A generative ai-based satellite semantic communication system

Abstract

This invention discloses a satellite semantic communication system based on generative AI, comprising: first, establishing a satellite semantic communication system based on generative AI; second, setting parameters for the satellite semantic communication system; third, training the satellite semantic communication system; and finally, completing the construction of the satellite semantic communication system and commencing communication. This invention proposes a satellite semantic communication system based on generative AI that can map user image data from a bit space to a more concise semantic space. While ensuring semantic information consistency, it reduces the amount of data transmitted, improves spectrum and power resource utilization, and enhances transmission efficiency. Furthermore, even with low satellite communication channel quality, using generative AI to generate bit data from the semantic space can effectively resist noise interference, improving the transmission quality and success rate of image data.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a satellite semantic communication system based on generative AI. Background Technology

[0002] Satellite communication systems suffer from poor link quality, significant path loss, and highly variable channel quality under different weather conditions, especially poor quality under rain attenuation, leading to unreliable communication quality and sometimes only the transmission of voice or SMS messages. Semantic communication, as an emerging communication theory, offers several advantages. It focuses on the content and meaning of information, not just the bit stream. This allows for more efficient information transmission, reduced redundancy, and improved communication efficiency. By transmitting only key information and semantic content, semantic communication can achieve effective information delivery under lower bandwidth conditions. This is particularly important in bandwidth-constrained environments. Semantic communication is better resistant to noise and interference because it focuses on the meaning of information, not the specific bit string. This characteristic improves communication quality in noisy environments. Semantic communication enhances information understanding between humans and machines, improving the quality of interaction between devices and users.

[0003] Chinese patent document CN202411509802.5 proposes a satellite semantic communication system for speech-to-text conversion, but it does not address the field of image transmission.

[0004] Chinese patent document CN202510074433.X proposes a lightweight image semantic communication method, which uses a deep learning model to extract the texture semantics of the source image and a traditional visual algorithm to extract the color semantics of the source image, but does not use generative AI technology for satellite semantic communication. Summary of the Invention

[0005] Traditional satellite communication systems face challenges such as long channel links, high path loss, and poor channel quality, leading to unreliable communication quality and sometimes limiting transmission to voice or SMS messages. This invention proposes a satellite semantic communication system based on generative AI. This solution maps user image data from a bit space to a more concise semantic space, eliminating redundant bits and improving the transmission efficiency of the satellite communication system. Furthermore, it effectively utilizes contextual information to resist noise interference, improving the transmission quality and success rate of image data, thus promoting the intelligent development of satellite communication in the 6G era.

[0006] The technical solution of this invention is as follows: A satellite semantic communication system based on generative AI, comprising: S1, Establish a satellite semantic communication system based on generative AI; S11 establishes the communication architecture of a satellite semantic communication system for generative AI, including a transmitter semantic encoder, a channel, a receiver semantic decoder, and a semantic discriminator. S12, Transmitter Semantic Encoder: Composed of a CNN convolutional neural network and a Transformer Encoder neural network, used to map images from bit space to semantic space; Where S represents the source data. X is the semantic encoder at the transmitting end, and X is the encoded semantic data; Convert the integer part of X to the real part and the fractional part to the imaginary part to obtain the complex number: Based on the power constraint, Z is normalized: Where k is the signal length and P is the transmission power.

[0007] S13, Channel: The channel is a Gaussian channel, a Ricean channel, or a channel used in actual satellite communication; Where Y is the received signal, H is the channel model, and N is Gaussian noise.

[0008] S14, receiver semantic decoder, consists of Unet neural network and Transformer Decoder neural network; in For the Unet neural network, For Transformer Decoder neural networks, For the restoration of the faith.

[0009] S15, Semantic Discriminator: Semantic Discriminator It can be a large AI model or other task-oriented specialized intelligent agent, which uses a semantic discriminator to determine the consistency of the semantics of the transmitted data. S2, set the parameters of the satellite semantic communication system; S3, training satellite semantic communication system; S31, determine the training set to be used as the information source, such as the CIFAR10 dataset or the COCO dataset; S32, set the parameters of the transmitter semantic encoder, including the number of input channels, the number of output channels, the kernel size, the stride of the CNN neural network, the input dimension, the output dimension, the number of layers, and the number of attention heads of the Transformer Encoder neural network; The parameters of the receiver semantic decoder are set, including the number of input channels, the number of output channels, the number of upsampling and downsampling layers, and the number of channels per sampling layer of the Unet neural network; and the input dimension, output dimension, number of layers, and number of attention heads of the Transformer Decoder neural network. Setting training hyperparameters includes learning rate, batch size, and number of training epochs; S33, the loss function is defined as follows: in, During training, a value is randomly sampled in the interval [0,1] as t. for linear interpolation, This is a Unet neural network whose output values ​​are used to approximate... Through the loss function Unet neural networks can be trained for noise reduction; S34, after training is complete Then, calculate from Y through m iterations. : S35, Reconstruct Transmitted Data: in, For mean square error loss, For divergence, Let X be the mean and variance, respectively. The loss follows a normal distribution. This loss indicates that, for the semantic discriminator, the transmitted data and the generated received data have the same semantics.

