A quantum key distribution system based on artificial neural networks

Abstract

This invention relates to a quantum key distribution system based on an improved artificial neural network, belonging to the field of quantum network security technology. The system includes a transmitter, a receiver, and a transmission channel. The quantum information generation module at the transmitter prepares quantum states and loads information to obtain quantum state light pulses loaded with key information, which are then sent to the receiver. The quantum state measurement module at the receiver receives the quantum state light pulses and decodes and detects them. After basis pairing is completed at the transmitter and receiver, the neural network analysis units at both ends detect the network security of the quantum key distribution system based on the transmitter and receiver data provided by the management and control modules at each end. The system determines whether to continue subsequent key distribution operations based on the detection results. This invention's system can automatically and efficiently complete network security detection during the quantum key distribution process with high accuracy.

Description

Technical Field

[0001] This invention belongs to the field of quantum network security technology, and in particular relates to a quantum key distribution system based on an improved artificial neural network. Background Technology

[0002] Currently, researchers, laboratories, and engineering practices widely use QKD systems based on the decoy state BB84 protocol for quantum key distribution. However, this QKD system suffers from significant errors: Alice, the QKD transmitter, uses a pulsed light source to provide the quantum state optical signal pulses, while Bob, the receiver, uses a single-photon detector to detect the single-photon horizontal optical signal. This process introduces state preparation errors. Alice uses a quantum state modulation module to modulate the quantum state signal, while Bob uses a quantum state demodulation module. This demodulation process introduces modulation errors. Furthermore, Alice uses the arrival time of the transmitted synchronization light pulse at Bob as a benchmark to measure the arrival time of the single-photon signal. The entire optical channel includes multiplexers, optical switches, and demultiplexers, leading to delays, attenuation, and overshoot at Bob's end, resulting in channel errors. Additionally, potential quantum channel eavesdroppers like Eve can intercept and retransmit the QKD system. The QKD protocol requires that qubits can communicate through a public channel when the bit error rate is below a certain threshold. Eve can exploit the errors in the QKD system to intercept key information in the quantum channel even with a bit error rate below the threshold. Therefore, further research is needed on the error and attack analysis of the QKD system.

[0003] Current technologies often rely on manual, iterative calculations to verify the errors and attack scenarios of QKD systems. This process can only be performed after quantum key distribution is complete, which results in delays and low efficiency for applications that have already been distributed. Summary of the Invention

[0004] Based on the above analysis, this invention aims to provide an improved quantum key distribution system based on artificial neural networks. By introducing artificial neural network units into existing QKD systems based on the decoy state BB84 protocol, an improved quantum key distribution system is obtained. This system automatically performs network security detection, which can be completed immediately after basis pairing, without waiting for the key information transmission process to finish. The system includes a transmitter, a receiver, and a transmission channel. The transmitting end includes a quantum state information generation module, a first control processing module, and a first neural network analysis unit; the receiving end includes a quantum state measurement module, a second control processing module, and a second neural network analysis unit. The quantum information generation module is used to realize quantum state preparation and information loading to obtain quantum state light pulses loaded with key information and send them to the receiving end through the transmission channel; The first control processing module is used to acquire the transmitting end data generated by the quantum information generation module and provide it to the first neural network analysis unit, and provide it to the second neural network analysis unit through the transmission channel; The quantum state measurement module is used to receive quantum state light pulses and to decode and detect the received quantum state light pulses; The second control processing module is used to acquire the receiving data generated by the quantum state measurement module and provide it to the second neural network analysis unit, and to provide it to the first neural network analysis unit through the transmission channel; The first neural network analysis unit and the second neural network analysis unit are used to detect network security of the system based on the data sent from the transmitting end and the data received from the receiving end.

[0005] Furthermore, the transmitting end and the receiving end perform base pairing based on the classical channel of the transmission channel; the network security of the detection system based on the data from the transmitting end and the data from the receiving end includes: After the base pairing is completed, the artificial neural networks in the first and second artificial neural network units work synchronously to perform network security detection based on the data from the sending end and the data from the receiving end, respectively, and feed back the detection results to the first control processing module and the second control processing module, respectively. When the detection result is a failure, the first control processing module controls the sending end to stop sending data, and the second control processing module controls the receiving end to discard the received data. The sending end and the receiving end will no longer perform subsequent key distribution operations. When the detection result is a success, the sending end and the receiving end will continue with subsequent key distribution operations.

[0006] Furthermore, the sending end data and the receiving end data include: The encoding parameters of the transmitting end {a n} and encoding base parameter {x n}; The measurement result parameter {b} of the receiving end n} and measurement basis parameters {y n}; The real-time state parameters of the quantum state information generation module and the quantum state measurement module when communicating through the quantum channel.

