Low-Latency Active Noise Control Using Attentive Recurrent Network.

Area of Science:

• Acoustics and Signal Processing
• Artificial Intelligence and Machine Learning

Background:

• Processing latency is a critical challenge in active noise control (ANC) systems due to causality constraints.
• Deep learning approaches for ANC often introduce additional latency, exacerbating the problem.

Purpose of the Study:

• To develop and evaluate deep learning-based strategies for achieving low-latency active noise control.
• To mitigate the causality constraints in ANC systems by reducing algorithmic latency.

Main Methods:

• Employed a time-domain attentive recurrent network (ARN) for deep ANC with reduced frame sizes.
• Introduced delay-compensated training to enable ANC operation using predicted noise.
• Utilized a revised overlap-add method to prevent latency during signal resynthesis.

Main Results:

• Demonstrated the effectiveness of the proposed strategies in achieving low-latency deep ANC.
• Combined strategies resulted in zero or even negative algorithmic latency.
• Minimal impact on overall ANC performance was observed.

Conclusions:

• The proposed strategies significantly reduce algorithmic latency in deep ANC systems.
• Achieving near-zero or negative latency alleviates critical causality constraints in ANC design.
• This work paves the way for more efficient and effective real-time ANC applications.