Active Road Noise Control Based on Data-Driven Predictions of Passenger Ear Noise Signal

This work proposes an active road noise control (RNC) system that effectively controls non-stationary noise inside the cabin by using a data-driven model to predict primary noise in passenger's ears from sensors on the car chassis, with a fixed filter design integrated into headrest speakers. We develop STFNet, a novel network that fully exploits spatial, temporal, and spectral information between reference noise recorded by the multiple accelerometers and primary noise present in the passenger's ears. Extensive testing with real-world recordings at various driving speeds demonstrates that our system not only achieves higher noise reduction, but also demonstrates significantly faster convergence performance compared with traditional multi-channel RNC system.