Deep learning-based cv-qkd system optimal parameter prediction method, device, equipment and medium

Abstract

This invention relates to the field of quantum key distribution technology, and provides a method, apparatus, device, and medium for predicting optimal parameters of a CV-QKD system based on deep learning. This method utilizes a deep learning neural network to accelerate the parameter optimization process of the CV-QKD system. An online predictor is trained from offline data and then used to quickly search for optimal parameters online, avoiding the inefficient process of exhaustive parameter tuning in experiments. In implementation, this invention can run offline before experiments to provide optimization guidance parameters; it can also be embedded in the CV-QKD system for real-time operation, dynamically updating the optimal settings by continuously acquiring new channel parameters to achieve closed-loop feedback control.

Description

Technical Field

[0001] This invention relates to the field of quantum key distribution technology, and more specifically, to a method, apparatus, device, and medium for predicting optimal parameters of a deep learning-based CV-QKD system. Background Technology

[0002] Quantum Key Distribution (QKD) technology provides unconditionally secure keys to both communicating parties, representing a significant innovation in information security. Continuous Variable Quantum Key Distribution (CV-QKD) systems utilize mature classical coherent optical communication devices, offering high secure code rates over short to medium distances and good compatibility with classical optical communication networks, making them an important branch of QKD. However, in practical CV-QKD systems, channel noise dynamically changes with environmental factors. If system parameters (such as modulation variance and error correction matrix selection) are not optimized accordingly, the secure code rate will decrease. Traditionally, optimizing CV-QKD system parameters relies on offline simulation or exhaustive / iterative search methods. These methods are computationally intensive and time-consuming, often preventing practical systems from adjusting to optimal parameters in real-time operation, forcing the use of suboptimal or even unoptimized fixed parameters. In practical systems, the time allotted for a single round of key distribution is limited, and exhaustive search for optimal parameters significantly increases latency, making it impossible to adjust system parameters to the optimal state in a timely manner. Simultaneous parameter optimization in large-scale quantum network nodes such as access networks is virtually infeasible. Therefore, existing CV-QKD systems face severe challenges in real-time adaptive optimization.

[0003] In recent years, the introduction of deep learning, especially neural network-based methods, has provided new solutions for real-time adaptive optimization of CV-QKD systems. Research shows that pre-trained neural network models can directly predict near-globally optimal parameter combinations based on given channel and system parameters, with computation speeds hundreds of times faster than traditional local search algorithms. Combining deep learning with CV-QKD to automatically predict and adjust system parameters can significantly improve the system's secure code rate and stable operation, representing a key approach to meeting the real-time optimization needs of future large-scale quantum communication.

[0004] However, existing methods are mostly targeted at specific single parameters (such as laser intensity control or modulation variance optimization), or are limited to software-level verification in terms of implementation. There is no complete method + device solution to cover the entire process from channel monitoring, deep learning prediction to automatic device parameter tuning. Summary of the Invention

[0005] To address the challenges of parameter tuning difficulties, sensitivity to channel changes, and low optimization efficiency in CV-QKD systems, this invention provides a deep learning-based method, apparatus, device, and medium for predicting optimal parameters in CV-QKD systems. In dynamic channel environments, based on real-time acquired channel parameters, it intelligently predicts the modulation variance and error correction matrix combination that maximizes the safe code rate. This avoids the lag in manual adjustments and the inefficiency of simulation searches, thereby achieving adaptive optimization and rapid convergence of system parameters, significantly improving the performance and stability of CV-QKD systems.

[0006] In a first aspect, the present invention provides a method for predicting optimal parameters of a CV-QKD system based on deep learning, comprising: Obtain the FER-SNR performance curves of the CV-QKD system under different error correction matrices; Establishing modulation variance based on FER-SNR performance curve V A Mapping relationship with SKR (Secure Bitrate) ; Using mapping relationships The constructed offline optimizer outputs labels (H * , V A * SKR * ), including the optimal error correction matrix H * Optimal modulation variance V A * and optimal safe bitrate SKR * ; Using the labels (H) output by the offline optimizer * , V A * SKR * Training an online predictor based on a neural network model; Call the online predictor to infer the globally optimal parameter combination And confidence level, based on the confidence level, decide whether to adopt the globally optimal parameter combination. ; Do not adopt the globally optimal parameter combination The offline optimizer is invoked to optimize the online predictor. Based on the adopted globally optimal parameter combination Output results.

