A method for realizing V2V link rate and maximization based on a graph neural network

Abstract

The application discloses a kind of based on graph neural network realizes V2V link rate and maximization method, including steps as follows: with small scale fading and large scale fading modeled vehicle networking transmission channel model with first-order Gaussian Markov process modeling;Modeling graph neural network model;Establish the total transmission rate maximization model of V2V link, propose to solve vehicle communication transmit power allocation problem with graph neural network optimization weighted least mean square error algorithm, and guarantee constraint condition such as delay and reliability;Graph neural network optimization weighted least mean square error algorithm not only retains the key data-driven components of weighted least mean square error algorithm to solve vehicle transmit power allocation, but also optimizes the key parameters in weighted least mean square algorithm through graph neural network, accelerates the convergence performance of network and improves the total transmission rate of V2V link.

Description

Technical Field

[0001] This invention relates to the field of vehicle-to-everything (V2V) communication resource allocation, and specifically to a method for maximizing V2V link speed and rate in a vehicle-to-everything (V2V) network based on a graph neural network, while ensuring V2V link reliability and latency. Background Technology

[0002] Vehicle-to-everything (V2X) communication technology relies on various vehicle devices, such as sensors and roadside units, to perceive information around the vehicle and interact with other nearby vehicles. These services can reduce traffic accidents, improve the driving experience, and make vehicle operation safer and more environmentally friendly. Vehicle-to-vehicle (V2V) communication is one of the important vehicle communication methods in V2X, enabling information transmission and exchange between vehicles. During the transmission of vehicle communication information, under strict requirements for latency and reliability, it is necessary to maximize the throughput of vehicle transmission to meet the actual needs of vehicle users such as video streaming and application data transmission. However, the high speed of vehicles in the V2V network environment is a reason for the difference between V2V networks and traditional cellular networks. Reasonable and efficient resource allocation for V2V link communication in high-speed moving vehicle networks is a crucial issue in V2V link system design.

[0003] In vehicle communication, power and bandwidth are fundamental resources. Allocating these resources in the randomly changing channel environment of a vehicle is crucial for the smooth operation of the vehicle system. Power and bandwidth resources also play a significant role in determining the transmission rate of the vehicle channel. In recent years, new energy vehicles have also developed rapidly. The vehicle is not only a means of transportation but also possesses the function of an intelligent terminal. Efficient power allocation strategies are important not only for reducing interference between multiple vehicle users but also for improving vehicle battery life.

[0004] The management and allocation of vehicle communication resources play a crucial role in vehicle-to-everything (V2X) networks. In high-speed moving vehicles and complex lane environments, the vehicle power allocation problem, while ensuring latency and reliability, is often NP-hard. Therefore, suboptimal algorithms for solving this problem have been explored to reduce computational complexity, including greedy heuristics and sequential connection selection algorithms. However, these suboptimal algorithms have multiple local optima, making performance difficult to control. Furthermore, the high complexity and long execution time of these algorithms make them unsuitable for the high real-time requirements of latency-sensitive V2X transmission systems. In practical applications, these algorithms require highly accurate channel coefficients. In dense vehicle transmission systems, estimating numerous transmission channel parameters is both time-consuming and resource-intensive, making them impractical for real-world vehicle transmission systems.

[0005] Currently, research strategies for vehicle power allocation based on graph neural networks are gradually developing. However, shortcomings still exist. While using GNNs to extract low-dimensional features of vehicle users and then employing reinforcement learning for resource allocation (Resource Allocation based on Graph Neural Networks in Vehicular Communications, Ziyan He, Liang Wang1, Hao Ye*, Geoffrey Ye Li, and Biing-Hwang Fred Juang, Department of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA), the reliability and latency of the vehicle communication process are not considered. For high-speed driving environments, ensuring both latency and reliability for vehicle users is particularly crucial.

[0006] Existing research on V2V link power allocation rarely utilizes the combination of graph neural networks (GNNs) and the WMMSE algorithm for power allocation strategies. This paper proposes using a DNN network combined with the WMMSE algorithm to apply a power allocation strategy to V2V links, maximizing the total transmission rate for both V2V and V2I users (Deep Neural Network Based Resource Allocation for V2x Communications, Jin Gao, Muhammad RAKhandaker, Faisal Tariq, Kai-Kit Wong, and Risala T. Khan, School of Engineering and Physical Sciences, Heriot-Watt University, Edinburgh, United Kingdom). Compared to DNNs, which cannot utilize the underlying network topology of vehicular networks, the GNN method fully leverages information from vehicle users and nearby vehicles, and experimental results demonstrate its superior effectiveness. Summary of the Invention

[0007] The purpose of this invention is to overcome the aforementioned deficiencies in the prior art and to propose a method for optimizing the WMMSE algorithm using a graph neural network to maximize V2V link speed.

[0008] The present invention is achieved by at least one of the following technical solutions.

