Method for realizing RIS self-adaptive reconstruction of multi-user channel based on Vision Transform network

Abstract

The invention provides a method for realizing self-adaptive reconstruction of a multi-user channel by an RIS (Remote Information System) based on a Vision Transform network. The method comprises the following steps: generating a channel impulse response matrix under a random user position based on a channel model, and solving a numerical solution of an RIS phase shift matrix when the maximum user receiving power is ensured through a deep neural network; generating a channel impulse response diagram in the local area based on the solved RIS phase shift matrix and a channel model, and training a Vision Transform network according to the channel impulse response diagram and the corresponding RIS phase shift matrix; and constructing a required channel impulse response diagram according to the user position and the mask model, and inputting the channel impulse response diagram into the trained network to realize the self-adaptive solution of the RIS phase. According to the method, the RIS phase matrix can be solved according to the required multi-user channel impulse response diagram, so that RIS self-adaptive solving under the scenes of different user positions and user numbers is realized, and an important basis is provided for realizing intelligent adjustment of a communication environment based on the RIS.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communications, and in particular to a method for implementing RIS adaptive reconstruction of multi-user channels based on a Vision Transformer network. Background Art

[0002] As one of the enabling technologies for 6G communications, RIS (Reconfigurable Intelligent Surface) introduces a transmitter-RIS-receiver communication link, which enhances the adjustability of the channel. By controlling the amplitude, phase, polarization, and frequency of its units, dynamic intelligent adjustment of the transceiver channel can be achieved. Research on adaptive reconstruction of wireless channels in multi-user scenarios is crucial for the development of 6G wireless communications. In multi-user scenarios, the optimization of the RIS phase shift matrix is ​​closely related to the dimension and distribution of the transceiver channel impulse response matrix. To achieve adaptive solution of the RIS phase shift matrix for any channel with random user positions and numbers, it is necessary to unify the representation of channel dimensions and distribution. However, most of the existing artificial intelligence optimization algorithms build a network to describe the correspondence between the channel matrix and the RIS phase shift matrix, and solve the RIS phase shift matrix to ensure the maximum channel capacity or coverage of the transceiver. For scenarios where the number of users changes or the user positions change, the optimization algorithm needs to be iteratively solved again, and it is difficult to adaptively solve the RIS phase shift matrix in any channel scenario.

[0003] For multi-user channels, it is particularly important to conduct research on RIS adaptive solution methods. In a channel environment where the number of users changes and the user positions are arbitrary, the adaptive RIS phase shift matrix solution method can ensure that multiple users within the target range are assisted at the same time. Therefore, how to build a network to mine the relationship between the channel impulse response matrix and the RIS phase shift matrix under different channel conditions is a hot topic in the research of RIS adaptive reconstruction of wireless channels. In addition, how to unify the input dimension of the network under different channel conditions is a major problem in RIS adaptive solution. At the same time, RIS adaptive reconstruction channel research is also helpful for the design of multi-user or mobile user beam tracking and sensor positioning functions, and can increase channel capacity and reduce coverage blind spots.

[0004] At present, in the existing technology, some scholars have adopted optimization algorithms such as deterministic and heuristic algorithms to solve RIS-related non-convex problems to ensure the maximization of multi-user channel capacity. However, the non-convex constraints and constant mode constraints in the optimization problem in this method make it challenging to solve the RIS phase shift matrix. In addition, deep learning is used to explore the mapping relationship between the channel matrix and the RIS phase shift to solve the complex RIS phase optimization problem. However, these research results require re-iterative training of the network model when the number and position of users are random, and cannot well guarantee the adaptive solution of the RIS phase in the channel scenario with any number and position of users. Summary of the invention

[0005] The invention provides a method for realizing RIS adaptive reconstruction of multi-user channels based on a Vision Transformer network, so as to realize effective utilization of RIS adaptive reconstruction of multi-user channels.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme.

