Particle Swarm Intelligence Based Univariate Parameter Tuning of Recursive Least Square Algorithm for Optimal Heart Sound Signal Filtering

Abstract

Highlights•For diagnosis of heart related diseases, the lung sound is considered as noisy component. •Suppression of lung sound signal from heart sound signal using Recursive least square algorithm.•The forgetting factor of the RLS algorithm is tuned using DNL-PSO algorithm.•The performance of RLS algorithm is compared with Least Mean Square (LMS) adaptive algorithm. Article Info AbstractHeart Sound Signal (HSS) is considered as one of the important bio-signals. It carries vital information about the heart functions. For bio-acoustic observations, the HSS is diagnosed and recorded with auscultatory procedures. During auscultation, the noisy components gets added along with the reading. The physician's individual diagnostic experience, ecological noise and the intersection of heart and lung sound signal (LSS) are considered as the major noisy components in HSS diagnosis. Suppression of LSS from the HSS is a challenging task. Due to its quasi stationary nature, adaptive filtering techniques are used for the noise removal. In this paper, Recursive Least Square (RLS) adaptive algorithm is proposed to obtain the HSS from the noisy mixture. Faster convergence is a benefit in selecting RLS algorithm over other adaptive algorithms. The forgetting factor is one of the important parameters of RLS which defines the convergence. The RLS performance is improved by choosing an optimal forgetting factor. A Particle Swarm Optimization (PSO) based search algorithms are deployed for optimization. To enhance the implementation time, a Dynamic Neighbourhood Learning Particle Swarm Optimizer (DNL-PSO) is analysed. In DNL-PSO, each particle studies from its knowledge in dynamically varying neighbourhood that prevents early convergence. The normal HSS with different LSS interference is taken to assess the RLS filter performance. In this paper, the RLS algorithm performance is compared with Least Mean Square (LMS) adaptive algorithms. Various metrics are used to compare the performance of both RLS and optimization algorithms.