Digital predistortion of power amplifier based on deep neural network model for wireless communication system

To support high data rate in wireless communication, power amplifiers (PAs) are required to process wideband signals. The nonlinearity of the PAs will cause great interference to adjacent channels. To compensate the adverse effects brought by the nonlinear power amplifier, this paper proposes an accurate model for the memory nonlinearity of broadband PAs. Based on the theory of neural network, a network model with nested structure, namely the behavior model of channel pipeline recurrent neural network (CPRNNBM), is designed. Compared with the generalized memory polynomial model (GMP) model and the traditional recurrent neural network structure model, the proposed model has stronger computing and memory ability. The simulation results show that CPRNNBM can approximate the working characteristics of the actual power amplifier. Based on CPRNNBM, the inverse power amplifier model is constructed to compensate the nonlinearity of a Class-F power amplifier, and the analysis is carried out from the two perspectives of time domain and frequency domain. The adjacent channel power ratio (ACPR) after digital pre-distortion is 19.1 dB better than the original power amplifier output signal, and 4.88 dB better than the traditional GMP model.