Beam Hopping: Dynamic Satellite Resource Allocation

• Beam hopping is a time-domain strategy for multi-beam satellite networks, activating a subset of beams in each time slot to dynamically match varying ground traffic demands.
• Mathematical formulations employ mixed-integer nonconvex models, conflict graphs, and optimization heuristics to balance throughput, fairness, and interference mitigation.
• Applications span high-throughput satellites, NGSO networks, and joint terrestrial-satellite systems, improving broadband access, positioning, and signaling efficiency.

Beam hopping is a time-domain resource allocation technique in multi-beam satellite communication systems. Rather than simultaneously illuminating all spot beams, only a subset of available beams is activated in each time slot according to a hopping schedule, dynamically matching resource supply to non-uniform and time-varying ground traffic demand, while mitigating interference and reducing hardware complexity. Beam hopping frameworks underpin high-throughput satellites (HTS) and non-geostationary orbit (NGSO) networks, supporting not only broadband access but also random access, positioning, and joint terrestrial-satellite operations.

1. Beam Hopping Fundamentals and System Models

Beam hopping (BH) exploits the time, space, and—in advanced variants—power/frequency degrees of freedom of modern satellite payloads, particularly those with phased array antennas and digital payload units. At its core, BH partitions the service area into a set of ground cells or beams, only a subset of which ($K \ll N$ for $N$ total beams) is illuminated at any one instant. Illumination patterns are dynamically scheduled across a finite repetition window ("hopping frame") to favor regions of high demand or critical functions (e.g., common signaling).

Mathematically, BH is characterized by the binary variables $x_{b,t}\in\{0,1\}$, where $x_{b,t} = 1$ denotes beam $b$ is active in slot $t$. The satellite's instantaneous resource allocation is subject to per-slot and aggregate constraints (e.g., maximum beams per slot, power budgets, interference constraints). Optimization objectives range from throughput maximization and backlog minimization to fairness, power efficiency, and QoS compliance (Sun et al., 2024, Kibria et al., 2019, Zhang et al., 2022, Artiga et al., 16 Dec 2025, Kibria et al., 2022).

In NGSO and LEO architectures, beam hopping further enables adaptation to the fast topological changes imposed by satellite movement, while in GEO systems, it reduces payload costs by allowing high-power amplifiers and RF chains to be shared across beams using TDM multiplexing (Sun et al., 2024, Kibria et al., 2022, Wang et al., 2024).

2. Mathematical Formulation and Optimization

Beam hopping problems are inherently mixed-integer and nonconvex. Classical formulations encode variables for beam-slot illumination ($x_{b,t}$), physical layer controls (power $p_{b,t}$, rate selection, carrier assignment), user associations, and demand backlogs (Sun et al., 2024, Wang et al., 2021, Ha et al., 2022). Representative objectives and constraints include:

• Demand-Backlog Minimization: $\min \sum_c (D_c^{f,t} - Q_c^{f,t})^2$ captures the instant mismatch between supplied and requested service per cell (cf. Lyapunov drift formulations) (Sun et al., 2024).
• Resource Constraints:
  • $\sum_{b} x_{b,t} \leq K$ (active beam limit),
  • $N$0 (per-slot beam hardware constraint),
  • $N$1, ensuring each beam serves at most one cell per slot (Sun et al., 2024).
• Interference and Feasibility Constraints:
  • Preclusion of adjacent/interfering beam activations using conflict graphs or distance-limited scheduling (Sun et al., 2024, Zhang et al., 2022),
  • Power allocation limits ($N$2), and
  • SINR or rate constraints per user (Ha et al., 2022, Wang et al., 2024).

The NP-hardness of general formulations is established via reductions from multidimensional matching and scheduling problems (Wang et al., 2021). Recent trends include explicit integration of multi-objective fairness metrics, e.g., max-min rate matching (Kibria et al., 2019), and queue/delay-based objectives (Zhao et al., 2024, Xie et al., 4 Jan 2025).

3. Algorithmic Approaches for Beam Hopping

3.1 Conflict Graph and Greedy WMIS Solvers

Conflict graph scheduling encodes mutual incompatibilities among beam–slot assignments, mapping the resource allocation problem to finding weighted maximum independent sets (WMIS) in constructed graphs. Vertices represent feasible service assignments, edges encode resource or interference constraints, and vertex weights are designed to drive backlog reduction or throughput (Sun et al., 2024). Greedy WMIS selection heuristics, prioritized by weight ratios that balance individual backlog relief against neighbor impact, yield computational complexity of $N$3 over $N$4 slots, providing order-of-magnitude improvements over naive greedy or exact (exponential) solvers while nearly saturating interference-free capacity (Sun et al., 2024).

