A multi-cell wireless network distributed resource optimization method based on a graph neural network

Abstract

The application discloses a multi-cell wireless network distributed resource optimization method based on a graph neural network, and comprises the following steps: establishing the total rate and optimization problem of all user terminal devices according to the communication model of the multi-cell wireless network; designing a message passing graph neural network model containing edge feature embedding, and mapping various channel information to the nodes and edges of the graph; centrally training the graph neural network model, maximizing the optimization problem by using the parameterization scheme of the graph neural network, and distributing the trained graph neural network on each cell to realize the distributed resource optimization of the multi-cell wireless network based on the graph neural network.

Description

Technical Field

[0001] This invention relates to the field of wireless communication, and in particular to a distributed resource optimization method for multi-cell wireless networks based on graph neural networks. Background Technology

[0002] This section provides only background information relevant to this disclosure and is not necessarily prior art.

[0003] Since its inception, cellular network architecture has gradually become the core architecture of commercial communications due to its advantages of simple implementation and high-frequency band resource utilization. With the development of mobile communication technology, technologies such as wider bandwidth, improved radio interfaces, and network encryption have been used to cope with the massive increase in traffic. Considering different cost-effectiveness factors, the contribution of massive MIMO (Multiple Input Multiple Output) technology is particularly significant. As a key 5G technology, massive MIMO constitutes a centralized solution for encrypted networks. It provides extremely high beamforming gain by using a large array of antennas to direct high-gain antenna beams toward one or more users; in addition, the highly directional beam not only improves the received power level, but also reduces interference from adjacent users and transmission points in space. Through spatial multiplexing operations, it achieves spatial reuse of scarce time / frequency resources, thereby improving the system's spectral efficiency and energy efficiency.

[0004] Multi-cell cooperative communication refers to the collaboration among multiple cells in a 5G communication system to achieve higher network capacity and a better user experience. In traditional wireless communication, different cells are often independent of each other. However, in multi-cell cooperative communication, cells can share and optimize resources, thereby improving communication efficiency and reliability. Currently, the centralized optimization schemes widely used in multi-cell networks have high computational complexity in large-scale networks and cannot adapt to dynamic changes in the number of cells or base stations or users. Summary of the Invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to provide a distributed resource optimization method for multi-cell wireless networks based on graph neural networks, which addresses the shortcomings of the existing technology.

[0006] Graph Neural Networks (GNNs) are neural network models specifically designed for processing graph-structured data. They aim to learn representations of nodes or graphs and perform complex graph data analysis and reasoning. Unlike traditional neural networks, which primarily process vector or matrix data, GNNs effectively capture the relationships and topological structures between nodes, making them suitable for various complex graph data analysis tasks. Unlike conventional neural network algorithms, GNNs extend existing neural network methods to the domain of non-Euclidean data, i.e., graph data processing. As a neural network model designed for graph data, GNNs possess three major advantages: strong expressive power, good generalization ability, and distributed operation, enabling them to address the challenges of intelligent wireless network design in 5G.

[0007] In graph neural networks (Graph Neural Networks), each node represents an entity or object in the graph, and each edge represents the relationship between nodes. The goal of Graph Neural Networks is to learn the representation of each node to enable information propagation, feature aggregation, and prediction across the entire graph structure. By learning the interactions and structural information between nodes, Graph Neural Networks can perform various tasks such as node classification, link prediction, and graph generation.

[0008] Graph neural networks typically comprise two main aspects: node representation learning and graph representation learning. Node representation learning aims to learn a low-dimensional representation of each node, allowing it to retain its neighbor relationships and feature information within the graph structure. Graph representation learning, on the other hand, maps the entire graph structure to a low-dimensional representation space for analysis and inference. Its core idea is to leverage the connections and local structural information between nodes, using multi-layer neural networks for information propagation and feature aggregation, thereby obtaining a more global graph representation. The advantage of graph neural networks lies in their ability to handle various complex graph data structures, including data from social networks, recommender systems, and bioinformatics. By learning the interactions and structural information between nodes, graph neural networks can achieve more accurate and effective graph data analysis and mining.

