A CVQKD system phase compensation method and device based on machine learning
By constructing a support vector machine regression model in the CVQKD system to predict the phase drift angle, the problem of difficult phase drift in the CVQKD system is solved, real-time compensation for slow drift and key rate improvement is achieved, and system complexity and redundancy are reduced. sex.
Patent Information
- Application Number
- CN202211245150.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-12
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-10-12
Smart Images

Figure CN115643011B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of phase compensation for quantum key distribution, and in particular to a method and device for phase compensation of a CVQKD system based on machine learning. Background Art
[0002] Quantum secure communication technology is currently the most practical unconditionally secure communication technology. Quantum secure communication solutions are divided into quantum secure direct communication technology and quantum key distribution technology. Quantum key distribution has the advantages of high communication rate and strong scalability. The implementation methods of quantum key distribution are mainly divided into continuous variable quantum key distribution (CVQKD) and discrete variable quantum key distribution (DVQKD), both of which transmit information through quantum state encoding. The advantage of CVQKD is that it has a higher communication rate, can be integrated with traditional laser communication equipment, and has stronger resistance to background noise interference. CVQKD means that in the entire key distribution process, the Hilbert space where the physical observable used to encode the key information is infinite-dimensional and continuous, and the information carrier can be various types of light field quantum states, such as coherent states, compressed states, entangled states, etc.
[0003] The CVQKD system often uses a method based on Gaussian modulated coherent states (GMCS). Two beams of light are generated at the transmitter: signal light and local oscillator light. The signal is recovered at the receiver based on the local oscillator light of the transmitter. In the actual process, due to the influence of other factors such as quantum channel attenuation, fiber jitter, temperature, etc., the phase difference between the local oscillator light and the signal light will change over time, thereby affecting the key rate of the system. The sources of the phase change of the local oscillator light of the CVQKD system include fast drift, slow drift and phase noise. The fast drift comes from the phase jitter and center frequency change of the laser at the transmitter, which has random uncertainty and local correlation; the slow drift comes from the slow change caused by the delay of the transmission optical path and the synchronization error of the laser, and the relationship between data blocks is approximately linear. Phase noise is also called Gaussian white noise, which mainly comes from system noise, channel noise, temperature and the attack noise of the eavesdropper Eve. After the quantum signal light is interfered, the phase noise amplification will lead to an increase in the system excess noise, but the quantum signal intensity is too small and the phase noise and other system noise are not easy to distinguish, so it is difficult to specifically compensate for the phase noise of the quantum signal light.
[0004] Most of the commonly used phase compensation schemes currently require the introduction of a real-time monitoring module, which increases the complexity of the system. However, using strong light to compensate the phase of quantum signals under low signal-to-noise ratio conditions will leave Eve with an attack vulnerability. Summary of the invention
[0005] In view of this, the present invention provides a phase compensation method and device for a CVQKD system, which collects phase drift values as training data to obtain a support vector machine regression model and predict the future change trend of the phase drift angle.
[0006] The technical method adopted by the present invention is:
[0007] A CVQKD system phase compensation method based on machine learning comprises the following steps:
[0008] Before the effective sequence data of the CVQKD system starts to be transmitted, a standard sequence known to the sender and the receiver is inserted, and the repetition frequency of the standard sequence is consistent with the repetition frequency of the CVQKD system;
[0009] At the transmitting end, the standard sequence data is first transmitted. The beam splitter splits the optical pulse into quantum signal light and local oscillator light. The quantum signal light is Gaussian modulated by the amplitude and phase modulator, polarization multiplexing is completed by the polarization beam splitter and Faraday mirror, and the local oscillator light is time-division multiplexed by the delay line. Finally, the signal light reaches another polarization beam splitter through the attenuator.
[0010] At the receiving end, the polarization beam splitter splits the quantum signal light and the local oscillator light. The local oscillator light completes the measurement basis selection and time delay, and arrives at the beam splitter at the same time as the quantum signal light to complete the beam combination. The balanced homodyne detector measures the combined signal and converts it into an electrical signal. The signal acquisition device obtains the phase angle from the electrical signal and obtains the phase drift angle by comparing it with the data at the sending end.
[0011] At the receiving end, a regression prediction model of time t and phase drift angle Δφ(t) is constructed through the support vector machine regression algorithm to predict the phase drift angle at future time; when transmitting a valid sequence, the phase drift angle at future time is loaded onto the phase modulator at the transmitting end as the phase compensation angle for phase compensation.
