A Practical Distributed Active Noise Control Algorithm Overcoming Communication Restrictions

Junwei Ji, Dongyuan Shi, Zhengding Luo, Xiaoyi Shen, Woon-Seng Gan

TL;DR

This paper introduces a novel distributed active noise control (DMCANC) algorithm that uses "compensation filters" to decouple control nodes. By allowing each node to update its weights independently while still accounting for inter-node interference, the system achieves centralized-level performance while remaining highly robust to network latency and intermittent communication.

What Problem Does This Paper Address?

Active Noise Control (ANC) works by generating "anti-noise" to cancel out unwanted sounds. In large spaces, you need multiple speakers (secondary sources) and microphones (error sensors), a setup known as Multichannel ANC (MCANC).

Traditionally, these systems are centralized: one "brain" processes every signal. This is computationally expensive and doesn't scale well. Distributed systems (DMCANC) spread the work across multiple nodes, but they face two major hurdles:

  1. Cross-talk: The anti-noise from Node A is picked up by Node B’s sensor, confusing Node B’s controller.
  2. Communication Constraints: Real-world wireless networks aren't perfect. Delays (latency) and dropped packets (interruptions) can cause distributed algorithms to lag or become unstable.

The authors aim to create a distributed system that performs as well as a centralized one but can survive the "messiness" of real-world networking.

Methods

The core innovation is the use of compensation filters ($c_{km}$) instead of traditional topology-based combination rules.

  • Compensation Filters: During an offline modeling phase, the system learns the relationship between "self-secondary paths" (the path from a node's speaker to its own sensor) and "cross-secondary paths" (the path from a neighbor's speaker to that same sensor).
  • Decoupled Updating: Unlike previous distributed methods that require immediate data from neighbors to update local filters, this algorithm updates local control filters ($\psi_k$) based only on the local error signal.
  • Global Control Filter: The "global" output of a node is a combination of its local filter and the compensation-filtered versions of its neighbors' filters. Because the neighbor information is used to construct the output rather than update the internal weights, the system can tolerate old or missing data from the network.

The authors validated this using the Filtered-X Least Mean Square (FxLMS) framework, a standard in the ANC field.

Datasets

The paper relies on numerical simulations rather than a public static dataset. The simulation environment parameters include:

  • Primary Noise: Broadband signal (100 Hz to 1000 Hz).
  • Sampling Frequency: 16,000 Hz.
  • Path Modeling: Band-pass filters (50 Hz to 5000 Hz) with tap lengths of 320 for acoustic paths and 512 for control filters.
  • Network Conditions: Simulated latencies ranging from 0 to 4000 sampling periods and communication frequencies as low as 0.01%.

Key Insights and Ideas

  • The "Aha" Moment - Decoupling: By moving the neighbor interaction out of the weight-update loop and into a compensation structure, the authors effectively "buffered" the algorithm against network jitter. If a neighbor's update is late, the local node doesn't stall; it just uses the last known state.
  • Cross-talk as a Filter Problem: Instead of treating interference from other nodes as random noise to be ignored, they treat it as a deterministic path difference that can be mathematically "compensated" for.
  • Latency Tolerance: The discovery that the system remains stable even with a latency of 187.5 ms is significant, as standard Wi-Fi latency is typically under 100 ms, suggesting this is ready for off-the-shelf wireless hardware.

Key Results

  • Performance Parity: In ideal conditions, the proposed DMCANC achieved a steady-state Mean Squared Error (MSE) identical to the gold-standard centralized McFXLMS algorithm.
  • Robustness to Latency: The system successfully converged with communication delays up to 3000 sampling periods (approx. 187.5 ms). Performance only began to degrade significantly at 4000 samples.
  • Robustness to Interruption: Remarkably, the system maintained near-optimal noise reduction even when the communication frequency was reduced to 0.1% (exchanging data only ~16 times per second). This proves the algorithm is highly resilient to packet loss and intermittent connectivity.

Limitations and Future Work

  • Offline Modeling: The compensation filters currently require offline modeling. In a dynamic environment where speakers or sensors move (e.g., a person moving their head), these filters would need to be updated online, which adds complexity.
  • Computational Overhead of Filters: While it reduces the central load, each node must now manage compensation filters for its neighbors. As the number of nodes increases, the local memory and processing requirements for these filters will grow.
  • Real-world Hardware: The study is entirely simulation-based. The next logical step, as noted by the authors, is implementation on physical distributed hardware using standard wireless protocols to verify these results in a 3D acoustic space.

Summary updated

Abstract

By assigning the massive computing tasks of the traditional multichannel active noise control (MCANC) system to several distributed control nodes, distributed multichannel active noise control (DMCANC) techniques have become effective global noise reduction solutions with low computational costs. However, existing DMCANC algorithms simply complete the distribution of traditional centralized algorithms by combining neighbour nodes' information but rarely consider the degraded control performance and system stability of distributed units caused by delays and interruptions in communication. Hence, this paper develops a novel DMCANC algorithm that utilizes the compensation filters and neighbour nodes' information to counterbalance the cross-talk effect between channels while maintaining independent weight updating. Since the neighbours' information required barely affects the local control filter updating in each node, this approach can tolerate communication delay and interruption to some extent. Numerical simulations demonstrate that the proposed algorithm can achieve satisfactory noise reduction performance and high robustness to real-world communication challenges.
BibTeX
@inproceedings{ji2023practical,
  author = {Junwei Ji and Dongyuan Shi and Zhengding Luo and Xiaoyi Shen and Woon-Seng Gan},
  title = {{A Practical Distributed Active Noise Control Algorithm Overcoming Communication Restrictions}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2023},
  doi = {10.1109/icassp49357.2023.10097013},
}

Loading related content...

Figure