Source-linked AI summary
Topology-Aware Cooperative Beam-Hopping Scheduling for Efficient Resource Allocation in LEO Satellite Systems
Shuang Zheng, Xing Zhang, Quan Z. Sheng, Wenbo Wang
TL;DR
Dynamic topology, heterogeneous traffic, and local-only satellite observations complicate scalable cooperative beam hopping. The paper combines load-balanced associations, a two-phase topology encoder, and recurrent MAPPO for decentralized joint scheduling and power control. Simulations report stronger energy efficiency and throughput than MAPPO across traffic loads, with competitive fairness.
Problem
Dynamic topology and heterogeneous traffic demands complicate cooperative beam hopping, while rapidly changing coverage and interference patterns and local-only observations challenge scalable scheduling.
Source: §I · p. 1Method
A load-balancing association scheme induces a satellite–user graph for a two-phase GNN, whose topology representations inform recurrent MAPPO agents performing decentralized joint beam scheduling and power control.
Source: §I · p. 1, §III · p. 3Results
The proposed scheme consistently achieves higher energy efficiency and throughput than MAPPO as traffic load increases, outperforms MA-SAC and heuristic schemes overall, and maintains competitive fairness.
Source: §IV · p. 5Takeaways & Limitations
The topology-enhanced scheme achieves the highest reward among the compared methods, whereas using only satellite positions provides no effective improvement over MAPPO.
Source: §IV · p. 5The topology encoder augments MAPPO inputs without changing its objective or update rule, so the method inherits standard MAPPO’s convergence conditions and limitations.
Source: §III · p. 4
Abstract
from arXiv · showhide
The dynamic topology and heterogeneous traffic demands of multi-satellite systems present substantial challenges for beam-hopping (BH) resource management. This letter develops a topology-aware cooperative BH framework for multi-satellite systems under partial observability. A lightweight load-balancing scheme first determines the serving relationships, which are then exploited to construct an association-induced satellite-user graph. A two-phase graph neural network (GNN) extracts episode-level topology-aware structural identities for parameter-sharing agents, while a recurrent multi-agent proximal policy optimization (MAPPO) policy handles slot-level dynamics for joint beam scheduling and onboard power control. Simulation results demonstrate notable gains in energy efficiency, throughput, and fairness over the baselines.
I. INTRODUCTION
Cooperative beam hopping is motivated by dynamic traffic and constellation topology, while local observations and restrictive prior methods limit scalable coordination. The proposed framework uses load-balanced associations and topology-aware representations for decentralized scheduling.
I. INTRODUCTION
- Rapidly changing coverage overlaps and interference patterns, together with satellites’ local-only observations, complicate cooperative scheduling and limit centralized methods’ scalability and real-time applicability.Beam hopping offers flexible beam adaptation to diverse traffic demands, while single-satellite schemes cannot fully exploit constellation-wide resources.Source: §I · p. 1
- Prior approaches retain constraints: has restricted inter-satellite information exchange,, [6] use centralized gateway decisions, and have restrictive action spaces, and omits explicit topology information.The text associates centralized satellite–ground interactions with latency and signaling overhead that reduce scheduling flexibility.Source: §I · p. 1
- The framework balances heterogeneous traffic across visible satellites, builds an association-induced satellite–user graph, and uses a two-phase GNN with recurrent MAPPO for decentralized joint beam scheduling and power control.The agents share policy parameters and operate under centralized training and decentralized execution.Source: §I · p. 1
A. System Model
The system model considers forward-link cooperative beam hopping across multiple LEO satellites, with bounded active beams, heterogeneous traffic arrivals, and modeled interference and throughput.