[0007] In a preferred embodiment, the mapping relationship between modulation variance and secure code rate (SKR) is established based on the FER-SNR performance curve. ,include: By training a neural network to approximate the FER-SNR performance curve, an FER predictor for any signal-to-noise ratio (SNR) is obtained. Combining FER predictor and signal-to-noise ratio (SNR) with modulation variance V A mapping relationship Based on the secure code rate formula of the CV-QKD system, the modulation variance is established. V A Mapping relationship with SKR (Secure Bitrate) .

[0008] In a preferred embodiment, modulation variance V A Mapping relationship with SKR (Secure Bitrate) Represented as:

[0009] in, This indicates the valid data during the key extraction process. n Total data N proportion, Represents the mutual information and modulation variance between Alice and Bob. V A The functional relationship, Indicates coordination efficiency and modulation variance V A The functional relationship, This represents the amount of information Eve can acquire and the modulation variance. V A The functional relationship.

[0010] In a preferred embodiment, the use of mapping relationships The constructed offline optimizer outputs labels (H * , V A * SKR * ),include: Random sampling system parameters; For each randomly sampled group of system parameters, the modulation variance is calculated using the secure bit rate formula. V A Establish modulation variance within the range of values. V A The mapping relationship between the error correction matrix H and each error correction matrix H; Based on this mapping relationship, each error correction matrix is ​​evaluated one by one. The optimal modulation variance corresponding to each error correction matrix is ​​obtained in this way. V A *and optimal safe bitrate SKR * ; Then, the global optimum is selected among the various error correction matrices H to obtain the label (H) under each set of system parameters. * , V A * SKR * ).

[0011] In a preferred embodiment, the label (H) output by the offline optimizer is used * , V A * SKR * Training an online predictor based on a neural network model includes: The online predictor, built on a neural network model, consists of two fully connected layers sharing a backbone and three outputs; these three outputs are the optimal error correction matrix H. * Classification, optimal modulation variance V A * Regression and Optimal Safe Bitrate SKR * return; The normalized vector of input system parameters and the labels output by the offline optimizer (H * , V A * SKR * The online predictor is trained using the Adam optimizer and early stopping; the loss function during training is: ; Where L is the total loss and CE is the optimal error correction matrix H. * Classification loss, For optimal modulation variance V A * Regression loss, SKR for optimal safe bitrate * Regression loss, For optimal modulation variance V A * Weights of regression loss SKR for optimal safe bitrate * Weights of regression loss This is the weight decay term. This is the weight decay coefficient. For model parameters The square of the L2 norm.

[0012] In a preferred embodiment, the step of invoking the offline optimizer to optimize the online predictor includes: Perform local or global refinement to quickly evaluate SKR using the FER predictor. V A And take the optimal modulation variance that maximizes the secure code rate SKR. V A * Then adaptively scale the optimal modulation variance. V A * The search range and step size are determined, with the search step size adaptively varying with the confidence level; finally, the optimal secure code rate (SKR) is calculated within the search range. * The maximum value is taken as the refined result, and the optimal safe bitrate (SKR) of the refined result is... * If the optimal secure bitrate (SKR) is significantly better than the prediction result, then the optimal secure bitrate will be used. * The corresponding sample is used as the new label (H) * , V A * SKR * )Cache, participate in the incremental updates of the online predictor.

[0013] In a preferred embodiment, based on the adopted globally optimal parameter combination When outputting the results: If there is only one error correction matrix, then directly output the corresponding optimal modulation variance and optimal secure code rate; If multiple error correction matrices exist, select the maximum secure bit rate and the corresponding optimal modulation variance output.

