A secure transmission method and system based on intelligent correction channel

Abstract

The present invention discloses a secure transmission method and system based on intelligent correction of channels, which obtains channel reciprocity degradation data and reference data that are not affected by channel reciprocity degradation, then corrects the channel reciprocity based on the autoencoder neural network model, and finally evaluates the security effect of the intelligent correction channel: the theoretical correction effect of the trained autoencoder model is obtained through simulation, a channel reciprocity correction model test experimental platform is designed and implemented, and the channel reciprocity correction effect of the trained model is verified through experiments. The channel reciprocity correction model disclosed by the present invention can significantly reduce the influence of multiple factors such as the difference in radio frequency devices at the transmitting and receiving ends and the asymmetry of uplink and downlink interference on the uplink and downlink channel differences, greatly enhance the ability of the transmitter to intelligently correct the channel reciprocity, adaptively learn the characteristics of channel differences, greatly save the feedback overhead of resource-constrained communication links, and effectively improve the fast and secure communication performance of TDD unmanned systems.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and in particular relates to a secure transmission method and system based on intelligent correction channels. Background Art

[0002] The wireless communication field has developed rapidly under the impetus of Multiple Input Multiple Output (MIMO) technology, but the broadcast nature of wireless communication allows unauthorized receivers to easily capture communication signals, obtain transmission information, and interfere with the wireless communication process, which has brought many security issues. Traditional cryptography-based security technology is a common defense method, but the increasing speed of computer computing has brought an increasingly obvious threat to cryptography-based security technology, and traditional cryptography security methods may be forcibly deciphered by computing power.

[0003] The advantages of physical layer communication security technology are becoming more and more obvious, especially the high frequency band characteristics and MIMO technology under the 5G standard provide critical intrinsic security physical resources for wireless physical layer communication security technology. Physical layer communication security technology is based on the difference in channel state information (CSI) between the sender and the legitimate receiver and eavesdropper to achieve communication security. The time-varying nature of the wireless channel causes the CSI to change all the time, which prevents the physical layer security technology from being cracked by the increasing computing power of computers.

[0004] As allocatable spectrum resources are gradually decreasing, spectrum allocation schemes are becoming increasingly complex, and there are more and more discrete spectrum resources in different frequency bands. The unique advantage of time division duplex (TDD) technology in effectively utilizing discrete spectrum resources is becoming increasingly important. By utilizing the reciprocity of uplink and downlink channels, the TDD system does not require the receiver to feedback channel state information. In the TDD-MIMO secure communication system assisted by uplink and downlink physical channel characteristics, the transmitter uses the uplink CSI for physical layer security transmission design, and the receiver uses the downlink CSI for information decoding. Ideally, the uplink CSI is exactly the same as the downlink CSI. However, in actual systems, channel reciprocity is limited by environmental factors such as RF gain and channel interference asymmetry, which will reduce the accuracy of the obtained channel state information, and there will be differences between the uplink CSI and the downlink CSI. Poor channel reciprocity will directly lead to serious interference when the signal is processed at both ends of the transmitter and receiver, reducing the quality of normal communication and significantly reducing the security transmission performance. For example, in a TDD system with non-ideal channel reciprocity, the transmitter uses the uplink CSI to process artificial noise signals, and the transmitted signal inevitably leaks to the legitimate user end, affecting the normal communication of the legitimate transceiver, and further deteriorating the channel accuracy obtained by the receiving end, seriously affecting the secure transmission effect based on physical characteristics. Summary of the invention

[0005] The purpose of the present invention is to intelligently calibrate the acquired uplink CSI data in a non-ideal channel environment through a channel reciprocity correction method based on an autoencoder neural network model, remove the influence of noise, improve the channel reciprocity in the TDD-MIMO communication system, and then enhance the physical layer security transmission based on physical characteristics; design and implement a channel reciprocity correction model test experimental platform, and verify the channel reciprocity correction effect of the trained model through experiments. The present invention can better improve the security transmission performance of the TDD-MIMO communication system assisted by physical characteristics.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] A secure transmission method based on intelligent correction channel, the method is implemented based on a TDD-MIMO communication system, the system comprises three communication nodes: a transmitter A, a legal receiver B and an eavesdropping end; the transmitter is composed of an industrial computer, four USRP2943Rs and an external clock source, and can realize 8 transmission and 8 reception functions at most; the legal receiver is composed of an industrial computer, two USRP2943Rs and an external clock source, and can realize 4 transmission and 4 reception functions at most; the eavesdropping end is composed of an industrial computer, two USRP 2943Rs and an external clock source, and can realize 4 transmission and 4 reception functions at most; the industrial computer and the USRP device are connected through a high-speed serial computer expansion bus standard bus cable; the method comprises the following steps:

