Research On Information Entropy Adaptive Filtering Algorithm Based On Conjugate Gradient

The rapid development of adaptive filtering algorithm in the past few decades is con-sidered to be of great significance and has been widely used.It mainly includes optimiza-tion criteria and optimization methods.Mean square error is one of the most commonly used criteria.It works well in Gaussian noise,but its performance tends to deteriorate in the widely existing non Gaussian noise environment.The criterion based on information entropy can effectively deal with non Gaussian noise,which mainly includes maximum correlation entropy and minimum error entropy.Stochastic gradient method is widely used in many scenes because of its simplicity and efficiency.However,its convergence speed is slow and the filtering accuracy often needs to be improved.The convergence speed of recursive least squares method is much faster than that of random gradient method,but its computational complexity is high and there is numerical instability.Conjugate gradient method makes a trade-off between convergence speed and computational complexity.Its convergence performance is similar to that of recursive least squares method,but it has lower computational complexity than the latter.In order to effectively deal with the non Gaussian noise in the signal,the minimum error entropy conjugate gradient algorithm is derived and proposed,and the computational complexity of the new algorithm compared with the existing algorithms is studied.Then,the performance of the algorithm is tested in four typical non-Gaussian noise environ-ments,and the influence of algorithm parameters on it is studied,as well as the compar-ison between the new algorithm and some existing algorithms.Finally,the application effect of the algorithm in actual data is studied.Simulation and application show that the proposed algorithm performs well in non Gaussian noise environment,and the compre-hensive performance is better than the existing algorithms.In order to further improve the performance of the minimum error entropy conjugate gradient algorithm,combined with the mixed entropy strategy,a mixed error entropy con-jugate gradient algorithm is proposed.Considering the high computational complexity of the new algorithm,combined with the quantization entropy strategy,a quantized mixed error entropy conjugate gradient algorithm is proposed,and the comparison of the oper-ation cost between the two new algorithms and the existing algorithms is studied.Then,the performance of the algorithm in four non-Gaussian noise environments is tested,and the influence of some parameters of the algorithm on the algorithm and the basis of pa-rameter selection are studied.The performance of the new algorithm is compared with the minimum error entropy conjugate gradient algorithm.Finally,the application effect of the algorithm in practical data is studied.Simulation and application show that the hy-brid error entropy conjugate gradient algorithm improves the filtering accuracy compared with the existing algorithms,and the quantization hybrid error entropy conjugate gradient algorithm improves the filtering accuracy and reduces the operation cost compared with the existing algorithms.