[0007] A method for implementing RIS adaptive reconstruction of multi-user channels based on a Vision Transformer network, comprising:

[0008] Construct a Vision Transformer network, use the user's impulse response graph and RIS label to train the Vision Transformer network, and obtain a trained Vision Transformer network;

[0009] According to the location and number of users to be assisted, the required impulse response diagram is drawn;

[0010] The required impulse response graph is input into the trained Vision Transformer network, and the Vision Transformer network performs RIS phase adaptive solution to obtain the RIS phase shift matrix of the user to be assisted.

[0011] Preferably, the construction of the Vision Transformer network includes:

[0012] A Vision Transformer network including block segmentation, block embedding, position embedding and Transformer encoding modules is built. The Transformer encoding module consists of L Transformer encoders, each of which includes a multi-head attention mechanism, a multi-layer perceptron and a normalization layer, and is connected using residuals.

[0013] Preferably, the method of training the Vision Transformer network using the impulse response graph and RIS label of the user to obtain the trained Vision Transformer network includes:

[0014] Multiple user locations are randomly generated. The horizontal and vertical coordinates of the user locations are uniformly distributed in the interval [5m, 15m] and the interval [-5m, 5m], respectively. The channel impulse response matrix is ​​generated based on the channel model. The channel impulse response matrix from the transmitting antenna to the kth user is:

[0015]

[0016] in, is the RIS phase shift matrix, is the phase shift of the nth element of RIS, N is the number of RIS elements, H BS-RIS and denote the base station-RIS and RIS-kth user channel matrices respectively;

[0017] The channel impulse response diagram is calculated according to the channel impulse response matrix. The channel impulse response diagram is composed of the channel impulse response amplitudes in the horizontal range [5m, 15m] and the vertical range [-5m, 5m]. The channel impulse response diagram I imag The value at (x,y) is:

[0018]

[0019] Among them, H RIS-xy is the channel matrix from RIS to (x,y) position;

[0020] Adjust the size of the input channel impulse response graph to Divide the conditioned impulse response image into a series of image blocks H and W are the height and width of the image, C is the number of channels, N c =HWP is the number of image blocks, P is the height and width of the image block;

[0021] The image block I P Perform block embedding operation and transform the size of convolution to The number of input channels of the convolution layer is C, the number of output channels is D, the stride and kernel size of the convolution kernel are both P, and the block embedding tensor is added to the learnable parameter x class ∈R 1×D , and the result is

[0022] A learnable position embedding operation is introduced to distinguish the relative positions of image blocks and obtain the input of the Transformer encoding module:

[0023]

[0024] The encoder of the Transformer encoding module is described as:

[0025] I l ′=E MHA (E Norm (I l-1 ))+I l-1 ,l=1,...,L (5)

[0026] I l =E MLP (E Norm (I l ′))+I l ′,l=1,...,L (6)

[0027] in, E MHA (·), E Norm (·) and E MLP (·) denotes multi-head attention, layer normalization, and multi-layer perceptron layer, respectively. The Q of the i-th head in the multi-head attention i , K i and V i They are:

[0028]

[0029] in, are learnable parameters, And d k =D / N s , parameter N s Represents the number of heads in multi-head attention, and the weight matrix A of the i-th head i It is expressed as:

[0030]

[0031] Where i∈1,2,...,N s , based on the weight matrix A of the i-th head i , the output of the multi-head attention layer is expressed as:

[0032]

[0033] where U∈R D×D is a linear projection;

[0034] After being processed by the multi-head attention mechanism and the L-layer transform encoder, the matrix I is obtained L , take the matrix I LThe first dimension of is normalized, and the RIS phase is obtained through a multilayer perceptron. The number of nodes in the input layer, hidden layer, and output layer of the multilayer perceptron are D, 4D, and D, respectively, expressed as:

[0035]

[0036] in, is a multi-layer perceptron;

[0037] The channel impulse response graph and the corresponding RIS phase shift matrix are used as labels, and the VisionTransformer network is trained using supervised learning. During the training process, the Adam optimizer is used to update the network parameters, and the learning rate is 0.001. The loss function of the VisionTransformer network is:

[0038]

[0039] Among them, N ba is the batch size, Θ and The trained VisionTransformer network is obtained by using the label value and the predicted RIS phase shift value.

