Free-Space Quantum Networks and Optimized Fiber-Reinforcement

Abstract: Free-space quantum communication provides a flexible complement to fiber-based quantum networks, but its point-to-point capacity is fundamentally limited by diffraction, atmospheric extinction and beam wandering induced by turbulence. In this work, we study the end-to-end performance of large-scale free-space quantum networks connecting randomly distributed fixed or mobile users, modelled as Waxman random graphs. We derive the mean network capacity, edge consumption and connectivity phase transitions for both single-path and multi-path (flooding) routing. We also study router-centered star networks, deriving the full distribution of end-to-end capacities as a function of the router's coverage radius. We then consider how performance may be improved by reinforcing free-space networks with a small number of optimally placed fiber-based backbone nodes. We prove that any optimal backbone configuration must correspond to a capacity-maximizing Voronoi tessellation of the network region, and show that this can be efficiently approximated by a centroidal Voronoi tessellation via Lloyd's algorithm, with backbone nodes connected according to a Delaunay triangulation. Numerical results show that even a modest number of backbone nodes substantially improves end-to-end capacity and reduces edge consumption for both mobile and fixed users.

• The paper develops analytical capacity models for fixed and mobile free-space quantum networks, including exact star-network distributions, Waxman-graph phase transitions, and high-density flooding limits.
• Free-space links have much shorter useful ranges than fiber—about 7 km for fixed users and 0.3 km for mobile users versus roughly 100 km for fiber—because diffraction, atmospheric loss, noise, and turbulence limit capacity.
• The paper shows that capacity-maximizing Voronoi placements, approximated with Lloyd’s algorithm and connected through Delaunay fiber links, substantially improve single-path capacity and reduce edge consumption without significantly shifting connectivity thresholds.

Overview and motivation

This paper by Fletcher, Pedone, and Pirandola develops an analytical framework for evaluating large-scale free-space quantum communication networks, in which randomly distributed fixed or mobile users are interconnected via free-space quantum links whose performance is fundamentally constrained by diffraction, atmospheric extinction, thermal noise, and turbulence-induced beam wandering. The central contributions are threefold: (i) a characterization of end-to-end network capacities for Waxman random graphs and router-centered star networks under both single-path (widest-path) and multi-path (flooding) routing; (ii) a derivation of the full capacity distribution of router-centered star networks; and (iii) a provably structured method for reinforcing a free-space network with a small number of fiber-optic backbone nodes, based on capacity-maximizing Voronoi tessellations approximated by centroidal Voronoi tessellations (CVTs) via Lloyd's algorithm.

The paper's quantitative premise is stark. Fiber-based channels sustain positive two-way assisted capacities to distances of order 100 km, whereas the corresponding limits for fixed and mobile free-space channels are approximately 7 km and 0.3 km, respectively, under the assumed parameters (800 nm wavelength, worst-case cloudy daytime background irradiance of $H_\lambda \approx 1.9\times10^{18}$ photons m$^{-2}$ sr$^{-1}$ s$^{-1}$ nm$^{-1}$). Beyond short ranges (~2 km fixed; ~20 m mobile), fiber capacity bounds exceed the equivalent free-space bounds by more than an order of magnitude. This asymmetry motivates the hybrid architecture: since backbone fiber links are comparatively cheap relative to quantum-memory-equipped nodes, a modest number of optimally placed repeaters can act as shortcuts that substantially reduce effective user separation.

Channel models

All link rates are computed from two-way assisted capacity bounds ($Q_2$, $D_2$, or $K_2$) of bosonic thermal-loss channels. For pure-loss channels the PLOB bound $\mathcal{C} = -\log_2(1-\eta)$ is exact; with thermal noise $\bar{n}$, the REE-based bound applies, vanishing when $^{-2}$0.

Two free-space regimes are distinguished:

Fixed (long-range) users: characterized by large apertures (5 cm), narrow field of view ($^{-2}$1 sr), and strong turbulence ($^{-2}$2). The channel transmissivity combines long-term beam-waist spreading $^{-2}$3 — incorporating the Rytov number, characteristic transmission length $^{-2}$4, and inner scale $^{-2}$5 — with Beer–Lambert extinction and detector efficiency. Turbulence dominates over pointing error here, so the latter does not enter the fixed-channel expressions.

Mobile (short-range) users: characterized by wide field of view ($^{-2}$6 sr), small spot size (1 mm) and aperture (1 cm). Over short ranges turbulence is negligible and pointing error dominates; beam wandering is resolvable as slow fading with Weibull-distributed transmissivity, yielding fading-PLOB-type capacity bounds with a misalignment correction $^{-2}$7.

A limitation worth noting: the strict monotonicity of capacity in distance fails once the thermal-loss bound reaches zero. Consequently the analytical capacity distribution derived for star networks holds only within the "thermal non-zero region" (≈100 km fiber, ≈9 km fixed, ≈0.3 m–km scale mobile), and general cases require numerical evaluation.

Router-centered star networks: exact capacity distributions

For a single router at the center of a circular region of radius $^{-2}$8 with uniformly distributed users, the unique path between any pair forces single-path routing, so end-to-end capacity equals $^{-2}$9. Exploiting the fact that $^{-1}$0 has pdf $^{-1}$1, and applying a change-of-variables proposition for strictly decreasing capacity functions, the authors derive a closed-form pdf for the end-to-end capacity under the PLOB bound:

$^{-1}$2

supported on $^{-1}$3. Numerically generated histograms with $^{-1}$4 users agree with this analytic density for fiber ($^{-1}$5 km), fixed free-space ($^{-1}$6 km), and mobile free-space ($^{-1}$7 m) configurations. This result gives network designers direct access to the full statistical distribution — not just the mean — of achievable end-to-end rates as a function of coverage radius.

