Study On The Blind Equalization Algorithms Based On Twin Support Vector Regressor

In a communication system, due to the limited channel bandwidth and some other noises, the transmitted signal is prone to cause inter-symbol interference to result the output signals can not be recognized. Channel equalization is a key technology to solve the inter-symbol interference. Traditional adaptive equalization technology needs to adjust the coefficient of equalizer by sending the training sequence, which occupies limited channel bandwidth and results in lower communication efficiency. Blind equalization is a kind of emerging adaptive equalization technology that the training sequence is not required and just use a priori information of the received sequence itself to achieve channel compensation and the output sequence try to approximate transmission sequence, it is one of the key technologies in digital communication systems to be solved currently.The classic blind equalization methods, such as the Bussgang blind equalization algorithms, because of its simple principle, easy realization, good robustness and strong applicability for different system, it become a kind of the most commonly used blind equalization algorithms. However, due to its limited length non-ideal filter, the cost function is non-convex which causes the phenomenon of local optima and suffers from false convergence. Therefore, some scholars proposed blind equalization method based on support vector machine (SVM) to solve the shortcomings of non-convex cost function in the Bussgang blind equalization algorithms. However, the classical SVM is obtained by solving a quadratic programming function to achieve the global optimization, which makes the computational complexity is too high and does not suitable for practical applications.Twin support vector Machines (TSVM) is developed based on the theory of SVM, it converts a large convex quadratic programming problem of support vector machine to two smaller convex quadratic programming problems and for the training sample set with the equal size, the training speed of TSVM is faster 4 times than SVM and it can be used to solve the classification and regression problems. This paper used excellent small sample learning ability and global optimization features of TSVR to construct blind equalization algorithm. This algorithm is based on the theoretical framework of TSVR to solve the local convergence problem of the Bussgang algorithms and used iterative re-weighted least square (IRWLS) method to improve the convergence rate.This paper studied the basic principle of the blind equalization and analyzed the traditional blind equalization methods with their advantages and disadvantages. Then, the principle of blind equalization based on SVR was elaborated and the experimental analysis was presented. We First proposed blind equalization algorithms based on TSVR, including the classic iterative re-weighted quadratic programming blind equalization algorithms and improved iterative re-weighted least squares algorithm and compared algorithm experiments of blind equalization algorithms based on classic SVR, TSVR and improved SVR, TSVR. It was found that the classic and improved TSVR blind equalization method can achieve a successful channel equalization, and the improved iterative re-weighted least squares based TSVR blind algorithms greatly reduces the computational complexity of the classical blind algorithms based on TSVR and obtain very short running time. In addition, the blind equalization algorithms based on TSVR have better stability than that of SVR. Finally, combining data reuse technology and the advantages of the TSVR based blind algorithms, we proposed the blind equalization method for the short burst signal.