Privacy enhancement semantic communication method based on trusted execution environment security reasoning
Through the combination of TEE and REE, model parameter disassembly and data confusion are used to solve the problem of neural network parameter leakage in semantic communication, and efficient and confidential data transmission and calculation are achieved.
Patent Information
- Application Number
- CN202510563484.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-08
AI Technical Summary
When existing semantic communication technologies are deployed on a large scale, neural network parameters are prone to leak, resulting in information leakage and white box attacks, and the calculation overhead of existing protection methods is too large.
The method of combining the trusted execution environment TEE and the rich execution environment REE is adopted. TEE performs model inference and obfuscates the intermediate data. REE performs linear operations, TEE recovers the calculation results, and uses the disassembly and obfuscation methods of model parameters to protect model parameters and intermediate results.
It effectively avoids information leakage, improves the confidentiality and computing efficiency of data transmission, and realizes efficient model inference operations.
Smart Images
Figure CN120456003A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of wireless security technology, and in particular to a privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment. Background Art
[0002] Among related technologies, semantic communication, as a new communication paradigm, combines artificial intelligence with a focus on transmitting semantic information rather than the communication symbols themselves, performing joint analysis and processing of the source and channel. This technology improves the transmission efficiency and capabilities of wireless communication systems, adapting to increasingly complex and diverse application scenarios. It has attracted widespread attention from academia and industry and has experienced rapid development. Semantic communication, as distinct from grammatical communication, is communication that extracts and transmits the information required for a specific communication task.
[0003] The semantic encoder is responsible for performing semantic analysis on the transmitted data and extracting semantic features; the channel coder comprehensively analyzes the semantic features and channel conditions and compresses the semantic features into wireless signals suitable for transmission; the channel decoder restores the received signal to semantic features; the semantic decoder analyzes the semantic features and outputs the results to be transmitted.
[0004] The communication process consists of two phases: model inference and wireless transmission. The model inference phase involves the various neural network components performing corresponding operations on a given input; the wireless transmission phase involves the transmission of the channel encoder's output wireless signal symbols across the wireless channel. However, large-scale deployment of semantic communication systems requires directly deploying the neural network model on the vast number of nodes that need to communicate, which can result in the leakage of neural network parameters. Summary of the Invention
[0005] To overcome the problems existing in the related art, the present disclosure provides a privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment.
[0006] According to a first aspect of an embodiment of the present disclosure, a privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment is provided, comprising:
[0007] Input the data to be transmitted into the trusted execution environment TEE;
[0008] Performing a first model inference operation on the data to be transmitted using the TEE;
[0009] In a case where the first model inference operation requires determining a linear operation result on the intermediate data x and the model parameter w, performing obfuscation processing on the intermediate data x according to a first obfuscation method to obtain an obfuscated result; the intermediate data is extracted from the data to be transmitted;
[0010] Inputting the obfuscated result into a rich execution environment (REE);
[0011] controlling the REE to perform a linear operation according to a linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part of the model parameter w, thereby obtaining an obfuscated linear operation result; the model parameter w includes a first part and a second part, and the first part and the second part are obtained by pre-decomposing the model parameter w;
[0012] Inputting the obfuscated linear operation result into the TEE, and controlling the TEE to determine the linear operation result before obfuscation based on the first obfuscation mode, the first part, and the model parameter w;
[0013] When the first model reasoning operation is completed, the semantic features obtained by the first model reasoning operation are sent to the target recipient, so that the target recipient performs a second model reasoning operation based on the semantic features to obtain target data; the operation mode of the second model reasoning operation is the same as the operation mode of the first model reasoning operation.
[0014] In some embodiments of the present disclosure, performing obfuscation processing on x according to the first obfuscation method to obtain an obfuscation result includes:
[0015] Generate a random number n;
[0016] The random number n is added to the intermediate data x to obtain the obfuscation result.
[0017] In some embodiments of the present disclosure, the tensor shape of the random number n is the same as the tensor shape of the intermediate data x.
[0018] In some embodiments of the present disclosure, when the first model reasoning operation is completed, sending the semantic features obtained by the first model reasoning operation to a target recipient includes:
[0019] When the first model inference operation is completed, encrypting the semantic feature using a secret key;
[0020] The encrypted semantic feature is sent to a target recipient, so that the target recipient uses the secret key to decrypt the encrypted semantic feature to obtain the semantic feature.
