Encrypted semantic communication method for end-to-end wireless communication scene

Abstract

The invention discloses an end-to-end wireless communication scene-oriented encrypted semantic communication method, which comprises the following steps: constructing an encrypted semantic communication system, and carrying out system network training based on a preset network total loss function which is obtained based on a loss function of a communication receiving end and a loss function of an attacker network model. After network training is completed, original text information is mapped to high-dimensional vector representation to obtain an embedded vector, extracted semantic features and session keys are input into an encryption module to obtain a ciphertext message, and the ciphertext message is input into a channel encoder to generate a symbol stream transmitted on a wireless channel; and receiving a signal corresponding to the symbol stream by using a communication receiving end, and enabling the signal to sequentially pass through a channel decoder, a decryption module and a semantic decoder to obtain reconstructed original text information. According to the method, confidentiality protection can be provided for key semantic information transmitted by legal users on the premise that extraction and interaction of semantic features are not affected.

Description

Technical Field

[0001] The present invention relates to the technical field of communication security research, and in particular to an encrypted semantic communication method for end-to-end wireless communication scenarios. Background Art

[0002] Semantic communication technology combines natural language processing, machine learning, and artificial intelligence technologies, focusing on the transmission of semantic information rather than the communication symbols themselves, and analyzing and processing the transmitted information at the semantic level. However, the inherent openness and accessibility of wireless channels make the semantic feature information transmitted in plain text in the channel extremely vulnerable to eavesdropping attacks. Eavesdropping devices in wireless channels use the characteristics of electromagnetic waves that are transmitted freely in space to adjust the receiving frequency of the device to intercept the information transmitted in the channel, resulting in the confidentiality of key feature information transmitted by legitimate users being destroyed.

[0003] In order to prevent the transmitted semantic feature data from being attacked by eavesdropping, one of the commonly used key technologies is encryption. The legitimate communication party encrypts the transmitted content, and only the legitimate receiver can decrypt the information to prevent the information from being obtained or accessed by unauthorized personnel, providing confidentiality protection for the transmitted data, thereby ensuring the transmission security of the wireless communication system. Under the deep learning network architecture, an encryption communication algorithm based on adversarial training network has been proposed and widely used. The basic principle of the adversarial encryption scheme is to achieve confidentiality protection of the transmitted message by reasonably designing the loss function between the legitimate communication parties and the attacker. During the training process, the legitimate receiver can accurately reconstruct the message and increase the attacker's information reconstruction error, thereby improving the security of the wireless communication system.

[0004] As an emerging wireless communication paradigm, the security situation faced by semantic communication technology has not been fully studied. The existing research on encrypted semantic communication technology is still insufficient, and it is difficult to effectively resist security threats in practical applications.

[0005] In end-to-end semantic communication networks, existing research has focused on adding additional artificial noise to the bit stream transmitted in the wireless channel or converting computationally difficult problems into composite noise during the transmission process to prevent information leakage, but the additional interference introduced will affect the accuracy and robustness of the original semantic communication system. Research related to encrypted semantic communication based on deep network architecture and adversarial training networks encrypts the original information and extracts semantic features from the ciphertext information, ignoring the rich semantic information contained in the original information and losing the original interpretable data; or independently training the semantic communication system and security module, resulting in the overall performance of the system being limited by the independent training of a single module. Summary of the invention

[0006] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0007] To this end, in the semantic communication scenario, for the semantic feature information transmitted in the wireless channel, the present invention proposes an encrypted semantic communication method for end-to-end wireless communication scenarios, which provides confidentiality protection for the key semantic information transmitted by legitimate users without affecting the extraction and interaction of semantic features.

[0008] A second object of the present invention is to provide a computer device.

[0009] A third object of the present invention is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above object, the present invention proposes, on one hand, an encrypted semantic communication method for end-to-end wireless communication scenarios, comprising:

[0011] Constructing an encrypted semantic communication system using a key processing module, a semantic codec, an encryption and decryption module, and a channel codec, and training a network of the encrypted semantic communication system based on a preset network total loss function; wherein the network total loss function is obtained based on a loss function of a communication receiving end and a loss function of an attacker's network model;

[0012] After the network training is completed, the original text information sent by the communication sender is mapped to a high-dimensional vector representation through a semantic encoder to obtain an embedded vector, and semantic features are extracted from the embedded vector. The semantic features and the session key passed through the key processing module are input into the encryption module to obtain a ciphertext message. Finally, the ciphertext message is input into the channel encoder to generate a symbol stream transmitted on the wireless channel.