[0040] Preferably, the step of drawing the required impulse response diagram according to the position and number of users to be assisted includes:

[0041] The mask model M(r) is defined as the average value of the channel impulse response amplitude r units away from the user position in the channel impulse response diagram, which is defined as:

[0042]

[0043] Where K is the number of impulse response graphs in the training set, p (k) =(x k ,y k ,z k ) is the impulse response image I of the kth channel (k) Medium user location, is the channel impulse response image I (k) The three-dimensional position of any point in the , parameter r∈{1,2,...,R}, R is the target point boundary around the user position, d(p (k) ,q (a) ) describes the distance between the user's location and surrounding points, Nr is the number of points with a distance r between the user position and the surrounding points in the impulse response graph, and the distance d(p (k) ,q (a) ) is defined as:

[0044]

[0045] Determine the positions of multiple users to be optimized in the impulse response diagram, generate channel impulse response values ​​with a distance of r units from the user according to the mask model M(r), and r∈{1,2,...,R}, remove the remaining values ​​defined by the mask model in the impulse response diagram, and use a random generation method with uniform distribution in the interval [-45dB,-40dB] to fill them in, and obtain the required impulse response diagram

[0046] Preferably, the step of inputting the required impulse response graph into a trained Vision Transformer network, wherein the Vision Transformer network performs RIS phase adaptive solution to obtain the RIS phase shift matrix of the user to be assisted, comprises:

[0047] The desired impulse response diagram Input into the trained Vision Transformer network, and obtain the RIS phase shift vector of the user to be assisted as θ:

[0048]

[0049] in The relationship with the RIS phase shift matrix Θ in formula (1) is: F ViT is the trained vision Transformer network, θ ViT are the VisionTransformer network parameters.

[0050] It can be seen from the technical solutions provided by the above embodiments of the present invention that the method of the present invention can well solve the adaptive solution of the RIS phase shift matrix in a channel scenario with any number of users and random positions, and has high network generalization, thereby serving the development of RIS communication.

[0051] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0053] Figure 1A processing flow chart of a method for implementing RIS adaptive reconstruction of multi-user channels based on a Vision Transformer network provided in an embodiment of the present invention.

[0054] Figure 2 A schematic diagram of the Vision Transformer network framework structure in the training phase provided by an embodiment of the present invention.

[0055] Figure 3 A flow chart of constructing a training data set provided by an embodiment of the present invention.

[0056] Figure 4 A RIS phase shift matrix prediction flow chart is provided in an embodiment of the present invention.

[0057] Figure 5 A schematic diagram of comparison between required channels and regenerated channels of two users at arbitrary positions provided by an embodiment of the present invention.

[0058] Figure 6 A schematic diagram of comparison between required channels and regenerated channels when three users are at any positions is provided in an embodiment of the present invention.

[0059] Figure 7 A schematic diagram of the minimum channel impulse response amplitude result corresponding to a RIS phase shift matrix predicted by different networks provided in an embodiment of the present invention.

[0060] Figure 8 A schematic diagram of a comparison result of similarity between a required impulse response diagram and a regenerated impulse response diagram using different prediction networks is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0062] 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 invention 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 intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0063] 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 in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0064] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0065] How to realize the adaptive solution of RIS phase shift matrix for any channel scenario with random number and position of users is the key to RIS adaptive reconstruction of channels. The present invention generates a channel impulse response diagram in a local interval based on the channel model, builds a Vision Transformer network to mine the mapping relationship between the channel impulse response diagram and the RIS phase shift matrix, and proposes a method for realizing RIS adaptive reconstruction of multi-user channels. This method generates the required channel impulse response diagram for random user positions, and uses the trained Vision Transformer network to adaptively predict the RIS phase, which lays the foundation for RIS adaptive reconstruction of wireless channels, and provides an important basis for the application of RIS in communication systems.