[0021] In some embodiments of the present disclosure, when the first model reasoning operation is completed, sending the semantic features obtained by the first model reasoning operation to a target recipient includes:
[0022] When the first model inference operation is completed, inputting the semantic feature into a channel encoder to obtain a wireless signal output by the channel encoder;
[0023] Mapping the wireless signal to a constellation diagram to obtain coordinates of symbol points corresponding to the wireless signal in the constellation diagram, and using the coordinates of the symbol points as quantized wireless signals to be transmitted;
[0024] The quantized wireless signal to be transmitted is sent to the target receiver.
[0025] In some embodiments of the present disclosure, the first part and the second part are obtained by decomposing the model parameter w through the following formula:
[0026] w=a·b
[0027] Among them, a is the first part and b is the second part.
[0028] According to a second aspect of an embodiment of the present disclosure, a privacy-enhanced semantic communication device based on secure reasoning in a trusted execution environment is provided, comprising:
[0029] An input unit, used to input data to be transmitted into the trusted execution environment TEE;
[0030] a model inference unit, configured to perform a first model inference operation on the data to be transmitted using the TEE;
[0031] an obfuscation unit configured to, when the first model inference operation requires determining a linear operation result on the intermediate data x and the model parameter w, perform obfuscation processing on the intermediate data x according to a first obfuscation method to obtain an obfuscated result; the intermediate data is extracted from the data to be transmitted;
[0032] An input unit, further configured to input the obfuscation result into a rich execution environment (REE);
[0033] a linear operation unit, configured to control the REE to perform a linear operation according to a linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part of the model parameter w, to obtain an obfuscated linear operation result; the model parameter w includes a first part and a second part, the first part and the second part being obtained by pre-decomposing the model parameter w;
[0034] a determining unit, configured to input the obfuscated linear operation result into the TEE, and control the TEE to determine the linear operation result before obfuscation based on the first obfuscation mode, the first part, and the model parameter w;
[0035] A sending unit is used to send the semantic features obtained by the first model reasoning operation to a target recipient when the first model reasoning operation is completed, so that the target recipient performs a second model reasoning operation based on the semantic features to obtain target data; the operation mode of the second model reasoning operation is the same as the operation mode of the first model reasoning operation.
[0036] According to a third aspect of an embodiment of the present disclosure, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described in any one of the first aspects is implemented.
[0037] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method according to any one of the first aspects is implemented.
[0038] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program implements the method as described in any one of the first aspects when executed by a processor.
[0039] The technical solution provided by the embodiments of the present disclosure may include the following beneficial effects: by inputting the data to be transmitted into the trusted execution environment (TEE); using the TEE to perform a first model reasoning operation on the data to be transmitted; when the first model reasoning operation requires determining the linear operation result for the intermediate data x and the model parameter w, the intermediate data x is obfuscated according to the first obfuscation method to obtain an obfuscated result; the intermediate data is extracted from the data to be transmitted; the obfuscated result is input into the rich execution environment (REE); controlling the REE to perform a linear operation according to the linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part of the model parameter w to obtain an obfuscated linear operation result; inputting the linear operation result into the TEE, and controlling the TEE to determine the linear operation result before obfuscation based on the first obfuscation method, the first part, and the model parameter w; when the first model reasoning operation is completed, sending the semantic features obtained by the first model reasoning operation to the target recipient, thereby utilizing the combination of the TEE and the REE to efficiently perform the first model reasoning operation while keeping the model parameters and intermediate results in the reasoning process confidential, effectively avoiding information leakage caused by semantic feature extraction of the data to be transmitted, and improving the confidentiality of data transmission.
[0040] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0042] Figure 1 The present invention is a flowchart showing a privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to an exemplary embodiment.
[0043] Figure 2 The figure is a schematic diagram showing an existing data transmission method according to an exemplary embodiment.
[0044] Figure 3 The figure is a schematic diagram of a model reasoning process according to an exemplary embodiment.
[0045] Figure 4 The figure is a data transmission flowchart according to an exemplary embodiment.
[0046] Figure 5 The present invention is a block diagram showing a privacy-enhanced semantic communication device based on secure reasoning in a trusted execution environment according to an exemplary embodiment.
[0047] Figure 6 The present invention is a block diagram showing an apparatus for a privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to an exemplary embodiment. DETAILED DESCRIPTION
[0048] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.
[0049] The terms used in the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure. The singular forms "a", "an" and "the" used in the present disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.
