GFANC-RL

This repository contains the code for the paper "GFANC-RL: Reinforcement Learning-based Generative Fixed-filter Active Noise Control," accepted by the Neural Networks journal. You can find the paper at Paper Link

This is a collaborative research work between the Digital Signal Processing Lab at NTU and the AI Lab at NUS. The RL algorithm used to train the 1D CNN in the GFANC-RL method is provided here.

The framework of the GFANC-RL method

           
Parameter update for the critic and actor in the RL algorithm.

Highlights

1. GFANC-RL employs RL techniques to address challenges associated with GFANC innovatively.
2. This paper formulates the GFANC problem as a Markov Decision Process (MDP) from a decision-making perspective, laying a theoretical foundation for using RL algorithms.
3. In the GFANC-RL method, an RL algorithm based on Soft Actor-Critic (SAC) is developed to train the CNN using unlabelled noise data and improve the exploration ability of the CNN model.
4. Experimental results show that the GFANC-RL method effectively attenuates real-recorded noises and exhibits good robustness and transferability in different acoustic paths.

Usage

1. Obtaining sub control filters

• The optimal control filter of broadband noise, whose frequency band contains our interested components, is first chosen as the full-band response. Subsequently, it is decomposed into M orthogonal sub bands as the desired sub control filters.

  

2. Training the CNN using RL

• A synthetic noise dataset is used to train the CNN, containing 80,000 noise instances for training. The noise instances are generated by filtering white noise through various bandpass filters with randomly chosen center frequencies and bandwidths. Each noise instance has a 1-second duration. The noise dataset is available at - Training dataset
• It is worth noting that no data labels are used in the training phase.

3. Real-time Noise Cancellation

• After training via the RL algorithm, the GFANC-RL method is used for real-time noise control.

Applying to New Environments

• Transferring the GFANC-RL method to new systems involves only updating the system-specific sub control filters, with the trained 1D CNN remaining unchanged, thus simplifying implementation across various scenarios.

Related Publications

1. Deep Generative Fixed-Filter Active Noise Control
   Conference: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2023)
   Paper Link: IEEE   Code Link: GitHub

2. Delayless Generative Fixed-filter Active Noise Control based on Deep Learning and Bayesian Filter
   Journal: IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2024, 32: 1048-1060.
   Paper Link: IEEE   Code Link: GitHub

3. GFANC-Kalman: Generative Fixed-Filter Active Noise Control with CNN-Kalman Filtering
   Journal: IEEE Signal Processing Letters, 2024, 31: 276-280.
   Paper Link: IEEE   Code Link: GitHub

4. GFANC-RL: Reinforcement Learning-based Generative Fixed-filter Active Noise Control
   Journal: Neural Networks, 2024, 106687. Paper Link: ScienceDirect   Code Link: GitHub

5. Unsupervised Learning based End-to-end Delayless Generative Fixed-filter Active Noise Control Conference: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2024)
   Paper Link: IEEE   Code Link: GitHub

6. Real-time implementation and explainable AI analysis of delayless CNN-based selective fixed-filter active noise control
   Journal: Mechanical Systems and Signal Processing, 2024, 214: 111364.
   Paper Link: ScienceDirect   Code Link: GitHub

7. A Hybrid SFANC-FxNLMS Algorithm for Active Noise Control based on Deep Learning
   Journal: IEEE Signal Processing Letters, 2022, 29: 1102-1106. Paper Link: IEEE   Code Link: GitHub

8. Performance Evaluation of Selective Fixed-filter Active Noise Control based on Different Convolutional Neural Networks
   Conference: The 51st International Congress and Exposition on Noise Control Engineering (Inter-Noise 2022) Paper Link: arXiv

If you are interested in our works, please consider citing our papers. Thanks! Have a great day!