Electromagnetic signal transmission method and system and computer readable storage medium

Abstract

The invention relates to an electromagnetic signal transmission method and system and a computer readable storage medium, and the method comprises the steps: inputting an obtained electromagnetic signal to a first noise generator, carrying out the noise addition of the electromagnetic signal based on the first learning interference resistance noise of the first noise generator, and generating a noise interference electromagnetic signal; the first noise generator inputs the obtained first training data into the initial neural network model, injects the first preset interference noise into each network level in the initial neural network model, and performs multi-level fusion processing on the first preset interference noise and the first training data through each network level; iteration training generation is carried out; the first anti-learning interference noise is obtained by performing minimum learning on the first preset interference noise in the iterative training process of the initial neural network model; and transmitting the noise interference electromagnetic signal to a signal receiving device. According to the invention, the problem of low security of electromagnetic signal transmission is solved.

Description

Technical Field

[0001] The present application relates to the field of signal transmission, in particular to an electromagnetic signal transmission method, system, and computer-readable storage medium. Background Art

[0002] With the advent of the information age, the technical level and business scale in the field of electromagnetic signal transmission have achieved leapfrog growth. Electromagnetic signals have been significantly improved in terms of speed, stability, distance, and efficiency. The deep integration of artificial intelligence technology has further promoted the intelligence and convenience of electromagnetic signal processing. Especially in tasks such as signal recognition, modulation and demodulation, and interference suppression, deep neural networks have shown better performance than traditional algorithms, significantly meeting the actual needs of people for communication services. However, the inherent "black box property" and "adversarial vulnerability" of deep learning models make them vulnerable to targeted attacks, resulting in difficulty in meeting the electromagnetic signal data protection requirements in related technologies and being unable to be used for secure transmission of communication signals, thus affecting the security of electromagnetic signal transmission.

[0003] Currently, no effective solution has been proposed for the problem of low security in electromagnetic signal transmission in related technologies. Summary of the Invention

[0004] Embodiments of the present application provide an electromagnetic signal transmission method, system, and computer-readable storage medium to at least solve the problem of low security in electromagnetic signal transmission in related technologies.

[0005] In a first aspect, an embodiment of the present application provides an electromagnetic signal transmission method, which is applied to a signal sending device. The method includes:

[0006] Obtain an electromagnetic signal to be transmitted;

[0007] Input the electromagnetic signal into a trained first noise generator, and perform noise addition processing on the electromagnetic signal based on a first anti-learning interference noise carried by the first noise generator to generate a noise-interfered electromagnetic signal;

[0008] Wherein, the first noise generator is generated by inputting obtained first training data into an initial neural network model, injecting a first preset interference noise into each network layer of the initial neural network model, performing multi-level fusion processing on the first preset interference noise and the first training data through each network layer, and iteratively training; the first anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training of the initial neural network model;

[0009] Transmit the noise-interfered electromagnetic signal to a signal receiving device.

[0010] In some of these embodiments, the signal receiving device is deployed with a first interference remover for performing noise removal processing on the noise-interfered electromagnetic signal. The training process of the first noise generator and the first interference remover includes:

[0011] Input the first training data into the first autoencoder model; wherein, the initial neural network model is set as the first autoencoder model;

[0012] Via the first encoder network of the first autoencoder model, perform multi-layer feature extraction processing on the first training data to obtain hierarchical feature data corresponding to each network layer in the first encoder network, and at each network layer in the first encoder network, fuse the injected first preset interference noise with the hierarchical feature data, and output the first fused training feature;

[0013] Input the first fused training feature into the first decoder network of the first autoencoder model for decoding processing to obtain the first reconstructed feature data;

[0014] Based on the first training data and the first reconstructed feature data, calculate the result of the first reconstruction loss function, and backpropagate the gradient of the result of the first reconstruction loss function to the first autoencoder model for iterative training to obtain the first target encoder model; the first target encoder model includes the first noise generator and the first interference remover.

[0015] In some of these embodiments, the calculating the result of the first reconstruction loss function based on the first training data and the first reconstructed feature data, and backpropagating the gradient of the result of the first reconstruction loss function to the first autoencoder model for iterative training to obtain the first target encoder model includes:

[0016] In the first training stage, calculate the result of the first reconstruction loss function;

[0017] In the second training stage, input a preset stage perturbation into the first autoencoder model, and use the first autoencoder model to fuse the first preset interference noise, the stage perturbation and the hierarchical feature data, output the second fused training feature, and input the second fused training feature into the first decoder network for decoding processing to obtain the second reconstructed feature data;

[0018] Calculate the result of the second reconstruction loss function according to the first training data and the second reconstructed feature data;

[0019] Backpropagate the gradients of the first reconstruction loss function result and the second reconstruction loss function result to the first autoencoder model for iterative training to obtain the first target encoder model.

[0020] In some embodiments, the first training data carries classification labels; the using the first autoencoder model to fuse the first preset interference noise, the stage perturbation and the hierarchical feature data, and output the second fused training feature includes:

[0021] Using the first autoencoder model to fuse the first preset interference noise, the stage perturbation and the hierarchical feature data, output the second fused training feature, and perform classification processing based on the second fused training feature to output a classification result;

[0022] The backpropagating the gradients of the first reconstruction loss function result and the second reconstruction loss function result to the first autoencoder model for iterative training to obtain the first target encoder model includes:

[0023] Calculating a classification loss function result based on the classification result and the classification label, and calculating a joint loss function result according to the classification loss function result and the second reconstruction loss function result;

[0024] Backpropagate the gradients of the first reconstruction loss function result and the joint loss function result to the first autoencoder model for iterative training to obtain the first target encoder model.