[0050] It should be understood that although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present disclosure, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the embodiments of the present disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0051] Furthermore, the various forms of processes shown in the embodiments of this disclosure may be used to reorder, add, or delete steps. For example, the steps described in this application may be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not a limitation herein.
[0052] Among related technologies, semantic communication, as a new communication paradigm, combines artificial intelligence with a focus on transmitting semantic information rather than the communication symbols themselves, performing joint analysis and processing of the source and channel. This technology improves the transmission efficiency and capabilities of wireless communication systems, adapting to increasingly complex and diverse application scenarios. It has attracted widespread attention from academia and industry and has experienced rapid development. Semantic communication, as distinct from grammatical communication, is communication that extracts and transmits the information required for a specific communication task.
[0053] The semantic communication system architecture is shown in the attached Figure 2 ,An end-to-end semantic communication system consists of a semantic codec and a channel codec. Both of them are implemented by deep learning neural networks.
[0054] The semantic encoder is responsible for performing semantic analysis on the transmitted data and extracting semantic features; the channel coder comprehensively analyzes the semantic features and channel conditions and compresses the semantic features into wireless signals suitable for transmission; the channel decoder restores the received signal to semantic features; the semantic decoder analyzes the semantic features and outputs the results to be transmitted.
[0055] The communication process is divided into two phases: the model inference phase and the wireless transmission phase. The model inference phase refers to the calculations performed by each neural network component on a given input; the wireless transmission phase refers to the transmission of the output wireless signal symbols of the channel encoder over the wireless channel.
[0056] It should be noted that large-scale deployment of semantic communication systems requires directly deploying neural network models on the massive number of nodes that need to communicate. This can lead to the leakage of neural network parameters, which vendors providing closed-source models are reluctant to disclose.
[0057] Furthermore, the leakage of neural network parameters and intermediate results of model inference facilitates white-box attacks. A white-box attack in neural network attacks involves an attacker fully understanding the target model's internal structure, parameters, and training data, and using this information to eavesdrop (obtain information about transmitted data), steal models (obtain model parameters), and reverse engineer models (obtain information about the training dataset).
[0058] In addition, the open nature of wireless communications allows attackers to easily obtain transmitted wireless signals. The wireless signals transmitted by semantic communication contain rich information, which helps attackers carry out attacks.
[0059] Moreover, although multi-party secure computing and homomorphic encryption can protect model parameters from being leaked, the computational overhead is too high.
[0060] In order to solve the above problems, the present disclosure provides a privacy-enhanced semantic communication method based on secure reasoning of a trusted execution environment, which comprises the following steps: inputting data to be transmitted into a trusted execution environment (TEE); utilizing TEE to perform a first model reasoning operation on the data to be transmitted; utilizing TEE to decompose the model parameter w into a first part and a second part; when the first model reasoning operation needs to determine the result of a linear operation on intermediate data x and model parameter w, obfuscating the intermediate data x according to a first obfuscation method to obtain an obfuscated result; extracting the intermediate data from the data to be transmitted; inputting the obfuscated result and the second part into a rich execution environment (REE); and controlling REE to perform a first obfuscation operation on the data based on the obfuscation result and the first obfuscation method. In the second part, the linear operation is performed according to the linear calculation type corresponding to the linear operation result to obtain the obfuscated linear operation result; the linear operation result is input into the TEE, and the TEE is controlled to determine the linear operation result before obfuscation based on the first obfuscation method, the first part and the model parameter w; when the first model inference operation is completed, the semantic features obtained by the first model inference operation are sent to the target recipient, thereby utilizing the combination of TEE and REE to efficiently execute the first model inference operation while keeping the model parameters and intermediate results in the inference process confidential, effectively avoiding information leakage caused by semantic feature extraction of the transmitted data, and improving the confidentiality of data transmission.
[0061] Figure 1 FIG is a flowchart showing a privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to an exemplary embodiment. Figure 1 As shown, it should be noted that the privacy-enhancing semantic communication method based on trusted execution environment security reasoning of the embodiment of the present disclosure is applied to the privacy-enhancing semantic communication device based on trusted execution environment security reasoning. Figure 1 As shown, the method may include the following steps:
[0062] Step 101: Input the data to be transmitted into the Trusted Execution Environment (TEE).
[0063] It should be noted that TEE is a secure execution environment based on hardware isolation that protects the confidentiality and integrity of sensitive data.