[0066] Example 1

[0067] The processing flow of a method for implementing RIS adaptive reconstruction of multi-user channels based on a Vision Transformer network provided by an embodiment of the present invention is as follows: Figure 1 As shown, the processing steps include the following:

[0068] Step S1: Generate a channel matrix with random user numbers and user positions, and use the channel matrix to solve the user's RIS phase based on a deep neural network.

[0069] Multiple user locations are randomly generated, and the horizontal and vertical coordinates of the user locations are uniformly distributed in the interval [5m, 15m] and the interval [-5m, 5m] respectively. The channel impulse response matrix is ​​generated based on the channel model. The channel impulse response matrix from the transmitting antenna to the kth user is:

[0070]

[0071] in, is the RIS phase shift matrix, is the phase shift of the nth element of RIS, N is the number of RIS elements, H BS-RIS and denote the base station-RIS and RIS-kth user channel matrices respectively;

[0072] In order to ensure the maximum received power of users, deep neural networks are built for different numbers of users and unsupervised learning is used to solve the RIS phase numerical solution. During the training process, the Adam optimizer is used to update the network parameters with a learning rate of 0.001. The loss function of the network is:

[0073]

[0074] Among them, N K is the batch size, α=50 is the penalty factor to ensure that each user has a good channel condition, ||·|| and |·| are the binary norm and absolute value operations respectively, To sum by column.

[0075] Step S2: construct RIS-assisted channel impulse response diagrams under different numbers of users using the RIS phases of the users.

[0076] In order to unify the representation form of the channel under different numbers of users, a channel impulse response diagram is constructed to represent the channels with different numbers of users and user positions. The channel impulse response diagram is calculated based on the channel impulse response matrix solved in step S1. The channel impulse response diagram consists of the channel impulse response amplitudes in the horizontal range [5m, 15m] and the vertical range [-5m, 5m]. Channel impulse response diagram The value at (x,y) is:

[0077]

[0078] Among them, H RIS-xy is the channel matrix from RIS to the (x,y) position.

[0079] Step S3: Construct the Vision Transformer network architecture based on the channel impulse response graph.

[0080] Build a Vision Transformer network including block segmentation, block embedding, position embedding and Transformer encoding modules. The structure of the Vision Transformer network is as follows Figure 2 First, adjust the size of the input channel impulse response graph to I imag ∈R H×W×C , divide the adjusted impulse response image into a series of image blocks Where H and W are the height and width of the image, and C is the number of channels. c =HWP is the number of image blocks, and P is the height and width of the image block.

[0081] Secondly, the image block I P Perform block embedding operation and transform the size of convolution to The number of input channels of the convolution layer is C, the number of output channels is D (the numerical size is equal to the number of RIS units N), and the stride and kernel size of the convolution kernel are both P. After the block is embedded, the tensor is added to the learnable parameter x class ∈R 1×D , and the result is

[0082] In addition, a learnable position embedding operation is introduced to distinguish the relative positions of image blocks and obtain the input of the Transformer encoding module:

[0083]

[0084] Finally, build the Transformer encoding block network, the network structure is as follows Figure 2 As shown on the right, it consists of L Transformer encoders. Each encoder includes a multi-head attention mechanism, a multi-layer perceptron, and a normalization layer, connected using residuals. The encoder can be described as:

[0085] I l ′=E MHA (E Norm (I l-1 ))+I l-1 ,l=1,...,L (5)

[0086] I l =E MLP (E Norm (I l ′))+I l ′,l=1,...,L (6)

[0087] in, E MHA (·), E Norm (·) and E MLP (·) denotes multi-head attention, layer normalization, and multi-layer perceptron layer, respectively. Q of the i-th head in the multi-head attention i , K i and V i They are:

[0088]

[0089] in, are learnable parameters, And d k =D / N s Parameter N s Represents the number of heads in multi-head attention. The weight matrix A of the i-th head i It can be expressed as:

[0090]

[0091] Where i∈1,2,...,N s . Based on the weight matrix A of the i-th head i , the output of the multi-head attention layer can be expressed as:

[0092]

[0093] where U∈R D×D is a linear projection.

