A Study On The Constant Modulus Algorithm And Its Initialization

At present, blind equalization is a new and developing self-adaptive equalization technology. It is a hot research area for its adaptive equalization without training sequence. The Bassgang class algorithm is widely used due to its all-around theory, computing efficiency and easy application. Constant modulus algorithm (CMA) is the one that is used mostly and is of the best performance among the class.The main contribution of this thesis is: it systematically analyses the basic theorise of blind equalization and Bassgang class algorithm. CMA is emphasized. Author makes a study on constant modulus algorithm and analyzes its mean square error (MSE) performance systematically to locate the position of constantmodulus algorithm's minimal spot and the upper bound of the mean square error roughly. What is more, the condition of ill-convergence of the CMA, the coefficients that will cause ill-convergence occurring, its stable conditions and the results of ill-convergence are analyzed thoroughly.The constant modulus algorithm is sensitive to its initialization condition. Due to its ill-convergence, improper initialization will lead to the algorithm converges to the minimal local, so ISI cannot be eliminated and the whole system will not recover the input signals. As to this problem, three methods are used: first, make an improvement on the traditional CMA by using tricepstrum. The computer simulation proves that the MSE of the start point of this method is superior to the traditional method's such as the convergence speed is quicker and convergence performance is better. Second, make an improvement on the traditional CMA by using of the relationship between CMA equalizer and Weiner equalizer. It is proved that the MSE of this method's after-converging is better than CMA and the convergence performance is stable. However, convergence speed is not as quickas CMA for the complex computering. Finally, make an improvement on the traditional CMA by applying whited filter. The computer simulation proves that the MSE is small after converging and the convergence performance is better than the traditional CMA.