Nonlinear modeling of MIMO antenna array power amplifiers based on time-delay neural network

An equivalent complex baseband model and time-delay neural networks (TDNN) were employed for modeling power amplifiers (PAs). The effectiveness of different algorithms was compared, and the impact of various parameters on the model's generalization ability was discussed. Ultimately, the Levenberg-Marquardt algorithm was chosen for modeling. Furthermore, this paper investigates the distortion characteristics and crosstalk issues in multiple-input multiple-output (MIMO) systems. By mitigating crosstalk, the system's normalized mean square error (NMSE) was achieved at -65.5 dB, indicating a significant improvement in performance after crosstalk elimination. After that a PA system was established by MATLAB/Simulink. Firstly, a reference PA system was constructed and tested using a 16-QAM signal with a bandwidth of 60 MHz. Subsequently, a transmit system was built based on the NXP Airfast LDMOS Doherty power amplifier, operating at a center frequency of 3.7 GHz. The PA model was tested with dual-tone signals and a wideband 5G Orthogonal Frequency Division Multiplexing (OFDM) signal with a bandwidth of 100 MHz.The results showed that in the dual-tone test, the NMSE of the PA model was -44.44 dB. In comparison, in the OFDM test, the NMSE of the PA model was -47.48 dB, while the NMSE of the memoryless model and the MP model provided by MATLAB were -33.65 dB and -43.30 dB, respectively. The neural network PA model exhibited an improvement of 4.18 dB compared to the MP model and 13.83 dB compared to the memoryless model.