[0025] In some embodiments, after obtaining the electromagnetic signal to be transmitted, the method further includes:

[0026] Performing semantic feature extraction processing on the electromagnetic signal to obtain a semantic feature vector;

[0027] Input the semantic feature vector into the trained second noise generator, and perform noise addition processing on the electromagnetic signal based on the second anti-learning interference noise carried by the second noise generator to generate a noise interference semantic vector;

[0028] Wherein, the second noise generator is generated by inputting the obtained second training data into an initial neural network model, injecting second preset interference noise into each network layer of the initial neural network model, performing multi-level fusion processing on the second preset interference noise and the second training data through each network layer, and performing iterative training; the second anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training of the initial neural network model;

[0029] Transmit the noise interference semantic vector to the signal receiving device.

[0030] In some embodiments, the semantic feature extraction process for the electromagnetic signal to obtain a semantic feature vector includes:

[0031] Input the electromagnetic signal into a trained signal encoder for semantic feature extraction processing to obtain the semantic feature vector.

[0032] In some embodiments, the signal receiving device is deployed with a signal decoder for feature restoration of the noise interference semantic vector; the training process of the signal encoder and the signal decoder includes:

[0033] Obtain third training data;

[0034] Input the third training data into the second encoder network in the second autoencoder model, and perform encoding processing on the third training data to obtain training semantic feature data;

[0035] Input the training semantic feature data into the second decoder network in the second autoencoder model, and perform reconstruction processing on the training semantic feature data to obtain training reconstruction data;

[0036] Based on the third training data and the training reconstruction data, calculate the result of the semantic feature loss function;

[0037] Backpropagate the gradient of the result of the semantic feature loss function to the second autoencoder model for iterative training, and generate a second target encoder model; the second target encoder model includes the signal encoder and the signal decoder.

[0038] In some embodiments, the transmitting the noise interference electromagnetic signal to the signal receiving device includes:

[0039] Obtain a preset random encryption seed;

[0040] Based on the random encryption seed, perform a channel dynamic sorting operation on the noise interference electromagnetic signal to obtain a random encrypted signal, and transmit the random encrypted signal to the signal receiving device.

[0041] In a second aspect, an embodiment of the present application provides an electromagnetic signal transmission system, including a signal sending device and a signal receiving device;

[0042] The signal sending device is communicatively connected to the signal receiving device and is configured to execute the electromagnetic signal transmission method as described in the first aspect above.

[0043] Fourthly, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the electromagnetic signal transmission method as described in the first aspect above.

[0044] Compared with the related art, the electromagnetic signal transmission method, system and computer-readable storage medium provided by the embodiments of the present application obtain the electromagnetic signal to be transmitted; input the electromagnetic signal into the trained first noise generator, and perform noise addition processing on the electromagnetic signal based on the first anti-learning interference noise carried by the first noise generator to generate a noise-interfered electromagnetic signal; wherein, the first noise generator is generated by inputting the obtained first training data into the initial neural network model, injecting the first preset interference noise into each network layer of the initial neural network model, performing multi-level fusion processing on the first preset interference noise and the first training data through each network layer, and iteratively training; the first anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training of the initial neural network model; and the noise-interfered electromagnetic signal is transmitted to the signal receiving device.

[0045] Based on this, via the first noise generator, the first anti-learning interference noise is added to the electromagnetic signal, and the noise will be input through multiple levels and fused with the electromagnetic signal at multiple scales, so as to retain information as much as possible, realize dynamic injection of multi-level anti-learning interference noise, and the generated anti-learning interference noise combined with the original features can effectively prevent data from being attacked and stolen to generate attack samples, effectively enhancing the anti-interference and anti-eavesdropping capabilities of the original electromagnetic signal, realizing secure transmission of electromagnetic signals, and thus solving the problem of low security of electromagnetic signal transmission.

[0046] The details of one or more embodiments of the present application are set forth in the following drawings and description to make the other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation of the present application. In the drawings:

[0048] Figure 1 is a hardware structure block diagram of a terminal for an electromagnetic signal transmission method according to an embodiment of the present application;

[0049] Figure 2 is a flowchart of an electromagnetic signal transmission method according to an embodiment of the present application;

[0050] Figure 3 is a schematic structural diagram of a first autoencoder model according to an embodiment of the present application;

[0051] Figure 4 is a flowchart of a training method for a first target encoder model according to an embodiment of the present application;

[0052] Figure 5 is a schematic structural diagram of a second autoencoder model according to an embodiment of the present application;

[0053] Figure 6 is a schematic diagram of an electromagnetic signal transmission system architecture according to an embodiment of the present application;

[0054] Figure 7 is a flowchart of a method for constructing an electromagnetic signal transmission system architecture according to an embodiment of the present application. Detailed implementation manners

[0055] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the present application will be described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments provided in the present application without creative efforts fall within the scope of protection of the present application.

[0056] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application. For those of ordinary skill in the art, without creative efforts, the present application can also be applied to other similar scenarios based on these drawings. In addition, it can also be understood that although the efforts made in this development process may be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing or production changes made based on the technical content disclosed in the present application are only conventional technical means and should not be understood as insufficient disclosure of the content of the present application.

[0057] Referring to "embodiment" in the present application means that the specific features, structures or characteristics described in combination with the embodiment can be included in at least one embodiment of the present application. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those of ordinary skill in the art explicitly and implicitly understand that the embodiments described in the present application can be combined with other embodiments without conflict.

[0058] Unless otherwise defined, technical terms or scientific terms involved in this application shall have the ordinary meanings understood by those with ordinary skills in the technical field to which this application belongs. The words such as "a", "an", "one", "the" and the like involved in this application do not indicate a limitation in quantity and may represent a singular or plural number. The terms "include", "comprise", "have" and any variations thereof involved in this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may further include steps or units not listed, or may further include other steps or units inherent to these processes, methods, products or devices. The words such as "connect", "be connected", "couple" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "plurality" involved in this application means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, "A and / or B" may represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. The terms "first", "second", "third" and the like involved in this application are only used to distinguish similar objects and do not represent a specific order of the objects.