[0012] Furthermore, the regression prediction model of time t and phase drift angle Δφ(t) is constructed by the support vector machine regression algorithm, specifically in the following manner:
[0013] The data points of a part of the phase drift angle are used as the training set, and the remaining data points are used as the test set of the model;
[0014] The regression prediction model of time t and phase drift angle Δφ(t) is constructed by support vector machine regression algorithm; the regression prediction model is:
[0015]
[0016] In the formula, α i is the Lagrange multiplier, is its optimal solution, b is the displacement term, K <x i ,x> is the Gaussian kernel function, and the physical model is i=1,2,…N,j=1,2,…N,x i and x j is the data set in the original space, N is the total number of data sets, σ>0 is the bandwidth of the Gaussian kernel, and ||Δ|| is the norm.
[0017] A CVQKD system phase compensation device based on machine learning includes a transmitting end and a receiving end; the transmitting end includes a laser, an amplitude modulator, a phase modulator, a beam splitter, a polarization beam splitter, a Faraday mirror and an attenuator, and the receiving end includes a polarization beam splitter, a phase modulator, a beam splitter, a balanced homodyne detector and a data acquisition device;
[0018] At the transmitting end, before transmitting the effective sequence, the standard sequence data is transmitted first. The beam splitter splits the optical pulse into quantum signal light and local oscillator light. The quantum signal light is Gaussian modulated by the amplitude and phase modulator, polarization multiplexing is completed by the polarization beam splitter and Faraday mirror, and the local oscillator light is time-division multiplexed by the delay line. Finally, the signal light reaches another polarization beam splitter through the attenuator.
[0019] At the receiving end, the polarization beam splitter splits the quantum signal light and the local oscillator light. The local oscillator light completes the measurement basis selection and time delay, and arrives at the beam splitter at the same time as the quantum signal light to complete the beam combination. The balanced homodyne detector measures the combined signal and converts it into an electrical signal. The signal acquisition device obtains the phase angle from the electrical signal and obtains the phase drift angle by comparing it with the data at the transmitting end. In addition, a regression prediction model of time t and phase drift angle Δφ(t) is constructed through the support vector machine regression algorithm to predict the phase drift angle at future time. When transmitting a valid sequence, the phase drift angle at future time is loaded as the phase compensation angle onto the phase modulator at the transmitting end for phase compensation.
[0020] Furthermore, before the effective sequence data of the CVQKD system starts to be transmitted, a standard sequence known to the sending end and the receiving end is inserted, and the repetition frequency of the standard sequence is consistent with the repetition frequency of the CVQKD system.
[0021] Furthermore, the regression prediction model of time t and phase drift angle Δφ(t) is constructed by the support vector machine regression algorithm, specifically in the following manner:
[0022] The data points of a part of the phase drift angle are used as the training set, and the remaining data points are used as the test set of the model;
[0023] The regression prediction model of time t and phase drift angle Δφ(t) is constructed by support vector machine regression algorithm; the regression prediction model is:
[0024]
[0025] In the formula, α i is the Lagrange multiplier, is the optimal solution, b is the displacement term, K <x i ,x> is a polynomial kernel function, and the physical model is K(x i ,x j )=(x i ·x j +1) p ,i=1,2,…N,j=1,2,…N,x i and x j is the data set of the original space, N is the total number of data sets, and P>1 is the degree of the polynomial.
[0026] The beneficial effects of the present invention are:
[0027] 1. The present invention adopts a support vector machine regression model, which can have a good prediction effect on the slow drift phase error angle of the local oscillator light. The predicted phase drift angle can be used to directly perform phase compensation on the data at the transmitting end, thereby improving the security key rate of the system.
[0028] 2. The support vector machine regression model proposed in the present invention can effectively compensate for the slow drift of the phase in real time, and does not need to add additional training frames to the sent data frames, thereby reducing the redundancy of the system and having better practicality.
[0029] 3. The present invention adopts machine learning to analyze the system phase drift value well and can be used as a phase compensation algorithm for optimization. In addition, the original data is directly reconstructed through the predicted value, and the phase drift value does not need to be calculated in real time, which reduces the system overhead to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 : is a block diagram of the system device proposed by the present invention, wherein the dotted line represents the optical path and the solid line represents the circuit.
