Research On Identification And Equalization Algorithms For Hammerstein Nonlinear Channel

In satellite communication, the nonlinearity of power amplifier will introduce intersymbol interference and affect the transmission characteristics of the system. In order to eliminate the nonlinear distortion and improve communication performance, a distortion compensator is required at the receving end. The design of distortion compensator and compensation algorithm is closely related to channel characteristics, thus memory characteristic and nonlinearity should be judged firstly. Channel identification and equalization are effective measures to compensate for the channel distortion, reduce inter-symbol interference and improve the reliability of communication systems. The paper focuses mainly on three issues, the algorithm of memory characteristic and nonlinearity judgment, the iteration identification algorithm with complex valued input and the equalization algorithm based on kernel method for Hammerstein model.Firstly, in order to eliminate the nonlinear distortion caused by channel nonlinearity, algorithm of memory characteristic and nonlinearity judgment for Hammerstein channel is proposed. By integrating the plural higher-order cumulant, the memory block identification is derived based on GM algorithm and then, memory characteristic is determined by the estimated memory depth. For the judged memory channel with estimated linear coefficients, the parameter fitting model of input and output signals is constructed based on the least-squares algorithm and the relationship of performance with linear and nonlinear channels under different fitting orders is deduce. On this basis, memory characteristic and nonlinearity of Hammerstein channel is determined. Simulation results show that the proposed algorithm has a good recognition performance for Hammerstein channel with higher max nonlinear intensity.Secondly, to obtain the iteration identification algorithm for Hammerstein model with complex valued input, the extended stochastic gradient algorithm and the extended hierarchical multi-innovation stochastic gradient algorithm are proposed by extending the real domain algorithms to complex domain. Then based on the identified model, the transmitted signals are recovered with Wiener model equalizer. Experimental simulations show that the extended algorithms can effectively identify the nonlinear model and recover the transmitted signals. Meanwhile, by introducing innovation length, the extended hierarchical multi-innovation stochastic gradient algorithm has better performance than the extended stochastic gradient algorithm at the expense of computational complexity.Finally, to improve the equalization performance for the complex kernel least mean square algorithm, the Sigmoid quadratic membership function is used as variable step size in feature space and the variable step size complex kernel least mean square algorithm based on Sigmoid quadratic membership function is proposed. Simulation results show that, comparing with the complex kernel least mean square algorithm, the proposed algorithm can reduce the steady-state error and get stronger equalization performance on the premise of increasing a small amount of computational complexity. Meanwhile, the variable step size algorithm is more flexible by introducing new controllable variables.