The Research Of Diffusion Robust Algorithms For System Identification Applications

Distributed adaptive filtering is an extension of adaptive filtering into distributed wireless sensor networks.Diffusion adaptive filtering has become a research hot spot of recent years due to its stable structure and strong adaptability.Plenty of scenarios that require system identification to recognize transmission channels.Therefore,the application of system identification that based on adaptive filtering has become a very important research topic in communication systems.The diffusion algorithm is the core of diffusion adaptive filtering.Due to the widespread presence of impulse noise in practical environments,traditional diffusion algorithms assume that the system to be identified is affected by Gaussian noise,resulting in severe performance degradation and lack of robustness.This thesis mainly focuses on the diffusion robust algorithm suitable for system identification under impulse noise and further improve the relevant performance of the algorithm.The main research content of this thesis is following:(1)In order to solve the problem that the performance of traditional diffusion algorithms deteriorates in impulse noise environment and that the diffusion robust algorithm cannot take into account the convergence rate and steady-state error,two variable scaling factor diffusion robust algorithms are proposed respectively.On the one hand,two diffused robust algorithms are constructed by using two nonlinear loss functions are insensitive to the error outliers under impulse noise interference.On the other hand,the scaling factor of the cost function of the two algorithms was analyzed.In view of the problem that the fixed scaling factor could not adapt to the change of the error quantity,the variable scaling strategy was introduced to construct two variable scaling factor diffusion algorithms.The mean performance and convergence of relevant parameters in the respective variable-scale functions of the two algorithms were analyzed.Simulation experiments are carried out on the system identification applications of the two variable scale factor algorithms in different impulse noise environments.Simulation experiments show that the variable-scale-factor strategy makes the performance of the algorithm better.(2)To improve the problems of poor tracking performance and high steady-state error of the existing diffusion robust algorithm.In this thesis,the error P-power strategy is introduced into the existing algorithm,proposed a diffusion scaling factor least logarithmic absolute difference base on P-power.While ensuring the robustness of the algorithm,the steepness and extremum of the cost function are adjusted by the error P-power strategy and the scale factor,so that the new algorithm can not only have lower steady-state error in the case of small error,but also have faster convergence speed and tracking performance in the case of large error in the case of system mutation.And analyzed the algorithm of P-power selection principle,mean performance and steady-state error.Simulation experiments are carried out on the system identification of the proposed algorithm in different impulse noise environments.Simulation experiments show that the proposed algorithm has better performance.