Adaptive Channel Equalization Using Tuning-Free Non-Parametric NLMS with Variable Step Size

This paper presents a novel, nonparametric, normalized least mean squares (NLMS) algorithm for applications in adaptive channel equalization to mitigate the effects of inter symbol interference (ISI). A novel, tuning-free, easy-to-implement, variable step-size scheme is proposed, which works by adding the square roots of the magnitudes of the input vector. We denote this new algorithm using the abbreviation TFVSS-NLMS, which stands for Tuning-Free Variable Step Size NLMS algorithm. The proposed algorithm has been benchmarked extensively against existing least mean squares (LMS) and NLMS algorithms. The bit error rate (BER) performance of TFVSS-NLMS significantly outperformed the conventional fixed step-size LMS and the classical NLMS algorithms at typical levels of signal-to-noise ratio (SNR). In comparison to more sophisticated variable step-size (VSS-NLMS) algorithms, TFVSS-NLMS has registered a significantly better BER performance. Moreover, the proposed TFVSS-NLMS offered comparable or faster convergence speeds than both the classical and variable step-size NLMS algorithms. As opposed to many existing VSS-NLMS algorithms that require tuning some parameters to achieve satisfactory performance at each noise level, the proposed TFVSS-NLMS is a tuning-free approach characterized by a better BER performance and a faster convergence rate than current algorithms regardless of the level of channel noise.