[0009] In the field of wireless communication, graph neural networks (GNNs) have broad prospects and potential. GNNs can be used to model and analyze the topology of communication networks, enabling optimized allocation of network resources and highlighting the necessity and advantages of applying GNNs in scheduling. By learning the spectrum utilization of wireless channels, intelligent spectrum sensing and allocation can be achieved. Modeling and optimizing the transmission, reception, and demodulation of wireless signals using GNNs can improve the performance and efficiency of communication systems. Furthermore, modeling and analyzing users, devices, and base stations in mobile networks using GNNs can enable intelligent management and scheduling of network resources.

[0010] A distributed resource optimization method for multi-cell wireless networks based on graph neural networks includes the following steps:

[0011] Step 1: With the goal of maximizing the total transmission rate of user terminal devices in multiple cells, and in conjunction with the maximum power constraint of cell base stations, establish an optimization problem;

[0012] Step 2: For the optimization problem, design a graph model as the input to the graph neural network. This graph model includes base station information of neighboring cells and base station and user information of the current cell.

[0013] Step 3: Establish a message-passing graph neural network learning framework that includes edge embeddings;

[0014] Step 4: Using the maximization of the system user transmission rate as the loss function, train and optimize the learning parameters of the graph neural network designed in Step 3.

[0015] Step 5: Distribute graph neural networks on each cell base station to design beamforming matrices, thereby achieving distributed resource optimization of multi-cell wireless networks based on graph neural networks.

[0016] In this invention, uppercase bold letters represent matrices, lowercase bold letters represent vectors, and lowercase letters represent scalars. Representation matrix The i-th column, Representation matrix The element in the i-th column and j-th row. Representing vectors The first element. Expressing expectations, Let A represent a complex field of dimension A*B, and let diag(A) represent a vector consisting of the diagonal elements of A. Represents a K-dimensional identity matrix, with superscript... T , H They represent the transpose and conjugate transpose of a matrix, respectively, with superscripts... * , t These represent the optimal solution and the value at iteration t, respectively. A backslash \ indicates exclusion from the set.

[0017] Further, step 1 includes: setting up a network with L cells, each cell equipped with one base station, and each base station serving N cells. l There are (l≤L) users, where the base stations in each cell transmit signals to the user terminal equipment via the downlink; each base station has N... t One antenna, and a single antenna for user terminal equipment.

[0018] Data symbols for all user terminal devices Represented as in The channel from the base station of cell l to user i within the cell is used It means that i≤Nl The channel from the base station of cell j to user i outside the cell uses express; Let y represent the beam vector when the base station of cell l sends a signal to the terminal device of user i within cell l. Then, the signal y received by the terminal device of user i within cell l is... li It can be represented as:

[0019]

[0020] Signal-to-interference-plus-noise ratio (SINR) received by user i's terminal device within cell l li Represented as:

[0021]

[0022] in The additive white Gaussian noise represents the noise emitted by the i-th user in the l-th cell. This represents the variance of the noise.

[0023] The overall optimization problem of this network model is specifically manifested as the problem of maximizing the total rate of all users:

[0024]

[0025] Where st represents the set of constraints imposed on the optimization problem. SINR is the maximum transmit power that the base station in cell l can use. li The signal-to-interference-plus-noise ratio (SIR) received by the terminal device of user i within cell l. It is the beamforming matrix of the base station in cell l.

[0026] Step 2 includes: using a graph neural network to model the original distributed problem into the following form:

[0027]

[0028] A heterogeneous graph model G is established based on the communication topology model of cell l. l ={N TX N RX，l E l}, where N TX ={N l |l≤L} represents the transmitting nodes of the entire multi-cell communication system, including intra-cell base stations and inter-cell base stations. Represents the variables on the TX node. N represents the characteristic matrix of the TX node; RX,l ={N i |i≤N l Let {l ≤ L} represent the receiving nodes, i.e., user nodes, within cell l. Represents the variables on the RX node. E represents the characteristic matrix of the RX node; l This represents the edges between the nodes in the graph represented by the current cell l. Represents the variables on the edge. Represents the edge features of each edge; d TX d RX and d E They represent the feature dimensions, where It is a mapping function from features to variables.