[0009] A method for maximizing V2V link speed based on graph neural networks includes the following steps:

[0010] Establish vehicle network communication scenarios;

[0011] Construct a convolutional graph neural network; embed the graph neural network into the WMMSE algorithm model;

[0012] This combines the reliability and latency constraints of V2V link transmission with the WMMSE algorithm model optimized by graph neural networks;

[0013] The WMMSE algorithm model optimized by graph neural network is trained centrally and implemented in a distributed manner on V2V links.

[0014] Furthermore, the vehicle network communication scenario includes a vehicle-to-everything (V2X) model and a channel model. Establishing a V2X communication scenario includes the following steps:

[0015] The vehicle-to-everything (V2X) scenario is a two-way multi-lane model: a communication base station is deployed on one side of the road, which covers the vehicles within the lane. There are M vehicles traveling at an average speed V in the lane, and among the M vehicles, there are K V2V links for communication.

[0016] A channel power gain model is established, in which the formula for calculating large-scale fading is:

[0017]

[0018] In the formula: L i，j This represents the large-scale channel fading during V2V communication between transmitter i and receiver j; A is the path fading coefficient; β i，j The shadow fading coefficient d follows a log-normal distribution. i，j λ is the distance from transmitter i to receiver j in the V2V link; λ is the path fading index.

[0019] A small-scale fading channel model is modeled using a first-order Gaussian Markov process in each transmission slot:

[0020]

[0021] Where: g i，j G′ represents small-scale channel fading. i，j To predict the channel gain, an exponential distribution with unit mean is used; e i，j The estimation error also follows an exponential distribution with a unit mean, similar to the predicted channel gain g′. i，j They are mutually independent; the coefficient ∈ is used to characterize the lack of correlation between two consecutive time slot channels;

[0022] The channel increment of a V2V link is represented as:

[0023] h i，j =Li，j g i，j

[0024] Each V2V link shares its channel information with the others. The input to the graph neural network constructed from this is the channel gain matrix H of the V2V link, including the channel gain h of the V2V link itself. ii Channel gain h of neighboring nodes ij (i≠j), where the channel gain of the neighboring node is the interference channel gain transmitted by other V2V links;

[0025]

[0026] The signal-to-interference-plus-noise ratio (SIR) of a V2V link is γ k The calculation formula is as follows:

[0027]

[0028] Where: h k p represents the channel power gain from the vehicle transmitter to the vehicle receiver in the k-th V2V link; k σ represents the transmit power of the k-th V2V link; 2 h represents the power of additive white Gaussian noise; k′ p represents the interference channel power gain from the transmitter in other V2V links to the receiver in the k-th V2V link; k′ Interference power;

[0029] According to Shannon's formula, the transmission rate C of the k-th V2V link... k for:

[0030] C k =Blog2(1+γ) k )

[0031]

[0032] In the formula: B is the bandwidth allocated by the base station to the k-th V2V link; γ k For signal-to-interference-plus-noise ratio;

[0033] The optimization problem is to maximize the total transmission rate of the V2V link:

[0034]

[0035] stC1:Pr{γ k ≤γ0}≤P0

[0036] C2:Pr{t k ≥T0}≤P0

[0037] C3:∑ k′≠k h k′p k′ ≤I0

[0038] C4: 0 ≤ p k ≤P max

[0039] The first constraint C1 is the transmission reliability requirement of the V2V link, γ0 is the signal-to-interference-plus-noise ratio threshold specified for the V2V link at the transmitting end, and P0 represents the maximum interruption probability threshold of the V2V link.

[0040] The second constraint C2 is the communication latency requirement for the V2V link, t k Let T0 be the communication delay of the k-th V2V link, and T0 be the maximum communication transmission delay.

[0041] The third constraint C3 requires that the interference from other V2V links be less than the interference threshold, where I0 is the interference threshold.

[0042] The fourth constraint, C4, is the maximum power limit for the V2V link transmitter, p k p is the transmit power of the k-th V2V link. max This represents the maximum transmit power allowed at the V2V link transmitter.

[0043] Furthermore, the coefficient ∈ is expressed as follows:

[0044] ∈=J0(2πf D T)

[0045] In the formula, J0 is a Bessel function of the first kind, order 0; f D =vf c / c represents the maximum Doppler frequency shift, f c Where is the carrier frequency, v is the vehicle speed; c = 3 × 10 8 Let be the wave propagation speed; T be the periodic feedback delay.

[0046] Furthermore, the latency of the V2V link is analyzed, including the following steps:

[0047] Delay constraint analysis:

[0048] The delay t of the k-th V2V link k The calculation formula is:

[0049]

[0050] In the formula, D k C represents the number of bits required to transmit the message in the k-th V2V link communication; k Let t be the transmission rate of the k-th V2V link; V2V links need to meet latency requirements for communication, otherwise communication will fail. The latency t of the k-th V2V link is given. kThe following constraints must be met:

[0051] Pr{t k ≥T max}≤P0

[0052] In the formula, T max Pr{t} represents the maximum transmission delay threshold of a V2V link. k ≤T max} represents the probability that the transmission delay of the k-th V2V link is less than the maximum transmission delay threshold, and P0 is the maximum interruption probability threshold.

