A codebook-free near-field beamforming method, device, equipment, medium and product based on deep learning

Abstract

The present application discloses a codeless near-field beamforming method, device, equipment, medium and product based on deep learning, which relates to the field of near-field communication. The method includes constructing and training a convolutional neural network model; the convolutional neural network model includes multiple feature extraction modules, each feature extraction module includes two convolution layers, two regularization layers and two activation layers; in the first convolution layer, the padding is set to 1 and the convolution kernel size is 2×2, and in the second convolution layer, the padding is set to 0 and the convolution kernel size is 2×2; the current channel state information is input into the trained convolutional neural network model to determine the optimal beamforming vector; based on the optimal beamforming vector, the transmit signal of the antenna array is adjusted to form a directional beam to complete the near-field beamforming. The present application can not only effectively reduce the overhead caused by beam training, but also obtain higher beamforming accuracy.

Description

Technical Field

[0001] The present application relates to the field of near-field communication, and in particular to a codebook-free near-field beamforming method, apparatus, equipment, medium, and product based on deep learning. Background Art

[0002] Near-field communication has the potential to improve communication coverage and signal quality, but it also faces a series of technical challenges:

[0003] Complex channel modeling: Near-field channels have more propagation paths and complex signal attenuation characteristics. Traditional far-field channel models are difficult to accurately describe the channel characteristics in near-field environments.

[0004] Complex channel state information: Near-field channel state information is more complex, and traditional methods find it difficult to fully learn the diverse channel characteristics, resulting in inaccurate channel estimation.

[0005] High beam training overhead: Since the near-field beam codebook contains angle and distance information, it is more complex than the far-field codebook that is only related to angle, resulting in a significant increase in the beam training overhead. Summary of the Invention

[0006] The purpose of this application is to provide a codebook-free near-field beamforming method, device, equipment, medium and product based on deep learning, which can efficiently extract and estimate features of near-field channels and reduce the overhead of beam training.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a codebook-free near-field beamforming method based on deep learning, comprising:

[0009] Constructing and training a convolutional neural network model; the convolutional neural network model includes multiple feature extraction modules, each of which includes two convolution layers, two regularization layers, and two activation layers; in the first convolution layer, the padding is set to 1 and the convolution kernel size is 2×2, and in the second convolution layer, the padding is set to 0 and the convolution kernel size is 2×2;

[0010] Input the current channel state information into the trained convolutional neural network model to determine the optimal beamforming vector;

[0011] The transmit signal of the antenna array is adjusted based on the optimal shaping vector to form a directional beam, thereby completing near-field beamforming.

[0012] In a second aspect, the present application provides a codebook-free near-field beamforming device based on deep learning, comprising:

[0013] A model building and training module for building and training a convolutional neural network model; the convolutional neural network model includes multiple feature extraction modules, each of which includes two convolutional layers, two regularization layers, and two activation layers; in the first convolution layer, the padding is set to 1 and the convolution kernel size is 2×2, and in the second convolution layer, the padding is set to 0 and the convolution kernel size is 2×2;

[0014] An optimal beamforming vector determination module is used to input the current channel state information into the trained convolutional neural network model to determine the optimal beamforming vector;

[0015] The near-field beamforming module is used to adjust the transmission signal of the antenna array based on the optimal beamforming vector to form a directional beam, thereby completing near-field beamforming.

[0016] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned deep learning-based codebook-free near-field beamforming methods.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the above-mentioned deep learning-based codebook-free near-field beamforming methods.

[0018] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned deep learning-based codebook-free near-field beamforming methods.

[0019] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0020] This application provides a codebook-free near-field beamforming method, apparatus, device, medium, and product based on deep learning. By cleverly designing the convolution kernel and padding size in a convolutional neural network model, the convolutional neural network model can effectively extract features of channel state information, thereby deriving the optimal beamforming vector. Leveraging the powerful data processing capabilities and efficient parallel computing capabilities of the convolutional neural network model, this application not only effectively reduces the overhead of beam training, but also achieves higher beamforming accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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.