[0031] Figure 2 It is the CVQKD system data frame structure diagram and standard sequence format provided by the present invention. DETAILED DESCRIPTION
[0032] The technical solution of the present invention will be described in more detail below in conjunction with the accompanying drawings and embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but the protection scope of the present invention is not limited thereto.
[0033] A CVQKD phase compensation method based on machine learning, which inserts a standard sequence known to the signal transmitter and the signal receiver before the effective data of the CVQKD system starts to be transmitted, converts the quantum state into an electrical signal after being measured by a balanced homodyne detector, obtains the original data through a signal acquisition device, and uses the phase drift value obtained by comparing it with the data of the transmitter as a training set for support vector machine regression, and obtains a prediction model of time and phase drift value based on the support vector machine regression algorithm. When the system transmits an effective sequence, the phase drift value is predicted by this model and loaded as a phase compensation angle onto the phase modulator of the transmitter for phase compensation.
[0034] Similarly, the displacement and momentum components can be measured simultaneously through a balanced heterodyne detector. Because the receiving end Bob measures the fixed phase, there is no need to dynamically select the basis like a balanced homodyne detector, which simplifies the key distribution. However, the balanced homodyne detector is more effective in actual systems.
[0035] Specifically, at the transmitting end Alice, the optical pulse generated by the laser is split into quantum signal light and local oscillator light by a beam splitter. The quantum signal light is Gaussian modulated by an amplitude and phase modulator, polarization multiplexing is performed by a polarization beam splitter and a Faraday mirror, and time division multiplexing is performed by a delay line for the local oscillator light. Finally, the signal light reaches the polarization beam splitter through an attenuator.
[0036] After Bob receives the signal at the receiving end, the quantum signal light and the local oscillator light are split into beams by a polarization beam splitter. The local oscillator light completes the measurement basis selection and time delay, and arrives at the beam splitter at the same time as the quantum signal light. After the signal is measured by a balanced homodyne detector, it is converted into an electrical signal, and the original data is obtained through the signal acquisition equipment.
[0037] When the standard sequence is transmitted, the receiving end converts it into an electrical signal after being measured by a balanced homodyne detector. After being collected by the data acquisition device, it is compared with the standard sequence data of the transmitting end to obtain the phase drift value. According to the above algorithm, a support vector machine regression prediction model of time t and phase drift value Δφ(t) is constructed. When the system transmits a valid sequence, the phase drift value is predicted by this model.
[0038] The phase drift value in the future is predicted according to the model and loaded onto the phase modulator at the transmitting end as the phase compensation angle for phase compensation. The phase-compensated data can be processed and analyzed to obtain a security key, which can improve the key rate of the system in real time.
[0039] In this method, the repetition frequency of the standard sequence is consistent with the repetition frequency of the CVQKD system, and the standard sequence is known at both the transmitting and receiving ends before transmission. The phase drift value is calculated based on the data received by the standard sequence at the receiving end. The more data the standard sequence has, the more accurate the support vector machine regression model is.
[0040] Here are more specific examples:
[0041] like Figure 1 As shown, a CVQKD phase compensation device based on machine learning includes a transmitting end Alice and a receiving end Bob, wherein:
[0042] At the transmitting end, the laser generates an optical pulse through amplitude modulator 1, which is split into quantum signal light and local oscillator light through beam splitter 1. The quantum signal light is Gaussian modulated through amplitude modulator 2 and phase modulator 1, polarization multiplexing is completed through polarization beam splitter 1 and Faraday mirror, and local oscillator light is time-division multiplexed through delay line. Finally, the signal light reaches polarization beam splitter 2 through attenuator.
[0043] After the signal is received at the receiving end, the quantum signal light and the local oscillator light are split by the polarization beam splitter 3. The local oscillator light completes the measurement basis selection and time delay, and arrives at the beam splitter 2 at the same time as the quantum signal light. After the signal is measured by the balanced homodyne detector, it is converted into an electrical signal, and the original data is obtained by the signal acquisition equipment.
[0044] like Figure 2 As shown in the figure, before the effective data of the traditional CVQKD system starts to be transmitted, a standard sequence known to the signal transmitter Alice and the signal receiver Bob is inserted. The phase drift value obtained by comparing with the data of the transmitter is used as the training set of the support vector machine regression. The repetition frequency of the standard sequence is consistent with the repetition frequency of the CVQKD system. The data points of 3 / 4 of the measured phase drift values are used as the training set, and 1 / 4 of the data points are used as the test set of the model. The prediction model function of the training time t and the phase drift value Δφ(t) is trained through the support vector machine regression, and the phase drift value Δφ(t+Δt) corresponding to the future time t+Δt can be predicted, so as to perform phase compensation on the phase modulator of the transmitter.