[0094] After the multi-head attention mechanism and L-layer transform encoder processing, the matrix I can be obtained L . Take the matrix I L The first dimension of is normalized and the RIS phase is obtained through a multilayer perceptron. The number of nodes in the input layer, hidden layer, and output layer of the multilayer perceptron are D, 4D, and D, respectively. It can be expressed as:

[0095]

[0096] in, is a multi-layer perceptron.

[0097] Step S4: Train the Vision Transformer network according to the impulse response graph and the RIS label to obtain a trained Vision Transformer network.

[0098] The channel impulse response graph and the corresponding RIS phase shift matrix constructed in step 2 are used as labels to train the network. During the training process, the Adam optimizer is used to update the network parameters with a learning rate of 0.001. The loss function of the Vision Transformer network is:

[0099]

[0100] Among them, N ba is the batch size. Θ and is the label value and the predicted RIS phase shift value.

[0101] Step S5: Draw the required impulse response diagram according to the location and number of users to be assisted

[0102] The mask model M(r) is defined as the average value of the channel impulse response amplitude r units away from the user position in the channel impulse response diagram, which is defined as:

[0103]

[0104] Where K is the number of impulse response graphs in the training set, p (k) =(x k ,y k ,z k ) is the impulse response image I of the kth channel (k) Medium user location, is the channel impulse response image I (k) The three-dimensional position of any point in the user position; parameter r∈{1,2,...,R}, R is the target point boundary around the user position, d(p (k) ,q (a) ) describes the distance between the user's location and surrounding points, N r is the number of points with a distance r between the user position and the surrounding points in the impulse response graph, and the distance d(p (k) ,q (a) ) is defined as:

[0105]

[0106] Determine the position in the impulse response diagram according to the number and position of the users to be assisted, and generate impulse response values ​​around the users based on the mask model to obtain the required impulse response diagram The specific generation process is as follows: first, determine the positions of multiple users to be optimized in the impulse response diagram; second, generate channel impulse response values ​​with a distance of r units from the user according to the mask model M(r), and r∈{1,2,...,R}; finally, for the remaining values ​​in the impulse response diagram except those defined by the mask model, use a random generation method with uniform distribution in the interval [-45dB,-40dB] to fill them.

[0107] Step S6: The required impulse response diagram The input is sent to the trained Vision Transformer network, which performs RIS phase adaptive solution to obtain the RIS phase shift matrix of the user to be assisted.

[0108] By inputting the arbitrary impulse response graph constructed in step 5 To the trained Vision Transformer network, the RIS phase shift vector θ can be obtained in any channel scenario:

[0109]

[0110] in The relationship with the RIS phase shift matrix Θ in formula (1) is: F ViT is the trained vision Transformer network, θ ViT are the VisionTransformer network parameters.

[0111] The present invention discloses a method for implementing RIS adaptive reconstruction of wireless channels. In this method, a Vision Transformer network is built to mine the corresponding relationship between a channel impulse response diagram and a RIS phase shift matrix. First, a channel impulse response matrix is ​​generated under a random user position based on a channel model, and a numerical solution of the RIS phase shift matrix is ​​solved by a deep neural network to ensure that the user's received power is maximum. Secondly, a channel impulse response diagram in a local area is generated based on the solved RIS phase shift matrix and the channel model, and a Vision Transformer network is trained according to the channel impulse response diagram and the corresponding RIS phase shift matrix. Finally, the required channel impulse response diagram is input into the trained network to realize the adaptive solution of the RIS phase. The present invention unifies the channel representation under different user numbers and position scenarios, builds a Vision Transformer network to realize the adaptive solution of RIS in a multi-user scenario, and provides an important basis for realizing intelligent adjustment of the communication environment through RIS design.