[0059] The method embodiment provided in this embodiment can be executed on a terminal, a computer or a similar computing device. Taking running on a terminal as an example, Figure 1 is a hardware structure block diagram of a terminal for an electromagnetic signal transmission method according to an embodiment of this application. As Figure 1 shown, the terminal may include one or more ( Figure 1 only one is shown in the figure) processors 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the above terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above terminal. For example, the terminal may further include more or fewer components than those shown in Figure 1 the figure, or have a different configuration from that shown in Figure 1 the figure.

[0060] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the electromagnetic signal transmission method in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implements the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and combinations thereof.

[0061] The transmission device 106 is used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by a communication provider of the terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one instance, the transmission device 106 can be a radio frequency (Radio Frequency, abbreviated as RF) module, which is used to communicate with the Internet wirelessly.

[0062] This embodiment also provides an electromagnetic signal transmission method. Figure 2 is a flowchart of an electromagnetic signal transmission method according to an embodiment of the present application, as Figure 2 shown, and the process includes the following steps:

[0063] Step S210, obtain the electromagnetic signal to be transmitted.

[0064] The acquisition of the electromagnetic signal is usually completed through sensors or antenna arrays, etc. Its essence is to modulate information (such as voice, image, data, etc.) onto electromagnetic waves in a specific manner, and complete signal acquisition, preprocessing, and encoding.

[0065] It should also be supplemented that after the electromagnetic signal emitted by the signal source is obtained, a series of signal preprocessing such as signal modulation can also be performed on it. Specifically, taking the electromagnetic signal after IQ modulation as an example. IQ modulation means that the data is divided into two paths and carrier modulation is performed separately, and the two carriers are orthogonal to each other. Among them, I represents in-phase, and Q represents quadrature. Then, preprocess the IQ signal to be sent, splice the I component and Q component matrices of the signal to obtain the electromagnetic signal to be transmitted.

[0066] Step S220: Input the electromagnetic signal into the trained first noise generator, and perform noise addition processing on the electromagnetic signal based on the first anti-learning interference noise carried by the first noise generator to generate a noise-interfered electromagnetic signal.

[0067] Among them, the first noise generator is generated by inputting the obtained first training data into the initial neural network model, injecting the first preset interference noise into each network layer in the initial neural network model, performing multi-level fusion processing on the first preset interference noise and the first training data through each network layer, and iteratively training; the first anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training process of the initial neural network model.

[0068] In this step, it is mainly to use the trained neural network model (the first noise generator) to generate specific interference noise and inject it into the electromagnetic signal. The training process of the first noise generator will be described in detail below:

[0069] The initial neural network model usually adopts a deep learning model, such as a U-shaped Network (UNet), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Generative Adversarial Networks (GAN), or an autoencoder, etc., and contains multiple hidden layers (convolutional layers, fully connected layers, etc.). Its core design is that during the training phase, the first preset interference noise (such as Gaussian white noise, impulse noise) is used as an additional input and injected into multiple layers of the neural network, rather than a single layer, to achieve dynamic adaptation of noise and signal through multi-level fusion. More specifically, the first training data (typical electromagnetic signal samples, such as voice, image, or modulation signal) is input into the initial neural network model. In the forward propagation of each layer of the network, the preset interference noise is fused with the feature map (Feature Map) of the current layer (such as adding pixel by pixel or channel splicing). The noise and signal features interact repeatedly in multiple layers of the neural network, forming deep coupling; for example, in the convolutional layer, the noise affects local feature extraction (such as edges, textures), and in the fully connected layer, the noise interferes with global feature mapping (such as the classification decision boundary). Then, the network weights are adjusted through the backpropagation algorithm to make the generated noise (i.e., the first anti-learning interference noise) meet the following goals: minimizing learnability, that is, the noise pattern needs to be difficult to be modeled or removed by a third-party neural network; retaining signal availability, that is, the signal after adding noise still needs to restore the original information at the receiving end through reverse processing (such as a noise separation network).

[0070] After the above-mentioned first noise generator is trained and deployed in the signal transmitting device, during the actual application stage, the electromagnetic signal (in analog or digital form) obtained through the above-mentioned step S210 is input into the first noise generator; the trained first noise generator dynamically generates matching anti-learning interference noise according to the statistical characteristics of the input signal (such as spectral distribution, time-domain waveform); the noise and the signal are fused in multiple network levels to form a noise-interfered electromagnetic signal. In the finally generated noise-interfered electromagnetic signal, the noise masks the original signal characteristics, reducing the risk of being wiretapped or actively interfered; even if the attacker knows the noise injection mechanism, it is difficult to eliminate the influence of the noise through reverse training because the noise is deeply bound to the signal, thus realizing anti-deep learning attack.

[0071] Step S230, transmit the noise-interfered electromagnetic signal to the signal receiving device.

[0072] Among them, the noise-interfered electromagnetic signal generated through the above-mentioned step S220 is transmitted to the signal receiving device through the physical channel. The signal receiving device uses the neural network paired with the noise generator to reversely separate the dynamic noise and restore the original signal. At the same time, through dynamic key negotiation and physical layer authentication, the transmission security is ensured, and finally, high-reliability and low-intercept-probability electromagnetic signal transmission is realized in a complex electromagnetic environment.

[0073] In the related art, the transmission system is vulnerable to attacks, and the signal data is easily used for unauthorized model training to generate attack samples, reducing the model performance and leaving attack vulnerabilities. Therefore, there is a problem threatening the secure and reliable transmission of signals.

[0074] In contrast, through the above-mentioned electromagnetic signal transmission method in the embodiment of the present application, via the first noise generator, the first anti-learning interference noise is added to the electromagnetic signal, and the noise will be input through multiple levels and achieve multi-scale fusion with the electromagnetic signal, so as to retain information as much as possible, realize dynamic injection of multi-level anti-learning interference noise, and the generated anti-learning interference noise combined with the original features can effectively prevent the data from being attacked and stolen to generate attack samples, and can effectively enhance the anti-interference and anti-wiretapping capabilities of the original electromagnetic signal, realize the secure transmission of electromagnetic signals, and solve the problem of low security of electromagnetic signal transmission.