[0064] In one embodiment, steps 101 to 107 may be performed using a pre-trained neural network model.
[0065] As an example, TEE can be configured in all communication nodes, and users of each communication node can use mature technologies to securely transmit the model to TEE.
[0066] Step 102: Use TEE to perform a first model inference operation on the data to be transmitted.
[0067] It is understandable that the purpose of model reasoning is to extract the required semantic features from the data to be transmitted.
[0068] Step 103 , when the first model inference operation needs to determine the linear operation result for the intermediate data x and the model parameter w, the intermediate data x is obfuscated according to the first obfuscation method to obtain an obfuscated result.
[0069] The intermediate data is extracted from the data to be transmitted.
[0070] It should be noted that the model inference process requires multiple linear and nonlinear calculations, with the nonlinear calculations being performed directly by the TEE. Furthermore, the intermediate data x and model parameters w mentioned above can be data associated with any linear operation. In other words, whenever a linear operation is required, obfuscation is performed, allowing the REE to perform linear calculations based on the obfuscated results.
[0071] In one embodiment, the TEE is first used to encrypt and obfuscate the data to be unloaded, and then the REE is responsible for performing linear calculations. Finally, the TEE restores the required calculation results based on the linear calculation results. In this way, even if the data in the REE is leaked, the attacker cannot obtain the real w and x.
[0072] In some embodiments of the present disclosure, step 104 may specifically include the following steps:
[0073] Generate a random number n;
[0074] Add the random number n to the intermediate data x to get the obfuscated result.
[0075] In some embodiments of the present disclosure, the tensor shape of the random number n is the same as the tensor shape of the intermediate data x.
[0076] Step 104: Input the obfuscated result into a Rich Execution Environment (REE).
[0077] It should be noted that REE refers to a conventional operating system environment, which is not very secure but has more powerful computing capabilities.
[0078] In one embodiment, the model parameter w, the intermediate result x, the split value a, and the random number n are stored in the TEE, and only the obfuscated b and x+n are stored in the REE. Even if the data in the REE is leaked, the attacker cannot obtain the real w and x.
[0079] Step 105 : Control the REE to perform a linear operation according to the linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part of the model parameter w, to obtain an obfuscated linear operation result.
[0080] The model parameter w includes a first part and a second part, and the first part and the second part are obtained by decomposing the model parameter w in advance.
[0081] As an example, the linear calculation type may be a matrix multiplication calculation or a convolution calculation.
[0082] In a possible embodiment, the second parts corresponding to different model parameters and the corresponding linear calculation types may be stored in the REE in advance.
[0083] In one embodiment, for a linear calculation w×x, by offloading it to the REE, the TEE only needs to calculate a·rw×n. Since w×n is independent of intermediate results, it can be calculated in advance. Therefore, during real-time inference, the TEE only needs to calculate one matrix multiplication and one matrix subtraction, which effectively improves the TEE's computational efficiency and, in turn, the efficiency of model inference.
[0084] In one embodiment, the above-mentioned model parameter w may be a model parameter of a convolutional neural network model.
[0085] In some embodiments of the present disclosure, the first part and the second part are obtained by decomposing the model parameter w through the following formula:
[0086] w=a·b
[0087] Among them, a is the first part and b is the second part.
[0088] In one embodiment, to reduce the complexity of model reasoning, the model parameters are decomposed so that the REE performs linear operations on a portion of them, effectively reducing the amount of computation required by the TEE during real-time reasoning.
[0089] In step 106 , the linear operation result after obfuscation is input into the TEE, and the TEE is controlled to determine the linear operation result before obfuscation based on the first obfuscation mode, the first part, and the model parameter w.
[0090] In one embodiment, the TEE stores the first obfuscation mode, the first part, and the model parameter w, so the linear operation result before obfuscation can be obtained by reverse deduction based on the linear operation and the above information.
[0091] For example, if Figure 3 As shown in the figure, after receiving the data x+n, the REE performs a linear operation r=b×(x+n) and sends r to the TEE. The TEE uses the confidentially stored a, w, and n to calculate a·rw×n. This result is the required w×x result because:
[0092] a·rw×n=a·b×(x+n)-w×n=c×b×(x+n)-w×n=w×(x+n)-w×n
[0093] =w×x
[0094] Here, c satisfies c×b=a·b=w.
[0095] For example, when the operation · is a scalar multiplication, the operation × is a matrix multiplication, a is a scalar, and b is a matrix c=aI, I is the unit matrix, and this condition holds.
