Researchers from NUST MISIS, HSE University, and the Russian company QRate have introduced a new approach to predicting quantum error rates in quantum key distribution systems using machine learning algorithms.
“As part of the Priority 2030 national program, a research team at NUST MISIS led by Professor Alexey Ustinov, a globally recognized scientist, is implementing the strategic technological project ‘Quantum Internet.’ One of its key objectives is to create the conditions necessary for transitioning quantum technologies from laboratories into industry and developing competitive products with export potential. The new machine learning—based algorithm enables dynamic optimization of error correction in quantum key distribution systems, improving operational stability under non-ideal conditions. This development is an important step toward building scalable and practical quantum networks,” NUST MISIS Rector Alevtina Chernikova.
Quantum cryptography provides a very high level of data protection because any attempt to intercept information alters the quantum state of the system and cannot go undetected. However, the technology is highly sensitive to noise and equipment instability.
In high-speed quantum key distribution (QKD) systems, data streams must be processed almost in real time. This requires fast error correction codes that reveal as little information as possible about the key over the public channel. Selecting the optimal code depends, among other factors, on accurately predicting the initial error rate in the distributed key. The researchers proposed a new solution by training an algorithm to analyze QKD system performance and dynamically predict quantum error rates based on telemetry data.
“At the end of a QKD session, legitimate users obtain ‘raw’ keys that should be identical. However, due to natural noise or potential eavesdropping, these keys always contain errors, which are detected and corrected using special error correction codes. The keys are divided into small blocks, and checksums—known as syndromes—are exchanged over a public channel for each block. This makes it possible to identify and correct mismatched bits without revealing their values. The more auxiliary information required for this exchange, the slower and more vulnerable the process becomes. Our algorithm analyzes system telemetry in real time and selects the optimal error correction mode for each block,” Andrey Tayduganov, Head of the Laboratory of Quantum Communications Theory at NUST MISIS.
“We systematically evaluated modern methods using real-world datasets, which significantly expanded our available toolkit. The key advantage of our method is that it has been validated on real experimental data and is directly applicable to specific physical setups. Most approaches described in the literature are tested only in simulations, allowing them to achieve formally high performance before being validated with actual data,” Denis Derkach, Head of the Research and Training Laboratory for Big Data Analysis Methods at HSE University.
The new model takes into account not only the history of error rate fluctuations but also a range of additional system parameters, enabling it to quickly adapt to unexpected changes. Detailed results of the study are published in the scientific journal Physics of Particles and Nuclei.
The algorithm also analyzes error rates and detection probabilities of decoy laser pulses, which do not contribute to key generation but play an essential role in estimating parameters required to calculate the length of the final secret key. This makes it possible to detect sudden changes in the quantum channel or single-photon detectors at the receiver and incorporate this information for more accurate prediction of signal pulse error rates.
The research was carried out as part of the NUST MISIS strategic technological project Quantum Internet under the Priority 2030 program of the Ministry of Science and Higher Education of Russia (National Project “Youth and Children”), project No. K1-2022-027.
