Study On The Blind Equalization Algorithm Based On Gaussian Process Regression

Channel equalization is one of the three major antifading technology(diversity reception,channel equalization and channel coding)in communication system,To solve the intersymbol interference problem caused due to the nonlinear and time-varying channel,Its essence is to compensate the features of the channel or the entire transmission system.The traditional adaptive equalization technology need to send a training sequence regularly to adjust the parameters of the equalizer,which waste the limited channel bandwidth.Blind equalization as a adaptive equalization technology that does not require a training sequence,only use priori information of the receive sequence itself to compensate the channel,has been the research hotspot in the field of signal processing in recent years.Bussgang blind equalization algorithm is currently the most commonly used type of blind equalization algorithm,because of their non-ideal and finite length filter causing non-convex cost function,thus causes the solution of the equalizer coefficients trap in local optimal solution,and leads to wrong convergence phenomenon.To this end,scholars have put forward blind equalization method based on Support Vector Regression Machine(SVR),overcoming the shortcomings of non-convex cost function in the Bussgang class blind equalization algorithm.However,the selection of parameter in the blind equalization algorithm of SVR is usually specified by experience,or through cross validation program to determine,this increases the complexity of the algorithm.This paper mainly study a new blind equalization algorithm based on gaussian process regression(GPR),it optimize super parameters and obtain the optimal equalizer output for the same process,and do not need to cross validation procedures,thus improve the efficiency of the algorithm.In this paper,the basic theory of blind equalization,Bussgang blind equalization algorithm and the principle of SVR blind equalization algorithm are introduced,and verified by experiments that the performance advantages of the SVR blind equalization Algorithm,which compared with the classical Bussgang blind equalization Algorithm,as Constant Modulus Algorithm and multimode Algorithm.On this basis,this paper expounds the blind equalization based on gaussian process regression theory and the Constant Modulus signal blind equalization algorithm based on the GPR,and then put forward the framework of the multimode signal blind equalization algorithm based on gaussian process regression,presents a complete theoretical derivation and experimental verification.In order to prove the advantage of the GPR blind equalization algorithm is studied in the text,we compared with the CMA,MMA and SVR blind equalization algorithm through the simulation experiment,Results show that compared with the SVR blind equalization algorithm,the advantages of the GPR blind equalization algorithm as follows:(1)Parameter selection without cumbersome cross validation process;(2)the super parameters optimized and the optimal equalizer output obtained for the same process;(3)fast convergence and perform the steady ISI.In addition,in this paper,based on gaussian process regression framework of multi-mode blind equalization algorithm is more suitable for the equilibrium of high-order QAM signal.