Edge offshore channel state transmission method and system based on low-rank adaptation
By introducing a state encoding and decoding model of low-rank adaptation algorithm and attention mechanism between user equipment and base stations, the problem of difficulty and accuracy of channel state information acquisition in offshore environments is solved, and efficient and robust channel state transmission is achieved.
Patent Information
- Application Number
- CN202411916653.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-24
- Publication Date
- 2025-04-08
AI Technical Summary
In complex offshore environments, the acquisition of channel state information is difficult and the accuracy is poor, which affects the reliability and efficiency of the communication system.
Using a transmission method based on the edge offshore channel state based on low rank adaptation, the channel matrix is obtained through the user equipment and input it into a preset state encoding model, and the channel matrix is converted into codewords using attention mechanism and low rank adaptation algorithm, and transmitted to the base station. The base station reconstructs the channel matrix through the state decoding model.
It improves the accuracy of the channel matrix, reduces the consumption of computing resources, enhances the system's adaptability and robustness to environmental changes, and supports efficient collaborative communication management in complex offshore areas.
Smart Images
Figure CN119788138A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of communication technology, and in particular to a transmission method and system based on low-rank adaptive edge near-shore channel state. Background Art
[0002] With the rapid development of contemporary wireless communication technology, the operation mode of various industries is undergoing profound changes. Especially in complex offshore areas, such as smart ports, offshore scientific detection stations, coastal resource detection areas and coastal economic corridors, efficient and reliable communication systems are particularly important. These highly automated environments need to process a large amount of control information and multi-band data transmission, which puts higher requirements on communication technology.
[0003] However, in these complex environments, channel state information will be affected by the complex environment, which will increase the difficulty of obtaining channel state information and affect the accuracy of the obtained channel state information. For example, the frequently changing ocean climate will cause the characteristics of the line signal propagation path to change frequently, affecting the accuracy of the channel state information; various physical obstacles such as waves, tides, ships and port facilities, and electromagnetic interference will affect the propagation of wireless signals, affecting the accuracy of the channel state information, etc.
[0004] Therefore the prior art still needs to be improved and enhanced. Summary of the invention
[0005] The technical problem to be solved by the present application is to provide a transmission method and system for edge near-shore channel states based on low-rank adaptation in view of the deficiencies in the prior art.
[0006] In order to solve the above technical problems, the first aspect of the present application provides a transmission method of edge offshore channel state based on low-rank adaptation, wherein the edge offshore channel state optimization method based on low-rank adaptation technology specifically includes:
[0007] The user equipment obtains a channel matrix, and inputs the channel matrix into a preset state coding model;
[0008] The user equipment converts the channel matrix into a codeword through the state coding model, wherein the state coding model includes an attention mechanism, and the attention mechanism integrates a low-rank adaptation algorithm;
[0009] The user equipment transmits the codeword to the base station.
[0010] The transmission method of the edge near-sea channel state based on low-rank adaptation, wherein the inputting of the channel matrix into a preset state coding model specifically comprises:
[0011] Performing a two-dimensional discrete Fourier transform on the channel matrix to convert the channel matrix into a truncated matrix, wherein the matrix dimension of the truncated matrix is smaller than the matrix dimension of the channel matrix;
[0012] The truncated matrix is input into a preset state coding model.
[0013] The transmission method based on low-rank adaptive edge near-shore channel state, wherein the channel matrix is determined by the user equipment based on the pilot signal, which is expressed as:
[0014]
[0015] in, represents the channel matrix, Indicates the nth th The channel vector of subcarriers, Indicates the number of subcarriers, N t (>>1) indicates the number of transmitting antennas configured in the base station.
[0016] The transmission method of the edge near-sea channel state based on low-rank adaptation, wherein the method further comprises:
[0017] The base station inputs the received codeword into a preset state decoding model, wherein the state decoding model includes an attention mechanism, and the attention mechanism integrates a low-rank adaptation algorithm;
[0018] The reconstructed channel matrix is determined by the state decoding model.