[0013] The communication receiving end is used to receive the signal corresponding to the symbol stream, and the signal is sequentially passed through a channel decoder, a decryption module and a semantic decoder to obtain the reconstructed original text information.

[0014] The encrypted semantic communication method for end-to-end wireless communication scenarios in the embodiments of the present invention may also have the following additional technical features:

[0015] In one embodiment of the present invention, the semantic encoder and the encryption module are respectively constructed using the encoding layer of Transformer, and the semantic decoder and the decryption module are respectively constructed using the decoding layer of Transformer; the channel codec is composed of dimensionally symmetric fully connected layers.

[0016] In one embodiment of the present invention, the communication sending end and the communication receiving end use the same session key k and transmit the original text information based on the encrypted semantic communication system. The original text information m is passed through the semantic encoder Se α Get semantic feature f; f is used as encryption module Enc βThe ciphertext message c is obtained from the input; finally c passes through the channel encoder Ce γ The symbol stream x transmitted on the wireless channel is obtained, that is:

[0017] x=Ce γ (Enc β (Se α (m),K κ (k)))(13)

[0018] Among them, α, β, γ are the parameter sets of semantic encoder, encryption module and channel encoder respectively; K k () is a key processing module shared between two legal communicating parties and is used to establish the same session key in the symmetric encryption algorithm. k is the parameter set of the corresponding network module.

[0019] In one embodiment of the present invention, the communication sending end and the communication receiving end are both legal communication parties. As a legal receiving party, the signal received by the communication receiving end is set to It is expressed as follows:

[0020]

[0021] Among them, h is the wireless channel coefficient between the sender and the receiver, n is the power σ 2 Additive Gaussian white noise;

[0022] After passing through the channel decoder Cd χ , decryption module Dec δ and semantic decoder Sd θ After that, the received signal is decoded to obtain the reconstructed original text information. It is expressed as:

[0023]

[0024] Among them, θ, δ, χ are the parameter sets of the semantic decoder, decryption module and channel decoder respectively.

[0025] In one embodiment of the present invention, the attacker network model is used to intercept information transmitted in the channel and based on a random key Reconstruct the original text information, namely:

[0026]

[0027] In one embodiment of the present invention, the semantic error is quantified using a cross entropy function, and the loss function of the communication receiving end is defined based on semantic channel joint coding:

[0028]

[0029] The cross entropy function is:

[0030]

[0031] Among them, p(w l ),q(w l ) represent the true probability of the lth word appearing in message m1 and the predicted probability of appearing in message m2 respectively.

[0032] In one embodiment of the present invention, the training of the network of the encrypted semantic communication system based on a preset network total loss function includes:

[0033] Initialize the parameters of each network module of the encrypted semantic communication system;

[0034] Generate the session key of the legitimate communicating parties and the attacker's random key;

[0035] The semantic part loss function of the semantic encoder and decoder is defined as formula (8), and the semantic encoder and decoder are trained independently according to the defined semantic part loss function:

[0036]

[0037] Among them, s is the batch data in the training process;

[0038] The encryption part loss function of the encryption and decryption module is defined as formula (9), and the encryption and decryption modules of the two legitimate communication parties are independently trained according to the defined encryption part loss function:

[0039]

[0040] The joint loss function at the communication receiving end is defined as:

[0041]

[0042] The loss function of the attacker network model is defined as:

[0043]

[0044] Based on the principle of adversarial cryptography, the joint loss function of the communication receiving end and the loss function of the attacker's network model are combined to define the joint loss function of the encrypted semantic communication system network as shown in formula (12). The entire network module of the legitimate communication parties is jointly trained according to the defined joint loss function of the encrypted semantic communication system network:

[0045]

[0046] Among them, λ is used as a hyperparameter of the adversarial training network.

[0047] The encrypted semantic communication method for end-to-end wireless communication scenarios of the embodiments of the present invention implements symmetric encryption of semantic feature information transmitted between two legitimate communicating parties based on the basic principles of adversarial cryptography, thereby achieving end-to-end secure transmission.

[0048] To achieve the above-mentioned purpose, the second aspect embodiment of the present application proposes a computer device, comprising: a processor and a memory; wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the encrypted semantic communication method for end-to-end wireless communication scenarios as described in the first aspect embodiment.