[0075] In some of these embodiments, the above-mentioned signal receiving device is deployed with a first interference remover for performing noise removal processing on the noise-interfered electromagnetic signal. The training process of the first noise generator and the first interference remover may further include the following steps:

[0076] Input the first training data into the first autoencoder model; wherein, the initial neural network model is set as the first autoencoder model. The specific structures of the first encoder and the first decoder in the first autoencoder model may not be limited; the first encoder is responsible for abstracting the input data (such as electromagnetic signals) layer by layer into high-level features (latent representations), and the first decoder is responsible for reconstructing the original data from the high-level features. Additionally, the above first training data may be a data set constructed based on the collected signal samples. All signal samples are collected from ultra-short wave radio stations, and the acquisition conditions are set as direct wave and diffraction; in this embodiment, preprocessing is performed on the IQ signals in the samples. First, for an IQ signal with a length of L, the component and the component matrix are concatenated to obtain the input signal :

[0077] X = concat(I, Q);

[0078] where and the component sizes are (1, L), and after concatenation, the size is (2, L). Then, is divided into K segments, each segment with a size of (2, N), satisfying the relationship L = K × N. The signal segments are flattened into a size of (1, 2 N); the divided segments are recombined into the input matrix X K,2N , which is used as the above first training data. This operation can save computing resources and simultaneously explore the correlations between different time series segments.

[0079] Through the first encoder network of the first autoencoder model, perform multi-layer feature extraction processing on the first training data to obtain hierarchical feature data corresponding to each network layer in the first encoder network, and at each network layer in the first encoder network, fuse the injected first preset interference noise with the hierarchical feature data to output the first fused training feature.

[0080] Specifically, please refer to Figure 3 , the input data (the first training data) passes through the multi-layer convolutional network of the first encoder to extract features layer by layer. At each encoder layer, the preset interference noise is fused with the feature map of the current layer. The fusion methods may include: element-wise addition, directly adding the noise to the feature map to simulate the additive interference in the channel; channel concatenation, concatenating the noise as an additional channel with the feature map to increase the feature dimension; or, an attention mechanism can also be introduced to dynamically adjust the influence weight of the noise on the features through learnable parameters. The fused features (the first fused training features) contain both the original signal features and the dynamic noise and are used as the input to the decoder.

[0081] Next, the first fused training feature is input into the first decoder network of the first autoencoder model for decoding to obtain the first reconstructed feature data; based on the first training data and the first reconstructed feature data, the result of the first reconstruction loss function is calculated, and the gradient of the result of the first reconstruction loss function is backpropagated to the first autoencoder model for iterative training to obtain the first target encoder model. Among them, the first decoder receives the fused feature, attempts to restore the original input data from it, and outputs the first reconstructed feature data. This process forces the decoder to learn the function of the first interference remover, that is, to extract effective information from the noise-polluted feature. By comparing the reconstructed data (the first reconstructed feature data) with the original data (the first training data), the reconstruction loss (such as mean square error MSE or cross-entropy loss) is calculated. The gradient of the reconstruction loss function is backpropagated to the entire autoencoder, and the weights of the first encoder and the first decoder are updated through an optimizer (such as Adaptive Moment Estimation Adam) for iterative training until the iterative stop condition of reaching the number of iterative training times, the duration of iterative training, or the model tending to converge is satisfied. At this time, the iteration is stopped and the trained first target encoder model is generated.

[0082] The above first target encoder model includes a first noise generator (i.e., the trained first encoder) and a first interference remover (i.e., the trained first decoder). At this time, the trained first noise generator and the first interference remover are respectively deployed at the signal sending device end and the signal receiving device end.

[0083] Through the above embodiments, by performing iterative training on the first autoencoder model, the noise generator and the interference remover are jointly trained, so that the model complexity can be reduced while ensuring performance.

[0084] In some of these embodiments, the above steps of calculating the result of the first reconstruction loss function based on the first training data and the first reconstructed feature data, and backpropagating the gradient of the result of the first reconstruction loss function to the first autoencoder model for iterative training to obtain the first target encoder model may further include the following steps:

[0085] In the first training stage, calculate the result of the first reconstruction loss function; in the second training stage, input the preset stage perturbation into the first autoencoder model. Using the first autoencoder model, fuse the first preset interference noise, the stage perturbation, and the hierarchical feature data, output the second fused training feature, and input the second fused training feature into the first decoder network for decoding to obtain the second reconstructed feature data; calculate the result of the second reconstruction loss function according to the first training data and the second reconstructed feature data; backpropagate the gradients of the result of the first reconstruction loss function and the result of the second reconstruction loss function to the first autoencoder model for iterative training to obtain the first target encoder model.

[0086] Specifically, for the above-mentioned first autoencoder model, the training adopts a two-stage iterative training framework; obtain training data from the training dataset, add anti-learning interference noise and then perform signal encoding in the network, obtain the reconstructed signal through reconstruction, calculate the reconstruction loss according to the reconstructed signal and the input signal, obtain the classification result through classification, and calculate the classification loss according to the classification result and the classification label for calculating the gradient, and use the projected gradient descent method as the optimization algorithm.