[0019] The transmission method of the edge near-sea channel state based on low-rank adaptation, wherein the determining the reconstructed channel matrix through the state decoding model specifically includes:
[0020] Decoding the codeword by the state decoding model to obtain an output channel matrix;
[0021] An inverse discrete Fourier transform is performed on the output channel matrix to obtain a reconstructed channel matrix corresponding to the channel matrix.
[0022] The transmission method of the edge near-shore channel state based on low-rank adaptation, wherein the state encoding model includes a first convolution module, a Transformer-LoRA module and a first fully connected module cascaded in sequence; the state decoding model deployed by the base station includes a second fully connected module, a Transformer-LoRA module, a second convolution module and an activation function layer cascaded in sequence, the Transformer-LoRA module multi-head attention unit, a residual fusion unit and a feedforward network, the input and output items of the multi-head attention unit are the input items of the residual fusion unit, and the output items of the residual fusion unit are the input items of the feedforward network, wherein the multi-head attention unit introduces two low-rank matrices A and low-rank matrix B through low-rank adaptation, so that the linear transformation matrix of the multi-head attention unit is expressed as:
[0023] W=W+AB
[0024] Among them, W represents the linear transformation matrix, and A and B both represent low-rank matrices.
[0025] The transmission method of the marginal near-shore channel state based on low-rank adaptation, wherein the multi-head attention unit adopts a Transformer structure, and when fine-tuning the state encoding model and / or the state decoding model, the linear transformation matrix is kept unchanged, and the low-rank matrix A and the low-rank matrix B are updated.
[0026] The transmission method of the edge near-shore channel state based on low-rank adaptation, wherein, during the training process of the state coding model and / or the state decoding model, the low-rank matrix A is initialized with random Gaussian values, and the low-rank matrix B is initialized with zero, so that the linear transformation matrix change ΔW=BA is zero at the beginning of training, and the linear transformation matrix change ΔW is scaled during the training process.
[0027] A second aspect of the present application provides a transmission system based on low-rank adaptive edge near-shore channel state, wherein the transmission system based on low-rank adaptive edge near-shore channel state includes a user equipment and a base station;
[0028] The user equipment is used to obtain a channel matrix, input the channel matrix into a preset state coding model, convert the channel matrix into a codeword through the state coding model, and transmit the codeword to a base station, wherein the state coding model integrates an attention mechanism and low-rank adaptation.
[0029] The low-rank adaptive edge near-shore channel state transmission system, wherein the base station is used to input the received codeword into a preset state decoding model, and determine the reconstructed channel matrix through the state decoding model, wherein the state decoding model integrates the attention mechanism and low-rank adaptation.
[0030] Beneficial effects:
[0031] 1. This application provides an attention mechanism to capture and process the dynamic changes of channel states in offshore environments, thereby improving the accuracy of the channel matrix sent to the base station.
[0032] 2. This application integrates the low-rank adaptation algorithm into the attention mechanism to achieve effective compression and optimization of the channel matrix, reduce the computing resources required for processing and transmission, improve the system's adaptability and robustness to environmental changes, and provide support for efficient collaborative communication management in complex offshore areas. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0034] Figure 1 A flowchart of a transmission method for edge near-shore channel states based on low-rank adaptation provided in an embodiment of the present application.
[0035] Figure 2 This is a flow chart showing the principle of the coordination between the state encoding model and the state decoding model.
[0036] Figure 3 This is the principle block diagram of the Transformer-LoRA module.
[0037] Figure 4 A functional block diagram of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0038] The embodiment of the present application provides a transmission method and system for edge near-shore channel states based on low-rank adaptation. In order to make the purpose, technical solution and effect of the present application clearer and more specific, the present application is further described in detail with reference to the accompanying drawings and examples. 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.
[0039] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.
[0040] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.
[0041] It should be understood that the sequence numbers and sizes of the steps in this embodiment do not mean the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.
[0042] The application content is further explained below through the description of embodiments in conjunction with the accompanying drawings.