[0008] S1. Acquisition of channel reciprocity degradation data: In a TDD-MIMO system under the influence of non-ideal channel reciprocity, a physical feature security verification experimental platform is built to obtain CSI data in a real channel environment, namely, source CSI data. Specifically, a legal receiver transmits a channel pilot signal to a transmitter, and the channel pilot signal is received by the transmitter after passing through the wireless channel. The transmitter obtains the uplink channel state information using a channel estimation method. The uplink CSI is used as training data for the autoencoder to correct the channel reciprocity model, namely, source CSI data x.

[0009] S2. Acquisition of reference data without channel reciprocity degradation: Acquisition of instant and accurate CSI data, i.e., target CSI data, through the TCP channel feedback link between the transmitter and the legitimate receiver;

[0010] S3. A channel reciprocity correction method based on an autoencoder to correct a channel reciprocity model: based on the target CSI data, the source CSI data is calibrated by correcting the channel reciprocity model through an autoencoder;

[0011] S4. Security effect evaluation based on intelligent correction channel: The theoretical correction effect of the trained autoencoder on the channel reciprocity model is obtained through simulation, and a channel reciprocity correction model test platform is designed and implemented. The channel reciprocity correction effect of the trained model is verified through experiments.

[0012] The step S2 specifically includes the following sub-steps:

[0013] S201: The transmitter transmits a channel pilot signal to a legitimate receiver. The channel pilot signal is received by the legitimate receiver through a wireless channel. The legitimate receiver obtains downlink channel information using a channel estimation method.

[0014] S202: A TCP channel feedback link is established between the transmitter and the legitimate receiver. The legitimate receiver feeds back the downlink CSI to the transmitter in real time and accurately through the TCP channel feedback link. The transmitter stores the downlink CSI as the neural network reference data that is not affected by the channel reciprocity degradation, that is, the target CSI data x target ;

[0015] S203. The transmitter broadcasts a publicly known channel pilot signal, which is received by the eavesdropping end through the wireless channel. The eavesdropping end uses a channel estimation method to obtain the transmitter-eavesdropping end link channel information for deciphering the confidential signal.

[0016] The step S3 is specifically as follows:

[0017] S301: Data preprocessing: Standardizing the real and imaginary parts of the source CSI data and the target CSI data, respectively, as shown below:

[0018]

[0019] Where, X norm is the standardized CSI data, X is the CSI data to be processed, and X min is the minimum value of the original CSI data, X max is the maximum value of the original CSI data, b and s are the minimum and maximum values ​​of the target range respectively;

[0020] S302: The source CSI data is used as training data and divided into a training set, a validation set and a test set. The training set is passed through an autoencoder model consisting of an encoding layer, a hidden layer and a decoding layer to obtain reconstructed source CSI data. The model is then optimized through a back propagation algorithm to minimize the loss function.

[0021] In each batch of training of the autoencoder model, the mean square error is used as the loss function to calculate the loss value between the reconstructed data and the target CSI data:

[0022]

[0023] Among them, L(x target ,x′) represents the loss function between the reconstructed data and the target CSI data, x target represents the target CSI data, and x′ represents the reconstructed source CSI data;

[0024] S303: Use the Adam optimization algorithm to minimize the loss function. After verification by the verification set, when the training meets the set requirements, stop the training and obtain the trained autoencoder to correct the channel reciprocity model.

[0025] The step S4 is specifically as follows:

[0026] S401: The transmitter sends the acquired downlink CSI data to the trained autoencoder to correct the channel reciprocity model, and the model reconstructs the source CSI data and transmits it back to the transmitter in real time;

[0027] S402: The transmitter performs mathematical matrix decomposition on the reconstructed source CSI to generate a physical layer security transmission signal based on physical characteristics.

