Load-Balanced and Congestion-Aware Routing for LEO Laser Satellite Networks Based on Deep Reinforcement Learning

GT-PAR: Graph Transformer-Aided Traffic Prediction and Adaptive Routing for Dynamic LEO Satellite Networks

Low Earth Orbit (LEO) satellites are essential for 6G non-terrestrial networks due to their global coverage and low-latency communication. However, the highly dynamic topology and uneven traffic distribution cause routing inefficiencies. This letter proposes a Graph Transformer–aided Traffic Prediction and Adaptive Routing (GT-PAR) scheme to capture topology-dependent spatial coupling and long-range link-utilization dynamics. The ground segment periodically broadcasts lightweight link-utilization predictions, and the satellites select the next routing hops using the congestion-aware cost analyzed in Lemmas 1 and 2. The simulation results show that GT-PAR can significantly reduce the packet loss and end-to-end delay when compared with representative routing schemes.

Distributed Routing for LEO Satellite Networks: A Multi-Agent Deep Reinforcement Learning Approach With State Information Lag

Multi-agent deep reinforcement learning (MADRL) offers a promising solution for routing in low Earth orbit (LEO) satellite networks. However, large inter-satellite propagation delays lead to severe state information lag in agent interactions, giving rise to decision biases and degraded routing timeliness. To this end, this paper proposes a distributed routing algorithm named time-aware prediction and dynamic attention routing (TAP-DAR). Specifically, it constructs a delay compensation model that incorporates ephemeris data and queue prediction to generate near real-time neighbor state estimates. In addition, a multi-head attention fusion mechanism considering temporal reliability is designed to achieve adaptive aggregation of asynchronous neighbor states. Simulation results demonstrate that across various constellation configurations and network load conditions, the proposed algorithm achieves a maximum reduction of 16.16% in end-to-end (E2E) latency, an average decrease of nearly 30% in packet loss rate, and a maximum improvement of 19.41% in throughput compared to the baseline. Moreover, it substantially curtails communication overhead by more than 90% relative to the global state flooding mechanism.

Reliable Session-Oriented Multi-Path Routing for LEO Satellite Networks: A Multi-Agent Learning Approach

Link-Aware Routing in Multi-Tier LEO Mega-Constellations

In multi-tier low-Earth orbit (LEO) mega-constellations, the mobility of satellites across different orbital altitudes leads to dynamic changes in network topology and inter-satellite link (ISL) states, including ISL duration and capacity. These changes often result in unstable connectivity and disrupted end-to-end data transmission. To address this issue, we formulate a routing optimization problem to determine the optimal ISL path between two end users by maximizing the average ISL utility, considering both ISL duration and capacity. To solve this problem, we propose a routing method that first prunes unstable ISLs using a graph neural network (GNN) and then selects optimal end-to-end paths on the pruned graph using a heuristic routing algorithm. Simulation results demonstrate that the proposed algorithm achieves higher throughput and a lower packet loss rate compared to benchmark methods.

Hierarchical Scheduling for Federated Learning Optimization over Deterministic Satellite Networks

Attention-Enhanced DRPO Reinforcement Learning for SFC Routing and Deployment in Satellite-Terrestrial Integrated Networks

To address service function chain routing challenges in satellite-terrestrial integrated networks, an attention-enhanced Direct Reward Policy Optimization(DRPO) method is proposed. Simulations show the proposed method achieves higher utility and admission-rate than Greedy with similar delay.