Method for optimizing quantization operation in continuous variable quantum key distribution information negotiation based on deep learning

Abstract

The invention provides a method for optimizing quantization operation in continuous variable quantum key distribution information negotiation based on deep learning. The method comprises the following steps: a quantum sending end and a quantum receiving end respectively obtain a first original Gaussian sequence and a second original Gaussian sequence; the quantum receiving end randomly generates a random Gaussian sequence, obtains a new Gaussian sequence and sends the new Gaussian sequence to the quantum sending end; the quantum sending end obtains a target Gaussian sequence; the quantum receiving end obtains an optimized quantization function based on the trained deep neural network, quantifies and splits the random Gaussian sequence into m layers of original key strings through the optimized quantization function, directly sends the first k layers of original key strings to the quantum sending end, and sends checkers of the m-k layers of original key strings after calculation to the quantum sending end; the quantum sending end carries out information negotiation; according to the method, the quantization function is optimized through the deep neural network, the quantization process is adaptively adjusted, errors in the quantization process are minimized, information loss in the quantization process is reduced, and quantization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of quantum communication, to continuous variable quantum key distribution, and in particular to quantization, and specifically to a method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning. Background Art

[0002] With the rapid development of quantum communication technology, continuous variable quantum key distribution (CV-QKD), as an emerging quantum encryption method, has gradually become a research hotspot in the field of quantum communication. CV-QKD uses the basic principles of quantum mechanics to distribute keys through continuous variables of optical signals (such as amplitude, phase, etc.) and encrypts them through quantum channels to ensure the security of communication. However, although the CV-QKD system has shown great potential in theory and experiments, it still faces many challenges in practical applications, one of which is how to efficiently quantize signals and generate keys.

[0003] Slice negotiation scheme and multi-dimensional negotiation scheme are the two most commonly used negotiation schemes. Multi-dimensional negotiation can achieve high negotiation efficiency in long-distance CV-QKD systems with extremely low signal-to-noise ratios, but its post-processing algorithm is complex, and each pulse extracts less than 1 bit of key information, so the system's key rate is low. Slice negotiation uses a univariate vector function to convert a single data into multiple negotiation codes. Compared with multi-dimensional negotiation, it is more suitable for short-distance CV-QKD systems where the quantum channel is a high signal-to-noise ratio transmission environment. However, Slice negotiation is susceptible to noise. When the deviation between the original data is large, the accuracy of the quantization estimate is very low, resulting in serious quantization loss in hierarchical negotiation, especially in the case of high noise or weak signals. The quantization error may increase significantly, resulting in a decrease in the quality of the key. Therefore, how to improve the efficiency of the quantization process and reduce information loss has become an important research direction for CV-QKD systems. Summary of the invention

[0004] The purpose of the present invention is to provide a method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning, so as to solve the problems pointed out in the above background technology.

[0005] The present invention provides a method for optimizing quantization operation in continuous variable quantum key distribution information negotiation based on deep learning, and the method comprises: step 1, in a reverse slice negotiation mechanism, a quantum transmitting end obtains a first original Gaussian sequence x, and a quantum receiving end obtains a second original Gaussian sequence y; the first original Gaussian sequence x and the second original Gaussian sequence y both obey Gaussian distribution; step 2, the quantum receiving end randomly generates a random Gaussian sequence c, and adds the random Gaussian sequence c and the second original Gaussian sequence y to obtain a new Gaussian sequence w=c+y, and then sends the new Gaussian sequence w to the quantum transmitting end; step 3, the quantum transmitting end subtracts the first original Gaussian sequence x from the new Gaussian sequence w to obtain a target Gaussian sequence v=wx; step 4, the quantum receiving end obtains an optimized quantization function based on a trained deep neural network; step 5, the quantum receiving end quantizes and cuts the random Gaussian sequence c into m layers of original key strings L1~L through the optimized quantization function m , and the first k layers of original key strings L1~L k It is directly sent to the quantum transmitter, and the original key string L of the mk layer is calculated according to the preset error correction code and optimized quantization function. k+1 ~L m The checksum S k+1 ~S m , and the checksum S k+1 ~S m Send to the quantum transmitter; 1≤k<m, m≥2, and k, m are both integers; Step 6, after the quantum transmitter obtains the target Gaussian sequence v, it sets a maximum number of iterations within a layer t and a maximum number of iterations between layers T; t≥1, T≥1, t≥T, and t, T are both integers; Step 7, the quantum transmitter uses the target Gaussian sequence v and the original key strings L1~L1 of the first k layers m Calculate the log-likelihood ratio of the k+ith layer, and then, based on the calculated log-likelihood ratio of the k+ith layer and the calibration factor S k+1 ~S mPerform k+i layer decoding; i∈[1,mk], and i is an integer; Step 8, in the quantum transmitter, when the k+i layer decoding is completed, the decoding result of the k+i layer is passed to the next layer for estimation and decoding until the m-th layer decoding is completed; after the m-th layer decoding is completed, if all layers are successfully decoded, the decoding success key is saved and the intra-layer iterative decoding is terminated; otherwise, the intra-layer iterative decoding is performed until the maximum number of iterations t in the layer is reached, the intra-layer iterative decoding is terminated, and the decoding result of the m-th layer is transmitted. Pass it to the k+1th layer for inter-layer iterative decoding; the decoding results corresponding to the successfully decoded layers are all the same; the decoding success key is the decoding result of successful decoding of all layers; step nine, in the inter-layer iterative decoding, at the end of each layer decoding, compare the inter-layer iterative decoding result of each layer with the decoding result of the corresponding layer in the step eight, if they are the same, output the decoding success key and the inter-layer iterative decoding ends; otherwise, the inter-layer iterative decoding is performed until the maximum number of inter-layer iterations T is reached, and the inter-layer iterative decoding ends.

