Abstract This supplementary package provides synthetic datasets, Python code, and computational results for a proof-of-concept study of telemetry-based anomaly detection in satellite quantum key distribution (QKD) systems with optical phased arrays (OPAs). It supports reproducible comparisons of monitoring methods under simulated disturbances and benign changes in operating conditions. The dataset comprises five computational replicates generated using fixed random seeds, each containing 19,200 observation windows across separate training, calibration, test, and benign distribution-shift partitions. The simulated telemetry represents photon counts, quantum bit error rate, pointing jitter, background yield, beacon transmission, and normalized count and error residuals. Test scenarios include injected background, loss, error, and coupled perturbations. Three detection approaches are compared: Isolation Forest applied to seven telemetry features, Isolation Forest applied to two residual features, and a non-learned rule based on the maximum absolute residual. Detection thresholds are calibrated exclusively on nominal data, without tuning against test labels. The package includes per-seed metrics, aggregate statistics, illustrative link-budget calculations, OPA geometry metadata, and exhaustive enumeration of 128 Boolean assignments for a conjunctive authorization rule. The residual rule performs best overall for the specified perturbations; this finding does not establish general superiority or transferability to other environments. All observations are synthetic and do not represent measured satellite telemetry or verified attack attribution. The package does not implement finite-key cryptographic security, and its logical checks do not constitute a security proof. Reproduction instructions, pinned dependencies, and file checksums are included.
Abstract This supplementary package provides synthetic datasets, Python code, and computational results for a proof-of-concept study of telemetry-based anomaly detection in satellite quantum key distribution (QKD) systems with optical phased arrays (OPAs). It supports reproducible comparisons of monitoring methods under simulated disturbances and benign changes in operating conditions. The dataset comprises five computational replicates generated using fixed random seeds, each containing 19,200 observation windows across separate training, calibration, test, and benign distribution-shift partitions. The simulated telemetry represents photon counts, quantum bit error rate, pointing jitter, background yield, beacon transmission, and normalized count and error residuals. Test scenarios include injected background, loss, error, and coupled perturbations. Three detection approaches are compared: Isolation Forest applied to seven telemetry features, Isolation Forest applied to two residual features, and a non-learned rule based on the maximum absolute residual. Detection thresholds are calibrated exclusively on nominal data, without tuning against test labels. The package includes per-seed metrics, aggregate statistics, illustrative link-budget calculations, OPA geometry metadata, and exhaustive enumeration of 128 Boolean assignments for a conjunctive authorization rule. The residual rule performs best overall for the specified perturbations; this finding does not establish general superiority or transferability to other environments. All observations are synthetic and do not represent measured satellite telemetry or verified attack attribution. The package does not implement finite-key cryptographic security, and its logical checks do not constitute a security proof. Reproduction instructions, pinned dependencies, and file checksums are included.
SatQKD-Sentinel is a self-contained Python research testbed integrating satellite quantum key distribution (QKD) channel modeling, optical phased array (OPA) analysis, synthetic telemetry generation, and machine-learning-based state classification and anomaly detection. The software also provides geostationary coverage calculations, geometric contact planning, and simulations of key generation, storage, and demand. The telemetry simulator represents ten operating states, comprising nominal conditions, five fault or environmental scenarios, and four synthetic attack signatures. Twenty-four causal features are extracted from telemetry without using scenario labels or future observations. Gradient boosting, random forest, and logistic regression classifiers are trained and compared using independently generated training, validation, and test partitions. Model selection uses validation macro-F1, followed by evaluation on the held-out test set. Isolation Forest and Mahalanobis-distance detectors provide complementary novelty detection based on nominal training data. The package includes a command-line interface and a notebook designed for Google Colab, with embedded project modules and controlled execution of dependency setup and numerical computations. Outputs include synthetic datasets, predictions, evaluation metrics, provenance records, physical-model tables, thirteen figures, and a standalone HTML report. Source code, tests, executed notebooks, and verification records support inspection and reproducibility. All observations are synthetic, and classification results characterize the simulated scenarios rather than demonstrated performance on operational satellite systems. Finite-size key estimates are illustrative and do not constitute a composable cryptographic security proof. No real cryptographic keys are generated or authorized for release. Geometric visibility does not establish QKD feasibility, and federated learning remains outside the implemented scope.