[0049] To achieve the above-mentioned purpose, the third aspect embodiment of the present application proposes a non-temporary computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the encrypted semantic communication method for end-to-end wireless communication scenarios as described in the first aspect embodiment.

[0050] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0052] Figure 1 is a flow chart of an encrypted semantic communication method for an end-to-end wireless communication scenario according to an embodiment of the present invention;

[0053] Figure 2 is a schematic diagram of the architecture of an encrypted semantic communication system according to an embodiment of the present invention;

[0054] Figure 3 is a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0056] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0057] The following describes an encrypted semantic communication method, device, and storage medium for end-to-end wireless communication scenarios proposed according to embodiments of the present invention with reference to the accompanying drawings.

[0058] Figure 1 is a flow chart of an encrypted semantic communication method for end-to-end wireless communication scenarios according to an embodiment of the present invention. Figure 1 As shown, including:

[0059] S1, constructing an encrypted semantic communication system using a key processing module, a semantic codec, an encryption and decryption module, and a channel codec, and training a network of the encrypted semantic communication system based on a preset network total loss function; wherein the network total loss function is obtained based on a loss function of a communication receiving end and a loss function of an attacker's network model;

[0060] S2, after the network training is completed, the original text information sent by the communication sender is mapped to a high-dimensional vector representation through a semantic encoder to obtain an embedded vector, and the semantic features are extracted from the embedded vector, and the semantic features and the session key passed through the key processing module are input into the encryption module to obtain the ciphertext message, and finally the ciphertext message is input into the channel encoder to generate a symbol stream transmitted on the wireless channel;

[0061] S3, using the communication receiving end to receive the signal of the corresponding symbol stream, and passing the signal through the channel decoder, decryption module and semantic decoder in sequence to obtain the reconstructed original text information.

[0062] It is understandable that in the existing communication field, Alice and Bob are often used as representatives of senders and receivers. In the context of secure communication, they are often used to describe the process of encryption and decryption.

[0063] The following are the general steps for Alice and Bob to perform a text message transmission task:

[0064] Key Exchange: Alice and Bob first need a shared secret key. This can be achieved by using a key exchange protocol such as the Diffie-Hellman protocol.

[0065] Message Encryption: Alice uses the shared key to encrypt the text message to be sent. This can be done using a symmetric encryption algorithm such as AES.

[0066] Information transmission: The encrypted information is sent to Bob through an insecure channel.

[0067] Message Decryption: After Bob receives the encrypted message, he decrypts it using the same shared key to recover the original text.

[0068] Therefore, in semantic communication systems, semantic feature information transmitted in wireless channels is usually transmitted in plain text, and sensitive information is easy to leak.

[0069] like Figure 2 As shown, the communication sending end Alice and the communication receiving end Bob of the present invention perform text information transmission tasks as the legitimate communication parties, and Eve, as an attacker, intercepts key information transmitted in the wireless channel and attempts to reconstruct the semantic information.

[0070] In one embodiment of the present invention, Alice and Bob use the same session key k and transmit text information based on the designed encrypted semantic communication system. The text information m is used as the original message and passes through the semantic encoder Se α Get semantic feature f; f is used as encryption module Enc β The ciphertext message c is obtained from the input; finally c passes through the channel encoder Ce γ Get the symbol stream x transmitted on the wireless channel, that is

[0071] x=Ce γ (Enc β (Se α (m),K κ (k)))(25)

[0072] Among them, α, β, γ are the parameter sets of semantic encoder, encryption module and channel encoder respectively; K k () is a key processing module shared between two legal communicating parties and is used to establish the same session key in the symmetric encryption algorithm, and κ is the parameter set of the network module.

[0073] Symmetrically, as a legitimate receiver, the signal received by Bob is set to It is expressed as follows:

[0074]

[0075] Where h is the wireless channel coefficient between the sender and the receiver (assumed to be a constant in a quasi-static channel); n is the power σ 2 Additive White Gaussian Noise (AWGN). After passing through the channel decoder Cd χ , decryption module Dec δ and semantic decoder Sd θ After that, the received signal is decoded to reconstruct the original text information, which is expressed as:

[0076]

[0077] Among them, θ, δ, χ are the parameter sets of the semantic decoder, decryption module and channel decoder respectively.