[0006] In the present invention, the quantization function is optimized through a deep neural network to achieve adaptive adjustment of the quantization process to minimize the error in the quantization process, thereby reducing information loss in the quantization process and improving quantization efficiency.

[0007] In one implementation of the present invention, in step 1, the quantum receiving end obtains the second original Gaussian sequence y, including: the quantum transmitting end sends the first original Gaussian sequence x to the quantum receiving end through a quantum channel, so that the quantum receiving end measures the first original Gaussian sequence x to obtain the second original Gaussian sequence y associated with the quantum transmitting end; wherein the first original Gaussian sequence is the modulation variance of the quantum transmitter; the noise z of the quantum channel obeys Gaussian distribution, is the noise variance of the quantum channel; y=x+z,

[0008] In one implementation of the present invention, in step 2, the random Gaussian sequence c obeys a Gaussian distribution. is the variance of the random Gaussian sequence c.

[0009] In an implementation of the present invention, before step 4, the method further includes: the quantum receiving end builds an original deep neural network; the quantum receiving end trains the original deep neural network based on a training set to obtain the trained deep neural network; the training set includes multiple training Gaussian sequences.

[0010] In one implementation of the present invention, in the process of training the original deep neural network based on the training set at the quantum receiving end, a loss function is used for training; the calculation formula of the loss function is:

[0011] L total =αL MSE +βL IM ;

[0012]

[0013] Among them, L total is the loss value calculated by the loss function; L MSE is the quantization error loss; L IM is the mutual information loss; α and β are hyperparameters; M is the total number of training Gaussian sequences contained in the training set; c j is the jth training Gaussian sequence in the training set; S(c j ) is based on c j When training the original deep neural network, the corresponding output of the original deep neural network; I(v); S(c j )) represents v, S(c j ) between them.

[0014] In one implementation of the present invention, in step four, the trained deep neural network includes: an input layer, an output layer and multiple hidden layers; wherein the input layer is used to input the random Gaussian sequence c; the multiple hidden layers are used to perform nonlinear transformation on the random Gaussian sequence c to extract high-dimensional features from the random Gaussian sequence c; the output layer is used to output the optimized quantization function.

[0015] In one implementation of the present invention, in step 5, the quantum receiving end maps the random Gaussian sequence c into a discrete binary sequence through the optimized quantization function, and multiplies the binary sequence with the preset error correction code to obtain the syndrome S k+1 ~S m ; Wherein, the binary sequence is the m-layer original key string L1~L m .

[0016] In an implementation of the present invention, in step 2, the quantum receiving end sends the new Gaussian sequence w to the quantum sending end through a classical channel.

[0017] As described above, the method for optimizing the quantization operation in the continuous variable quantum key distribution information negotiation based on deep learning according to the present invention has the following beneficial effects:

[0018] Compared with the prior art, the present invention uses a deep neural network to adaptively optimize the quantization function, minimizes the information loss in the quantization process, and makes the mutual information after quantization as close as possible to the mutual information before quantization, thereby effectively improving the quantization efficiency, increasing the rate of quantum key generation, and improving the efficiency of information negotiation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Shown is a flowchart of a method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to an embodiment of the present invention.