[0007] Furthermore, the subsequent operations of the key distribution include error correction by the sending end and the receiving end, and security enhancement to generate the final key.

[0008] Furthermore, the quantum state information generation module includes: Pulsed light source, used to generate coherent optical pulse signals; A decoy state modulation module is used to perform decoy state amplitude modulation on the optical pulse signal to obtain a decoy state modulated optical pulse signal; The quantum state modulation module is used to modulate the decoy-state modulated optical pulse using two sets of mutually conjugate orthogonal bases as encoding basis vectors to obtain four quantum state optical signals; An optical path adaptation monitoring module is used to attenuate the power of the quantum state optical signal and transmit it. A random number generator is used to generate random numbers as control signals for decoy state modulation and quantum state modulation. The quantum state measurement module includes: The optical path adaptation monitoring module is used to receive signals sent by the transmitting end optical path adaptation monitoring module, perform optical isolation and input optical power detection, and obtain the received quantum state optical signal; A quantum state demodulation module is used to randomly select a set of measurement basis vectors corresponding to the encoded basis vectors to demodulate the received quantum state optical signal, thereby obtaining a demodulated optical pulse signal; A single-photon detector is used to detect the demodulated optical pulse signal to obtain the received optical pulse signal.

[0009] Furthermore, the first and second artificial neural network units have the same artificial neural network structure and the same training method, the training method including: Construct various fault scenarios for the quantum key distribution system, as well as scenarios with and without eavesdropping when the system is fault-free; record the transmitting and receiving data when the system transmits quantum signals under each scenario, and construct training and validation sets respectively; The artificial neural network to be trained is trained and validated based on the training set and the validation set to obtain a trained artificial neural network.

[0010] Furthermore, the construction of various fault scenarios for the quantum key distribution system, as well as scenarios with and without eavesdropping when the system is fault-free, includes: The system is constructed by considering the following scenarios: failure of the pulsed light source module, quantum state modulation module, and optical path switch of the quantum channel. Specifically, failure of the pulsed light source module will cause state preparation errors during quantum signal transmission; failure of the quantum state modulation module will cause adjustment errors during quantum signal transmission; and failure of the optical path switch will cause channel errors during quantum signal transmission. When the system is fault-free and there is no monitoring, there is no error when the system transmits quantum signals. In the case of a listener present even when the system is fault-free, the listener may attempt to obtain key information through interception and retransmission attacks. In the case of a situation where the pulsed light source module, quantum state modulation module, and optical path switch of the quantum channel of the constructed system all fail simultaneously and there is an eavesdropper, the system is simultaneously introduced with state preparation error, modulation error, channel error, and interception and retransmission attack.

[0011] Furthermore, the process of recording the transmitting and receiving data during quantum signal transmission in each of the aforementioned scenarios to construct training and validation sets includes: Record the sending and receiving data in each of the above situations as input data for the training or validation set, respectively; The corresponding network security result data under each of the aforementioned conditions are recorded as output data for the training set or validation set, respectively. The network security result data includes bit error rate, state preparation error correlation, modulation error correlation, channel error correlation, interception and retransmission attack correlation, and network security detection results. The bit error rate is calculated by the system based on the basis result. The state preparation error correlation, modulation error correlation, channel error correlation, and interception and retransmission attack correlation are set to preset values. The network security detection results are determined based on the bit error rate and a set threshold.

[0012] Furthermore, during training, the method for determining the network security detection results includes: When an interception and retransmission attack is introduced, the network security test result is "failed". When state preparation error, modulation error, channel error and interception and retransmission attack are introduced, the network security test result is "failed". When one of the state preparation error, modulation error and channel error is introduced, the network security test result is determined based on the bit error rate and a set threshold. If the bit error rate is less than the set threshold, the network security test result is "passed"; otherwise, it is "failed".

[0013] Furthermore, the transmitting end also includes a synchronization transmission module for transmitting synchronization signals; the receiving end also includes a synchronization signal receiving module for receiving and detecting synchronization signals, and restoring the system clock signal synchronized with the transmitting end at the receiving end.

[0014] The present invention can achieve at least one of the following beneficial effects: An improved quantum key distribution system is obtained by introducing artificial neural network units to improve the existing QKD system. This system can automatically perform network security detection based on artificial neural networks, and the detection can be performed immediately after basis pairing, taking only a few seconds and not requiring the completion of key information transmission. Compared with the existing technology where manual error and attack analysis is performed after the key information transmission process is completed, and the entire key data needs to be discarded once an attack is detected, the method of this invention is more efficient and real-time.

