Adaptive Detection Algorithms for Multiple Targets Combining Latent Variable Models and Sparsity

This paper considers the challenge of detecting multiple targets in a radar scenario, where several point-like targets with distinct angles of arrival (AoAs) coexist within the same range bin. Traditional methods often experience performance degradation due to assumptions of identical AoAs or reliance on prior knowledge of the targets. To address these limitations, we propose an approach that combines a suitable clustering algorithm with sparsity-based reconstruction techniques. The former classifies the range bins with targets while adaptively estimating the interference covariance matrices and target amplitudes using maximum a posteriori criteria. Concurrently, the sparse recovery methods, which exploit sparsity-promoting priors, estimate AoAs without requiring precise steering vector information, thus enhancing the algorithm f lexibility in unknown environments. Remarkably, the proposed detection architectures eliminate the need for prior knowledge of the number or (angular and range) location of targets, enabling robust detection in scenarios with homogeneous interference. Performance analysis confirms the effectiveness of the proposed architecture, demonstrating its ability to accurately cluster multiple target signals and provide (at least rough) estimates of the AoAs.