Millimeter wave communication system beam alignment method based on intelligent angle map

Abstract

The invention belongs to the field of wireless communication, and relates to an intelligent angle map-based millimeter wave communication system beam alignment method, which comprises the following steps of: constructing an intelligent reflecting surface-assisted millimeter wave communication system model; constructing a multivariable coupled wireless resource optimization problem taking the sum rate of the maximized model as a target; converting a wireless resource optimization problem into an optimization problem about angle domain parameters; constructing a data set according to an optimization problem about the angle domain parameters; a Transform model is constructed, and the Transform model is trained according to the data set; obtaining user position information, inputting the user position information into the trained Transform model to obtain estimated angle information, and completing beam alignment according to the estimated angle information; according to the method, active-passive beamforming related parameters are completely converted into an angle form, so that beam angle information estimation is realized, and the complexity of a problem is reduced to a great extent.

Description

Technical Field

[0001] The invention belongs to the field of wireless communications and relates to a beam alignment method for a millimeter wave communication system based on an intelligent angle map. Background Art

[0002] The millimeter wave band has a very wide range of spectrum resources. Using millimeter waves for signal transmission can effectively alleviate the problem of spectrum resource shortage. However, the propagation characteristics of millimeter waves also make its transmission loss in space extremely high, and it is very easy to be blocked by obstacles during transmission. In recent years, based on beamforming technology, the directional gain brought by large-scale antenna arrays and the reinforcement function of intelligent reflective surfaces (Reconfigurable Intelligent Surface, RIS) in expanding signal coverage can effectively make up for such shortcomings. By using beamforming technology at the transmitting and receiving ends, the signal energy can be concentrated in a specific direction (that is, forming a "beam"), thereby making up for the defect of large millimeter wave path loss. In order to ensure the quality of the millimeter wave communication link, it is necessary to estimate the angle of the beam and achieve beam alignment.

[0003] In addition, in the field of communications, in order to visualize and analyze the characteristics of wireless channels in specific areas or conditions, researchers have proposed the important concept of channel maps. The so-called channel map is a graphical tool that displays the characteristics of wireless channels in a certain geographical area, which can help people better understand and optimize wireless networks. The indicators that can be shown in the map usually include information such as signal strength, signal-to-noise ratio, path loss, etc. Knowing this information is helpful for network planning, optimization, troubleshooting, and performance evaluation. Using traditional methods to draw channel maps is very resource-intensive and has many limitations in accuracy and coverage. The use of machine learning methods can make the drawn channel maps more practical and save a lot of costs through automated processing, precise modeling, data fusion, dynamic adaptation, etc.

[0004] The industry has achieved a lot of research results on beam alignment technology in high-frequency communication systems, and many of them are related to RIS or channel maps. However, for the scenario of RIS-assisted millimeter-wave communication, the efficiency and accuracy of beam alignment using channel maps are not high enough. Summary of the invention

[0005] In order to solve the above-mentioned prior art problems, the present invention adopts a beam alignment method of a millimeter wave communication system based on an intelligent angle map, comprising:

[0006] S1. Construct a multi-user millimeter wave communication system assisted by an intelligent reflector;

[0007] S2. Construct a multivariable coupled wireless resource optimization problem with the goal of maximizing the sum rate of a multi-user millimeter wave communication system;

[0008] S3, transforming the multi-variable coupled wireless resource optimization problem into an optimization problem about angle domain parameters;

[0009] S4. constructing a data set according to an optimization problem regarding angle domain parameters, wherein the data set includes user position information and user angle information;

[0010] S5. Build a Transformer model, train the Transformer model according to the data set, and obtain a trained Transformer model;

[0011] S6. Obtain user location information, input the user location information into the trained Transformer model, obtain estimated angle information, and complete beam alignment according to the estimated angle information.

[0012] The multi-user millimeter wave communication system includes: a millimeter wave base station, a smart reflector RIS and N users; the blocking probability between the millimeter wave base station and user n when When , there is a direct link between the millimeter wave base station and user n; when If the direct link between the millimeter wave base station and user n is blocked, user n is a blocked user.

