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

Abstract

Sonic crystal noise barriers (SCNB) have emerged as a promising solution for mitigating traffic noise pollution. These barriers utilize periodic structures to selectively reflect acoustic waves at specific target frequencies, offering the advantage of being permeable to light and wind. However, their installation and maintenance costs have hindered widespread adoption. In contrast, active noise control (ANC) systems leverage speakers and microphones to generate opposing sound waves that cancel out incoming noise, presenting a potentially cost-effective alternative. The efficacy of ANC, however, hinges on the precision of noise prediction models and control algorithms. Reinforcement Learning (RL) technique, an interdisciplinary area of machine learning, has shown promise in enhancing ANC systems by enabling them to adapt to changing noise conditions and achieve superior noise reduction, particularly in enclosed spaces. Despite these advancements, several challenges remain in applying RL to ANC systems for SCNB. This paper explores these challenges and proposes an RL-based solution for autonomous ANC systems within the context of SCNB, utilizing a Finite Difference Time Domain (FDTD) simulation environment to address low-frequency, moving sources, and outdoor propagation noise scenarios.

Keywords

• Reinforcement learning
• noise barriers
• finite-difference time-domain
• sonic crystals