Wireless terminal encryption authentication method and system based on hearing and tactile interaction

Abstract

The present invention provides a wireless terminal encryption authentication method and system based on auditory and tactile interaction, which relates to the field of encryption authentication technology, including collecting user's voice instructions through a microphone array, obtaining compact feature representation, generating keyword vectors, extracting time-frequency energy distribution features, and forming a seed key vector; collecting continuous touch trajectory data through a flexible tactile sensor array, constructing a six-dimensional touch gesture feature manifold, performing spatiotemporal sequence modeling based on a graph attention network, extracting the spatiotemporal evolution pattern of the gesture manifold, generating a tactile key vector based on a manifold arrangement learning algorithm, and fusing to generate a composite key vector; generating an authentication request data packet, sending it to an authentication server through an authentication channel, demodulating it through continuous quantum state detection, and extracting the composite key vector; evaluating cosine similarity respectively through deep metric learning, determining authentication, and sending authentication credentials, which are stored in a blockchain account book.

Description

Technical Field

[0001] The present invention relates to the technical field of encryption authentication, and in particular to a wireless terminal encryption authentication method and system based on auditory and tactile interaction. Background Art

[0002] With the rapid development of wireless communication technology and the widespread popularity of smart mobile devices, wireless terminals have become an indispensable tool in people's daily life and work. However, the open wireless network environment where wireless terminals are located faces many security threats, such as illegal access, identity spoofing, information theft, etc. In order to ensure the information security and privacy protection of wireless terminal users, a secure and reliable wireless terminal authentication solution is urgently needed.

[0003] Among the traditional wireless terminal authentication methods, password authentication is easy to implement, but there is a risk that the password may be guessed, forgotten or stolen. SMS verification code authentication confirms the identity by sending a one-time verification code to the user's mobile phone, but the verification code may be intercepted or forwarded, and the user needs to enter the verification code manually, which leads to a poor user experience. Single biometric authentication, such as fingerprint recognition and face recognition, is convenient and fast, but the biometric feature may be copied and forged, and once the biometric feature is leaked, it cannot be changed, and security is challenged. With the continuous advancement of artificial intelligence and machine learning technologies, continuous authentication schemes based on behavioral characteristics are gradually emerging. Behavioral characteristics reflect the dynamic characteristics of the user's interaction with the device. Compared with biometric features, behavioral characteristics have stronger time-varying and randomness, and are difficult to imitate and copy.

[0004] To sum up, the wireless terminal authentication method in the prior art has problems such as insufficient security, limited authentication accuracy, and poor user experience. By utilizing the dynamic and random nature of behavioral characteristics, a safe, reliable, and easy-to-use wireless terminal encryption authentication method is provided. Therefore, the present invention can solve the problems in the prior art. Summary of the invention

[0005] The embodiments of the present invention provide a wireless terminal encryption authentication method and system based on auditory and tactile interaction, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provided is a wireless terminal encryption authentication method based on auditory and tactile interaction, comprising:

[0008] The wireless terminal collects the user's voice command through a microphone array, extracts features of the voice command, adaptively weights the voice feature sequence through the set attention pooling layer, obtains a compact feature representation of the voice command, learns the probability density distribution of the voice command through variational inference based on the compact feature representation, generates a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, extracts the time-frequency energy distribution features of the keyword vector through multi-scale adaptive signal decomposition, and forms a seed key vector;

[0009] On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array to construct a six-dimensional touch gesture feature manifold, and the six-dimensional touch gesture feature manifold is modeled in time and space sequence based on a graph attention network. The time and space evolution pattern of the gesture manifold is extracted through a graph attention layer and a temporal convolution layer. Based on a manifold permutation learning algorithm, a tactile key vector consistent with the dimension of the seed key vector is generated, and the tactile key vector is fused with the seed key vector to generate a composite key vector;

[0010] Based on the composite key vector, the wireless terminal generates an authentication request data packet through a torus lattice cryptographic mechanism, and sends the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol. The authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation.

[0011] The composite key vector is split into a seed key vector and a tactile key vector, and the cosine similarities with the corresponding vectors in the standard auditory-tactile interaction model library pre-constructed on the authentication server are evaluated respectively through deep metric learning. If the cosine similarities exceed the preset similarity threshold, the authentication is judged to be successful, and the post-quantum blind signature scheme is used to send the authentication credential to the wireless terminal, and the authentication credential is stored in a threshold signature blockchain account book based on lattice cryptography to complete the wireless terminal encryption authentication process.

[0012] In an optional embodiment,

[0013] The wireless terminal collects the user's voice command through a microphone array, extracts features of the voice command, adaptively weights the voice feature sequence through the set attention pooling layer, obtains a compact feature representation of the voice command, learns the probability density distribution of the voice command through variational inference based on the compact feature representation, generates a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, extracts the time-frequency energy distribution features of the keyword vector through multi-scale adaptive signal decomposition, and forms a seed key vector including:

[0014] The wireless terminal collects the user's voice commands through a microphone array, divides the voice commands into frames, performs Fourier transform on each frame, determines the spectrum information, extracts filter energy through a scale filter group, performs weighted processing on the filter energy according to an equal loudness curve, combines logarithmic compression, and performs feature extraction through a preset linear prediction analysis method to obtain an initial feature sequence;

[0015] Learning the importance weights of the initial feature sequence at different time steps through an attention pooling layer, and performing weighted summation on the initial feature sequence based on the importance weights to obtain a compact feature representation of the voice command;

[0016] The compact feature representation is regarded as the output of the latent variable model, and the variational inference model is determined with the goal of minimizing the KL divergence between the variational distribution and the true posterior distribution. The latent variables are sampled from the prior distribution through the stochastic gradient variational Bayes algorithm, and the compact feature representation is reconstructed and output through the decoder. After iteratively updating the decoder parameters, the solution that is most similar to the true posterior distribution is obtained, and the final probability density distribution of the voice command is determined;

[0017] Based on the final probability density distribution, mapping the latent variables into keyword vectors through a decoder, and selecting the keyword vector most relevant to the voice command by minimizing the reconstruction error between the keyword vector and the compact feature representation;

[0018] Performing empirical mode decomposition on the keyword vector to generate eigenmode function components, performing Hilbert transform on the eigenmode function components to obtain analytical signals, and calculating instantaneous frequency and instantaneous amplitude of the eigenmode function components based on the analytical signals, and combining them to generate time-frequency energy distribution features;

[0019] The time-frequency energy distribution feature is quantized and encoded to obtain a quantized coding sequence, and the quantized coding sequence is hash mapped using a secure hash algorithm to obtain a hash value of a fixed length as a seed key vector.

[0020] In an optional embodiment,

[0021] On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array to construct a six-dimensional touch gesture feature manifold, and the six-dimensional touch gesture feature manifold is modeled in time and space sequence based on a graph attention network. The time and space evolution pattern of the gesture manifold is extracted through a graph attention layer and a time convolution layer. Based on a manifold permutation learning algorithm, a tactile key vector consistent with the dimension of the seed key vector is generated, including:

[0022] On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array, wherein the continuous touch trajectory data includes a touch coordinate sequence, a touch force sequence, a normal touch force component, a tangential touch force component, and a fluid impedance sequence;

[0023] Based on the continuous touch trajectory data, each touch data point in the continuous touch trajectory data is regarded as a sample point in the six-dimensional touch gesture feature manifold, and a local linear embedding algorithm is used to perform nonlinear dimensionality reduction on the sample point, and the six-dimensional touch gesture data is mapped to a low-dimensional manifold space by maintaining the local linear structure and reducing the dimension of the global nonlinear structure, so as to obtain a six-dimensional touch gesture feature manifold;

[0024] A graph attention network is constructed to perform spatiotemporal sequence modeling on the six-dimensional touch gesture feature manifold, wherein the graph attention network is formed by alternately stacking a plurality of graph attention layers and a plurality of temporal convolution layers, wherein in each graph attention layer, the neighborhood information of each sample point in the six-dimensional touch gesture feature manifold is adaptively aggregated through an attention mechanism, the dynamic association weights between the sample points are learned, the feature representation of the sample points is updated, and the spatial structure information is determined; in each temporal convolution layer, a causal one-dimensional convolution operation is used to extract the temporal dynamic features of the six-dimensional touch gesture feature manifold at different time scales, and the temporal evolution information is determined;

[0025] Based on the spatial structure information and the time evolution information, a spatiotemporal joint feature matrix is ​​determined. Based on the spatiotemporal joint feature matrix, a dimensionality transformation is performed through a manifold permutation learning algorithm to generate a tactile key vector that is consistent with the dimension of the seed key vector.

