A Testbed on Xilinx RF-SoC for Data-Driven RF Power Amplifier AI/ML-Model Development
Conference
Weerasooriya, H, Karunanayake, K, Gayanath, B et al. (2025). A Testbed on Xilinx RF-SoC for Data-Driven RF Power Amplifier AI/ML-Model Development
. 10.1109/MAPCON65020.2025.11426276
Weerasooriya, H, Karunanayake, K, Gayanath, B et al. (2025). A Testbed on Xilinx RF-SoC for Data-Driven RF Power Amplifier AI/ML-Model Development
. 10.1109/MAPCON65020.2025.11426276
ML based digital twins that model the nonlinear dynamic behavior of RF PA is important for simulation of fullduplex radios, native-AI RAN, 5G/6G wireless and electronic warfare (EW) applications. The digital twins of PAs should ideally be based on measurements of real amplifiers. For example, PA measurements are achieved via a hardware in the loop testbed for signal capture in real-time as part of the ML loop. We develop digital systems at Florida International University (FIU) towards a generalized framework for training AI/ML models of PAs having up to 1 GHz of baseband bandwidth. The models operate at baseband; and the RF carrier can be located at any band of interest. The instantaneous bandwidth goes up to 1 GHz. The testbed uses an 8-phase polyphase signal processing architecture with ADC/DACs operating at 4 GS/s. Digital data will be provided through standard ML platforms, such as Tensorflow and/or PyTorch.
