Quantum-Assisted Design of Space-Terrestrial Integrated Networks

TL;DR

This work addresses end-to-end design of Space-Terrestrial Integrated Networks (STINs) by formulating SSP, GSP, and SAP as a sequence of graph-based combinatorial problems. It leverages a neutral-atom quantum processor (Aquila) to solve SSP mapped to $MWIS$ via the Quantum Adiabatic Algorithm ($QAA$), with a Distance Encoder Network (DEN) for hardware-aware unit-disk embeddings and subsequent postprocessing. Benchmarking on 165 remote-region instances shows that $QAA$ solutions closely match classical exact solvers and outperform greedy baselines, with downstream GSP and SAP remaining robust to SSP differences. The results indicate that quantum optimization can be competitive for larger or more complex STIN subproblems and are enabled by a practical hybrid quantum-classical workflow, including neural embedding and classical refinement. The work also introduces a detailed hardware-aware embedding and scheduling framework that could inform future quantum-accelerated network design problems.

Abstract

Achieving ubiquitous global connectivity requires integrating satellite and terrestrial networks, particularly to serve remote and underserved regions. In this work, we investigate the design and optimization of Space-Terrestrial Integrated Networks (STINs) using a hybrid quantum-classical approach. We formalize three key combinatorial optimization problems: the Satellite Selection Problem (SSP), the Gateway Selection Problem (GSP), and the Spectrum Assignment Problem (SAP), each capturing critical aspects of network deployment and operation. Leveraging neutral-atom quantum processors, we map the SSP onto a Maximum Weight Independent Set problem, embedding it onto the Aquila platform and solving it via the Quantum Adiabatic Algorithm (QAA). Postprocessing ensures feasible solutions that guide downstream GSP and SAP optimization. Benchmarking across 165 realistic remote regions shows that QAA solutions closely match classical exact solvers and outperform greedy heuristics, while subsequent GSP and SAP outcomes remain largely robust to differences in initial satellite selection. These results demonstrate that quantum optimization achieves performance broadly comparable to classical approaches for end-to-end STIN design, with rare instances where it can even surpass state-of-the-art solvers. This suggests that, while not yet consistently superior, quantum methods may offer competitive advantages for larger or more complex instances of the underlying combinatorial subproblems.

Figures (6)

• Figure 1: Relative improvement of QAA compared to the greedy heuristic (orange) and exact solver (blue) for each graph instance.
• Figure 2: Scatter plot showing the number of non-independent bitstrings as a function of the number of vertices.
• Figure 3: Comparison of GSP and SAP objective distributions when varying the SSP input satellites obtained from either the exact solver or the quantum method (QAA).
• Figure 4: DEN architecture for a 5-vertex embedding example.
• Figure 5: Adiabatic pulse schedule on Aquila for a three-vertex graph. Global Rabi and detuning sweeps are overlaid with local detuning offsets (color-coded by vertex weight).