[0112] Embodiment 2

[0113] Table 1 summarizes the settings of channel simulation parameters. The transmitter position is (0m, 0m, 10m), the RIS position is (10m, 5m, 7m), the number of RIS units is 100, the number of transmitter antennas is 3, the number of users is 2 or 3, the user height is 1.5m, and the users are uniformly distributed in the interval [5m, 15m] and the interval [-5m, 5m] in the horizontal and vertical coordinates respectively.

[0114] Table 1 Channel simulation parameter settings

[0115]

[0116] The core of the invention is to realize RIS adaptive reconstruction of multi-user channels. Under this method, the adaptive solution of RIS phase shift matrix is ​​divided into three steps:

[0117] 1. Construct a training data set, generate a channel impulse response matrix under random positions and numbers of users based on the channel model, solve the RIS phase shift matrix value through unsupervised learning of deep neural networks, and generate a channel impulse response diagram based on the required RIS phase shift matrix and channel model;

[0118] 2. Build the Vision Transformer network, take the channel impulse response atlas as input, the RIS phase shift matrix set as label, and train the network using supervised learning;

[0119] 3. Based on the user position, the mask model is introduced to generate the required channel impulse response diagram, which is input into the trained Vision Transformer network to achieve adaptive solution of the RIS phase.

[0120] Firstly, the channel impulse response atlas and RIS phase shift matrix set for communication scenarios with different numbers of users and user locations are established. Figure 3 Build a flow chart for the dataset. The network architecture is shown in Table 2. The RIS phase shift matrix label set is generated by a deep neural network. For different numbers of users, the input matrix of the deep neural network is Unsupervised learning is used to solve the RIS phase shift matrix as shown in equation (2). Combined with the solved RIS phase shift matrix, the local impulse response diagram in the horizontal interval [5m, 15m] and the vertical interval [-5m, 5m] is constructed. The channel impulse response diagram can uniformly represent the channel under different user numbers and positions. Figure 3 The elements in are the channel impulse response amplitudes at any locations within the transmitter-RIS-interval.

[0121] Table 2. Deep neural network architecture in dataset construction

[0122]

[0123] Secondly, build the Vision Transformer network architecture as follows Figure 2 As shown in the figure, the Vision Transformer network is trained by inputting the impulse response graph set and the corresponding RIS phase shift matrix set. The trained network can realize the RIS phase adaptive solution in any channel scenario. The network architecture includes segmentation of input image data, block embedding, position embedding, Transformer encoding and multi-layer perception operation. The introduction of multi-head attention mechanism and residual connection in the Transformer encoding network can effectively capture the beam information in the impulse response graph and improve the training stability.

[0124] Finally, the impulse response graph required for optimizing the user position is constructed, and the mask model is introduced to determine the channel impulse response value around the user position. The rest of the channel impulse response graph is randomly generated using a uniform distribution. The required impulse response graph is input into the trained Vision Transformer network to achieve adaptive solution of the RIS phase. Figure 4 Flowchart for implementing RIS phase shift matrix prediction based on Vision Transformer network. Figure 5 The figure is a schematic diagram of the comparison results of the required channel impulse response diagram and the regenerated channel impulse response diagram when two users are at arbitrary positions. (a) and (b) respectively represent the required channel impulse response diagram generated based on the mask model and the impulse response diagram based on solving the RIS phase regeneration. It can be seen that there are two users in the shown interval, located at (8.6m, -1.5m) and (7.6m, 3.5m) respectively. It can be concluded from the figure that in the communication scenario of two users, the RIS phase predicted by the Vision Transformer network can achieve the desired beam in the required impulse response diagram. Figure 6 Figures (a) and (b) show the channel impulse response diagrams required for three users at any position and the impulse response diagrams based on solving the RIS phase regeneration. The user positions are (11m, -2m), (9.8m, 5m) and (7.6m, -3m). It can be concluded from the figure that in the communication scenario of three users, the RIS phase predicted by the Vision Transformer network can obtain the desired beam, and the network can realize the adaptive solution of the RIS phase shift matrix under different numbers of users.