[0013] The multivariable coupled wireless resource optimization problem is:

[0014]

[0015] Among them, C1 is the constraint of the blocking probability between the millimeter wave base station and user n, C2 is the constraint of the user's arrival angle, the transmission angle of the millimeter wave base station, the arrival angle of the RIS and the azimuth angle of the transmission angle of the RIS, C3 is the constraint of the user's arrival angle, the transmission angle of the millimeter wave base station, the arrival angle of the RIS and the elevation angle of the transmission angle of the RIS, C4 is the phase constraint of the reflection unit of the RIS, C5 is the power constraint allocated to the user, C6 is the constraint of the beamforming vector at the base station end, C7 is the constraint of the beamforming vector at the user end, R represents the system and rate, represents the azimuth matrix between the base station and the user, θ n =(θ R,1 ,θ T,1 ,...,θ R,L ,θ T,L ) is the azimuth vector of user n, represents the azimuth matrix between the user and RIS, N2 represents the number of blocked users, is the azimuth vector of the RIS involved in the blocked user v, represents the elevation angle matrix of the base station and the user, is the pitch angle vector of user n, represents the elevation angle matrix between user n and RIS, is the elevation angle vector of the RIS involved in the blocked user v, Φ represents the RIS phase shift matrix, represents the allocated power set, P n represents the power allocated to user n, P max Indicates the total power of the base station, represents the set of transmission paths, L represents the total number of transmission paths, and transmission path l=1 represents a direct path. and Respectively represent the set of RIS reflection unit numbers in the horizontal and vertical directions, is the blocking probability between the millimeter wave base station and user n, θ in,l represents the azimuth of the arrival angle of the RIS under the lth path, θ out,l represents the azimuth of the launch angle of the RIS under the lth path, θ R,l represents the azimuth angle of the user's arrival angle under the lth path, θ T,l represents the azimuth of the base station's transmission angle under the lth path, represents the pitch angle of the arrival angle of RIS under the lth path, represents the pitch angle of the launch angle of RIS under the lth path, represents the pitch angle of the user's arrival angle under the lth path, represents the elevation angle of the base station's transmission angle under the lth path, Indicates the mth x Line m z The phase of the reflection unit on the column, w n 、f n denote the beamforming vectors at the user end and the base station end, respectively. A collection of users.

[0016] The system and rate R is:

[0017]

[0018] in, is the variance of the noise, the superscript H is the transposed conjugate operation, m is the index of the user, and Ω n is the equivalent channel, H n,LoS is the direct channel between base station and user, H n,NLoS It is the cascade channel of base station-RIS-user.

[0019] Cascade channel H of base station-RIS-usern,NLoS =h n,NLoS,2 Φh NLoS,1 , where h NLoS,1 is the communication channel between the millimeter wave base station and RIS, h n,NLoS,2 is the communication channel between RIS and user n.

[0020] Converting the multivariable coupled wireless resource optimization problem into an optimization problem of angle domain parameters includes:

[0021] S31. Construct the optimal phase shift matrix of RIS

[0022] S32. Construct auxiliary variables Using χ to simplify the optimal phase shift matrix, we get

[0023] S33, Solution Get the optimal phase shift matrix Φ opt , Φ opt The phase of each reflection unit is expressed as

[0024] S34, the optimal phase shift matrix Φ opt Substituting into the multi-variable coupled wireless resource optimization problem, we get the optimization problem of the angle domain parameters;

[0025] in, Respectively represent h n,NLoS,2 and h NLoS,1 The direct path portion, h NLoS,1 is the communication channel between the millimeter wave base station and RIS, h n,NLoS,2 is the communication channel between RIS and user n, a M (·) is the RIS array steering vector, is a constant, d is the antenna spacing of RIS, and λ is the wavelength.

[0026] The optimization problem of the angle domain parameters is:

[0027]

[0028] stC1,C2,C3,C4,C6,C7.

[0029] Where p is the average power allocated to the user.

[0030] Constructing a data set based on the optimization problem of the angle domain parameters includes: using DeepMIMO to generate data X of a multi-user millimeter wave communication system assisted by an intelligent reflector, and extracting user position information and user angle information from the data X User location information and user angle information Combine to get a sample set, combine multiple sample sets to get a data set; where, Θ * is the optimal azimuth matrix of the base station and the user, Ψ * is the optimal elevation angle matrix of the base station and the user, is the optimal user-RIS azimuth matrix, is the optimal user and RIS elevation angle matrix.

