Quantum key distribution (QKD) is sometimes marketed as secure against computational attacks, as its security is based on the laws of quantum mechanics for detecting eavesdroppers, rather than the difficulty of mathematical problems. However, this security guarantee applies only to the quantum channel itself, and the physical hardware implementing these systems remain vulnerable. The electrical components controlling these systems generate electromagnetic emissions when active, and if those emissions carry information such as the polarization states being encoded, an attacker could recover key-relevant information without ever interacting with the quantum channel, bypassing the security guarantees of the protocol entirely. This work investigates whether such a vulnerability exists in a commercial QKD transmitter unit by applying various unsupervised and supervised machine-learning methods to near-field electromagnetic measurements, with the aim of determining whether device-level parameters and the prepared polarization states can be extracted from the observed emissions alone. The developed unsupervised machine-learning methods were unable to recover the distribution of polarization states from the electromagnetic measurements, partitioning a single dominant cluster into arbitrary groupings, rather than finding separate clusters. When trained on files with various distributions of key and check bases, the Ridge regression model exhibited strong predictive capability, accurately following the proportions of D-states during leave-one-out cross-validation, with a mean absolute error (MAE) of 4.9% and an R2 score of 0.946. During testing on new, unseen files, the MAE increased to 11.3% and the R2 score dropped to -0.12, largely attributed to a single file, with the remaining test files producing results more consistent with the cross-validation performance. Similarly, a convolutional neural network trained on fixed five-symbol sequences achieved an accuracy of 41.13% and a F1-score of 36.46%, which for the given symbol space is significantly above random guessing of 33.33%.
Quantum key distribution (QKD) is sometimes marketed as secure against computational attacks, as its security is based on the laws of quantum mechanics for detecting eavesdroppers, rather than the difficulty of mathematical problems. However, this security guarantee applies only to the quantum channel itself, and the physical hardware implementing these systems remain vulnerable. The electrical components controlling these systems generate electromagnetic emissions when active, and if those emissions carry information such as the polarization states being encoded, an attacker could recover key-relevant information without ever interacting with the quantum channel, bypassing the security guarantees of the protocol entirely. This work investigates whether such a vulnerability exists in a commercial QKD transmitter unit by applying various unsupervised and supervised machine-learning methods to near-field electromagnetic measurements, with the aim of determining whether device-level parameters and the prepared polarization states can be extracted from the observed emissions alone. The developed unsupervised machine-learning methods were unable to recover the distribution of polarization states from the electromagnetic measurements, partitioning a single dominant cluster into arbitrary groupings, rather than finding separate clusters. When trained on files with various distributions of key and check bases, the Ridge regression model exhibited strong predictive capability, accurately following the proportions of D-states during leave-one-out cross-validation, with a mean absolute error (MAE) of 4.9% and an R2 score of 0.946. During testing on new, unseen files, the MAE increased to 11.3% and the R2 score dropped to -0.12, largely attributed to a single file, with the remaining test files producing results more consistent with the cross-validation performance. Similarly, a convolutional neural network trained on fixed five-symbol sequences achieved an accuracy of 41.13% and a F1-score of 36.46%, which for the given symbol space is significantly above random guessing of 33.33%.
Abstract While quantum key distribution (QKD) is a theoretically secure way of growing quantum-safe encryption keys, many practical implementations are challenged due to various open attack vectors, resulting in many variations of QKD protocols. Side channels are one such vector that allows a passive or active eavesdropper to obtain QKD information leaked through practical devices. This paper assesses the feasibility and implications of extracting the raw secret key from far-field radiated emissions from the single-photon avalanche diodes used in a BB84 QKD quad-detector receiver. Enhancement of the attack was also demonstrated through the use of deep-learning model to distinguish radiated emissions due to the four polarized encoding states. To evaluate the severity of such side-channel attack, multi-class classification based on raw-data and pre-processed data is implemented and assessed. Results show that classifiers based on both raw-data and pre-processed features can discern variations of the electromagnetic emissions caused by specific orientations of the detectors within the receiver with an accuracy higher than 90%. This research proposes machine learning models as a technique to assess EM information leakage risk of QKD and highlights the feasibility of side-channel attacks in the far-field region, further emphasizing the need to utilise mechanisms to avoid electromagnetic radiation information leaks and measurement-device-independent QKD protocols.
While quantum key distribution (QKD) protocols are proven secure based on fundamental physical laws, side channels may allow an eavesdropper to obtain information unnoticeably. We present a profiled [1] side-channel attack using a deep convolutional neural network to analyze the classical, radio-frequency electromagnetic emissions from the electronics of a QKD sender. At a few centimeters from the device, we are able to recover virtually all information about the secret key. Furthermore, we can still observe traces of electromagnetic radiation from the device at distances of a few meters using a wideband antenna. Our methods are designed to be easily adaptable and may serve as a starting point for assessing the presence of this side channel for other devices.
