Distributed Estimation with Novel Adaptive Data Selection Based on a Cross-Matching Mechanism

Distributed estimation using general data selection (DS) has always been applicable for reducing calculation loads in many fields. However, the traditional general DS (GDS) mode can deteriorate algorithm performance and usually neglects solving the problem of communication cost. These issues arise because distributed estimation is extremely susceptible to selecting the fused data and requires swapping all data. To solve these problems, a diffusion least-mean-square (DLMS) algorithm with an adaptive DS (ADS) is proposed to improve the GDS mode. The proposed algorithm can choose more reliable information in the data fusion process and diminish the communication cost (by using the saved intermediate data of previous iteration) and the calculation load. In addition, in GDS mode, the DS factor (DSF) selects data based on noise statistics (NS), resulting in some loss of selection ability. To further improve this situation, a novel cross-matching mechanism is proposed to improve the design of the DSF based on an intermediate estimation error. The mean stability and mean-square performance of the proposed DLMS algorithm with the ADS mode are analyzed theoretically, which can derive a convergence condition based on the step-size. Theoretical verification and target localization simulations are implemented to illustrate the effectiveness and robustness of the proposed ADS algorithm under satisfying the convergence condition as compared to other related DS algorithms.