Deep Convolutional Linear Precoder Neural Network for Rate Splitting Strategy of Aerial Computing Networks

Aerial computing networks are facing the challenge of massive node access, where user devices generally have stringent latency and robustness requirements. Rate Splitting Multiple Access (RSMA) is a general and robust multiple access framework for the aerial computing communication architecture, which splits each user's message into common and private parts and superposes the common message and the private message for transmission to manage interference among multiple users. We propose a simple deep convolutional neural network to implement the linear precoder design for RSMA in aerial computing networks to reduce the average optimization time and thus improve the massive communication efficiency. And we also propose two patterns of combining the linear precoder design model with the Channel State Information (CSI) feedback self-encoder model, one use the CSI feedback model decoder output as the input of the precoder model, and the other is to extract the features directly from the feedback codeword without recovering the complete CSI at the base station side, which can help reduce the computational effort and time of the optimization solution. Simulations show that the proposed models are close to the communication rate of the traditional strategy in optimizing the linear precoder but have a substantially higher time efficiency.