[0087] Among them, in the two-stage iterative training, a batch of data is sampled from the training dataset in each round of training, and the perturbation starting point is initialized ; in each round of training, the first-stage training iteration is repeated times, and then the second-stage training iteration is repeated times, and then the model parameters are updated. Each round of iterative training is divided into:

[0088] In the first training stage, each iteration includes adding anti-learning interference noise, calculating the gradient, updating the direction, and projecting back to the constraint space, and its goal is to minimize the learnable information of the model after adding anti-learning interference noise, as shown in the following formula:

[0089] ;

[0090] In the above formula, is used to represent the calculation formula of the first reconstruction loss function; in this step, it can be the MSE loss or other loss calculation methods. X represents the input first training data. represents the added first preset interference noise, represents the noise 's maximum allowable amplitude. represents the first autoencoder model 's predicted output for the attacked input

[0091] In the second training stage, each iteration includes adding perturbations, calculating gradients, updating directions, and projecting back to the constraint space. The goal is to enhance the model's ability through adversarial learning, learn as much information as possible from the data, and eliminate the influence of noise. In other words, the preset stage perturbations and the first preset interference noise are input into the first autoencoder model. These perturbations and noises can be random or specifically designed to simulate various interferences that may be encountered in practical applications. In the first autoencoder model, the stage perturbations, the first preset interference noise, and the hierarchical feature data extracted in the first training stage are fused to generate the second fused training features. This step aims to enable the model to learn how to extract effective features in the presence of perturbations and noises. Then, the second fused training features are input into the first decoder network for decoding to obtain the second reconstructed feature data; according to the first training data and the second reconstructed feature data, the result of the second reconstruction loss function is calculated. The loss function in this second training stage, similar to that in the first training stage, also measures the difference between the reconstructed data and the original data, but it is calculated in the presence of perturbations and noises.

[0092] In the above two training stages, the gradients of the first reconstruction loss function and the second reconstruction loss function with respect to the model parameters are calculated respectively. These two-stage gradients are backpropagated into the first autoencoder model. This means that the model parameters will be updated according to the gradients of these two loss functions. By iterating the above process multiple times (i.e., repeating the operations of the first training stage and the second training stage, and calculating the losses and gradients), the parameters of the first autoencoder model are continuously optimized. Finally, when the model meets the preset stopping conditions (such as the loss function value no longer significantly decreases, the number of iterations reaches the upper limit, etc.), the first target encoder model is obtained.

[0093] Through the above embodiments, iterative training is carried out in two stages. The basic reconstruction ability is established in the first stage, and perturbations and noises are introduced in the second stage to enhance the model's robustness. Combining the gradient backpropagation of the dual loss functions enables the model to significantly improve its adaptability and generalization performance to complex interferences while maintaining the feature extraction accuracy.

[0094] It should also be noted that in the second training stage, the second reconstruction loss function can be calculated based on the first training data and the second reconstructed feature data, and the gradient of the calculated result of the second reconstruction loss function is backpropagated to the model for iterative training. Or, in another embodiment, stage training can also be combined with the classification results. Specifically, the above first training data carries classification labels, and at this time, the first training data is another batch of data with classification labels sampled from the above signal sample dataset; the process includes the following steps:

[0095] Using the first autoencoder model, the first preset interference noise, phase perturbation, and hierarchical feature data are fused to output the second fused training feature, and classification processing is performed based on the second fused training feature to output a classification result. Among them, please refer to Figure 3 In the second training stage, the input feature (i.e., the first training data carrying the classification label mentioned above) is input into each convolutional network layer of the first encoder of the first autoencoder model, and after multi-scale fusion with the first preset interference noise injected into each convolutional network layer, the fused feature is input into the batch normalization layer for normalization processing to obtain the processed second fused training feature. Subsequently, the second fused training feature is input into the fully connected layer serving as the classification head for classification processing, or it is input into the auxiliary classification network for classification processing to obtain the classification result for the electromagnetic signal.

[0096] Next, based on the classification result and the classification label, the result of the classification loss function is calculated, and based on the result of the classification loss function and the result of the second reconstruction loss function, the result of the joint loss function is calculated; the gradients of the first reconstruction loss function result and the joint loss function result are backpropagated to the first autoencoder model for iterative training to obtain the first target encoder model.

[0097] More specifically, at this time, as shown in the following formula:

[0098] ;

[0099] Among them, is the calculation formula of the second reconstruction loss function. represents the added phase perturbation, represents the perturbation 's maximum allowable amplitude. represents the first autoencoder model 's predicted output for the attacked input .

[0100] Through the above embodiments, by combining the reconstruction loss and the classification loss for iterative training of the first autoencoder model, compared with the training method that only uses the reconstruction loss constraint, using the idea of multi-task training can further constrain the model training and improve the model performance. At the same time, only introducing the classification loss in the second training stage, using unlabeled data to train the model weights to a certain extent first, and then using labeled data to add the classification loss to further constrain the training, can effectively reduce the label dependence and solve the problem of limited labeled data.

[0101] For ease of understanding, the training process of the first target encoder model is described below with reference to the accompanying drawings. Please refer to Figure 4 , and this process includes the following steps:

[0102] Step S401, input the first training data.

[0103] Step S402, initialize the model parameters, including: learning rate , the PGD parameters in the first training stage (perturbation range , step size , number of iterations ), the PGD parameters in the second training stage (perturbation range , step size , number of iterations ).

[0104] Step S403, determine whether it has been trained for m rounds currently. If so, end the training and generate the first target encoder model.

[0105] Step S404, if the judgment result in the above step S403 is no, perform data sampling and initialize the perturbation.

[0106] Step S405, execute the first training stage.

[0107] Step S406, execute the second training stage.

[0108] Step S407, calculate the gradient, update the parameters, and return to the above step S403 to determine whether to stop training.

[0109] In some of the embodiments, after obtaining the electromagnetic signal to be transmitted, the electromagnetic signal transmission method further includes the following steps:

[0110] Perform semantic feature extraction processing on the electromagnetic signal to obtain a semantic feature vector. Among them, through technical means such as signal preprocessing, multi-domain feature fusion, and deep learning coding, high-level semantic information reflecting its essential attributes is parsed from the original electromagnetic signal and converted into a structured vector representation for subsequent signal classification, recognition, or decision-making tasks.

