无线传感器网络中基于字典优化的压缩感知定位方法

Dictionary Refinement-based Localization Method Using Compressive Sensing in Wireless Sensor Networks

传统的压缩感知定位方法将物理空间离散化为一个固定网格,并假设所有目标准确地落在该网格上,从而将定位问题转化为稀疏重构问题。事实上,目标的随机性导致很难找到满足上述假设的固定网格,进而引起字典失配问题,使得定位性能急剧下降。针对该问题,文中提出一种基于字典优化的压缩感知定位方法,将稀疏字典建模为以网格为参数的参数化字典,通过动态调整网格不断优化稀疏字典,从而将定位问题转化为联合参数优化的稀疏重构问题,并在变分贝叶斯推理框架下解决该问题。仿真结果表明,与传统的压缩感知定位方法相比,所提方法具有更强的可靠性和鲁棒性。

Traditional Compressive Sensing (CS)-based localization methods divide physical space into a fixed grid and assume that all targets fall on the grid precisely,therefore formulating the localization problem into a sparse reconstruction problem.In fact,it is very difficult to find such a fix grid because of the randomness of targets.As a result,there always exists mismatch between the assumed and actual sparse dictionaries,deteriorating localization performance significantly.This paper addressed this problem and proposed a novel dictionary refinement-based localization method using CS.In this method,the true sparse dictionary is modeled as a parameterized dictionary which views grids as adjustable parameters.Based on the model,the sparse dictionary is gradually refined by dynamically adjusting the grid.Consequently,the localization problem is formulated into a joint parameter estimation and sparse reconstruction problem,and this problem is solved under variational Bayesian inference framework.Simulation results show that the proposed localization method is more efficient and robust compared with traditional CS-based methods.