Active Noise Control 

Meta-Learning Initialization of Adaptive Filter

Active noise control (ANC) is widely used to mitigate urban noise, but many advanced adaptive algorithms are not implemented in actual products due to their high computational complexity and slow convergence. With the rise of deep learning, Meta-learning-based initialization offers a promising solution for accelerating adaptive algorithms. We proposed a modified Model-Agnostic Meta-Learning (MAML) initialization for multichannel active noise control (MCANC), and developed the Monte-Carlo Gradient Meta-learning (MCGM) algorithm, which simplifies the process while achieving comparable performance. Numerical simulations with raw noise samples validate the effectiveness of these methods in speeding up the convergence of the multichannel-filtered reference least mean square algorithm (McFxLMS).

Generative Fixed-Filter Active Noise Control with CNN-Kalman/Bayesian Filtering

Selective Fixed-filter Active Noise Control (SFANC) is limited by choosing a single candidate from pre-trained filters, while Generative Fixed-filter Active Noise Control (GFANC) improves this by adaptively combining sub-control filters but can be inaccurate with dynamic noises. We propose a GFANC-Kalman/Bayesian approach, integrating a 1D CNN and a Kalman/Bayesian filter, to enhance GFANC by predicting and refining the weight vector for sub-control filter combination. This method improves adaptability to dynamic noises, as validated by numerical simulations demonstrating its efficacy in real-world scenarios.

Selective fixed-filter ANC based on CNN

Active noise control (ANC) technology is common in wearable audio devices due to its low computational complexity, robustness, and performance with dynamic noise. The traditional fixed-filter strategy, while central to real-time ANC, cannot optimize noise reduction for various noise types. We propose a selective fixed-filter ANC method using a simplified 2D CNN on a co-processor and a lightweight 1D CNN for real-time noise type classification, showing superior performance in reducing real-world non-stationary noise compared to conventional adaptive algorithms.

Transferable Latent of CNN-based Selective Fixed-filter Active Noise Control

Practical active noise control (ANC) systems, such as noise-canceling headphones, typically use control filters with preset coefficients to effectively reduce dynamic noise. The selection of the appropriate control filter is crucial and usually determined by trial and error. This paper introduces an efficient one-dimensional convolutional neural network that selects the best pre-trained control filter for various noise types, using a similarity matching method for better generalization and incorporating L-softmax to enhance performance, validated through numerical simulations.

Ten questions concerning active noise control in the built environment

Urban noise pollution is a significant public health threat that requires urgent attention. While passive noise control measures are less effective for low-frequency noise and can be bulky, active noise control (ANC) has shown promise in various applications, including automobiles, construction machines, and noise barriers, but remains rare in the built environment. Recent advancements in digital signal processing, artificial intelligence, and embedded electronics, along with the demand for naturally ventilated buildings, are driving renewed interest and research in implementing ANC solutions for environmental noise mitigation.

Adaptive active noise control algorithms overcoming the output saturation effect

The saturation effect introduces nonlinear elements into the adaptive algorithm, impacting the ANC system's performance and stability due to excessive output power of the control signal. Two categories of adaptive algorithms have been developed to address this issue: one constrains the output signal to manage saturation, and the other employs nonlinear ANC algorithms to model inherent signal nonlinearity. Output constraint algorithms are more computationally efficient and robust, making them a more practical choice for handling output saturation in ANC systems.

MOV-Modified-FxLMS Algorithm With Variable

Penalty Factor in a Practical Power Output

Constrained Active Control System

Practical Active Noise Control (ANC) systems need to limit their maximum output power to prevent system instability. The MOV-FxLMS algorithm, which offers optimal control under output constraints, requires offline knowledge of disturbance power and is susceptible to variations in disturbance power. This research introduces a variable penalty factor in the Modified-FxLMS (MFxLMS) algorithm, enabling real-time adaptation to disturbance changes and ensuring the system consistently meets maximum power output constraints, unlike the fixed penalty factor.