Baseline Waxman networks and phase transitions

Free-space Waxman networks use connection probability $^{-1}$8 with $^{-1}$9 and $^{-1}$0 chosen so that $^{-1}$1 bits/use. Several structural results emerge:

• High-density saturation: single-path capacities converge across region sizes — to the zero-distance free-space bound in the fixed case, and to a common but still-increasing function in the mobile case. Multi-path (flooding) capacities converge to a neighborhood-cut upper bound $^{-1}$2, where $^{-1}$3, extending the fiber result of prior work on random quantum networks to thermal-loss channels.
• Connectivity phase transition: at low nodal densities, both single-path and flooding capacities exhibit a sharp transition mirroring the giant-component transition of Waxman graphs.
• Edge consumption anomaly: single-path edge consumption exhibits a fall–rebound–decay structure at low densities, attributable to disconnected instances contributing nothing until a giant component forms, followed by a power-law decay $^{-1}$4 with $^{-1}$5, consistent across mobile and fixed setups. Since edge consumption quantifies resource usage per communicating pair, these scalings bear directly on how many concurrent user pairs a free-space network can support.

Optimal backbone design: Voronoi structure theorem

The core theoretical contribution defines backbone optimality through a functional

$^{-1}$6

motivated by the observation that fiber shortcuts reduce effective separation between users, and bottlenecks reside almost surely in the free-space layer whenever backbone fiber spans remain moderate. Two propositions then pin down the optimal geometry:

1. Voronoi optimality: a necessary condition for optimality is that the service regions be the Voronoi cells of the backbone nodes and each node sit at its cell's capacity-maximizing point. Notably, the argument extends to merely non-increasing capacities (relevant for thermal-loss channels beyond their cutoff distance), guaranteeing existence of a Voronoi optimal tessellation though not uniqueness.
2. Centroid reduction: for convex regions with two mirror-symmetry axes, the capacity-maximizing point coincides with the centroid, proven via an isometry/reflection argument. Since interior Voronoi cells of an asymptotically hexagonal CVT possess such symmetry, Lloyd's algorithm converging to a CVT provides an efficient approximation — explicitly acknowledged by the authors as heuristic near region boundaries and wherever the capacity ceases to be strictly decreasing.

Backbone nodes are interconnected by their Delaunay triangulation, chosen for linear edge scaling ($^{-1}$7), nearest-neighbor connectivity, and stretch factor below 2. Access edges obey the same Waxman rule as the base network.

This framework refines earlier Voronoi-based QKD topology optimization [Alleaume et al.] by optimizing generator positions directly for end-to-end capacity rather than deployment cost under fixed infrastructure.

Backbone-assisted performance results

Numerical experiments (100 instances × 50 user pairs per point, thermal-loss links throughout, fiber backbone with $^{-1}$8) support four findings:

• Single-path gains: increasing backbone node count improves low-density capacity in both regimes; in the mobile case improvement persists across the full density range, while in the fixed case capacities saturate at the maximal point-to-point free-space capacity, confining backbone benefits to the sub-saturated regime.
• Consumption reduction: backbone-induced shortcuts offset the power-law consumption curves downward, interpretable as an effective shrinkage of the network region.
• Flooding gains are bounded: multi-path improvements occur only below the density at which the neighborhood-cut upper bound binds, after which the backbone provides no advantage.
• No shift in the connectivity phase transition: because backbone connections obey the same probabilistic Waxman rule, adding $^{-1}$9 nodes shifts the critical density only heuristically by $^{-1}$0, which is negligible when $^{-1}$1. This is a substantive caveat: backbones enhance throughput on connected networks but do not materially extend the connectivity regime.

Limitations and open questions

The authors identify several constraints on their results. The CVT approximation to the capacity-maximizing tessellation rests on symmetry assumptions valid primarily in tessellation interiors, and the analysis assumes constant nodal density — appropriate for mobile users over long timescales, but potentially suboptimal for fixed users, where clustering algorithms could improve placements; determining density thresholds where such refinement becomes necessary remains open. Backbone nodes are modeled as ideal and costed only by count; imperfect memories or multiplexing limits, and budget-constrained optimization of $^{-1}$2 itself, are unaddressed. Edge consumption serves only as a proxy for multi-user contention — explicit multi-pair simulations would clarify topological requirements on the backbone. Finally, flooding results constitute upper bounds; intermediate scalable multi-path strategies, and extensions to scale-free random graph models, remain unexplored.

Conclusion

This work supplies both analytical tools — exact star-network capacity distributions, high-density flooding asymptotics $^{-1}$3, phase-transition characterizations — and a practical design principle: optimal fiber reinforcement of a free-space quantum network corresponds to a capacity-maximizing Voronoi tessellation, efficiently approximated by Lloyd-relaxed centroidal tessellations with Delaunay fiber interconnection. The numerical evidence shows that a small number of backbone nodes yields substantial single-path capacity gains and consumption reductions for both mobile and fixed users, while leaving connectivity transitions essentially unchanged. The framework extends naturally to other regions, radial densities, and random-graph ensembles, providing concrete guidance for metropolitan-scale hybrid free-space/fiber quantum networks.