Adaptive Exponential Trigonometric Functional Link Neural Network-based Filter Proportionate Normalized Least-Mean-Square Algorithm

The presence of passive components, noise, and distortions can render systems nonlinear. To effectively address this nonlinearity, the Adaptive Exponential Trigonometric Functional Link Neural Network (AETFLN) was developed. Additionally, many nonlinear physical systems exhibit sparse impulse responses. The recently proposed Filtered Proportionate Normalized Least Mean Square (FPNLMS) algorithm has demonstrated superior performance in sparse system identification under Gaussian noise by adjusting its weight coefficients proportionally to the system characteristics. This paper proposes an AETFLN-based FPNLMS algorithm that simultaneously accounts for system nonlinearity and sparsity to enhance performance under non-Gaussian noise conditions. The effectiveness of the proposed method is demonstrated through two nonlinear system applications: acoustic echo cancellation (AEC) and adaptive system identification.