[0014] Secondly, the present invention provides an optimal parameter prediction device for a deep learning-based CV-QKD system, comprising: The data acquisition module is used to acquire the FER-SNR performance curves of the CV-QKD system under different error correction matrices; The fitting module is used to establish the modulation variance based on the FER-SNR performance curve. V A Mapping relationship with SKR (Secure Bitrate) ; Training data generation module, used to utilize mapping relationships The constructed offline optimizer outputs labels (H * , V A * SKR * ), including the optimal error correction matrix H * Optimal modulation variance V A* and optimal safe bitrate SKR * ; The training module is used to utilize the labels (H) output by the offline optimizer. * , V A * SKR * Training an online predictor based on a neural network model The inference module is used to call the online predictor to infer the globally optimal combination of parameters. And confidence level, based on the confidence level, decide whether to adopt the globally optimal parameter combination. ; The optimization module is used to avoid adopting the globally optimal parameter combination. The offline optimizer is invoked to optimize the online predictor. Output module for using the adopted globally optimal parameter combination Output results.

[0015] Thirdly, the present invention provides an electronic device, comprising: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor executes the instructions stored in the memory to perform the method described above.

[0016] Fourthly, the present invention provides a computer-readable storage medium for storing instructions that, when executed, cause the above-described method to be implemented.

[0017] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This invention utilizes neural networks to rapidly predict optimal parameters, achieving millisecond-level optimization and significantly improving experimental and operational efficiency. Simultaneously, it can dynamically adjust parameters based on channel conditions, enhancing the system's real-time performance, robustness, and stability, and avoiding performance and security risks caused by parameter fluctuations. Compared to traditional methods relying on manual debugging and complex hardware, this invention requires only a general-purpose computing unit to achieve automated, low-cost parameter optimization, reducing labor costs and hardware overhead.

[0018] In summary, this invention realizes the transformation of CV-QKD system parameter tuning from experience-driven to data-driven, maximizing the secure code rate while ensuring security. It has significant practical value and promotion prospects. This method is not only applicable to metropolitan area network CV-QKD systems, but can also be extended to access network CV-QKD systems. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an optimal parameter prediction method for a CV-QKD system based on deep learning, provided as an embodiment of the present invention.

[0020] Figure 2 Modulation variance as an example of the present invention V A The output results of a single random sampling of SKR with a secure bit rate are shown in the figure.

[0021] Figure 3 This is a schematic diagram of the structure of an optimal parameter prediction device for a deep learning-based CV-QKD system, provided in an embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0025] like Figure 1 As shown, this embodiment of the invention provides a method for predicting optimal parameters of a CV-QKD system based on deep learning, including: S100, obtain the FER-SNR performance curves of the CV-QKD system under different error correction matrices; S200, establishing modulation variance based on FER-SNR performance curve V A Mapping relationship with SKR (Secure Bitrate) ; S300, based on mapping relationship The constructed offline optimizer outputs labels (H * , V A * SKR * ), including the optimal error correction matrix H * Optimal modulation varianceV A * and optimal safe bitrate SKR * ; S400, using the labels (H) output by the offline optimizer * , V A * SKR * Training an online predictor based on a neural network model; S500 invokes the online predictor to infer the globally optimal parameter combination. And confidence level, based on the confidence level, decide whether to adopt the globally optimal parameter combination. ; S600 does not adopt the globally optimal parameter combination. The offline optimizer is invoked to optimize the online predictor. S700, based on the adopted globally optimal parameter combination Output results.

[0026] First, you need to set the range of channel parameter values, including transmission distance. d Channel transmittance Excessive noise Detection efficiency Electrical noise ele The range of values ​​for the channel parameters (derived from real-time estimation or experimental calibration). The specific implementation of the above method is as follows: S100, obtain the FER-SNR performance curves of the CV-QKD system under different error correction matrices.

[0027] For the error correction code scheme (such as LDPC parity-check matrix) used in the CV-QKD system, the frame error rate (FER) of different error correction matrices is collected within the signal-to-noise ratio (SNR) range corresponding to their characteristic curves. Collecting error correction matrices with different code rates generates a continuously covering FER-SNR performance curve. For example, the frame error rate (FER) corresponding to the SNR under several error correction matrices can be recorded through simulation or experiments (this FER can be the average frame error rate). The collected FER-SNR performance curves are preprocessed and used as input / output pairs for subsequent model training. The input is the SNR, and the output is the frame error rate (FER). The wider the SNR range covered by the data collection, the better the model's generalization ability.