[0026] In an optional embodiment,

[0027] The manifold permutation learning algorithm includes:

[0028] Performing row normalization processing on the spatiotemporal joint feature matrix to obtain a normalized feature matrix;

[0029] Based on the normalized feature matrix, the eigenvalues ​​corresponding to the eigenvectors in the normalized feature matrix are calculated by eigenvalue decomposition, and the eigenvectors corresponding to the largest d eigenvalues ​​are taken according to the preset number of features d to construct a main eigenvector matrix;

[0030] Based on the main eigenvector matrix, the normalized feature matrix is ​​projected, and each sample point in the normalized feature matrix is ​​projected from the feature space of the original dimension to the low-dimensional subspace spanned by the main eigenvector to obtain a reduced-dimensional feature matrix;

[0031] Perform row normalization on the reduced dimension feature matrix to obtain a normalized reduced dimension feature matrix;

[0032] Based on the normalized reduced-dimensionality feature matrix, the tactile key vector is obtained through manifold permutation learning, including: randomly generating an initial permutation vector consistent with the dimension of the seed key vector, optimizing the permutation loss function, taking maximizing the inner product between the initial permutation vector and each sample point in the normalized reduced-dimensionality feature matrix as the goal, iteratively generating the optimal permutation vector, and determining the tactile key vector.

[0033] In an optional embodiment,

[0034] Based on the composite key vector, the wireless terminal generates an authentication request data packet through a torus lattice cryptographic mechanism, sends the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol, the authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation, including:

[0035] Based on the torus lattice cryptographic mechanism, the composite key vector is mapped to a key point on the torus lattice, and a mask point on the torus lattice is randomly generated, the key point and the mask point are added on the torus lattice to obtain a masked key point, the masked key point is converted into a binary code sequence to constitute the payload part of the authentication request data packet, and control information is added to the header of the authentication request data packet to construct the authentication request data packet;

[0036] The quantum secure key negotiation protocol includes a quantum key distribution protocol and a quantum secure communication protocol. The wireless terminal and the authentication server generate a shared security key based on the quantum key distribution protocol, and use the shared security key to establish a quantum secure authentication channel through the quantum secure communication protocol.

[0037] Based on the shared security key, encrypt the authentication request data packet to obtain a ciphertext of the authentication request data packet, modulate it into a continuous quantum state signal, and send it to the authentication server through the quantum secure authentication channel;

[0038] The authentication server demodulates the received continuous quantum state signal using coherent demodulation technology, obtains a demodulated signal in complex form by synchronously demodulating the orthogonal amplitude and phase components of the continuous quantum state signal, performs analog-to-digital conversion on the demodulated signal, converts the complex form into a real form, determines a level signal, performs a binary decision on the level signal according to a preset decision threshold, and obtains a binary symbol sequence in combination with symbol mapping;

[0039] Based on the preset error correction code, correct the channel transmission error, decode the binary symbol sequence to obtain binary data, perform format parsing and data extraction on the binary data based on the preset format of the authentication request data packet, extract the ciphertext of the authentication request data packet, and decrypt it based on the shared security key through a preset symmetric key algorithm to obtain the authentication request data packet;

[0040] The authentication server extracts the composite key vector from the authentication request data packet through the reverse operation, extracts the binary code sequence from the payload part of the authentication request data packet and converts it into the mask key point on the torus lattice, regenerates the mask point using the pre-negotiated mask point generation algorithm, performs a subtraction operation on the torus lattice to obtain the key point on the torus lattice, and maps the key point back to the real number space to obtain the composite key vector.

[0041] In an optional embodiment,

[0042] The composite key vector is split into a seed key vector and a tactile key vector, and the cosine similarity with the corresponding vectors in the standard auditory tactile interaction model library pre-built on the authentication server is evaluated by deep metric learning. If the cosine similarities exceed the preset similarity threshold, the authentication is determined to be passed, including:

[0043] Construct a twin network containing two sub-networks with the same structure, each of which consists of multiple convolutional layers, pooling layers, fully connected layers and an embedding layer, wherein the embedding layer maps the output of the fully connected layer to a low-dimensional embedding space, and introduces an attention mechanism. An attention module is set after each convolutional layer, and by generating an attention weight matrix, the attention to different areas in the convolutional feature map is adaptively adjusted to extract key features;

[0044] The twin network is trained, and based on the preset category labels of the training samples, the category samples are determined, and the category clustering is constructed by minimizing the same-category embedding distance between the same-category sample pairs and maximizing the heterogeneous embedding distance between the heterogeneous category sample pairs. The center loss function is combined as an auxiliary loss, and the intra-class aggregation and inter-class separation are maximized by minimizing the distance between each category sample and the corresponding category clustering center to generate a discriminative embedding space;

[0045] Based on a standard auditory-tactile interaction model library, determine a standard seed key vector and a standard tactile key vector, wherein the seed key vector and the standard seed key vector constitute a first vector comparison pair, and the tactile key vector and the standard tactile key vector constitute a second vector comparison pair;

[0046] Inputting the first vector comparison pair and the second vector comparison pair into the sub-networks in the twin network respectively, obtaining the first feature representation vector pair and the second feature representation vector pair in the discriminant embedding space respectively, calculating the cosine value of the angle respectively by the cosine similarity function, and determining the first cosine similarity and the second cosine similarity;

[0047] The first cosine similarity and the second cosine similarity are weightedly fused to obtain a comprehensive similarity, and the comprehensive similarity is compared with a preset similarity threshold. When the comprehensive similarity exceeds the threshold, it is determined that the authentication is passed; otherwise, it is determined that the authentication fails.

[0048] According to a second aspect of the embodiments of the present invention,

[0049] Provided is a wireless terminal encryption authentication system based on auditory and tactile interaction, comprising:

[0050] The first unit is used for the wireless terminal to collect the user's voice command through the microphone array, perform feature extraction on the voice command, obtain a compact feature representation of the voice command by adaptively weighting the voice feature sequence through the set attention pooling layer, learn the probability density distribution of the voice command through variational inference based on the compact feature representation, generate a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, and extract the time-frequency energy distribution characteristics of the keyword vector through multi-scale adaptive signal decomposition to form a seed key vector;

[0051] The second unit is used to collect continuous touch trajectory data on the curved touch interface of the wireless terminal through a flexible tactile sensor array, construct a six-dimensional touch gesture feature manifold, perform spatiotemporal sequence modeling on the six-dimensional touch gesture feature manifold based on a graph attention network, extract the spatiotemporal evolution pattern of the gesture manifold through a graph attention layer and a temporal convolution layer, generate a tactile key vector consistent with the dimension of the seed key vector based on a manifold permutation learning algorithm, and fuse it with the seed key vector to generate a composite key vector;

[0052] A third unit is configured to generate an authentication request data packet by the wireless terminal through a torus lattice cryptographic mechanism based on the composite key vector, and send the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol, wherein the authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation;

[0053] The fourth unit is used to split the composite key vector into a seed key vector and a tactile key vector, and evaluate the cosine similarity of the corresponding vectors in the standard auditory-tactile interaction model library pre-constructed on the authentication server through deep metric learning. If the cosine similarity exceeds the preset similarity threshold, the authentication is determined to be successful, and the post-quantum blind signature scheme is used to send the authentication certificate to the wireless terminal, and the authentication certificate is stored in a threshold signature blockchain account book based on lattice cryptography to complete the wireless terminal encryption authentication process.

[0054] According to a third aspect of the embodiments of the present invention,

[0055] An electronic device is provided, comprising:

[0056] processor;

[0057] a memory for storing processor-executable instructions;

[0058] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0059] A fourth aspect of the embodiments of the present invention is:

[0060] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0061] In an embodiment of the present invention, voice commands are collected through a microphone array, and feature extraction and adaptive weighting are performed on them, so that voice features can be obtained more accurately, thereby improving the accuracy of voice recognition; variational inference is used to generate keyword vectors that are most semantically relevant to voice commands, so that the system can more accurately capture user intentions and improve the accuracy of keyword extraction; time-frequency energy distribution features of keyword vectors are extracted through multi-scale adaptive signal decomposition, which can better cope with voice signals in complex environments and enhance the robustness of the system under different background noises and voice changes; touch gesture features are modeled through spatiotemporal sequences through a graph attention network, and the spatiotemporal evolution pattern of gestures is extracted, which can more accurately capture the user's personalized gesture features, provide a more accurate basis for identity authentication, and reduce the risk of misauthentication; touch gesture feature manifolds are processed through a manifold permutation learning algorithm, which can adapt to the touch habits and changes of different users and improve the system's robustness to different The adaptability to users and the environment makes the key generation process more flexible and robust; by generating authentication request packets based on the torus lattice cryptographic mechanism and combining the authentication channel established by the quantum secure key negotiation protocol, high-strength encryption that is resistant to quantum computing attacks is provided, ensuring the confidentiality and anti-eavesdropping capabilities of data during transmission; by generating and extracting composite key vectors through reverse operations, the integrity and uniqueness of the key throughout the authentication process are guaranteed to prevent the key from being tampered with or forged; the authentication server accurately demodulates the authentication request packet through continuous quantum state detection to ensure accurate verification of the key during the authentication process; by evaluating the cosine similarity between the composite key vector and the standard model through deep metric learning, it ensures that the authentication process can accurately distinguish between legitimate users and potential attackers; combined with the post-quantum blind signature scheme and the threshold signature blockchain ledger based on lattice cryptography, it provides strong security protection against quantum computing attacks to prevent the authentication credentials from being forged or tampered with. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of a flow chart of a wireless terminal encryption authentication method based on auditory and tactile interaction according to an embodiment of the present invention;

[0063] Figure 2 The present invention is a schematic diagram of the structure of a wireless terminal encryption authentication system based on auditory and tactile interaction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 are within the scope of protection of the present invention.

