基于噪声自适应加权的GCC时延估计算法

A GCC Time Delay Estimation Algorithm Based on Noise-Adaptive Weighting

摘要: 在强噪声环境下，传统GCC时延估计方法性能会急剧下降，主要原因是噪声会干扰信号的互相关函数，造成旁瓣抬升和主瓣展宽。此时，峰值所对应的时延会出现偏差。广义互相关(Generalized Cross-Correlation, GCC)通过引入加权函数来减轻噪声带来的影响，例如，PHAT加权通过相位归一化减轻信号幅度对时延估计的影响，但它对所有频率分量一视同仁，不能有效区分高信噪比的有效信号频带和噪声占主导的低信噪比频带。为此，本文提出了一种噪声自适应加权函数(Noise-Robust Adaptive Weighting, NRAW)，对GCC-PHAT的输出进行二次频域加权，具体的权重根据局部信噪比来动态调节，高信噪比分量保持PHAT特性，低信噪比分量则被大幅抑制。该方法无需预设的噪声模型，仅利用信号起始段的静音假设，就能针对性地抑制噪声频带。实验结果显示，在SNR = −5 dB的强噪声条件下，NRAW方法的平均绝对误差较GCC的7.14˚降低到4.57˚。同时，在5˚误差容限内，方位角估计准确率从50%提高到68%，这充分验证了该自适应加权策略在抑制噪声干扰和提升定位鲁棒性上的显著效果。

Abstract: In strong noise environments, the performance of traditional GCC time delay estimation methods degrades sharply, primarily because noise interferes with the signal’s cross-correlation function, leading to sidelobe elevation and mainlobe broadening. As a result, deviations occur in the time delay corresponding to the peak value. The Generalized Cross-Correlation (GCC) mitigates the impact of noise by introducing weighting functions; for instance, PHAT weighting reduces the influence of signal amplitude on time delay estimation through phase normalization. However, it treats all frequency components equally and fails to effectively distinguish between effective signal bands with high Signal-to-Noise Ratio (SNR) and noise-dominated bands with low SNR. To address this, this paper proposes a Noise-Robust Adaptive Weighting (NRAW), which applies secondary frequency-domain weighting to the GCC-PHAT output. The specific weights are dynamically adjusted based on local SNR: high-SNR components retain PHAT characteristics, while low-SNR components are significantly suppressed. This method requires no preset noise models and achieves targeted suppression of noise frequency bands solely by leveraging the silence assumption in the initial segment of the signal. Experimental results show that, under strong noise conditions with SNR = −5 dB, the NRAW method reduces the mean absolute error from 7.14˚ for GCC to 4.57˚. Meanwhile, within a 5˚ error tolerance, the azimuth estimation accuracy improves from 50% to 68%. This fully validates the significant effectiveness of the proposed adaptive weighting strategy in suppressing noise interference and enhancing localization robustness.