[0028] S200, establishing modulation variance based on FER-SNR performance curve V A Mapping relationship with SKR (Secure Bitrate) : By training a neural network (a one-dimensional convolutional network is used in this embodiment of the invention) to approximate the FER-SNR performance curve, a fast and generalizable FER predictor for any signal-to-noise ratio (SNR) is obtained. This process can be performed online or offline. Combining FER predictor and signal-to-noise ratio (SNR) with modulation variance V A mapping relationship Based on the secure code rate formula of the CV-QKD system, the modulation variance is established. V A Mapping relationship with SKR (Secure Bitrate) , is represented as:

[0029] in, This indicates the valid data during the key extraction process. n Total data N proportion, Represents the mutual information and modulation variance between Alice and Bob. V A The functional relationship, Indicates coordination efficiency and modulation variance V A The functional relationship, usually 0 < ≤1, This represents the amount of information Eve (the eavesdropper) can obtain and the modulation variance. V A The functional relationship.

[0030] S300, utilizing a mapping-based relationship The constructed offline optimizer outputs labels (H * , V A * SKR * ), including the optimal error correction matrix H * Optimal modulation variance V A * and optimal safe bitrate SKR * : Based on the set range of channel parameters, randomly sample system parameters ( d , , , , ele ...); For each randomly sampled group of system parameters, the modulation variance is calculated using the secure bit rate formula. VA Establish modulation variance within the range of values. V A The mapping relationship between the error correction matrix H and each error correction matrix H; Based on this mapping relationship, each error correction matrix is ​​evaluated one by one. The optimal modulation variance corresponding to each error correction matrix is ​​obtained in this way. V A * and optimal safe bitrate SKR * ,get( V A * SKR * ), Then, the global optimum is selected among the various error correction matrices H to obtain the label (H) under each set of system parameters. * , V A * SKR * ).

[0031] S400, using the labels (H) output by the offline optimizer * , V A * SKR * Training an online predictor based on a neural network model: The online predictor, built on a neural network model, consists of two fully connected layers (128-128, ReLU) sharing a backbone and three outputs; these three outputs are the optimal error correction matrix H. * Classification, optimal modulation variance V A * Regression and Optimal Safe Bitrate SKR * The regression is as follows: Head 1: Optimal Error Correction Matrix H * Classification (Softmax, CE loss); Head 2: Optimal Modulation Variance V A * Regression (Sigmoid output, MSE loss); First 3: Optimal Secure Bitrate SKR * Regression (Sigmoid output, MSE loss).

[0032] Input system parameters [transmission distance] d Channel attenuation coefficient α Excessive noise Detection efficiency Electrical noise ele , The normalized vectors of the detection method [hom / het, ...] and the labels (H) output by the offline optimizer. * , V A * SKR * The online predictor was trained using the Adam optimizer and early stopping; the loss function during training was: ; Where L is the total loss and CE is the optimal error correction matrix H. * Classification loss, For optimal modulation variance V A * Regression loss, SKR for optimal safe bitrate * Regression loss, For optimal modulation variance V A * Weights of regression loss SKR for optimal safe bitrate * Weights of regression loss This is the weight decay term. This is the weight decay coefficient, used to prevent overfitting. For model parameters The square of the L2 norm is used for L2 regularization.

[0033] Steps S300-S400 construct an offline optimizer-online predictor mechanism to achieve fast approximation and online prediction. If multiple error correction matrices / bitrates exist, FER predictors are trained separately for each. The offline optimizer generates corresponding labels, while the online predictor learns simultaneously through the classification head which error correction matrix is ​​optimal.

[0034] S500 invokes the online predictor to infer the globally optimal parameter combination. And confidence level, based on the confidence level, decide whether to adopt the globally optimal parameter combination. .

[0035] During deployment, the online predictor is invoked by default to obtain the globally optimal parameter combination in a single forward pass. At the same time, output the classification probability. The confidence level is determined by the regression residual estimate.

[0036] Threshold method based on confidence level: If the confidence level is higher than the confidence threshold, then the globally optimal parameter combination is adopted. ; If the confidence level is lower than the confidence threshold, the globally optimal parameter combination will not be adopted. .

[0037] The confidence threshold can be set according to needs and actual circumstances.

