Semantic Communication-Aware End-to-End Routing in Large-Scale LEO Satellite Networks

Enhanced by inter-satellite links, large-scale satellite networks (SNs) hold promise to deliver low-latency services globally. However, given the scarcity of available spectrum and the capacity limitations of Shannon’s classic information theory, supporting the ever-growing multimedia communication services poses a challenge for SNs. A promising direction is to employ semantic communication (SC), which utilizes artificial intelligence (AI) to extract and transmit the "meaning" of raw data, thereby reducing bandwidth consumption. We identify that SC requires the raw data to be transmitted to a satellite equipped with the AI encoder supporting the same knowledge base for successful decoding at the destination. This advanced functionality renders traditional routing schemes such as contact graph routing (CGR) not seamlessly suitable. Hence, we develop the SC-aware routing (SCR) strategy aimed at optimizing SC’s delay and bandwidth consumption. Firstly, we define the problem using integer linear programming (ILP) and observe its intractability in practice. Then, we use the temporal graph to represent diverse knowledge bases, semantic encoders and transmission resources of SNs, and design SCR based on it. Simulations conducted on the Starlink constellation show that SCR runs much faster than the ILP solver. Additionally, we verify that SC can bring significant performance gains in terms of higher data throughput and lower average end-to-end delay when compared to CGR.