[0111] Input the semantic feature vector into the trained second noise generator, and perform noise addition processing on the electromagnetic signal based on the second anti-learning interference noise carried by the second noise generator to generate a noise interference semantic vector; among them, the second noise generator is generated by inputting the obtained second training data into the initial neural network model and injecting the second preset interference noise into each network layer of the initial neural network model, and performing multi-level fusion processing of the second preset interference noise and the second training data through each network layer and iterative training; the second anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training process of the initial neural network model; transmit the noise interference semantic vector to the signal receiving device.

[0112] It should be understood that the training process of the above-mentioned second noise generator is similar to that of the first noise generator, except that the training data input during the training process is different for the two. Among them, the first training data for training the first noise generator is electromagnetic signal data, while the second training data for training the second noise generator is the training data obtained by converting the electromagnetic signal data into the semantic feature vector Feat; that is, the training data set containing the second training data input into the training framework , represents the i-th training data extracted as the semantic feature vector, and N represents the number of samples in the data set.

[0113] Through the above embodiments, the electromagnetic signal is extracted as a high-dimensional semantic feature and used to replace the original signal for transmission. In this way, only the local features related to the signal and the task need to be transmitted each time, avoiding information redundancy and excessive resource occupation, and greatly improving the transmission efficiency of the communication system; at the same time, through the transmission of semantic features, the data information is projected from the low-dimensional space to the high-dimensional space, which is an effective data encryption method to protect sensitive information from being stolen, thereby further improving the security of signal transmission.

[0114] In some of the embodiments, the above-mentioned semantic feature extraction process for the electromagnetic signal to obtain the semantic feature vector includes:

[0115] The electromagnetic signal is input into the trained signal encoder for semantic feature extraction processing to obtain the semantic feature vector. Specifically, first, the original electromagnetic signal is framed, windowed, and denoised (such as wavelet packet transform or adaptive filtering) to suppress spectral leakage and enhance the information of key frequency bands. Subsequently, the signal is converted into a time-frequency spectrogram (such as a two-dimensional time-frequency matrix generated by short-time Fourier transform) as the input representation of the deep learning model. Finally, the trained signal encoder (such as a convolutional neural network or an autoencoder) automatically extracts abstract features through multi-layer non-linear transformation.

[0116] Through the above embodiments, through the end-to-end training of the deep learning model, the traditional time-frequency analysis and semantic modeling are integrated into a unified framework, significantly improving the automation degree and generalization ability of feature extraction.

[0117] In some of the embodiments, the above-mentioned signal receiving device is deployed with a signal decoder for feature restoration of the noise interference semantic vector; the training process of the above-mentioned signal encoder and signal decoder includes:

[0118] Obtain the third training data; among them, the third training data can use the same batch of electromagnetic signal data as the above-mentioned first training data, or different batches of training data.

[0119] Next, input the third training data into the second encoder network in the second autoencoder model, and perform encoding processing on the third training data to obtain training semantic feature data; input the training semantic feature data into the second decoder network in the second autoencoder model, and perform reconstruction processing on the training semantic feature data to obtain training reconstruction data; based on the third training data and the training reconstruction data, calculate the result of the semantic feature loss function; backpropagate the gradient of the semantic feature loss function result to the second autoencoder model for iterative training, and generate the second target encoder model; the second target encoder model includes a signal encoder and a signal decoder.

[0120] Specifically, input the third training data into the second encoder network of the second autoencoder model, extract high-order semantic features through non-linear mapping to generate training semantic feature data; then input the feature data into the second decoder network, and use deconvolution or transposed convolution operations to reconstruct the signal and output the training reconstruction data; based on the difference between the original third training data and the reconstruction data (such as mean square error or cross-entropy loss), calculate the value of the semantic feature loss function; through the gradient backpropagation algorithm, backpropagate the loss gradient to the encoder and decoder networks to synchronously optimize their parameters; after multiple rounds of iterative training, the model converges to the second target encoder model, whose signal encoder can accurately capture the essential semantic information of the electromagnetic signal, and the signal decoder has a high-fidelity reconstruction ability, forming an end-to-end semantic feature extraction and reconstruction system as a whole.

[0121] Please refer to Figure 5 , the input data (the third training data) passes through the multi-layer convolutional network of the second encoder, layer by layer extracts features from low-level (such as edges, textures) to high-level (such as semantic information), and outputs the training semantic feature data. This training semantic feature data is then input into the multi-layer deconvolution network of the second decoder for layer-by-layer deconvolution operations to output the training reconstruction data. Calculate the loss function based on the input third training data and the output training reconstruction data to minimize the difference between the input signal and the reconstruction result, and achieve iterative training.

[0122] Through the above embodiments, by performing iterative training on the second autoencoder model and jointly training the signal encoder and the signal decoder, the encoder and the decoder can promote each other and optimize synergistically during the training process. Specifically, under the feedback of the decoder, the encoder can learn more accurate and discriminative semantic feature representations, which can better capture the essential attributes of the electromagnetic signal; at the same time, under the feature input of the encoder, the decoder can continuously improve the accuracy and fidelity of signal reconstruction, minimizing the difference between the reconstructed signal and the original signal. This way of joint training not only helps to improve the performance of the entire autoencoder model, but also enables the model to exhibit stronger robustness and generalization ability in electromagnetic signal processing tasks.

[0123] In some of these embodiments, the above-mentioned transmission of the noise-interfered electromagnetic signal to the signal receiving device may further include the following steps:

[0124] Obtain a preset random encryption seed; based on the random encryption seed, perform a channel dynamic sorting operation on the noise-interfered electromagnetic signal to obtain a random encrypted signal, and transmit the random encrypted signal to the signal receiving device.