[0065] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0066] Figure 1 FIG. 1 is a flow chart of a wireless terminal encryption authentication method based on auditory and tactile interaction according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0067] S101. The wireless terminal collects the user's voice command through a microphone array, extracts features of the voice command, adaptively weights the voice feature sequence through the set attention pooling layer, obtains a compact feature representation of the voice command, learns the probability density distribution of the voice command through variational inference based on the compact feature representation, generates a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, extracts the time-frequency energy distribution features of the keyword vector through multi-scale adaptive signal decomposition, and forms a seed key vector;

[0068] In this embodiment, by collecting voice commands through a microphone array, performing feature extraction and adaptive weighting on them, voice features can be obtained more accurately, thereby improving the accuracy of voice recognition; variational inference is used to generate keyword vectors that are most semantically relevant to the voice command, so that the system can more accurately capture user intentions and improve the accuracy of keyword extraction; time-frequency energy distribution features of keyword vectors are extracted through multi-scale adaptive signal decomposition, which can better cope with voice signals in complex environments and enhance the robustness of the system under different background noises and voice changes; by generating a seed key vector, based on the time-frequency energy distribution features of voice features, a more secure foundation can be provided for subsequent voice encryption or identity authentication, thereby enhancing the security and anti-attack capabilities of the system.

[0069] In an optional embodiment, the wireless terminal collects the user's voice command through a microphone array, performs feature extraction on the voice command, adaptively weights the voice feature sequence through the set attention pooling layer, obtains a compact feature representation of the voice command, learns the probability density distribution of the voice command through variational inference based on the compact feature representation, generates a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, extracts the time-frequency energy distribution characteristics of the keyword vector through multi-scale adaptive signal decomposition, and forms a seed key vector including:

[0070] The wireless terminal collects the user's voice commands through a microphone array, divides the voice commands into frames, performs Fourier transform on each frame, determines the spectrum information, extracts filter energy through a scale filter group, performs weighted processing on the filter energy according to an equal loudness curve, combines logarithmic compression, and performs feature extraction through a preset linear prediction analysis method to obtain an initial feature sequence;

[0071] Learning the importance weights of the initial feature sequence at different time steps through an attention pooling layer, and performing weighted summation on the initial feature sequence based on the importance weights to obtain a compact feature representation of the voice command;

[0072] The compact feature representation is regarded as the output of the latent variable model, and the variational inference model is determined with the goal of minimizing the KL divergence between the variational distribution and the true posterior distribution. The latent variables are sampled from the prior distribution through the stochastic gradient variational Bayes algorithm, and the compact feature representation is reconstructed and output through the decoder. After iteratively updating the decoder parameters, the solution that is most similar to the true posterior distribution is obtained, and the final probability density distribution of the voice command is determined;

[0073] Based on the final probability density distribution, mapping the latent variables into keyword vectors through a decoder, and selecting the keyword vector most relevant to the voice command by minimizing the reconstruction error between the keyword vector and the compact feature representation;

[0074] Performing empirical mode decomposition on the keyword vector to generate eigenmode function components, performing Hilbert transform on the eigenmode function components to obtain analytical signals, and calculating instantaneous frequency and instantaneous amplitude of the eigenmode function components based on the analytical signals, and combining them to generate time-frequency energy distribution features;

[0075] The time-frequency energy distribution feature is quantized and encoded to obtain a quantized coding sequence, and the quantized coding sequence is hash mapped using a secure hash algorithm to obtain a hash value of a fixed length as a seed key vector.

[0076] The scale filter group is a group of filters with different scales or frequency bands, which can separate and analyze different frequency components of the input signal. By applying these filters, the energy information of the signal in different frequency ranges can be extracted, thereby obtaining rich spectral features. The scale filter group is widely used in the fields of speech processing, image processing, etc. for multi-scale analysis.

[0077] The equal loudness curves describe the sound pressure levels required for different frequencies to be perceived as the same loudness in human ear perception. These curves reflect the sensitivity of the human ear to different frequencies, which is usually most sensitive in the mid-frequency area, about 1kHz to 5kHz, and less sensitive in the low-frequency and high-frequency areas. In the speech processing process, the equal loudness curves are used for weighted processing to simulate the human ear's perception characteristics of sounds of different frequencies, so as to more accurately extract speech features that are consistent with human auditory perception.

[0078] The latent variables are variables that are not directly observed or measured in the model, but they usually affect the observed data. The latent variable model assumes that the observed data is generated by the latent variables, and by inferring the latent variables, we can better understand the generation mechanism of the data. In speech recognition, latent variables can represent the potential structure of speech features and capture the complex patterns and relationships behind the observed data.

[0079] The wireless terminal collects the user's voice commands through a microphone array. The collected voice command signals are divided into frames, and the length of each frame is generally 10-30 milliseconds. Each frame of voice data is fast Fourier transformed to convert the time domain signal into a frequency domain signal to obtain the spectrum information of the corresponding frame.

[0080] Filter energy extraction and weighted processing: Input the spectrum information into a set of scale filters, each of which corresponds to a different frequency sub-band. Calculate the spectrum energy in each sub-band to obtain a set of filter energy values. According to the auditory characteristics of the human ear, the filter energy values ​​are weighted with equal loudness to highlight the frequency components that match the sensitivity of the human ear.

[0081] Logarithmic compression and linear prediction analysis: logarithmic compression is performed on the weighted filter energy value to reduce the dynamic range of the energy value, making it more consistent with the perception characteristics of the human ear. Then, the linear prediction analysis method is used to extract the linear prediction coefficient reflecting the pronunciation characteristics of the speech by estimating the linear combination relationship between the current speech frame and several past speech frames. The linear prediction coefficient is used as the initial feature sequence of the speech instruction.

[0082] Attention pooling and compact feature representation, use the attention pooling layer to process the initial feature sequence. By learning the importance weight of the feature at each time step, the contribution of the features at different times is adaptively adjusted. According to the importance weight, the initial feature sequence is weighted and summed to obtain a compact feature vector as a high-level representation of the voice command.

[0083] Variational inference and probability density estimation, treats the compact feature representation as the output of the latent variable model, and aims to estimate the posterior probability distribution of the latent variables. The variational inference method is used to find the optimal variational distribution parameters by minimizing the KL divergence between the variational distribution and the true posterior distribution. The stochastic gradient variational Bayes algorithm is used to sample the latent variables from the prior distribution and reconstruct the compact feature representation through the decoder. The decoder parameters are iteratively optimized to minimize the reconstruction error and obtain the variational solution that is closest to the true posterior distribution.

[0084] Using the trained variational decoder, the sampled latent variables are mapped to the keyword vector space. By minimizing the reconstruction error between the keyword vector and the compact feature representation, the keyword vector that can restore the original voice command information to the greatest extent is found. A set of keyword vectors that are most relevant to the voice command is selected.

[0085] Empirical mode decomposition and Hilbert transform, perform empirical mode decomposition on the selected keyword vectors and decompose them into several intrinsic mode function components. Perform Hilbert transform on each intrinsic mode function component to obtain the corresponding analytical signal. The analytical signal consists of the real part (the original eigenmode function component) and the imaginary part (the Hilbert transform of the eigenmode function component), which can reflect the instantaneous characteristics of the signal.

[0086] Instantaneous feature extraction and combination, based on the analytical signal, calculate the instantaneous frequency and instantaneous amplitude of each eigenmode function component. The instantaneous frequency reflects the main frequency components of the signal at different times, and the instantaneous amplitude reflects the energy intensity of the signal at different times. The instantaneous frequency and instantaneous amplitude are combined to generate a two-dimensional time-frequency energy distribution feature matrix.

[0087] Quantization coding and hash mapping, quantization coding of the time-frequency energy distribution feature matrix, discretization of continuous energy values ​​into several coding symbols, and obtaining a quantization coding sequence. A secure hash algorithm, preferably SHA-256, is used to perform hash mapping on the quantization coding sequence and convert it into a hash value of fixed length. The hash value has high randomness and irreversibility and can be used as the final seed key vector.

