Abstract
Most of the literature focuses on the development of the linear active noise control (ANC) techniques. However, ANC systems might have to deal with some nonlinear components and the performance of linear ANC techniques may degrade in this scenario. To overcome this limitation, nonlinear ANC (NLANC) algorithms were developed. In Part II, we review the development of NLANC algorithms during the last decade. The contributions of heuristic ANC algorithms are outlined. Moreover, we emphasize recent advances of NLANC algorithms, such as spline ANC algorithms, kernel adaptive filters, and nonlinear distributed ANC algorithms. Then, we present recent applications of ANC technique including linear and nonlinear perspectives. Future research challenges regarding ANC techniques are also discussed.
Introduction: The Challenge of Nonlinearity
The "gold standard" for noise cancellation has long been the Filtered-x Least Mean Square (FxLMS) algorithm. While it performs exceptionally well in linear environments, real-world acoustic scenarios often introduce complexities that linear models cannot resolve. Nonlinearities typically arise from three primary sources: high-pressure sound propagation in ducts (primary path), hardware saturation in amplifiers and loudspeakers (secondary path), and the chaotic nature of industrial noise sources like fans or grinders.
When these nonlinear factors are present, standard linear ANC performance degrades significantly. This review surveys specialized NLANC algorithms designed to model these complex relationships and maintain effective noise attenuation in non-ideal conditions.
NLANC System Architecture
To address these challenges, NLANC systems incorporate specialized components to handle non-linear signal transformations. A typical architecture involves a reference microphone capturing the noise source, followed by a nonlinear expansion block that prepares the signal for filtering.
As shown in the diagram, the system employs an adaptive algorithm to update the digital filter weights based on an error signal $e(n)$ captured by an error microphone. This feedback loop allows the system to compensate for the secondary path $S(z)$ and its estimate $\hat{S}(z)$, ensuring the anti-noise generated by the active loudspeaker effectively cancels the primary noise $d(n)$.
Algorithmic Methodologies
The landscape of NLANC algorithms has expanded significantly over the last decade, focusing on balancing computational complexity with modeling accuracy:
- Volterra Filters: These use series expansions to model nonlinearities. While powerful, they are computationally intensive due to the "curse of dimensionality."
- Functional Link Artificial Neural Networks (FLANN): These offer a compromise by using single-layer networks with trigonometric expansions, providing nonlinear modeling without the weight of deep learning.
- Kernel Adaptive Filters (KAF): These utilize the "kernel trick" to map data into high-dimensional spaces. The paper notes that "quantized schemes" or sparsification are essential to manage the growing data dictionary during long-term operation.
- Heuristic Algorithms: Techniques like Particle Swarm Optimization (PSO) and Bacterial Foraging Optimization (BFO) allow for "blind" optimization, reducing the need for precise mathematical models of the secondary path.
- Spline Adaptive Filters (SAF): These use adaptive look-up tables and polynomial interpolation, offering a high-performance, low-overhead alternative to neural networks.
Key Insights and Results
Research indicates a clear trade-off between complexity and performance. While Second-Order Volterra (SOV) filters often yield the best reduction, their cost is high for consumer hardware. Conversely, bio-inspired algorithms like BFO have shown steady-state noise reduction improvements of nearly 5dB over standard Genetic Algorithms.
Practical applications have demonstrated significant success:
- Medical Imaging: Hybrid algorithms have achieved noise attenuation exceeding 40dB in fMRI environments.
- Automotive: Multi-channel systems have reached a 4dB broadband reduction in vehicle interior noise up to 1kHz.
- Hardware Efficiency: Specialized VLSI implementations have achieved a 41% reduction in energy consumption per sample compared to standard architectures.
Future Research Directions
Despite these advancements, several challenges remain. Most advanced NLANC algorithms are still in the theoretical or laboratory stage. The primary engineering hurdle is shrinking these complex models to fit on low-power, real-time Digital Signal Processors (DSPs). Additionally, future work must address theoretical gaps in impulsive noise environments and explore the integration of ANC into Internet of Things (IoT) and Internet of Vehicles (IoV) ecosystems to create distributed "quiet zones" in smart cities.
Summary updated