[0020] Figure 2 Shown is a schematic diagram of a method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0022] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0023] In recent years, with the rapid development of deep learning technology, deep neural networks (DNNs) have been widely used in various signal processing tasks. The powerful ability of deep learning enables it to automatically learn complex mapping relationships from large amounts of data, which has great potential in quantization problems. Traditional quantization methods usually rely on preset rules and fixed thresholds, while deep learning can adaptively adjust the quantization process through training data, thereby reducing information loss and improving quantization efficiency.

[0024] See also Figure 1 and Figure 2The following embodiments of the present invention provide a method for optimizing the quantization operation in the information negotiation of continuous variable quantum key distribution based on deep learning. Compared with the prior art, the present invention uses a deep neural network to adaptively optimize the quantization function, minimize the information loss in the quantization process, and make the mutual information after quantization as close as possible to the mutual information before quantization, thereby effectively improving the quantization efficiency, increasing the rate of quantum key generation, and improving the efficiency of information negotiation.

[0025] The technical solutions in the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0026] like Figure 1 and Figure 2 As shown, in one embodiment, the present invention provides a method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning, the method comprising:

[0027] Step S1: In the reverse Slice negotiation mechanism, the quantum transmitter obtains a first original Gaussian sequence x, and the quantum receiver obtains a second original Gaussian sequence y.

[0028] It should be noted that both the first original Gaussian sequence x and the second original Gaussian sequence y obey Gaussian distribution.

[0029] Reverse Slice negotiation refers to a method of negotiating the original key in a quantum key distribution system. During the quantum key distribution process, due to channel noise and eavesdropper interference, the original keys obtained by the legitimate communication parties may be asymmetric. In order to obtain a symmetric key, the original key needs to be negotiated during data post-processing.

[0030] like Figure 1 and Figure 2 As shown, in one embodiment, in the step S1, the quantum receiving end obtains the second original Gaussian sequence y, including: the quantum transmitting end sends the first original Gaussian sequence x to the quantum receiving end through a quantum channel, so that the quantum receiving end measures the first original Gaussian sequence x to obtain the second original Gaussian sequence y associated with the quantum transmitting end.

[0031] Among them, the first original Gaussian sequence is the modulation variance of the quantum transmitter; the noise z of the quantum channel obeys Gaussian distribution, is the noise variance of the quantum channel; y=x+z,

[0032] It should be noted that the quantum channel is used to transmit information-carrying or random quantum states (corresponding to the first original Gaussian sequence x).

[0033] It should be noted that after the above-mentioned quantum transmitting end sends the first original Gaussian sequence x to the quantum receiving end, the quantum transmitting end still contains the first original Gaussian sequence x.

[0034] Step S2: The quantum receiving end randomly generates a random Gaussian sequence c, and adds the random Gaussian sequence c to the second original Gaussian sequence y to obtain a new Gaussian sequence w=c+y, and then sends the new Gaussian sequence w to the quantum sending end.

[0035] In one embodiment, in step S2, the random Gaussian sequence c obeys a Gaussian distribution. is the variance of the random Gaussian sequence c.

[0036] like Figure 1 and Figure 2 As shown, in one embodiment, in step S2, the quantum receiving end sends the new Gaussian sequence w to the quantum sending end through a classical channel.

[0037] It should be noted that the classical channel is used to ensure that some necessary information can be exchanged between the quantum sender and the quantum receiver.

[0038] In one embodiment, a random Gaussian sequence c is randomly generated by a quantum random number generator (full name: Quantum Random Number Generator, English abbreviation: QRNG).

[0039] It should be noted that QRNG uses quantum physics processes to quickly and massively generate true random numbers guaranteed by quantum randomness principles such as quantum superposition collapse theory.

[0040] Step S3: The quantum transmitting end subtracts the first original Gaussian sequence x from the new Gaussian sequence w to obtain a target Gaussian sequence v=wx.

[0041] Step S4: The quantum receiving end obtains an optimized quantization function based on the trained deep neural network.

[0042] In one embodiment, before step S4, the method further includes: the quantum receiving end builds an original deep neural network; the quantum receiving end trains the original deep neural network based on a training set to obtain the trained deep neural network.

