Research On Distributed LMS Algorithm

The distributed adaptive filtering algorithm is a new important branch in the signal processing discipline that combines the concept of wireless sensor networks with adaptive filtering theory.During the data information collection and transmission process of the distributed network,the sensor nodes in the network process the collected data information.Then,the collected information is extracted by useful information according to different purposes,and the estimation of unknown target parameters is realized.Compared with the traditional single-node adaptive filtering algorithm,the distributed adaptive filtering algorithm has the advantages of lower steady-state error and faster convergence.Therefore,it has been widely used in radar,communication technology,adaptive control,wireless sensor networks and biomedical fields.The thesis focuses on the most basic distributed Diffusion Least Mean Square(DLMS)algorithm in distributed adaptive filtering algorithm,which has the characteristics of small computation and simple structure.However,since such an algorithm is based on second-order moment statistics,when the algorithm is applied to the background of non-Gaussian noise,its performance cannot be guaranteed or even the algorithm is completely invalid.In order to solve the above problems,the thesis introduces the relevant theory of entropy,and chooses the minimum error entropy and the maximum corretropy criterion to replace the traditional mean square error criterion.In addition,considering the sparse characteristics of systems such as hands-free calling,and the idea of the proportionate matrix in the Proportionate Normalized LMS(PNLMS)algorithm,the distributed Diffusion Proportionate Minimum Error Entropy(DPMEE)algorithm and the Diffusion Proportionate Maximum Corretropy Criterion(DPMCC)algorithm is proposed.The two algorithms combine the good convergence characteristics of the sparse channel with the distributed diffusion proportionate adaptive filtering algorithm and the ability of the entropy algorithm to suppress the non-Gaussian noise.Theoretical analysis and simulation results show that the proposed algorithm has good convergence performance,steady-state performance and tracking performance under non-Gaussian noise and Gaussian noise,which verifies the effectiveness and feasibility of the new algorithm.