[0078] like Figure 2 As shown in the figure, the network architecture of the attacker Eve is similar to that of the legitimate receiver Bob, but Eve is unaware of the session key k between the legitimate communicating parties. She intercepts the information transmitted in the channel and tries to use random keys to Reconstruct the original data to achieve the purpose of eavesdropping, namely:

[0079]

[0080] Under the assumptions of the above-mentioned legitimate participants and attacker models, the proposed encrypted semantic communication scheme aims to maximize Bob's reconstruction accuracy to accurately recover the required text information while ensuring the confidentiality of the transmitted information, that is, reducing Eve's reconstruction accuracy.

[0081] It can be understood that the present invention encrypts the semantic feature information transmitted in the wireless channel and does not interfere with the semantic communication system's extraction and interpretation of the semantic level information.

[0082] In one embodiment of the present invention, based on the goal of the above-mentioned scheme, the main architecture of the network of the encrypted semantic communication system proposed by the present invention is based on the Transformer network, and the Transformer encoding layer is used to construct the semantic encoder and encryption module, and the symmetrical decoding layer is used to construct the semantic decoder and decryption module; the channel codec is composed of dimensionally symmetrical fully connected layers, and the specific parameter design can be adjusted according to the channel conditions.

[0083] Exemplarily, the processing process of the encrypted semantic communication method for end-to-end wireless communication scenarios proposed in the present invention may include the following steps:

[0084] (1) The embedding layer of the semantic codec first transforms the input original text information M and maps it to a high-dimensional vector representation to obtain an embedded vector;

[0085] (2) The semantic encoder (Transformer encoding layer-1) extracts semantic features F from the embedding vector;

[0086] (3) The semantic feature F and the session key K passed through the key processing module are used as inputs to the encryption module (Transformer encoding layer-2) to obtain the ciphertext message C;

[0087] (4) Before entering the wireless channel for transmission, the ciphertext message C is input into the channel encoder and generates a symbol stream X.

[0088] At the same time, the processing process of the communication receiving end is symmetrical with that of the communication sending end. The received signal Y is reconstructed through the channel decoder, decryption module (Transformer decoding layer-2) and semantic decoder (Transformer decoding layer-1) to obtain the original text information.

[0089] It is understandable that according to the Transformer network architecture, during the training process, in addition to the output of the channel encoder In addition to the session key K, the target result of the input message is used as part of the input to the decryption module to calculate the loss and update the parameters of the model.

[0090] Based on the Transformer network architecture proposed in this invention, according to the goal of the above scheme, a reasonable loss function is designed to achieve the expected purpose. Considering that the transmitted data is in text form, the cross entropy (CE) function is used to quantify the semantic error. Based on the idea of ​​semantic channel joint coding, the loss function of the legitimate receiver Bob is defined as:

[0091]

[0092] Among them, the cross entropy function is:

[0093]

[0094] Among them, p(w l ),q(w l ) represent the true probability of the lth word appearing in message m1 and the predicted probability of it appearing in message m2 respectively.

[0095] It can be understood that Semantic Channel Joint Coding (SCJC or JSCC) is an advanced communication coding technology that combines the functions of traditional source coding (used to compress and optimize information transmission) and channel coding (used for error detection and correction) to improve the efficiency and reliability of data transmission, especially in communication systems with limited bandwidth and noisy environments. The core idea of ​​Semantic Channel Joint Coding is to take into account both the semantics (i.e., meaning) of the source information and the characteristics of the channel in the coding process, so as to make more efficient use of communication resources.

[0096] In one embodiment of the present invention, training a network of an encrypted semantic communication system based on a preset network loss function specifically includes the following steps:

[0097] Step 1: To support the independent use of semantic encoder and decoder, the semantic part loss function is designed as follows:

[0098]

[0099] Among them, s is the batch data in the training process.

[0100] Step 2: Similar to the above training process, in order to support the independent use of the encryption and decryption modules, the encryption part loss function is designed as follows:

[0101]

[0102] During the training process, in order to further improve the security of the system, the network module training follows the one-time-one-key principle and uses a unique key for each individual message.

[0103] According to the independent training of step 1 and step 2, for both parties of legitimate communication, the basic network modules of the proposed encrypted semantic communication system can be used independently and realize the corresponding semantic feature extraction, text source message reconstruction and encryption and decryption functions. Focusing on the semantic communication architecture and encryption and decryption modules, the training and update of the channel codec and its parameter set are included in the joint training process.

[0104] Step 3: In order to further reduce the redundancy generated by the separate coding scheme and jointly train the entire encrypted semantic communication system, on the basis of independently training each network module, a joint source-encryption-channel coding architecture is designed. The joint loss function of the legitimate receiver Bob is defined as:

[0105]

[0106] Among them, m, are the original message at the sending end and the reconstructed message at the receiving end defined in formulas (1) and (3) respectively.