[0010] S36, take different data from the dataset, transmit images under different signal-to-noise ratio conditions, and calculate the loss function; S37 uses a stochastic gradient descent algorithm to jointly optimize the transmitter and receiver neural networks until the number of iterations exceeds epochs or the loss function stabilizes.

[0011] S4, complete the construction of the satellite semantic communication system and carry out communication.

[0012] The advantages of this invention compared to existing technologies are as follows: This invention proposes a satellite semantic communication system based on generative AI, which can map user image data from bit space to a more concise semantic space, thereby eliminating redundant bit data. Under the condition of ensuring semantic information consistency, it reduces the amount of data transmitted, improves the utilization rate of spectrum and power resources, and improves the transmission efficiency of satellite communication systems. Furthermore, when the satellite communication channel quality is low, using generative AI to generate bit data from the semantic space can effectively resist noise interference, improve the transmission quality and success rate of image data, and promote the intelligent development of satellite communication in the 6G era. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the implementation of the satellite semantic communication system based on generative AI in this invention. Detailed Implementation

[0014] This invention proposes a satellite semantic communication system based on generative AI.

[0015] like Figure 1 As shown, this invention describes a satellite semantic communication method based on generative AI, with the following specific steps: S1, Establish a satellite semantic communication system based on generative AI; S11 establishes the communication architecture of a satellite semantic communication system for generative AI, including a transmitter semantic encoder, a channel, a receiver semantic decoder, and a semantic discriminator. S12, Transmitter Semantic Encoder: Composed of a CNN convolutional neural network and a Transformer Encoder neural network, used to map images from bit space to semantic space; Where S represents the source data. X is the semantic encoder at the transmitting end, and X is the encoded semantic data; Convert the integer part of X to the real part and the fractional part to the imaginary part to obtain the complex number: Based on the power constraint, Z is normalized: Where k is the signal length and P is the transmission power.

[0016] S13, Channel: The channel is a Gaussian channel, a Ricean channel, or a channel used in actual satellite communication; Where Y is the received signal, H is the channel model, and N is Gaussian noise.

[0017] S14, receiver semantic decoder, consists of Unet neural network and Transformer Decoder neural network; in For the Unet neural network, For Transformer Decoder neural networks, For the restoration of the faith.

[0018] S15, Semantic Discriminator: Semantic Discriminator It can be a large AI model or other task-oriented specialized intelligent agent, which uses a semantic discriminator to determine the consistency of the semantics of the transmitted data. S2, set the parameters of the satellite semantic communication system; S3, training satellite semantic communication system; S31, determine the training set to be used as the information source, such as the CIFAR10 dataset or the COCO dataset; S32, set the parameters of the transmitter semantic encoder, including the number of input channels, the number of output channels, the kernel size, the stride of the CNN neural network, the input dimension, the output dimension, the number of layers, and the number of attention heads of the Transformer Encoder neural network; The parameters of the receiver semantic decoder are set, including the number of input channels, the number of output channels, the number of upsampling and downsampling layers, and the number of channels per sampling layer of the Unet neural network; and the input dimension, output dimension, number of layers, and number of attention heads of the Transformer Decoder neural network. Setting training hyperparameters includes learning rate, batch size, and number of training epochs; S33, the loss function is defined as follows: in, During training, a value is randomly sampled in the interval [0,1] as t. for linear interpolation, This is a Unet neural network whose output values ​​are used to approximate... Through the loss function Unet neural networks can be trained for noise reduction; S34, after training is complete Then, calculate from Y through m iterations. : S35, Reconstruct Transmitted Data: in, For mean square error loss, For divergence, Let X be the mean and variance, respectively. The loss follows a normal distribution. This loss indicates that, for the semantic discriminator, the transmitted data and the generated received data have the same semantics.