[0088] In this embodiment, the spectrum information of the speech signal can be accurately captured through Fourier transform, scale filter group and equal loudness curve processing, and weighted processing is performed according to human ear perception, so as to extract more accurate and human auditory characteristics of speech features; the feature sequence is weighted and summed through the attention pooling layer to further extract key features and obtain compact feature representation. Compact feature representation can reduce redundant information and improve the robustness of the speech recognition system in a noisy environment; the final probability density distribution of the voice command is determined by the variational inference method, and the keyword vector is generated by the hidden variable model, so that the keyword extraction is more accurate and relevant, and can better reflect the user's intention; by quantizing and encoding the time-frequency energy distribution characteristics of the keyword vector, and combining the secure hash algorithm to generate a seed key vector, the vector can be used for subsequent encryption or identity authentication, thereby improving the security of the system. The key generated by this process is unique and secure, and effectively prevents being cracked or forged; by iteratively updating the decoder parameters, the variational inference model can gradually approach the true posterior distribution, and by sampling the hidden variables through the stochastic gradient variational Bayes algorithm, the computational complexity can be reduced while maintaining the model accuracy, and the real-time performance of the system can be improved.

[0089] S102. On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array to construct a six-dimensional touch gesture feature manifold, and spatiotemporal sequence modeling is performed on the six-dimensional touch gesture feature manifold based on a graph attention network. The spatiotemporal evolution pattern of the gesture manifold is extracted through a graph attention layer and a temporal convolution layer. Based on a manifold permutation learning algorithm, a tactile key vector consistent with the dimension of the seed key vector is generated, and the tactile key vector is fused with the seed key vector to generate a composite key vector;

[0090] In this embodiment, by collecting the touch trajectory data of the user on the curved touch interface of the wireless terminal, and combining the tactile gesture features to construct a six-dimensional touch gesture feature manifold, a unique tactile key vector is generated, and a composite key vector is generated, so that the key is not only based on the voice feature, but also incorporates the touch behavior feature, improves the uniqueness and security of the key, and prevents the key from being copied or cracked; the touch gesture is a unique behavioral feature of the user and is difficult to be imitated by others. Through the graph attention network, the spatiotemporal sequence modeling of the touch gesture features is performed, and the spatiotemporal evolution pattern of the gesture is extracted, which can more accurately capture the personalized gesture features of the user, provide a more accurate basis for identity authentication, and reduce the risk of misauthentication; the touch gesture feature manifold is processed by the manifold arrangement learning algorithm, which can adapt to the touch habits and changes of different users, improve the system's adaptability to different users and environments, and make the key generation process more flexible and robust; by fusing the seed key vector generated by the touch gesture and the voice command, the generated composite key vector integrates multimodal information, so that the system can provide stronger resistance when facing various attack methods, and further improve the overall security.

[0091] In an optional embodiment, on a curved touch interface of a wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array to construct a six-dimensional touch gesture feature manifold, and spatiotemporal sequence modeling is performed on the six-dimensional touch gesture feature manifold based on a graph attention network. The spatiotemporal evolution pattern of the gesture manifold is extracted through a graph attention layer and a temporal convolution layer. Based on a manifold permutation learning algorithm, a tactile key vector consistent with the dimension of the seed key vector is generated, including:

[0092] On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array, wherein the continuous touch trajectory data includes a touch coordinate sequence, a touch force sequence, a normal touch force component, a tangential touch force component, and a fluid impedance sequence;

[0093] Based on the continuous touch trajectory data, each touch data point in the continuous touch trajectory data is regarded as a sample point in the six-dimensional touch gesture feature manifold, and a local linear embedding algorithm is used to perform nonlinear dimensionality reduction on the sample point, and the six-dimensional touch gesture data is mapped to a low-dimensional manifold space by maintaining the local linear structure and reducing the dimension of the global nonlinear structure, so as to obtain a six-dimensional touch gesture feature manifold;

[0094] A graph attention network is constructed to perform spatiotemporal sequence modeling on the six-dimensional touch gesture feature manifold, wherein the graph attention network is formed by alternately stacking a plurality of graph attention layers and a plurality of temporal convolution layers, wherein in each graph attention layer, the neighborhood information of each sample point in the six-dimensional touch gesture feature manifold is adaptively aggregated through an attention mechanism, the dynamic association weights between the sample points are learned, the feature representation of the sample points is updated, and the spatial structure information is determined; in each temporal convolution layer, a causal one-dimensional convolution operation is used to extract the temporal dynamic features of the six-dimensional touch gesture feature manifold at different time scales, and the temporal evolution information is determined;

[0095] Based on the spatial structure information and the time evolution information, a spatiotemporal joint feature matrix is ​​determined. Based on the spatiotemporal joint feature matrix, a dimensionality transformation is performed through a manifold permutation learning algorithm to generate a tactile key vector that is consistent with the dimension of the seed key vector.

[0096] The touch coordinate sequence refers to the trajectory of the position change of the finger when the user continuously touches the curved touch interface, which is usually composed of a series of coordinate points, recording the changes in the position of the touch point on the interface over time.

[0097] The touch force sequence specifically refers to the change in pressure applied by the user on the interface during the touch process, and records the change in the magnitude of the touch force applied by the user as the touch time passes, reflecting the strength of the user's touch action.

[0098] The normal touch force component specifically refers to the touch force component in a direction perpendicular to the touch interface, and generally reflects the change in force when the user presses or releases the touch interface.

[0099] The tangential touch force component specifically refers to a touch force component in a direction parallel to the touch interface, and generally reflects a change in force when a user slides, drags or rubs the touch interface.

[0100] The fluid impedance sequence is usually related to the fluid mechanics characteristics in the flexible touch interface or tactile sensor, and records the resistance or impedance changes caused by the fluid medium (such as liquid in the flexible material) when the user touches. The fluid impedance sequence can reflect the dynamic resistance changes of the touch action.

[0101] An array of multiple flexible tactile sensors is arranged on the curved touch interface of the wireless terminal. When the user performs continuous touch operations on the touch interface, the tactile sensor array collects a series of touch data points in real time. Each touch data point contains the coordinate information of the touch position, the touch force, the components of the touch force in the normal and tangential directions, and the fluid impedance value at the touch point. The collected touch data points are arranged in chronological order to form a continuous touch trajectory data sequence.

[0102] Each touch data point in the continuous touch trajectory data is regarded as a sample point in a six-dimensional space. The six dimensions correspond to the touch coordinates (x, y), touch force magnitude, normal touch force component, tangential touch force component and fluid impedance value, respectively. Due to the high-dimensional characteristics of the six-dimensional space, the computational complexity of directly processing the sample points is high. Therefore, a local linear embedding algorithm is used to perform nonlinear dimensionality reduction on the sample points. The local linear embedding algorithm maps the sample points in the six-dimensional space to a low-dimensional manifold space by maintaining the linear relationship of the sample points in the local neighborhood while reducing the global nonlinear structure. The mapped low-dimensional manifold can better maintain the intrinsic structure and characteristics of the original data.

[0103] In order to model the spatiotemporal sequence of the six-dimensional touch gesture feature manifold, a graph attention network consisting of multiple graph attention layers and temporal convolution layers is constructed. In each graph attention layer, the association weight between each sample point in the manifold space and other sample points in its neighborhood is calculated through the attention mechanism. The association weight reflects the importance of different sample points to the feature representation of the current sample point. According to the association weight, the features of the sample points in the neighborhood are weighted and aggregated to update the feature representation of the current sample point. By stacking multiple graph attention layers, the network can adaptively learn the dynamic association relationship between sample points and capture the spatial structure information of the manifold. In each temporal convolution layer, a causal one-dimensional convolution operation is used to convolve the feature sequence of the sample points along the time axis. By setting different convolution kernel sizes, the network can extract the temporal dynamic features of the manifold at different time scales and characterize the temporal evolution information of the manifold.

[0104] After being processed by the graph attention network, two types of features are obtained to characterize the spatial structure and temporal evolution of the six-dimensional touch gesture feature manifold: spatial structure information and temporal evolution information. These two types of features are combined in the order of sample points to generate a spatiotemporal joint feature matrix. Each row of the matrix corresponds to a sample point, and each column corresponds to a spatiotemporal feature dimension. The spatiotemporal joint feature matrix integrates the feature information of touch gestures in both spatial and temporal dimensions, and can more comprehensively characterize the characteristics of touch behavior.

[0105] In order to transform the spatiotemporal joint feature matrix into a tactile key vector with the same dimension as the seed key vector, the manifold permutation learning algorithm is used for dimensional transformation. The manifold permutation learning algorithm rearranges the rows (sample points) in the spatiotemporal joint feature matrix by optimizing a permutation matrix, so that similar sample points are clustered together in space in the permuted matrix. At the same time, by setting the target dimension, the permuted matrix is ​​reduced in dimension to obtain a low-dimensional feature matrix with the same dimension as the seed key vector. The elements in the low-dimensional feature matrix are extracted in a certain order and recombined into a vector to obtain the final tactile key vector. The tactile key vector inherits the information of the spatiotemporal joint feature matrix and can effectively characterize the user's continuous touch behavior characteristics.