[0029] Step 3 includes: constructing a graph neural network model with the following structure:

[0030] The graph neural network model consists of three parts: an initialization module, a node update module, and a normalization module. The initialization module first processes the features of complex numbers. Convert to real number characteristics

[0031]

[0032] in Indicates the real part, The imaginary part is represented; the initial representations of TX node features, RX node features, and edge features are transformed using a single-layer MLP, wherein the single-layer MLP uses a rectified linear unit as the activation function:

[0033]

[0034] in These are learnable network parameters;

[0035] The node update module consists of a T-layer message passing layer. Each message passing layer includes an aggregation layer and a combination layer. The aggregation layer collects features of neighboring nodes and edge features. The combination layer combines the features obtained from the aggregation layer and the node features to obtain and update the current node features.

[0036] In the node update module, the update strategy for the features of node TX in the t-th layer of the neural network is as follows:

[0037]

[0038] in yes The m-th line, yes The kth row, Let represent the set of neighboring RX nodes of the m-th TX node. It is the value output by the aggregation layer after aggregating adjacent RX nodes and edge features. AGG represents the aggregation function. They are two different MLPs.

[0039] The update strategy for the features of the RX node in the t-th layer of the neural network is as follows:

[0040]

[0041] in Let represent the set of neighboring TX nodes of the m-th RX node. It is the value output by the aggregation layer after aggregating adjacent TX nodes and edge features. AGG represents the aggregation function. They are two different MLPs.

[0042] The update strategy for edge features in the t-th layer of the neural network is as follows:

[0043]

[0044] in yes The (m,k)th element, It is the value output by the aggregation layer after aggregating the node features of adjacent TX and RX nodes. These are three MLPs with different parameters.

[0045] In the normalization module, the features of each node and edge in the graph are first represented using a single-layer MLP. Mapped to output variables The single-layer MLP uses rectified linear units as the activation function, specifically:

[0046]

[0047] These are trainable parameters;

[0048] Next, the output variable needs to be... Transform into the required complex variable Specifically:

[0049]

[0050] Finally, the output variables... Normalization is performed to ensure that the power meets the constraints, and the beamforming matrix of the current cell base station is finally obtained as follows:

[0051]

[0052] in P is the power of the signal transmitted by the base station within cell l according to the current beamforming matrix. max It is the maximum power that each base station can use to transmit signals.

[0053] Step 4 includes:

[0054] A large amount of random sample data is generated as training and test sets. The input of the dataset is the channel information of user terminals and base stations in the current cell, and the transmit power constraints of each base station. The output of the graph neural network is the beamforming matrix of the base stations in the current cell, and the loss function is the sum of the total rate of all users.

[0055] Applying the graph neural network framework to the problem of maximizing users and rate yields the following problem statement:

[0056]

[0057] The TX node features are set to the base station transmit power, the RX node features are set to the user-end noise, and the edge features are set to the channel information between antennas. The beamforming matrix output by each cell is applied to the overall optimization problem to output the beamforming matrix of each base station.

[0058] With the continuous development of mobile communication technology, 5G networks have brought faster data transmission speeds, lower latency, and greater network capacity. In this next-generation mobile communication network, multi-cell cooperative technology is one of the key factors in realizing its potential.

[0059] Multi-cell cooperative communication refers to the collaboration among multiple cells in a 5G communication system to achieve higher network capacity and a better user experience. In traditional wireless communication, different cells are often independent of each other. However, in multi-cell cooperative communication, cells can share and optimize resources, thereby improving communication efficiency and reliability. This invention proposes a distributed optimization scheme based on graph neural networks. By adjusting the graph neural network parameter output of each cell, it solves the beamforming design problem in downlink transmission, ensuring approximate performance while significantly reducing computation time and network overhead. This invention also proposes a machine learning method based on graph neural networks. Simulation verification shows that this method has good performance and can maximize users and data rates under power constraints.

