Abstract
This paper presents a novel delayless block individual-weighting-factors sign subband adaptive filter (DB-IWF-SSAF) designed for robust active noise control (ANC) in environments plagued by impulsive noise. By combining a block-processing architecture with an improved band-dependent variable step-size (I-BDVSS) algorithm, the proposed method achieves superior noise reduction and faster convergence with significantly lower computational overhead than traditional delayless subband filters.
Introduction
Active Noise Control (ANC) is a critical technology for real-time audio applications, such as noise-canceling headphones. However, traditional ANC algorithms, like the filtered-x least mean square (FxLMS), are optimized for Gaussian noise and often fail when encountering impulsive noise—sudden, sharp acoustic spikes (e.g., machinery clanks or gunshots). These spikes violate the $l_2$-norm assumptions of standard filters.
To address this, researchers utilize Sign Subband Adaptive Filters (SSAF), which employ the more robust $l_1$-norm. While effective, standard SSAFs introduce signal path delays that are unacceptable for real-time use. Previous "delayless" versions solved the latency issue but remained computationally expensive. This paper proposes a hybrid approach that optimizes both efficiency and robustness.
Proposed Methodology
The authors introduce two major innovations to the SSAF framework:
- Block Implementation with Overlap-Save: Instead of updating filters on a sample-by-sample basis, the proposed DB-IWF-SSAF uses block processing. As illustrated in Figure 1, the filtered-x signal is generated in the frequency domain using Fast Fourier Transforms (FFTs) and the overlap-save technique. This allows for high-speed signal generation while keeping the weight updates in the time domain, where sign-based logic remains computationally trivial.
- Improved Band-Dependent Variable Step-Size (I-BDVSS): Subband filtering splits the signal into various frequency ranges. The proposed I-BDVSS exploits "auxiliary past gradients" that are naturally generated during block processing. By averaging these gradients, the algorithm dynamically adjusts the step-size for each band, leading to smoother and more rapid convergence without adding significant mathematical complexity.
Experimental Setup
The algorithm was tested using simulations modeled on the Symmetric Alpha Stable ($S\alpha S$) distribution to represent varying levels of impulsive noise. The characteristic exponent ($\alpha$) was varied from 1.0 (less impulsive) to 0.4 (highly impulsive). The primary and secondary acoustic paths were modeled using high-order Finite Impulse Response (FIR) filters to simulate realistic acoustic environments.
Key Results and Discussion
- Computational Efficiency: For a filter with $M=1024$ taps, the proposed block method reduced the multiplications required for filtered-x signal generation from 1024 down to approximately 104—a nearly 10x improvement in efficiency.
- Enhanced Noise Reduction: In highly impulsive environments ($\alpha = 0.6$), the proposed I-BDVSS achieved an Averaged Noise Reduction (ANR) of ~22 dB, compared to the ~17 dB achieved by the original BDVSS algorithm.
- Robust Stability: While conventional algorithms often become unstable or fail to converge under extreme impulsivity ($\alpha = 0.4$), the DB-IWF-SSAF remained stable and effective, proving its suitability for harsh real-world noise environments.
Conclusion
The DB-IWF-SSAF offers a powerful balance of robustness and efficiency for real-time ANC. By leveraging frequency-domain block processing and an intelligent variable step-size, it overcomes the limitations of previous delayless subband filters. Future research will focus on formal theoretical convergence analysis and hardware implementation for industrial and consumer audio devices.
Summary updated