[0106] In this embodiment, by collecting multi-dimensional touch data, including touch position, strength, directional component of force and fluid impedance, the touch behavior is fully described to ensure that feature extraction is more accurate and detailed; the graph attention network is used to perform spatiotemporal sequence modeling of touch gestures, which can adaptively aggregate the neighborhood information of sample points, extract complex spatiotemporal dynamic features, and improve the system's understanding and analysis capabilities of user touch behavior; through dimensionality reduction and manifold permutation learning algorithms, high-dimensional touch data is mapped to low-dimensional space, and a tactile key vector matching the seed key vector is generated. The key generated by combining multi-dimensional touch features and spatiotemporal dynamic information is more unique and secure; through precise spatiotemporal joint feature matrix modeling, the system can better adapt to the touch habits of different users and environmental changes, and improve the stability of touch recognition and key generation.

[0107] In an optional embodiment, the manifold arrangement learning algorithm includes:

[0108] Performing row normalization processing on the spatiotemporal joint feature matrix to obtain a normalized feature matrix;

[0109] Based on the normalized feature matrix, the eigenvalues ​​corresponding to the eigenvectors in the normalized feature matrix are calculated by eigenvalue decomposition, and the eigenvectors corresponding to the largest d eigenvalues ​​are taken according to the preset number of features d to construct a main eigenvector matrix;

[0110] Based on the main eigenvector matrix, the normalized feature matrix is ​​projected, and each sample point in the normalized feature matrix is ​​projected from the feature space of the original dimension to the low-dimensional subspace spanned by the main eigenvector to obtain a reduced-dimensional feature matrix;

[0111] Perform row normalization on the reduced dimension feature matrix to obtain a normalized reduced dimension feature matrix;

[0112] Based on the normalized reduced-dimensionality feature matrix, the tactile key vector is obtained through manifold permutation learning, including: randomly generating an initial permutation vector consistent with the dimension of the seed key vector, optimizing the permutation loss function, taking maximizing the inner product between the initial permutation vector and each sample point in the normalized reduced-dimensionality feature matrix as the goal, iteratively generating the optimal permutation vector, and determining the tactile key vector.

[0113] The spatiotemporal joint feature matrix is ​​row-normalized so that the sum of the elements in each row is 1. Specifically, each element in the matrix is ​​divided by the sum of the elements in its row to obtain the normalized feature matrix. Row normalization can eliminate the scale differences between different rows, so that each sample point contributes equally to the subsequent feature extraction.

[0114] Perform eigenvalue decomposition on the normalized feature matrix. First, calculate the covariance matrix of the normalized feature matrix. Then, perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. Sort the eigenvectors according to the size of the eigenvalues, select the eigenvectors corresponding to the first d largest eigenvalues, and construct the main eigenvector matrix. The main eigenvector matrix describes the most important d directions in the normalized feature matrix and can capture the main change patterns of the data.

[0115] The normalized feature matrix is ​​projected using the principal eigenvector matrix. Each sample point in the normalized feature matrix is ​​multiplied by the principal eigenvector matrix through matrix multiplication to map it from the original high-dimensional feature space to the d-dimensional subspace spanned by the principal eigenvector. The matrix obtained after the projection transformation is called a reduced-dimensional feature matrix, in which each row corresponds to the coordinate representation of a sample point in the low-dimensional subspace. The reduced-dimensional feature matrix retains the main features of the original data while reducing the dimension of the data for easy subsequent processing.

[0116] The reduced dimension feature matrix is ​​row normalized so that the sum of the elements in each row is 1. Row normalization can eliminate the scale differences between different sample points, so that each sample point has a similar impact on the subsequent tactile key vector generation.

[0117] Based on the normalized reduced dimension feature matrix, the tactile key vector is generated by the manifold permutation learning algorithm. First, an initial permutation vector with the same dimension as the seed key vector is randomly generated. Then, a permutation loss function is defined to measure the similarity between the permutation vector and the normalized reduced dimension feature matrix. The permutation loss function measures the similarity between the permutation vector and each sample point in the normalized reduced dimension feature matrix by calculating the inner product between them. The larger the inner product, the higher the similarity. With the maximization of the inner product sum of the permutation vector and the normalized reduced dimension feature matrix as the optimization goal, the initial permutation vector is updated by an iterative optimization algorithm (such as gradient descent), and the element value of the permutation vector is continuously adjusted until the inner product sum reaches the maximum or meets the convergence condition. The optimal permutation vector obtained by optimization is the generated tactile key vector. The tactile key vector can effectively capture and characterize the user's touch behavior characteristics by comparing the similarity with the sample points in the normalized reduced dimension feature matrix.

[0118] In this embodiment, the original high-dimensional data is projected into a low-dimensional subspace through eigenvalue decomposition and main eigenvector matrix construction, which effectively retains the main information of the data, reduces the influence of redundant data, and improves the accuracy of feature extraction; through manifold permutation learning, the permutation loss function is optimized, and the optimal permutation vector is iteratively generated, so that the generated tactile key vector not only has high security, but also can be consistent with the dimension of the seed key vector, ensuring the uniqueness and complexity of the key, and is difficult to crack or copy; through dimensionality reduction processing, the data dimension is reduced, the computational complexity is reduced, the processing efficiency of the feature matrix is ​​improved, and the key generation process is more efficient while maintaining the integrity of the information; the feature matrix is ​​normalized to ensure the comparability of data at different scales, reduce the influence of noise and outliers, and thus improve the stability and robustness of the system in different environments.

[0119] S103. Based on the composite key vector, the wireless terminal generates an authentication request data packet through a torus lattice cryptographic mechanism, and sends the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol. The authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation.

[0120] In this embodiment, by generating an authentication request data packet based on a torus lattice cryptographic mechanism and combining an authentication channel established with a quantum secure key negotiation protocol, high-strength encryption that resists quantum computing attacks is provided, ensuring the confidentiality and anti-eavesdropping capabilities of data during transmission; by generating a composite key vector and extracting it through reverse operations, the integrity and uniqueness of the key in the entire authentication process are guaranteed, preventing the key from being tampered with or forged; the authentication server accurately demodulates the authentication request data packet through continuous quantum state detection, ensuring accurate verification of the key during the authentication process, and improving the accuracy and reliability of identity authentication; by using a quantum secure key negotiation protocol, common communication security threats such as man-in-the-middle attacks are effectively prevented, ensuring the security of the authentication channel.

[0121] In an optional embodiment, based on the composite key vector, the wireless terminal generates an authentication request data packet through a torus lattice cryptographic mechanism, sends the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol, the authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation, including:

[0122] Based on the torus lattice cryptographic mechanism, the composite key vector is mapped to a key point on the torus lattice, and a mask point on the torus lattice is randomly generated, the key point and the mask point are added on the torus lattice to obtain a masked key point, the masked key point is converted into a binary code sequence to constitute the payload part of the authentication request data packet, and control information is added to the header of the authentication request data packet to construct the authentication request data packet;

[0123] The quantum secure key negotiation protocol includes a quantum key distribution protocol and a quantum secure communication protocol. The wireless terminal and the authentication server generate a shared security key based on the quantum key distribution protocol, and use the shared security key to establish a quantum secure authentication channel through the quantum secure communication protocol.

[0124] Based on the shared security key, encrypt the authentication request data packet to obtain a ciphertext of the authentication request data packet, modulate it into a continuous quantum state signal, and send it to the authentication server through the quantum secure authentication channel;

[0125] The authentication server demodulates the received continuous quantum state signal using coherent demodulation technology, obtains a demodulated signal in complex form by synchronously demodulating the orthogonal amplitude and phase components of the continuous quantum state signal, performs analog-to-digital conversion on the demodulated signal, converts the complex form into a real form, determines a level signal, performs a binary decision on the level signal according to a preset decision threshold, and obtains a binary symbol sequence in combination with symbol mapping;

[0126] Based on the preset error correction code, correct the channel transmission error, decode the binary symbol sequence to obtain binary data, perform format parsing and data extraction on the binary data based on the preset format of the authentication request data packet, extract the ciphertext of the authentication request data packet, and decrypt it based on the shared security key through a preset symmetric key algorithm to obtain the authentication request data packet;

[0127] The authentication server extracts the composite key vector from the authentication request data packet through the reverse operation, extracts the binary code sequence from the payload part of the authentication request data packet and converts it into the mask key point on the torus lattice, regenerates the mask point using the pre-negotiated mask point generation algorithm, performs a subtraction operation on the torus lattice to obtain the key point on the torus lattice, and maps the key point back to the real number space to obtain the composite key vector.

[0128] The torus lattice cryptographic mechanism is a cryptographic method based on a mathematical structure called a lattice, especially a lattice on a ring structure. A torus lattice is a special type of lattice in which the points are located on a torus structure, usually a modular ring. The torus lattice cryptographic mechanism uses the complexity and difficulty of decoding on the torus lattice to construct a cryptographic system, and its security is believed to be able to resist quantum computing attacks. The main operations of the torus lattice cryptographic mechanism include mapping keys to lattice points, performing mathematical operations (such as addition and subtraction) on the torus lattice, and using the results of these operations to encrypt and decrypt data.