[0107] Step 4: For the attacker Eve, who has the same network structure as the legitimate receiver but different model parameters, the training and updating of its network model is similar to that of Bob. It attempts to reconstruct the original text information through the data eavesdropped in the channel. The loss function is defined as:

[0108]

[0109] Since Eve, as an eavesdropping attacker, only intercepts the data transmitted in the wireless channel without actively interfering with the channel or legitimate users, it does not affect the network parameter update process of the legitimate communicating parties Alice and Bob during the training process.

[0110] Step 5: Based on the principle of adversarial cryptography, combined with the loss functions of the legitimate receiver and the attacker, the joint loss function of the encrypted semantic communication network is defined as:

[0111]

[0112] Among them, λ is a hyperparameter of the adversarial training network, which is used to limit Eve's reconstruction ability. This parameter can be adaptively adjusted according to the attacker's assumptions to achieve a trade-off between the availability and security of the entire system.

[0113] Understandably, the core principle of adversarial cryptography is to ensure that cryptographic systems can protect the confidentiality, integrity, and availability of data even in the face of motivated and capable attackers.

[0114] The present invention designs an overall loss function for adversarial training, which can be adaptively adjusted according to the assumptions of the legitimate communicating parties and the attacker, thereby limiting the accuracy of information reconstructed by an eavesdropping attacker in a wireless channel.

[0115] The present invention jointly designs the source-encryption-channel to improve the mutual dependence between various network modules in the system, and through joint training and iterative updating, ensures the independent use of various modules of the encrypted semantic communication system and the overall performance of the system.

[0116] Through the above loss function definition, the training steps of the network of the encrypted semantic communication system of the present invention can be the following process:

[0117] S11, initializing parameters of each network module of the encrypted semantic communication system;

[0118] S21, repeat the training until all epochs are completed or the early stopping condition is met, including:

[0119] S2.11 Generate the session key of the legitimate communicating parties and the attacker's random key;

[0120] S2.21 Independently train the semantic codecs of the legitimate communicating parties according to formula (8);

[0121] S2.31 independently trains the encryption and decryption modules of both parties in legal communication according to formula (9);

[0122] S2.41 jointly trains the entire network module of both parties of the legitimate communication according to formula (12);

[0123] S2.51 trains the attacker network module against the formula (11).

[0124] S31, returning the system model parameters of the two legal communicating parties.

[0125] Thus, based on the training of each network module, a trained encrypted semantic communication system is obtained, and based on the trained encrypted semantic communication system, the text information transmission task taking the above-mentioned end-to-end wireless communication scenario as an example is performed.

[0126] The encrypted semantic communication scheme proposed in the embodiment of the present invention is combined with a semantic-channel joint coding architecture, and a joint training loss function is designed based on the basic idea of ​​adversarial cryptography. On the basis of the legitimate sender extracting the semantic features of the text source and the legitimate receiver accurately restoring the text semantic information, the malicious receiver is agnostic to the transmission content, and confidentiality protection is provided for the key feature information transmitted by the semantic communication system, thereby achieving end-to-end secure transmission.

[0127] According to the encrypted semantic communication method for end-to-end wireless communication scenarios in the embodiments of the present invention, the present invention proposes to encrypt the transmitted semantic features, and provide confidentiality protection for the wireless communication system on the basis of retaining the interpretable data of the original message. On the basis of joint source-channel coding, the encryption and decryption module is combined with the original semantic communication network architecture, and from the perspective of both parties of legitimate communication, the overall performance of the encrypted semantic communication system and the attack capability of eavesdroppers in the wireless channel are considered, and the semantic codec, encryption and decryption device, and channel codec are jointly trained to minimize the reconstruction loss of semantic feature information and avoid interference introduced in the process of semantic information extraction and interpretation due to the independence of modules. In addition, based on the designed loss function, by reasonably selecting the network model hyperparameters, it is possible to implement encrypted semantic communication schemes under different transmission and attacker assumptions, and provide security services for semantic communication systems in end-to-end wireless transmission scenarios.

[0128] In order to implement the method of the above embodiment, the present invention also provides a computer device, such as Figure 3 As shown, the computer device 600 includes a memory 601 and a processor 602; wherein the processor 602 runs a program corresponding to the executable program code by reading the executable program code stored in the memory 601, so as to implement each step of the method described above.