3.2 Combinatorial and Integer Programming

Beam hopping pattern optimization is amenable to ILP/MILP and MINLP formulations. Notable examples include max-min fairness integer programs for cluster-based hopping (CH), where clusters are groups of beams concurrently illuminated with full frequency reuse and intra-cluster MMSE precoding (Kibria et al., 2019), and MILP models for joint BH and carrier aggregation (CA), which maximize user/beam satisfaction ratios under time–space–frequency constraints (Kibria et al., 2022). Primal heuristics and relaxations (e.g., reweighted $N$5 sparsity for beam actuation cost (Ha et al., 2022)) facilitate scalable solutions.

3.3 Monte Carlo Tree Search, Bisection, and Alternating Optimization

For massive-scale HTS, hybrid computational frameworks exploit both rapid greedy algorithms for provisional patterns and more sophisticated search methods (e.g., Monte Carlo Tree Search, MCTS-BH) for high-quality offline refinement. MCTS-BH, together with sliding window-based interference estimation and action pruning, produces nearly optimal illumination patterns for $N$6 cells within practical timescales, outperforming DRL-based approaches for large problem sizes (Yang et al., 10 Dec 2025). For uplink grant-free random access, alternating optimization (collision avoidance allocation via bisection, pattern design via ADMM with $N$7-box relaxation) efficiently solves min-max success probability problems (Jeon et al., 5 Aug 2025).

3.4 DRL and Game-Theoretic Schemes

Multi-agent and distributed reinforcement learning (A3C, MADDPG, PPO) have been successfully applied to multi-satellite beam hopping and power allocation in LEO constellations, enabling operation in high-dimensional, dynamic environments with overlapping coverage, time-varying demand, and complex interference (Zhao et al., 2024, Xie et al., 4 Jan 2025). Deep RL architectures with hybrid (discrete+continuous) action spaces jointly optimize beam pattern scheduling and power allocation, balancing throughput, delay, and fairness metrics. Game-theoretic approaches, such as potential game-based beam scheduling paired with penalty interior-point methods for power optimization, achieve fast convergence to Nash equilibria with low computational overhead (Zheng et al., 2023).

3.5 Joint Designs and Extensions

Advanced frameworks integrate BH with other spectral and access techniques:

• Joint BH and NOMA (both code- and power-domain), where resource assignment is formulated as a nonconvex MINLP, handled via fractional programming and Dinkelbach-type variable techniques to minimize the demand-capacity gap (Zhang et al., 2024, Wang et al., 2021);
• Cluster hopping with intra-cluster full-frequency MMSE precoding, achieving demand-adaptive, high spectral efficiency (Kibria et al., 2019);
• Joint BH–CA, solved as a bi-objective MILP to optimize fairness and demand matching (Kibria et al., 2022);
• Joint beamforming (hybrid or fully digital) and illumination pattern design for sum-rate maximization under hardware constraints (Wang et al., 2024).

4. Performance Evaluation and Practical Guidelines

Extensive numerical experiments across system scales and operating conditions demonstrate that:

• Conflict-graph WMIS-based and advanced RL/decomposition methods achieve service satisfaction and queue stability approaching the interference-free or exact solution benchmarks, with computational complexity orders suitable for on-board or near-real-time scheduling (Sun et al., 2024, Zhao et al., 2024, Yang et al., 10 Dec 2025).
• Greedy and round-robin methods, while fast, suffer up to $N$8–$N$9 higher queue/backlog penalties, poor fairness, and degraded delay performance, highlighting the importance of interference-aware, demand-adaptive scheduling (Sun et al., 2024, Zhang et al., 2022).
• Hybrid computation frameworks (greedy+MCTS) provide millisecond-layer provisional responses with asynchronous global refinement, enabling real-time operation in HTS with hundreds to thousands of beams (Yang et al., 10 Dec 2025, Gaudry et al., 2024).
• The integration of BH with NOMA, carrier aggregation, or joint cluster hopping augments system throughput, fairness (user and beam Jain indices up to 0.99), and power efficiency over conventional OMA/BH-only regimes (Kibria et al., 2019, Kibria et al., 2022, Zhang et al., 2024).

