Dynamic Dictionary Algorithm for CS-based Localization in Wireless Sensor Networks

Traditional Compressive Sensing (CS)-based localization methods assume all targets fall on a pre- sampled and fixed grid. There will be mismatch between the adopted and actual sparsifying dictionaries when some targets fall off the grid, leading these methods to perform poorly. To address this problem, an efficient dynamic dictionary algorithm is developed for CS-based localization. To achieve this, the actual sparsifying dictionary is modeled as a parameterized dictionary with the grid viewed as adjustable parameters. By doing so, the localization problem is formulated as a joint sparse reconstruction and parameter estimation problem. Additionally, the non-convex parameter optimization problem is transformed into a tractable convex problem by approximating the actual sparsifying dictionary with its first Taylor expansion. Extensive simulation results show that the proposed dynamic dictionary algorithm provides better performance than the state-of-the-art fixed dictionary algorithms.