[0060] Beneficial effects: Currently, centralized optimization schemes widely used in multi-cell networks require high computational complexity when applied to large-scale networks. Compared with traditional centralized optimization algorithms, the graph neural network method proposed in this invention can deliver better performance. The graph neural network-based learning method has excellent computational speed. Although the network training time is relatively long, the trained network can quickly output beamforming matrices under different channel environments.

[0061] This method also exhibits good generalization performance, with the trained model effectively adapting to changes in the number of cells or base stations and users. For network structures with good generalization performance, small changes in the number of base stations (BS) and user equipment (UE) will not result in significant performance loss. Furthermore, it closely resembles production environments in relatively stable communication settings.

[0062] In summary, the contribution of this invention aims to address some key challenges in implementing multi-cell MISO (Multiple Input Single Output) systems, such as collaborative power allocation optimization. By using advanced graph neural network algorithms and considering base station power constraints, the proposed solution can provide a practical solution for realizing high-performance, large-scale multi-cell MISO networks. Attached Figure Description

[0063] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0064] Figure 1 This is a simulation diagram of the application scenario of the method of the present invention.

[0065] Figure 2 This is a schematic diagram of the method of the present invention.

[0066] Figure 3 This is a schematic diagram of the overall GNN architecture of the method of this invention.

[0067] Figure 4 This is a comparative diagram of the cumulative distribution function of the method of the present invention.

[0068] Figure 5 This is a schematic diagram illustrating the generalization capability of the method of the present invention for the number of users within a cell.

[0069] Figure 6 This is a schematic diagram illustrating the generalization capability of the method of the present invention for the number of cells. Detailed Implementation

[0070] This invention primarily studies the problem of maximizing user transmission rate in multi-cell communication systems. It improves network communication performance by optimizing the beamforming matrix of base stations and employs a low-complexity algorithm based on graph neural networks to distribute and maximize network performance, including the following steps:

[0071] Step 1, Problem Description:

[0072] Consider as Figure 1The diagram shows a multi-cell wireless communication network, where one cell has L cells, each cell is equipped with one base station, and each base station serves N cells. l There are (l≤L) users, and the base station transmits signals to user terminal equipment via a downlink; each base station has N... t One antenna, and a single antenna for user terminal equipment.

[0073] Data symbols for all user terminal devices Represented as in The channel from the base station of cell l to user i within the cell is used It means that i≤N l The channel from the base station of cell j to user i outside the cell uses express; Let represent the beam vector when the base station of cell l sends a signal to the terminal device of user i within cell l. Then, the signal received by the terminal device of user i within cell l can be represented as:

[0074]

[0075] Signal-to-interference-plus-noise ratio (SINR) received by user i's terminal device within cell l li Represented as:

[0076]

[0077] in The additive white Gaussian noise represents the noise emitted by the i-th user in the l-th cell. This represents the variance of the noise.

[0078] The optimization problem is defined as follows: In order for the base station to improve the total data rate of all users in all cells under limited power constraints, the method of this invention addresses the following problem:

[0079]

[0080] Where st represents the set of constraints imposed on the optimization problem. This is the maximum transmit power that the base stations in cell l can use. The constraint ensures that the transmit power of each base station does not exceed its maximum power limit.

[0081] Step 2: This step proposes the graph structure model corresponding to the communication model of each cell.