[0031] The Transformer model includes: an encoder and a decoder; the encoder includes multiple encoding modules with the same structure, and the encoding modules are connected in series; the decoder includes multiple decoding modules with the same structure, and the decoding modules are connected in series; training the Transformer model according to the data set includes:

[0032] S51, inputting the user position information into an encoder to obtain a coding matrix;

[0033] S52, inputting the coding matrix and the user angle information into a decoder to obtain estimated angle information;

[0034] S53. Calculate the loss function value according to the estimated angle information, and update the parameters of the Transformer model according to the loss function value. When the loss function value is minimum, a trained Transformer model is obtained.

[0035] The encoding module includes: embedding layer, attention layer, normalization and residual connection layer, and feedforward network layer; the encoding module processes user location information including:

[0036] S511, inputting the user location information into the embedding layer to obtain an embedding vector of the user location;

[0037] S512, input the embedding vectors of the user positions into the self-attention layer respectively;

[0038] S513, normalizing the output input of the self-attention layer and connecting it to the residual connection layer;

[0039] S514, input the output of the normalization and residual connection layer into the feedforward network layer, and input the output of the feedforward network layer into the normalization and residual connection layer again to obtain a coding matrix.

[0040] Beneficial effects of the present invention:

[0041] The present invention realizes beam angle information estimation by completely converting active-passive beamforming related parameters into the form of angle information function, completes beam alignment between millimeter wave base stations, RIS and multiple users, and greatly reduces the complexity of the problem; in addition, compared with traditional algorithms, the present invention uses deep learning to train the relationship model between the user's location information and the optimal transmit-receive beam angle corresponding to the location, thereby drawing a beam angle map of the user's location area. Using the obtained angle map, it is possible to quickly estimate the beam angle information when the user moves, and can be effectively applied to more complex communication environments, thereby achieving efficient beam alignment. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart of a beam alignment method for a millimeter wave communication system based on an intelligent angle map provided in an embodiment of the present invention;

[0043] Figure 2 A model diagram of a millimeter wave communication system assisted by an intelligent reflective surface provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] like Figure 1 , Figure 2 As shown, the present invention adopts a beam alignment method for a millimeter wave communication system based on an intelligent angle map, comprising:

[0046] In this embodiment, a RIS-assisted multi-user millimeter wave communication network is considered, including: a millimeter wave base station, a RIS, and N communication users; wherein, there is a line-of-sight link between N1 users and the base station, and N2 users are blocked by obstacles and need to establish a cascade channel through the RIS to communicate with the base station. All users, the millimeter wave base station, and the RIS can be regarded as equipped with a uniform planar antenna array.

[0047] The specific implementation steps of the plan are as follows:

[0048] S1. Modeling RIS-assisted multi-user millimeter-wave communication system;

[0049] A multi-user millimeter wave communication system includes: a millimeter wave base station, a RIS and N communication users. The millimeter wave base station can provide services for N users at the same time; among them, there are direct links between N1 users and the base station, and the direct links between N2 users and the base station are blocked by obstacles. It is necessary to establish a cascade link with the base station through the RIS for communication, and N=N1+N2.

[0050] The millimeter wave base station is equipped with a A uniform planar antenna array, and Represents the number of antennas in the horizontal and vertical directions respectively. All users are equipped with antennas of size A uniform planar antenna array, and Represent the number of antennas in the horizontal and vertical directions respectively. The size of the reflection unit array of RIS is M = M x ×M z , M x and M z Represent the number of antennas in the horizontal and vertical directions respectively.

[0051] For Users For a user collection, define is the blocking probability between the millimeter wave base station and user n; when When there is a direct link between the base station and user n, the channel in this state is defined as when This means that the direct link between user n and the base station is blocked. At this time, user n needs to communicate through the cascade channel between base station-RIS-user. The cascade channel between base station-RIS-user is defined as H n,NLoS =h n,NLoS,2 Φh NLoS,1 ;

[0052] in, represents the communication channel between the base station and RIS, is the communication channel between RIS and user n, Then it represents the RIS phase shift matrix, Indicates the mth x Line m z The phase of the reflection unit on the column, L represents the number of paths between the base station and user n, α l represents the complex channel gain under the lth path, a R (·) and a T (·) represent the array steering vectors of the signal receiving end and the signal transmitting end, respectively. represents the azimuth (elevation angle) under the lth path, R, T, in, and out represent the user's arrival angle, the base station's transmission angle, the RIS's arrival angle, and the RIS's transmission angle, respectively.