Table: Comparative Metrics for Selected Schemes

Algorithm/Framework	Objective/Metric	Gain vs. Baseline
Conflict-graph WMIS (LEO BH)	Avg. queue length, drift	$x_{b,t}\in\{0,1\}$0–$x_{b,t}\in\{0,1\}$1
Tyche (MCTS-BH)	Throughput, comp. time	$x_{b,t}\in\{0,1\}$2, $x_{b,t}\in\{0,1\}$31s @ 127 cells
RL (A3C, MADDPG, PPO)	Throughput, delay, fairness	$x_{b,t}\in\{0,1\}$4 load-gap, $x_{b,t}\in\{0,1\}$5 delay
Joint BH–NOMA	Demand-capacity mismatch	$x_{b,t}\in\{0,1\}$6 (E-JPBT vs. OMA)
BH–CA MILP	Unused capacity, Jain index	$x_{b,t}\in\{0,1\}$7, $x_{b,t}\in\{0,1\}$8
Beam scheduling + game theoretic	SOD, fairness, complexity	$x_{b,t}\in\{0,1\}$9 throughput, $x_{b,t} = 1$0 GA
ℓ₂-Box ADMM (random access BH)	Min success prob. (CDF)	$x_{b,t} = 1$1–$x_{b,t} = 1$2 at 30th percentile

5. Extensions: Positioning, Signaling, System Integration

Cooperative Beam Hopping for Positioning

For high-precision positioning in ultra-dense LEO networks, cooperative BH adapts the set of beams illuminating a user across neighboring satellites, minimizing the aggregated Cramér-Rao lower bound (CRLB) on TDOA localization error. Flexible BH control algorithms iteratively solve association, layout (via Voronoi partitioning), and power allocation via convex SDP, achieving substantial CRLB reductions versus communication-centric designs (Wang et al., 2021).

3GPP NTN Common Signaling and Earth-Fixed BH

BH is also employed to optimize downlink broadcast of 3GPP NR NTN common signaling (SSB, PDCCH, etc.) over large footprints under EIRP constraints. Pragmatic earth-fixed grid designs with phased-array beams and scheduled hopping patterns minimize the number of beams (e.g., 451 vs. 1723), maximize signaling efficiency (up to 80.6% slot utilization), and ensure 100% user coverage, nearly saturating paging/RA capacity under stringent periodicity (Artiga et al., 16 Dec 2025).

Joint Beamforming and Illumination Pattern Optimization

The interplay between BH and (hybrid) beamforming is critical for sum-rate maximization under hardware-limited RF chains. Alternating optimization (beam patterns and beamformers) and fractional programming with quadratic transforms yield near-optimal throughput, with extra hopping slots conferring greater gains than increasing simultaneous beams per slot given fixed beam-position counts (Wang et al., 2024).

6. Design Insights, Trade-Offs, and Future Directions

Across architectures and applications, key BH design insights include:

• Conflict-graph and decoupled/staged algorithms strike effective trade-offs between complexity and schedule optimality, often operating within $x_{b,t} = 1$3–$x_{b,t} = 1$4 of interference-free or optimal benchmarks.
• Virtual queueing and drift-based Lyapunov frameworks yield provably stable policies with explicit control over backlog versus handover/complexity (Sun et al., 2024).
• Slot-by-slot scheduling, while potentially suboptimal globally, offers essential computational scalability for large-scale, fast-moving constellations (Yang et al., 10 Dec 2025, Gaudry et al., 2024).
• Advanced RL/game theoretic approaches can gracefully handle spatial-temporal demand heterogeneity and support multi-satellite cooperation, though training/sample efficiency and robustness to rapidly changing environment states remain open challenges (Zhao et al., 2024, Xie et al., 4 Jan 2025, Zheng et al., 2023).
• Joint optimization with other multiple-access or aggregation technologies (e.g., NOMA, CA, full-reuse precoding) consistently outperforms BH or OMA baselines on throughput, fairness, and power—demonstrated by analytic results, simulation, and hardware-constrained emulations (Zhang et al., 2024, Kibria et al., 2022, Kibria et al., 2019).
• ILP/MILP-based approaches remain unscalable beyond $x_{b,t} = 1$5 beams but provide meaningful optimality benchmarks; decomposition-by-powers-of-two and compressed sensing techniques enable near-real-time BH pattern construction for HTS with over $x_{b,t} = 1$6 beams with $x_{b,t} = 1$720% capacity-matching error (Gaudry et al., 2024).

Emerging research directions focus on integration of online learning for rapid traffic adaptation, joint terrestrial–satellite spectrum and interference management, multi-agent/multi-satellite coordination, and robust operation under imperfect or delayed CSI. The modular, optimization-driven structure of BH frameworks also aligns naturally with future digital regenerative payload architectures and in-situ on-board AI/ML inference.

---

Principal References (arXiv IDs):

(Sun et al., 2024, Kibria et al., 2019, Zhao et al., 2024, Yang et al., 10 Dec 2025, Wang et al., 2021, Kibria et al., 2022, Artiga et al., 16 Dec 2025, Zhang et al., 2022, Zheng et al., 2023, Xie et al., 4 Jan 2025, Wang et al., 2021, Wang et al., 2024, Ha et al., 2022, Jeon et al., 5 Aug 2025, Gaudry et al., 2024, Zhang et al., 2024)