[0125] Among them, a random encryption seed with high entropy value is pre-generated (such as a pseudo-random sequence generated based on a linear feedback shift register), and this seed determines the initial state and dynamic change law of the subsequent channel sorting algorithm. The seed can be distributed to both communication parties through an encryption protocol (such as a key exchange algorithm) to ensure synchronization and security. In an electromagnetic environment with noise interference, the channel priority of signal transmission is dynamically adjusted according to the random encryption seed. For example, when adopting a dual-channel architecture, the pseudo-random sequence generated by the seed controls the switching timing of the main / backup channels, so that the noise-interfered signals are randomly distributed among different channels. Specifically, by real-time monitoring the channel quality (such as signal-to-noise ratio, bit error rate), and combining the sorting rules defined by the seed (such as polling, weighted random allocation), the channels severely affected by interference are temporarily blocked or their weights are reduced, forming a dynamically adaptive anti-interference channel matrix. After channel sorting, the transmitted signal is embedded in the randomized transmission path determined by the seed to form a randomly encrypted signal with physical layer encryption. The signal receiving device restores the channel sorting rule through the synchronized seed and accurately extracts the effective signal, while illegal eavesdroppers are difficult to analyze the signal path due to the lack of seed information, thus preventing data leakage or tampering.

[0126] More specifically, a trained random encryptor and an inverse random decryptor are respectively deployed at the signal sending device and the signal receiving device. The input of the random encryptor F is the electromagnetic signal with added noise output by the first noise generator, or the semantic feature with added noise output by the second signal noise generator, and the output is concatenated with the random encryption seed and sent through the signal sending unit. In this embodiment, the random transmitter further encrypts the signal by using the random channel shuffling method. For the semantic feature signal with added anti-learning interference noise, its channel dimension is randomly shuffled, and the random encryption seed is . Then the final signal to be sent Signal is:

[0127] Signal = [shuffle(x; r), r];

[0128] In the above formula, x represents the input signal input to the random encryptor, and shuffle(·) represents randomly shuffling the channel dimension of the original semantic feature signal, and its shuffling rule is controlled by the random encryption seed r.

[0129] Through the above embodiments, a dynamic sorting rule is generated by obtaining a preset random encryption seed, and channel rearrangement encryption is performed on the electromagnetic signal carrying noise interference, so that the signal presents a randomized spectrum distribution characteristic during the transmission process. Finally, the randomly encrypted signal that is difficult to be parsed by the eavesdropper is transmitted to the signal receiving device, thereby realizing the concealment of the true transmission path of the signal through the channel switching mechanism driven by the random encryption seed, and effectively improving the anti-interception and anti-interference capabilities of the electromagnetic signal.

[0130] It should be noted that the steps shown in the above process or the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0131] This embodiment also provides an electromagnetic signal transmission system, including a signal sending device and a signal receiving device; the signal sending device is communicatively connected to the signal receiving device and is used to execute the electromagnetic signal transmission method described in any of the above embodiments.

[0132] The present application will be described in detail below in conjunction with specific embodiments. Figure 6 is a schematic diagram of an electromagnetic signal transmission system architecture according to an embodiment of the present application, as Figure 6 shown, the system includes a signal sending device and a signal receiving device.

[0133] The signal sending device includes a data pre-processor, a signal encoder, a second noise generator, a random encryptor, and a signal transmitter. Among them, the data pre-processor is used to pre-process the electromagnetic signal data to meet the input requirements of the transmission system; the signal encoder is used to extract the feature representation of the electromagnetic signal and obtain the high-dimensional semantic feature vector of the signal; the second noise generator is used to generate an anti-learning interference noise vector and perform multi-scale fusion with the semantic features; the random encryptor is used to further randomly encrypt the output result of the noise generator according to the random seed, and splice the encryption result with the encryption seed to obtain the complete signal to be sent; the signal transmitter is used to send the signal. In this way, the electromagnetic signal and the random encryption seed are input into the signal sending device, and are processed and signal sending operations are performed through each module in the signal sending device to realize signal transmission.

[0134] The signal receiving device includes a signal receiver, an inverse random decryptor, a second interference remover, and a signal decoder. Among them, the signal receiver is used to receive the signal; the inverse random decryptor is used to decode the random encryption seed to realize the reverse random encryption process; the second interference remover is used to exclude the interference information in the received signal; the signal decoder is used to restore the information and finally obtain the output signal.

[0135] It is understandable that the above signal transmitter and signal receiver can adopt the transmitters and receivers commonly used in communication networks.

[0136] Furthermore, for the above electromagnetic signal transmission system architecture, its construction process can be referred to Figure 7 , and this process includes the following steps:

[0137] Step S701, collect samples and construct a signal data set.

[0138] Step S702, build an electromagnetic signal secure transmission system.

[0139] Step S703, train the signal encoder and the signal decoder.

[0140] Step S704, train the second noise generator and the second interference eliminator.

[0141] Step S705, complete the training and perform signal transmission.

[0142] This embodiment also provides a computer device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0143] Optionally, the above computer device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0144] Optionally, in this embodiment, the above processor may be configured to execute the following steps through a computer program:

[0145] S1, obtain the electromagnetic signal to be transmitted.

[0146] S2, input the electromagnetic signal into the trained first noise generator, and perform noise addition processing on the electromagnetic signal based on the first anti-learning interference noise carried by the first noise generator to generate a noise-interfered electromagnetic signal; wherein, the first noise generator is generated by inputting the obtained first training data into an initial neural network model, injecting the first preset interference noise into each network layer of the initial neural network model, performing multi-level fusion processing on the first preset interference noise and the first training data through each network layer, and iteratively training; the first anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training process of the initial neural network model.

[0147] S3, transmit the noise-interfered electromagnetic signal to the signal receiving device.

[0148] It should be noted that for the specific examples in this embodiment, reference may be made to the examples described in the above embodiments and optional implementation manners, and details thereof will not be elaborated herein.

[0149] In addition, in combination with the electromagnetic signal transmission method in the above embodiments, an embodiment of the present application can provide a storage medium to implement it. A computer program is stored on the storage medium; when the computer program is executed by a processor, any one of the electromagnetic signal transmission methods in the above embodiments is implemented.

[0150] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties.