[0028] The output dimensions of the encoding layer and the decoding layer of the autoencoder modified channel reciprocity model are the same as the input dimensions, which are both the number of sample features.

[0029] The encoder is a neural network, including a fully connected layer and a nonlinear activation function. The function of the encoder is to map the input data x to the hidden layer to obtain data z. The encoding process of the original data x from the input layer to the hidden layer is expressed as:

[0030] z=f E (W E x+b E )

[0031] Among them, x is the input data of the autoencoder neural network, z is the encoded data, and W E is the weight matrix of the encoder, b E is the encoder bias vector, f E (·) is the activation function of the encoder;

[0032] The decoder is a back-propagation neural network, and the decoding process from the hidden layer to the output layer is expressed as:

[0033]

[0034] in, is the decoded data, W D is the weight matrix of the decoder, b D is the decoder bias vector, f D (·) is the activation function of the decoder.

[0035] The present invention also provides a secure transmission system based on intelligent correction channels, which is implemented based on a TDD-MIMO communication system. The system includes three communication nodes: a transmitter, a legal receiver, an eavesdropping end, and a host. The transmitter is composed of an industrial computer, four USRP 2943Rs, and an external clock source, and can achieve a maximum of 8 transmit and 8 receive functions. The legal receiver Bob is composed of an industrial computer, two USRP 2943Rs, and an external clock source, and can achieve a maximum of 4 transmit and 4 receive functions. The eavesdropping end Eve is composed of an industrial computer, two USRP 2943Rs, and an external clock source, and can achieve a maximum of 4 transmit and 4 receive functions. The industrial computer and the USRP device are connected via a high-speed serial computer expansion bus standard bus cable. The host and the transmitter are connected via Ethernet.

[0036] The transmitter transmits the acquired CSI data under the real channel environment to the host via Ethernet. The host includes a program capable of running the trained autoencoder to correct the channel reciprocity model as described in claim 1, obtaining the reconstructed source CSI data, and then transmitting the reconstructed source CSI data to the transmitter in real time.

[0037] The beneficial effects of the present invention are as follows: the present invention utilizes the autoencoder neural network channel error feature learning method to be able to offline learn the differences in radio frequency devices at the transmitting and receiving ends and the asymmetry of uplink and downlink interference and other factors that affect the uplink and downlink channel differences, enhance the transmitter's ability to intelligently correct channel reciprocity in real time, and greatly improve the fast and secure communication performance of the TDD-MIMO system. Through simulation and experimental verification, the intelligent channel correction method proposed in the present invention can achieve the reciprocity of the wireless channel environment, and at the same time promote the application of physical layer security methods based on physical features in the field of wireless communications, which has important application value and development potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the hardware configuration of the three-communication node TDD-MIMO communication system of the present invention.

[0039] Figure 2 This is a physical picture of the three-communication node TDD-MIMO communication system.

[0040] Figure 3 This is an example of CSI data collected by the system.

[0041] Figure 4 Modify the channel reciprocity model structure for the autoencoder.

[0042] Figure 5 The curve of the training loss function of the autoencoder corrected channel reciprocity model changing with the training batch.

[0043] Figure 6These are the curves of Bob's receiving bit error rate changing with the artificial noise power allocation ratio in the three experimental systems (logarithmic vertical coordinates).

