CAN ACTIVE NOISE CONTROL BENEFIT FROM DEEP REINFORCEMENT LEARNING?

Can Active Noise Control algorithms benefit from Deep Reinforcement Learning?

To answer this question we used DRL to supervise the performance and dynamically adjust a parameter of an Engine Order Control algorithm. The DRL-supervised ANC algorithm is trained on acoustic data recorded in a car, and augmented using granular synthesis. When comparing the performance of this approach with a reference system using a manually tuned fixed parameter, we found that the noise reduction is improved and that the DRL learns to adjust the controlled parameter to different driving conditions. In the future, this type of approach may be used not only to improve the performance of ANC systems and the robustness to vehicle variations, but also to reduce the need for initial tuning of the system.

Read the full article here:
https://www.researchgate.net/publication/349104882_Deep_reinforcement_learning_applied_to_active_noise_control_a_smart_adaptive_variable_step_size_FxLMS_algorithm_design