Coded Semantic-Aware Coordinated Transmission in Cohesive Clustered Satellite Systems: An Incremental MADRL Approach

Cohesive clustered satellite (CCS) systems offer a promising solution for high-capacity and continuous communication through multi-satellite coordinated transmission (MSCT). However, existing MSCT schemes compromise the semantic awareness of transmitted data, limiting information utility. In this paper, we propose a coded semantic-aware coordinated transmission (CSCT) scheme to improve both semantic awareness and transmission reliability in scalable CCS systems. The CSCT scheme uses concatenated coding with T-order codebook to enable satellites to encode semantic messages and coordinate their transmission to the ground station. We design a randomized Kaczmarz based successive cancellation, followed by prioritized joint decoding (RKZ-SCPJD) decoder to recover satellite signals with low complexity. Next, we introduce the utility of information (UoI) metric to quantify the utility of semantic-aware transmission and formulate an optimization problem for joint scheduling of spot beams and semantic messages, solved using a multi-agent deep reinforcement learning (MADRL) approach. Additionally, by leveraging incremental learning, we extend the MADRL approach to an incremental MADRL framework to accommodate varying numbers of satellites due to mobility or scalability. Extensive simulation results demonstrate the accuracy of theoretical derivations, and highlight the superior performance of our CSCT scheme over existing schemes across various settings, particularly in enhancing UoI and enabling scalable MSCT.