[0044] Figure 7 This is a constellation comparison diagram of the decoded confidential signal by the legitimate receiver Bob and the illegal eavesdropper Eve in the channel reciprocity correction model test platform when the ratio of artificial noise power to total transmission power is 20%. Figure 7 (a) is the constellation diagram of the legitimate receiver Bob receiving and decoding the secret signal. Figure 7 (b) in the figure is the constellation diagram used by the eavesdropper Eve to decipher the confidential signal. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0046] like Figure 1 , Figure 2 As shown in the figure, the TDD-MIMO system consists of three communication nodes: transmitter Alice (i.e., security control center), legal receiver Bob, and eavesdropper Eve. These three communication nodes are all composed of industrial computers and Universal Software Radio Peripheral Reconfigurable Input / Output (USRP RIO) devices provided by National Instruments (NI). Alice consists of an industrial computer, four USRP 2943Rs, and an external clock source, and can achieve up to 8 transmit and 8 receive functions. Bob consists of an industrial computer, two USRP 2943Rs, and an external clock source, and can achieve up to 4 transmit and 4 receive functions. Eve consists of an industrial computer, two USRP 2943Rs, and an external clock source, and can achieve up to 4 transmit and 4 receive functions. The industrial computer and USRP devices are connected via a high-speed serial computer expansion bus standard (Peripheral Component Interconnect Express, PCIe) bus cable.

[0047] A secure transmission method and system based on intelligent correction channel, comprising the following steps:

[0048] S1. Channel reciprocity degradation data acquisition: In a TDD unmanned system under the influence of non-ideal channel reciprocity, a physical feature safety verification experimental platform is built to obtain CSI in a real channel environment.

[0049] S101. The legitimate receiver Bob transmits a channel pilot signal to the transmitter Alice. The channel pilot signal is received by Alice after passing through the wireless channel. Alice obtains the uplink channel state information using the channel estimation method. The uplink CSI (which can be regarded as inaccurate CSI) is used as the training data for the autoencoder to correct the channel reciprocity model, which is called "source CSI data x".

[0050] S2. Acquisition of reference data that is not affected by channel reciprocity degradation: providing instant and accurate CSI data required for the autoencoder to correct the channel reciprocity model to correct the channel reciprocity.

[0051] S201. Bob first initializes the program to enter the waiting state for pilot signal capture.

[0052] S202. Alice initializes the USRP and the Transmission Control Protocol (TCP) client.

[0053] S203. Transmitter Alice transmits a publicly known channel pilot signal to legal receiver Bob. The channel pilot signal is received by Bob after passing through the wireless channel. Bob obtains downlink channel information using a channel estimation method.

[0054] S204. Transmitter Alice broadcasts a publicly known channel pilot signal, which is received by eavesdropper Eve after passing through the wireless channel. Eve obtains the channel state information of the Alice-Eve link using a channel estimation method, and stores the estimated illegal CSI for subsequent decryption of the confidential signal.

[0055] S205. A TCP channel feedback link is established between the transmitter Alice and the legitimate receiver Bob, and the server of the TCP link is set in the Bob program, and the TCP client is set in the Alice program. The program runs for a single delay of 100 milliseconds, that is, the server Bob scans the connection request or message from the client Alice 10 times per second, and also sends CSI data to the client Alice at most 10 times per second.

[0056] S206. Bob feeds back the downlink CSI to the transmitter Alice in real time and accurately through the TCP channel feedback link. The retransmission mechanism of TCP communication ensures that the program on Alice's end can successfully receive the downlink CSI data.

[0057] S207. Alice stores the downlink CSI as the reference data for the autoencoder to correct the channel reciprocity model without being affected by the channel reciprocity degradation. That is, the downlink CSI (which can be regarded as the accurate CSI) is used as the reference data for the autoencoder to correct the channel reciprocity model, which is called the “target CSI data x target ”.

[0058] S3. A channel reciprocity correction method based on an autoencoder to correct a channel reciprocity model: the uplink CSI data (source CSI data) is calibrated by correcting the channel reciprocity model through an autoencoder.

[0059] S301. Alice configures a high-performance host Cal to solve the high computing resource problem of the sending host Alice training the autoencoder to correct the channel reciprocity model. Cal and Alice are connected via Ethernet wired, and Alice transmits the stored "source CSI data" and "target CSI data" to the high-performance host Cal.

[0060] S302. In the scenario where Alice uses 4 transmitting antennas and Bob uses 2 receiving antennas, 40,000 sets of source CSI data and target CSI data are collected. The training data is divided into a training set, a validation set, and a test set, which contain 35,000, 3,000, and 2,000 data samples, respectively.

