An arXiv version-1 preprint dated 20 August 2026 reports an experiment in which locally generated continuous-variable quantum key distribution (CV-QKD) was tested over a deployed link combining fibre and free space. The study used Gaussian-modulated coherent states and squeezed states, while adaptive post-processing grouped data by transmission conditions, handled residual fading and used rate-adaptive blind reconciliation, the stage that helps turn noisy shared measurements into matching keys.
Under the reported asymptotic security analysis, the coherent-state protocol produced a secret-key rate of 0.42 Mbit/s, while the squeezed-state protocol produced 0.93 Mbit/s. The full adaptive chain extracted up to 318 kbit from a single fluctuating hybrid-channel measurement.
The study’s central question was whether a locally generated local-oscillator CV-QKD system could operate across a deployed hybrid fibre-free-space channel and whether adaptive post-processing could accommodate its fluctuating transmission. The authors describe this as a way to move adaptation to the transmission medium from the quantum-optical layer into post-processing.
A link with two distinct legs
The deployed testbed used a 620-m free-space segment. The short configuration had approximately 10 to 18 dB of total loss, while the long configuration used 2 km of fibre and had approximately 13 to 21 dB of total loss.
The two protocol analyses were not identical. The coherent-state protocol used Gaussian-modulated no-switching under a trusted-detector assumption. The squeezed-state analysis estimated squeezing and anti-squeezing parameters from measured data.
The report covered four representative data sets: coherent-state short- and long-channel data, a squeezed-state short-channel data set, and a jointly processed group of four coherent-state short-channel measurements.
The reported processing frames were 100 µs for coherent-state data and 1 ms for squeezed-state data.
Handling a changing channel in post-processing
Rather than apply one fixed treatment to the whole record, the adaptive pipeline estimated transmittance and clustered the data around changing transmission conditions. For fading that remained after clustering, it used covariance-matrix averaging or de-fading, then applied rate-adaptive blind reconciliation.
In practical terms, the approach handled the measurements in stages: first sorting data according to transmission, then accounting for remaining fluctuations, and finally reconciling the data at an adaptive rate.
That strategy produced different contributions in different channel regimes. In the reported short-channel regime, the threshold cluster retained an average 85% of the data and supplied 91% of the total secret-key rate. In the intermediate long-channel regime, the threshold cluster retained 42% of the data and supplied 70% of the total rate, while additional long-channel clusters supplied 30% of the rate and captured 29% more data.
The figures show a clear trade-off: a threshold cluster carried most of the reported rate in both summaries, but additional clusters made a larger contribution to the long-channel data share.
More data survived the tested alternatives
In a comparison of processing strategies, adaptive clustering exceeded the best tested threshold, linear-binning and dB-binning approaches by 27%, 31% and 41%, respectively. It retained 78.6% of the acquired data, compared with 60.5% for threshold selection, 65.0% for linear binning and 49.6% for dB binning.
The reported comparison combined a higher weighted total secret-key rate with a larger retained-data fraction. Its figures describe the tested measurements and do not by themselves establish a ranking beyond those channel conditions.
The error-correction step changed the totals
The study also compared blind reconciliation with one-shot decoding. Across three measurements, the reported total-secret-key-rate differences were 12.06%, 16.77% and 18.79%, with corresponding key-length differences of 11.68%, 16.50% and 17.72%. The supplied analysis reports blind reconciliation as the higher-performing option in those comparisons.
At the end of the complete chain, the maximum reported extracted key length was 318 kbit from one fluctuating hybrid-channel measurement in the asymptotic regime. That result belongs to the combined adaptive pipeline, not to the clustering step alone.
The finite-data test changes the picture
The strongest qualification comes from the study’s finite-size composable-security analysis, the part of the evaluation that asks whether a finite block of measured symbols yields a positive secure key. The largest high-transmittance cluster contained 2.15 × 10^8 symbols and remained insufficient for positive composable key extraction under the experimental parameters.
Put simply, the experiment reported positive asymptotic rates but not positive finite-size composable security under the conditions tested. The 318-kbit extraction was also reported in the asymptotic regime.
That limitation means the coherent and squeezed rates should not be read as a universal head-to-head verdict. The protocols used different security analyses: the coherent-state analysis assumed trusted detectors, while the squeezed-state analysis estimated squeezing and anti-squeezing from measured data.
A result with a narrower reach
The authors interpret the experiment as evidence that adaptation to a changing transmission medium can be moved from the quantum-optical layer to post-processing. They also point to a possible common architecture spanning fibre, terrestrial free-space and future satellite links.
That broader possibility is still a proposal, not a result measured across all of those settings. The supplied analysis describes direct evidence from the reported deployed hybrid testbed and identifies stronger turbulence, different deployments and satellite-relevant geometries as areas for further testing.
Other unresolved questions include whether longer acquisitions or improved hardware could produce positive composable finite-size keys, and whether rate-adaptive reconciliation and its associated classical communication can operate in real time.
What the preprint establishes
The work provides a deployed-channel test of adaptive processing for coherent- and squeezed-state CV-QKD, with its principal secret-key rates calculated in the asymptotic regime. Its finite-size test did not reach a positive composable key.
The authors state that the data underlying the results are available in reference [49]. The acknowledgments report European Union Horizon Europe support through the Quantum Secure Networks Partnership.
Paper data and sources
Original title: Squeezed- and coherent-state quantum key distribution over a deployed hybrid fibre-free-space channel
Authors: Dnan A. E. Hajomer, Huy Q. Nguyen, Ivan Derkach et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text