[0096] Step 107, when the first model reasoning operation is completed, the semantic features obtained by the first model reasoning operation are sent to the target recipient, so that the target recipient performs a second model reasoning operation based on the semantic features to obtain target data.
[0097] The operation mode of the second model reasoning operation is the same as that of the first model reasoning operation.
[0098] In some embodiments of the present disclosure, after receiving the semantic feature signal sent by the sender, the target receiver needs to sequentially use a channel decoder to restore the signal to semantic features, and further use a semantic decoder to perform semantic analysis on the semantic features. In other words, the channel decoder and the semantic decoder are used to perform a second model inference operation in the same manner as the first model inference operation, thereby obtaining the target data. The second model inference operation process includes the following steps:
[0099] Step a1: When the second model inference operation requires determining a linear operation result on the intermediate data x and the model parameter w, the intermediate data x is obfuscated according to the first obfuscation method to obtain an obfuscated result; the intermediate data is extracted from the data to be transmitted;
[0100] Step a2: input the obfuscated result into the rich execution environment (REE);
[0101] Step a3: Controlling the REE to perform a linear operation based on the obfuscated result and the second part of the model parameter w according to the linear calculation type corresponding to the linear operation result, thereby obtaining an obfuscated linear operation result; the model parameter w includes a first part and a second part, and the first part and the second part are obtained by pre-decomposing the model parameter w;
[0102] In step a4, the linear operation result after obfuscation is input into the TEE, and the TEE is controlled to determine the linear operation result before obfuscation based on the first obfuscation method, the first part, and the model parameter w.
[0103] It should be noted that the specific implementation of the above steps a1-a4 can be implemented using any implementation disclosed in the embodiments of the present disclosure.
[0104] In some embodiments of the present disclosure, step 107 may specifically include the following steps:
[0105] When the first model inference operation is completed, the semantic features are encrypted using the secret key;
[0106] The encrypted semantic features are sent to a target recipient, so that the target recipient uses a secret key to decrypt the encrypted semantic features to obtain the semantic features.
[0107] In some embodiments of the present disclosure, the data sender and the data receiver may perform key negotiation in advance to obtain the key P. An attacker can only obtain the encrypted wireless signal and cannot decrypt the wireless signal.
[0108] In some embodiments of the present disclosure, step 107 may specifically include the following steps:
[0109] When the first model inference operation is completed, the semantic feature is input into the channel encoder to obtain a wireless signal output by the channel encoder;
[0110] Mapping the wireless signal to a constellation diagram to obtain coordinates of symbol points corresponding to the wireless signal in the constellation diagram, and using the coordinates of the symbol points as quantized wireless signals to be transmitted;
[0111] The quantized wireless signal to be transmitted is sent to the target receiver.
[0112] In one embodiment, after the first model inference operation is completed, the semantic features are input into a channel encoder, which comprehensively analyzes the semantic features and channel conditions and compresses the semantic features into a wireless signal suitable for transmission.
[0113] It's important to note that in communications theory, a constellation diagram is a graphical tool used to represent digitally modulated signals. It visually displays the signal's distribution on the complex plane (IQ plane), with each point corresponding to a specific symbol, used to transmit digital information.
[0114] In one embodiment, signal symbols can be mapped to the closest symbol point in the constellation diagram to obtain the symbol point coordinates, which are then used as the quantized wireless signal to be transmitted. The constellation diagram is designed through machine learning. The forward propagation process of model training simulates this mapping process and adds corresponding training objectives, such as minimizing the distance between the channel encoder output symbol and the closest symbol point in the constellation diagram; minimizing the loss function of the semantic communication model task; minimizing the information entropy of the signal symbol, and so on.
[0115] In addition, various methods can be used to simplify wireless signal symbols and achieve a constellation diagram with limited symbol points. First, quantization can be used. In addition to quantizing single scalars, such as 8-bit and 16-bit quantization, vector quantization can also be performed on the entire vector.
[0116] In one embodiment, the sender encrypts the quantized wireless signal Sig to be transmitted based on the key p to obtain Enc p (Sig).
[0117] In the disclosed embodiment, a quantization mechanism is utilized to operate the semantic communication system on a constellation diagram of finite symbol points, which can significantly reduce the requirements for hardware accuracy of the communication equipment. In addition, the transmission result can eliminate interference in the wireless channel to facilitate real-time encryption and decryption solutions.