[0015] By introducing a dual-hidden-layer artificial neural network that achieves an accuracy rate of over 90% for classification and identification problems, the accuracy of using artificial neural networks for network security detection can be guaranteed. This clearly displays statistical information on sent and received data to users, greatly improving the security of the quantum key distribution system.

[0016] By introducing state preparation error, modulation error, channel error, and interception / retransmission attack during the training of the artificial neural network, the system simulates the identification of potential quantum channel eavesdroppers like Eve's interception / retransmission attack. Furthermore, during the learning process of multiple error attack identification iterations, the actual output of each training iteration is compared with the expected output, thereby correcting the weights and thresholds of each relevant neuron. As the number of training iterations increases, the error between the actual and expected outputs decreases, meaning the error attack identification becomes more accurate, thus improving the accuracy of network security detection.

[0017] By introducing artificial neural network units to improve the system, the system has strong scalability. For new errors and new attack patterns encountered in the development of QKD system, the method of this invention only needs to add new training sets to the already trained neural network units for system training to identify new errors and new attack patterns.

[0018] Other features and advantages of the invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained from what is particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Figure 1 This is a schematic diagram of the quantum key distribution system of the present invention; Figure 2 This is a schematic diagram of the decoy state BB84 protocol flow for introducing an artificial neural network analysis unit into the QKD system in this invention; Figure 3 This is a schematic diagram of the structure of the dual hidden layer artificial neural network of the present invention; Figure 4 This is a flowchart of the artificial neural network training process of the present invention; Detailed Implementation Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0020] One specific embodiment of the present invention discloses an improved quantum key distribution system based on artificial neural networks. By introducing artificial neural network units into the existing QKD system based on the decoy state BB84 protocol process, an improved quantum key distribution system is obtained. This system automatically realizes network security detection, and the detection can be completed after basis pairing without waiting for the key information transmission process to be completed. Compared with the existing technology, which requires waiting for the key information transmission to be completed before manual error and attack analysis, this greatly improves the efficiency of network security detection.

[0021] The system in this embodiment includes a transmitter, a receiver, and a transmission channel, wherein the transmission channel includes a quantum channel and a classical channel; The transmitting end includes a quantum state information generation module, a first control processing module, and a first neural network analysis unit; the receiving end includes a quantum state measurement module, a second control processing module, and a second neural network analysis unit. The quantum information generation module is used to prepare quantum states and load information to obtain quantum state light pulses loaded with key information and send them to the receiving end through a quantum channel. The first control processing module is used to acquire the transmitting end data generated by the quantum information generation module and provide it to the first neural network analysis unit, and provide it to the second neural network analysis unit through a classical channel; The quantum state measurement module is used to receive quantum state light pulses and to decode and detect the received quantum state light pulses; The second control processing module is used to acquire the receiving data generated by the quantum state measurement module and provide it to the second neural network analysis unit, and to provide it to the first neural network analysis unit through a classical channel; The first neural network analysis unit and the second neural network analysis unit are used to detect network security of the system based on the transmitting end data and the receiving end data; wherein, the transmitting end data and the receiving end data include encoded parameters {a n} and encoding base parameter {x n};Measurement result parameter {b n} and measurement basis parameters {y n The quantum state information generation module and the quantum state measurement module transmit real-time state parameters during communication via the quantum channel. For example... Figure 1 This is a schematic diagram of the quantum key distribution system of the present invention.

[0022] Furthermore, the transmitter and receiver are based on a classical channel pair; network security for the detection system based on transmitter and receiver data includes: After the base pairing is completed, the artificial neural networks in the first and second artificial neural network units work synchronously, performing network security checks based on the data from the sending end and the data from the receiving end, respectively, and feeding back the detection results to the first and second control processing modules, respectively. When the detection result is a failure, the first control processing module controls the sending end to stop sending data, and the second control processing module controls the receiving end to discard the received data. Neither the sending nor the receiving end will perform any further key distribution operations. When the detection result is a success, the sending and receiving ends continue with the subsequent key distribution operations, including error correction and security enhancement to generate the final key. Figure 2 A schematic diagram of the decoy state BB84 protocol flow for introducing an artificial neural network analysis unit into a QKD system.

[0023] It's important to note that before detecting network security vulnerabilities, the artificial neural networks in the first and second artificial neural network units need to be trained. These two artificial neural networks must have identical structures, use the same datasets, and employ the same training methods. Therefore, the two trained artificial neural networks, with identical structures and parameters, can produce the same output based on the same inputs and can operate synchronously. Consequently, during network security detection, the first and second artificial neural network units do not need to transmit data to each other, further improving detection efficiency.