[0022] Figure 1 A flowchart of a codebook-free near-field beamforming method based on deep learning provided in one embodiment of the present application;

[0023] Figure 2 Schematic diagram of a near field communication model between a base station and a user;

[0024] Figure 3 It is a structural diagram of the convolutional neural network model and feature extraction module;

[0025] Figure 4 This is a training diagram of the convolutional neural network model;

[0026] Figure 5 Schematic diagram of the achievable rate under different signal-to-noise ratios;

[0027] Figure 6 Schematic diagram of the effect of carrier frequency on achievable rate;

[0028] Figure 7 Schematic diagram of the achievable rate at different distances;

[0029] Figure 8 Schematic diagram of the achievable rate at different angles;

[0030] Figure 9 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0032] With the rapid development of mobile communications, 6G has become a research hotspot for next-generation wireless communication systems. 6G aims to achieve higher spectral efficiency, lower power consumption, and a higher number and density of antennas to meet the high speed, low latency, and high reliability demands of future intelligent applications and the Internet of Things. To achieve these goals, ultra-large-scale antenna arrays are considered a key technology for 6G.

[0033] In traditional wireless communication systems, antenna array technology is mainly focused on far-field communication, that is, when the signal propagation distance is long. In this case, the signal wavefront can be approximated as a plane wave, which simplifies the channel modeling and beamforming process. However, as the communication frequency band shifts to higher frequency bands (such as millimeter waves and terahertz), the communication distance becomes shorter, the number of antennas increases significantly, and the Rayleigh distance gradually increases, which causes the signal propagation to change from far-field to near-field characteristics. In the near-field environment, the signal wavefront exhibits spherical wave characteristics, and the traditional far-field channel model and beamforming method are no longer applicable. However, current near-field communications have problems such as complex channel modeling, complex channel state information, and high beam training overhead.

[0034] This application designs a convolutional neural network model that, through offline training, enables efficient extraction and processing of near-field channel features. Through offline training and online inference of the convolutional neural network model, fast beamforming is achieved, significantly reducing the overhead of beam training and addressing the aforementioned issues with existing near-field communications.

[0035] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0036] In an exemplary embodiment, Figure 1 As shown, a codebook-free near-field beamforming method based on deep learning is provided, comprising the following steps S1 to S3.

[0037] S1: Construct and train a convolutional neural network model; the convolutional neural network model includes multiple feature extraction modules, each feature extraction module includes two convolution layers, two regularization layers and two activation layers; in the first convolution layer, the padding is set to 1 and the convolution kernel size is 2×2, and in the second convolution layer, the padding is set to 0 and the convolution kernel size is 2×2.

[0038] S2: Input the current channel state information into the trained convolutional neural network model to determine the optimal beamforming vector.

[0039] S3: Based on the optimal shaping vector, the transmission signal of the antenna array is adjusted to form a directional beam, thereby completing near-field beamforming.

[0040] Before step S1, the process also includes: constructing a near field communication model between the base station and the user, and collecting channel state information. The constructed near field communication model is as follows: Figure 2 As shown, Figure 2 In, r n is the distance between the user and the nth antenna unit, r is the distance between the center antenna and the user, D nis the distance from the nth antenna unit to the center antenna, θ n is the angle between the nth antenna unit and the user, and θ is the angle between the center antenna and the user.

[0041] Specifically:

[0042] Executing entity: computer simulation software.

[0043] Input: number of users K, signal-to-noise ratio SNR, number of antenna units N, and normalized transmit power P.

[0044] Operation: We study downlink beam training for a massive multiple-input, single-output (MISO) system. The base station (BS) is equipped with a linear array (ULA) with N antenna elements (i.e., antenna elements) spaced d = λ / 2, where λ represents the carrier wavelength. Given that millimeter-wave communications (i.e., near-field communications) primarily rely on line-of-sight (LOS) links, this application considers a single LOS radio frequency (RF) chain.

[0045] When the distance between the antenna array and the receiver is less than the Rayleigh distance (RD), the receiver is in the near field region. The formula for the Rayleigh distance is:

[0046]

[0047] Where D is the maximum size of the antenna array, λ is the wavelength of the carrier, and f is the carrier frequency. Assume that the distance from the center antenna to the user is r and the angle is θ. Considering the characteristics of the near-field spherical wave, the distance r between the UE and the nth antenna element is n for:

[0048]

[0049] The downlink channel h of the subcarrier between the nth antenna element and the UE n It can be expressed as:

[0050]

[0051] Among them, β n is the path attenuation coefficient of the nth antenna unit.