[0129] Based on the torus lattice cryptographic mechanism, the wireless terminal maps the generated composite key vector to a key point on the torus lattice. At the same time, the wireless terminal randomly generates a mask point on the torus lattice. The key point and the mask point are added on the torus lattice to obtain the masked key point. The masked key point is converted into a binary code sequence according to the preset coding rule as the payload part of the authentication request data packet. Necessary control information, such as data packet type and length, is added to the header of the authentication request data packet to construct a complete authentication request data packet.

[0130] The wireless terminal and the authentication server negotiate to generate a shared security key through a quantum key distribution protocol, such as the BB84 protocol. Specifically, the wireless terminal randomly prepares a set of quantum states, such as single photon states, and randomly selects two sets of orthogonal measurement bases, such as horizontal / vertical bases and diagonal bases, to measure the quantum states. Then, the wireless terminal sends the prepared quantum state sequence to the authentication server through a quantum channel. The authentication server randomly selects the same measurement basis as the wireless terminal to measure the received quantum state and records the measurement results. The two parties compare a part of the measurement basis selection and measurement results through a public channel, and filter out the bit values ​​corresponding to the quantum states with the same measurement basis of both parties to form the original key. The two parties finally generate a secure shared key through post-processing steps such as error correction and privacy amplification.

[0131] The wireless terminal and the authentication server use the shared security key generated by negotiation to establish a quantum secure authentication channel through the quantum secure communication protocol. Both parties use the shared key to encrypt the authentication data, modulate the encrypted data into a quantum state, and transmit it through the quantum channel. The quantum secure communication protocol ensures the confidentiality and integrity of the authentication data during transmission.

[0132] The wireless terminal uses the shared security key generated by negotiation with the authentication server to encrypt the constructed authentication request data packet. The encryption process uses a preset symmetric key algorithm. After encryption, the authentication request data packet ciphertext is obtained. The wireless terminal modulates the authentication request data packet ciphertext into a continuous quantum state signal and sends it to the authentication server through the established quantum secure authentication channel.

[0133] The authentication server uses coherent demodulation technology to demodulate the received continuous quantum state signal. Coherent demodulation synchronously demodulates the orthogonal amplitude and phase components of the quantum state signal to obtain a demodulated signal in complex form. Then, the authentication server performs analog-to-digital conversion on the demodulated signal, converts the complex form into a real form, and obtains a level signal. The level signal is binarized according to a preset decision threshold to obtain a binary symbol sequence.

[0134] The authentication server decodes the received binary symbol sequence based on the preset error correction code to correct the bit errors that may occur during the channel transmission process. Binary data is obtained after decoding. The authentication server performs format parsing and data extraction on the binary data according to the preset format of the authentication request data packet to obtain the ciphertext of the authentication request data packet. Then, the authentication server uses the shared security key generated by negotiation with the wireless terminal to decrypt the ciphertext of the authentication request data packet through the preset symmetric key algorithm to obtain the original authentication request data packet.

[0135] The authentication server extracts the binary code sequence from the payload of the authentication request data packet and converts it into a mask key point on the torus lattice. The authentication server regenerates the mask point on the torus lattice using the mask point generation algorithm pre-negotiated with the wireless terminal. The mask key point is subtracted from the mask point on the torus lattice to obtain the key point on the torus lattice. Finally, the authentication server maps the key point from the torus lattice back to the real number space to obtain the composite key vector generated by the wireless terminal.

[0136] In this embodiment, the torus lattice cryptographic mechanism is based on a difficult mathematical problem that is difficult to crack even on a quantum computer, thereby providing extremely high security; by mapping the composite key vector to the torus lattice and performing masking processing, the generated masked key point increases the complexity of the encrypted data, ensuring that it is difficult to be cracked or tampered with during transmission; the authentication server restores the key point and extracts the composite key vector through reverse operations, thereby ensuring the integrity and accuracy of the key during transmission and decoding, and avoiding information loss or erroneous decryption; combining quantum key distribution and quantum secure communication protocols, a highly secure communication channel is provided to ensure the confidentiality and anti-eavesdropping capability of the authentication request data packet during transmission; through masking processing and secure channel transmission, the risk of key exposure during communication is avoided, further improving the security of the system.

[0137] S104. Split the composite key vector into a seed key vector and a tactile key vector, and evaluate the cosine similarity with the corresponding vectors in the standard auditory-tactile interaction model library pre-constructed on the authentication server through deep metric learning. If the cosine similarities exceed the preset similarity threshold, the authentication is determined to be successful, and the post-quantum blind signature scheme is used to send the authentication credential to the wireless terminal, and the authentication credential is stored in a threshold signature blockchain ledger based on lattice cryptography to complete the wireless terminal encryption authentication process.

[0138] In this embodiment, the cosine similarity between the composite key vector and the standard model is evaluated through deep metric learning to ensure that the authentication process can accurately distinguish between legitimate users and potential attackers; combined with the post-quantum blind signature scheme and the threshold signature blockchain ledger based on lattice cryptography, a strong security guarantee against quantum computing attacks is provided to prevent the authentication credentials from being forged or tampered with; the authentication credentials are stored on the blockchain, ensuring their immutability and traceability, and enhancing the authentication credibility of the wireless terminal; through automated similarity evaluation and encrypted storage mechanism, the authentication process can be completed efficiently while ensuring security.

[0139] In an optional embodiment, the composite key vector is split into a seed key vector and a tactile key vector, and the cosine similarities with the corresponding vectors in the standard auditory tactile interaction model library pre-built on the authentication server are respectively evaluated through deep metric learning. If the cosine similarities exceed a preset similarity threshold, then determining that the authentication is passed includes:

[0140] Construct a twin network containing two sub-networks with the same structure, each of which consists of multiple convolutional layers, pooling layers, fully connected layers and an embedding layer, wherein the embedding layer maps the output of the fully connected layer to a low-dimensional embedding space, and introduces an attention mechanism. An attention module is set after each convolutional layer, and by generating an attention weight matrix, the attention to different areas in the convolutional feature map is adaptively adjusted to extract key features;

[0141] The twin network is trained, and based on the preset category labels of the training samples, the category samples are determined, and the category clustering is constructed by minimizing the same-category embedding distance between the same-category sample pairs and maximizing the heterogeneous embedding distance between the heterogeneous category sample pairs. The center loss function is combined as an auxiliary loss, and the intra-class aggregation and inter-class separation are maximized by minimizing the distance between each category sample and the corresponding category clustering center to generate a discriminative embedding space;

[0142] Based on a standard auditory-tactile interaction model library, determine a standard seed key vector and a standard tactile key vector, wherein the seed key vector and the standard seed key vector constitute a first vector comparison pair, and the tactile key vector and the standard tactile key vector constitute a second vector comparison pair;

[0143] Inputting the first vector comparison pair and the second vector comparison pair into the sub-networks in the twin network respectively, obtaining the first feature representation vector pair and the second feature representation vector pair in the discriminant embedding space respectively, calculating the cosine value of the angle respectively by the cosine similarity function, and determining the first cosine similarity and the second cosine similarity;

[0144] The first cosine similarity and the second cosine similarity are weightedly fused to obtain a comprehensive similarity, and the comprehensive similarity is compared with a preset similarity threshold. When the comprehensive similarity exceeds the threshold, it is determined that the authentication is passed; otherwise, it is determined that the authentication fails.

[0145] Construct a twin network containing two sub-networks with the same structure. Each sub-network consists of multiple convolutional layers, pooling layers, fully connected layers and an embedding layer. The convolutional layer extracts local features of the input data through convolution operations. The pooling layer downsamples the output of the convolutional layer to reduce the size of the feature map and improve the robustness of the features. The fully connected layer flattens the features after convolution and pooling to extract and fuse high-level features. The embedding layer maps the output of the fully connected layer to a low-dimensional embedding space to learn compact feature representations. After each convolutional layer, an attention module is introduced. The attention module adaptively adjusts the attention to different areas in the convolutional feature map by generating an attention weight matrix, highlighting key features and suppressing redundant information.

[0146] The twin network is trained using a pre-constructed training sample set. The training samples include seed key vectors and tactile key vectors, as well as corresponding category labels. By minimizing the embedding distance between pairs of samples of the same category and maximizing the embedding distance between pairs of samples of different categories, samples of the same category are clustered together in the embedding space, and samples of different categories are separated in the embedding space. The center loss function is introduced as an auxiliary loss to further enhance the intra-class aggregation and inter-class separation by minimizing the distance between each category sample and the corresponding category clustering center. The network parameters are iteratively updated through optimization algorithms such as gradient descent, and finally a highly discriminative embedding space is generated.

