Scholarlink

Background/Introduction <p>Conventional active noise control (ANC) systems face challenges in complex automotive cabins, failing to adequately address psychoacoustic quality fixed-filter controllers are model-mismatch sensitive, fixed step the Filtered-x Least Mean Square (FxLMS) struggles with convergence steady state error trade-offs, and extraneous references degrade performance.</p> Purpose <p>To overcome these limitations, a sound quality-oriented hybrid ANC framework integrating pre-trained control filters with adaptive techniques is proposed, aiming to enhance in cabin sound quality and control robustness.</p> Methods <p>A multi-automobile, multi-condition in-cabin noise dataset (4 models, 4 road types, 5 speeds, 50 high-fidelity samples) was constructed via real road experiments. A convolutional neural network (CNN) was offline-trained to select near-optimal fixedfilter parameters, integrated into a novel Filtered-error Normalized Least Mean Square (FeNLMS)based architecture. Efficacy was validated via simulations and double-blind subjective evaluations.</p> Results <p>The results show that the pre-trained CNN achieves an optimal filter selection accuracy of 97.8%. Compared with standalone Filtered-x Normalized Least Mean Square (FxNLMS) and FeNLMS algorithms, the proposed hybrid algorithm accelerates the initial convergence speed to 0.1–0.5 s, with noise reduction gains of 7 dB and 6 dB respectively. It also achieves breakthrough optimizations in core psychoacoustic parameters: 1.5 sones lower loudness, 1.0 acum lower sharpness, and 0.4 asper lower roughness, solving the key problem of traditional ANC systems that "reduce noise but fail to improve sound quality". Subjective evaluations confirm the highest scores in perceived noise reduction, sound comfort, and overall preference (mean score 6.0 ± 0.8).</p> Conclusions <p>The proposed hybrid framework effectively combines rapid filter selection and adaptive optimization, providing a robust foundation for next-generation automotive active sound quality systems and supporting technological innovation in the automotive industry.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Active Control Method for Automotive Interior Sound Quality Using a Pre-trained Control Filter

  • 期刊论文

  • 发表日期：

    2026年3月2日

查看全文
  • Rui Li,
  • Zhao Tang,
  • Yangxue Hu,
  • Cheng Li,
  • Shuang Li

摘要

Background/Introduction

Conventional active noise control (ANC) systems face challenges in complex automotive cabins, failing to adequately address psychoacoustic quality fixed-filter controllers are model-mismatch sensitive, fixed step the Filtered-x Least Mean Square (FxLMS) struggles with convergence steady state error trade-offs, and extraneous references degrade performance.

Purpose

To overcome these limitations, a sound quality-oriented hybrid ANC framework integrating pre-trained control filters with adaptive techniques is proposed, aiming to enhance in cabin sound quality and control robustness.

Methods

A multi-automobile, multi-condition in-cabin noise dataset (4 models, 4 road types, 5 speeds, 50 high-fidelity samples) was constructed via real road experiments. A convolutional neural network (CNN) was offline-trained to select near-optimal fixedfilter parameters, integrated into a novel Filtered-error Normalized Least Mean Square (FeNLMS)based architecture. Efficacy was validated via simulations and double-blind subjective evaluations.

Results

The results show that the pre-trained CNN achieves an optimal filter selection accuracy of 97.8%. Compared with standalone Filtered-x Normalized Least Mean Square (FxNLMS) and FeNLMS algorithms, the proposed hybrid algorithm accelerates the initial convergence speed to 0.1–0.5 s, with noise reduction gains of 7 dB and 6 dB respectively. It also achieves breakthrough optimizations in core psychoacoustic parameters: 1.5 sones lower loudness, 1.0 acum lower sharpness, and 0.4 asper lower roughness, solving the key problem of traditional ANC systems that "reduce noise but fail to improve sound quality". Subjective evaluations confirm the highest scores in perceived noise reduction, sound comfort, and overall preference (mean score 6.0 ± 0.8).

Conclusions

The proposed hybrid framework effectively combines rapid filter selection and adaptive optimization, providing a robust foundation for next-generation automotive active sound quality systems and supporting technological innovation in the automotive industry.