Incremental Learning Based Adaptive Filter for Nonlinear Distributed Active Noise Control System

Ruchi Kukde,M Sabarimalai Manikandan,Ganapati Panda

doi:10.1109/ojsp.2020.2975768

Abstract

Active control of noise for multi-channel applications is affected by the existence of nonlinear primary and secondary paths. There is a degradation in the performance of linear multi-channel active noise control (LMANC) systems based on minimization of sum of squared errors obtained from multiple sensors in presence of nonlinear primary path (NPP) and nonlinear secondary path (NSP) conditions. The NPP and NSP problems are more prominent and challenging for multi-point noise control applications owing to different locations of silent zones and acoustic coupling between secondary sources and error sensors. In order to surmount this problem, an incremental strategy based nonlinear distributed ANC (NDANC) system is developed in this article. The adaptive exponential functional link network (AE-FLN) is employed as an adaptive control unit at the acoustic sensor nodes (ASNs) for the design of NDANC system. The incremental co-operation scheme is utilized to provide uniform noise cancellation in presence of NPP and NSP conditions. Simulation study is conducted extensively to demonstrate the efficiency of the proposed system for different practical NPP and NSP scenarios. The detailed computational load analysis and subjective evaluation of reduction in perceptual noise levels are performed for different real noise conditions.

Highlights

Active noise control (ANC) systems have gained immense importance to reduce undesirable noise from automobiles, transformers, consumer appliances, medical equipment, infant incubators and other outdoor sources [1]
With an objective to address the dual challenges of nonlinearities and need of uniform noise cancellation, this paper focuses on maneuvering incremental learning approach for distributed ANC (DANC) application
As the paper aims at mitigating noise in presence of nonlinear environments, particular focus of the simulation studies is on nonlinear primary path (NPP) and nonlinear secondary path (NSP) situations

Summary

INTRODUCTION

Active noise control (ANC) systems have gained immense importance to reduce undesirable noise from automobiles, transformers, consumer appliances, medical equipment, infant incubators and other outdoor sources [1]. The FLANN based controllers are employed to develop computationally efficient NANC systems using filtered-s least mean square (FsLMS) algorithm [20], [21] This approach for multi-channel nonlinear noise cancellation is reported in [22]. These MANC systems require re-structuring of update rules for every addition or deletion of loudspeaker and error sensor To resolve this problem, a decentralized distributed processing approach to design multi-point ANC schemes is developed in this paper. The distinct aspects that the paper focrusDesevoenloarpem: ent of nonlinear DANC scheme with adaptive exponential functional link network (AE-FLN) controller structure at individual ASNs and communication r using incremental learning across the ASNs. Performance evaluation of the proposed scheme in terms of noise cancellation and computational complexity as compared to existing centralized and distributed ANC r methods.

FUNCTIONAL EXPANSION BASED MULTI-CHANNEL

DERIVATION OF WEIGHT UPDATE RULES

DERIVATION OF BOUND ON CONVERGENCE PARAMETERS

RESULTS AND DISCUSSION

PERFORMANCE EVALUATION FOR NPP AND NSP ENVIRONMENTS

CONCLUSION