MATLAB Source Code for GCC Algorithm with Maximum Likelihood (ML) Weighting Function

The Generalized Cross-Correlation (GCC) method is a widely used time delay estimation technique in signal processing, where the choice of weighting function significantly impacts algorithm performance. The Maximum Likelihood (ML) form of weighting function effectively suppresses noise and reverberation interference, enhancing the accuracy of time delay estimation.

The core implementation of GCC-ML algorithm in MATLAB can be divided into three main stages:

Signal Preprocessing Stage First, input signals need to be segmented into frames and windowed, typically using Hanning or Hamming windows to reduce spectral leakage. The implementation involves using buffer() function for framing and applying window functions. Then, Fast Fourier Transform (FFT) is performed to convert time-domain signals to frequency domain using fft() function, obtaining the cross-power spectrum between signals.

Weighting Function Calculation The ML weighting function's characteristic lies in considering the noise power spectra of both channels. Implementation requires calculating auto-power spectra for both channels and constructing a signal-to-noise ratio (SNR) related weighting coefficient matrix. This stage requires special attention to numerical stability issues, typically by adding a small regularization term using eps or a small constant to prevent division by zero in the denominator.

Time Delay Estimation Stage The weighted cross-power spectrum is inverse Fourier transformed using ifft() to obtain the generalized correlation function. Time delay estimation is achieved by locating the peak position. To improve resolution, peak refinement methods like parabolic interpolation can be implemented using polynomial fitting techniques around the peak index.

Compared to other weighting forms like PHAT (Phase Transform), the ML weighting function's advantage is its effective suppression of interference in low-SNR frequency bands, making it particularly suitable for acoustic environments with colored noise. However, it's important to note that ML weighting is sensitive to the accuracy of noise power spectrum estimation, requiring the design of appropriate noise estimation algorithms in practical implementations, such as using minimum statistics or spectral subtraction methods.

This algorithm performs excellently in applications like sound source localization and echo cancellation. By adjusting frequency band weighting strategies, its performance in specific scenarios can be further optimized through parameter tuning in the weighting function implementation.