[0082] like Figure 2As shown, a heterogeneous graph model is established using the communication network of a single cell. The types of nodes include transmitting nodes representing base stations within the cell and those of neighboring cells, and receiving nodes representing users within the cell. The notation in the heterogeneous graph neural network is as follows: TX nodes correspond to transmitting nodes in the heterogeneous graph, and RX nodes correspond to receiving nodes. Let represent the feature vectors of the i-th TX node and the k-th RX node. Where, d TX and d RX These represent the corresponding feature dimensions. Therefore, the feature matrices of the TX and RX nodes can be represented as follows: and Similarly, the characteristics of an edge can also be represented by a tensor. Where d E This is the edge feature dimension. On the other hand, the variables on the TX and RX nodes can be represented as... and in d′ represents the variable on the i-th TX node and the k-th RX node. TX ,d′ RX These represent the corresponding variable feature dimensions. Similarly, edge variables are represented as: d′ E Indicates the dimension of the edge variable.

[0083] For a distributed heterogeneous graph problem, it can be modeled as follows:

[0084]

[0085] Where φ is the mapping function from features to variables.

[0086] Step 3 includes: The overall GNN architecture of this algorithm is as follows Figure 3 As shown;

[0087] The initialization module converts the input features into initial node and edge feature representations. These representations are updated according to the node and edge update mechanism in the T node update module. During the update process, the dimension of each feature remains unchanged. Finally, the normalization module converts the feature representations into the required variables and outputs them.

[0088] In the initialization module, since the current neural network is not suitable for handling complex numbers, the features that may be complex numbers must first be excluded. Convert to real number characteristics Specifically, by extracting and concatenating the real and imaginary parts of a complex number, we obtain the following representation:

[0089]

[0090] in Indicates the real part, The imaginary part is represented. Then, a single-layer MLP using the Rectified Linear Unit (ReLU) as the activation function is used to transform the initial representations of the TX nodes, RX nodes, and edges:

[0091]

[0092] These are all learnable parameters.

[0093] In the node update module, edge convolution requires updating both node features and edge features. Meanwhile, based on the message passing model of graph neural networks, updating both node features and edge features requires two steps: aggregation and combination.

[0094] In the node update module, the update strategy for the features of node TX in the t-th layer of the neural network is as follows:

[0095]

[0096] in yes The m-th line, yes The kth row, Let represent the set of neighboring RX nodes of the m-th TX node. It is the value output by the aggregation layer after aggregating adjacent RX nodes and edge features. AGG represents the aggregation function. They are two different MLPs.

[0097] The update strategy for the features of the RX node in the t-th layer of the neural network is as follows:

[0098]

[0099] in Let represent the set of neighboring TX nodes of the m-th RX node. It is the value output by the aggregation layer after aggregating adjacent TX nodes and edge features. AGG represents the aggregation function. They are two different MLPs.

[0100] The update strategy for edge features in the t-th layer of the neural network is as follows:

[0101]

[0102] in yes The (m, k)th element, It is the value output by the aggregation layer after aggregating the node features of adjacent TX and RX nodes. These are three MLPs with different parameters.

[0103] The update strategies in the above three graph neural networks have been proven by existing work to have permutational variability, which can greatly reduce the number of training samples.

[0104] In the normalization module, the features of each node and edge in the graph are first represented using a single-layer MLP. Mapped to output variables The single-layer MLP uses rectified linear units as the activation function, specifically:

[0105]

[0106]

[0107] These are trainable parameters;

[0108] Next, the output variable needs to be... Transform into the required complex variable Specifically:

[0109]

[0110] Then, the output variables Normalization is performed to ensure that the power meets the constraints, and the beamforming matrix of the current cell base station is finally obtained as follows:

[0111]

[0112] in P is the power of the signal transmitted by the base station within cell l according to the current beamforming matrix. max It is the maximum power that each base station can use to transmit signals.

[0113] Step 4 uses a graph neural network-based method to optimize the beamforming matrix of the base station.

[0114] Sample data is generated as a training set and a test set. The input of the dataset is the channel information of the user terminal and each base station in the current cell, and the transmit power constraints of each base station. The output of the graph neural network is the beamforming matrix of the base station in the current cell, and the loss function is the sum of the total rate of all users.