[0024] In this embodiment, an improved quantum key distribution system is obtained by introducing an artificial neural network unit to improve the existing QKD system. This improved system can automatically perform network security detection based on the artificial neural network, and the detection can be performed immediately after basis pairing, taking only a few seconds and not requiring the completion of key information transmission. Compared to existing technologies where manual error and attack analysis is performed after key information transmission, and where the entire key data needs to be discarded upon detection of an attack, the method of this invention is more efficient and real-time.

[0025] In one specific embodiment of the present invention, the quantum state information generation module includes: Pulsed light source, used to generate coherent optical pulse signals; A decoy state modulation module is used to perform decoy state amplitude modulation on the optical pulse signal to obtain a decoy state modulated optical pulse signal; The quantum state modulation module is used to modulate the decoy-state modulated optical pulse using two sets of mutually conjugate orthogonal bases as encoding basis vectors to obtain four quantum state optical signals; An optical path adaptation monitoring module is used to attenuate the power of the quantum state optical signal and transmit it. A random number generator is used to generate random numbers as control signals for decoy state modulation and quantum state modulation.

[0026] The quantum state measurement module includes: The optical path adaptation monitoring module is used to receive signals sent by the transmitting end optical path adaptation monitoring module, perform optical isolation and input optical power detection, and obtain the received quantum state optical signal; A quantum state demodulation module is used to randomly select a set of measurement basis vectors corresponding to the encoded basis vectors to demodulate the received quantum state optical signal, thereby obtaining a demodulated optical pulse signal; A single-photon detector is used to detect the demodulated optical pulse signal to obtain the received optical pulse signal.

[0027] The modules of the quantum state information generation module work in coordination to prepare quantum states and load information to obtain quantum state light pulses loaded with key information, which are then sent to the receiving end through the transmission channel. The modules of the quantum state measurement module work in coordination to receive quantum state light pulses and decode and detect the received quantum state light pulses.

[0028] In this embodiment, the QKD system is implemented modularly, giving it strong scalability. The modular QKD system incorporates artificial neural network units, making it easy to implement. Furthermore, for new errors and attack patterns encountered during the development of the QKD system, the artificial neural network units only need to add new training sets for system training to identify these new errors and attack patterns, demonstrating good scalability and practicality.

[0029] In one specific embodiment of the present invention, the training method for the artificial neural networks in the first and second artificial neural network units includes: Construct various fault scenarios for the quantum key distribution system, as well as scenarios with and without eavesdropping when the system is fault-free; record the transmitting and receiving data when the system transmits quantum signals under each scenario, and construct training and validation sets respectively; The artificial neural network to be trained is trained and validated based on the training set and the validation set to obtain a trained artificial neural network.

[0030] Furthermore, various fault scenarios of the quantum key distribution system are constructed, as well as scenarios with and without eavesdropping when the system is fault-free, including: The system is constructed by considering the following scenarios: failure of the pulsed light source module, quantum state modulation module, and optical path switch of the quantum channel. Specifically, failure of the pulsed light source module will cause state preparation errors during quantum signal transmission; failure of the quantum state modulation module will cause adjustment errors during quantum signal transmission; and failure of the optical path switch will cause channel errors during quantum signal transmission. When the system is fault-free and there is no monitoring, there is no error when the system transmits quantum signals. In the case of a listener present even when the system is fault-free, the listener may attempt to obtain key information through interception and retransmission attacks. In the case of a situation where the pulsed light source module, quantum state modulation module, and optical path switch of the quantum channel of the constructed system all fail simultaneously and there is an eavesdropper, the system is simultaneously introduced with state preparation error, modulation error, channel error, and interception and retransmission attack.

[0031] It should be noted that the method for constructing faults in each module is as follows: A component with partial hardware failure is used in the module, so that although the module is still running, it will cause corresponding transmission errors. For example, constructing a fault in the pulse source module is based on using the hardware related to the faulty pulse source, such as using a DFB laser with partial hardware failure; constructing a fault in the quantum state modulation module is based on the hardware related to the faulty quantum state adjustment module, such as using a random number generator with partial hardware failure; constructing a fault in the optical path switch is based on the hardware implementation of the faulty optical path switch, such as constructing poor contact between the optical path switch and the optical path demultiplexer.

[0032] Furthermore, the transmitting and receiving data during quantum signal transmission in each of the aforementioned cases are recorded to construct training and validation sets, respectively, including: Record the sending and receiving data in each of the above situations as input data for the training or validation set, respectively; The corresponding network security result data under each of the aforementioned conditions are recorded as output data for the training set or validation set, respectively. The network security result data includes bit error rate, state preparation error correlation, modulation error correlation, channel error correlation, interception and retransmission attack correlation, and network security detection results. The bit error rate is calculated by the system based on the basis result. The state preparation error correlation, modulation error correlation, channel error correlation, and interception and retransmission attack correlation are set to preset values. The network security detection results are determined based on the bit error rate and a set threshold.

