Super-resolution direction-of-arrival estimation based on generalized eigendecomposition and sparse reconstruction under near-field interference

Super-resolution methods based on sparse recovery often suffer significant performance degradation under strong near-field interference. Existing approaches typically involve either incorporating near-field steering vectors into the dictionary or applying pre-filtering techniques. However, the former approach introduces a severe basis coherence problem, which may lead to reconstruction failure. The latter may distort the target signal, especially when the interference subspace is high-dimensional. This paper proposes a preprocessing method based on generalized eigenvalue decomposition. Inspired by the concept of matrix filtering, the method aims to maximize the output power ratio between the subspace spanned by the desired steering vectors and the interference subspace. Formally, the proposed method can be viewed as a generalized extension of the optimal beamformer to a matrix filtering framework. Unlike zero-forcing methods, it offers a balanced trade-off between interference suppression and target preservation. Moreover, the computational efficiency is significantly lower than that of conventional matrix filtering approaches that rely on convex optimization. When combined with subsequent sparse recovery algorithms, the proposed method enables super-resolution direction-of-arrival estimation even under conditions of limited snapshots and low signal-to-noise ratios. Simulation and experimental results demonstrate the effectiveness of the proposed method in handling real-world strong near-field interference while maintaining better real-time performance.