we use normalization technique which is used in NLMS. The Experiment result shows that the proposed method has almost same performance as the conventional AFF-RLS method with relatively low-complexity.