[0033] Furthermore, the relevant parameters of the sending and receiving ends in each of the aforementioned cases are recorded as input data for the training set or validation set, including: The encoding parameters of the QKD system transmitter in each of the described cases {a n} and encoding base parameter {x n}; Measurement result parameters {b} of the QKD system receiver in each of the aforementioned cases n} and measurement basis parameters {y n}; Real-time state parameters of each module of the QKD system during quantum channel communication at the transmitting and receiving ends under each of the described conditions.

[0034] Preferably, the real-time status parameters of each module during quantum channel communication include: status parameters of the transmitting end control and processing module, status parameters of the transmitting end synchronization signal transmission module, status parameters of the transmitting end pulse light source, status parameters of the transmitting end quantum state modulation module, status parameters of the transmitting end optical path adaptation monitoring module, status parameters of the channel optical path switch, status parameters of the receiving end synchronization signal receiving module, status parameters of the receiving end control and processing module, status parameters of the receiving end optical path adaptation monitoring module, status parameters of the receiving end quantum state demodulation module, and status parameters of the receiving end single-photon detector.

[0035] Furthermore, the corresponding network security result data in each of the aforementioned situations are recorded as output data for the training set or validation set, respectively.

[0036] For example, in each of the aforementioned cases: when a single state preparation error is introduced, the state preparation error correlation value is preferably preset to 0.99, and the modulation error correlation, channel error correlation, and interception / retransmission attack correlation are preset to 0; when a single modulation error is introduced, the modulation error correlation is preferably preset to 0.99, and the state preparation error correlation, channel error correlation, and interception / retransmission attack correlation are preset to 0; when a single channel error is introduced, the channel error correlation is preferably preset to 0.99, and the state preparation error correlation, modulation error correlation, and interception / retransmission attack correlation are preset to 0; when an interception / retransmission attack is introduced, the interception / retransmission attack correlation is preferably preset to 0.99, and the state preparation error correlation, modulation error correlation, and channel error correlation are preset to 0; when state preparation error, modulation error, channel error, and interception / retransmission attack are introduced simultaneously, the preset value for each correlation is 0.25.

[0037] Specifically, the security detection results under each of the aforementioned conditions include: network security detection passed or network security detection failed. Specifically, when an interception and retransmission attack is introduced, the network security detection result is failed; when all error conditions and interception and retransmission attacks are introduced, the network security detection result is failed; for each condition introducing a single error, the network security monitoring result is determined based on the bit error rate of the corresponding condition. If the bit error rate in the corresponding condition is less than a set threshold, the network security detection result is passed; otherwise, it is failed.

[0038] It should be noted that the threshold value is determined based on transmission requirements. When the QKD system serves fields with high data transmission accuracy requirements, the threshold value is smaller, indicating that the lower the bit error rate, the higher the transmission accuracy. When the QKD system serves fields with low transmission accuracy requirements, the threshold value can be appropriately increased according to the needs.

[0039] Specifically, the process of constructing the training set in this embodiment includes steps S011 to S013: S011, the sending end Alice selects a random sequence {a} of length N.n},{x n}, based on these two sequences, prepare N single photons and send them to the receiver Bob; where {x n} determines the direction of the coding base, {a n} represents the encoding parameters; Bob declares reception after receiving a single photon and selects a random sequence {y} of length N. n} as the measurement base direction, for {a n} Perform the measurement and record the result as {b} n Simultaneously, it records the real-time state parameters and network security results data of each module during quantum channel communication.

[0040] S012. Repeat S011 multiple times to record all relevant parameters of the sending and receiving ends and the corresponding network security results data for each iteration, which will serve as the dataset D1 for the QKD system under fault-free and non-monitoring conditions.

[0041] S013. Construct the following scenarios for the QKD system: the pulse source module, quantum state modulation module, and optical path switch are faulty; the scenario where there is no fault and the eavesdropper intercepts and retransmits the signal; and the scenario where all the faults occur simultaneously and an eavesdropper is present. Repeat S011 to S012 under each scenario to construct the corresponding datasets D2, D3, D4, D5, and D6.

[0042] Furthermore, the process of constructing the validation set in this embodiment is the same as the process of constructing the training set.

[0043] Specifically, the artificial neural network is trained using datasets D1-D6. Preferably, the artificial neural network is a backpropagation (BP) neural network; alternatively, other types of neural networks may be used. Figure 3 This is a schematic diagram of the structure of the dual-hidden-layer artificial neural network used in this embodiment. This artificial neural network consists of d input neurons... l A multilayer feedforward network structure consisting of one output neuron and 2q hidden layer neurons.

