A Momentum Two-Gradient Direction Algorithm with Variable Step Size Applied to Solve Practical Output Constraint Issue for Active Noise Control

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

TL;DR

This paper introduces a "2GD-Momentum" algorithm for Active Noise Control (ANC) that prevents speaker saturation without sacrificing speed. By combining a two-gradient switching mechanism with a momentum factor and variable step size, the authors achieve faster convergence and smoother performance in both static and rapidly changing acoustic environments.

What Problem Does This Paper Address?

Active Noise Control (ANC) works by generating "anti-noise"—a sound wave with the same amplitude but opposite phase as the unwanted noise. However, real-world hardware has limits. If the required anti-noise is louder than what the secondary speaker can physically produce (output saturation), the system's performance degrades, and the adaptive algorithm can even diverge, making the noise worse.

Existing solutions, like the Two-Gradient Direction (2GD) algorithm, handle this by switching the "direction" of the filter updates when a limit is hit, effectively bouncing the system's weights off the boundary. While this prevents saturation, it often leads to two major issues:

  1. Oscillation: The weights "vibrate" at the boundary, creating steady-state error.
  2. Slow Convergence: To stop the vibration, researchers often use very small step sizes, which makes the system painfully slow to react when the environment changes (e.g., if a person moves or a door opens).

Methods

The authors propose the 2GD-Momentum algorithm with variable step size (VSS). Their approach builds on the standard Filtered-X Least Mean Square (FXLMS) framework but adds three key layers:

  • Two-Gradient Switching: If the output power is within limits, it uses the standard gradient to minimize error. If it exceeds the limit, it switches to a second gradient that pulls the weights back toward the constraint boundary.
  • Variable Step Size (VSS): The algorithm automatically reduces the step size when the system hits the saturation threshold. This acts as a "soft landing," significantly reducing the weight oscillations (vibrations) at the boundary.
  • Momentum Factor: To compensate for the slowdown caused by the VSS, the authors add a momentum term ($\kappa$). This uses the "inertia" of previous updates to accelerate convergence, which is critical in dynamic environments where the acoustic path changes over time.

The paper also provides a rigorous mathematical convergence analysis, defining the stable bounds for step sizes and the "time constant" of the system.

Datasets

The paper evaluates the algorithm through simulations using real-world measurements:

  • Acoustic Paths: The primary and secondary acoustic paths were measured from a physical air duct at a sampling frequency of 16 kHz.
  • Noise Types: The system was tested against Broadband noise (200–800 Hz) and White Gaussian Noise.
  • Scenarios: The authors tested both a static environment (fixed noise source and paths) and a time-varying environment (where the primary path suddenly changes during operation).

Key Insights and Ideas

  • The "Sub-optimal" Sweet Spot: When hardware limits are reached, the "perfect" mathematical solution is physically impossible. The 2GD-Momentum algorithm is designed to find the best possible "sub-optimal" solution that stays exactly on the edge of the speaker's capability.
  • Momentum as a Compensator: The most clever part of the design is using momentum to offset the sluggishness of the variable step size. It allows the system to be "cautious" at the boundary (low VSS) but "aggressive" when moving toward a new target (high momentum).
  • Avoiding Frequency Boosting: Standard FXLMS, when saturated, often accidentally "boosts" noise at frequencies outside its control band. The proposed method avoids this artifact, maintaining a cleaner output spectrum.

Key Results

  • Reduced Vibration: In static tests, the 2GD-Momentum method showed significantly lower weight fluctuation compared to standard 2GD and rescaling algorithms.
  • Faster Recovery: In the time-varied simulation (where the environment changes at $t=5$ seconds), the proposed method reached the new optimal operating point much faster than its predecessors.
  • Spectral Purity: As shown in the power spectrum plots, the 2GD-Momentum algorithm successfully attenuated noise in the 200–800 Hz band without the unwanted frequency spikes seen in unconstrained FXLMS.

Limitations and Future Work

While the results are promising, the paper focuses on simulations using measured paths. Future work would likely involve:

  • Real-time Hardware Implementation: Testing on low-power DSPs to see if the added computational complexity of the momentum and VSS calculations impacts real-time latency.
  • Online Secondary Path Modeling: The current study assumes the secondary path is modeled offline. In many real-world scenarios, the secondary path (the path from the speaker to the error mic) can also change, requiring an online modeling component.

Summary updated

Abstract

Active noise control (ANC) has been widely utilized to reduce unwanted environmental noise. The primary objective of ANC is to generate an anti-noise with the same amplitude but the opposite phase of the primary noise using the secondary source. However, the effectiveness of the ANC application is impacted by the speaker's output saturation. This paper proposes a two-gradient direction ANC algorithm with a momentum factor to solve the saturation with faster convergence. In order to make it implemented in real-time, a computation-effective variable step size approach is applied to further reduce the steady-state error brought on by the changing gradient directions. The time constant and step size bound for the momentum two-gradient direction algorithm is analyzed. Simulation results show that the proposed algorithm performs effectively in the time-unvaried and time-varied environment.
BibTeX
@inproceedings{shen2023momentum,
  author = {Xiaoyi Shen and Dongyuan Shi and Zhengding Luo and Junwei Ji and Woon-Seng Gan},
  title = {{A Momentum Two-Gradient Direction Algorithm with Variable Step Size Applied to Solve Practical Output Constraint Issue for Active Noise Control}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2023},
  doi = {10.1109/icassp49357.2023.10095950},
}

Loading related content...

Figure