[0053] definition is the signal sent by the millimeter wave base station to user n, is the signal received by user n, where w n 、f n represent the beamforming vectors at the user end and the base station end respectively, P n represents the power allocated to user n, represents Gaussian white noise, σ 2 is the variance of the noise.

[0054] In summary, the signal-to-interference-noise ratio of user n can be modeled as in, is the variance of the noise; further, the user achievable rate can be described as R n =log2(1+SINR n ). It is worth mentioning that the beamforming vectors on the user and base station sides are calculated based on the channel, and the channel can be regarded as a function of the angle, so both beamforming vectors can be described by a function of the angle, that is

[0055] S2. Construct a multivariable coupled wireless resource optimization problem with the goal of maximizing the rate of a multi-user millimeter wave communication system.

[0056] First, the achievable rate of a single user is given in step S1, then the system sum rate can be expressed as The corresponding optimization problem can be described as follows:

[0057]

[0058] Among them, C1 is the constraint of the blocking probability between the millimeter wave base station and user n, C2 is the constraint of the user's arrival angle, the transmission angle of the millimeter wave base station, the arrival angle of the RIS and the azimuth angle of the transmission angle of the RIS, C3 is the constraint of the user's arrival angle, the transmission angle of the millimeter wave base station, the arrival angle of the RIS and the elevation angle of the transmission angle of the RIS, C4 is the phase constraint of the reflection unit of the RIS, C5 is the power constraint allocated to the user, C6 is the constraint of the beamforming vector at the base station end, and C7 is the constraint of the beamforming vector at the user end.

[0059] S3. To reduce the computational complexity, the multivariable coupled wireless resource optimization problem is transformed into an optimization problem regarding the angle domain parameters.

[0060] First, define the optimal phase shift matrix of RIS as in, Respectively represent h n,NLoS,2 and h NLoS,1 For the direct path part, this equation expresses that when the spectrum efficiency of the blocked user is maximized, the phases of each reflection unit on the RIS constitute the optimal phase shift matrix.

[0061] Furthermore, define auxiliary variables h n,NLoS,2 With h NLoS,1 Bring in Φ opt In, we can get in, part is a constant, so we can get

[0062] Then direct the RIS array to vector a M Substituting the specific expression of (·) and the phase shift matrix Φ into χ, we can obtain Where j is the imaginary unit, k is a constant, d is the antenna spacing and λ is the wavelength.

[0063] Final solution The optimal phase shift matrix Φ can be obtained opt , where the phase of each reflection unit is expressed as:

[0064]

[0065] Among them, m x is the number of rows m z is the number of columns. At this point, the phase shift matrix of RIS is converted into a function that is only related to the angle

[0066] In addition, to make the calculation more concise, this scheme assumes that the base station allocates equal power to all users, and the average power is defined as Therefore, the original optimization problem is transformed into

[0067]

[0068] stC1,C2,C3,C4,C6,C7.

[0069] S4, constructing a data set according to an optimization problem regarding angle domain parameters;

[0070] In one embodiment, DeepMIMO is used to generate data X of a multi-user millimeter wave communication system assisted by a smart reflective surface, and user position information and user angle information are extracted from the data X of the multi-user millimeter wave communication system assisted by a smart reflective surface. User location information and user angle information Combine to get a sample set, combine multiple sample sets to get a data set; where, Θ * is the optimal azimuth matrix of base stations and users, Ψ * is the optimal elevation angle matrix of the base station and the user, is the optimal user-RIS azimuth matrix, is the optimal user and RIS pitch angle matrix. DeepMIMO (Deep Learning for Massive MIMO) is a dataset generation tool for wireless communication research, which can simulate the channel state information (CSI) of massive multiple input multiple output (Massive MIMO) systems. By using this tool, channel state information (CSI) in various scenarios can be generated, so as to test and optimize wireless communication algorithms in different environments.