[0115] Applying the graph neural network framework to the problem of maximizing users and rate yields the following problem statement:

[0116]

[0117] The TX node features are set to the base station transmit power, the RX node features are set to the user-end noise, and the edge features are set to the channel information between antennas. The beamforming matrix output by each cell is applied to the overall optimization problem to output the beamforming matrix of each base station.

[0118] Performance evaluation, using numerical results to demonstrate the effectiveness of the proposed multi-cell MISO communication system.

[0119] The specific simulation parameters are as follows: Consider a downlink wireless cellular network with uniformly distributed cell and user locations, where the base station is located at the center of each cell. The path loss is 30.5 + 36.7 log 10 (d)dB, where d is the distance between the base station antenna and the user antenna, and small-scale channels follow Rayleigh fading.

[0120] Unless otherwise specified, network training and simulation testing will be performed using the parameter settings in Table 1 below by default:

[0121] Table 1 Simulation Parameter Settings

[0122] parameter value Community area <![CDATA[1.5×1.5km 2 ]]> Cell radius 250m Base station height 30m User receiving antenna height 1.5m Noise power - -90dBm maximum transmission power of base station 30dBm Number of residential communities 4 Number of users in each community 4 Number of antennas per base station 10

[0123] For the graph neural network method, all aggregation functions are implemented using the maximum aggregation function, and all MLPs are implemented with three linear layers, each followed by a ReLU activation function. The node update module has three layers. During training, the number of epochs is set to 500, the batch size to 256, the training set size for the graph neural network is 10000, and the learning rate is set to 10. -4 The Adam optimizer was used. All experiments were implemented using PyTorch on an Intel Core i7-11800H (2.30GHz) and an NVIDIA GeForce RTX 3060 GPU (12GB).

[0124] Furthermore, to evaluate the proposed Distributed Edge Convolutional Graph Neural Network (DistributedEGNN) algorithm, this invention sets up three baseline methods for comparison: a Distributed Graph Neural Network (GNN) algorithm without edge convolutions and a Centralized Graph Neural Network (GNN) algorithm. The Distributed GNN algorithm without edge convolutions differs from the algorithm proposed in that its node update module focuses on message passing and aggregation at nodes, not at edges; that is, it only updates node features, not edge features. The Centralized Graph Neural Network (GNN) algorithm differs from the algorithm proposed in that it abstracts the entire multi-cell communication model into a graph model as the input for training the GNN, using a graph model that is not a single cell as the input.

[0125] Figure 4 This paper presents a comparison of the overall performance of three graph neural network (GNN)-based methods (Distributed EGNN, Distributed GNN, and Centralized GNN). Using the Weighted Minimum Mean-Square Error (WMMSE) algorithm as the benchmark, cumulative distribution curves for 200 random implementations are plotted for each of the three GNN-based algorithms. The horizontal axis represents the sum of all user rates, and the vertical axis represents the cumulative distribution probability. The graph shows that all three GNN-based algorithms produce higher sums of user rates compared to the WMMSE algorithm. Specifically, the Distributed EGNN algorithm improves the average sum of user rates by approximately 5.6% compared to the WMMSE algorithm, the Distributed GNN algorithm improves by approximately 2.7%, and the Centralized GNN algorithm improves by approximately 4.1%. The Distributed EGNN algorithm also shows a significant improvement in average users and speed compared to the Distributed GNN algorithm. This is because, compared to the message passing mode of traditional graph neural network algorithms, the introduction of edge convolution is more conducive to user nodes obtaining additional feature information of the base station serving them, such as interference channel information from the base station to other users. This message passing architecture can better adapt to the distributed multi-cell interference channel model, so that the message passing process can transmit useful features and interference features at the same time, and can better fit the mathematical modeling form of the proposed problem, thus showing better performance.