[0061] S303. When storing the source CSI data and the target CSI data, the complex channel state information data is divided into real and imaginary parts. Each CSI data has 512 rows and 4 columns, which contains the channel state information of 128 subcarriers used by Alice's 4 transmissions and orthogonal frequency division multiplexing (OFDM) and Bob's 2 receptions. Each data is accurate to 17 decimal places. The CSI data example is as follows: Figure 3 shown.

[0062] S304. The high-performance host Cal uses the "source CSI data" and the "target CSI data" to train the autoencoder neural network. The autoencoder designed by the present invention corrects the channel reciprocity model structure as shown in Figure 4 As shown. The autoencoder corrected channel reciprocity model is a neural network with the same input and learning target, that is, the input is "source CSI data" and the output is "reconstructed source CSI data". Its structure is divided into two parts: encoder and decoder. In order to ensure the reconstruction of the input source CSI data, the output dimension of the encoding layer and the decoding layer is the same as the input dimension, both of which are the number of sample features. The data is not compressed, and sufficient information is retained to accurately reconstruct the source CSI data and remove the influence of noise on channel reciprocity.

[0063] S305. Data preprocessing: Standardize the real and imaginary parts of the source CSI data and the target CSI data respectively, and scale the data to a specified range, in order to make the input data meet the training requirements of the autoencoder modified channel reciprocity model and promote model convergence. The Min-Max standardization operation is expressed as:

[0064]

[0065] Where, X norm is the standardized CSI data, X min is the minimum value of the original CSI data, X max is the maximum value of the original CSI data. b and s are the minimum and maximum values ​​of the target range, usually -1 and 1.

[0066] S306. The encoder is a 2-layer neural network, including 2 fully connected layers and nonlinear activation functions (such as ReLU). The function of the encoder part is to map the input data x to the hidden layer to obtain data z. The encoding process of the original data x from the input layer to the hidden layer can be expressed as:

[0067] z=f E (W E x+b E )

[0068] Among them, x is the input data of the autoencoder neural network, z is the encoded data, and W E is the weight matrix of the encoder, b E is the encoder bias vector, f E (·) is the activation function of the encoder.

[0069] S307. The decoder is also a multi-layer neural network with a similar structure to the encoder, but with back propagation. The output of the decoder is as close to the input data as possible to achieve source CSI data denoising and restore channel reciprocity. The decoding process from hidden layer to output layer:

[0070]

[0071] Among them, z is the data input from the hidden layer to the decoder, is the decoded data, W D is the weight matrix of the decoder, b D is the decoder bias vector, f D (·) is the activation function of the decoder.

[0072] S308. The training set is passed through the autoencoder composed of the encoding layer, hidden layer and decoding layer to correct the channel reciprocity model to obtain the reconstructed source CSI data, the purpose of which is to remove the influence of noise in the source CSI data, and then calculate the loss function, and optimize the model through the back propagation algorithm to minimize the loss function. Remove noise and generate meaningful representation;

[0073] In each batch of training, the mean square error is used as the loss function to calculate the loss value between the reconstructed data and the target CSI data:

[0074]

[0075] Among them, L(x target ,x′) represents the loss function between the reconstructed source CSI data and the target CSI data, x target represents the target CSI data, and x′ represents the reconstructed source CSI data.

[0076] The Adam optimization algorithm is used to minimize the loss function. The Adam optimization algorithm uses momentum and adaptive learning rate to speed up the model convergence and maintains the first-order moment estimate and second-order moment estimate of the gradient.

[0077] S309. The present invention designs an intelligent correction channel autoencoder to correct the channel reciprocity model training process is:

[0078] 1) Input: target CSI data x target , source CSI data x;

[0079] 2) Randomly initialize the autoencoder neural network encoder parameters W E 、b E , decoder parameters W D 、b D ;

[0080] 3) for each episode do

[0081] i. Perform forward propagation to obtain the output reconstructed source CSI data x′;

[0082] ii. Calculate the loss function

[0083] iii. Perform back-propagation and Adam algorithm optimization;

[0084] iv.if the loss function converges do

[0085] Fixed training parameters

[0086] else modify the training parameters

[0087] v. Input validation data set;

[0088] vi. Calculate the loss function;

[0089] vii.if the loss function is less than the selected reconstruction loss threshold do

[0090] Save training parameters;

[0091] else returns to the main program to restart the training process;

[0092] 4)end

[0093] 5) Output training parameters, namely channel characteristic error parameters.

