QKD Dataset

Overview

Background

Simulation Scenarios

Dataset Structure

Tasks

Files

Subject Keywords

Conflicts of Interest

References

To evaluate the performance of the QKD system, two distinct analytical models were employed to estimate the quantum bit error rate (QBER) and secret key rate (SKR). The first model is designed for coexistence scenarios, where classical and quantum channels are transmitted over the same fiber, and can also be reduced to the dark-fiber case by neglecting coexistence-induced impairments ‎[1]. The second model, developed by Fraunhofer Heinrich Hertz Institute (HHI), is specifically applicable to dark-fiber conditions. To generate the datasets, a simulation pipeline was developed to transform the outputs of these analytical models into structured datasets suitable for further analysis, performance evaluation, and data-driven QKD network-planning studies.

Two main scenarios are considered. The first scenario, corresponding to Dataset 01, models a dark-fiber QKD link, where the optical fiber is dedicated exclusively to quantum communication using the BB84 protocol. In this case, no classical optical channels are transmitted in the same fiber, so coexistence-related impairments such as Raman scattering, inter-channel crosstalk, and four-wave mixing are not considered. The performance of each point-to-point QKD link is mainly affected by fiber attenuation, link length, optical visibility, detector dark counts, detector afterpulsing, detector efficiency, and dead time. The second scenario, corresponding to Dataset 02, models a coexistence environment, where quantum and classical signals are transmitted through the same C-band optical fiber. In this case, the quantum channel shares the fiber with several classical WDM channels placed on a 100 GHz wavelength grid. To reduce the impact of classical traffic on the quantum signal, a 200 GHz symmetric guard band is reserved around the quantum wavelength, and classical channels are only allocated outside this protected region. Unlike the dark-fiber case, this scenario includes additional noise contributions caused by coexistence, especially Raman scattering and leakage from classical channels into the quantum channel. For both scenarios, the simulations are performed over multiple configurations by varying fiber types, fiber lengths, detector technologies, and component parameters. Three detector technologies are considered: SPADs, cooled SPADs, and SNSPDs, allowing the datasets to capture different levels of receiver performance. Each generated sample represents one point-to-point QKD link between two network nodes, and for each configuration, the analytical models compute the resulting QBER and SKR.

The dataset X ∈ RD×N consists of D samples x(d) ∈ RN, d=1, …, D. Each sample corresponds to a point-to-point BB84-based QKD link between two network nodes and is characterized by N features xn ∈ R, where n=1, …, N. These features describe the physical and operational properties of the QKD link and other component-level parameters. In the coexistence scenario, the feature vector is extended to include classical-channel-related parameters, such as number of coexisting classical channels, WDM grid spacing, guard-band configuration, Raman scattering contribution, and crosstalk contribution. Each sample is associated with output performance labels, including QBER and SKR. This representation allows the generated datasets to be used for data-driven modeling, performance prediction, and network-planning studies in QKD systems.

These datasets support the development of machine-learning-based automation for QKD networks and enable efficient and comparable QKD performance studies. They can be used for QKD performance estimation through regression tasks, such as predicting QBER and SKR, as well as classification tasks, such as identifying feasible links or categorizing QKD link performance based on predefined thresholds.

The dataset files are available for download on this webpage. If you are interested in receiving the dataset, please follow the link.

Quantum Key Distribution (QKD); BB84 Protocol; Secret Key Rate (SKR); Quantum Bit Error Rate (QBER); Dark Fiber; Classical–Quantum Coexistence; QKD Dataset; Machine Learning (ML) for QKD.

The authors have no conflicts of interest to declare.

[1]   P. Eraerds, N. Walenta, M. Legré, et. al., “Quantum key distribution and 1 Gbps data encryption over a single fibre,” New Journal of Physics, 12, 063027 (2010). doi: 10.1088/1367-2630/12/6/063027.