[0126] Figure 5This paper demonstrates the generalization performance of three graph neural network-based methods (Distributed EGNN, Distributed GNN, and Centralized GNN) on varying numbers of users within a cell. The neural networks were trained on a training set with 4 users per cell, and their performance was tested on test sets with 3, 5, 6, and 7 users per cell, respectively. Figure 4 In the graph, the horizontal axis represents the number of users within the cell, and the vertical axis represents the number of users and the data rate. Figure 5 As can be seen, all three graph neural network algorithms can adapt well to the increase or decrease in the number of users in the cell, demonstrating good generalization performance.

[0127] Figure 6 This paper demonstrates the generalization performance of three graph neural network-based methods (Distributed EGNN, Distributed GNN, and Centralized GNN) with varying cell counts. The neural networks were trained on a training set with 4 cells, and their performance was tested on test sets with 3, 5, 6, and 7 cells respectively. Figure 6 In the graph, the horizontal axis represents the number of cells, and the vertical axis represents the number of users and the data rate. Figure 6 As can be seen, all three graph neural network-based algorithms can adapt well to the increase or decrease of the number of cells, demonstrating good generalization performance. Moreover, compared with Centralized GNN, Distributed GNN performs better when the number of cells is 7. This is because as the number of cells increases, the network node size of the centralized algorithm increases much more than that of the distributed algorithm, which leads to a decrease in performance.

[0128] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of computer programs, i.e., software products. These computer program software products can be stored in a storage medium and include several instructions to cause a device containing a data processing unit (which may be a personal computer, server, microcontroller, MUU, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0129] This invention provides a distributed resource optimization method for multi-cell wireless networks based on graph neural networks. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A distributed resource optimization method for multi-cell wireless networks based on graph neural networks, characterized in that, Includes the following steps: Step 1: With the goal of maximizing the total transmission rate of user terminal devices in multiple cells, and in conjunction with the maximum power constraint of cell base stations, establish an optimization problem; Step 2: For the optimization problem established in Step 1, design a graph model as the input to the graph neural network. The graph model includes base station information of neighboring cells and base station and user information of the current cell. Step 3: Establish a message-passing graph neural network learning framework that includes edge embeddings; Step 4: Using the maximization of the system user transmission rate as the loss function, train and optimize the learning parameters of the graph neural network designed in Step 3. Step 5: Distribute graph neural networks on each cell base station to design beamforming matrices, thereby achieving distributed resource optimization of multi-cell wireless networks based on graph neural networks; Step 2 specifically includes: The problem can be modeled using graph neural networks as follows: A heterogeneous graph model G is established based on the communication topology model with cell l as the unit. l ={N TX N RX,l E L }, where N TX ={N l |l≤L} represents the transmitting node of the entire multi-cell communication system, i.e., the TX node. The transmitting node includes in-cell base stations and out-of-cell base stations. N TX =[n TX,1 ,...,n TX,L ] T Represents the variables on the TX node. N represents the characteristic matrix of the TX node; RX,l ={N i |i≤N l ,l≤L} represents the receiving nodes within cell l, i.e., RX nodes, i.e., user nodes, and N. RX,l =[n RX,l,1 ,...,n RX,l,L ] T Represents the variables on the RX node. E represents the characteristic matrix of the RX node; l This represents the edges between the nodes in the graph represented by the current cell l. Represents the variables on the edge. Represents the edge features of each edge; d TX d RX and d E They represent the feature dimensions, where It is a mapping function from features to variables; Step 3 includes: A graph neural network model based on a message passing model is constructed. The graph neural network model includes an initialization module, a node update module, and a normalization module. The graph neural network model converts complex features into real features and preprocesses them through an initialization module. It uses a node update module to aggregate neighborhood information to iteratively update the feature representations of TX nodes, RX nodes, and edges. The normalization module maps the output variables into a beamforming matrix in complex form while ensuring that it satisfies the constraints.

