混合稀疏水声信道的动态区分性压缩感知估计

由于水声传播过程中同时存在声信号直达、静态或动态边界反射的现象,水声信道会呈现不同动态特性的多径,形成具有混合稀疏的结构,即多径由静态或相对缓变的平稳多径分量和快速时变的动态多径分量混合组成。对于混合稀疏信道,经典的稀疏信道估计算法未考虑混合稀疏性,将导致算法失配、性能下降;以时变稀疏集为模型,动态压缩感知(DCS)结合卡尔曼滤波(KF-CS)可提高对时变多径分量的估计精度,但KF对静态稀疏分量的估计无法充分挖掘其稀疏性。通过将混合稀疏水声信道建模为由静态和时变支撑集所组成的稀疏集,提出一种动态区分性压缩感知(DDCS)方法。该算法首先结合同步正交匹配追踪(SOMP)和正交匹配追踪(OMP)将混合稀疏多径进行区分,分解为静态分量和时变分量;然后,分别用KF-CS和同步正交匹配追踪算法估计时变和静态多径的幅度;最后,将静态分量和时变分量的估计结果整合以得到整个水声信道的冲激响应。通过海试实验把所提DDCS算法与经典信道估计算法、压缩感知算法和DCS算法进行了比较,验证了所提算法的有效性。结果表明,对混合稀疏水声信道进行区分性稀疏估计可改善信道估计性能,进而可通过信道估计均衡器提升水声通信质量。 

Dynamic discriminative compressed sensing estimation of hybrid sparse underwater acoustic channel

Exploitation of hybrid sparsity for improving underwater acoustic communication was investigated.Through the process of multipath propagation,underwater acoustic signal will experience static and time varying reflections on static or dynamic boundaries such as sea bottom or windy sea surface,which generate static multipath arrivals and time varying multipath arrivals respectively.This type of hybrid sparsity renders the traditional sparse channel estimation strategies exceptionally difficult,as the hybrid sparsity is not considered under the framework of Compressed Sensing(CS) or Dynamic Compressed Sensing(DCS).In this paper,a Dynamic Discriminative Compressed Sensing(DDCS) approach is proposed to perform different sparse estimation strategy on different type of multipath components discriminatively.Firstly,hybrid multipath arrivals are initially decomposed into static and time varying components by applying Simultaneous Orthogonal Matching Pursuit(SOMP) among continuous data block for identifying static component and Orthogonal Matching Pursuit(OMP) on its residual error for obtaining time varying component.After that the time varying and static multipath is updated by Kalman filtering CS and SOMP,respectively.At last,the whole channel response is obtained by summing up the static and time varying components.Experimental results exhibit that the proposed algorithm outperforms the classic CS or DCS estimation approached at the presence of time variations,thus demonstrating the effectiveness of hybrid sparsity exploitation in improving channel estimation.