Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Co-Initialization of Control Filter and Secondary Path via Meta-Learning for Active Noise Control

Ziyi Yang, Li Rao, Zhengding Luo, Dongyuan Shi, Qirui Huang, Woon-Seng Gan

TL;DR

The paper addresses slow start-up in FxLMS-based active noise control when the acoustic environment changes by introducing a co-initialization strategy via Model-Agnostic Meta-Learning (MAML) to jointly initialize the control filter and the secondary-path model. The learned initializations are applied at deployment while preserving the existing runtime adaptive updates, enabling rapid adaptation through short inner loops that simulate identification and residual-noise reduction. Evaluations on an Online Secondary Path Modeling (OSPM) FxLMS testbed with measured in-ear headphone paths show faster convergence, lower early residuals, reduced auxiliary-noise energy, and quicker recovery after path changes compared to a non-reinitialized baseline. A diversity analysis demonstrates that training-path diversity, especially secondary-path diversity, improves initialization performance, highlighting practical guidance for data collection and meta-training.

Abstract

Active noise control (ANC) must adapt quickly when the acoustic environment changes, yet early performance is largely dictated by initialization. We address this with a Model-Agnostic Meta-Learning (MAML) co-initialization that jointly sets the control filter and the secondary-path model for FxLMS-based ANC while keeping the runtime algorithm unchanged. The initializer is pre-trained on a small set of measured paths using short two-phase inner loops that mimic identification followed by residual-noise reduction, and is applied by simply setting the learned initial coefficients. In an online secondary path modeling FxLMS testbed, it yields lower early-stage error, shorter time-to-target, reduced auxiliary-noise energy, and faster recovery after path changes than a baseline without re-initialization. The method provides a simple fast start for feedforward ANC under environment changes, requiring a small set of paths to pre-train.

Co-Initialization of Control Filter and Secondary Path via Meta-Learning for Active Noise Control

TL;DR

The paper addresses slow start-up in FxLMS-based active noise control when the acoustic environment changes by introducing a co-initialization strategy via Model-Agnostic Meta-Learning (MAML) to jointly initialize the control filter and the secondary-path model. The learned initializations are applied at deployment while preserving the existing runtime adaptive updates, enabling rapid adaptation through short inner loops that simulate identification and residual-noise reduction. Evaluations on an Online Secondary Path Modeling (OSPM) FxLMS testbed with measured in-ear headphone paths show faster convergence, lower early residuals, reduced auxiliary-noise energy, and quicker recovery after path changes compared to a non-reinitialized baseline. A diversity analysis demonstrates that training-path diversity, especially secondary-path diversity, improves initialization performance, highlighting practical guidance for data collection and meta-training.

Abstract

Active noise control (ANC) must adapt quickly when the acoustic environment changes, yet early performance is largely dictated by initialization. We address this with a Model-Agnostic Meta-Learning (MAML) co-initialization that jointly sets the control filter and the secondary-path model for FxLMS-based ANC while keeping the runtime algorithm unchanged. The initializer is pre-trained on a small set of measured paths using short two-phase inner loops that mimic identification followed by residual-noise reduction, and is applied by simply setting the learned initial coefficients. In an online secondary path modeling FxLMS testbed, it yields lower early-stage error, shorter time-to-target, reduced auxiliary-noise energy, and faster recovery after path changes than a baseline without re-initialization. The method provides a simple fast start for feedforward ANC under environment changes, requiring a small set of paths to pre-train.
Paper Structure (12 sections, 17 equations, 3 figures, 1 table, 1 algorithm)

This paper contains 12 sections, 17 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Online Secondary Path Modeling (OSPM)
  3. Co-initialization for Feedforward ANC based on Meta Learning
  4. Inner Loop — Training (per sampled task)
  5. Inner Loop — Testing
  6. Across-task MAML update
  7. Simulation
  8. Path Dataset and Diversity Evaluation
  9. Evaluation of online modeling FxLMS under path change
  10. Effect of training-set diversity on initialization performance
  11. Conclusion
  12. Acknowledgement

Figures (3)

  • Figure 1: Modified cross–updated online secondary–path modeling with auxiliary noise (with error jump detector). Dashed arrows indicate re–initialization and scheduling actions.
  • Figure 2: Online modeling FxLMS with auxiliary-noise power (paths switch at $t{=}60$ s & $t{=}120$ s).
  • Figure 3: Magnitude responses of the 46 measured paths from the PANDAR database. (a) Primary paths; (b) Secondary paths.