2. The distributed resource optimization method for multi-cell wireless networks based on graph neural networks according to claim 1, characterized in that, Step 1 includes: Imagine a network with L cells, each cell equipped with one base station, and each base station serving N cells. l There are (l≤L) users, and the base station transmits signals to user terminal equipment via a downlink; each base station has N... t One antenna, and a single antenna for user terminal equipment; The data symbol s for all user terminal devices is represented as in s li This represents the data sent by the base station of cell l to user i within the cell; the channel from the base station of cell l to user i within the cell is used... It means that i≤N l The channel from the base station of cell j to user i outside the cell uses Indicates; w li Let y represent the beam vector when the base station of cell l sends a signal to the terminal device of user i within cell l. Then, the signal y received by the terminal device of user i within cell l is... li It can be represented as: in The additive white Gaussian noise represents the signal-to-interference-plus-noise ratio (SINR) received by the terminal device of user i in cell l. li Represented as: in The additive white Gaussian noise represents the noise emitted by the i-th user in the l-th cell. Represents the variance of the noise; The problem of optimizing the total transmission rate of user terminal equipment is: Where st represents the set of constraints imposed on the optimization problem. SINR is the maximum transmit power that the base station in cell l can control. li The signal-to-interference-plus-noise ratio (SIR) received by the terminal device of user i within cell l. It is the beamforming matrix of the base station in cell l.

3. The distributed resource optimization method for multi-cell wireless networks based on graph neural networks according to claim 2, characterized in that, Step 3 includes: A graph neural network model based on a message passing model is constructed. The graph neural network model includes an initialization module, a node update module, and a normalization module. The initialization module first sets the complex feature (X) TX ,X RX,l ,Y l Convert to real number characteristics in Indicates the real part, The imaginary part is represented; the initial representations of TX node features, RX node features, and edge features are transformed using a single-layer multilayer perceptron (MLP), wherein the single-layer MLP uses a rectified linear unit as the activation function: in These are the learnable parameters of the network; The node update module consists of two parts: an aggregation layer and a combination layer. The aggregation layer collects features of adjacent nodes and edge features, and the combination layer combines the features obtained from the aggregation layer and the node features to obtain new features. In the node update module, the update strategy for the features of node TX in the t-th layer of the neural network is as follows: in yes The m-th line, yes The kth row, Let represent the set of neighboring RX nodes of the m-th TX node. It is the value output by the aggregation layer after aggregating adjacent RX nodes and edge features. AGG represents the aggregation function. They are two different MLPs; The update strategy for the features of the RX node in the t-th layer of the neural network is as follows: in Let represent the set of neighboring TX nodes of the m-th RX node. It is the value output by the aggregation layer after aggregating adjacent TX nodes and edge features. AGG represents the aggregation function. They are two different MLPs; The update strategy for edge features in the t-th layer of the neural network is as follows: in yes The (m, k)th element, It is the value output by the aggregation layer after aggregating the node features of adjacent TX and RX nodes. These are three MLPs with different parameters; In the normalization module, the features of each node and edge in the graph are first represented using a single-layer MLP. Mapped to output variables The single-layer MLP uses rectified linear units as the activation function, specifically: These are trainable parameters; Next, the output variable needs to be... Transform into the required complex variable Specifically: For output variables Normalize to (N) TX N RX，l E l This ensures that its power satisfies the constraints; Obtain the beamforming matrix W of the current cell base station l Specifically, this manifests as follows: in P is the power of the signal transmitted by the base station within cell l according to the current beamforming matrix. max It is the maximum power that each base station can use to transmit signals.

4. The distributed resource optimization method for multi-cell wireless networks based on graph neural networks according to claim 3, characterized in that, Step 4 includes: Sample data is generated as training and testing sets. The input of the dataset is the channel information of user terminals and base stations in the current cell, and the transmit power constraints of each base station. The output of the graph neural network is the beamforming matrix of the base stations in the current cell. The loss function is the sum of the total rates of all users. Applying the graph neural network framework to the problem of maximizing users and rate yields the following problem statement: The TX node feature is set to the base station's maximum transmit power, the RX node feature is set to user-end noise, and the edge feature is set to the channel information between antennas. The beamforming matrix output by each cell's network is applied to the overall optimization problem to output the beamforming matrix of each base station.

Images