Quantum Key Distribution (QKD) offers theoretically unbreakable security by leveraging quantum mechanics. However, practical implementation is challenged by environmental vulnerabilities, noise, and hardware imperfections. Recently, Machine Learning (ML) has emerged as a powerful tool to address these limitations and enhance the real-world viability of QKD systems. In this survey, we review ML techniques applied to improve QKD security and performance across five applications. First, parameter optimization, covering signal calibration, polarization alignment, phase stabilization, modulation state tuning, and post-processing enhancements to maximize secure key generation and minimize error rates. Second, attack detection, where ML models identify and classify quantum threats such as photon-number-splitting and Trojan-horse attacks. Third, protocol selection, leveraging ML to dynamically choose QKD protocols based on operational conditions. Fourth, key performance prediction of core metrics such as Secret Key Rate (SKR) and Quantum Bit Error Rate (QBER). Finally, quantum network management, optimizing large-scale QKD deployments through intelligent routing, node management, and resource allocation. Performance improvements are evaluated using accuracy, reduced QBER, and increased SKR. While ML shows significant potential for finance, government, and defense applications, challenges remain in scalability, computational demands, and real-world testing. Ongoing work should focus on lightweight, generalizable models and standardized benchmarks for practical ML-enhanced QKD deployment.
Quantum Key Distribution (QKD) promises theoretically unbreakable encryption by using fundamental quantum physics principles. QKD has emerged as a contender in the next generation of encryption protocols as the demand for more secure communication grows in our linked society. However, practical QKD implementations confront obstacles such as noise, interception attempts, and flaws in quantum hardware. We investigate the possibilities of Machine Learning (ML) techniques to improve the efficiency, resilience, and security of QKD protocols in this study. We provide a novel system that uses deep learning models to optimize photon polarization states and hence increase the key generation rate. Our approach detects and mitigates eavesdropping attempts using unsupervised learning algorithms, substantially enhancing the security of quantum communications. To achieve security, the BB84 protocol makes use of fundamental quantum mechanics features. Furthermore, we use reinforcement learning to adaptively alter system settings in real time to account for the quirks and noise profiles of different quantum channels. In comparison to typical QKD setups, experimental results show that our ML-enhanced QKD system is more resilient to known quantum attacks and provides a more consistent and efficient key distribution rate. Our findings indicate that the interaction of quantum physics and machine learning can pave the way for more secure and practical quantum communication networks in the near future.
Due to the fast-paced advancements of the quantum computing field, current public key cryptographic protocols need to be rethought. One promising solution is Quantum Key Distribution (QKD), whose security is robust to any computational advancement. The security of QKD relies on the laws of quantum mechanics, ensuring that any eavesdropper attempting to access the data will be detected. In particular, polarization-encoding QKD systems allow to generate secret keys in two remote locations to encode/decode information, using the State of Polarization (SOP) of single photons. Although its security has been theoretically proven, its practical implementation poses some challenges, part of which are the main focus of this PhD thesis. One relevant process in QKD systems is the synchronization between the transmitter and receiver, namely, the frame synchronization. This way, in this thesis, a frame synchronization method is developed and implemented, showing its potential for the use in prepare-and-measure QKD systems. The developed method has been shown to be effective, even with a low number of photons reaching the receiver. Furthermore, its architecture allows it to be adjusted for different attenuation regimes of the QKD system. The ability to encode data on the physical properties of single photons, in this case, polarization, highly depends on the capacity to accurately generate the required polarization states on demand. This way, an automatic SOP generation method, comprising an optical polarization monitoring system and a dedicated algorithm was developed and implemented, capable of generating four SOPs. This method reached an average Quantum Bit Error Rate (QBER) of 0.75%, which is well below the security threshold. After the SOP generation, and their transmission through the quantum channel, polarization alignment algorithms are required to compensate for the random rotations induced by birefringence in the optical fiber. In this thesis, a non-iterative polarization compensation method was implemented. This method requires only three QBER estimations to re-align the polarization, reaching average QBERs of 1.8%. Additionally, multiple iterative methods were implemented for comparison purposes, with the results clearly showing a better efficiency of the deterministic method. Beyond fiber-based transmissions, free-space transmissions were analysed due to their ability to reach sites without installed fiber and due to its potential for satellite-based quantum communications. This thesis demonstrates the feasibility of polarization-encoding QKD over free-space channels, by subjecting quantum signals to atmospheric turbulence to study its impact. In that sense, polarization variations over a 1.8 km outdoor free-space channel were monitored, to understand the effect of atmospheric turbulence on polarization. These results have shown that SOP variations in free-space links are slow, allowing their compensation. Furthermore, it was identified that, under conditions tested, the Secret Key Rate (SKR) could reach values up to 8.7×10ˉ⁴ bits per second, allowing QKD over free space for distances up to 2.31 km.
The Quantum Key Distribution (QKD) method leverages the principles of quantum mechanics to securely create and distribute private keys using quantum systems and an authenticated public classical channel. Despite offering informationtheoretical security, its physical implementations often suffer from unintended information leakage, known as side channels, which eavesdroppers can exploit to obtain the private key. This study introduces a classical-side-channel attack on the privacyamplification step of a general QKD protocol based on matrix hashing, utilizing machine-learning techniques to analyze powerconsumption leakage. Through multiple simulated scenarios, the study found that the gradient-boosting machine consistently outperformed other models, recovering the entire private key for high measuring-instrument sampling rates, regardless of the hashing matrix size and noise level tested. Additionally, the study proposes a strategy based on analyzing the confusion matrices of the model to facilitate a brute-force search for the critical feasible in the case of a non-perfect model. Furthermore, the study discusses countermeasures such as noise insertion, masking, and randomization techniques, potentially restoring the QKD protocol's information-theoretical security. This work demonstrates that machine-learning techniques can effectively characterize the leakages in a QKD implementation and create potent attacks.