[0044] Specifically, the training process of the artificial neural network in this embodiment includes steps S021 to S026: S021. Randomly initialize the connection weights and thresholds in the network; specifically, randomly initialize the connection weights and thresholds in the network within the range of (0, 1).

[0045] S022. Given a training set D=D1.

[0046] S023. Based on the input and output data of each sample in the training set D, the connection weights and thresholds of each layer in the network are corrected.

[0047] Specifically, assume that a single training example in the training set D is represented as ( x k, y k For training examples ( x k , y k ),calculate x k The result of the input layer entering the artificial neural network, passing through two hidden layers, and finally reaching the output layer. ;assumed The mean square error on an artificial neural network is expressed as: ; The error E in the above equation k Given a learning rate η(0,1), then: ; in, This represents the change in the connection weights of the j-th neuron in the output layer. It is the input of the j-th neuron in the output layer; , This is the output of the h-th neuron in the second hidden layer; The gradient term of the output layer neuron can be obtained as follows: ; Then the gradient term e of the second hidden layer neuron h and the gradient term c of the first hidden layer neurons h They are represented as follows: ; ; Then, the change in the connection weights of the output layer: ; Output layer threshold change: ; Changes in the connection weights of the second hidden layer: ; Threshold change in the second hidden layer: ; Changes in the connection weights of the first hidden layer: ; Threshold change in the first hidden layer: ; The connection weights and thresholds of each layer are adjusted to complete the training of this sample.

[0048] Specifically, training is completed on all samples in the training set.

[0049] S024. Given a training set D = D2∪D3∪D4∪D5, repeat S023 for training. This training set contains only a single error or attack scenario.

[0050] S025. Given a training set D=D6, repeat S023 for training. This training set is a training set under complex conditions that include various errors and attacks.

[0051] S026. Verify the accuracy, precision, and recall of the artificial neural network output based on the validation set; if they all meet the set threshold, determine that the artificial neural network is well trained; otherwise, repeat steps S022 to S026.

[0052] Specifically, the methods for calculating accuracy and precision are as follows: Accuracy:

[0053] Accuracy:

[0054] Where P refers to the number of positive samples, N refers to the number of negative samples, TP refers to the number of correctly predicted positive samples, FP refers to the number of negative samples predicted as positive samples, FN refers to the number of positive samples predicted as negative samples, and TN refers to the number of correctly predicted positive samples.

[0055] Specifically, Figure 4 This is a flowchart of the artificial neural network training process in this embodiment.

[0056] It should be noted that the standards typically adopted in the existing process of manually verifying the network security of QKD systems are: If there is an interception and retransmission attack in the QKD system during this communication, the identification result will be "fail". If there is no interception and retransmission attack in the QKD system during this communication, but the bit error rate is higher than the set threshold, the identification result is failure. If there is no interception and retransmission attack in the QKD system during this communication, and the bit error rate is lower than the set threshold, then the identification result is passed.

[0057] In this embodiment, by introducing state preparation error, modulation error, channel error, and intercept / retransmission attack into the training and learning process of an artificial neural network, the system simulates the detection of a potential quantum channel eavesdropper, Eve, through intercept / retransmission attacks. During the learning process of multiple error attack detection iterations, the actual output of each training iteration is compared with the expected output, thereby correcting the weights and thresholds of each relevant neuron. As the number of training iterations increases, the error between the actual output and the expected output decreases, meaning the error attack detection becomes increasingly accurate, achieving the goal of network security detection of a key distribution system based on an artificial neural network.

[0058] In this embodiment, by introducing a dual-hidden-layer artificial neural network that achieves an accuracy rate of over 90% for classification and identification problems, the accuracy of using artificial neural networks for network security detection can be guaranteed. The system clearly displays statistical information on sent and received data to the user, greatly improving the security of the quantum key distribution system.

[0059] In a specific embodiment of the present invention, detecting network security of the QKD system by transmitting and receiving data during quantum key distribution based on the QKD system refers to the process of identifying whether there is a potential quantum channel eavesdropper Eve through the artificial neural networks in the first and second artificial neural network units.