[0038] S600 does not adopt the globally optimal parameter combination. The offline optimizer is invoked to optimize the online predictor. Specifically, when the confidence level falls below a confidence threshold, or the features exceed the training distribution (drift detection), or the upper-layer policy requires verification, the offline optimizer is triggered to perform a fast grid search within a small neighborhood. Perform local refinement, and extend to global refinement if necessary, using the FER predictor. Quickly assess SKR ( V A And take the optimal modulation variance that maximizes the secure code rate SKR. V A * Then adaptively scale the optimal modulation variance. V A * The search range and step size are determined, with the step size adaptively varying with confidence (high confidence → small step size / narrow interval; low confidence → wider interval); finally, the optimal secure bitrate (SKR) is calculated within the search range. * The maximum value is taken as the refined result, and the optimal safe bitrate (SKR) of the refined result is... * If the optimal secure bitrate (SKR) is significantly better than the prediction result, then the optimal secure bitrate will be used. * The corresponding sample is used as the new label (H) * , V A * SKR * The cache participates in the incremental updates of the online predictor, thereby enabling adaptive optimization and continuous learning of the online predictor and improving its long-term robustness.

[0039] S700, based on the adopted globally optimal parameter combination Output result: If there is only one error correction matrix, then directly output the corresponding optimal modulation variance and optimal secure code rate; If multiple error correction matrices exist, select the maximum secure bit rate and the corresponding optimal modulation variance output.

[0040] The output results can be used for user reference or directly sent to the modulator, encoder, and other execution units of the CV-QKD system to adjust to the optimal configuration. Furthermore, the optimization results can be stored for long-term performance monitoring.

[0041] Simultaneously, it outputs model side-factor information, including: the probability distribution of matrix / rate selection and threshold determination results; and the modulation variance in the globally optimal parameter combination. Modulated variance after refinement V A * Deviation, secure code rate in the globally optimal parameter combination The difference between the measured / verified safe bitrate (SKR) and the actual safe bitrate; timestamps and feature distribution drift indicators. This side-module information is used for closed-loop monitoring and online retraining triggering, ensuring long-term stability and traceability. To verify the effectiveness of the method, these optimal prediction parameters can also be used for actual system testing to observe whether a near-safe bitrate is achieved. To further improve the training data and model.

[0042] Through the above steps, this invention accelerates the parameter optimization process of the CV-QKD system using deep learning neural networks. An online predictor is trained from offline data and then used to quickly search for optimal parameters online, avoiding the inefficient process of exhaustive parameter tuning in experiments. In implementation, this invention can run offline before experiments to provide optimization guidance parameters; it can also be embedded in the CV-QKD system for real-time operation, dynamically updating the optimal settings by continuously acquiring new channel parameters to achieve closed-loop feedback control. For example, data can be re-acquired, the model updated, and a new optimal modulation variance calculated at regular intervals. V A * This ensures that the CV-QKD system always operates close to its optimal point, guaranteeing a high level of secure bitrate.

[0043] The following example details the optimal parameter prediction method for the deep learning-based CV-QKD system.

[0044] Set the range of channel parameters (modulation variance) A Transmission distance: 0.01~5 d The range is 30-65km, with a step size of 1km, and a channel attenuation coefficient. α =0.2dB / km, channel transmittance T =10 -αd / 10 Excessive noise =0.01, detection efficiency =0.65, electrical noise ele =0.02, detection method is zero difference detection, etc.), the FER-SNR performance curves obtained by simulation of three error correction matrices R=0.11, R=0.1, and R=0.09.

[0045] Three error correction matrices at three different code rates are trained using a neural network to generate three FER predictors. The result is used as the fitting result of the error correction matrix. The modulation variance is established based on the secure code rate formula of the CV-QKD system. A Mapping relationship with SKR (Secure Bitrate) ,like Figure 2 As shown.

