Utah and Rice Test EntangledEdge, a Satellite-to-Ground Quantum Sensing and Learning System, on IBM Heron Processors With a Simulated Orbital Link: arXiv 2610.24353, 17 October 2026, TRL 3
Quentir Defense Monitor
Evidence-based insights for quantum defense and security. Published by Quentir Systems LLC · October 20, 2026.
On October 17, 2026, five researchers from the University of Utah and Rice University posted arXiv 2610.24353, a systems paper called EntangledEdge that asks a question no earlier satellite quantum demonstration has asked. If a quantum sensor in orbit holds its measurement as a quantum state, and that state can be teleported to the ground instead of being read out on board, can a fleet of ground-based quantum processors learn from it without ever collapsing it? The authors, Mohammad Abrarul Hasanat, Amit Samanta and Rohan Basu Roy in Salt Lake City and Jason Ludmir and Tirthak Patel in Houston, answer with an architecture, a hybrid prototype on real IBM superconducting processors and offline anomaly-detection numbers on stored telemetry from two real satellites. The arXiv listing says the paper is accepted at ATC 2026.
For a defense reader the interest is structural. A constellation of quantum magnetometers, gravimeters or clocks whose outputs can be fused coherently on the ground is what the capability map calls a distributed sensor array, separate instruments working as one. The paper does not use the word defense once; it frames its uses as geomagnetic field mapping, space-weather characterization and gravity-anomaly tracking for ice mass and hydrology. Those are the observables a navy reads for undersea search and a space force reads for satellite health, which is why the Monitor reads the development as dual-use.
What arXiv 2610.24353 proposes: quantum states teleported from orbit to four ground stations and classified by quantum autoencoders on IBM Heron and Nighthawk processors
The architecture has two planes. The quantum data plane carries the sensed state from the satellite to a ground station by teleportation over a pre-shared Bell pair, with the two-bit outcome of the on-board Bell measurement sent down by classical radio; the state itself never travels and is never measured in transit. The classical control plane carries model parameters, calibration signals from each ground processor and threshold updates between stations. The paper's argument for this shape is that a learner with coherent access to quantum states can in some settings need exponentially fewer samples than one working from classical readouts, and that advantage disappears the moment the satellite measures its own qubits.
Three design choices do the engineering work. Each ground station holds a capacity-diverse bank of quantum autoencoders, small circuits of three and four qubits at different depths, and routes each arriving event to the model that matches the channel it came through, estimating optical turbulence along the line of sight from the Rytov variance under a Hufnagel-Valley atmosphere profile. Because labeling at the satellite would require the destructive measurement the system exists to avoid, the detection threshold is set without labels: a two-component Gaussian mixture is refitted periodically on a rolling reservoir of anomaly scores and the crossing point between its components becomes the threshold. And the stations train jointly in a federated fashion, each station's contribution weighted by an expected success probability derived from its processor's calibrated gate errors and coherence times. Quentir's Medicine lane looked at the same federated pattern in a classical setting when it asked what quantum-inspired buys a federated ECG classifier; here the federation is between quantum processors that differ in hardware quality on the day.
The prototype is candid about what exists. The paper states that orbital quantum sensors which emit their state into a quantum link are not yet deployed. The authors therefore encode stored telemetry segments for offline classification into the amplitude vector a three- or four-qubit circuit expects, apply sensor noise characterized on QuEra's Aquila, the 256-qubit neutral-atom analog processor available through Amazon Braket since November 2022, and simulate the orbit-to-ground link with an SGP4 orbit propagator and the turbulence model above. The ground processors are real: four simulated stations at Fairbanks, Wallops, Madrid and Singapore each run on IBM Heron and Nighthawk superconducting processors, trained with simultaneous perturbation gradient estimates at 1,024 shots. The code and hardware results are deposited on Zenodo under an MIT license, a 64.7-megabyte artifact published on October 15, 2026.
Quantum pillar: networking (distributed sensor arrays). Use posture: dual-use. Technology readiness: TRL 3 of 9. Inference for the simulated ground stations ran on real IBM quantum processors, while the orbital sensor and the teleportation link were simulated and the sensed states were encoded from stored satellite telemetry for offline classification, so this is a proof of concept on hardware that exists plus stand-in data for the hardware that does not.
What the hybrid prototype measured on ESA OPS-SAT and NASA SMAP telemetry: F1 0.63 against 0.52 for isolated stations, 0.57 seconds per event and 51.6 percent processor utilization
The two telemetry sources are well documented. OPS-SAT was the European Space Agency's flying laboratory, launched on December 18, 2019 into a 515-kilometer polar orbit to test mission-control software in orbit, and it reentered in the night of May 22 to 23, 2024. Its housekeeping telemetry, with segments labeled normal or anomalous by domain experts, gives the paper its main task at an anomaly fraction of about 20 percent. The SMAP channels come from the NASA Jet Propulsion Laboratory telemetry set released with the telemanom repository and the 2018 KDD paper by Hundman and colleagues, which carries 105 labeled anomaly sequences across 82 channels from the Soil Moisture Active Passive satellite and the Curiosity rover. Neither dataset involves a quantum sensor; both are classical spacecraft measurements encoded into quantum states to stand in for sensed states that no satellite yet emits.