[0043] In this embodiment, the training set includes a plurality of training Gaussian sequences.

[0044] In one embodiment, the training Gaussian sequence is also a Gaussian sequence randomly generated by the quantum receiving end, which has the same principle as the above-mentioned random Gaussian sequence c.

[0045] In one embodiment, during the process of training the original deep neural network based on the training set at the quantum receiving end, the training is performed using a loss function.

[0046] It should be noted that, in order to optimize the quantization process, the present invention designs a composite loss function, taking into account the quantization error loss and the mutual information loss.

[0047] Among them, the quantization error loss (ie, MSE loss) L MSE : Used to measure the difference between the quantized signal S(cj) and the original signal cj, measured by mean square error (MSE).

[0048]

[0049] Mutual information loss L IM :The goal of the quantization process is to maximize the mutual information I(v; S(cj) between the quantized signal S(cj) and the shared value (i.e., the target Gaussian sequence v) j )).

[0050]

[0051] Loss function: It is the weighted sum of quantization error loss and mutual information loss.

[0052] L total =αL MSE +βL IM ;

[0053] L total is the loss value calculated by the loss function; L MSE is the quantization error loss; L IM is the mutual information loss; α and β are hyperparameters used to adjust the quantization error loss L MSE and mutual information loss L IM The trade-off between; M is the total number of training Gaussian sequences contained in the training set; c j is the jth training Gaussian sequence in the training set; S(c j ) is based on c j When training the original deep neural network, the corresponding output of the original deep neural network; I(v); S(c j )) represents v, S(c j ) between them.

[0054] It should be noted that mutual information is a useful information measure in information theory, which is used to measure the degree of mutual dependence between two random variables. Mutual information can be regarded as the amount of information contained in one random variable about another random variable, or the uncertainty of a random variable reduced by knowing another random variable; the calculation of mutual information adopts conventional technical means in the field, so it will not be described in detail here.

[0055] Specifically, by calculating the loss value L total , until the loss value L total No more descent, stop training to get the original deep neural network after training.

[0056] It should be noted that during the training process, a large amount of simulation data is used, including Gaussian sequences under different channel noise conditions and their corresponding quantized values. During the training process, the back propagation algorithm and the gradient descent optimization algorithm are used to minimize the loss function, thereby optimizing the parameters of the deep neural network so that the error of the quantization process is minimized and the mutual information is maximized.

[0057] In one embodiment, after the quantum receiving end trains the original deep neural network based on the training set, it will test the trained original neural network based on the test set to obtain the trained deep neural network.

[0058] It should be noted that the testing of the original deep neural network after training adopts conventional technical means in the field (network model field), so it will not be described in detail here.

[0059] In one embodiment, in step S4, the trained deep neural network includes: an input layer, an output layer and multiple hidden layers.

[0060] Among them, the input layer is used to input the random Gaussian sequence c; the multiple hidden layers are used to perform nonlinear transformation on the random Gaussian sequence c to extract high-dimensional features in the random Gaussian sequence c; and the output layer is used to output the optimized quantization function S(c).

[0061] It should be noted that in the process of training the original deep neural network, the training Gaussian sequence c is input through the input layer j ; Then, the training Gaussian sequence c is processed through multiple hidden layers j Perform nonlinear transformation to extract the training Gaussian sequence c j Finally, the output layer outputs S(c j ).

[0062] Specifically, after the trained deep neural network is obtained through the above steps, the random Gaussian sequence c is input into the trained deep neural network so that the trained deep neural network outputs an optimized quantization function S(c).

[0063] It should be noted that the optimized quantization function S(c) is used to quantize the received data c to ensure that the mutual information I(v; S(c)) after quantization is as close as possible to the mutual information I(x; y) before quantization (ie, the mutual information between x and y).

[0064] In one embodiment, the random Gaussian sequence c is preprocessed and then input into a trained deep neural network.

[0065] It should be noted that the quantum receiver adaptively optimizes the quantization function by training a deep neural network, and automatically learns the mapping relationship between the input c and the quantized output S(c) by training the deep neural network to minimize the quantization error and information loss, thereby improving the quantization efficiency.

