A Novel near Field Source Localization Algorithm Based on Information Theoretic Criteria

Second order statistics (SOS) matrix has been utilized in near field source localization with an extremely low complexity. However, in practice there exits a vector matching defect seriously deteriorating the robustness of the algorithm. In this paper, we introduce the Minimum Description Length (MDL) as a data pre-processing mechanism and generate a merging algorithm with signal detection which can effectively solve this problem. The integral algorithm merely draws in an additional covariance matrix formulation and the eigenvalue decomposition of it, which can be effectively fulfilled by parallel computing. Above all it is the robustness the algorithm gains that potentiates the practical use of the approach. Simulation outcome indicates that even in a highly adverse circumstance with only 4 sensors, the localization error proves to be constrained in 0.001, demonstrating the excellent overall performance of the proposed algorithm. Before that with specific matrix manipulation imposed, a generalized form of algorithm of the type is derived in terms of two delay factors. A concrete approach referring to single delay is proposed and verified to be accurate but highly unstable in unfavorable circumstance, even unable to localize with insufficient sensors available as stated previously.