[0071] In one embodiment, constructing a data set based on an optimization problem regarding angle domain parameters includes: obtaining user location information, solving an optimization problem regarding angle domain parameters, and obtaining an optimal angle. The optimal angle The labels and user location information are combined to obtain a sample set; multiple sample sets are combined to obtain a data set.

[0072] S5. Build a Transformer model, train the Transformer model according to the data set, and obtain a trained Transformer model;

[0073] The Transformer model includes: an encoder and a decoder; the encoder includes multiple encoding modules with the same structure, and the encoding modules are connected in series; the decoder includes multiple decoding modules with the same structure, and the decoding modules are connected in series; training the Transformer model according to the optimization problem of the angle domain parameters includes:

[0074] S51, inputting the user location information into the serially connected encoding modules in sequence to obtain the encoding matrix C;

[0075] S52, inputting the coding matrix C and the user angle information into each decoding module to obtain estimated angle information;

[0076] S53. Calculate the loss function value according to the estimated angle information, and update the parameters of the Transformer model according to the loss function value. When the loss function value is minimum, the trained Transformer model is obtained. The trained model will be saved on the base station side as the mapping function of the angle map.

[0077] Optionally, the loss function is a mean square error loss function, that is Among them, I is the number of sample users, r i is the true angle value of the i-th sample user, is the predicted angle value of the i-th sample user.

[0078] Both the encoding module and the decoding module include: an embedding layer, an attention layer, a normalization and residual connection layer, and a feedforward network layer; the encoding module processes the user location information including:

[0079] S511, input the user location information into the embedding layer to obtain the embedding vector I of the user location att ;

[0080] S512, inputting the embedding vector of the user position into the self-attention layer;

[0081] First, embedding vector I att Respectively with W q , W k and W v The query matrix is ​​obtained by multiplying the three weight matrices. Key-value matrix Sum Matrix N is the number of users, d k d e The parameters of the three weight matrices are training parameters and will change with model iterations. Then, the query matrix Q is multiplied by the transpose of the key matrix K and multiplied by the scaling factor. The multiplication result is then sent to a Softmax layer to calculate the attention score of the current input user position Finally, multiplying the attention score A by the value matrix V gives us the output matrix of the self-attention layer:

[0082] In one embodiment, multi-head attention is used in the model, that is, the same input sequence is sent to the attention layer modules with different weight matrices at the same time, and then the results of all attention layers are merged at the final output, that is, h is the number of attention layers used in a multi-head attention layer.

[0083] S513, normalizing the output input of the self-attention layer and connecting it to the residual connection layer;

[0084] In this layer, element-by-element normalization is first performed, that is, Among them, I ij The input matrix The element in row i and column j, μ j and Represents the input matrix The mean and variance of the jth column, ε is a very small constant to ensure Not 0. After normalization, residual connection is performed to obtain I prev It represents the input of the network layer before the normalization layer, which is equal to I in the first encoding module of the encoder. att .

[0085] S514, input the output of the normalization and residual connection layer into the feedforward network layer, and input the output of the feedforward network layer into the normalization and residual connection layer again to obtain a coding matrix.

[0086] The feed-forward network layer consists of two fully connected layers, where the first fully connected layer uses the Relu activation function, while the second layer does not. The specific expression is as follows: The output matrix of the first fully connected layer is O FC,1 =max(0,I FC W1+b1), the output of the second fully connected layer is O FC,2 =O FC,1 W2+b2,I FC Represents the input of the feedforward network layer, that is, the output matrix O of the previous layer AN , W1, W2 and b1, b2 are the weight matrices and bias vectors of the two fully connected layers respectively. Then, the output matrix O of the feedforward network layer FC,2 The result is the encoding matrix C when all encoding modules are completed.

[0087] The output matrix C of the encoder is input into each decoding module, and compared with the encoding module, the decoding module uses an additional attention layer and a normalization and residual connection layer. Specifically, taking the first decoding module in the decoder as an example, the decoding module processes the encoding matrix C including:

[0088] Input the user angle information into the embedding layer to obtain the embedding vector of the user angle;

[0089] The embedding vector of the user's angle is input into a masked attention layer, which also uses a multi-head attention structure. The difference between the masked attention and the conventional attention module is that the masked attention layer will first mask the input data, that is, add a diagonal matrix with all negative infinity in the upper right corner to the input. This process can be expressed as Mask(I mask )=I mask +tril(-inf), where I mask is the input angle information embedding vector, tril(·) represents a lower triangular matrix, -inf represents negative infinity, and tril(-inf) is a lower triangular matrix with a value of negative infinity. After masking, the remaining steps are exactly the same as the normal attention layer;