[0066] Step S5: The quantum receiving end quantizes and divides the random Gaussian sequence c into m layers of original key strings L1 to L2 by optimizing the quantization function. m , and the first k layers of original key strings L1~L k It is directly sent to the quantum transmitter, and the original key string L of the mk layer is calculated according to the preset error correction code and optimized quantization function. k+1 ~L m The checksum S k+1 ~S m , and the checksum S k+1 ~S m Send to the quantum sending end.

[0067] It should be noted that 1≤k<m, m≥2, and k and m are both integers.

[0068] Specifically, the random Gaussian sequence c is divided into multiple slices (i.e., m layers of original key strings L1 to L m ), then the low-order slices (i.e. the original key strings L1 to L k ) is sent directly to the quantum transmitter, and the high-bit slice (i.e. the original key string L of the later mk layer) is sent directly to the quantum transmitter. k+1 ~L m ) is encoded through a classical channel (i.e., the original key string L after calculating the mk layer k+1 ~L m The checksum S k+1 ~S m ) and transmission (that is, to check the sub-S k+1 ~S m sent to the quantum sending end).

[0069] In one embodiment, in step S5, the quantum receiving end maps the random Gaussian sequence c into a discrete binary sequence through the optimized quantization function S(c), and multiplies the binary sequence with the preset error correction code to obtain the syndrome S k+1 ~S m .

[0070] In this embodiment, the binary sequence is the m-layer original key string L1-L m .

[0071] Step S6: After obtaining the target Gaussian sequence v, the quantum transmitter sets a maximum number of iterations within a layer, t, and a maximum number of iterations between layers, T.

[0072] It should be noted that t≥1, T≥1, t≥T, and t and T are both integers.

[0073] For example, in one embodiment, t=100; T=30 is set.

[0074] Step S7: The quantum transmitter uses the target Gaussian sequence v and the first k layers of original key strings L1 to L k Calculate the log-likelihood ratio of the k+ith layer, and then, based on the calculated log-likelihood ratio of the k+ith layer and the calibration factor S k+1 ~S m Perform k+i-th layer decoding.

[0075] It should be noted that i∈[1,mk], and i is an integer.

[0076] Specifically, the k+i-th layer decoding is completed using the LLR-BP decoding algorithm.

[0077] It should be noted that LLR, the full name of which is Log-Likelihood Ratio, is translated into Chinese as log-likelihood ratio; BP decoding algorithm, as a classic soft decision method, corrects some sudden errors by continuously iteratively updating the posterior probability, and has higher error correction capability than hard decision; LLR-BP decoding algorithm is an improvement on BP decoding, which reduces the amount of computation and the complexity of algorithm description.

[0078] Step S8, in the quantum transmitting end, when the decoding of the k+i layer is completed, the decoding result of the k+i layer is passed to the next layer for estimation and decoding until the decoding of the m layer is completed; after the decoding of the m layer is completed, if all layers are successfully decoded, the decoding success key is saved and the intra-layer iterative decoding is terminated; otherwise, the intra-layer iterative decoding is performed until the maximum number of iterations t within the layer is reached, the intra-layer iterative decoding is terminated, and the decoding result of the m layer is passed to the k+1 layer for inter-layer iterative decoding.

[0079] Specifically, at the quantum transmitting end, the k+1th layer is first decoded. After the decoding is completed, the decoding result of the k+1th layer is passed to the k+2th layer for estimation and decoding. After the decoding is completed, the decoding result of the k+2th layer is passed to the k+3th layer for estimation and decoding... until the decoding of the mth layer is completed.

[0080] It should be noted that, in this embodiment, the decoding results corresponding to the successfully decoded layers are all the same; the decoding success key is the decoding result when all layers are successfully decoded.

[0081] In one embodiment, the LLR-BP decoding algorithm is used to calculate the quantum transmitter syndrome and compare it with the syndrome (i.e., S k+1 ~S m ) for comparison, and then determine whether the two are the same. If they are the same, it means successful decoding; otherwise, the decoding fails.

[0082] It should be noted that the use of the LLR-BP decoding algorithm to calculate the quantum transmitter syndrome adopts the existing technology, so it will not be described in detail here.

[0083] It should be noted that if each layer from the k+1th layer to the mth layer is successfully decoded, it means that the intra-layer decoding is successful. At this time, the decoding success key is saved and the intra-layer iterative decoding is terminated. If no intra-layer decoding is successful, the iterative decoding is continued until the maximum number of iterations t within the layer is reached, and the intra-layer iterative decoding is terminated.