[0060] This embodiment performs network security detection based on the decoy BB84 protocol process, including: The transmitter of the QKD system modulates the quantum state used to load key information onto the corresponding optical pulse; Alice, the transmitting end, modulates the quantum state used to load key information onto the corresponding optical pulse through quantum state preparation and information loading. Alice, the transmitter, sends light pulses to Bob, the receiver, via a quantum channel. Bob at the receiving end performs quantum state measurements; Alice, the sender, and Bob, the receiver, perform base pairing. A trained artificial neural network is used to perform network security detection based on the data sent by Alice at the sending end and the data received by Bob at the receiving end, and the corresponding network security result data is obtained. The process of determining whether to continue distributing keys based on the corresponding network security results includes: when the network security detection result is "failed", the receiving end discards the received data and stops subsequent operations; when the network security detection result is "passed", the sending end and receiving end perform error correction and security enhancement to generate the final key; wherein, error correction includes: the sending end and receiving end analyze the filtered key, estimate the qubit error rate and phase error rate, correct the errors, and obtain the error-correcting key.

[0061] In this embodiment, a trained artificial neural network is used to automatically perform network security detection on the key distribution system. This replaces the complex work of manual analysis of QKD system errors and attacks, as well as the large number of iterative calculations required for bit error rate indicators. This can effectively improve the efficiency of QKD system error and attack analysis.

[0062] In one specific embodiment of the present invention, the transmitting end of the quantum key distribution system based on artificial neural networks further includes a synchronization transmission module for transmitting a synchronization signal, that is, generating an optical signal for transmitting synchronization clock information under the control of the transmitting end clock; the receiving end further includes a synchronization signal receiving module for receiving and detecting the synchronization signal, and restoring the system clock signal synchronized with the transmitting end at the receiving end as the receiving end trigger control signal.

[0063] Furthermore, both the sending and receiving ends of the system include negotiation signal transceiver modules for processing after quantum key distribution.

[0064] Furthermore, both the sending and receiving ends of the system include a key interface module, which is used to output the key to the key management system or encryption application system after the quantum key is generated.

[0065] Furthermore, both the sending and receiving ends of the system include management interface modules, which are used to implement interface functions with the upper-layer network management system or local maintenance functions.

[0066] Furthermore, the quantum state measurement module at the receiver end of the system may optionally include a random number generator to generate random numbers as control signals for demodulating quantum state optical signals.

[0067] Furthermore, the system also includes a multiplexer and a demultiplexer to enable wavelength division multiplexing and demultiplexing reception of quantum state optical signals with other optical signals.

[0068] Furthermore, the system also includes an optical path switch, which enables the optical path switching of quantum state optical signals with other optical signals.

[0069] This embodiment discloses a quantum key distribution system based on an improved artificial neural network. It includes common working modules found in existing QKD systems, excluding the artificial neural network analysis unit. This enables the improved quantum key distribution system to complete quantum key distribution based on the decoy state BB84 protocol flow and to automatically perform network security detection based on the artificial neural network. Since the working modules in this embodiment are existing technologies, they will not be described in detail here.

[0070] It should be noted that the above embodiments are based on the same inventive concept, and any parts not described repeatedly can be referenced from each other.

[0071] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A quantum key distribution system based on an improved artificial neural network, comprising a transmitter, a receiver, and a transmission channel, characterized in that: The transmitting end includes a quantum state information generation module, a first control processing module, and a first neural network analysis unit; the receiving end includes a quantum state measurement module, a second control processing module, and a second neural network analysis unit. The quantum state information generation module is used to prepare quantum states and load information to obtain quantum state light pulses loaded with key information and send them to the receiving end through the transmission channel. The first control processing module is used to acquire the transmitting end data generated by the quantum state information generation module and provide it to the first neural network analysis unit, and provide it to the second neural network analysis unit through the transmission channel; The quantum state measurement module is used to receive quantum state light pulses and to decode and detect the received quantum state light pulses; The second control processing module is used to acquire the receiving data generated by the quantum state measurement module and provide it to the second neural network analysis unit, and to provide it to the first neural network analysis unit through the transmission channel; The first neural network analysis unit and the second neural network analysis unit are used to detect network security of the system based on the data sent from the transmitting end and the data received from the receiving end.

2. The quantum key distribution system according to claim 1, characterized in that, The transmitting end and the receiving end perform base pairing based on the classical channel of the transmission channel; the network security of the detection system based on the data from the transmitting end and the data from the receiving end includes: After the base pairing is completed, the artificial neural networks in the first and second artificial neural network units work synchronously to perform network security detection based on the data from the sending end and the data from the receiving end, respectively, and feed back the detection results to the first control processing module and the second control processing module, respectively. When the detection result is a failure, the first control processing module controls the sending end to stop sending data, and the second control processing module controls the receiving end to discard the received data. The sending end and the receiving end will no longer perform subsequent key distribution operations. When the detection result is a success, the sending end and the receiving end will continue with subsequent key distribution operations.

