For a practical quantum key distribution (QKD) system, parameter optimization -the choice of intensities and probabilities of sending them -is a crucial step in gaining optimal performance, especially when one realistically considers finite communication time. With the increasing interest in the field to implement QKD over free-space on moving platforms, such as drones, handheld systems, and even satellites, one needs to perform parameter optimization with low latency and with very limited computing power. Moreover, with the advent of the Internet of Things (IoT), a highly attractive direction of QKD could be a quantum network with multiple devices and numerous connections, which provides a huge computational challenge for the controller that optimizes parameters for a large-scale network. Traditionally, such an optimization relies on brute-force search, or local search algorithms, which are computationally intensive, and will be slow on low-power platforms (which increases latency in the system) or infeasible for even moderately large networks. In this work we present a new method that uses a neural network to directly predict the optimal parameters for QKD systems. We test our machine learning algorithm on hardware devices including a Raspberry Pi 3 single-board-computer (similar devices are commonly used on drones) and a mobile phone, both of which have a power consumption of less than 5 watts, and we find a speedup of up to 100-1000 times when compared to standard local search algorithms. The predicted parameters are highly accurate and can preserve over 95-99% of the optimal secure key rate. Moreover, our approach is highly general and not limited to any specific QKD protocol.
I. BACKGROUND
A. Parameter Optimization in QKDQuantum key distribution (QKD)[1-4] provides unconditional security in generating a pair of secure key between two parties, Alice and Bob. To address imperfections in realistic source and detectors, decoy-state QKD [5][6][7] uses multiple intensities to estimate single-photon contributions, and allows the secure use of Weak Coherent Pulse (WCP) sources, while measurement-deviceindependent QKD (MDI-QKD) [8] addresses susceptibility of detectors to hacking by eliminating detector side channels and allowing Alice and Bob to send signals to an untrusted third party, Charles, who performs the measurement.In reality, a QKD experiment always has a limited transmission time, therefore the total number of signals is finite. This means that, when estimating the single-photon contributions with decoy-state analysis, one would need to take into consideration the statistical fluctuations of the observables: the Gain and Quantum Bit Error Rate (QBER). This is called the finite-key analysis of QKD. When considering finite-size effects, the choice of intensities and probabilities of sending these intensities is crucial to getting the optimal rate. Therefore, we would need to perform optimizations for the search of parameters.Traditionally, the optimization of parameters is implemented as either a brute-for...