[0129] In order to implement the above embodiments, the present application also proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in the above embodiments is implemented.

[0130] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0131] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

Claims

1. An encrypted semantic communication method for end-to-end wireless communication scenarios, characterized in that: include: Constructing an encrypted semantic communication system using a key processing module, a semantic codec, an encryption and decryption module, and a channel codec, and training a network of the encrypted semantic communication system based on a preset network total loss function; wherein the network total loss function is obtained based on a loss function of a communication receiving end and a loss function of an attacker's network model; After the network training is completed, the original text information sent by the communication sender is mapped to a high-dimensional vector representation through a semantic encoder to obtain an embedded vector, and semantic features are extracted from the embedded vector. The semantic features and the session key passed through the key processing module are input into the encryption module to obtain a ciphertext message. Finally, the ciphertext message is input into the channel encoder to generate a symbol stream transmitted on the wireless channel. The communication receiving end is used to receive the signal corresponding to the symbol stream, and the signal is sequentially passed through a channel decoder, a decryption module and a semantic decoder to obtain the reconstructed original text information.

2. The method according to claim 1, characterized in that The semantic encoder and encryption module are constructed using the encoding layer of Transformer respectively, and the semantic decoder and decryption module are constructed using the decoding layer of Transformer respectively; the channel codec is composed of dimensionally symmetric fully connected layers.

3. The method according to claim 1, characterized in that The communication sending end and the communication receiving end use the same session key k and transmit the original text information based on the encrypted semantic communication system. The original text information m is passed through the semantic encoder Se α Get the semantic feature f; f is used as the encryption module Enc β The ciphertext message c is obtained from the input; finally c passes through the channel encoder Ce γ The symbol stream x transmitted on the wireless channel is obtained, that is: x=This γ (Enc β (Se α (m),K κ (k))) (1) Among them, α, β, γ are the parameter sets of semantic encoder, encryption module and channel encoder respectively; K κ () is a key processing module shared between two legal communicating parties and is used to establish the same session key in the symmetric encryption algorithm, and κ is a parameter set of the corresponding network module.

4. The method according to claim 3, characterized in that The communication sending end and the communication receiving end are both legal communication parties. As a legal receiving party, the signal received by the communication receiving end is set as It is expressed as follows: Among them, h is the wireless channel coefficient between the sender and the receiver, n is the power σ 2 Additive Gaussian white noise; After passing through the channel decoder Cd χ , decryption module Dec δ and semantic decoder Sd θ After that, the received signal is decoded to obtain the reconstructed original text information. It is expressed as: Among them, θ, δ, χ are the parameter sets of the semantic decoder, decryption module and channel decoder respectively.

5. The method according to claim 4, characterized in that The attacker network model is used to intercept the information transmitted in the channel and based on the random key Reconstruct the original text information, namely:

6. The method according to claim 5, characterized in that The semantic error is quantified using the cross entropy function, and based on the semantic channel joint coding, the loss function of the communication receiving end is defined as: The cross entropy function is: Among them, p(w l ),q(w l ) represent the true probability of the lth word appearing in message m1 and the predicted probability of appearing in message m2 respectively.

7. The method according to claim 6, characterized in that The method of training the network of the encrypted semantic communication system based on a preset network total loss function includes: Initialize the parameters of each network module of the encrypted semantic communication system; Generate the session key of the legitimate communicating parties and the attacker's random key; The semantic part loss function of the semantic encoder and decoder is defined as formula (8), and the semantic encoder and decoder are trained independently according to the defined semantic part loss function: Among them, s is the batch data in the training process; The encryption part loss function of the encryption and decryption module is defined as formula (9), and the encryption and decryption modules of the two legitimate communication parties are independently trained according to the defined encryption part loss function: The joint loss function at the communication receiving end is defined as: The loss function of the attacker network model is defined as: Based on the principle of adversarial cryptography, the joint loss function of the communication receiving end and the loss function of the attacker's network model are combined to define the joint loss function of the encrypted semantic communication system network as shown in formula (12). The entire network module of the legitimate communication parties is jointly trained according to the defined joint loss function of the encrypted semantic communication system network: Among them, λ is used as a hyperparameter of the adversarial training network.

8. A computer device, characterized in that: including a processor and a memory; The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the encrypted semantic communication method for end-to-end wireless communication scenarios as described in any one of claims 1-7.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the encrypted semantic communication method for end-to-end wireless communication scenarios as described in any one of claims 1-7 is implemented.

Images