[0084] Step S9, in the inter-layer iterative decoding, when each layer of decoding is completed, the inter-layer iterative decoding result of each layer is compared with the decoding result of the corresponding layer in the step S8. If they are the same, the decoding success key is output and the inter-layer iterative decoding is completed; otherwise, the inter-layer iterative decoding is performed until the maximum number of inter-layer iterations T is reached, and the inter-layer iterative decoding is terminated.

[0085] Specifically, in the inter-layer iterative decoding, when each layer of decoding is completed, the inter-layer iterative decoding result of the layer is compared with the decoding result of the corresponding layer in step S8. If the two are the same, the decoding success key is output and the inter-layer iterative decoding is ended. Alternatively, if the maximum number of inter-layer iterations T has been reached, the inter-layer iterative decoding will also be ended, and at the same time, this round of information negotiation ends.

[0086] It should be noted that, in step S8, not all layers can be decoded successfully. For those layers that fail to be decoded, the process will proceed to inter-layer iterative decoding in step S9.

[0087] It should be noted that the key for successful decoding is used as the shared key, and the key for failed decoding and the original key string L1 to L2 of the previous k layers are used as the shared key. k will be discarded.

[0088] Specifically, the quantum sender uses an inter-layer iterative decoding algorithm to decode and recover the original key. After multiple iterations, the successfully decoded key string (i.e., the successfully decoded key) is left as the key of the quantum sender, while the key that failed to be decoded and the low-order key string sent directly are discarded.

[0089] It should be noted that in the CV-QKD system, deep neural networks can be used to optimize the quantization function to minimize the error in the quantization process. Compared with traditional quantization methods, deep learning can more accurately fit the relationship between the input signal c and the output quantization value S(c) by learning the nonlinear characteristics of the input signal c. Specifically, the deep neural network can learn the probability distribution and channel characteristics of the input signal through training sample data, and then adaptively adjust the quantization function to minimize the error after quantization and maximize the mutual information of the signal.

[0090] The present invention adopts deep learning technology and optimizes the quantization process by training a deep neural network, thereby minimizing information loss while maintaining system robustness. Specifically, the quantization method of the present invention first pre-processes the Gaussian sequence (i.e., the first original Gaussian sequence x) transmitted through the quantum channel, and uses the trained deep neural network to automatically learn the quantization rule that best suits the current channel noise characteristics. By optimizing the quantization function, the mutual information I(v; S(c)) after quantization is close to the mutual information I(x; y) before quantization, thereby effectively reducing information loss and improving quantization efficiency. In the data processing after quantization, the present invention also uses a rateless code to encode the quantized information, and transmits the encoded data to the quantum transmitter through a classical channel. The quantum transmitter uses an LLR-BP algorithm for decoding to restore the original key. The optimization of the deep learning model makes the quantized key information more accurate, thereby improving the information negotiation efficiency and key generation rate of the entire CV-QKD system.

[0091] The protection scope of the method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning described in the embodiment of the present invention is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing or replacing steps in the prior art based on the principles of the present invention are included in the protection scope of the present invention.

[0092] Those of ordinary skill in the art should further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0093] The descriptions of the processes or structures corresponding to the above-mentioned figures have different emphases. For parts that are not described in detail in a certain process or structure, please refer to the relevant descriptions of other processes or structures.