[0053] Substituting and simplifying, we get:

[0054]

[0055] Simplifying the above equation, we get:

[0056]

[0057] The above formula yields the signal-to-interference-plus-noise ratio (SIR / NNR) threshold γ allowed for V2V communication. k To ensure the quality of V2V communication, the probability of interruption during communication must be less than P0. If the above reliability and latency requirements are not met, the V2V link will experience a communication interruption.

[0058] The interruption of a V2V link is defined as follows: when the signal-to-interference-plus-noise ratio (SINNR) of the V2V link is lower than the SINNR threshold γ0, and the message cannot be effectively demodulated during the communication process, the link is interrupted; when the communication delay of the V2V link is greater than the delay threshold T0, the link is interrupted. At the same time, the interference of other V2V links is ensured to be less than the interference threshold I0, otherwise the interference will be too large and affect the communication process.

[0059] To ensure the reliability and latency of V2V links, the signal-to-interference-plus-noise ratio (SINNR) threshold γ0 for the k-th V2V link is defined, and its magnitude satisfies the SINNR γ k and Between, it is represented as:

[0060]

[0061] Furthermore, the reliability of the V2V link is analyzed, including the following steps:

[0062] Series Theorem: Let z1, z2, ..., z n They are independent random variables that follow an exponential distribution, and their mean satisfies but

[0063]

[0064] In the formula: z i Let E[z] be the i-th random variable that follows an exponential distribution;i ] represents the i-th random variable z in the series theorem. i The mean; λ i λ1 is the reciprocal of the mean of the i-th random variable, and λ1 is the reciprocal of the mean of the 1-th random variable.

[0065] Substituting the signal-to-interference-plus-noise ratio (SIR) condition and the reliability requirement of the V2V link into the series theorem, we obtain:

[0066]

[0067] Further simplification:

[0068]

[0069]

[0070] V2V link transmit power lower bound p min for:

[0071]

[0072] From the above expression, we get p min The constraints are integrated and transformed, that is:

[0073] p min ≤p k ≤p max

[0074] In the formula: p max This represents the maximum transmit power allowed at the V2V link transmitter.

[0075] The nonlinear parameter α guarantees that the output parameter of the WMMSE algorithm model optimized by the graph neural network is within [p]. min p max Within the specified range, the reliability constraints of the V2V link are met.

[0076] Furthermore, the convolutional graph neural network includes a two-layer convolutional network GCN, expressed as follows:

[0077]

[0078] ψ=a2(diag(H)Zω 21 +HZω 22 )

[0079] In the formula: Z is the output variable of the first layer of the convolutional graph neural network; ψ is the output variable of the iterative convolutional graph neural network; X is the input matrix of the graph neural network, which is all 1s; H is the channel power gain matrix of the V2V link; diag(H) is the diagonal value of H, i.e., H ij;ω 11 ω 12 ω 21 ω 22 For weight parameters, the set of weight parameters [ω] of the two-layer graph neural network for each V2V link. 11 ω 12 ω 21 ω 22 In the vehicle networking system, it is shared; a1 is the activation function ReLU (the rectified linear unit); a2 is the sigmoid activation function, which ensures that the output is between 0 and 1.

[0080] Furthermore, the loss function of the convolutional graph neural network is as follows:

[0081]

[0082] In the formula: there are K V2V links in the vehicle-to-everything (V2V) system; the mean value E represents the total transmission rate and mean value of the V2V links; h k p represents the channel power gain from the vehicle transmitter to the vehicle receiver in the k-th V2V link; k σ represents the transmit power of the k-th V2V link; 2 h represents the power of additive white Gaussian noise; k′ p represents the interference channel power gain from the transmitter in other V2V links to the receiver in the k-th V2V link; k′ Interference power;

[0083] The loss function is defined as the negative of the total transmission rate and mean of the V2V link, and stochastic gradient descent is used to optimize the loss.

[0084] Furthermore, embedding the graph neural network into the WMMSE algorithm model includes the following steps:

[0085] The WMMSE algorithm model applied to V2V links is constructed as follows:

[0086]

[0087]

[0088]

[0089] C2:Pr{γ k ≤γ0}≤P0

[0090] C3:Pr{t k ≥T0}≤P0

[0091] C4:∑ k′≠k hk′ p k′ ≤I0

[0092] In the formula: I i M is the input variable for vehicle i; i An intermediate variable representing vehicle i; O i Output variables for vehicle i, O j Output variable for vehicle j; E i h is the mean square error value of vehicle i; i σ represents the channel power gain of vehicle i in the V2V link. 2 h represents the power of additive white Gaussian noise; ij Let p be the channel power gain transmitted from vehicle j to vehicle i; C1 to C4 represent different constraints, where: p min p max γk represents the rated minimum transmit power and rated maximum transmit power of the V2V link, respectively; γk is the signal-to-interference-plus-noise ratio (SIR) of the k-th V2V link, γ0 is the SIR threshold, and P0 represents the maximum outage probability threshold of the V2V link; t k Let T0 be the communication delay of the k-th V2V link, and T0 be the maximum communication transmission delay; I0 be the total interference threshold for other V2V links.