[0019] S36, take different data from the dataset, transmit images under different signal-to-noise ratio conditions, and calculate the loss function; S37 uses a stochastic gradient descent algorithm to jointly optimize the transmitter and receiver neural networks until the number of iterations exceeds epochs or the loss function stabilizes.

[0020] S4, complete the construction of the satellite semantic communication system and carry out communication.

[0021] This invention proposes a satellite semantic communication system based on generative AI, which can map user image data from bit space to a more concise semantic space, thereby eliminating redundant bit data. While ensuring semantic information consistency, it reduces the amount of data transmitted, improves the utilization of spectrum and power resources, and enhances the transmission efficiency of the satellite communication system. Furthermore, when the satellite communication channel quality is low, using generative AI to generate bit data from the semantic space can effectively resist noise interference and improve the transmission quality and success rate of image data.

[0022] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A satellite semantic communication system based on generative AI, characterized in that, It includes a transmitter semantic encoder, a satellite channel, a receiver semantic decoder, and a semantic discriminator; among which, The function of the transmitter semantic encoder is to map the source information S from the bit space to the semantic space to obtain semantic information. The transmitter semantic encoder includes a CNN convolutional neural network and a Transformer Encoder neural network; The function of a channel is to transmit semantic information. ; The function of the receiver semantic decoder is to decode the semantic information received by the receiver, which contains noise. The decoded information is recovered from the receiver. The receiver's semantic decoder includes a Unet neural network and a Transformer Decoder neural network. The function of the semantic discriminator is to determine whether the semantics of the recovered sink information and the source information are consistent. The transmitter semantic encoder and receiver semantic decoder of the satellite semantic communication system are trained; the trained satellite semantic communication system is then used for communication.

2. The system according to claim 1, characterized in that, The function of the transmitting semantic encoder is to map the image from the bit space to the semantic space to obtain semantic information. The specific functions of the transmitter semantic encoder are as follows: Where S represents the source information. X is the semantic encoder at the transmitter, and X is the encoded semantic information. Convert the integer part of X to the real part and the fractional part to the imaginary part to obtain the complex number Z: Based on the power constraint, Z is normalized to obtain semantic information. : Where k is the signal length and P is the transmission power.

3. The system according to claim 1, characterized in that, The function of the satellite channel is to transmit semantic information. ;in, Where Y is the received signal, H is the satellite channel model, and N is Gaussian noise; the type of satellite channel model H includes Gaussian channel, Ricean channel, or channel used in actual satellite communication.

4. The system according to claim 1, characterized in that, The function of the receiver semantic decoder is to decode the semantic information received by the receiver, which contains noise. The decoded and recovered destination information ,in,; in The Unet neural network is used to remove noise from the received signal Y to obtain the transmitted semantic information. ; For Transformer Decoder neural networks, used to... Decoding yields the recovered destination information. ; For the recovery of the recipient information.

5. The system according to claim 1, characterized in that, The specific method for training the transmitter semantic encoder and receiver semantic decoder of the satellite semantic communication system is as follows: Determine the training set to be used as the information source; Configure the parameters of the transmitter semantic encoder, including the number of input channels, the number of output channels, the kernel size, the stride of the CNN neural network, and the input dimension, output dimension, number of layers, and number of attention heads of the Transformer Encoder neural network. The parameters of the receiver semantic decoder are set, including the number of input channels, the number of output channels, the number of upsampling and downsampling layers, the number of channels per sampling layer of the Unet neural network, and the input dimension, output dimension, number of layers, and number of attention heads of the Transformer Decoder neural network. Setting training hyperparameters includes learning rate, batch size, and number of training epochs; The loss function is defined as follows: in, During training, a value is randomly sampled in the interval [0,1] as t. for linear interpolation, This is a Unet neural network whose output values ​​are used to approximate... Through the loss function To train the Unet neural network for noise reduction; After training Then, calculate from Y through m iterations. : Reconstructing transmitted data: Among them, semantic discriminator It is a large AI model, or other specialized intelligent agent for specific tasks, that uses a semantic discriminator to determine the semantic consistency of transmitted data. For mean square error loss, For divergence, Let X be the mean and variance, respectively. It follows a normal distribution. For hyperparameters, through the loss function Jointly train the semantic encoder and semantic decoder; Different data are used from the dataset, images are transmitted under different signal-to-noise ratio conditions, and the loss function is calculated; The transmitter and receiver neural networks are jointly optimized using stochastic gradient descent until the number of iterations exceeds epochs or the loss function is satisfied. If the condition is stable, then the training is complete.

Images