FPGA realization of RF-PA models with memory effects based on ANFIS

This article presents an adaptive approach system to model the RF power amplifier behavior, taking into account memory effects and nonlinearities. These models are based in an offline training by applying an ANFIS, additionally the model performance is compared with a MPM traditional technique. The ANFIS using 10 or more epochs achieves a lower NMSE. The evaluation of the proposed ANFIS learning system is conducted into an experimental testbed based in a DSP/FPGA hardware implementation, using DSP-Builder tool. A graphical interface developed allows the use of the test bed in a flexible solution, which is able to emulate in a digitally chain three different RF power amplifier behaviors from input-output data extracted, providing the AM-AM and AM-PM distortion curves. The modeling based on ANFIS demonstrates a suitable performance through a reduced NMSE, and an adequate hardware resources utilization. Finally, the obtained results and the experimental testbed enables an entire tool for a further application of power amplifier linearization. Finally, the obtained results and the experimental testbed offers an entire digital tool for a further application on power amplifier linearization.