[0093] When input variable I i intermediate variable M i and output variable O i When both are optimal, we get:

[0094]

[0095] That is, the optimal transmit power of the i-th V2V is the output variable value O. i The minimum mean square error problem is solved using the block coordinate descent method. It is proven that when two of the three variables (input parameter, output parameter, and intermediate parameter) are fixed, the objective is convex in each variable. This transforms the maximum rate problem in the V2V link into the minimum mean square error problem. This represents the optimal transmit power for the i-th V2V link.

[0096] Furthermore, the WMMSE algorithm model uses the square root of the maximum transmit power of the V2V link. Using the input as input, after l = {1...L} iterations, the optimal V2V link transmit power is obtained. The specific process of the l-th layer in the k-th V2V link is as follows:

[0097] (1) Input variables:

[0098]

[0099] In the formula: Let h be the input variable for the l-th iteration in the k-th V2V link; k It is represented as the channel gain coefficient of the k-th V2V link; σ is represented as the output variable in the (l-1)th iteration of the k-th V2V link; 2 Let k represent the additive white Gaussian noise power; k′ represents other V2V links, then This represents the output variables of the other V2V links in (l-1) iterations;

[0100] (2) Intermediate variables:

[0101]

[0102] In the formula: This is represented as an intermediate variable in the l-th iteration of the k-th V2V link; and parameters These are trainable parameters; h is the input variable. k It is represented as the channel gain coefficient of the k-th V2V link; This is represented as the output variable of the previous iteration in the k-th V2V link;

[0103] (3) Output variables:

[0104]

[0105] In the formula: Let h be the output variable of the l-th iteration in the k-th V2V link; α is the nonlinear parameter; k Let h be the channel gain coefficient of the k-th V2V link. k′k The interference channel gain of other V2V links on the k-th V2V link; Let be the input variable in the (l-1)th iteration of the k-th V2V link. This is represented as the input variable for the l-th iteration in other V2V links; Let be the intermediate variable in the l-th iteration of the k-th V2V link. For the intermediate variable in the l-th iteration of other V2V links;

[0106] In the three expressions above, the input variables depend entirely on the channel state of the V2V link receiver, while the parameters in the intermediate variables... and parameters These are trainable parameters; the nonlinear parameter α in the output variables ensures that the magnitude of the output parameters of each layer is within [p]. min p max Within the specified range, the constraints are guaranteed to be satisfied.

[0107] Furthermore, the WMMSE algorithm model, implemented in a distributed manner on V2V links, includes: both the transmitter and receiver of the V2V link possessing channel information for all V2V links; the receiver calculating the input variables. and intermediate variables The information is sent to the sending end; the sending end calculates the output variable. The optimal transmit power p for the V2V link is obtained based on the vehicle index information. * This ensures that the total transmission rate of the V2V link is maximized while meeting the reliability and latency requirements of the V2V link.

[0108] Compared with existing technologies, the beneficial effects of the present invention are as follows:

[0109] To address the power allocation problem in vehicular networks (V2V) systems while ensuring V2V link reliability and latency, a convolutional graph neural network-optimized WMMSE algorithm model is designed to maximize the total V2V transmission rate. V2V latency and reliability constraints are represented by interruption probabilities. Through series theorems and constraint transformations, these constraints are combined with the convolutional graph neural network-optimized WMMSE algorithm model. While maintaining V2V link communication reliability and latency, the convergence speed of the WMMSE algorithm is accelerated, increasing the total V2V link transmission rate while reducing the algorithm's execution time. Experimental results show that the graph neural network-optimized WMMSE algorithm achieves better overall transmission rate performance as vehicle speed and the number of V2V pairs change. Attached Figure Description

[0110] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0111] Figure 1 This is a graph showing the change in network training loss with the number of training iterations in this embodiment;

[0112] Figure 2 This is a comparison chart of the total transmission rate of the V2V link for each power allocation method in this embodiment;

[0113] Figure 3 This is a comparison graph showing the change in the total transmission rate of the V2V link with vehicle speed in this embodiment;

[0114] Figure 4 This is a statistical graph showing the change in the total V2V link transmission rate with vehicle speed using the graph neural network-optimized WMMSE algorithm model in this embodiment.

[0115] Figure 5This is a statistical graph showing the total transmission rate of the V2V link as a function of vehicle speed using the MLP algorithm model in this embodiment;

[0116] Figure 6 This is a statistical graph of the total V2V link transmission rate as a function of vehicle speed using the WMMSE algorithm model in this embodiment;

[0117] Figure 7 This is a graph comparing the total transmission rate of the V2V link with the number of V2V link pairs in this embodiment;

[0118] Figure 8 This is a comparison chart of the average V2V link rate with the number of V2V link pairs in this embodiment;

[0119] Figure 9 This is a statistical graph showing the change in the total transmission rate of the V2V link with the maximum interruption probability in this embodiment;

[0120] Figure 10 This is a statistical graph showing the change in the total transmission rate of the V2V link as a function of the total interference threshold in this embodiment.

