Joint AoI and Handover Optimization in Space-Air-Ground Integrated Network

Zifan Lang, Guixia Liu, Geng Sun, Jiahui Li, Jiacheng Wang, Weijie Yuan, Dusit Niyato, Dong In KimarXiv 2025 · DOI · arXiv

Abstract

Maintaining data freshness while ensuring connection stability is a critical challenge in Space-Air-Ground Integrated Networks (SAGIN). This paper proposes a three-layer architecture utilizing High-Altitude Platforms (HAPs) as intelligent relays between fast-moving Low Earth Orbit (LEO) satellites and ground users. To optimize the Age of Information (AoI) and minimize handover frequency, the authors introduce DD3QN-AS, a deep reinforcement learning algorithm enhanced by Transformer encoders and Diffusion Models. Experimental results show that the proposed method reduces handover frequency by 15% and improves data freshness compared to existing state-of-the-art baselines.

Introduction

Low Earth Orbit (LEO) satellites are pivotal for global connectivity, offering low latency and extensive coverage. However, their high orbital velocity presents two significant challenges: temporal discontinuity and handover instability. As satellites move rapidly out of range, the Age of Information (AoI)—a metric for data freshness—increases, and frequent handovers between satellites create signaling overhead and "ping-pong" effects. Existing optimization methods often struggle with the high-dimensional, non-convex nature of simultaneous transmit power allocation and satellite selection.

System Architecture

The authors propose a three-layer SAGIN architecture: LEO Satellites → HAP → Ground Terminals. In this model, the HAP serves as a quasi-stationary relay, bridging the gap between the space and ground segments. It utilizes high-capacity Free-Space Optical (FSO) links for satellite communication and robust Radio Frequency (RF) links for ground user connectivity.

Figure 1: The three-layer SAGIN architecture featuring LEO satellites, a HAP relay with a data buffer, and ground users.

The HAP includes a data buffer to manage intermittent connectivity, allowing it to store packets from passing satellites and transmit them to ground users at optimal intervals, effectively smoothing the AoI.

The DD3QN-AS Algorithm

To manage the dynamic environment, the paper introduces the DD3QN-AS algorithm. This framework integrates several advanced machine learning techniques to optimize network performance:

Figure 2: The DD3QN-AS algorithm architecture, highlighting the State Transformer Encoder, Diffusion Model-based Latent Prompt Generative Module, and Action Decomposition.
  • Action Decomposition: The optimization problem is split into discrete and continuous components. The RL agent handles discrete satellite selection, while a traditional convex optimization solver manages continuous power allocation.
  • State Transformer Encoder (STE): By leveraging Transformer-based architectures, the agent captures long-term temporal dependencies in satellite orbits and channel fluctuations more effectively than standard neural networks.
  • Diffusion Model-based Latent Prompt (DLPG): Borrowing from generative AI, a diffusion model "denoises" state-action representations. This makes the agent robust against the stochastic noise inherent in satellite-to-ground communications.

Experimental Evaluation

The study utilizes a high-fidelity simulation environment featuring 10 LEO satellites (altitudes 500km–1,800km), one HAP (20km), and 10 ground users. Orbital dynamics were modeled using Keplerian elements, while FSO and RF channels followed Gamma-Gamma and Nakagami-m distributions.

Key Results

  • Enhanced Data Freshness: DD3QN-AS reduced the average AoI by approximately 1.7% compared to the best-performing baseline (DDQN).
  • Improved Stability: The algorithm achieved a 15% reduction in satellite handover frequency, significantly lowering signaling overhead.
  • Faster Convergence: The inclusion of the Transformer encoder allowed the model to reach a stable steady-state reward faster than standard DRL methods.

Conclusion and Future Work

The integration of HAPs and generative denoising techniques provides a robust solution for SAGIN optimization. While the DD3QN-AS algorithm shows significant improvements in stability and freshness, future work is needed to address the computational overhead of diffusion models for real-time deployment on power-constrained hardware. Additionally, scaling the approach to "mega-constellations" with thousands of satellites remains an open research challenge.

Summary updated

Abstract

Despite the widespread deployment of terrestrial networks, providing reliable communication services to remote areas and maintaining connectivity during emergencies remains challenging. Low Earth orbit (LEO) satellite constellations offer promising solutions with their global coverage capabilities and reduced latency, yet struggle with intermittent coverage and limited communication windows due to orbital dynamics. This paper introduces an age of information (AoI)-aware space-air-ground integrated network (SAGIN) architecture that leverages a high-altitude platform (HAP) as intelligent relay between the LEO satellites and ground terminals. Our three-layer design employs hybrid free-space optical (FSO) links for high-capacity satellite-to-HAP communication and reliable radio frequency (RF) links for HAP-to-ground transmission, and thus addressing the temporal discontinuity in LEO satellite coverage while serving diverse user priorities. Specifically, we formulate a joint optimization problem to simultaneously minimize the AoI and satellite handover frequency through optimal transmit power distribution and satellite selection decisions. This highly dynamic, non-convex problem with time-coupled constraints presents significant computational challenges for traditional approaches. To address these difficulties, we propose a novel diffusion model (DM)-enhanced dueling double deep Q-network with action decomposition and state transformer encoder (DD3QN-AS) algorithm that incorporates transformer-based temporal feature extraction and employs a DM-based latent prompt generative module to refine state-action representations through conditional denoising. Simulation results highlight the superior performance of the proposed approach compared with policy-based methods and some other deep reinforcement learning (DRL) benchmarks.
BibTeX
@misc{lang2025joint,
  author = {Zifan Lang and Guixia Liu and Geng Sun and Jiahui Li and Jiacheng Wang and Weijie Yuan and Dusit Niyato and Dong In Kim},
  title = {{Joint AoI and Handover Optimization in Space-Air-Ground Integrated Network}},
  year = {2025},
  eprint = {2509.12716},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.12716},
}
Figure