Keywords: 
• 时频分析 /
• 盲波束形成 /
• 导向矢量最优估计 /
• Capon

Abstract: The existing blind beamforming methods are effective only under the condition that the source signals have some special statistical or structural characteristics. Additionally, the structure of cascade model is complicated and the stability of parallel model is poor when dealing with multi-target signals. To address these problems, a novel blind beamforming algorithm for multi-target signals based on time-frequency (TF) analysis is proposed in this paper. The received array signals are first transformed into time-frequency domain by using quadratic time-frequency distributions (TFDs). Then, the single-source auto-term TF points which show energy concentration at a single signal are extracted through three operations: (i) removing noise points by setting a reasonable threshold, (ii) separating auto-term TF points from cross-term points, and (iii) selecting the single-source auto-term TF points from the auto-term ones. Moreover, these single-source auto-term TF points are classified by the principal eigenvector of their spatial time-frequency distribution matrixes. For each class of TF points, the uncertain set of signal steering vector is given, whose radius is defined as the ultimate range between the center and the elements in the class. Within the uncertain set, an optimization algorithm is provided to get the optimal estimation of the signal steering vector. Finally, the blind beamforming for multi-target signals is achieved based on the Capon method, which can enhance the desired signals and suppress the noise and interference signals. In addition, the influence of parameters selection, the clustering method of unknown source number, and the computational complexity of the proposed algorithm are analyzed. The proposed algorithm can achieve parallel output of multi-target signals under the condition that the array manifold and the direction of arrival (DOA) are unknown. Also, the complex iterative solving processing may be avoided and special limitations on signal characteristics are unnecessary. As a result, it has wide applicability and superior stability compared with the existing blind beamforming methods. Simulations illustrate that the proposed algorithm can well separate multi-target signals which are TF-nondisjoint to a certain extent. It can achieve a higher output signal to interference plus noise ratio (SINR) compared with the constant modulus algorithm (CMA), the independent component analysis (ICA) algorithm, and the joint approximate diagolization of eigenmald (JADE) algorithm. Furthermore, the output performance of the proposed algorithm is close to the optimal Capon beamformer.