Application of reinforcement learning for active noise control

Abstract

Active noise control (ANC) systems are used to reduce the sound noise level by generating antinoise signals.M-Estimators are widely employed in ANC systems for updating the adaptive FIR filter taps used as the system controller.Observing the state-of-the-art M-estimators design shows that there is a need for further improvements. In this paper, a feedback ANC based on the reinforcement learning (RL) method is proposed. The sensitivity of the constant parameter in the RL method is checked. The effectiveness of the proposed method is proven by comparing the results with previous feedforward studies through computer simulations.

Improving Insertion Loss of Sonic Crystal Active Noise Barrier by Reinforcement Learning and Finite Difference Time Domain Simulations

“…There is a wide scope of applications that have demonstrated its efficiency optimizing its processes such as Blockchain [21] or additive manufacturing processes [22]. Deep Q-learning and actor-critic methods, two prominent branches of RL, have been applied successfully to ANC systems [23], enabling them to adapt dynamically to changing noise conditions and improve noise reduction performance, especially in enclosed spaces or noise signals without simulations [24]. In Table I, some of the most representative works concerning the principal components of this study are presented.…”