3. The quantum key distribution system according to claim 2, characterized in that, The transmitting end data and the receiving end data include: The encoding parameters of the transmitting end {a n } and encoding base parameter {x n }; The measurement result parameter {b} of the receiving end n } and measurement basis parameters {y n }; The real-time state parameters of the quantum state information generation module and the quantum state measurement module when communicating through the quantum channel.

4. The quantum key distribution system according to claim 3, characterized in that, The subsequent operations of the key distribution include error correction by the sending end and the receiving end, and security enhancement to generate the final key.

5. The quantum key distribution system according to claim 4, characterized in that, The quantum state information generation module includes: Pulsed light source, used to generate coherent optical pulse signals; A decoy state modulation module is used to perform decoy state amplitude modulation on the optical pulse signal to obtain a decoy state modulated optical pulse signal; The quantum state modulation module is used to modulate the decoy-state modulated optical pulse using two sets of mutually conjugate orthogonal bases as encoding basis vectors to obtain four quantum state optical signals; An optical path adaptation monitoring module is used to attenuate the power of the quantum state optical signal and transmit it. A random number generator is used to generate random numbers as control signals for decoy state modulation and quantum state modulation. The quantum state measurement module includes: The optical path adaptation monitoring module is used to receive signals sent by the transmitting end optical path adaptation monitoring module, perform optical isolation and input optical power detection, and obtain the received quantum state optical signal; A quantum state demodulation module is used to randomly select a set of measurement basis vectors corresponding to the encoded basis vectors to demodulate the received quantum state optical signal, thereby obtaining a demodulated optical pulse signal; A single-photon detector is used to detect the demodulated optical pulse signal to obtain the received optical pulse signal.

6. The quantum key distribution system according to claim 5, characterized in that, The first and second artificial neural network units have the same artificial neural network structure and the same training method, which includes: Construct various fault scenarios for the quantum key distribution system, as well as scenarios with and without eavesdropping when the system is fault-free; record the transmitting and receiving data when the system transmits quantum signals under each scenario, and construct training and validation sets respectively; The artificial neural network to be trained is trained and validated based on the training set and the validation set to obtain a trained artificial neural network.

7. The quantum key distribution system according to claim 6, characterized in that, The various fault scenarios of the quantum key distribution system, as well as the scenarios with and without eavesdropping when the system is fault-free, are described below: The system is constructed by considering the following scenarios: failure of the pulsed light source module, quantum state modulation module, and optical path switch of the quantum channel. Specifically, failure of the pulsed light source module will cause state preparation errors during quantum signal transmission; failure of the quantum state modulation module will cause adjustment errors during quantum signal transmission; and failure of the optical path switch will cause channel errors during quantum signal transmission. When the system is fault-free and there is no monitoring, there is no error when the system transmits quantum signals. In the case of a listener present even when the system is fault-free, the listener may attempt to obtain key information through interception and retransmission attacks. In the case of a situation where the pulsed light source module, quantum state modulation module, and optical path switch of the quantum channel of the constructed system all fail simultaneously and there is an eavesdropper, the system is simultaneously introduced with state preparation error, modulation error, channel error, and interception and retransmission attack.

8. The quantum key distribution system according to claim 7, characterized in that, The process of recording the transmitting and receiving data during quantum signal transmission in each of the aforementioned scenarios, and constructing training and validation sets respectively, includes: Record the sending and receiving data in each of the above situations as input data for the training or validation set, respectively; The corresponding network security result data under each of the aforementioned conditions are recorded as output data for the training set or validation set, respectively. The network security result data includes bit error rate, state preparation error correlation, modulation error correlation, channel error correlation, interception and retransmission attack correlation, and network security detection results. The bit error rate is calculated by the system based on the basis result. The state preparation error correlation, modulation error correlation, channel error correlation, and interception and retransmission attack correlation are set to preset values. The network security detection results are determined based on the bit error rate and a set threshold.

9. The quantum key distribution system according to claim 8, characterized in that, During training, the method for determining the network security detection results includes: When an interception and retransmission attack is introduced, the network security test result is "failed". When state preparation error, modulation error, channel error and interception and retransmission attack are introduced, the network security test result is "failed". When one of the state preparation error, modulation error and channel error is introduced, the network security test result is determined based on the bit error rate and a set threshold. If the bit error rate is less than the set threshold, the network security test result is "passed"; otherwise, it is "failed".

10. The quantum key distribution system according to any one of claims 1-9, characterized in that, The transmitting end also includes a synchronization transmission module for transmitting synchronization signals; the receiving end also includes a synchronization signal receiving module for receiving and detecting synchronization signals, and restoring the system clock signal synchronized with the transmitting end at the receiving end.

Images