Sparse representation perspective for source localization based on JSL0-SVD
基于稀疏表示的信号DOA估计

The source localization problem was cast as the problem of recovering a joint sparse representation. It used the singular value decomposition of the data matrix to summarize multiple time and frequency samples, then imposed the smoothed l0 norm to enforce sparsity and used a fixed-point iteration approach to solve the joint optimization problem. The proposed algorithm has the following advantages: improved robustness to noise, improved computation efficiency, robustness to limited number of samples, robustness to correlated sources, no requirement of accurate initialization. The performance of the proposed method was compared to standard spectrum based approaches and other sparse based methods.