[0043] This embodiment provides a transmission method based on low-rank adaptation of the edge offshore channel state. The method can be applied to complex offshore areas. Here, a smart port is used as an example for explanation. In the application scenario of the smart port, a network system in the application scenario is constructed based on a large-scale single-area MIMO frequency division duplex network. The network system includes a base station (BS) and multiple user equipment (UE). The base station is configured with N t (>>1) Transmitting antenna, the user equipment is equipped with a receiving antenna. Among them, the network system adopts The orthogonal frequency division multiplexing (OFDM) technology of subcarriers is used to improve the spectrum utilization and multi-user processing capability of the network system. The received signal y on the subcarrier n It can be expressed as:
[0044]
[0045] in, Indicates the nth th The channel vector of subcarriers; Represents the precoding vector, which is used to optimize the signal transmission direction and power distribution to overcome signal attenuation and interference caused by complex environments; represents the data symbol transmitted in the downlink, It is nth th The additive noise of each subcarrier includes the additive noise of noise and other interference.
[0046] It can be seen that the base station needs the channel state information (CSI) of each user device to assist in generating the precoding vector, and the channel in the large-scale MIMO system will change frequently, especially in the complex offshore area, the channel will be affected by the complex environment (for example, large mobile devices, complex physical shielding and special effects of the marine environment, etc.), which will lead to the problem that it is difficult to obtain and the accuracy is difficult to guarantee. To this end, in the embodiment of the present application, the Transformer architecture is combined with the LoRA algorithm, and the state coding model and the state decoding model are constructed through the Transformer architecture and the LoRA algorithm. Then, the channel matrix is encoded by the state coding model and transmitted to the base station. The base station decodes the state decoding model to obtain the channel state information. In this way, not only can the dynamic changes of the channel state in the offshore environment be efficiently captured and processed through the self-attention mechanism in the Transformer architecture, and the accuracy of CSI feedback can be improved, but also the LoRA algorithm can be used to achieve effective compression and training of CSI data, reduce the computing resources required for processing and transmission, and improve the adaptability and robustness of the system to environmental changes.
[0047] like Figure 1 As shown, the transmission method of the edge near-sea channel state based on low-rank adaptation provided in the embodiment of the present application specifically includes:
[0048] S10. The user equipment obtains a channel matrix, and inputs the channel matrix into a preset state coding model.
[0049] Specifically, the channel matrix is obtained through channel sampling, which includes signal propagation information transmitted by multiple transmit antennas. That is, the user equipment receives signals transmitted by multiple transmit antennas, and performs channel sampling on each received signal to form a channel matrix, that is, the channel matrix includes channel state information of signal propagation information from multiple transmit antennas to receive antennas, which can be expressed as:
[0050]
[0051] in, represents the channel matrix, Indicates the nth th The channel vector of subcarriers, Indicates the number of subcarriers, N t (>>1) indicates the number of transmitting antennas configured in the base station.
[0052] Furthermore, in a large-scale MIMO system, a large number of antennas result in a huge number of CSI matrix parameters. If the acquired channel matrix (i.e., the uncompressed CSI matrix) is directly fed back, a large amount of bandwidth resources will be consumed. Therefore, after acquiring the channel matrix, the channel matrix can be compressed first and then fed back to the base station. Accordingly, the inputting of the channel matrix into a preset state coding model specifically includes:
[0053] Performing a two-dimensional discrete Fourier transform on the channel matrix to convert the channel matrix into a truncated matrix, wherein the matrix dimension of the truncated matrix is smaller than the matrix dimension of the channel matrix;
[0054] The truncated matrix is input into a preset state coding model.
[0055] Specifically, the two-dimensional discrete Fourier transform (DFT) is used to compress the channel matrix. The two-dimensional discrete Fourier transform (DFT) can be used to sparse the channel matrix in the angular delay domain. The truncation matrix is determined based on the sparse channel matrix. The sparse channel matrix can be expressed as:
[0056]
[0057] Where H represents the sparse channel matrix, F d and F a The dimensions are and N t ×N t The DFT matrix of .
[0058] Furthermore, in the delay domain, due to the time delay between multipath arrivals, the previous The rows contain significant values. To this end, after obtaining H, the top rows of H can be retained. rows to form a real-valued matrix H a , and the real-valued matrix H a As a truncated matrix, where the real-valued matrix H a The dimension is N c ×N t .