[0121] Figure 11 This is a statistical graph showing the change in the total transmission rate of the V2V link as a function of the amount of data transmitted in this embodiment.

[0122] Figure 12 This is a flowchart of a method for maximizing V2V link speed based on a graph neural network in this embodiment. Detailed Implementation

[0123] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0124] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0125] This embodiment presents a method for maximizing V2V link speed based on graph neural networks. While ensuring V2V link latency and reliability, the graph neural network-optimized weighted minimum mean square error algorithm solves the problem of maximizing V2V link speed in vehicle-to-everything (V2V) networks, including the following steps:

[0126] S1. Establish a vehicle network communication scenario. In this embodiment, the scenario is a two-way multi-lane road. A communication base station is deployed on one side of the road, covering the vehicles within that lane. There are M vehicles traveling at an average speed V in the lane, and K V2V links are used for communication among these M vehicles. The V2V links within the base station's coverage area share subframe resources in the frequency band, thus causing interference during vehicle communication. In this channel environment, the channel gain considers the effects of large-scale and small-scale fading. A vehicle communication power allocation model is constructed, which calculates the signal-to-interference-plus-noise ratio (SIR) of the V2V links as γ. k To ensure that the probability of a vehicle's signal-to-interference-to-noise ratio (SNR) being less than the SNR threshold is constrained by the maximum outage probability threshold, and with the constraint that the vehicle's transmit power cannot exceed the maximum transmit power, a mathematical model is constructed with the optimization objective of maximizing the total transmission rate of the V2V link in the vehicle communication system. Based on this model, the transmit power of the vehicle's transmitter can be strategically planned, determining the continuous allocation vector in vehicle communication within the system, thus solving the power allocation problem in the vehicle-to-everything (V2V) system and maximizing the total transmission rate of the V2V link.

[0127] S2. Construct a graph neural network model. This model not only needs to be compatible with the WMMSE algorithm (by embedding it into the WMMSE model to improve the algorithm's computation and convergence speed), but also needs to adapt to the rapidly changing dynamic vehicle transmission communication channel model in the vehicular network environment. Based on these requirements, a graph convolutional network was designed with vehicle channel gain as input and the core parameters of the WMMSE algorithm as output. The loss function of the graph neural network is the negative of the average total transmission rate of the V2V link, and it is optimized using stochastic gradient descent.

[0128] S3. Embed the graph neural network into the WMMSE algorithm structure. The graph neural network-optimized WMMSE algorithm uses the output of the graph neural network as a parameter in each computational iteration. and parameters The value of is adjusted to accelerate the convergence speed of the WMMSE algorithm and reduce its execution time. The weighted minimum mean square error algorithm optimized by graph neural networks can obtain a suitable system power allocation vector. Applying this power allocation vector to vehicle networks yields the maximum total transmission rate of V2V links, and this method has achieved good results in specific experiments.

[0129] S4. The reliability and delay constraints of V2V link transmission are combined with the WMMSE algorithm optimized by a graph neural network. The signal-to-interference-plus-noise ratio (SIR) reliability constraint of the V2V link is simplified by solving a series theorem to obtain the V2V link transmit power range. The output variables in the graph neural network-optimized WMMSE algorithm are... Design a nonlinear parameter α to ensure that the V2V transmit power is within the reliability range.

[0130] S5, the WMMSE algorithm optimized by graph neural networks is trained centrally and implemented in a distributed manner on V2V links. The receiving end calculates the input variables. and intermediate variables The information is sent to the sending end; the sending end calculates the output variable. The optimal transmit power for the V2V link is obtained based on the vehicle index information.

[0131] As a preferred embodiment, step S1 is as follows:

[0132] S11. Construct a network model for vehicle-to-everything (V2X) communication, such as... Figure 1 As shown.

[0133] The channel power gain model established initially considers both large-scale and small-scale fading to better reflect real-world vehicle-to-everything (V2X) scenarios. Large-scale fading is primarily affected by path loss and shadowing, while small-scale fading is influenced by multipath propagation. The formula for calculating large-scale fading is as follows:

[0134]

[0135] In the formula: L i，j The large-scale channel gain is expressed as V2V for communication between transmitter i and receiver j; A is the path fading coefficient; β i，j The shadow fading coefficient d follows a log-normal distribution. i，j λ is the distance parameter from the transmitter i to the receiver j in the V2V link; λ is the path fading index.