[0052] By stacking the channels of all antenna elements into a vector, the channel state information h of the kth UE can be obtained k :

[0053] h k =[h1,h2,…,h n …,h N ]

[0054] It can be seen that the near-field antenna response is a function of angle and distance. Based on this, the near-field steering vector can be expressed as:

[0055]

[0056] The downlink received signal y of the kth UE k It can be expressed as:

[0057]

[0058] Where G represents the antenna unit power, w represents the beamforming vector, H represents the transpose, and h k is the channel state information collected for the kth user, s is the downlink transmission signal, and η is the noise and interference that obeys the complex Gaussian distribution. Assume that the downlink transmission signal is |s| 2 =1, the achievable rate of the kth UE can be expressed as:

[0059]

[0060] Among them, h i is the channel state information collected for the i-th user, is the signal-to-noise ratio of the user.

[0061] In the above formula, other non-serving UE interference is considered as noise. According to the above near-field communication model, the channel state information collected from the base station and the user is used as the input of the convolutional neural network model.

[0062] Due to the close correlation between the near-field channel and the angle and distance information, the near-field codebook is usually large in size, and a lot of computational complexity will be introduced when using codebook-based beam design. In practical scenarios, the main focus is on the beam gain toward the target UE position, rather than optimizing the beam gain of each grid in the entire space. This application does not require a predefined codebook, and effectively controls the beamforming shape through deep learning to maximize the achievable data transmission rate. Considering the normal mode constraint|w i | 2 =1, for i=1,…,N, the optimization problem of the beamforming vector w is expressed as:

[0063]

[0064] stw i ∈w,|w i | 2 =1

[0065] The complexity of channel state information in XL-MIMO systems makes the above problems significantly nonlinear and non-convex. This application introduces deep learning as a powerful tool to solve these problems. An unsupervised convolutional neural network model is designed to enable the beamforming vector to flexibly adapt and learn under complex channel conditions. The solution goal is to find the mapping function Φ of the parameter ξ ξ , which uses the channel state information to predict the optimal beamforming vector w'. The mapping function can be expressed as:

[0066] Φ ξ {h}→{w'}

[0067] Wherein, h represents the channel state information of the user.

[0068] In a specific embodiment, the training process of the convolutional neural network model in step S1 includes: inputting the collected channel state information into the convolutional neural network model, outputting the equivalent phase of the beamforming coefficient; converting the equivalent phase into a beamforming vector using the Euler formula; calculating the user achievable rate based on the beamforming vector and the collected channel state information; constructing a loss function based on the user achievable rate; adjusting the parameters of the convolutional neural network model through the loss function to complete the training of the convolutional neural network model. Specifically:

[0069] () Design a convolutional neural network model

[0070] Execution entity: Computer

[0071] Operation: The input data dimension is 2×N. By adding an extra dimension, it is converted to 1×2×N and then input into the convolutional neural network in the figure above to extract complex signal features.

[0072] like Figure 3 As shown, each feature extraction module consists of two convolutional layers, two regularization layers (BatchNorm), and two activation layers (ReLU). In the first convolution layer, padding is set to 1 and the convolution kernel size is 2×2. In the second convolution layer, padding is set to 0 and the convolution kernel size is 2×2. By cleverly designing the convolution kernel and padding size, the convolutional neural network effectively extracts the real and imaginary components of the channel state information, as well as the equivalent phase (real and imaginary combined), while ensuring that the data size remains unchanged before and after processing in the feature extraction module. After passing through a feature extraction module, the data is downsampled using average pooling. This process is repeated twice, reducing the data size to B×C×2×N / 4. Subsequently, it undergoes two rounds of upsampling and processing in the feature extraction module, restoring the final data size to B×1×2×N, the same as the original input data. The data is then flattened into a one-dimensional vector and, after passing through a fully connected layer, mapped to the equivalent phase using a "Tanh" layer as the output.

