Enhancing the Power Efficiency of the NLMS Adaptive IIR Filter with Fast Binary Reciprocal Approximation

Division is the primary computational bottleneck in the normalized least-mean-square (NLMS) algorithm. This paper introduces a novel method for binary reciprocal approximation. While the approximation introduces some errors, its effect on the robust NLMS adaptive filter's performance is negligible. To substantiate this claim, the study investigates the performance of the NLMS adaptive IIR filter employing the proposed method. The key advantages of using reciprocal approximation in place of traditional division in the NLMS adaptive IIR filter include reduced power consumption and fewer clock cycles per data sample. Compared to the Newton-Raphson approximation, the look-up table method, and the previous approach based on the canonical signed digit (CSD) representation, the proposed technique demonstrates superior efficiency in terms of both power usage and hardware resource requirements.