[0151] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0152] Those skilled in the art should understand that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0153] The above embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. An electromagnetic signal transmission method, characterized in that, Applied to a signal transmitting device, the method includes: Obtain an electromagnetic signal to be transmitted; Input the electromagnetic signal into a trained first noise generator, and perform noise addition processing on the electromagnetic signal based on the first anti-learning interference noise carried by the first noise generator to generate a noise-interfered electromagnetic signal; Wherein, the first noise generator is generated by inputting the obtained first training data into an initial neural network model, injecting a first preset interference noise into each network layer of the initial neural network model, performing multi-level fusion processing on the first preset interference noise and the first training data through each network layer, and iteratively training; the first anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training process of the initial neural network model; Transmit the noise-interfered electromagnetic signal to a signal receiving device.

2. The electromagnetic signal transmission method according to claim 1, characterized in that, The signal receiving device is deployed with a first interference remover for performing noise removal processing on the noise-interfered electromagnetic signal. The training processes of the first noise generator and the first interference remover include: Input the first training data into a first autoencoder model; wherein, the initial neural network model is set as the first autoencoder model; Perform multi-level feature extraction processing on the first training data through the first encoder network of the first autoencoder model to obtain level feature data corresponding to each network layer in the first encoder network, and at each network layer in the first encoder network, fuse the injected first preset interference noise with the level feature data to output a first fused training feature; Input the first fused training feature into the first decoder network of the first autoencoder model for decoding processing to obtain first reconstructed feature data; Based on the first training data and the first reconstructed feature data, calculate the result of a first reconstruction loss function, and backpropagate the gradient of the result of the first reconstruction loss function to the first autoencoder model for iterative training to obtain a first target encoder model; the first target encoder model includes the first noise generator and the first interference remover.

3. The electromagnetic signal transmission method according to claim 2, wherein The calculating the result of the first reconstruction loss function based on the first training data and the first reconstructed feature data, and backpropagating the gradient of the result of the first reconstruction loss function to the first autoencoder model for iterative training to obtain a first target encoder model includes: In a first training stage, calculate the result of the first reconstruction loss function; In a second training stage, input a preset stage perturbation into the first autoencoder model, use the first autoencoder model to fuse the first preset interference noise, the stage perturbation and the level feature data, output a second fused training feature, and input the second fused training feature into the first decoder network for decoding processing to obtain second reconstructed feature data; Calculate the result of the second reconstruction loss function according to the first training data and the second reconstructed feature data; Backpropagate the gradients of the first reconstruction loss function result and the second reconstruction loss function result to the first autoencoder model for iterative training to obtain the first target encoder model.

4. The electromagnetic signal transmission method according to claim 3, wherein The first training data carries classification labels; using the first autoencoder model to fuse the first preset interference noise, the stage perturbation and the hierarchical feature data, and output the second fused training feature, including: Using the first autoencoder model to fuse the first preset interference noise, the stage perturbation and the hierarchical feature data, output the second fused training feature, and perform classification processing based on the second fused training feature to output a classification result; The step of backpropagating the gradients of the first reconstruction loss function result and the second reconstruction loss function result to the first autoencoder model for iterative training to obtain the first target encoder model includes: Calculate the result of the classification loss function based on the classification result and the classification label, and calculate the result of the joint loss function according to the classification loss function result and the second reconstruction loss function result; Backpropagate the gradients of the first reconstruction loss function result and the joint loss function result to the first autoencoder model for iterative training to obtain the first target encoder model.

5. The electromagnetic signal transmission method according to claim 1, wherein After obtaining the electromagnetic signal to be transmitted, the method further includes: Performing semantic feature extraction processing on the electromagnetic signal to obtain a semantic feature vector; Input the semantic feature vector into the trained second noise generator, and add noise to the electromagnetic signal based on the second anti-learning interference noise carried by the second noise generator to generate a noise interference semantic vector; Wherein, the second noise generator is generated by inputting the obtained second training data into the initial neural network model, injecting the second preset interference noise into each network layer of the initial neural network model, performing multi-level fusion processing of the second preset interference noise and the second training data through each network layer, and performing iterative training; the second anti-learning interference noise is obtained by minimizing the first preset interference noise during the iterative training of the initial neural network model; Transmit the noise interference semantic vector to the signal receiving device.

6. The electromagnetic signal transmission method according to claim 5, wherein The step of performing semantic feature extraction processing on the electromagnetic signal to obtain a semantic feature vector includes: Input the electromagnetic signal into the trained signal encoder for semantic feature extraction processing to obtain the semantic feature vector.

7. The electromagnetic signal transmission method according to claim 6, wherein The signal receiving device is deployed with a signal decoder for feature restoration of the noise interference semantic vector; the training process of the signal encoder and the signal decoder includes: Obtain the third training data; Input the third training data into the second encoder network in the second autoencoder model, and perform encoding processing on the third training data to obtain training semantic feature data; Input the training semantic feature data into the second decoder network in the second autoencoder model, and perform reconstruction processing on the training semantic feature data to obtain training reconstruction data; Calculate the semantic feature loss function result based on the third training data and the training reconstruction data; Backpropagate the gradient of the semantic feature loss function result to the second autoencoder model for iterative training, and generate a second target encoder model; the second target encoder model includes the signal encoder and the signal decoder.

8. The electromagnetic signal transmission method according to any one of claims 1 to 7, characterized in that, The step of transmitting the noise-interfered electromagnetic signal to the signal receiving device includes: Obtain a preset random encryption seed; Based on the random encryption seed, perform a channel dynamic sorting operation on the noise-interfered electromagnetic signal to obtain a random encrypted signal, and transmit the random encrypted signal to the signal receiving device.

9. An electromagnetic signal transmission system, characterized in that, It includes a signal transmitting claim device and a signal receiving device; The signal transmitting device is communicatively connected to the signal receiving device and is configured to execute the electromagnetic signal transmission method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the electromagnetic signal transmission method according to any one of claims 1 to 8.

Images