[0045] The physical model of support vector machine regression is as follows:
[0046]
[0047] Among them, assuming ω T x+b=0 is the hyperplane that has been determined, and the normal vector can determine the direction as ω=(ω 1 ;ω 2 ;...;ω d ), b is the displacement term to determine the distance from the origin. In the sample space, the length from any point x to this plane (w, b) can be expressed as r = (ω·x i +b) / ||ω||,ξ i and is the slack variable, C can be called the harmonic coefficient, ε is called the tolerance deviation, and the interval is determined by the normal vector ω, which is equal to 2 / ||ω||. -1 Solving for the maximum value is equivalent to solving ||ω|| 2 The minimum value of , so that the maximum interval value can be obtained. The Lagrange function is introduced when solving:
[0048]
[0049] α i , μ i and is the Lagrange multiplier. Therefore, it can be transformed into the dual problem of the maximum and minimum problem:
[0050]
[0051] Find the optimal solution The final prediction model function
[0052] Since there is a nonlinear relationship between time t and phase drift value Δφ(t), it is necessary to introduce a kernel function to transform the nonlinear regression problem into an approximate linear regression problem. Then K(x i ,x j ) is the kernel function, is the mapping function.
[0053] For the application in the field of phase compensation, this method selects the Gaussian kernel function because its corresponding feature space is infinite-dimensional, so it can make the limited data samples linearly separable. The physical model of the Gaussian kernel function is:
[0054]
[0055] Therefore, the prediction model function of phase drift is transformed into:
[0056]
[0057] When the standard sequence is transmitted, the receiving end converts it into an electrical signal after being measured by a balanced homodyne detector. After being collected by the data acquisition device, it is compared with the standard sequence of the transmitting end to obtain the phase drift value. According to the above algorithm, a support vector machine regression prediction model of time t and phase drift value Δφ(t) is constructed. When the system transmits a valid sequence, the phase drift value is predicted by this model.
[0058] The phase drift value in the future is predicted according to the model and loaded onto the phase modulator at the transmitting end as the phase compensation angle for phase compensation. The phase-compensated data can be processed and analyzed to obtain a security key, which can improve the key rate of the system in real time.
[0059] In a preferred embodiment, the displacement and momentum components can be measured simultaneously by a balanced heterodyne detector, and because the receiving end Bob fixes the phase measurement, key distribution is simplified.
[0060] In a preferred embodiment, the kernel function may be selected as a polynomial kernel function, and the physical model of the polynomial kernel function is:
[0061] K(x i ,x j )=(x i ·x j +1) p (5)
[0062] The corresponding support vector machine is a P-order polynomial classifier, so the prediction model function of phase drift is transformed into
[0063] When the system data is transmitted, due to the slow changes introduced by the delay of the transmission optical path and the clock synchronization error, it can be assumed that the slow drift changes linearly during the working time of the system. Therefore, the support vector machine regression model proposed in the present invention can effectively compensate for the slow drift of the phase in real time, and there is no need to add additional training frames to the sent data frames, which reduces the redundancy of the system and has good practicality. However, the phase compensation effect on the drift error of randomness and local correlation is not obvious.
[0064] The above description is only a specific implementation of the present invention in the embodiment, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the field within the technical scope disclosed by the present invention, and any modifications or equivalent substitutions that do not deviate from the spirit and scope of the present invention should be covered within the protection scope of the claims attached to the present invention.
Claims
1. A CVQKD system phase compensation method based on machine learning, characterized in that: The following steps are involved: Before the effective sequence data of the CVQKD system starts to be transmitted, a standard sequence known to the sender and the receiver is inserted, and the repetition frequency of the standard sequence is consistent with the repetition frequency of the CVQKD system; At the transmitting end, the standard sequence data is transmitted first. The beam splitter splits the optical pulse into quantum signal light and local oscillator light. The quantum signal light is Gaussian modulated by the amplitude and phase modulator, and polarization multiplexing with the local oscillator light is completed by the polarization beam splitter and Faraday mirror. The local oscillator light is time-division multiplexed with the quantum signal light by the delay line. Finally, the signal light reaches another polarization beam splitter through the attenuator. At the receiving end, the polarization beam splitter splits the quantum signal light and the local oscillator light. The local oscillator light completes the measurement basis selection and time delay, and arrives at the beam splitter at the same time as the quantum signal light to complete the beam combination. The balanced homodyne detector measures the combined signal and converts it into an electrical signal. The signal acquisition device obtains the phase angle from the electrical signal and obtains the phase drift angle by comparing it with the data at the sending end. At the receiving end, a regression prediction model of time t and phase drift angle φ(t) is constructed through the support vector machine regression algorithm to predict the phase drift angle at future time; when transmitting a valid sequence, the phase drift angle at future time is loaded as the phase compensation angle onto the phase modulator at the transmitting end for phase compensation.