[0118] In some embodiments of the present disclosure, Figure 4As shown, when the sender needs to transmit data for communication, the semantic encoder is used to perform the model reasoning proposed in this disclosure to obtain semantic features. The channel encoder converts the semantic features into wireless signals. After the wireless signals are quantized according to the quantization processing method proposed in this disclosure, they are encrypted and the encrypted data is sent to the receiver through a wireless channel. The receiver decrypts the received data, then uses the channel decoder to decode the data, and uses the semantic decoder to perform semantic decoding to obtain the data to be sent by the sender. Since the receiver also has the need to send data to other parties, the receiver can also install TEE, thereby using a combination of TEE and REE to perform model reasoning.
[0119] According to the privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment (TEE) proposed in an embodiment of the present disclosure, the method includes inputting data to be transmitted into a trusted execution environment (TEE); utilizing the TEE to perform a first model reasoning operation on the data to be transmitted; utilizing the TEE to decompose the model parameter w into a first part and a second part; when the first model reasoning operation requires determining a linear operation result for intermediate data x and model parameter w, obfuscating the intermediate data x according to a first obfuscation method to obtain an obfuscated result; extracting the intermediate data from the data to be transmitted; inputting the obfuscated result and the second part into a rich execution environment (REE); controlling the REE to perform a linear operation according to a linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part to obtain an obfuscated linear operation result; inputting the linear operation result into the TEE, and controlling the TEE to determine the linear operation result before obfuscation based on the first obfuscation method, the first part, and the model parameter w; and upon completion of the first model reasoning operation, sending the semantic features obtained from the first model reasoning operation to a target recipient. Thus, the method utilizes a combination of the TEE and the REE to efficiently perform the first model reasoning operation while keeping the model parameters and intermediate results in the reasoning process confidential, effectively avoiding information leakage caused by semantic feature extraction on the data to be transmitted and improving the confidentiality of data transmission.
[0120] Figure 5 This is a block diagram of a privacy-enhanced semantic communication device based on secure reasoning in a trusted execution environment according to an exemplary embodiment. Figure 5 The device includes an input unit 501, a model inference unit 502, a confusion unit 503, an input unit 504, a linear operation unit 505, a determination unit 506 and a sending unit 507.
[0121] The input unit 501 is used to input the data to be transmitted into the trusted execution environment TEE;
[0122] A model inference unit 502 is configured to perform a first model inference operation on the data to be transmitted using the TEE;
[0123] The obfuscation unit 503 is configured to, when the first model inference operation requires determining a linear operation result on the intermediate data x and the model parameter w, perform obfuscation processing on the intermediate data x according to the first obfuscation method to obtain an obfuscated result; the intermediate data is extracted from the data to be transmitted;
[0124] The input unit 504 is further configured to input the obfuscated result and the second part into the rich execution environment REE;
[0125] A linear operation unit 505 is used to control the REE to perform a linear operation according to the linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part, to obtain an obfuscated linear operation result;
[0126] a determination unit 506 configured to input the obfuscated linear operation result into the TEE and control the TEE to determine the pre-obfuscated linear operation result based on the first obfuscation mode, the first part, and the model parameter w;
[0127] The sending unit 507 is used to send the semantic features obtained by the first model reasoning operation to the target recipient when the first model reasoning operation is completed, so that the target recipient performs a second model reasoning operation based on the semantic features to obtain target data; the operation mode of the second model reasoning operation is the same as the operation mode of the first model reasoning operation.
[0128] In some embodiments of the present disclosure, the obfuscation unit 503 may be specifically configured to:
[0129] Generate a random number n;
[0130] Add the random number n to the intermediate data x to get the obfuscated result.
[0131] In some embodiments of the present disclosure, the tensor shape of the random number n is the same as the tensor shape of the intermediate data x.
[0132] In some embodiments of the present disclosure, the sending unit 507 may be specifically configured to:
[0133] When the first model inference operation is completed, the semantic features are encrypted using the secret key;
[0134] The encrypted semantic features are sent to a target recipient, so that the target recipient uses a secret key to decrypt the encrypted semantic features to obtain the semantic features.