[0094] like Figure 5 As shown in the figure, the curve of the training loss function of the autoencoder corrected channel reciprocity model with the training batch reflects the ability of the trained model to reconstruct the source CSI data and make it close to the target CSI data. During the model training process, the loss function shows a convergence trend with the increase of the number of training rounds; at the beginning of the training, the loss function increases instead of decreases due to the random initialization of the model parameters; after nearly 10 rounds of model training, the loss function begins to decrease; after about 30 rounds of model training, the loss function begins to converge rapidly; after 120 rounds of training, the mean square error loss function tends to be stable; when the number of training rounds reaches 180, the loss function is the smallest, and the minimum value is about 1.573. Compared with the untrained source CSI data, the loss function of the reconstructed source CSI data and target CSI data through model training decreased by about 92.58%.

[0095] S4. Security effect evaluation based on intelligent correction channel: The theoretical correction effect of the channel reciprocity model corrected by the autoencoder trained by S309 is obtained through simulation, and a channel reciprocity correction model test experimental platform is designed and implemented. The channel reciprocity correction effect of the trained model is verified through experiments.

[0096] S401. The high-performance host Cal transmits the reconstructed source CSI data to the host Alice in real time via an Ethernet cable.

[0097] S402. After obtaining high-precision reconstructed source CSI data, Alice performs mathematical matrix decomposition on the reconstructed source CSI to generate a physical layer security transmission signal based on physical characteristics.

[0098] Next, the specific embodiments of the physical layer secure transmission signal based on physical characteristics are all based on artificial noise signals as examples. Alice uses the autoencoder to correct the reconstructed source CSI matrix output by the channel reciprocity model, calculates the null space of the legal reconstructed source CSI matrix through singular value decomposition (SVD), and generates a null space artificial noise signal, while broadcasting a confidential message signal (referred to as "confidential signal").

[0099] S403. The physical layer security transmission signal based on physical characteristics is transmitted in the wireless channel environment and received by Bob and Eve respectively.

[0100] S404. The legitimate receiver Bob knows the reconstructed source CSI between the Alice-Bob link that satisfies high channel reciprocity and can correctly decode the confidential signal. Its receiving error performance directly reflects the accuracy of the channel reciprocity model modified by the autoencoder. The eavesdropper Eve does not know the CSI of the Alice-Bob link and can only use the illegal CSI estimated in the early stage to decipher the confidential signal. The uniqueness of the channel determines that Eve cannot decipher the confidential signal generated based on physical characteristics.

[0101] Artificial noise signals and confidential signals are received by Bob through the wireless channel environment. Since the reconstructed source CSI and the downlink CSI maintain a high channel reciprocity, the artificial noise signal has almost no effect on Bob. After receiving the signal, Bob uses the downlink CSI to complete the decoding of the confidential signal. Its error performance directly reflects the accuracy of the channel correction by the autoencoder neural network. In order to verify the security performance of the physical layer security transmission method based on physical characteristics, the eavesdropper Eve needs to be added. Eve uses the estimated illegal CSI to demodulate and decode the confidential signal interfered by the physical layer security transmission method. Since the illegal CSI is inconsistent with the reconstructed source CSI, the artificial noise signal will inevitably interfere with Eve's receiving performance. The degree of interference of the confidential signal obtained by Eve can be reflected in the security performance of the physical layer security transmission method based on physical characteristics in the communication system.

[0102] S405. The present invention uses the bit error rate (BER) of the receiving end as an evaluation index for measuring the reciprocity correction effect of the uplink and downlink channels and a security transmission performance index.