[0073] (2) Design loss function

[0074] Executing entity: Computer

[0075] Operation: Apply the Euler formula to convert the equivalent phase α output by the convolutional neural network into the required beamforming vector. The complex output is given by the following formula:

[0076] w'=exp(j·α·π)=cos(πα)+j·sin(πα)

[0077] in, As can be seen, πα has a clear physical meaning, and its elements correspond to the equivalent phases of the beamforming coefficients in w'. Since this application uses a codebook-free deep learning method, the neural network training is guided by a loss function directly related to the beamforming objective. The loss function for the task is defined as the negative of the achievable rate:

[0078]

[0079] Where Q represents the total number of training samples, that is, the total number of channel state information collected, q is the number of channel state information collected, and K represents the total number of users. is the user's signal-to-noise ratio, and w' is the beamforming vector obtained based on the convolutional neural network model during training.

[0080] (3) Model training

[0081] Execution entity: High-performance computer or GPU server

[0082] Operation: The collected channel state information is input into the convolutional neural network model. After regularization, the real and imaginary parts are stacked and used as input data for training. The convolutional neural network model architecture includes multiple convolutional layers and fully connected layers to capture complex channel characteristics. The output is converted into an equivalent phase through a "Tanh" layer. The equivalent phase is converted into a beamforming vector using the Euler formula, and the loss function formula is used to calculate the loss. The convolutional neural network model parameters are then updated through backpropagation. As training progresses, the loss function decreases, corresponding to an increase in the user achievable rate. When the convolutional neural network model converges, the loss function tends to stabilize and the user achievable rate is maximized. The convolutional neural network model parameters are saved for subsequent deployment.

[0083] In a specific embodiment, the trained convolutional neural network is deployed online. Specifically:

[0084] Executing entity: base station

[0085] Operation: As Figure 4As shown in the figure, the trained convolutional neural network model is deployed in the base station. During communication, the current channel state information is input in real time. The base station can then quickly output the optimal beamforming vector that achieves the maximum sum rate based on the trained convolutional neural network model. Based on this output, the base station adjusts the transmitted signal to form a directional beam, completing beamforming.

[0086] This application validates the technical effectiveness of the above method through simulation experiments. An N = 256 equally spaced half-wave linear arrays are deployed at the base station. This application is compared with near-field stratification, far-field stratification, and an exhaustive beamforming scheme with a computational overhead limited to 256 beams. The reachability is used as the evaluation metric.

[0087] Figure 5 The figure shows how the reachability changes with the signal-to-noise ratio (SNR) under different beam training schemes, where the SNR increases from -20dB to 20dB. Figure 5 The results clearly demonstrate that the proposed scheme consistently outperforms existing far-field and near-field beam training schemes across a wide range of signal-to-noise ratio (SNR) values. At a 20dB SNR, the proposed scheme achieves approximately 18%, 30%, and 120% higher reachability compared to the near-field layered beamforming scheme, the far-field layered beamforming scheme, and the exhaustive search scheme, respectively. Notably, when the SNR exceeds -5dB, the proposed scheme exhibits significant performance gains compared to the comparison schemes.

[0088] Carrier frequency is a key factor affecting communication rate. Figure 6 Shown are the achievable rate curves of the scheme proposed in this application and the comparative scheme at different carrier frequencies. For all carrier frequencies, the scheme proposed in this application is always superior to other schemes in terms of achievable rate. For the far-field layered scheme, as the carrier frequency increases, near-field propagation gradually dominates, resulting in the average achievable rate of the far-field scheme basically no longer increasing when the carrier is greater than 40 GHz. The best performance of all solutions was observed at a carrier frequency of 50 GHz. This is because the free space propagation loss generated by high-frequency signals during transmission increases. As the carrier frequency increases, the performance gain obtained by increasing the number of antennas will not be enough to offset the increasing free space propagation loss. When fc>50 GHz, the performance of the comparative scheme decreases, while the performance of the scheme proposed in this application remains relatively stable, which confirms the robustness of the solution proposed in this application.