[0125] Figure 7 A schematic diagram of the minimum impulse response amplitude obtained by using the RIS phase shift matrix predicted by different networks is given. The minimum impulse response amplitude of the Vision Transformer network, CNN network and RIS element phase design is zero in a multi-user scenario, and 500 sets of data are counted. Figure 7It can be concluded that the RIS phase shift matrix predicted by Vision Transformer can obtain the maximum impulse response amplitude, proving that the RIS phase shift matrix predicted by the constructed network can achieve the best channel reconstruction effect in any scenario.

[0126] Figure 8 A schematic diagram of the similarity comparison between the regenerated impulse response graph and the required impulse response graph under different prediction networks is given. The weighted RMSE results of the Vision Transformer network, CNN network and RIS element phase design zero in multi-user scenarios are statistically analyzed, where the weighted RMSE is defined as:

[0127]

[0128] in, and They represent the required impulse response diagram and the impulse response diagram regenerated according to the solved RIS phase. ij is the weight factor at position (i, j), which is 10 within the boundary B and 1 beyond the boundary. im and M im Represent the horizontal dimension and vertical dimension of the image respectively. It can be concluded from the figure that the impulse response image regenerated by the RIS phase shift matrix predicted by the Vision Transformer network has the highest similarity with the input image, which shows that the RIS phase shift matrix predicted by the proposed Vision Transformer network method has higher accuracy in reconstructing multi-user channels.

[0129] In summary, the embodiment of the present invention can solve the RIS phase matrix by constructing the required multi-user channel impulse response diagram, thereby realizing the dynamic solution of the RIS phase matrix under different user positions and user number scenarios, and realizing adaptive adjustment of the wireless transmission environment. This adaptability improves the flexibility and adaptability of the communication environment and is applicable to a variety of complex scenarios. At the same time, the neural network constructed in the solution process of the present invention has an efficiency advantage, can quickly solve the RIS phase matrix, and helps to reduce system delays and computing resource consumption. In addition, compared with the CNN network, the constructed Vision Transformer can obtain a larger impulse response amplitude at the required position, and the impulse response image regenerated by the RIS phase shift matrix predicted by the network has a high similarity with the input image. This method provides an important basis for realizing intelligent adjustment of the communication environment based on RIS.

[0130] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0131] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0132] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0133] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for adaptively reconstructing multi-user channels using RIS based on a Vision Transformer network, characterized in that: include: Construct a Vision Transformer network, use the user's impulse response graph and RIS label to train the Vision Transformer network, and obtain a trained Vision Transformer network; According to the location and number of users to be assisted, the required impulse response diagram is drawn; The required impulse response graph is input into the trained Vision Transformer network, and the Vision Transformer network performs RIS phase adaptive solution to obtain the RIS phase shift matrix of the user to be assisted.

2. The method according to claim 1, characterized in that: The construction of the Vision Transformer network includes: A Vision Transformer network including block segmentation, block embedding, position embedding and Transformer encoding modules is built. The Transformer encoding module consists of L Transformer encoders, each of which includes a multi-head attention mechanism, a multi-layer perceptron and a normalization layer, and is connected using residuals.

