Signal Processing and Deep Learning for Practical Active Noise Control 

Abstract:

Digital Active Noise Control (ANC) systems have surged in popularity, particularly with advancements in digital electronics like embedded processors and ADC/DAC converters. These systems offer a high level of tunability and adaptability, making them highly sought-after for a variety of applications ranging from ANC headphones to road noise cancellation in automobiles. The design and development process has become more accessible, extending their reach across a multitude of functional areas in both academia and industry. This progress has spurred the creation of complex, multi-channel, distributed, and efficient control algorithms for ANC systems, leading to heightened performance and enhanced noise management capabilities across diverse applications. Moreover, their straightforward implementation and cost-effectiveness make them an attractive solution for addressing noise pollution challenges in multiple sectors[1].

In this presentation, I will delve into the practicalities of engineering ANC systems for both headphones and windows, showcasing some prototype systems from the Digital Signal Processing Lab at Nanyang Technological University, Singapore. I will illustrate how machine learning and digital signal processing techniques have been adeptly harnessed to identify noise patterns and masterfully mitigate noise. This discussion will include insights into the design of ANC systems based on theoretical studies that ensure stability and robustness, enabling the deployment of autonomous ANC systems in real-world settings.

Furthermore, I will introduce an innovative approach to ANC design that departs from the traditional method of employing deep neural networks directly in the secondary path for anti-noise generation—a method often too complex and impractical for real-world application. Our approach centers on a delayless method in selective[2] or generative[3] control filters that can effectively manage in-situ noise conditions. I will elucidate this novel DNN-driven methodology and provide perspectives on how this rapid-response strategy can address dynamic noise variations and the associated changes in noise paths and characteristics.

[1] B Lam, WS Gan, DY Shi, M Nishimura, S Elliott, “Ten questions concerning active noise control in the built environment,” Elsevier Building and Environment Vol 200, Aug 2021, 107928, https://doi.org/10.1016/j.buildenv.2021.1079281] 

[2] ZD Luo, DY Shi, WS Gan, “A Hybrid SFANC-FxNLMS Algorithm for Active Noise Control based on Deep learning,” IEEE Signal Processing Letters, Vol 29, 2022, pp1102-1106.

[3] ZD Luo, DY Shi, WS Gan, "Delayless Generative Fixed-filter Active Noise Control based on Deep Learning and Bayesian Filter," IEEE Transaction on Audio, Speech, and Language Processing (2023)