Noise Suppression for Time Difference of Arrival: Performance Evaluation of a Generalized Cross-Correlation Method Using Mean Signal and Inverse Filter

TL;DR

This paper introduces GCC-MSIF, a new generalized cross-correlation method that enhances TDOA estimation accuracy in noisy environments by reconstructing signals and suppressing noise, outperforming conventional techniques.

Contribution

GCC-MSIF is a novel method combining mean signal estimation and inverse filtering to improve TDOA accuracy under low SNR conditions, demonstrating superior performance over existing methods.

Findings

GCC-MSIF outperforms GCC-PHAT and GCC-SCOT in low SNR environments.

The method's accuracy improves with more array elements.

GCC-MSIF achieves robustness comparable to or better than GCC-ML.

Abstract

This paper proposes a novel generalized cross-correlation (GCC) method, termed GCC-MSIF, to improve time difference of arrival (TDOA) estimation accuracy in noisy environments. Conventional GCC methods often suffer from performance degradation under low signal-to-noise ratio (SNR) conditions, particularly when the signal bandwidth is unknown. GCC-MSIF introduces a "mean signal" estimated from multi-channel inputs and an "inverse filter" to virtually reconstruct the source signal, enabling adaptive suppression of out-of-band noise. Numerical simulations simulating a small-scale array demonstrate that GCC-MSIF significantly outperforms conventional methods, such as GCC-PHAT and GCC-SCOT, in low SNR regions and achieves robustness comparable to or exceeding the maximum likelihood (GCC-ML) method. Furthermore, the estimation accuracy improves scalably with the number of array elements. These results suggest that GCC-MSIF is a promising solution for robust passive localization in practical blind environments.