[0046] Offline optimizer - online predictor mechanism, offline optimizer: in parameter space ( d Random sampling is performed within the range of 30~65km (in actual systems, the ranges of other parameters can be set and added to the random sampling process). The optimal modulation variance is calculated for the three error correction matrices R=0.11, R=0.1, and R=0.09. V A *and the corresponding SKR*, and take the global optimum among the error correction matrices to form the label. For example, taking the 50km condition as an example: (R=0.11, V A *=2.968500, SNR*=0.177684, SKR*=0.005712, d =50); (R=0.10, V A *=2.688500, SNR*=0.160924, SKR*=0.005056, d =50); (R=0.09, V A *=2.384500, SNR*=0.142727, SKR*=0.005491, d =50). Online predictor: Employs a two-layer fully connected (128-128, ReLU) multilayer perceptron (MLP) with a shared trunk, and features multi-tasking three-head output: Classification head: Predicts the optimal error correction matrix H* (Softmax, cross-entropy loss) Regression Head 1: Predicting the Optimal Modulation Variance V A * (Sigmoid output, MSE loss) Regression Head 2: Predict the optimal safe bitrate SKR*(Sigmoid output, MSE loss) Input features are [ d , α , ξ , η , V ele The normalized vector of [, detection method, ...], and the total loss function is a weighted sum of three parts.

[0047] During the deployment phase, the online predictor is prioritized for rapid inference to output the globally optimal parameter combination. When the confidence level is insufficient or the features exceed the training distribution, a fast grid search of the offline optimizer is triggered for local refinement. If necessary, this is expanded to global refinement, and the new results are used for incremental learning to improve long-term robustness.

[0048] The final output is the globally optimal parameter combination. It also includes probability distribution, rollback records and monitoring indicators to achieve closed-loop control and traceability.

[0049] Based on the same technical concept, this invention also provides a deep learning-based CV-QKD system optimal parameter prediction device corresponding to the above method. For example... Figure 3 As shown, the device includes the following modules: The data acquisition module is used to acquire the FER-SNR performance curves of the CV-QKD system under different error correction matrices; The fitting module is used to establish the modulation variance based on the FER-SNR performance curve. V A Mapping relationship with SKR (Secure Bitrate) ; Training data generation module, used to utilize mapping relationships The constructed offline optimizer outputs labels (H * , V A * SKR * ), including the optimal error correction matrix H * Optimal modulation variance V A * and optimal safe bitrate SKR * ; The training module is used to utilize the labels (H) output by the offline optimizer. * , V A * SKR * Training an online predictor based on a neural network model The inference module is used to call the online predictor to infer the globally optimal combination of parameters. And confidence level, based on the confidence level, decide whether to adopt the globally optimal parameter combination. ; The optimization module is used to avoid adopting the globally optimal parameter combination. The offline optimizer is invoked to optimize the online predictor. Output module for using the adopted globally optimal parameter combination Output results.

[0050] The functional modules in the above-mentioned device work together, and the specific working principle can be referred to the description in the foregoing method embodiments. This realizes the automated execution of the method of the present invention and can be integrated into the CV-QKD system as an intelligent auxiliary unit.

[0051] Based on the same technical concept, embodiments of the present invention also provide an electronic device that can implement the optimal parameter prediction method for the deep learning-based CV-QKD system provided in the above embodiments of the present invention. In one embodiment, the electronic device can be a server, a terminal device, or other electronic devices. Figure 4 As shown, the electronic device may include: At least one processor and a memory connected to the at least one processor. In this embodiment of the invention, the specific connection medium between the processor and the memory is not limited. Figure 4 The example used is the connection between the processor and memory via a bus. The bus... Figure 4 The connections between other components are indicated by thick lines and are for illustrative purposes only, not as limiting information. Buses can be categorized into address buses, data buses, control buses, etc., but for ease of representation, [the specific bus type is not shown here]. Figure 4 The processor is represented by a single thick line, but this does not imply that there is only one bus or one type of bus. Alternatively, a processor can also be called a controller; there are no restrictions on the name.

[0052] In this embodiment of the invention, the memory stores instructions that can be executed by at least one processor. By executing the instructions stored in the memory, at least one processor can execute the optimal parameter prediction method for a deep learning-based CV-QKD system discussed above.

[0053] The processor is the control center of the device. It can connect to various parts of the control device through various interfaces and lines. By running or executing instructions stored in memory and calling data stored in memory, it can monitor the device's various functions and process data, thereby enabling overall monitoring of the device.

[0054] In an alternative design, the processor may include one or more processing units. The processor may integrate an application processor and a modem processor, wherein the application processor primarily handles the operating system, user interface, and applications, while the modem processor primarily handles wireless communication. It is understood that the modem processor may also not be integrated into the processor. In some embodiments, the processor and memory may be implemented on the same chip; in some embodiments, they may also be implemented separately on separate chips.

