Research On Distributed Subband Adaptive Filtering Algorithm

Distributed adaptive networks have received considerable attention owing to their wide range of applications in the fields of environmental monitoring and spectrum sensing.Among them,the Diffusion Subband Adaptive Filtering(DSAF)algorithm uses the diffusion strategy to exchange information with neighbor nodes and utilizes the subband technology to solve the problem of input signal correlation in the distributed networks.However,the non-Gaussian noise with impulsive characteristics widely exsiting in the practical application scenarios will lead to the performance degradation of the traditional DSAF algorithm or even complete failure.In this thesis,the influence of non-Gaussian noise on the DSAF algorithm is researched deeply.The Maximum Correntropy Criterion(MCC)is used as a new cost function to derive the Maximum Correntropy Criterion Diffusion Subband Adaptive Filtering(MCC-DSAF)algorithm.In the MCC-DSAF algorithm,when the error signal is greatly affected by non-Gaussian noise,the weight coefficient of the adaptive filter is almost not updated,which ensures the stability of the algorithm;when the error signal changes little,the weight coefficient of the filter is updated in a large step to ensure the convergence speed of the algorithm.Furthermore,considering the sparse characteristics of underwater acoustic communication,conference call and other systems,this thesis combines the adaptive gain matrix technology with MCC-DSAF algorithm,and the Maximum Correntropy Criterion Improved Proportionate Diffusion Subband Adaptive Filtering(MCC-IPDSAF)algorithm for sparse system is developed.The MCC-IPDSAF algorithm adjusts each element of the weight vector proportionally according to the weight of each moment in the process of updating the weight coefficient,thus,the convergence speed of the algorithm in sparse system is further accelerated.Based on the reasonable hypothesis,this thesis makes a theoretical analysis of MCC-DSAF algorithm from the aspects of convergence and steady-state performance.A large number of simulation experiments show that the algorithms proposed in this thesis have better convergence performance,steady-state performance and tracking performance compared with some existing DSAF algorithms,and they have good adaptability under different background noise environments in different sparse systems.