A. System Model
- The modeled forward-link scenario covers Nu users across Nm beam positions served by Ns LEO satellites over Ts consecutive time slots.Each satellite has sets of activated beam positions and associated users.Source: A. System Model · p. 1
- Each satellite may activate at most K beams per slot, with K no greater than its available beam positions.The passage specifies the bound as K ≤ |Ms|.Source: A. System Model · p. 2
- User traffic follows a Poisson arrival process with user- and satellite-specific rate λu,s, and buffers retain only packets arriving within the previous Tttl slots.Tttl denotes packet time to live.Source: A. System Model · p. 2
- Hotspot users have two receiving antennas while users outside hotspots have one; transmit power is evenly divided across active beams and their subchannels, with full frequency reuse.The model assumes each beam is associated with at most one beam position.Source: A. System Model · p. 2
- The SINR model includes both intra-satellite and inter-satellite co-channel interference, and the model then defines user achievable data rate and throughput.Allocated bandwidth is identified as the bandwidth corresponding to the user’s throughput.Source: A. System Model · p. 2
B. Problem Formulation
The formulation separates user–satellite association from subsequent cooperative beam scheduling, using association as preprocessing and addressing load imbalance before scheduling.
B. Problem Formulation
- The problem is divided into user–satellite association as a preprocessing step and coordinated multi-satellite, multi-beam scheduling once serving relationships are established.The second phase exploits coordinated satellite decisions to enhance overall system performance.Source: B. Problem Formulation · p. 2
- The cooperative scheduling and onboard power-allocation subproblem targets a trade-off among energy efficiency, system throughput, and user-satisfaction fairness.The fairness metric is the standard deviation of service satisfaction across users, with lower values indicating more balanced satisfaction.Source: B. Problem Formulation · p. 2
- Concentrating users on a subset of satellites creates load imbalance that degrades cooperative scheduling, motivating preliminary load balancing.The association subproblem is framed around balancing satellite traffic loads.Source: 1) User-satellite association subproblem · p. 2
1) User-satellite association subproblem:
The association subproblem balances traffic loads across satellites while respecting binary serving assignments and satellite support limits.
1) User-satellite association subproblem:
- The association objective is to minimize traffic-load differences among satellites.The subproblem’s stated goal is load balancing before beam scheduling.Source: 1) User-satellite association subproblem · p. 2
- Each user–satellite association is represented by a binary variable indicating whether that satellite serves the user.The constraints also accommodate different antenna counts for ordinary and high-demand users.Source: 1) User-satellite association subproblem · p. 2
- The formulation bounds the maximum number of beam positions and users supported by each satellite.The passage names these limits as Nmax_s and U_s.Source: 1) User-satellite association subproblem · p. 2
- The broader cooperative scheduling and power-allocation objective balances energy efficiency, throughput, and user-satisfaction fairness.The fairness measure is the standard deviation of satisfaction across users, where lower values mean more balanced service.Source: 2) Multi-satellite coordination BH… · p. 2
2) Multi-satellite coordination BH subproblem:
The subproblem couples beam scheduling and onboard power allocation under constraints, with service satisfaction used to assess fairness.
2) Multi-satellite coordination BH subproblem:
- The standard deviation of users' service-satisfaction ratios is the fairness metric, with lower values indicating more balanced service.The ratio compares cumulative delivered data with cumulative traffic demand over the same period.Source: 2) Multi-satellite coordination BH… · p. 2
- Each satellite may serve at most K beam positions simultaneously, while onboard power is bounded by prescribed minimum and maximum levels.The objective's three weights sum to one.Source: 2) Multi-satellite coordination BH… · pp. 2–3
- The association procedure sorts users by demand and assigns them in order of satellite load, subject to its stated association checks.High-demand users may be associated with two satellites; other users with one.Source: 2) Multi-satellite coordination BH… · p. 3
- P2 is a nonconvex mixed-integer nonlinear program because it combines binary beam scheduling, continuous power control, and interference-dependent rates.Each satellite's power is constrained between ηPmax and Pmax.Source: 2) Multi-satellite coordination BH… · p. 3
III. USER-SATELLITE ASSOCIATION AND TOPOLOGY-AWARE COOPERATIVE BEAM SCHEDULING
The framework first establishes user–satellite associations and then uses partial-observation agents with topology-aware representations to coordinate beam scheduling and power control. Its graph encoder captures local service context and cross-satellite structure, while recurrent MAPPO handles slot-level decisions.