[0135] In some embodiments of the present disclosure, the sending unit 507 may be specifically configured to:
[0136] When the first model inference operation is completed, the semantic feature is input into the channel encoder to obtain a wireless signal output by the channel encoder;
[0137] Mapping the wireless signal to a constellation diagram to obtain coordinates of symbol points corresponding to the wireless signal in the constellation diagram, and using the coordinates of the symbol points as quantized wireless signals to be transmitted;
[0138] The quantized wireless signal to be transmitted is sent to the target receiver.
[0139] In some embodiments of the present disclosure, the first part and the second part are obtained by decomposing the model parameter w through the following formula:
[0140] w=a·b
[0141] Among them, a is the first part and b is the second part.
[0142] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0143] According to the privacy-enhanced semantic communication device based on secure reasoning in a trusted execution environment (TEE) proposed in an embodiment of the present disclosure, the device inputs data to be transmitted into a trusted execution environment (TEE); uses the TEE to perform a first model reasoning operation on the data to be transmitted; uses the TEE to decompose the model parameter w into a first part and a second part; when the first model reasoning operation requires determining a linear operation result for intermediate data x and model parameter w, the intermediate data x is obfuscated according to a first obfuscation method to obtain an obfuscated result; the intermediate data is extracted from the data to be transmitted; the obfuscated result and the second part are input into a rich execution environment (REE); the REE is controlled to perform a linear operation according to a linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part to obtain an obfuscated linear operation result; the linear operation result is input into the TEE, and the TEE is controlled to determine the linear operation result before obfuscation based on the first obfuscation method, the first part, and the model parameter w; when the first model reasoning operation is completed, the semantic features obtained by the first model reasoning operation are sent to the target recipient, thereby efficiently performing the first model reasoning operation while keeping the model parameters and intermediate results in the reasoning process confidential by combining the TEE and the REE, effectively avoiding information leakage caused by semantic feature extraction on the data to be transmitted and improving the confidentiality of data transmission.
[0144] Figure 6This is a block diagram of an apparatus for a privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment, according to an exemplary embodiment. For example, apparatus 600 may be an electronic device, such as a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, or the like.
[0145] Reference Figure 6 , apparatus 600 may include one or more of the following components: a processing component 602 , a memory 604 , a power component 606 , a multimedia component 608 , an audio component 610 , an input / output (I / O) interface 612 , a sensor component 614 , and a communication component 616 .
[0146] The processing component 602 generally controls the overall operation of the device 600, such as operations associated with display, phone calls, data communications, camera operation, and recording operations. The processing component 602 may include one or more processors 620 to execute instructions to perform all or part of the steps of the above-described method. In addition, the processing component 602 may include one or more modules to facilitate interaction between the processing component 602 and other components. For example, the processing component 602 may include a multimedia module to facilitate interaction between the multimedia component 608 and the processing component 602.
[0147] The memory 604 is configured to store various types of data to support operations on the device 600. Examples of such data include instructions for any application or method operating on the device 600, contact data, phone book data, messages, pictures, videos, etc. The memory 604 can be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.
[0148] The power component 606 provides power to the various components of the device 600. The power component 606 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device 600.
[0149] The multimedia component 608 includes a screen that provides an output interface between the device 600 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor can not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 608 includes a front camera and / or a rear camera. When the device 600 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have a focal length and optical zoom capability.
[0150] The audio component 610 is configured to output and / or input audio signals. For example, the audio component 610 includes a microphone (MIC), which is configured to receive external audio signals when the device 600 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 604 or transmitted via the communication component 616. In some embodiments, the audio component 610 also includes a speaker for outputting audio signals.
[0151] I / O interface 612 provides an interface between processing component 602 and peripheral interface modules, such as a keyboard, click wheel, buttons, etc. These buttons may include but are not limited to: a home button, volume buttons, a start button, and a lock button.
[0152] The sensor assembly 614 includes one or more sensors for providing various aspects of the status assessment of the device 600. For example, the sensor assembly 614 can detect the open / closed state of the device 600, the relative positioning of components, such as the display and keypad of the device 600. The sensor assembly 614 can also detect changes in the position of the device 600 or a component of the device 600, the presence or absence of user contact with the device 600, the orientation or acceleration / deceleration of the device 600, and temperature changes of the device 600. The sensor assembly 614 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 614 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 614 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
[0153] The communication component 616 is configured to facilitate wired or wireless communication between the device 600 and other devices. The device 600 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 616 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 616 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
[0154] In an exemplary embodiment, the apparatus 600 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.
[0155] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 604 including instructions, which can be executed by the processor 620 of the apparatus 600 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0156] In an exemplary embodiment, a computer program product is also provided, comprising a computer program, which implements the above method when executed by the processor 620 of the apparatus 600 .