A Frequency-domain Output-constrained Active

Noise Control Algorithm Based on An Intuitive

Circulant Convolutional Penalty Factor

Least mean square (LMS)–based algorithms are commonly used in active noise control (ANC) applications due to their computational efficiency, but implementing advanced ANC functionalities like selective frequency cancellation is hindered by complexity issues. Time-domain adaptive algorithms, meant to simplify the process, still face increased complexities. This paper introduces a circulant convolutional penalty factor to assist the extended leaky filtered-reference LMS (FxLMS) algorithm in achieving frequency constraints without frequency-domain transformations, and uses the coordinate descent method to reduce computations, making real-time implementation feasible and effective as demonstrated by numerical simulations.

Abating Urban Noise Through a Holistic approach of noise monitoring, analytic and control (MAC)

The main research objective of this project is to combine different enabling technologies developed by NTU to continuously monitor, analyze, and perform active noise control on aircraft, rail, and road traffic noises affecting the liveable residential environment. With rapid urban development, the annoyance and dissatisfaction caused by excessive urban noise is familiar to almost all Singaporeans; the negative health impact is less well known but is nonetheless very significant. 

Active Contol of Low-Frequency Noise through a Single Top-Hung Window in a Full-sized Room

The push for urban sustainability has intensified the search for noise mitigation solutions that allow natural ventilation into buildings. To improve low-frequency attenuation in active noise control (ANC) systems for open windows, four passive radiator-based speakers were added around a top-hung ventilation window. Testing in a mock-up apartment showed that active control achieved 8-12 dB of attenuation directly in front of the window and up to 10.5 dB across the room, performing similarly to or better than passive insulation for various noise types, including low-frequency compressor noise and jet aircraft flyby.

Wireless reference microphone enhancing noise reduction performance in multi-channel active noise control windows

Multi-channel active noise control (ANC) effectively cancels noise over a large area but is limited by the quality of the reference signal. Using wireless reference microphones in multi-channel ANC improves reference signal quality and noise reduction performance. Theoretical and experimental results show that the multi-channel adjoint least mean square (LMS) algorithm benefits from higher reference signal-to-interference ratio (RSIR) provided by wireless microphones, enhancing steady-state performance.

Recent advances on active noise control: open issues and innovative applications

The problem of acoustic noise is becoming increasingly serious with the growing use of industrial and medical equipment, appliances, and consumer electronics. Active noise control (ANC), based on the principle of superposition, was developed in the early 20th century to help reduce noise. However, ANC is still not widely used owing to the effectiveness of control algorithms, and to the physical and economical constraints of practical applications. In this paper, we briefly introduce some fundamental ANC algorithms and theoretical analyses, and focus on recent advances on signal processing algorithms, implementation techniques, challenges for innovative applications, and open issues for further research and development of ANC systems.

Active Noise Control System for Headphone Applications

This early research work presented the design and implementation of an adaptive feedback active noise control (ANC) system for headphone applications. The optimal position of the error microphone in the ear-cup was determined through experimental studies, and music signals were utilized for adaptive system identification of the secondary path. The ANC headphone was implemented using a digital signal processor for real-time experiments, and its performance was evaluated against a high-end commercial ANC headphone using the same set of primary noises, including real-world engine noises. Experimental results demonstrate that the proposed ANC headphone achieves superior noise cancellation, particularly for low-frequency harmonics.

Improving Convergence of the NLMS Algorithm Using Constrained Subband Updates

We propose a new design criterion for subband adaptive filters (SAFs) based on the principle of minimal disturbance with multiple constraints on updated subband filter outputs. This criterion leads to a subband adaptive filtering algorithm that converges faster than the classical fullband least-mean-square (LMS) algorithm under colored excitation and has a simple recursive tap-weight adaptation comparable to the normalized LMS (NLMS) algorithm. The effectiveness of the proposed criterion and algorithm is validated through mathematical analysis and simulation.