基于具有时序结构的稀疏贝叶斯学习的水声目标 DOA 估计研究

摘要

现有的基于CS-MMV（Compressed Sensing-Multiple Measurement Vectors）模型的DOA估计一般都假定信号源为独立同分布（ i．i．d），算法建立在信号的空间结构上进行分析，而当处理具有时序结构的源信号时表现出性能和鲁棒性差的问题，为此该文提出一种具有时序结构的稀疏贝叶斯学习的DOA算法，该方法通过建立一阶自回归过程（ AR）来描述具有时序结构的水声信号，将信号源的时间结构特性充分应用到DOA估计模型中，然后采用针对多测量矢量的稀疏贝叶斯学习（ Muti-vectors Sparse Bayesian Learning ）算法重构信号空间谱，建立多重测量向量中恢复未知稀疏源的信号的CS（ Compressed Sensing ）模型，最终完成DOA估计．仿真结果表明该方法相对于传统的算法具有更高的空间分辨率和估计精度的特点，且抗干扰能力强．%Assuming independently but identically distributed sources,most existing DOA algorithms based on the CS-MMV model are analyzed according to the spatial structure of the signals.The temporal correlation between the sources,how-ever,results in poor performance and robustness.To overcome this problem,we propose a DOA estimation algorithm based on Sparse Bayesian Learning ( SBL) with temporally correlated source vectors.In this method,an underwater acoustic source is regarded as a first-order autoregressive process,with time structure characteristics being applied to DOA estimation model.Af-ter that,the multi-vector SBL algorithm is used to reconstruct the signal spatial spectrum.Then the CS-MMV model of the un-known sparse vector signal sources is established to estimate the DOA.Through simulation,it shows that the proposed algo-rithm provides a higher spatial resolution and estimation accuracy in comparison to many other current algorithms.