3. The method according to claim 2, characterized in that The method of training the Vision Transformer network using the impulse response graph and RIS label of the user to obtain the trained Vision Transformer network includes: Multiple user locations are randomly generated. The horizontal and vertical coordinates of the user locations are uniformly distributed in the interval [5m, 15m] and the interval [-5m, 5m], respectively. The channel impulse response matrix is ​​generated based on the channel model. The channel impulse response matrix from the transmitting antenna to the kth user is: in, is the RIS phase shift matrix, is the phase shift of the nth element of RIS, N is the number of RIS elements, H BS-RIS and denote the base station-RIS and RIS-kth user channel matrices respectively; The channel impulse response diagram is calculated according to the channel impulse response matrix. The channel impulse response diagram is composed of the channel impulse response amplitudes in the horizontal range [5m, 15m] and the vertical range [-5m, 5m]. The channel impulse response diagram I imag The value at (x,y) is: Among them, H RIS-xy is the channel matrix from RIS to (x,y) position; Adjust the size of the input channel impulse response graph to Divide the conditioned impulse response image into a series of image blocks H and W are the height and width of the image, C is the number of channels, N c =HWP is the number of image blocks, P is the height and width of the image block; The image block I P Perform block embedding operation and transform the size of convolution to The number of input channels of the convolution layer is C, the number of output channels is D, the stride and kernel size of the convolution kernel are both P, and the block embedding tensor is added to the learnable parameter x class ∈R 1 ×D , and the result is A learnable position embedding operation is introduced to distinguish the relative positions of image blocks and obtain the input of the Transformer encoding module: The encoder of the Transformer encoding module is described as: THE l ′=And MHA (AND Norm (THE l-1 ))+I l-1 ,l=1,...,L (5) THE l =And MLP (AND Norm (THE l ′))+I l ′,l=1,...,L (6) in, E MHA (·), E Norm (·) and E MLP (·) denotes multi-head attention, layer normalization, and multi-layer perceptron layer, respectively. The Q of the i-th head in the multi-head attention i , K i and V i They are: in, are learnable parameters, And d k =D / N s , parameter N s Represents the number of heads in multi-head attention, and the weight matrix A of the i-th head i It is expressed as: Where i∈1,2,...,N s , based on the weight matrix A of the i-th head i , the output of the multi-head attention layer is expressed as: where U∈R D×D is a linear projection; After being processed by the multi-head attention mechanism and the L-layer transform encoder, the matrix I is obtained L , take the matrix I L The first dimension of is normalized, and the RIS phase is obtained through a multilayer perceptron. The number of nodes in the input layer, hidden layer, and output layer of the multilayer perceptron are D, 4D, and D, respectively, expressed as: in, is a multi-layer perceptron; The channel impulse response graph and the corresponding RIS phase shift matrix are used as labels, and the VisionTransformer network is trained using supervised learning. During the training process, the Adam optimizer is used to update the network parameters, and the learning rate is 0.

001. The loss function of the VisionTransformer network is: Among them, N ba is the batch size, Θ and The trained VisionTransformer network is obtained by using the label value and the predicted RIS phase shift value.

4. The method according to claim 3, characterized in that The described method of drawing the required impulse response diagram according to the position and number of users to be assisted includes: The mask model M(r) is defined as the average value of the channel impulse response amplitude r units away from the user position in the channel impulse response diagram, which is defined as: Where K is the number of impulse response graphs in the training set, p (k) =(x k ,y k ,z k ) is the impulse response image I of the kth channel (k) Medium user location, is the channel impulse response image I (k) The three-dimensional position of any point in the , parameter r∈{1,2,...,R}, R is the target point boundary around the user position, d(p (k) ,q (a) ) describes the distance between the user's location and surrounding points, Nr is the number of points with a distance r between the user position and the surrounding points in the impulse response graph, and the distance d(p (k) ,q (a) ) is defined as: Determine the positions of multiple users to be optimized in the impulse response diagram, generate channel impulse response values ​​with a distance of r units from the user according to the mask model M(r), and r∈{1,2,...,R}, remove the remaining values ​​defined by the mask model in the impulse response diagram, and use a random generation method with uniform distribution in the interval [-45dB,-40dB] to fill them in, and obtain the required impulse response diagram 5. The method according to claim 4, characterized in that The step of inputting the required impulse response graph into a trained Vision Transformer network, wherein the Vision Transformer network performs RIS phase adaptive solution to obtain the RIS phase shift matrix of the user to be assisted, comprises: The desired impulse response diagram Input into the trained Vision Transformer network, and obtain the RIS phase shift vector of the user to be assisted as θ: in The relationship with the RIS phase shift matrix Θ in formula (1) is: F ViT is the trained vision Transformer network, θ ViT are the VisionTransformer network parameters.

Images