[0055] The processor can be a general-purpose processor, such as a CPU, digital signal processor, application-specific integrated circuit, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the optimal parameter prediction method for a deep learning-based CV-QKD system disclosed in the embodiments of this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0056] Memory, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory can include at least one type of storage medium, such as flash memory, hard disk, multimedia cards, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), and electrically erasable programmable read-only memory (EPROM). Only memory (EEPROM), magnetic storage, magnetic disks, optical disks, etc. A memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures, and accessible by a computer, but is not limited thereto. The memory in embodiments of this invention can also be a circuit or any other device capable of performing storage functions for storing program instructions and / or data.

[0057] By designing and programming the processor, the code corresponding to the optimal parameter prediction method for a deep learning-based CV-QKD system described in the foregoing embodiments can be embedded into the chip, enabling the chip to execute the steps of the method described in the foregoing embodiments during runtime. How to design and program the processor is a technique well-known to those skilled in the art and will not be elaborated upon here.

[0058] Based on the same inventive concept, embodiments of the present invention also provide a storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the aforementioned deep learning-based CV-QKD system optimal parameter prediction method.

[0059] In some alternative embodiments, the present invention also provides a method for predicting optimal parameters of a deep learning-based CV-QKD system, which can also be implemented as a program product including program code. When the program product is run on a device, the program code is used to cause the control device to perform the steps in the method for predicting optimal parameters of a deep learning-based CV-QKD system according to various exemplary embodiments of the present invention as described above.

[0060] It should be noted that although several units or sub-units of the apparatus have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the invention, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units. Furthermore, although the operation of the method of the invention is described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0061] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can be implemented in one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs) containing computer-usable program code. The form of a computer program product implemented on ROM, optical memory, etc.

[0062] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a server, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0063] Program code for performing the operations of this invention can be written using any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0064] In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting optimal parameters of a CV-QKD system based on deep learning, characterized in that, include: Obtain the FER-SNR performance curves of the CV-QKD system under different error correction matrices; Establishing modulation variance based on FER-SNR performance curve V A Mapping relationship with SKR (Secure Bitrate) ; Using mapping relationships The constructed offline optimizer outputs labels (H * , V A * SKR * ), including the optimal error correction matrix H * Optimal modulation variance V A * and optimal safe bitrate SKR * ; Using the labels (H) output by the offline optimizer * , V A * SKR * Training an online predictor based on a neural network model; Call the online predictor to infer the globally optimal parameter combination And confidence level, based on the confidence level, decide whether to adopt the globally optimal parameter combination. ; Do not adopt the globally optimal parameter combination The offline optimizer is invoked to optimize the online predictor. The process of calling the offline optimizer to optimize the online predictor includes: performing local or global refinement to quickly evaluate the SKR using the FER predictor. V A And take the optimal modulation variance that maximizes the secure code rate SKR. V A * Then adaptively scale the optimal modulation variance. V A * The search range and step size are determined, with the search step size adaptively varying with the confidence level; finally, the optimal secure code rate (SKR) is calculated within the search range. * The maximum value is taken as the refined result, and the optimal safe bitrate (SKR) of the refined result is... * If the optimal secure bitrate (SKR) is significantly better than the prediction result, then the optimal secure bitrate will be used. * The corresponding sample is used as the new label (H) * , V A * SKR * )Cache, participate in the incremental updates of the online predictor; Based on the adopted globally optimal parameter combination Output results.

2. The optimal parameter prediction method for a deep learning-based CV-QKD system according to claim 1, characterized in that, The mapping relationship between modulation variance and secure code rate (SKR) is established based on the FER-SNR performance curve. ,include: By training a neural network to approximate the FER-SNR performance curve, an FER predictor for any signal-to-noise ratio (SNR) is obtained. Combining FER predictor and signal-to-noise ratio (SNR) with modulation variance V A mapping relationship Based on the secure code rate formula of the CV-QKD system, the modulation variance is established. V A Mapping relationship with SKR (Secure Bitrate) .

3. The optimal parameter prediction method for a deep learning-based CV-QKD system according to claim 1, characterized in that, Modulation variance V A Mapping relationship with SKR (Secure Bitrate) Represented as: in, This indicates the valid data during the key extraction process. n Total data N proportion, Represents the mutual information and modulation variance between Alice and Bob. V A The functional relationship, Indicates coordination efficiency and modulation variance V A The functional relationship, This represents the amount of information Eve can acquire and the modulation variance. V A The functional relationship.