[0147] Based on the auditory-tactile interaction data of a large number of legitimate users, a standard auditory-tactile interaction model library is constructed. Representative seed key vectors and tactile key vectors are selected from the model library as standard seed key vectors and standard tactile key vectors. The standard seed key vector and the standard tactile key vector form a first vector comparison pair and a second vector comparison pair with the seed key vector and the tactile key vector to be authenticated, respectively.

[0148] The first vector comparison pair is input into a sub-network of the twin network, and after being processed by the convolution layer, attention module, pooling layer, fully connected layer and embedding layer, the first feature representation vector pair in the discriminative embedding space is obtained. Similarly, the second vector comparison pair is input into another sub-network of the twin network to obtain the second feature representation vector pair in the discriminative embedding space.

[0149] The cosine similarity function is used to calculate the cosine value of the angle between the first feature representation vector pair and the second feature representation vector pair, respectively, to obtain the first cosine similarity and the second cosine similarity. Cosine similarity can measure the consistency of the direction of two vectors, and the value range is between -1 and 1. The closer the cosine similarity is to 1, the more consistent the directions of the two vectors are, and the higher the similarity is; the closer the cosine similarity is to -1, the more opposite the directions of the two vectors are, and the lower the similarity is.

[0150] The first cosine similarity and the second cosine similarity are weightedly fused to obtain a comprehensive similarity. The weighted fusion can assign different weight coefficients according to the degree of influence of the seed key and the tactile key on the authentication result. The comprehensive similarity is compared with the preset similarity threshold. If the comprehensive similarity exceeds the threshold, the authentication is judged to be passed, that is, it is considered that the auditory tactile interaction data to be authenticated has a sufficiently high similarity with the data in the standard model library and belongs to a legitimate user. If the comprehensive similarity does not exceed the threshold, the authentication is judged to have failed, that is, it is considered that the auditory tactile interaction data to be authenticated has insufficient similarity with the data in the standard model library and belongs to an illegal user or attacker.

[0151] In this embodiment, through the structure and attention mechanism of the twin network, the key features in the input data can be accurately extracted, and the quality of the embedded representation of the seed key vector and the tactile key vector can be improved; by optimizing the training process by using the distance between similar sample pairs and heterogeneous sample pairs and combining the center loss function, the intra-class aggregation and inter-class separation can be effectively improved to generate a more discriminative embedding space; through cosine similarity calculation and weighted fusion comprehensive similarity, the standard seed key vector and the standard tactile key vector can be more accurately matched, thereby improving the accuracy and robustness of the authentication judgment; through precise similarity comparison and threshold judgment mechanism, unauthorized access can be effectively prevented to ensure the security of the authentication system; the mechanism of comprehensive similarity and threshold comparison makes the authentication decision process fast and reliable, suitable for real-time application scenarios.

[0152] Figure 2 FIG. 1 is a schematic diagram of a wireless terminal encryption authentication system based on auditory and tactile interaction according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0153] The first unit is used for the wireless terminal to collect the user's voice command through the microphone array, perform feature extraction on the voice command, obtain a compact feature representation of the voice command by adaptively weighting the voice feature sequence through the set attention pooling layer, learn the probability density distribution of the voice command through variational inference based on the compact feature representation, generate a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, and extract the time-frequency energy distribution characteristics of the keyword vector through multi-scale adaptive signal decomposition to form a seed key vector;

[0154] The second unit is used to collect continuous touch trajectory data on the curved touch interface of the wireless terminal through a flexible tactile sensor array, construct a six-dimensional touch gesture feature manifold, perform spatiotemporal sequence modeling on the six-dimensional touch gesture feature manifold based on a graph attention network, extract the spatiotemporal evolution pattern of the gesture manifold through a graph attention layer and a temporal convolution layer, generate a tactile key vector consistent with the dimension of the seed key vector based on a manifold permutation learning algorithm, and fuse it with the seed key vector to generate a composite key vector;

[0155] A third unit is configured to generate an authentication request data packet by the wireless terminal through a torus lattice cryptographic mechanism based on the composite key vector, and send the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol, wherein the authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation;

[0156] The fourth unit is used to split the composite key vector into a seed key vector and a tactile key vector, and evaluate the cosine similarity of the corresponding vectors in the standard auditory-tactile interaction model library pre-constructed on the authentication server through deep metric learning. If the cosine similarity exceeds the preset similarity threshold, the authentication is determined to be successful, and the post-quantum blind signature scheme is used to send the authentication certificate to the wireless terminal, and the authentication certificate is stored in a threshold signature blockchain account book based on lattice cryptography to complete the wireless terminal encryption authentication process.

[0157] According to a third aspect of the embodiments of the present invention,

[0158] An electronic device is provided, comprising:

[0159] processor;

[0160] a memory for storing processor-executable instructions;

[0161] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0162] According to a fourth aspect of the embodiments of the present invention,

[0163] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0164] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wireless terminal encryption authentication method based on auditory and tactile interaction, characterized in that: include: The wireless terminal collects the user's voice command through a microphone array, extracts features of the voice command, adaptively weights the voice feature sequence through the set attention pooling layer, obtains a compact feature representation of the voice command, learns the probability density distribution of the voice command through variational inference based on the compact feature representation, generates a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, extracts the time-frequency energy distribution features of the keyword vector through multi-scale adaptive signal decomposition, and forms a seed key vector; On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array to construct a six-dimensional touch gesture feature manifold, and the six-dimensional touch gesture feature manifold is modeled in time and space sequence based on a graph attention network. The time and space evolution pattern of the gesture manifold is extracted through a graph attention layer and a temporal convolution layer. Based on a manifold permutation learning algorithm, a tactile key vector consistent with the dimension of the seed key vector is generated, and the tactile key vector is fused with the seed key vector to generate a composite key vector; Based on the composite key vector, the wireless terminal generates an authentication request data packet through a torus lattice cryptographic mechanism, and sends the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol. The authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation. The composite key vector is split into a seed key vector and a tactile key vector, and the cosine similarity with the corresponding vector in the standard auditory-tactile interaction model library pre-built on the authentication server is evaluated by deep metric learning. If the cosine similarity exceeds the preset similarity threshold, the authentication is determined to be successful, and the post-quantum blind signature scheme is used to send the authentication certificate to the wireless terminal, and the authentication certificate is stored in a threshold signature blockchain account book based on a lattice password, completing the wireless terminal encryption authentication process; The composite key vector is split into a seed key vector and a tactile key vector, and the cosine similarity with the corresponding vectors in the standard auditory tactile interaction model library pre-built on the authentication server is evaluated by deep metric learning. If the cosine similarities exceed the preset similarity threshold, the authentication is determined to be passed, including: Construct a twin network containing two sub-networks with the same structure, each of which consists of multiple convolutional layers, pooling layers, fully connected layers and an embedding layer, wherein the embedding layer maps the output of the fully connected layer to a low-dimensional embedding space, and introduces an attention mechanism. An attention module is set after each convolutional layer, and by generating an attention weight matrix, the attention to different areas in the convolutional feature map is adaptively adjusted to extract key features; The twin network is trained, and based on the preset category labels of the training samples, the category samples are determined, and the category clustering is constructed by minimizing the same-category embedding distance between the same-category sample pairs and maximizing the heterogeneous embedding distance between the heterogeneous category sample pairs. The center loss function is combined as an auxiliary loss, and the intra-class aggregation and inter-class separation are maximized by minimizing the distance between each category sample and the corresponding category clustering center to generate a discriminative embedding space; Based on a standard auditory-tactile interaction model library, determine a standard seed key vector and a standard tactile key vector, wherein the seed key vector and the standard seed key vector constitute a first vector comparison pair, and the tactile key vector and the standard tactile key vector constitute a second vector comparison pair; Inputting the first vector comparison pair and the second vector comparison pair into the sub-networks in the twin network respectively, obtaining the first feature representation vector pair and the second feature representation vector pair in the discriminant embedding space respectively, calculating the cosine value of the angle respectively by the cosine similarity function, and determining the first cosine similarity and the second cosine similarity; The first cosine similarity and the second cosine similarity are weightedly fused to obtain a comprehensive similarity, and the comprehensive similarity is compared with a preset similarity threshold. When the comprehensive similarity exceeds the threshold, it is determined that the authentication is passed; otherwise, it is determined that the authentication fails.