Because quantum computing can break the hard mathematical problems that protect RSA and ECC, classical cryptography is entering a period of risk. QKD offers a different kind of protection, using quantum physics itself to guarantee secrecy. Bringing QKD to satellites helps cover long distances, but that link must withstand disturbances from the atmosphere, precision-related optical challenges, and sophisticated adversaries. In this paper, we explore how artificial intelligence can support satellite QKD. Our framework predicts channel changes using machine learning, tunes system parameters through reinforcement learning, and watches for suspicious activity using anomaly detection. When tested in simulations, the approach improved secure key generation by about 35%, cut QBER from 8% to 5%, and held up strongly against quantum-level attacks. These results demonstrate that integrating AI techniques with satellite-based QKD provides a resilient and scalable solution for secure global communications.
Abstract Practical quantum key distribution (QKD) is secure only to the extent that a deployed implementation remains compatible with the physical and statistical assumptions used in its security analysis. Scalar quantum bit error rate (QBER) monitoring cannot, by itself, distinguish legitimate channel drift from all implementation-level attacks, and conventional supervised classifiers usually assume that every test-time attack family was represented during training. This work presents QKD-XGuard, a simulation-validated monitoring method for decoy-state BB84 that combines physics-informed residuals, temporal feature encoding, metric-structured latent representations, class-conditional conformal prediction, and explicit open-set rejection. The method models expected signal/decoy gains, error rates, single-photon behavior, key-rate consistency, and detector statistics; it then learns from deviations between measured and physically expected telemetry rather than from raw measurements alone. A reproducible stochastic block-level simulator was used to generate 3,500 developmentregime sessions spanning normal operation and four attack families, with 16 telemetry blocks per session. Entire attack families were excluded from training in four leave-one-family-out folds. Across these folds, the physics-residual model achieved known-class macro-F1 of 0.952 ± 0.007, unknown-attack AUROC of 0.945 ± 0.049, unknown recall of 0.876 ± 0.137, and empirical class-conditional conformal coverage of 0.910 ± 0.013 at nominal 90% coverage. Under a disjoint 75–115 km channel regime, macro-F1 remained 0.962 ± 0.004, whereas an otherwise identical raw-telemetry ablation fell to 0.796 ± 0.068. A deliberately stealthoriented adaptive composite attack was rejected in 69.2% ± 5.7% of sessions. These results support physics-residual modeling as a useful mechanism for separating channel variation from unsupported attack behavior, while also showing that open-set performance varies materially by attack family. QKD-XGuard is an operational monitoring layer; it does not replace composable QKD security proofs or finite-key analysis.