4. The optimal parameter prediction method for a deep learning-based CV-QKD system according to claim 1, characterized in that, The use of mapping relationship The constructed offline optimizer outputs labels (H * , V A * SKR * ),include: Random sampling system parameters; For each randomly sampled group of system parameters, the modulation variance is calculated using the secure bit rate formula. V A Establish modulation variance within the range of values. V A The mapping relationship between the error correction matrix H and each error correction matrix H; Based on this mapping relationship, each error correction matrix is ​​evaluated one by one. The optimal modulation variance corresponding to each error correction matrix is ​​obtained in this way. V A * and optimal safe bitrate SKR * ; Then, the global optimum is selected among the various error correction matrices H to obtain the label (H) under each set of system parameters. * , V A * SKR * ).

5. The optimal parameter prediction method for a deep learning-based CV-QKD system according to claim 1, characterized in that, The label (H) output by the offline optimizer is used. * , V A * SKR * Training an online predictor based on a neural network model includes: The online predictor, built on a neural network model, consists of two fully connected layers sharing a backbone and three outputs; these three outputs are the optimal error correction matrix H. * Classification, optimal modulation variance V A * Regression and Optimal Safe Bitrate SKR * return; The normalized vector of input system parameters and the labels output by the offline optimizer (H * , V A * SKR * The online predictor is trained using the Adam optimizer and early stopping; the loss function during training is: ； Where L is the total loss and CE is the optimal error correction matrix H. * Classification loss, For optimal modulation variance V A * Regression loss, SKR for optimal safe bitrate * Regression loss, For optimal modulation variance V A * Weights of regression loss SKR for optimal safe bitrate * Weights of regression loss This is the weight decay term. This is the weight decay coefficient. For model parameters The square of the L2 norm.

6. The method for predicting optimal parameters of a CV-QKD system based on deep learning according to claim 1, characterized in that, Based on the adopted globally optimal parameter combination When outputting the results: If there is only one error correction matrix, then directly output the corresponding optimal modulation variance and optimal secure code rate; If multiple error correction matrices exist, select the maximum secure bit rate and the corresponding optimal modulation variance output.

7. A device for predicting optimal parameters of a deep learning-based CV-QKD system, characterized in that, include: The data acquisition module is used to acquire the FER-SNR performance curves of the CV-QKD system under different error correction matrices; The fitting module is used to establish the modulation variance based on the FER-SNR performance curve. V A Mapping relationship with SKR (Secure Bitrate) ; Training data generation module, used to utilize mapping relationships The constructed offline optimizer outputs labels (H * , V A * SKR * ), including the optimal error correction matrix H * Optimal modulation variance V A * and optimal safe bitrate SKR * ; The training module is used to utilize the labels (H) output by the offline optimizer. * , V A * SKR * Training an online predictor based on a neural network model; The inference module is used to call the online predictor to infer the globally optimal combination of parameters. And confidence level, based on the confidence level, decide whether to adopt the globally optimal parameter combination. ; The optimization module is used to avoid adopting the globally optimal parameter combination. The offline optimizer is invoked to optimize the online predictor. The process of calling the offline optimizer to optimize the online predictor includes: performing local or global refinement to quickly evaluate the SKR using the FER predictor. V A And take the optimal modulation variance that maximizes the secure code rate SKR. V A * Then adaptively scale the optimal modulation variance. V A * The search range and step size are determined, with the search step size adaptively varying with the confidence level; finally, the optimal secure code rate (SKR) is calculated within the search range. * The maximum value is taken as the refined result, and the optimal safe bitrate (SKR) of the refined result is... * If the optimal secure bitrate (SKR) is significantly better than the prediction result, then the optimal secure bitrate will be used. * The corresponding sample is used as the new label (H) * , V A * SKR * )Cache, participate in the incremental updates of the online predictor; Output module for using the adopted globally optimal parameter combination Output results.

8. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which executes the instructions stored in the memory to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store instructions that, when executed, cause the method as described in any one of claims 1-6 to be implemented.

Images