Joint Deep Secondary Path Estimation and Adaptive Control for Active Noise Cancellation

Accurate secondary path estimation (SPE) is crucial for the stability and performance of active noise cancellation (ANC) systems, especially under time-varying acoustic conditions. Classical adaptive algorithms converge slowly and degrade in dynamic environments. This paper proposes an end-to-end deep learning framework that jointly performs secondary path estimation and adaptive control. A Deep Secondary Path Estimator (DeepSPE) leverages convolutional, recurrent, and attention layers to predict the secondary path in real time. The estimated path conditions an ANC-Net controller, which generates anti-noise signals through dynamic filter selection without iterative adaptation. Experiments on real and simulated impulse responses demonstrate that the proposed system achieves superior noise attenuation, reduced latency, and improved robustness compared to both traditional adaptive filters and recent deep ANC approaches.