III. USER-SATELLITE ASSOCIATION AND TOPOLOGY-AWARE COOPERATIVE BEAM SCHEDULING
- Under bounded regional overlap, Algorithm 1 has complexity O(NuNs log Ns); accepted associations pass visibility, capacity, cardinality, and angular-separation tests.The feasibility constraints are preserved whenever sufficient feasible candidates exist.Source: §III · p. 3
- The beam-scheduling subproblem is a partially observable Markov decision process in which each satellite acts as an autonomous agent.Each satellite observes local information about its served users.Source: §III · p. 3
- Under CTDE, recurrent MAPPO uses a centralized critic during training and a shared policy for decentralized joint beam scheduling and onboard power control.The PPO clipped surrogate objective is used to update the policy.Source: §III · p. 3
- Phase I aggregates associated users' traffic and link features into local satellite embeddings, and Phase II propagates them to encode cross-satellite structural relations.Both phases are needed to represent local service context and inter-satellite coupling.Source: §III · p. 3
- The association-induced satellite–user graph uses traffic-arrival features, satellite location and altitude, and initialization-time channel gains on visible-user links.The graph is constructed at the beginning of each episode.Source: §III · pp. 3–4
- The graph encoder runs once per episode, and shared actors execute in parallel with per-satellite inference cost independent of the number of satellites.The topology encoder augments policy and value inputs without changing MAPPO's objective or update rule.Source: §III · p. 4
- The topology-augmented method inherits the convergence conditions and limitations of standard MAPPO.The encoder does not modify MAPPO's objective or update rule.Source: §III · p. 4
IV. NUMERICAL RESULTS
Across traffic demands, the proposed scheme improves energy efficiency and throughput over MAPPO, while the association and convergence comparisons also favor topology-aware scheduling and load balancing.
IV. NUMERICAL RESULTS
- The simulation uses a Ka-band system with eight satellites, each supporting up to 80 users, 60 beam positions, and 16 simultaneous beams.Each satellite has 250 W onboard power and 200 MHz bandwidth; traffic arrivals follow a Poisson process.Source: §IV · p. 4
- Algorithm 1 yields almost uniform load distribution close to SCIP; conventional schemes have load gaps of 0.350 and 0.325.Against the two conventional association strategies, EE increases by 57.14% and 48.42%, throughput by 79.27% and 69.52%, and service-satisfaction standard deviation falls from 0.41 to 0.36.Source: Figure 4 · p. 5, §IV · p. 5, Table II · p. 5
- As traffic load increases, the proposed scheme achieves higher EE and throughput than MAPPO, outperforms MA-SAC and heuristic schemes overall, and maintains competitive fairness.Greedy delay maintains relatively stable fairness by prioritizing waiting delay.Source: Figure 5 · p. 5, §IV · p. 5
- The proposed scheme consistently achieves the highest reward; ST-MAPPO is nearly level with MAPPO, while F-MAPPO converges to a much lower reward.The reward combines EE, throughput, and fairness; F-MAPPO uses full-power allocation, whereas MAPPO jointly schedules beams and controls power.Source: Figure 3 · p. 4, §IV · p. 5
V. CONCLUSIONS
The conclusion presents a topology-aware cooperative beam-hopping framework that combines load-balanced associations, graph-derived structural identities, and decentralized joint scheduling and power control. Simulations show improvements in energy efficiency, throughput, and fairness.
V. CONCLUSIONS
- The framework uses load-balanced serving relationships to form an episode-specific satellite–user graph, then applies two-phase message passing to generate structural identities for parameter-sharing agents.Combined with slot-level local observations, the representations support decentralized joint beam scheduling and onboard power control under CTDE.Source: §V · p. 5