[0059] The embodiment of the present application performs a two-dimensional discrete Fourier transform on the sampled channel matrix to effectively convert the channel matrix into a low-dimensional truncated matrix. This not only simplifies the data, but also reduces the burden on the subsequent encoding process of the state coding model and the training process of the state coding model.
[0060] S20. The user equipment converts the channel matrix into a codeword through the state coding model, wherein the state coding model includes an attention mechanism, and the attention mechanism integrates a low-rank adaptation algorithm.
[0061] Specifically, the state coding model is a trained deep learning model that integrates an attention mechanism and a low-rank adaptation algorithm. That is, the state coding model includes an attention mechanism that integrates a low-rank adaptation algorithm, and by integrating the attention mechanism and the low-rank adaptation algorithm, it is possible to efficiently capture and process the dynamic changes of the channel state in the offshore environment on the basis of low computational complexity.
[0062] like Figure 2 and Figure 3 As shown, the state coding model includes a first convolution module, a Transformer-LoRA module and a first fully connected module cascaded in sequence. Among them, the first convolution module may include a convolution layer and a batch normalization layer. The Transformer-LoRA module includes a multi-head attention unit, a residual fusion unit and a feedforward network, the input and output items of the multi-head attention unit are the input items of the residual fusion unit, and the output items of the residual fusion unit are the input items of the feedforward network. The first fully connected module includes a fully connected layer. Based on this, the conversion of the channel matrix into a codeword through the state coding model is specifically: inputting the truncated matrix into the first convolution module, performing convolution calculation and batch normalization processing on the truncated matrix through the first convolution module to perform efficient feature learning to capture the deep features of the channel, and then inputting the deep features into the Transformer-LoRA for attention learning to obtain attention features, and then inputting the attention features into the first fully connected module, and converting the attention features into a codeword matrix through the first fully connected module.
[0063] The Transformer-LoRA module integrates the Transformer architecture and the LoRA algorithm. Specifically, the multi-head attention unit introduces two low-rank matrices A and B through low-rank adaptation, so that the linear transformation matrix of the multi-head attention unit is expressed as: W = W + AB, where W represents the linear transformation matrix, and A and B both represent low-rank matrices. In this way, on the one hand, long-distance dependencies can be captured through the self-attention mechanism, and through the high-efficiency low-rank structure adjustment of LoRA, the model reduces the computational complexity without losing performance.
[0064] In one implementation, the state encoding model can be jointly trained with the state decoding model used by subsequent base stations, wherein the state decoding model includes a second fully connected module, a Transformer-LoRA module, a second convolutional module, and an activation function layer cascaded in sequence, the Transformer-LoRA module multi-head attention unit, a residual fusion unit, and a feedforward network, the input and output items of the multi-head attention unit are the input items of the residual fusion unit, and the output items of the residual fusion unit are the input items of the feedforward network. The second fully connected module may include a fully connected layer, the second convolutional module may include a convolutional layer and a batch normalization layer, and the activation function layer may use a Sigmoid function.
[0065] Specifically, an end-to-end learning strategy is adopted to train all layers of the state encoding model and the state decoding model, and the parameters are updated by the ADAM optimization algorithm. The unsupervised learning algorithm is used for training to reduce the dependence on labeled data. The input and output channel matrices are normalized to adapt to the learning range of the network. In addition, during the training process of the state encoding model and the state decoding model, the low-rank matrix A is initialized with random Gaussian values, and the low-rank matrix B is initialized with zero, so that the linear transformation matrix change ΔW=BA is zero at the beginning of training, and the linear transformation matrix change ΔW is scaled during the training process. During the fine-tuning process of the state encoding model and the state decoding model, the linear transformation matrix is kept unchanged, and the low-rank matrix A and the low-rank matrix B are updated, which keeps the main body of the model stable and allows small-scale updates to effectively adapt to specific tasks or data sets.