[0094] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning, characterized in that: The method comprises: Step 1: In the reverse Slice negotiation mechanism, the quantum transmitter obtains a first original Gaussian sequence x, and the quantum receiver obtains a second original Gaussian sequence y; both the first original Gaussian sequence x and the second original Gaussian sequence y obey Gaussian distribution; Step 2: The quantum receiving end randomly generates a random Gaussian sequence c, and adds the random Gaussian sequence c to the second original Gaussian sequence y to obtain a new Gaussian sequence w=c+y, and then sends the new Gaussian sequence w to the quantum sending end; Step 3: The quantum transmitter subtracts the first original Gaussian sequence x from the new Gaussian sequence w to obtain a target Gaussian sequence v=wx; Step 4: The quantum receiving end obtains an optimized quantization function based on the trained deep neural network; Step 5: The quantum receiving end quantizes and divides the random Gaussian sequence c into m layers of original key strings L1 to L2 by optimizing the quantization function. m , and the first k layers of original key strings L1~L k It is directly sent to the quantum transmitter, and the original key string L of the mk layer is calculated according to the preset error correction code and optimized quantization function. k+1 ~L m The checksum S k+1 ~S m , and the checksum S k+1 ~S m Send to the quantum transmitting end; 1≤k<m, m≥2, and k and m are both integers; Step 6: After the quantum transmitter obtains the target Gaussian sequence v, it sets a maximum number of iterations within a layer t and a maximum number of iterations between layers T; t≥1, T≥1, t≥T, and t and T are both integers; Step 7: The quantum transmitter uses the target Gaussian sequence v and the original key strings L1 to L2 of the first k layers. k Calculate the log-likelihood ratio of the k+ith layer, and then, based on the calculated log-likelihood ratio of the k+ith layer and the calibration factor S k+1 ~S m Perform k+i-th layer decoding; i∈[1,mk], and i is an integer; Step 8, in the quantum transmitting end, when the decoding of the k+i layer is completed, the decoding result of the k+i layer is passed to the next layer for estimation and decoding until the decoding of the m layer is completed; after the decoding of the m layer is completed, if all layers are successfully decoded, the decoding success key is saved and the intra-layer iterative decoding is terminated; otherwise, the intra-layer iterative decoding is performed until the maximum number of iterations t in the layer is reached, the intra-layer iterative decoding is terminated, and the decoding result of the m layer is passed to the k+1 layer for inter-layer iterative decoding; the decoding results corresponding to the successfully decoded layers are all the same; the decoding success key is the decoding result of all layers successfully decoded; Step nine, in the inter-layer iterative decoding, when each layer of decoding is finished, the inter-layer iterative decoding result of each layer is compared with the decoding result of the corresponding layer in the step eight. If they are the same, the decoding success key is output and the inter-layer iterative decoding is finished; otherwise, the inter-layer iterative decoding is performed until the maximum number of inter-layer iterations T is reached, and the inter-layer iterative decoding is finished.

2. The method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to claim 1 is characterized in that: In the step 1, the quantum receiving end obtains the second original Gaussian sequence y, which includes: the quantum transmitting end sends the first original Gaussian sequence x to the quantum receiving end through a quantum channel, so that the quantum receiving end measures the first original Gaussian sequence x to obtain the second original Gaussian sequence y associated with the quantum transmitting end; wherein, The first original Gaussian sequence is the modulation variance of the quantum transmitter; the noise z of the quantum channel obeys Gaussian distribution, is the noise variance of the quantum channel; y=x+z, 3. The method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to claim 1 is characterized in that: In the step 2, the random Gaussian sequence c obeys the Gaussian distribution. is the variance of the random Gaussian sequence c.

4. The method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to claim 1 is characterized in that: Before step 4, the method further includes: The quantum receiving end builds an original deep neural network; The quantum receiving end trains the original deep neural network based on a training set to obtain the trained deep neural network; the training set includes multiple training Gaussian sequences.

5. The method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to claim 4 is characterized in that: In the process of training the original deep neural network based on the training set at the quantum receiving end, the training is performed using a loss function; the calculation formula of the loss function is: L total =αL MSE +βL IM ; Among them, L total is the loss value calculated by the loss function; L MSE is the quantization error loss; L IM is the mutual information loss; α and β are hyperparameters; M is the total number of training Gaussian sequences contained in the training set; c j is the jth training Gaussian sequence in the training set; S(c j ) is based on c j When training the original deep neural network, the corresponding output of the original deep neural network; I(v); S(c j )) represents v, S(c j ) between them.

6. The method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to any one of claims 1 to 5, characterized in that: In step 4, the trained deep neural network includes: an input layer, an output layer and multiple hidden layers; wherein, The input layer is used to input the random Gaussian sequence c; The plurality of hidden layers are used to perform nonlinear transformation on the random Gaussian sequence c to extract high-dimensional features from the random Gaussian sequence c; The output layer is used to output the optimized quantization function.

7. The method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to claim 1 is characterized in that: In step 5, the quantum receiving end maps the random Gaussian sequence c into a discrete binary sequence through the optimized quantization function, and multiplies the binary sequence with the preset error correction code to obtain the syndrome S k+1 ~S m ; Wherein, the binary sequence is the m-layer original key string L1~L m .

8. The method for optimizing quantization operations in continuous variable quantum key distribution information negotiation based on deep learning according to claim 1 is characterized in that: In the step 2, the quantum receiving end sends the new Gaussian sequence w to the quantum transmitting end through a classical channel.

Images