Joint Training of Predistortion, Power Back-Off and Constellation for Satellite Power Amplifiers Using Neural Networks

The deployment of satellite mega constellations may enable global coverage, even for direct transmission from satellite to handheld device. Such transmissions come with increased demands in power efficiency. The traveling wave-tube amplifier (TWTA) in satellite payloads fundamentally limits the transmit power and causes distortions to the transmit signal when power efficient transmission close to amplifier saturation is desired. This work introduces a novel joint training paradigm of constellation, amplifier power back-off and data predistortion to maximize the throughput of single carrier transmission over transparent satellite links. The joint design is enabled by means of communication autoencoders, where transmitter and receiver components are adapted together to achieve the lowest bit error rate (BER). We show how constrained constellation optimization can improve performance on selected configurations from the broadcasting standard DVB-S2X. Results are presented in terms of coded BER and information rates.