[0066] The embodiment of the present application combines the low-rank adaptation algorithm into the Transformer architecture, and indirectly trains the state encoding model and the state decoding model by updating the low-rank matrix A and the low-rank matrix B, which significantly reduces the computational work required to update the matrix during model training, and enables the model to quickly adapt to specific fields or different environments. At the same time, in addition to reducing computing resource consumption and ensuring model stability, only a small number of parameters need to be updated, thereby reducing the risk of overfitting and enhancing the anti-interference ability of the state encoding model and the state decoding model.
[0067] It should be noted that in addition to introducing two low-rank matrices in the multi-head attention unit in the Transformer architecture, two low-rank matrices can also be introduced in the feedforward network in the Transformer architecture. The introduction process and training process are the same as those of introducing two low-rank matrices in the multi-head attention unit, so they will not be repeated here.
[0068] S30: The user equipment transmits the codeword to the base station.
[0069] Specifically, after the channel matrix is converted into codewords through a state encoder, the codewords are transmitted to a base station, wherein after receiving the codewords, the base station decodes the received codewords through a decoder to obtain a final channel matrix in the spatial frequency domain.
[0070] Exemplarily, the transmission method of the edge near-sea channel state based on low-rank adaptation also includes:
[0071] The base station inputs the received codeword into a preset state decoding model, wherein the state decoding model includes an attention mechanism, and the attention mechanism integrates a low-rank adaptation algorithm;
[0072] The reconstructed channel matrix is determined by the state decoding model.
[0073] Specifically, the state decoding model processes the codewords through the second fully connected layer module, and then inputs the processed codewords into the Transformer-LoRA module, and restores the characteristic information of the channel through the Transformer-LoRA module, and finally outputs the decoded channel matrix through the second convolution module and the activation function layer. In addition, after obtaining the decoded channel matrix, the decoded channel matrix will also be Fourier transformed to restore the dimension of the channel matrix to obtain the reconstructed channel matrix. Based on this, the determination of the reconstructed channel matrix through the state decoding model specifically includes: decoding the codewords through the state decoding model to obtain the output channel matrix; performing an inverse discrete Fourier transform on the output channel matrix to obtain the reconstructed channel matrix corresponding to the channel matrix. To this end, in practical applications, after decoding and obtaining the reconstructed channel matrix, the base station can generate an assisting precoding vector through the reconstructed channel matrix to facilitate the determination of the signal received on each subcarrier.
[0074] In summary, this embodiment provides a transmission method for the edge offshore channel state based on low-rank adaptation, the method comprising a user device obtaining a channel matrix, and inputting the channel matrix into a preset state coding model; the user device converts the channel matrix into a codeword through the state coding model, wherein the state coding model includes an attention mechanism, and the attention mechanism integrates a low-rank adaptation algorithm; the user device transmits the codeword to the base station. The embodiment of the present application integrates the Transformer structure and the rank adaptation algorithm, constructs a state coding model and a state decoding model through the Transformer architecture and the LoRA algorithm, and then encodes the channel matrix through the state coding model and transmits it to the base station, and the base station decodes it through the state decoding model to obtain the channel state information, so that not only can the dynamic changes of the channel state in the offshore environment be efficiently captured and processed through the self-attention mechanism in the Transformer architecture, and the accuracy of CSI feedback can be improved, but also the effective compression and training of CSI data can be realized through the LoRA algorithm, reducing the computing resources required for processing and transmission, and improving the adaptability and robustness of the system to environmental changes.
[0075] Based on the above-mentioned transmission method based on low-rank adaptive edge near-shore channel state, this embodiment provides a transmission system based on low-rank adaptive edge near-shore channel state, such as Figure 4 As shown, the transmission system based on low-rank adaptation of edge near-shore channel state includes a user equipment and a base station;
[0076] The user equipment is used to obtain a channel matrix, input the channel matrix into a preset state coding model, convert the channel matrix into a codeword through the state coding model, and transmit the codeword to a base station, wherein the state coding model integrates an attention mechanism and low-rank adaptation.
[0077] In one implementation, the base station is used to input the received codeword into a preset state decoding model, and determine the reconstructed channel matrix through the state decoding model, wherein the state decoding model integrates an attention mechanism and low-rank adaptation.
