FPGA-Deployed Lightweight Adaptive Hybrid DPD Implementation for PA Linearization in LEO STIN

Abstract

Power amplifiers (PAs) are essential components of Space–Terrestrial Integrated Networks (STINs), where they compensate for the high propagation losses associated with long-distance satellite communication links. To maximize power efficiency, PAs are typically operated near saturation, resulting in nonlinear distortions that degrade signal quality and communication reliability. This work presents a lightweight adaptive hybrid Memory Polynomial Model–Long Short-Term Memory (MPM–LSTM) digital predistortion (DPD) framework with an FPGA-oriented implementation for PA linearization in Low Earth Orbit (LEO)-based STINs. The proposed DPD is experimentally validated using a hardware testbed comprising a Xilinx Zynq-7000 FPGA, a GaN Class-AB power amplifier, a vector signal generator, and a vector signal analyzer. The experimentally obtained predistorted waveform is subsequently employed as the input to a digital twin of a realistic LEO-STIN communication channel incorporating Doppler shift, free-space path loss, Rician fading, shadowing, rain attenuation, and additive white Gaussian noise (AWGN) to evaluate end-to-end communication performance. Experimental RF validation achieves an average adjacent channel power ratio (ACPR) of −50.19 dBc and an error vector magnitude (EVM) of −20.23 dB after 8 bit fixed-point integer (INT8) quantization. Using the measured DPD output within the digital twin, the communication evaluation demonstrates EVM improvement from 3.00 dB to 0.63 dB, throughput enhancement from 29.29 Mbps to 44.99 Mbps, and spectral efficiency improvement from 0.586 to 0.902 bits/s/Hz over an SNR range of 0–30 dB. Furthermore, FPGA implementation reduces processing latency, demonstrating the suitability of the proposed architecture for real-time LEO-STIN communication systems.