[0157] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the invention that follow from the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the following claims.
[0158] It should be understood that the present invention is not limited to the exact construction described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.
Claims
1. A privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment, characterized in that: include: Input the data to be transmitted into the trusted execution environment TEE; Performing a first model inference operation on the data to be transmitted using the TEE; In a case where the first model inference operation needs to determine a linear operation result on the intermediate data x and the model parameter w, performing obfuscation processing on the intermediate data x according to a first obfuscation method to obtain an obfuscated result; The intermediate data is extracted from the data to be transmitted; Inputting the obfuscated result into a rich execution environment (REE); controlling the REE to perform a linear operation according to a linear calculation type corresponding to the linear operation result based on the obfuscated result and the second part of the model parameter w, to obtain an obfuscated linear operation result; The model parameter w includes a first part and a second part, wherein the first part and the second part are obtained by disassembling the model parameter w in advance; Inputting the obfuscated linear operation result into the TEE, and controlling the TEE to determine the linear operation result before obfuscation based on the first obfuscation mode, the first part, and the model parameter w; When the first model reasoning operation is completed, the semantic features obtained by the first model reasoning operation are sent to a target recipient, so that the target recipient performs a second model reasoning operation based on the semantic features to obtain target data; The second model inference operation operates in the same manner as the first model inference operation.
2. The privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to claim 1, characterized in that: Obfuscating x according to the first obfuscation method to obtain an obfuscated result includes: Generate a random number n; The random number n is added to the intermediate data x to obtain the obfuscation result.
3. The privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to claim 2, characterized in that: The tensor shape of the random number n is the same as the tensor shape of the intermediate data x.
4. The privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to claim 1, characterized in that: When the first model reasoning operation is completed, sending the semantic features obtained by the first model reasoning operation to a target recipient includes: When the first model inference operation is completed, encrypting the semantic feature using a secret key; The encrypted semantic feature is sent to a target recipient, so that the target recipient uses the secret key to decrypt the encrypted semantic feature to obtain the semantic feature.
5. The privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to claim 1, characterized in that: When the first model reasoning operation is completed, sending the semantic features obtained by the first model reasoning operation to a target recipient includes: When the first model inference operation is completed, inputting the semantic feature into a channel encoder to obtain a wireless signal output by the channel encoder; Mapping the wireless signal to a constellation diagram to obtain coordinates of symbol points corresponding to the wireless signal in the constellation diagram, and using the coordinates of the symbol points as quantized wireless signals to be transmitted; The quantized wireless signal to be transmitted is sent to the target receiver.
6. The privacy-enhanced semantic communication method based on secure reasoning in a trusted execution environment according to claim 1, characterized in that: The first and second parts are obtained by decomposing the model parameter w through the following formula: w=a·b Among them, a is the first part and b is the second part.
7. A privacy-enhanced semantic communication device based on secure reasoning in a trusted execution environment, characterized in that: include: An input unit, used to input data to be transmitted into the trusted execution environment TEE; a model inference unit, configured to perform a first model inference operation on the data to be transmitted using the TEE; an obfuscation unit configured to, when the first model inference operation requires determining a linear operation result on the intermediate data x and the model parameter w, perform obfuscation processing on the intermediate data x according to a first obfuscation method to obtain an obfuscated result; the intermediate data is extracted from the data to be transmitted; An input unit, further configured to input the obfuscation result into a rich execution environment (REE); a linear operation unit, configured to control the REE to perform a linear operation based on the obfuscated result and the second part of the model parameter w according to a linear calculation type corresponding to the linear operation result, to obtain an obfuscated linear operation result; The model parameter w includes a first part and a second part, wherein the first part and the second part are obtained by disassembling the model parameter w in advance; a determining unit, configured to input the obfuscated linear operation result into the TEE, and control the TEE to determine the linear operation result before obfuscation based on the first obfuscation mode, the first part, and the model parameter w; a sending unit, configured to send the semantic features obtained by the first model reasoning operation to a target recipient when the first model reasoning operation is completed, so that the target recipient performs a second model reasoning operation based on the semantic features to obtain target data; The second model inference operation operates in the same manner as the first model inference operation.
8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.
10. A computer program product comprising a computer program, characterized in that The computer program implements the method according to any one of claims 1 to 6 when executed by a processor.