[0136] In vehicular network (V2L) communication, channel state information remains constant within the transmission time slots. Therefore, it is assumed that the base station can obtain large-scale fading information relatively well. However, because vehicles travel at high speeds in V2L systems, small-scale fading is significantly affected by Doppler shift. The base station can only obtain the estimated channel gain and estimation error distribution for small-scale fading. A small-scale fading channel model is modeled using a first-order Gaussian Markov process in each transmission time slot, as follows:

[0137]

[0138] In the formula: g′ i，j To predict the channel gain, an exponential distribution with unit mean is used; e i，j The estimation error also follows an exponential distribution with a unit mean, similar to the predicted channel gain g′. i，j They are mutually independent. The coefficient ∈ is used to characterize the uncorrelation between two consecutive time slot channels. Assuming that the coefficient ∈ remains consistent in the vehicle-to-everything (V2V) system and can be accurately known by both the receiver and transmitter in the V2V link, the specific expression for the coefficient ∈ is given below:

[0139] ∈=J0(2πf D T)

[0140] In the formula, J0 is a Bessel function of the first kind, order 0; f D =vf c / c represents the maximum Doppler frequency shift, f c The carrier frequency is v; the vehicle speed is c = 3 × 10⁻⁶. 8 Let be the wave propagation speed; T be the periodic feedback delay;

[0141] The channel increment of a V2V link is represented as:

[0142] h i，j =L i，j g i，j

[0143] The channel gain matrix H is constructed as the input to the graph neural network. Each V2V link shares its channel information with the others; therefore, the input to the constructed graph neural network is the channel gain matrix H of the V2V link, including its own channel gain h. ii Channel gain h of neighboring nodes ij (i≠j), where the channel gain of the neighboring node is the interference channel gain transmitted by other V2V links.

[0144]

[0145] The channel gain matrix H is constructed as the input to the graph neural network. Each V2V link shares its channel information with the others; therefore, the input to the constructed graph neural network is the channel gain matrix H of the V2V link, including its own channel gain H. ii Channel gain H of neighboring nodes ij (i≠j), where the channel gain of the neighboring node is the interference channel gain transmitted by other V2V links.

[0146] The signal-to-interference-plus-noise ratio (SIR) γ of the S12 and V2V links k The calculation formula is as follows:

[0147]

[0148] In the above formula, h k p represents the channel power gain from the vehicle transmitter to the vehicle receiver in the k-th V2V link; k σ represents the transmit power of the k-th V2V link; 2 h represents the power of additive white Gaussian noise; k′ p represents the interference channel power gain from the transmitter in other V2V links to the receiver in the k-th V2V link; k′ Interference power coefficient;

[0149] S13. According to Shannon's formula, the transmission rate C of the k-th V2V link... k for:

[0150] C k =Blog2(1+γ) k )

[0151]

[0152] In the above formula, B is the bandwidth allocated by the base station to the k-th V2V link, and γ k For signal-to-interference-plus-noise ratio.

[0153] S14. The objective of this invention is to find a suitable instantaneous transmit power for V2V links to maximize the overall transmission rate of the vehicle network. Therefore, the power allocation problem is summarized as follows:

[0154] The optimization problem is to maximize the total transmission rate of the V2V link:

[0155]

[0156] stC1:Pr{γ k ≤γ0}≤P0

[0157] C2:Pr{t k ≥T0}≤P0

[0158] C3:∑ k′≠k h k′ p k′ ≤I0

[0159] C4: 0 ≤ p k ≤p max

[0160] The constraint C1 is the transmission reliability requirement of the V2V link, γ0 is the signal-to-interference-plus-noise ratio threshold specified for the transmitter of the V2V link, and P0 represents the maximum outage probability threshold of the V2V link.

[0161] Constraint C2 is the communication latency requirement for the V2V link, t k Let T0 be the communication delay of the k-th V2V link, and T0 be the maximum communication transmission delay.

[0162] Constraint C3 requires that interference from other V2V links be less than a specific threshold, where I0 is the interference threshold.

[0163] Constraint C4 is the maximum power limit of the V2V link transmitter, p max This indicates the maximum power allowed at the V2V link transmitter.

[0164] Traditional deep learning methods have achieved great success in extracting features from Euclidean space, but their performance on non-Euclidean data is less than satisfactory. This is because graphs are irregular in shape, with varying numbers of nodes in each graph. Furthermore, deep learning algorithms require samples to be independent, while graph data samples are interconnected.

[0165] Constructing a convolutional graph neural network includes the following steps:

[0166] As another preferred embodiment, a graph neural network model was constructed, consisting of a two-layer GCN (Graph Neural Network) to solve the problem, as shown in the following expression:

[0167]

[0168] ψ=a2(diag(H)Zω 21 +HZω 22 )

[0169] In the formula: X is the input matrix of the graph neural network, consisting of all 1s; ω 11 ω 12 ω 21 ω 22 For weight parameters, the set of weight parameters [ω] of the two-layer graph neural network for each V2V link. 11 ω 12 ω 21 ω 22 In the vehicle-to-everything (V2V) system, this function is shared; a1 is the ReLU activation function (the rectified linear unit); a2 is the sigmoid activation function, ensuring the output is between 0 and 1; H is the channel power gain of the V2V link; diag(H) is the diagonal value of H, i.e., H2. ij ;

[0170] The loss function is as follows:

[0171]

[0172] As a preferred embodiment, the loss function is defined as the negative of the total transmission rate and mean of the V2V link, and the loss is optimized using stochastic gradient descent.

