Orbital edge computing (OEC) is crucial for supporting space intelligence applications within satellite networks. However, individual satellites face resource constraints, and implementing distributed processing techniques, such as federated learning (FL), across multiple satellites introduces significant scheduling complexity. To address these challenges, we first model the key factors influencing complex satellite networks, including satellite constellations, regional resource demands, inter-satellite communication and routing, energy consumption, and battery aging—a novel aspect invoked by OEC operations. We propose an adaptive aggregation method to fundamentally improve communication efficiency in OEC-based FL. To enhance scheduling performance, we formulate a unified optimization problem that jointly considers data partitioning, resource allocation, and aggregation transmission tasks within a decentralized partially observable Markov decision process (Dec-POMDP) framework. Furthermore, we introduce an episodic-phase-recalling reward shaping (EPRS) method to correlate the influences across these phases. Inspired by multi-task learning, we propose an efficient multi-agent reinforcement learning (MARL) algorithm featuring a multi-head actor-critic (MH-AC) network structure and task-equalized adaptation (TEA) technology, designed to optimize latency, energy consumption, network traffic, and battery aging. Extensive experiments validate the effectiveness of the proposed method, showing a 29.9% reduction in total training time, an 11.5% reduction in network traffic, and superior overall performance compared to rule-based methods.

Hybrid automatic repeat request (HARQ) is a potential technique for communication systems in non-terrestrial networks (NTN) to ensure reliability. However, the long round-trip time (RTT) within NTN diminishes the information timeliness when applying HARQ. Moreover, the absence of perfect channel state information (CSI) poses challenges to the precise design of HARQ strategies. In this paper, we introduce a novel coded HARQ scheme for NTN, in which the age of coded information (C-AoI) is proposed as a metric of information timeliness to guide the search for the optimal coded HARQ strategy. The rate-compatible inherited bits codeword construction (RCIC) is introduced for HARQ process to enhance the transmission efficiency. By applying Markov decision process (MDP) and Q-learning algorithm in the optimization, the obtained coded HARQ scheme significantly outperforms the existing schemes in terms of information timeliness and number of transmissions. The throughput of satellite-terrestrial links is also improved.

Non-geostationary orbit (NGSO) satellites, combined with beam hopping (BH) technology, play a vital role in wide-area communications, sensing, and positioning. However, the uneven distribution of ground users and the high mobility of NGSO satellites pose more significant challenges to multi-beam scheduling. This paper proposes a multi-NGSO BH scheduling framework considering the traffic demand of ground users and inter-satellite interference. We decompose the complex optimization problem into two sub-problems. First, we reallocate the satellite service cells to achieve a balanced distribution of satellite service coverage. Then, we propose a multi-NGSO BH algorithm based on soft actor-critic (SAC) to achieve real-time and traffic-driven beam scheduling. Simulation results show that the proposed algorithm outperforms other benchmarks in throughput and service failure rate (SFR).

The low earth satellite networks are envisioned to be an indispensable part of next-generation network due to the seamless Internet access. Deploying distributed computation into LEO satellite networks can decrease the latency of transmitting satellite-terrestrial computing jobs, which is crucial to expanding the service capability. Distributed computation depends on the exchange of data flows between worker nodes, as a type of concurrent and interrelated flows called coflow. However, the mesh-shaped topologies of LEO satellite networks make coflows prone to bandwidth competition on multi-hop links, which impedes the efficiency of distributed computation. In this paper, we formulated the multi-hop coflow scheduling process in LEO satellite networks as a routing and bandwidth allocation problem. Then, we simplified the problem to a coflow routing problem, and proposed the coflow routing greedy scheduling (CRGS) algorithm on the basis of the characteristics of multi-hop networks. Finally, we simulated in an SDN environment, where CRGS was deployed in an SDN controller. Compared with several existing algorithms, the CRGS algorithm is proved to reduce the coflow completion time (CCT) more effectively.

Network selection is one of the important techniques in cognitive radio networks (CRNs). With the development of network convergence technology and the popularity of heterogeneous networks, multiple primary CRNs interacting with multiple authorized networks are becoming possible, which can provide secondary users with more spectrum resources by network selection. Network selection is the key to spectrum sharing between CRNs and multiple primary networks. However, the spectrum sensing results, highly complex system state, and unsystematic research framework make the research of network selection very challenging. Traditional network selection algorithms are offline selection methods that are based on prior knowledge of primary networks. However, in the complex network environment, it is impossible to get prior knowledge from multiple primary networks, because the offline network selection methods lack efficiency. In order to meet these challenges, this article aims at improving the quality of service of cognitive users, and based on reinforcement learning method and the achievements of dynamic spectrum access of cognitive radio in single primary network environment, proposed a deep reinforcement learning based online network selection method of CRNs with multiple primary networks.