Floating-point adaptive filter architectures for the cancelling of harmonics power line interference

This work presents floating-point dedicated architectures of LMS (Least Mean Square) and NLMS (Normalized Least Mean Square) adaptive filtering algorithms for the harmonics power line interference cancelling. Among the various algorithms of adaptation, the (LMS is surely the most popular, mainly due to its low computational cost, robustness and tracking ability. On the other hand, the normalized version of the LMS algorithm, named NLMS presents as main characteristics the increase of speed of adaptation and small sensitivity to the power reference signal. This work explores these tradeoffs by developing floating-point dedicated architectures for both algorithms, and applying them to the harmonics cancelling structure. The developed architectures were described in hardware description language, and synthesized to 45 nm Nangate Open Cell using Cadence RTL Compiler. The main results show that the interference canceller with NLMS filters presents faster convergence at the cost of an increased area and power consumption as compared with the structure with LMS filters.