Research On Neural Network Based Nonlinear Models For RF Power Amplifiers

Due to its outstanding approximation ability, the neural network is gradually becoming a more and more charming scheme for the power amplifier modeling. Although we don’t understand the internal structure of the power amplifier, the neural network can learn the amplifier internal behavior of the circuit through the input and output data. Therefore, this paper starts from the nonlinear characteristics of a single band power amplifier, and then we describe the memoryless model of several kinds of power amplifiers; then analyze the memory effects of power amplifier and the reasons of memory effect, and introduce several common power amplifier models with memory effect; Next we introduce the NN modeling, and use the RBF neural network to model RF amplifier; Finally, we research the nonlinear characteristics of dual band power amplifiers, and the modeling of this kind of RF power amplifiers. The main research includes:(1) The dynamic nonlinear behavior model of single-band RFPA for broadband application is researched. Then we analyze the root cause of RF power amplifier Then we analyze the classification of memory effect and cause of the memory effect, if the signal bandwidth is wider,the memory effect is more obvious for power amplifier. At this time AM/AM and AM/PM curve will be more and more divergent, So the influence of memory effect on model precision is very high. This theory supplies theoretical foundation for the modeling of the following:(2) We study on the behavior model of single band RF power amplifier and put forward two kinds of single band RF power amplifier behavior models: the first kind is cross-item real-valued time-delay radial basis function neural network, second kind is recurrent type radial basis function neural network. The first model joins the polynomial cross and delay term, cross term is used for strong nonlinear of power amplifier, delay term is used for the simulation of PA memory effects.Experimental results show that the model can be suitable for the behavioral model of RF power amplifier. Compared with the real time delay neural network of BP(back propagation) network,This model can greatly short the time of training and has fast convergence speed. For the purpose of further improving the modeling accuracy, and then we propose a recurrent RBF neural network model. The model adds output feedback terms in the input. The experiment consequence test verifies that the effect of the model for strong nonlinear RF power amplifier modeling has high precision and fast convergence speed.(3) We study on the behavior model of dual band RF power amplifiers, First analyze the nonlinear characteristics of dual band, and then elaborate the significance of dual band for themodern communication technology, then elaborate the problem how to synchronize and obtain dual waveband signals, finally we put forward a dual band RF power amplifier behavioral model based on radial basis function network.