2. A CVQKD system phase compensation method based on machine learning according to claim 1, characterized in that: The support vector machine regression algorithm is used to construct a regression prediction model of time t and phase drift angle φ(t), specifically in the following manner: The data points of a part of the phase drift angle are used as the training set, and the remaining data points are used as the test set of the model; The regression prediction model of time t and phase drift angle φ(t) is constructed by support vector machine regression algorithm; the regression prediction model is: In the formula, α i is the Lagrange multiplier, is its optimal solution, b is the displacement term, K <x i ,x> is the Gaussian kernel function, and the physical model is x i and x j is the data set in the original space, N is the total number of data sets, σ>0 is the bandwidth of the Gaussian kernel, and || || is the norm.
3. A CVQKD system phase compensation device based on machine learning, comprising a transmitting end and a receiving end; characterized in that: The transmitting end includes a laser, an amplitude modulator, a phase modulator, a beam splitter, a polarization beam splitter, a Faraday mirror and an attenuator, and the receiving end includes a polarization beam splitter, a phase modulator, a beam splitter, a balanced homodyne detector and a data acquisition device; Before the effective sequence data of the CVQKD system starts to be transmitted, a standard sequence known to the sender and the receiver is inserted, and the repetition frequency of the standard sequence is consistent with the repetition frequency of the CVQKD system; At the transmitting end, before transmitting the effective sequence, the standard sequence data is transmitted first. The beam splitter splits the optical pulse into quantum signal light and local oscillator light. The quantum signal light is Gaussian modulated by the amplitude and phase modulator, and polarization multiplexing with the local oscillator light is completed by the polarization beam splitter and Faraday mirror. The local oscillator light is time-division multiplexed with the quantum signal light by the delay line. Finally, the signal light reaches another polarization beam splitter through the attenuator. At the receiving end, the polarization beam splitter splits the quantum signal light and the local oscillator light. The local oscillator light completes the measurement basis selection and time delay, and arrives at the beam splitter at the same time as the quantum signal light to complete the beam combination. The balanced homodyne detector measures the combined signal and converts it into an electrical signal. The signal acquisition device obtains the phase angle from the electrical signal and obtains the phase drift angle by comparing it with the data at the transmitting end. In addition, a regression prediction model of time t and phase drift angle φ(t) is constructed through the support vector machine regression algorithm to predict the phase drift angle at future time. When transmitting a valid sequence, the phase drift angle at future time is loaded as the phase compensation angle onto the phase modulator at the transmitting end for phase compensation.
4. A CVQKD system phase compensation device based on machine learning according to claim 3, characterized in that: Before the effective sequence data of the CVQKD system starts to be transmitted, a standard sequence known to the sender and the receiver is inserted. The repetition frequency of the standard sequence is consistent with the repetition frequency of the CVQKD system.
5. The CVQKD system phase compensation device based on machine learning according to claim 3, characterized in that: The support vector machine regression algorithm is used to construct a regression prediction model of time t and phase drift angle φ(t), specifically in the following manner: The data points of a part of the phase drift angle are used as the training set, and the remaining data points are used as the test set of the model; The regression prediction model of time t and phase drift angle φ(t) is constructed by support vector machine regression algorithm; the regression prediction model is: In the formula, α i is the Lagrange multiplier, is the optimal solution, b is the displacement term, K <x i ,x> is a polynomial kernel function, and the physical model is K(x i ,x j )=(x i ·x j +1) p ,i=1,2,···N,j=1,2,···N,x i and x j is the data set of the original space, N is the total number of data sets, and P>1 is the degree of the polynomial.
Citation Information
Patent Citations
Method and system for improving phase compensation precision and communication efficiency of CVQKD system
CN111769881A
Deep learning-based holdover compensation, and related systems, methods and devices
WO2022173522A1