[0089] Figure 7The correlation between the achievable rate and the distance from the UE to the BS is shown. When the UE is close to the BS, the scheme proposed in this application has superior rate performance compared with the comparative scheme. For the far-field scheme, as the distance decreases, the advantage of near-field propagation causes the average transmission rate of the far-field scheme to drop rapidly. Due to the limitation of the maximum training cost, the exhaustive scheme suffers from severe degradation in the near-field area, which hinders the effective search of the near-field position. In addition, the near-field layered beam training scheme can moderately alleviate the decline in the average achievable rate, but its performance gradually decreases when the distance exceeds 30m. In addition, when the distance is less than 20 meters, its performance lags significantly behind the scheme proposed in this application. In contrast, this application demonstrates the ability to search for the optimal beamforming vector with minimal pilot overhead. This confirms the robustness of the scheme proposed in this application in near-field and far-field communication scenarios.

[0090] Figure 8 The achievable rate as a function of angle is demonstrated. Here, θ is gradually increased from -60° to 60°. For the far-field layered scheme, the performance degrades significantly as θ approaches 0 degrees. This degradation is attributed to the enhanced near-field effect near zero angle. In addition, the near-field layered beam training scheme shows significant fluctuations. This is because the near-field hybrid scheme creates a near-field codebook by uniformly sampling angles and distances in a Cartesian coordinate system, which is challenging to achieve stable beamforming performance in the entire near-field environment. In contrast, the scheme proposed in this application consistently achieves the highest average achievable rate over all considered angle ranges. This result highlights the robustness and effectiveness of this application in dealing with the challenges posed by angle changes in near-field communication scenarios.

[0091] Based on the same inventive concept, the embodiments of the present application also provide a device for implementing the aforementioned deep learning-based codebook-free near-field beamforming method. The implementation solution provided by the device is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of the one or more embodiments of the deep learning-based codebook-free near-field beamforming device provided below can be found in the above-mentioned limitations of the deep learning-based codebook-free near-field beamforming method, and will not be repeated here.

[0092] In an exemplary embodiment, a codebook-free near-field beamforming apparatus based on deep learning is provided, comprising:

[0093] A model construction and training module is used to construct and train a convolutional neural network model; the convolutional neural network model includes multiple feature extraction modules, each feature extraction module includes two convolution layers, two regularization layers and two activation layers; in the first convolution layer, the padding is set to 1 and the convolution kernel size is 2×2, and in the second convolution layer, the padding is set to 0 and the convolution kernel size is 2×2.

[0094] The optimal beamforming vector determination module is used to input the current channel state information into the trained convolutional neural network model to determine the optimal beamforming vector.

[0095] The near-field beamforming module is used to adjust the transmission signal of the antenna array based on the optimal beamforming vector to form a directional beam, thereby completing near-field beamforming.

[0096] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-mentioned method embodiments. The computer device can be a server or a terminal, and its internal structure can be as shown in FIG. Figure 9 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data to be processed. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a codebook-free near-field beamforming method based on deep learning is implemented.

[0097] Those skilled in the art will understand that Figure 9 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps of the above-mentioned method embodiments when executing the computer program.

[0098] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0099] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0100] It should be noted that the user information (including but not limited to user information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0101] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0102] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0103] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0104] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A codebook-free near-field beamforming method based on deep learning, characterized in that: include: Construct and train a convolutional neural network model; the convolutional neural network model includes multiple feature extraction modules, each of which includes two convolution layers, two regularization layers, and two activation layers; in the first convolution layer, the padding is set to 1 and the convolution kernel size is 2×2, and in the second convolution layer, the padding is set to 0 and the convolution kernel size is 2×2; the convolutional neural network can effectively extract the real part information, imaginary part information, and equivalent phase of the channel state information, while ensuring that the data size remains unchanged before and after processing by the feature extraction module; The equivalent phase is the real part information combined with the imaginary part information of the channel state information; Input the current channel state information into the trained convolutional neural network model to determine the optimal beamforming vector; Adjusting the transmit signal of the antenna array based on the optimal shaping vector to form a directional beam, thereby completing near-field beamforming; The expression of channel state information is: h k =[h1,h2,…,h n …,h N ] Among them, h k is the channel state information collected for the kth user, h n is the downlink channel of the subcarrier between the nth antenna unit and the user, N is the number of antenna units, β n is the path attenuation coefficient of the nth antenna unit, r is the distance between the center antenna and the user, d is the spacing between the antenna units, and θ is the angle between the center antenna and the user; Before building and training the convolutional neural network model, it also includes: Build a near-field communication model between the base station and the user, and collect channel state information; The training process of the convolutional neural network model includes: Inputting the collected channel state information into the convolutional neural network model and outputting the equivalent phase of the beamforming coefficient; The equivalent phase is converted into a beamforming vector using the Euler formula; Calculating a user achievable rate based on the beamforming vector and the collected channel state information; constructing a loss function based on the user achievable rate; Adjusting the parameters of the convolutional neural network model by using the loss function to complete the training of the convolutional neural network model; The method of adjusting the parameters of the convolutional neural network model by the loss function is specifically as follows: using the loss function to calculate the loss, and then updating the convolutional neural network model parameters by back propagation.