[0103] In order to demonstrate the higher channel reciprocity correction effect of the autoencoder corrected channel reciprocity model, Figure 6The experimental platform tests the artificial noise-assisted secure transmission system. The legitimate receiver Bob receives the confidential signal BER as the proportion of artificial noise power changes. It includes three sets of experimental verification systems. The "Autoencoder neural network reconstructs the source CSI data" in the legend represents the performance curve of the physical feature secure transmission using the reconstructed source CSI based on the autoencoder neural network channel reciprocity correction model test experimental platform. The "target CSI data" in the legend represents the performance curve of Alice directly using the feedback target CSI data for secure transmission. The "source CSI data" in the legend represents the performance curve of Alice using the source CSI data for secure transmission. It can be seen from the figure that the autoencoder neural network channel reciprocity correction model has a good correction effect, and the received BER is almost close to 0, which reflects that the reconstructed source CSI trained by Alice using the autoencoder correction channel reciprocity model is basically consistent with the target CSI data, basically meeting the channel reciprocity requirements, and Bob's receiving BER curve is close to the best performance "target CSI data" solution.

[0104] Figure 7 The constellation diagrams of the confidential signal received by the legitimate receiver Bob and the eavesdropper Eve are shown when the proportion of artificial noise power is 20%. As can be seen from the figure, the security performance of the legitimate receiver Bob is basically not affected by the physical layer security transmission of physical characteristics using the reconstructed source CSI data obtained by the autoencoder channel reciprocity correction model, and the receiving end constellation diagram converges very well, meeting the channel reciprocity requirements; the constellation diagram of the signal after receiving and decoding by the eavesdropper Eve diverges, and the confidential signal cannot be correctly decoded, achieving the purpose of secure transmission of the communication system, and verifying the effectiveness of the channel correction based on the autoencoder neural network proposed in the present invention.

[0105] The above is a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention, and should be within the scope of protection of the claims attached to the present invention.

Claims

1. A secure transmission method based on intelligent correction channel, characterized in that: The method is implemented based on a TDD-MIMO communication system, which includes three communication nodes: a transmitter A, a legal receiver B and an eavesdropping end E; the transmitter is composed of an industrial computer, four universal software radio peripherals USRP 2943R and an external clock source, and can achieve 8 transmission and 8 reception functions at most; the legal receiver is composed of an industrial computer, two USRP 2943R and an external clock source, and can achieve 4 transmission and 4 reception functions at most; the eavesdropping end is composed of an industrial computer, two USRP 2943R and an external clock source, and can achieve 4 transmission and 4 reception functions at most; the industrial computer and the USRP device are connected through a high-speed serial computer expansion bus standard bus cable; the method includes the following steps: S1. Channel reciprocity degradation data acquisition: In a TDD-MIMO system under the influence of non-ideal channel reciprocity, a physical feature security verification experimental platform is built to obtain CSI data in a real channel environment, namely, source CSI data; S2. Acquisition of reference data without channel reciprocity degradation: Acquisition of instant and accurate CSI data, i.e., target CSI data, through the TCP channel feedback link between the transmitter and the legitimate receiver; S3. A channel reciprocity correction method based on an autoencoder to correct a channel reciprocity model: Based on the target CSI data, the source CSI data is calibrated by correcting the channel reciprocity model through an autoencoder; the details are as follows: S301: Data preprocessing: The real part and the imaginary part of the source CSI data and the target CSI data are respectively subjected to standardization operations, which are expressed as follows: Where, X norm is the standardized CSI data, X is the CSI data to be processed, and X min is the minimum value of the original CSI data, X max is the maximum value of the original CSI data, b and s are the minimum and maximum values ​​of the target range respectively; S302: The source CSI data is used as training data and divided into a training set, a validation set and a test set. The training set is passed through an autoencoder composed of a coding layer, a hidden layer and a decoding layer to correct the channel reciprocity model to obtain reconstructed source CSI data. The model is then optimized through a back propagation algorithm to minimize the loss function. In each batch of training of the autoencoder corrected channel reciprocity model, the mean square error is used as the loss function to calculate the loss value between the reconstructed source CSI data and the target CSI data: Among them, L(x target ,x′) represents the loss function between the reconstructed source CSI data and the target CSI data, x target represents the target CSI data, and x′ represents the reconstructed source CSI data; S303: Using the Adam optimization algorithm to minimize the loss function, after verification by the verification set, when the training meets the set requirements, the training is stopped to obtain the trained autoencoder corrected channel reciprocity model; S4. Security effect evaluation based on intelligent correction channel: The theoretical correction effect of the trained autoencoder to correct the channel reciprocity model is obtained through simulation, and a channel reciprocity correction model test experimental platform is designed and implemented. The channel reciprocity correction effect of the trained autoencoder to correct the channel reciprocity model is verified through experiments.

