Research On Blind Equalization Algorithm Based On Support Vector Regressor

Blind equalization is a new adaptive technology without resorting to a training sequence, which only utilizes the prior information of transmitted signals to equalize the channel character, and makes sure that the output sequence approximates the transmitted signals as accurate as possible. It is a hot research issue in communicating, signal processing and detecting, which is very important for instructing both the theory and the applications in communication, radar, sonar, control engineering, biomedicine engineering and so on.Statistical learning theory is a special theory of researching machine learning rules on the condition of small sample. Because of the theory is more and more maturity and the neural network is lacking of substantive progress in the theory, statistical learning theory is becoming more and more important. It provides a unified framework for solving the limited sample learning.The major contribution of this paper is summarized as follow:(1) This paper systematically describes the basic theory of blind equalization, analysises several kind of blind equalization algorithms that commonly used: Blind equalization algorithm based on Bussgang; Blind equalization algorithm based on higher order statistics; Blind equalization algorithm based on neural network; Blind equalization algorithm based on wavelet transform and Blind equalization based on support vector machine.(2) Overviews the basic theory of support vector machine which is the new machine learning method based on statistical learning theory and structural risk minimization theory, it not only considers the fit of the training samples but also taking into account the complexity of training samples, the support vector regressor has good performance generalization.(3) Summarizes the progress of how to select the support vector regresson model. the computer simulation results of one-dimensional regression problems describes the advantages of RBF kernel function and a set of optimum model parameters. Model options include the type of kernel function, the model regularization parameter C, not sensitive parameterεand so on.(4) Because of the good generalization properties the support vector regressor presents, two kinds of blind equalizations based on support vector regressor were developed, the algorithm carries out the model and some procedures were proposed for attaining the solutions. Then the simulation results are provided to show higher convergence speed and accuracy of the proposed algorithms compared with the traditional blind equalization algorithm.