2. The method according to claim 1, characterized in that The wireless terminal collects the user's voice command through a microphone array, extracts features of the voice command, adaptively weights the voice feature sequence through the set attention pooling layer, obtains a compact feature representation of the voice command, learns the probability density distribution of the voice command through variational inference based on the compact feature representation, generates a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, extracts the time-frequency energy distribution features of the keyword vector through multi-scale adaptive signal decomposition, and forms a seed key vector including: The wireless terminal collects the user's voice commands through a microphone array, divides the voice commands into frames, performs Fourier transform on each frame, determines the spectrum information, extracts filter energy through a scale filter group, performs weighted processing on the filter energy according to an equal loudness curve, combines logarithmic compression, and performs feature extraction through a preset linear prediction analysis method to obtain an initial feature sequence; Learning the importance weights of the initial feature sequence at different time steps through an attention pooling layer, and performing weighted summation on the initial feature sequence based on the importance weights to obtain a compact feature representation of the voice command; The compact feature representation is regarded as the output of the latent variable model, and the variational inference model is determined with the goal of minimizing the KL divergence between the variational distribution and the true posterior distribution. The latent variables are sampled from the prior distribution through the stochastic gradient variational Bayes algorithm, and the compact feature representation is reconstructed and output through the decoder. After iteratively updating the decoder parameters, the solution that is most similar to the true posterior distribution is obtained, and the final probability density distribution of the voice command is determined; Based on the final probability density distribution, mapping the latent variables into keyword vectors through a decoder, and selecting the keyword vector most relevant to the voice command by minimizing the reconstruction error between the keyword vector and the compact feature representation; Performing empirical mode decomposition on the keyword vector to generate eigenmode function components, performing Hilbert transform on the eigenmode function components to obtain analytical signals, and calculating instantaneous frequency and instantaneous amplitude of the eigenmode function components based on the analytical signals, and combining them to generate time-frequency energy distribution features; The time-frequency energy distribution feature is quantized and encoded to obtain a quantized coding sequence, and the quantized coding sequence is hash mapped using a secure hash algorithm to obtain a hash value of a fixed length as a seed key vector.

3. The method according to claim 1, characterized in that On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array to construct a six-dimensional touch gesture feature manifold, and the six-dimensional touch gesture feature manifold is modeled in time and space sequence based on a graph attention network. The time and space evolution pattern of the gesture manifold is extracted through a graph attention layer and a time convolution layer. Based on a manifold permutation learning algorithm, a tactile key vector consistent with the dimension of the seed key vector is generated, including: On the curved touch interface of the wireless terminal, continuous touch trajectory data is collected through a flexible tactile sensor array, wherein the continuous touch trajectory data includes a touch coordinate sequence, a touch force sequence, a normal touch force component, a tangential touch force component, and a fluid impedance sequence; Based on the continuous touch trajectory data, each touch data point in the continuous touch trajectory data is regarded as a sample point in the six-dimensional touch gesture feature manifold, and a local linear embedding algorithm is used to perform nonlinear dimensionality reduction on the sample point, and the six-dimensional touch gesture data is mapped to a low-dimensional manifold space by maintaining the local linear structure and reducing the dimension of the global nonlinear structure, so as to obtain a six-dimensional touch gesture feature manifold; A graph attention network is constructed to perform spatiotemporal sequence modeling on the six-dimensional touch gesture feature manifold, wherein the graph attention network is formed by alternately stacking a plurality of graph attention layers and a plurality of temporal convolution layers, wherein in each graph attention layer, the neighborhood information of each sample point in the six-dimensional touch gesture feature manifold is adaptively aggregated through an attention mechanism, the dynamic association weights between the sample points are learned, the feature representation of the sample points is updated, and the spatial structure information is determined; in each temporal convolution layer, a causal one-dimensional convolution operation is used to extract the temporal dynamic features of the six-dimensional touch gesture feature manifold at different time scales, and the temporal evolution information is determined; Based on the spatial structure information and the time evolution information, a spatiotemporal joint feature matrix is ​​determined. Based on the spatiotemporal joint feature matrix, a dimensionality transformation is performed through a manifold permutation learning algorithm to generate a tactile key vector that is consistent with the dimension of the seed key vector.

4. The method according to claim 3, characterized in that The manifold permutation learning algorithm includes: Performing row normalization processing on the spatiotemporal joint feature matrix to obtain a normalized feature matrix; Based on the normalized feature matrix, the eigenvalues ​​corresponding to the eigenvectors in the normalized feature matrix are calculated by eigenvalue decomposition, and the eigenvectors corresponding to the largest d eigenvalues ​​are taken according to the preset number of features d to construct a main eigenvector matrix; Based on the main eigenvector matrix, the normalized feature matrix is ​​projected, and each sample point in the normalized feature matrix is ​​projected from the feature space of the original dimension to the low-dimensional subspace spanned by the main eigenvector to obtain a reduced-dimensional feature matrix; Perform row normalization on the reduced dimension feature matrix to obtain a normalized reduced dimension feature matrix; Based on the normalized reduced-dimensionality feature matrix, the tactile key vector is obtained through manifold permutation learning, including: randomly generating an initial permutation vector consistent with the dimension of the seed key vector, optimizing the permutation loss function, taking maximizing the inner product between the initial permutation vector and each sample point in the normalized reduced-dimensionality feature matrix as the goal, iteratively generating the optimal permutation vector, and determining the tactile key vector.

5. The method according to claim 1, characterized in that Based on the composite key vector, the wireless terminal generates an authentication request data packet through a torus lattice cryptographic mechanism, sends the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol, the authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation, including: Based on the torus lattice cryptographic mechanism, the composite key vector is mapped to a key point on the torus lattice, and a mask point on the torus lattice is randomly generated, the key point and the mask point are added on the torus lattice to obtain a masked key point, the masked key point is converted into a binary code sequence to constitute the payload part of the authentication request data packet, and control information is added to the header of the authentication request data packet to construct the authentication request data packet; The quantum secure key negotiation protocol includes a quantum key distribution protocol and a quantum secure communication protocol. The wireless terminal and the authentication server generate a shared security key based on the quantum key distribution protocol, and use the shared security key to establish a quantum secure authentication channel through the quantum secure communication protocol. Based on the shared security key, encrypt the authentication request data packet to obtain a ciphertext of the authentication request data packet, modulate it into a continuous quantum state signal, and send it to the authentication server through the quantum secure authentication channel; The authentication server demodulates the received continuous quantum state signal using coherent demodulation technology, obtains a demodulated signal in complex form by synchronously demodulating the orthogonal amplitude and phase components of the continuous quantum state signal, performs analog-to-digital conversion on the demodulated signal, converts the complex form into a real form, determines a level signal, performs a binary decision on the level signal according to a preset decision threshold, and obtains a binary symbol sequence in combination with symbol mapping; Based on the preset error correction code, correct the channel transmission error, decode the binary symbol sequence to obtain binary data, perform format parsing and data extraction on the binary data based on the preset format of the authentication request data packet, extract the ciphertext of the authentication request data packet, and decrypt it based on the shared security key through a preset symmetric key algorithm to obtain the authentication request data packet; The authentication server extracts the composite key vector from the authentication request data packet through the reverse operation, extracts the binary code sequence from the payload part of the authentication request data packet and converts it into the mask key point on the torus lattice, regenerates the mask point using the pre-negotiated mask point generation algorithm, performs a subtraction operation on the torus lattice to obtain the key point on the torus lattice, and maps the key point back to the real number space to obtain the composite key vector.

6. A wireless terminal encryption authentication system based on auditory and tactile interaction, used to implement the method described in any one of claims 1 to 5, characterized in that: include: The first unit is used for the wireless terminal to collect the user's voice command through the microphone array, perform feature extraction on the voice command, obtain a compact feature representation of the voice command by adaptively weighting the voice feature sequence through the set attention pooling layer, learn the probability density distribution of the voice command through variational inference based on the compact feature representation, generate a keyword vector most relevant to the semantics of the voice command through random sampling and reconstruction error minimization, and extract the time-frequency energy distribution characteristics of the keyword vector through multi-scale adaptive signal decomposition to form a seed key vector; The second unit is used to collect continuous touch trajectory data on the curved touch interface of the wireless terminal through a flexible tactile sensor array, construct a six-dimensional touch gesture feature manifold, perform spatiotemporal sequence modeling on the six-dimensional touch gesture feature manifold based on a graph attention network, extract the spatiotemporal evolution pattern of the gesture manifold through a graph attention layer and a temporal convolution layer, generate a tactile key vector consistent with the dimension of the seed key vector based on a manifold permutation learning algorithm, and fuse it with the seed key vector to generate a composite key vector; A third unit is configured to generate an authentication request data packet by the wireless terminal through a torus lattice cryptographic mechanism based on the composite key vector, and send the authentication request data packet to an authentication server through an authentication channel established based on a quantum secure key agreement protocol, wherein the authentication server demodulates the authentication request data packet through continuous quantum state detection, and extracts the composite key vector through a reverse operation; The fourth unit is used to split the composite key vector into a seed key vector and a tactile key vector, and evaluate the cosine similarity of the corresponding vectors in the standard auditory-tactile interaction model library pre-constructed on the authentication server through deep metric learning. If the cosine similarity exceeds the preset similarity threshold, the authentication is determined to be successful, and the post-quantum blind signature scheme is used to send the authentication certificate to the wireless terminal, and the authentication certificate is stored in a threshold signature blockchain account book based on lattice cryptography to complete the wireless terminal encryption authentication process.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images