Recurrent Neural Network Data Predistortion for Transparent Satellite Channels

Citation
IEEE Global Communications Conference 2024
Contribution to Conference
IEEE Global Communications Conference 2024
Publisher Link
https://globecom2024.ieee-globecom.org/program/technical-program-day-2
Satellite transponders operate with traveling wavetube amplifiers (TWTAs) which nonlinearly distort the transmit signal, particularly when transmission power close to amplifier saturation is desired. Hence, the concatenation of pulse shaping filter, input multiplexer (IMUX) filter, TWTA, output multiplexer (OMUX) filter and receive filter causes distorted decision regions and severe intersymbol interference (ISI) at the receive filter output. The standards DVB-S2 and DVB-S2X recommend setups for transparent satellite channels. Here, the distortion is traditionally compensated using a lookup tablebased centroid data predistortion algorithm. In this work, we propose an improved approach for data predistortion based on bidirectional recurrent neural networks (BRNNs). We show the superiority of BRNN data predistortion over centroidbased predistortion by comparing information rates and bit error rates on recommended setups from the DVB-S2 standard. Furthermore, we show how BRNN predistortion enables ISI compensation for high order modulations from DVB-S2X, where the traditional lookup table approach fails due to infeasible memory requirements.
Subjects
MLE@TUHH