[0173] The WMMSE algorithm model applied to V2V links is as follows:

[0174]

[0175]

[0176] stC1:Pr{γ k≤γ0}≤P0

[0177] C2:Pr{t k ≥T0}≤P0

[0178] C3:∑ k′≠k h k′ p k′ ≤I0

[0179] C4: 0 ≤ p k ≤p max

[0180] In the above expression: I i M is the input variable for vehicle i; i An intermediate variable representing vehicle i; O i Output variable for vehicle i; E i h is the mean square error value of vehicle i; i h represents the channel power gain of vehicle i in the V2V link. ij Let be the channel power gain from vehicle j to vehicle i;

[0181] The constraint C1 is the transmission reliability requirement of the V2V link, γ0 is the signal-to-interference-plus-noise ratio threshold specified for the transmitter of the V2V link, and P0 represents the maximum outage probability threshold of the V2V link.

[0182] Constraint C2 is the communication latency requirement for the V2V link, t k Let T0 be the communication delay of the k-th V2V link, and T0 be the maximum communication transmission delay.

[0183] Constraint C3 requires that interference from other V2V links be less than a specific threshold, where I0 is the interference threshold.

[0184] Constraint C4 is the maximum power limit of the V2V link transmitter, p max This indicates the maximum power allowed at the V2V link transmitter.

[0185] When input variable I i intermediate variable M i and output variable O i When all are optimal, we can obtain:

[0186]

[0187] That is, the optimal transmit power for the i-th V2V is the square of the output variable value. The minimum mean square error problem can be solved using the block coordinate descent method. It can be proven that when two of the three variables (input parameter, output parameter, and intermediate parameter) are fixed, the objective is convex in each variable. Therefore, the problem of maximizing the rate in a V2V link can be transformed into a minimum mean square error problem.

[0188] The S42 and WMMSE algorithm models use the square root of the maximum transmit power of the V2V link. Using the input as input, after l = {1...L} iterations, the optimal V2V link transmit power is obtained. The specific process of the lth iteration in the kth V2V link is as follows:

[0189] (1) Input variables:

[0190]

[0191] In the formula: Let h be the input variable for the l-th iteration in the k-th V2V link; k It is represented as the channel gain coefficient of the k-th V2V link; σ is represented as the output variable of the previous iteration in the k-th V2V link; 2 Let k represent the additive white Gaussian noise power; k′ represents other V2V links, then This represents the output variable of the previous iteration of other V2V links;

[0192] (2) Intermediate variables:

[0193]

[0194] In the formula: This is represented as an intermediate variable in the l-th iteration of the k-th V2V link; and parameters These are trainable parameters; h is the input variable. k It is represented as the channel gain coefficient of the k-th V2V link; This is represented as the output variable of the previous iteration in the k-th V2V link;

[0195] (3) Output variables:

[0196]

[0197] In the formula: Let h be the output variable of the l-th iteration in the k-th V2V link; α is the nonlinear parameter; k Let h be the channel gain coefficient of the k-th V2V link. k′k The interference channel gain of other V2V links on the k-th V2V link; Let be the input variable in the (l-1)th iteration of the k-th V2V link. This is represented as the input variable for the l-th iteration in other V2V links; Let be the intermediate variable in the l-th iteration of the k-th V2V link. For the intermediate variable in the l-th iteration of other V2V links;

[0198] In the three expressions above, the input variables It depends entirely on the channel state at the V2V link receiver. (Intermediate variable) Parameters in and parameters As trainable parameters, they are derived from the training results of the graph neural network, which can accelerate network convergence while ensuring good network performance; output variables The nonlinear parameter α ensures that the magnitude of the output parameter of each layer is within [p]. min p max Within the specified range, the constraints are guaranteed to be satisfied;

[0199] The weighted minimum mean square error algorithm optimized by graph neural networks can obtain the optimal vehicle transmission power allocation strategy through hierarchical iteration to maximize the total transmission power of the V2V link.

[0200] The classic weighted least mean square error algorithm can effectively solve the power allocation problem in V2V links, but it requires multiple iterative calculations, resulting in high time complexity and execution time. Therefore, reducing the operator complexity of this algorithm is crucial, and a graph neural network is proposed to optimize the training parameters.

[0201] By combining the reliability constraints of V2V link transmission with the WMMSE algorithm optimized by graph neural networks, the constraints are simplified.

