Broadband Coherent Source DOA Estimation

Broadband coherent source DOA (Direction of Arrival) estimation is a crucial problem in array signal processing, primarily used to estimate the arrival directions of multiple coherent signal sources. The Coherent Signal Subspace Method (CSSM) is a widely used broadband DOA estimation approach, whose core concept involves focusing signal subspaces at different frequencies onto a reference frequency through spatial resampling, thereby overcoming the non-stationary nature of broadband signals.

### Algorithm Implementation

The key step in spatial resampling is constructing focusing matrices, which transform covariance matrices from different frequency points to the reference frequency. The specific implementation logic includes:

Frequency Band Division: First, divide the broadband signal in the frequency domain, typically using Short-Time Fourier Transform (STFT) or filter bank methods, to obtain multiple narrowband signal components. In MATLAB implementation, this can be achieved using spectrogram() function with appropriate windowing parameters.

Covariance Matrix Calculation: For each frequency point, compute the corresponding array covariance matrix to describe signal statistical characteristics. The code implementation typically involves matrix multiplication of signal vectors: R = X*X'/N, where X represents the signal matrix and N is the number of snapshots.

Focusing Matrix Construction: Select a reference frequency (usually the center frequency), then construct a focusing matrix for each non-reference frequency point. This matrix aligns signal subspaces from different frequencies to the reference frequency. The focusing matrix T(f) can be computed using optimization techniques like least-squares or maximum likelihood criteria.

Subspace Focusing: Transform covariance matrices from various frequency points to the reference frequency using focusing matrices, then average them to obtain the focused covariance matrix. The mathematical operation is: R_focused = Σ T(f_i)R(f_i)T(f_i)^H.

DOA Estimation: Perform eigenvalue decomposition on the focused covariance matrix and use subspace algorithms like MUSIC or ESPRIT to estimate signal source directions. The MUSIC algorithm implementation involves computing the noise subspace and searching for spectrum peaks using steering vectors.

### Key Technical Points

Spatial Resampling: Since the array manifold of broadband signals varies with frequency, directly applying narrowband DOA estimation methods leads to performance degradation. Spatial resampling adjusts signal subspaces at different frequencies through focusing matrices to maintain consistency at the reference frequency.

Focusing Matrix Optimization: The design of focusing matrices directly affects algorithm performance. Common optimization objectives include minimizing signal subspace alignment errors or maximizing the signal-to-noise ratio after focusing. Implementation often involves solving constrained optimization problems.

Computational Efficiency: Due to multi-frequency data processing, the algorithm has high computational complexity. Typically, dimensionality reduction techniques or parallel computing approaches are incorporated to enhance efficiency, such as using principal component analysis for covariance matrix approximation.

### Application Scenarios

This method is suitable for broadband signal processing scenarios including radar, sonar, and wireless communications, particularly in environments with severe multipath effects, where it effectively improves DOA estimation accuracy and robustness.