[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A transmission method for edge near-shore channel state based on low-rank adaptation, characterized in that: The edge offshore channel state optimization method based on low-rank adaptation technology specifically includes: The user equipment obtains a channel matrix, and inputs the channel matrix into a preset state coding model; The user equipment converts the channel matrix into a codeword through the state coding model, wherein the state coding model includes an attention mechanism, and the attention mechanism integrates a low-rank adaptation algorithm; The user equipment transmits the codeword to the base station.
2. The transmission method of edge near-shore channel state based on low-rank adaptation according to claim 1, characterized in that: The inputting the channel matrix into a preset state coding model specifically includes: Performing a two-dimensional discrete Fourier transform on the channel matrix to convert the channel matrix into a truncated matrix, wherein the matrix dimension of the truncated matrix is smaller than the matrix dimension of the channel matrix; The truncated matrix is input into a preset state coding model.
3. The transmission method of edge near-shore channel state based on low-rank adaptation according to claim 1 or 2, characterized in that: The channel matrix is determined by the user equipment based on the pilot signal, which is expressed as: in, represents the channel matrix, Indicates the nth th The channel vector of subcarriers, Indicates the number of subcarriers, N t (>>1) indicates the number of transmitting antennas configured on the base station.
4. The transmission method of edge near-shore channel state based on low-rank adaptation according to claim 1, characterized in that: The method further comprises: The base station inputs the received codeword into a preset state decoding model, wherein the state decoding model includes an attention mechanism, and the attention mechanism integrates a low-rank adaptation algorithm; The reconstructed channel matrix is determined by the state decoding model.
5. The transmission method of edge near-shore channel state based on low-rank adaptation according to claim 4 is characterized in that: Determining the reconstructed channel matrix by using the state decoding model specifically includes: Decoding the codeword by the state decoding model to obtain an output channel matrix; An inverse discrete Fourier transform is performed on the output channel matrix to obtain a reconstructed channel matrix corresponding to the channel matrix.
6. The transmission method based on low-rank adaptive edge near-shore channel state according to claim 1 or 4, characterized in that: The state encoding model includes a first convolution module, a Transformer-LoRA module and a first fully connected module cascaded in sequence; the state decoding model deployed by the base station includes a second fully connected module, a Transformer-LoRA module, a second convolution module and an activation function layer cascaded in sequence, the Transformer-LoRA module multi-head attention unit, a residual fusion unit and a feedforward network, the input and output items of the multi-head attention unit are the input items of the residual fusion unit, and the output items of the residual fusion unit are the input items of the feedforward network, wherein the multi-head attention unit introduces two low-rank matrices A and low-rank matrix B through low-rank adaptation, so that the linear transformation matrix of the multi-head attention unit is expressed as: W=W+AB Among them, W represents the linear transformation matrix, and A and B both represent low-rank matrices.
7. The transmission method of edge near-sea channel state based on low-rank adaptation according to claim 6 is characterized in that: The multi-head attention unit adopts a Transformer structure, and when fine-tuning the state encoding model and / or the state decoding model, the linear transformation matrix is kept unchanged and the low-rank matrix A and the low-rank matrix B are updated.
8. The transmission method of edge near-shore channel state based on low-rank adaptation according to claim 6, characterized in that: During the training process of the state encoding model and / or the state decoding model, the low-rank matrix A is initialized with random Gaussian values, and the low-rank matrix B is initialized with zero, so that the linear transformation matrix change ΔW=BA is zero at the beginning of training, and the linear transformation matrix change ΔW is scaled during the training process.
9. A transmission system based on low-rank adaptive edge near-shore channel state, characterized in that: The transmission system based on low-rank adaptive edge near-shore channel state includes a user equipment and a base station; The user equipment is used to obtain a channel matrix, input the channel matrix into a preset state coding model, convert the channel matrix into a codeword through the state coding model, and transmit the codeword to a base station, wherein the state coding model integrates an attention mechanism and low-rank adaptation.
10. The low-rank adaptive edge near-shore channel state transmission system according to claim 9, characterized in that: The base station is used to input the received codeword into a preset state decoding model, and determine the reconstructed channel matrix through the state decoding model, wherein the state decoding model integrates an attention mechanism and low-rank adaptation.