On OPS-SAT the full system reaches an F1 score of 0.63, with precision 0.59, recall 0.66 and accuracy 0.83, and the standard deviation across 100 simulated calibration days is 4.3 percent of F1. On SMAP the F1 is 0.48, with recall 0.67 and precision 0.37. The comparison that matters is against the paper's own alternatives on the same hardware. A station working alone on its own data reaches 0.52 on OPS-SAT; shipping everything to one central station gives 0.44; plain federated averaging gives 0.52; a quorum rule gives 0.44. The authors summarize the gain as a 21 to 27 percent F1 improvement over the isolated baseline, mostly from precision, with a small loss of recall. An ablation removes each design choice in turn: dropping the label-free threshold costs the most, followed by replacing the fidelity-weighted aggregation with plain averaging, and then collapsing the three-model bank to one.
The systems numbers are where the paper is most at home. Each event costs 0.57 seconds end to end, against 0.30 seconds for a single-model station and 0.78 seconds for the centralized design, and that figure stays flat when the sensing rate is raised fourfold and when the fleet grows from a two-by-two to an eight-by-eight arrangement of satellites and stations, over which the total event count scales from roughly 300 to 3,700. Active-window utilization of the quantum processors is 51.6 percent, against 26 percent isolated and 21 percent centralized. The queue is a single-server, single-slot model: a state that arrives while a processor is busy is lost, because there is no quantum buffer to hold it. Growing the autoencoders from four to ten qubits raises the federation gain from 8.4 to 18.3 percent, which the authors read as the architecture becoming more useful as hardware grows.
Two cautions belong beside the numbers. The paper compares only against quantum variants of itself, never against a classical baseline. Quentir has seen this pattern before on satellite data, when IonQ and Capella's 20-qubit circuit Born machine reached a filtered F1 of 0.32 on radar change detection. The second caution is the paper's own: qubit coherence times on today's processors are orders of magnitude shorter than the gaps between satellite contact windows, so a sensed state cannot wait for its station to come into view. Both cautions keep the result at TRL 3.
What a quantum data plane from orbit would let a force do, who gains from it, and what stands between the paper and a program office
Read as capability, the paper is a blueprint for the orbital quantum sensing case the Monitor has watched from the ground up. Quantum magnetometers reach the femtotesla scale where fluxgates stop near a picotesla, as the paper itself notes, and gravity gradiometers and optical clocks hold a similar margin over their classical counterparts. A constellation of such instruments whose states are fused coherently on the ground is, in principle, a single planet-sized sensor: entanglement-enhanced gradiometry and networked clock comparison are the two tasks the authors name. For an undersea force, that is magnetic and gravity anomaly mapping at a resolution a single satellite cannot reach. For a space force, it could support a continuous, label-free watch on the health of its own fleet, although the prototype performed offline classification of stored telemetry. For a navigation program, it is clock comparison across a network that does not depend on the satellite navigation signal an adversary can jam.
Who gains follows from who owns the pieces. The architecture needs a satellite that can emit a sensed state into a quantum link, space-to-ground entanglement distribution at useful rates, and a geographically spread set of ground quantum processors. The first does not exist. The second has been demonstrated once at scale, by the Micius mission that distributed entangled pairs to two ground sites 1,200 kilometers apart, which the paper cites. The third exists today only as cloud access to a handful of vendors' machines. A state that can field all three gains a sensing and fusion capability, which is why the posture is dual-use: the identical system that maps ice mass for a science agency maps magnetic anomalies for a navy.
What stands between the arXiv listing and a program office relying on it is a list the paper largely writes for the reader. The sensor side is a simulation: no orbital instrument has ever handed a quantum state to a link, and the Aquila noise model stands in for a space-qualified neutral-atom payload that would face radiation, thermal cycling and launch loads no laboratory array sees. The link side assumes pre-shared Bell pairs at the event rate, which no current satellite supplies, and a teleportation fidelity through turbulence that the paper models. The timing side is the hardest: a system in which a state is lost whenever its receiver is busy, and in which coherence expires long before the next pass, needs quantum memory in orbit or on the ground before it can be more than a demonstration. The learning side needs a comparison against the classical detectors a mission operator already runs, on the same channels, before anyone can say the quantum path buys accuracy.
The sources establish a coherent systems design, an open artifact that runs on real superconducting processors from four simulated ground sites, and a measured gain over the obvious alternatives in offline classification of two sets of stored spacecraft telemetry. They do not establish an orbital quantum sensor, a quantum link at mission rates or an advantage over the classical tools in use. A buyer should file EntangledEdge as the first end-to-end sketch of what a quantum sensing constellation's ground segment would have to do, and watch for a flown sensor payload that emits into a link, a measured teleportation fidelity on a real downlink, and a head-to-head against the classical anomaly detectors on the same OPS-SAT channels.
Sources
Primary source: Mohammad Abrarul Hasanat, Amit Samanta, Jason Ludmir, Tirthak Patel and Rohan Basu Roy, 'EntangledEdge: A Distributed System for Orbital Quantum Sensing and Learning,' arXiv 2610.24353, submitted October 17, 2026 (University of Utah and Rice University), with its Zenodo artifact. Other material: ESA's OPS-SAT mission page; the NASA JPL telemanom repository and the 2018 KDD paper by Hundman and colleagues; QuEra's Aquila product page. The capability reading, the comparison with earlier satellite-data results and the buyer questions are Quentir analysis.