2. The codebook-free near-field beamforming method based on deep learning according to claim 1, characterized in that: The calculation formula of the user achievable rate is: Among them, R k is the achievable rate of the kth user, w is the beamforming vector, H represents the transpose, G is the antenna unit power, h i is the channel state information collected for the i-th user, is the signal-to-noise ratio of the user.

3. The codebook-free near-field beamforming method based on deep learning according to claim 1, characterized in that: The expression of the loss function Loss is: Where Q is the total amount of channel state information collected, q is the number of channel state information collected, K is the total number of users, R k is the achievable rate of the kth user, w' is the beamforming vector obtained based on the convolutional neural network model during training, H represents the transpose, and h k is the channel state information collected for the kth user, h i is the channel state information of the i-th user collected, m is the subcarrier number, is the signal-to-noise ratio of the user.

4. A codebook-free near-field beamforming device based on deep learning, characterized in that: include: A model construction and training module for constructing and training a convolutional neural network model; the convolutional neural network model includes multiple feature extraction modules, each of which includes two convolutional layers, two regularization layers, and two activation layers; in the first convolution layer, the padding is set to 1 and the convolution kernel size is 2×2, and in the second convolution layer, the padding is set to 0 and the convolution kernel size is 2×2; the convolutional neural network can effectively extract the real part information, imaginary part information, and equivalent phase of the channel state information, while ensuring that the data size remains unchanged before and after processing by the feature extraction module; The equivalent phase is the real part information combined with the imaginary part information of the channel state information; An optimal beamforming vector determination module is used to input the current channel state information into the trained convolutional neural network model to determine the optimal beamforming vector; A near-field beamforming module, configured to adjust the transmit signal of the antenna array based on the optimal beamforming vector to form a directional beam, thereby completing near-field beamforming; The expression of channel state information is: h k =[h1,h2,…,h n …,h N ] Among them, h k is the channel state information collected for the kth user, h n is the downlink channel of the subcarrier between the nth antenna unit and the user, N is the number of antenna units, β n is the path attenuation coefficient of the nth antenna unit, r is the distance between the center antenna and the user, d is the spacing between the antenna units, and θ is the angle between the center antenna and the user; Before building and training the convolutional neural network model, it also includes: Build a near-field communication model between the base station and the user, and collect channel state information; The training process of the convolutional neural network model includes: Inputting the collected channel state information into the convolutional neural network model and outputting the equivalent phase of the beamforming coefficient; The equivalent phase is converted into a beamforming vector using the Euler formula; Calculating a user achievable rate based on the beamforming vector and the collected channel state information; constructing a loss function based on the user achievable rate; Adjusting the parameters of the convolutional neural network model by using the loss function to complete the training of the convolutional neural network model; The method of adjusting the parameters of the convolutional neural network model by the loss function is specifically as follows: using the loss function to calculate the loss, and then updating the convolutional neural network model parameters by back propagation.

5. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the codebook-free near-field beamforming method based on deep learning according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the codebook-free near-field beamforming method based on deep learning described in any one of claims 1 to 3 are implemented.

7. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the codebook-free near-field beamforming method based on deep learning described in any one of claims 1 to 3 are implemented.

Images