[0090] Normalize the output input of the mask attention layer and process it with the residual connection layer;

[0091] The matrix output by the normalization and residual connection layer is input into the encoder-decoder attention layer. The encoder-decoder attention layer is also a multi-head structure. In the encoder-decoder attention layer, only the query matrix Q is calculated based on the output matrix of the previous layer, while the key matrix K and value matrix V are calculated based on the encoding matrix C output by the encoder. Other than this, the other steps of the encoder-decoder attention are the same as the conventional attention.

[0092] The output of the encoder-decoder attention layer is input into the feed-forward network layer, and the output of the feed-forward network layer is normalized and input into the residual connection layer. The relevant calculations of these two layers are consistent with those in the encoding module.

[0093] Finally, when all decoding modules have been run to completion, the decoder will input a fully connected layer and a Softmax layer before output. These two layers are used to restore the embedded information and determine the specific value of the estimated angle information.

[0094] S6. Obtain user location information, input the user location information into the trained Transformer model, obtain estimated angle information, and complete beam alignment according to the estimated angle information.

[0095] In the actual angle estimation stage, since the user angle information is completely unknown, the input of the decoder part will be replaced by a blank sequence with only the start symbol. The model will estimate the angle information based on the correspondence between the learned spatial position information and the angle information, and complete the beam alignment based on the estimated transmit-receive beam alignment corresponding angle information.

[0096] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A beam alignment method for a millimeter wave communication system based on an intelligent angle map, characterized in that: include: S1. Construct a multi-user millimeter wave communication system assisted by an intelligent reflector; S2. Construct a multivariable coupled wireless resource optimization problem with the goal of maximizing the sum rate of a multi-user millimeter wave communication system; S3, transforming the multi-variable coupled wireless resource optimization problem into an optimization problem about angle domain parameters; S4. constructing a data set according to an optimization problem regarding angle domain parameters, wherein the data set includes user position information and user angle information; S5. Build a Transformer model, train the Transformer model according to the data set, and obtain a trained Transformer model; S6. Obtain user location information, input the user location information into the trained Transformer model, obtain estimated angle information, and complete beam alignment according to the estimated angle information.

2. According to claim 1, a beam alignment method for a millimeter wave communication system based on an intelligent angle map is characterized in that: The multi-user millimeter wave communication system includes: a millimeter wave base station, a smart reflector RIS and N users; the blocking probability between the millimeter wave base station and user n when When , there is a direct link between the millimeter wave base station and user n; when If the direct link between the millimeter wave base station and user n is blocked, user n is a blocked user.

3. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 2, characterized in that: The multivariable coupled wireless resource optimization problem is: Among them, C1 is the constraint of the blocking probability between the millimeter wave base station and user n, C2 is the constraint of the user's arrival angle, the transmission angle of the millimeter wave base station, the arrival angle of the RIS and the azimuth angle of the transmission angle of the RIS, C3 is the constraint of the user's arrival angle, the transmission angle of the millimeter wave base station, the arrival angle of the RIS and the elevation angle of the transmission angle of the RIS, C4 is the phase constraint of the reflection unit of the RIS, C5 is the power constraint allocated to the user, C6 is the constraint of the beamforming vector at the base station end, C7 is the constraint of the beamforming vector at the user end, R represents the system and rate, represents the azimuth matrix between the base station and the user, θ n =(θ R,1 ,θ T,1 ,...,θ R,L ,θ T,L ) is the azimuth vector of user n, represents the azimuth matrix between the user and RIS, N2 represents the number of blocked users, is the azimuth vector of the RIS involved in the blocked user v, represents the elevation angle matrix of the base station and the user, is the pitch angle vector of user n, represents the elevation angle matrix between user n and RIS, is the elevation angle vector of the RIS involved in the blocked user v, Φ represents the RIS phase shift matrix, represents the allocated power set, P n represents the power allocated to user n, P max Indicates the total power of the base station, represents the set of transmission paths, L represents the total number of transmission paths, and transmission path l=1 represents a direct path. and Respectively represent the set of RIS reflection unit numbers in the horizontal and vertical directions, is the blocking probability between the millimeter wave base station and user n, θ in,l represents the azimuth of the arrival angle of the RIS under the lth path, θ out,l represents the azimuth of the launch angle of the RIS under the lth path, θ R,l represents the azimuth angle of the user's arrival angle under the lth path, θ T,l represents the azimuth of the base station's transmission angle under the lth path, represents the pitch angle of the arrival angle of RIS under the lth path, represents the pitch angle of the launch angle of RIS under the lth path, represents the pitch angle of the user's arrival angle under the lth path, represents the elevation angle of the base station's transmission angle under the lth path, Indicates the mth x Line m z The phase of the reflection unit on the column, w n 、f n denote the beamforming vectors at the user end and the base station end, respectively. A collection of users.

4. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 3, characterized in that: The system and rate R is: in, is the variance of the noise, the superscript H is the transposed conjugate operation, m is the index of the user, and Ω n is the equivalent channel, H n,LoS is the direct channel between the base station and the user, H n,NLoS It is the cascade channel of base station-RIS-user.

5. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 4, characterized in that: Cascade channel H of base station-RIS-user n,NLoS =h n,NLoS,2 Φh NLoS,1 , where h NLoS,1 is the communication channel between the millimeter wave base station and RIS, h n,NLoS,2 is the communication channel between RIS and user n.

6. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 3, characterized in that: Converting the multivariable coupled wireless resource optimization problem into an optimization problem of angle domain parameters includes: S31. Construct the optimal phase shift matrix of RIS S32. Construct auxiliary variables Using χ to simplify the optimal phase shift matrix, we get S33, Solution Get the optimal phase shift matrix Φ opt , Φ opt The phase of each reflection unit is expressed as S34, the optimal phase shift matrix Φ opt Substituting into the multi-variable coupled wireless resource optimization problem, we get the optimization problem of the angle domain parameters; in, Respectively represent h n,NLoS,2 and h NLoS,1 The direct path portion, h NLoS,1 is the communication channel between the millimeter wave base station and RIS, h n,NLoS,2 is the communication channel between RIS and user n, a M (·) is the RIS array steering vector, is a constant, d is the antenna spacing of RIS, and λ is the wavelength.

7. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 6, characterized in that: The optimization problem of the angle domain parameters is: stC1,C2,C3,C4,C6,C7. Where p is the average power allocated to the user.

8. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 7, characterized in that: Constructing a data set based on the optimization problem of the angle domain parameters includes: using DeepMIMO to generate data X of a multi-user millimeter wave communication system assisted by an intelligent reflector, and extracting user position information and user angle information from the data X User location information and user angle information Combine to get a sample set, combine multiple sample sets to get a data set; where Θ * is the optimal azimuth matrix of the base station and the user, Ψ * is the optimal elevation angle matrix of the base station and the user, is the optimal user-RIS azimuth matrix, is the optimal user and RIS elevation angle matrix.

9. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 1, characterized in that: The Transformer model includes: an encoder and a decoder; the encoder includes multiple encoding modules with the same structure, and the encoding modules are connected in series; the decoder includes multiple decoding modules with the same structure, and the decoding modules are connected in series; training the Transformer model according to the data set includes: S51, inputting the user position information into an encoder to obtain a coding matrix; S52, inputting the coding matrix and the user angle information into a decoder to obtain estimated angle information; S53. Calculate the loss function value according to the estimated angle information, and update the parameters of the Transformer model according to the loss function value. When the loss function value is minimum, a trained Transformer model is obtained.

10. The beam alignment method of a millimeter wave communication system based on an intelligent angle map according to claim 9, characterized in that: The encoding module includes: embedding layer, attention layer, normalization and residual connection layer, and feedforward network layer; the encoding module processes user location information including: S511, inputting the user location information into the embedding layer to obtain an embedding vector of the user location; S512, input the embedding vectors of the user positions into the self-attention layer respectively; S513, normalizing the output input of the self-attention layer and connecting it to the residual connection layer; S514, input the output of the normalization and residual connection layer into the feedforward network layer, and input the output of the feedforward network layer into the normalization and residual connection layer again to obtain a coding matrix.

Images