Research On Distributed Diffusion Estimation Algorithm

Distributed estimation is an important aspect of distributed signal processing.It is to estimate some parameters of interest from noise environment by cooperation.Because distributed estimation is flexible,robust,and efficient,distributed estimation has been considered as an important application of wireless sensor networks(WSNs)and is widely applied in practical applications such as precision agriculture,environmental monitoring,military and air surveillance.Wireless sensor networks are the basis of distributed estimation.Distributed estimation algorithms are the core of distributed estimation.Diffusion Least Mean Square Algorithm(DLMS)is a classical distributed estimation method.However,there are also some disadvantages with the DLMS algorithm.For example,nodes exchange information only with their direct neighbors.The accuracy of the estimation will be improved with the range of neighbor nodes expanded.However,it will cause a significant increase in communication traffic.The DLMS algorithm can achieve a good estimation effect when the background noise is drawn from Gaussian distribution.However,the algorithm may degrade considerably in non-Gaussian situations,especially for the impulsive-noise.In this paper,the Multi-hop RC-DLMS algorithm is proposed to solve the problem that communication load of DLMS will significantly increase with expanding the communication range of the neighbor nodes.The proposed algorithm expands the range of neighbor nodes and receives the intermediate estimates from only a subset of neighbors.The proposed algorithm not only improves the accuracy of the DLMS estimation algorithm,but also achieves an effective balance between the communication load and the estimation accuracy.The performance of the algorithms based on MSE such as,DLMS,DRLS,etc.may degrade considerably in non-Gaussian environment.In this paper,the generalized maximum correntropy criterion is introduced.Furthermore,diffusion generalized maximum correntropy criterion algorithm based on multitask is proposed to solve multitask estimation problems.The relatedness of tasks among nodes is also studied to uncover its impact on estimation performance.Simulations illustrate that the proposed algorithm achieves desirable performance and outperforms other related methods.