Cloud infrastructures face a long-horizon threat from cryptographically relevant quantum computers, motivating both post-quantum cryptography (PQC) and quantum key distribution (QKD) as complementary defenses. QKD additionally offers a physical, information-theoretic signal of eavesdropping through the quantum bit error rate (QBER) of the sifted key. Prior work has applied machine learning across the QKD pipeline, chiefly for parameter optimization and physical-layer stabilization, but has given less systematic attention to whether learned, distribution-aware statistics improve on a simple aggregate-QBER threshold for detecting partial intercept-resend eavesdropping in a cloud key-rotation setting. We simulate BB84 key exchange under uniform and time-concentrated (“bursty”) partial interception across two monitoring-window lengths, and compare a static aggregate-QBER threshold against a logistic-regression detector that additionally uses block-wise QBER variance and maxima, at a matched false-positive rate. Across all four tested conditions, the aggregate threshold matched or exceeded the block-wise detector, while monitoring-window length had a substantially larger, quantifiable effect on detection sensitivity than the choice of detector. For this attack model, cloud QKD key-management pipelines therefore gain more from shortening or adaptively sizing the QBER monitoring window than from adding distributional machine-learning features to the detection statistic, which also helps explain why reported machine-learning gains in QKD concentrate in parameter optimization rather than eavesdropping detection Keywords: machine learning; Quantum Key Distribution; BB84; eavesdropping detection; quantum bit error rate; cloud security
As more aspects of modern society depend on digital communication, we increasingly rely on infrastructure that ensures the privacy and security of this communication. Classically, this has been provided by cryptographic protocols such as public-key encryption, in which secrets called keys are exchanged between different parties to enable secure communication. The rapid development of quantum algorithms which violate the assumptions of these protocols, however, poses a security challenge to modern cryptography. \n \nQuantum resources can also be used to strengthen cryptographic security, particularly the security of key exchange protocols. This approach, QKD, can be implemented by encoding in quantum systems such as single photons sent through free-space or a fiber. Fiber based QKD devices are already commercially available, but are fundamentally limited to distributing keys over a few hundred kilometers. To address this distance limitation, research QKD systems are being developed to exchange keys through free-space to satellites. This work considers practical challenges to building and testing both types of QKD devices. \n \nFirstly, we consider modeling and mission analysis for airborne demonstrations of QKD to stratospheric balloons and aircraft to simulate a satellite. Based on the mission parameters available for both platforms, we found aircraft platforms were more promising for testing prototype QKD satellite systems. We developed a mission planning tool to help design the flight geometries for testing the device. \n \nNext, we developed three new components for a QKD satellite prototype. The requirements for electro-optical devices in orbit are very different from lab environments, mandating new approaches to designing QKD devices. We developed a quad single photon detector package to meet the requirements for free-space links to low earth orbit. Moreover, we designed and built optical systems for analyzing the polarization of photons and an adaptive optics unit to increase the efficiency of collecting the encoded photons. All three devices were tested in conditions that simulated the time and loss of a satellite pass. \n \nFinally, we demonstrated a laser damage attack on a live commercial QKD system. Our attack injected additional optical power into the sender device to modify security-critical components. Specifically, our attack damaged the PIN diodes which monitor the encoded photon number, reducing their sensitivity or completely blinding them. Our damage could compromise the entire key, and was performed during system operation while raising no alarms. \n \nIn summary, this work shows the trade-offs of testing QKD payloads on different airborne platforms, develops components for a satellite QKD payload, and demonstrates a security vulnerability in a commercial QKD system that can fully compromise the key. These results help address practical challenges to building QKD devices, improving the security of modern cryptography.
In modern communication networks, particularly satellite-based systems, data security faces significant challenges from vulnerabilities such as signal interception, jamming, and latency during long distance transmissions. Traditional cryptographic methods are increasingly vulnerable to quantum computing threats, underscoring the need for advanced solutions to protect data integrity, confidentiality, and availability. This research investigates the fusion of quantum cryptography and Machine Learning (ML) to improve security in satellite communication. The Quantum Key Distribution (QKD), which is grounded in quantum mechanics, enables unbreakable encryption by detecting eavesdropping via quantum state disturbances. The CatBoost ML algorithm is applied to a dataset of 10,000 records featuring categorical attributes for prioritizing security elements such as anomaly detection, encryption types, and access controls. The model yields an accuracy of 89.23% and Area under Curve the Receiver Operating Characteristic (AUC-ROC) score of 94.56%, effectively predicting threat levels. Feature importance reveals anomaly detection (28.5%) and quantum encryption (22.3%) as primary contributors. While hurdles such as high implementation costs and transmission range limitations persist, this quantum ML synergy provides a proactive, adaptive framework for resilient, future-ready communication networks.