Links — fibre, hollow-core, satellite, free-space, microwave, HAPs, VBG

Standard single-mode fibre (SMF)

ITU-T G.652.D defines the canonical telecom fibre: attenuation 0.20 dB/km at 1550 nm, in the C-band where the loss minimum sits ^{ITU-T G.652.D} . SMF is deployed in every existing telecom network, with installed base measured in millions of kilometres. Polarisation drifts over long distances and requires active compensation; chromatic dispersion is manageable; the medium is mature.

Practical reach for unaided quantum links: roughly 100–200 km before the channel-loss budget bottoms out under realistic detector dark-count rates and source brightness. Beyond that, repeaters (memory-based or otherwise) are needed.

Hollow-core fibre (HCF)

Light propagates in an air- (or vacuum-) filled core rather than silica, avoiding silica's intrinsic loss floor. The best academic demonstration to date is Jasion et al.'s 2022 OFC postdeadline at 0.174 dB/km in a double-nested antiresonant nodeless fibre (DNANF) ^{Jasion et al. 2022} . The commercial state of the art is Lumenisity / Microsoft's 0.091 dB/km hollow-core fibre, now in early data-centre interconnect deployments ^{Microsoft / Lumenisity 2024} .

Two practical advantages over SMF: lower loss extends the unaided reach, and the in-fibre speed is ~99.7 % of c rather than ~67 % (silica's index). That second figure matters for latency-sensitive interconnects and for keeping the classical-correction round-trip inside the memory-coherence budget on memory-based repeaters. Manufacturing is harder than SMF; deployment is early but accelerating.

Satellite

A satellite-borne photon source flies above the dense atmosphere; the loss budget for an uplink/downlink is dominated by roughly 6 km of dense atmospheric layer plus beam divergence over thousands of kilometres. Single-pass loss to LEO is in the 30–50 dB range depending on elevation angle and weather. The rest of the path is near-vacuum, so atmospheric attenuation only applies near the ground station.

Yin et al. 2017 (the Micius satellite) demonstrated entanglement distribution over 1200 km between two ground stations, with one photon delivered to each — the canonical proof that satellite links work as a quantum channel ^{Yin et al. 2017} . Operational constraints include pointing acquisition, daytime background-light rejection, and pass-scheduling (a LEO satellite is overhead for only a few minutes per orbit). Above intercontinental distance, satellite is currently the only practical medium — fibre and HCF can't yet bridge the gap.

Free-space optical (FSO)

A telescope-to-telescope link through the atmosphere. Clear-sky loss is roughly 0.5 dB/km in good weather, much higher in fog or rain. Atmospheric turbulence introduces beam scintillation and beam wander; daytime background light is a noise source that limits usable link bandwidth.

The canonical demonstration is Schmitt-Manderbach et al.'s 2007 decoy-state QKD over a 144 km horizontal link between La Palma and Tenerife ^{Schmitt-Manderbach et al. 2007} . FSO fits where fibre isn't available — across rivers, across building gaps in dense urban environments, or to mobile platforms — but weather dependence makes it a poor solo backbone.

Cryogenic microwave links

A cryogenic microwave link is a superconducting coaxial cable or waveguide running at millikelvin temperatures between two dilution refrigerators, each housing a superconducting QPU. Superconducting qubits emit and absorb microwave photons natively, so a transmon at one end can hand a quantum state to a transmon at the other end with no frequency conversion in between. The price is that the entire channel is part of the fridge plant: the inner conductor is typically NbTi or aluminium, the line is shielded against stray fields, and the centre of the link sits below ~100 mK so that thermal blackbody photons from a 4 K stage do not swamp the quantum signal. Bending the cable through room temperature is not an option; the link is rack-to-rack inside a quantum computing facility, with a practical reach of tens of metres.

The benchmark demonstration is Magnard et al. (PRL 125, 260502, 2020), a 5 m cryogenic waveguide connecting two transmons in separate fridges with deterministic state transfer at 85.8 % fidelity and on-demand remote entanglement ^{Magnard et al. 2020} . Storz et al. (Nature 617, 265, 2023) extended this to a 30 m link and used it to close the locality loophole in a Bell test between two superconducting qubits — the first loophole-free Bell violation with a macroscopic, microwave-controlled platform, rejecting the local-hidden-variable null at P < 10^{−108 Storz et al. 2023} . Two 2025 results sharpen the engineering envelope: Yam et al. (npj Quantum Information 11, 87, 2025) demonstrated continuous-variable entanglement distribution over a 6.6 m link held below 52 mK at its centre ^{Yam et al. 2025} , and Mollenhauer et al. (Nature Electronics, 2025) showed a detachable plug-and-play interconnect with inter-module SWAP efficiency above 99 % in under 100 ns, easing the path to modular multi-fridge systems ^{Mollenhauer et al. 2025} .