2. A secure transmission method based on intelligent correction channel according to claim 1, characterized in that: The steps for obtaining the source CSI data are as follows: The legitimate receiver transmits a channel pilot signal to the transmitter. The channel pilot signal is received by the transmitter after passing through the wireless channel. The transmitter obtains the uplink channel state information using the channel estimation method. The uplink CSI is used as the training data for the autoencoder to correct the channel reciprocity model, that is, the source CSI data.

3. A secure transmission method based on intelligent correction channel according to claim 1, characterized in that: The step S2 specifically includes the following sub-steps: S201: The transmitter transmits a channel pilot signal to a legitimate receiver. The channel pilot signal is received by the legitimate receiver through a wireless channel. The legitimate receiver obtains downlink channel information using a channel estimation method. S202: A TCP channel feedback link is established between the transmitter and the legitimate receiver. The legitimate receiver feeds back the downlink CSI to the transmitter in real time and accurately through the TCP channel feedback link. The transmitter stores the downlink CSI as reference data that is not affected by channel reciprocity degradation, that is, target CSI data. S203. The transmitter broadcasts a publicly known channel pilot signal, which is received by the eavesdropping end through the wireless channel. The eavesdropping end uses a channel estimation method to obtain the transmitter-eavesdropping end link channel information for deciphering the confidential signal.

4. A secure transmission method based on intelligent correction channel according to claim 1, characterized in that: The step S4 is specifically as follows: S401: The transmitter sends the acquired downlink CSI data to the trained autoencoder to correct the channel reciprocity model, and the model reconstructs the source CSI data and transmits it back to the transmitter in real time; S402: The transmitter performs mathematical matrix decomposition on the reconstructed source CSI to generate a physical layer security transmission signal based on physical characteristics.

5. A secure transmission method based on intelligent correction channel according to claim 4, characterized in that: The output dimensions of the encoding layer and the decoding layer of the autoencoder modified channel reciprocity model are the same as the input dimensions, which are both the number of sample features.

6. A secure transmission method based on intelligent correction channel according to claim 5, characterized in that: The autoencoder corrected channel reciprocity model structure is divided into two parts: encoder and decoder; the encoder is a neural network, including a fully connected layer and a nonlinear activation function. The function of the encoder is to map the input data to the hidden layer to obtain data z. The encoding process of the original data from the input layer to the hidden layer is expressed as: z=f E (W E x+b E ) Among them, x is the input data of the autoencoder neural network, z is the encoded data, and W E is the weight matrix of the encoder, b E is the encoder bias vector, f E (·) is the activation function of the encoder; The decoder is a back-propagation neural network, and the decoding process from the hidden layer to the output layer is expressed as: in, is the decoded data, W D is the weight matrix of the decoder, b D is the decoder bias vector, f D (·) is the activation function of the decoder.

7. A secure transmission system based on an intelligent correction channel, characterized in that: The system is a TDD-MIMO communication system, which includes four communication nodes: a transmitter, a legal receiver, an eavesdropping end, and a host; the transmitter is composed of an industrial computer, four USRP 2943Rs, and an external clock source, and can achieve a maximum of 8 transmit and 8 receive functions; the legal receiver Bob is composed of an industrial computer, two USRP 2943Rs, and an external clock source, and can achieve a maximum of 4 transmit and 4 receive functions; the eavesdropping end Eve is composed of an industrial computer, two USRP 2943Rs, and an external clock source, and can achieve a maximum of 4 transmit and 4 receive functions; the industrial computer and the USRP device are connected through a high-speed serial computer expansion bus standard bus cable; the host and the transmitter are connected through Ethernet; The transmitter transmits the acquired CSI data under the real channel environment to the host via Ethernet. The host runs a program to implement the secure transmission method based on the intelligent correction channel as described in any one of claims 1 to 6, obtains the reconstructed source CSI data, and then transmits the reconstructed source CSI data to the transmitter in real time.

Images