The strategic position is narrow but important. For two superconducting QPUs inside one facility, a microwave link is the only option that avoids the microwave-to-optical transducer — currently the lowest-efficiency element in any inter-QPU optical link (see the transduction subject). Below ~100 m, microwave wins on fidelity and rate; beyond that, the cryogenic plant cost and thermal load make further extension impractical, and transduction-equipped optical fibre takes over. Treat microwave as the rack-to-rack and fridge-to-fridge tier of a layered network: the protocols built on top — entanglement generation, swapping, purification — are the same as on optical links (see entanglement and swapping); only the physical medium changes.

High-altitude platforms (HAPs)

HAPs are airborne carriers that hold a quantum-photon payload above part of the atmosphere: stratospheric balloons at 18–25 km, fixed-wing aircraft and long-endurance UAVs in the 5–20 km band, and rotary-wing drones flying a few hundred metres above the ground. The classical analogue is Project Loon's stratospheric internet relay. For quantum links the appeal is the same — sit above the densest, most turbulent layer of air; hover or loiter over a defined service area; deploy and recover on hours, not launch windows.

The canonical balloon demonstration is Wang et al. 2013, in which the Pan group flew a decoy-state BB84 transmitter on a hot-air balloon and ran a 40 km link to a moving ground station — a deliberate ground-based simulation of the orbital dynamics later flown on Micius, including attitude change, vibration, and high-loss operation up to 96 km equivalent channel ^{Wang et al. 2013} . On the drone side, Liu et al. 2020 mounted an entangled-photon source on a 35 kg octocopter and distributed polarisation entanglement between two ground sites 200 m apart, in daytime and through rain ^{Liu et al. 2020} . A follow-up the next year added a second drone as an optical relay, extending the entanglement-distribution baseline to about 1 km ^{Liu et al. 2021} .

HAPs fit between terrestrial FSO and full LEO satellites. The advantages are persistence over a fixed region, lower latency than a LEO pass, and far cheaper deployment than a launch — useful for transient or contingency coverage, military and emergency-response use cases, and as a mobile trusted relay where neither fibre nor a satellite pass is available. The disadvantages are daytime sky background, limited payload mass and power for cryogenic sources or large telescopes, station-keeping cost in wind, and the much shorter reach of a single platform compared to a satellite footprint. HAPs are a niche between FSO and satellite, not a substitute for either.

Vacuum beam guide (VBG)

A vacuum beam guide is an evacuated tube with an array of relay lenses spaced kilometres apart, guiding a Gaussian beam between two endpoints without total internal reflection in glass. Huang et al. (Phys Rev Lett 133, 020801, 2024) analyse a confocal design with lens spacing L₀ = 4 km, lens radius R = 10 cm, and a telecom-band beam waist w₀ ≈ 3 cm, in a tube held below ~1 Pa. Under those parameters they project a total attenuation as low as 5 × 10⁻⁵ dB/km — an effective attenuation length around 80,000 km, roughly three orders of magnitude below the best silica-core fibre (~0.148 dB/km) ^{PRL 133, 020801} .

The loss budget has three terms: lens absorption / scattering / reflection (kept below 10⁻⁴ dB/km with good AR coatings and R ≥ 10 cm to suppress diffraction), residual-gas absorption (<10⁻⁴ dB/km at 1 Pa for telecom wavelengths, per HITRAN), and misalignment of the lens array. Transverse alignment is the binding constraint: σ_s < 0.2 mm per lens keeps the misalignment term below 10⁻⁴ dB/km; relaxing to 0.6 mm still beats SMF by an order of magnitude. The authors point to LIGO's kilometre-scale evacuated beam tubes as the existence proof that the supporting civil and vacuum infrastructure is buildable.

The strategic claim is that a VBG could carry quantum traffic over thousands of kilometres without repeaters, at a frequency-integrated channel capacity above 10¹³ qubit/s — orders of magnitude above current satellite QKD rates. None of this is built. The open engineering problems are active alignment of an N-thousand-lens array against seismic drift, vacuum-pump spacing and maintenance over continental distances, and the capital cost of an evacuated right-of-way competitive with installing HCF in existing ducts. Treat the VBG as a long-horizon alternative to space-segment backbones rather than a near-term procurement option.

Side-by-side comparison

Each link has a regime where it wins. The metrics subject covers loss-budget composition (how the per-km numbers turn into end-to-end photon-survival probabilities); the repeaters subject covers what happens past the unaided reach.