[0202] Series Theorem: Assume z1, z2...z n They are independent random variables that follow an exponential distribution, and their mean satisfies but

[0203]

[0204] Substituting the signal-to-interference-plus-noise ratio (SIR) condition and the reliability constraint of the V2V link into the series theorem, we obtain:

[0205]

[0206] Further simplification:

[0207]

[0208]

[0209] The lower bound p of the V2V link transmit power is readily available. min for:

[0210]

[0211] Delay constraint analysis:

[0212] The message transmission rate of the k-th V2V link can be expressed by Shannon's formula as follows:

[0213] C k =Blog2(1+γ) k )

[0214] Then the delay t of the k-th V2V link k The calculation formula is:

[0215]

[0216] In the formula, D k This represents the number of bits required to transmit the message for the k-th V2V link communication. V2V links must meet latency requirements for communication; excessive latency will prevent communication. Therefore, the latency t of the k-th V2V link is... k The following constraints must be met:

[0217] Pr{t k ≥T max}≤P0

[0218] In the formula, T max Pr{t} represents the maximum transmission delay threshold of a V2V link. k ≤T max} represents the probability that the transmission delay of the k-th V2V link is less than the maximum transmission delay threshold, and P0 is the maximum interruption probability threshold.

[0219] Substituting, we get:

[0220]

[0221] Simplifying the above equation, we get:

[0222]

[0223] The signal-to-interference-plus-noise ratio (SIR) threshold γ allowed for V2V communication can be obtained. k To ensure the quality of V2V communication, the probability of interruption during communication must be less than P0. If the above reliability and latency requirements are not met, the V2V link will experience a communication interruption.

[0224] Here, the interruption of a V2V link is defined as follows: First, the V2V link is interrupted when the signal-to-interference-plus-noise ratio (SINNR) is low and demodulation is impossible during communication. Second, the V2V link is interrupted when the communication delay is large. Simultaneously, it is necessary to ensure that interference from other V2V links is below a certain threshold; otherwise, excessive interference will affect the communication process.

[0225] To ensure the reliability and latency of V2V links, the signal-to-interference-plus-noise ratio (SIN / NNR) threshold γ0 of the k-th V2V link needs to be defined, and its magnitude must satisfy the SIN / NNR γ0 threshold. k and Between, it is represented as:

[0226]

[0227] p can be obtained min And since it satisfies the time delay constraint, the constraint conditions can be integrated and transformed, that is:

[0228] p min ≤p k ≤p max

[0229] In the WMMSE algorithm output parameters, the nonlinear parameter α ensures that the magnitude of the output parameter of each layer is within [p]. min p max Within the specified range, to meet the reliability constraints of the V2V link;

[0230] The WMMSE algorithm, optimized based on graph neural networks, is implemented in a distributed manner on V2V links:

[0231] Training the WMMSE algorithm optimized by graph neural networks requires channel gain information for all V2V links, necessitating centralized training.

[0232] In the specific implementation of the WMMSE algorithm optimized by graph neural networks in vehicle-to-everything (V2V) networks: both the transmitter and receiver of the V2V link have channel information for all V2V links; the receiver calculates the input variables. and intermediate variables The information is sent to the sending end; the sending end calculates the output variable. The optimal transmit power p for the V2V link is obtained based on the vehicle index information. * .

[0233] The WMMSE algorithm optimized by graph neural networks can effectively utilize vehicle computing resources and improve hardware computing efficiency in distributed computing within V2V links. However, some problems exist. For example, if one vehicle in a V2V link fails, the power allocation algorithm for that link will fail, requiring vehicles to have a certain ability to handle distributed deployment problems. Simultaneously, information transmission between distributed deployments also consumes some communication resources between V2Vs. A proposed solution is to use the maximum transmit power allocation method when communication or hardware failure occurs between V2V vehicles.

[0234] Furthermore, the power allocation algorithm for vehicle-to-everything (V2V) networks based on weighted minimum mean square error optimization using graph neural networks was verified in performance simulation experiments. The simulation parameters were set according to the highway model in 3GPP, and the parameters for this embodiment are shown in Table 1. The vehicles in the V2V system are distributed according to a spatial Poisson distribution, where vehicle density is determined by vehicle speed. The transmitter of the V2V link is randomly selected, and the vehicle closest to the transmitter is chosen as the receiver. It is assumed that the bandwidth resources obtained by multiple V2V links are equal.

[0235] Table 1 Simulation parameters for the vehicle networking environment

[0236]

[0237]

[0238] The vehicle speed and the V2V link signal-to-interference-plus-noise ratio threshold are variables that vary within a specified range. The specific values ​​of these variables under these simulation conditions will be given in the simulation description below.

[0239] The proposed graph neural network model consists of four iterative layers. Each layer contains two layers of graph neural networks (GCNs) to obtain training parameters a and b, thereby accelerating model convergence. The hidden graph layers in the GCN network model have a dimension of 4. Training takes 10,000 steps, with each step executed on the channel states of a batch of 64 V2V links, at a learning rate of 10. -3 .

[0240] The training loss of the graph neural network-improved WMMSE algorithm under the conditions of a V2V link signal-to-interference-plus-noise ratio threshold γ0 = 1 dB and a vehicle speed of 70 km / h is as follows: Figure 1 As shown. Figure 1 The graph shows the training loss, with the horizontal axis representing the number of training iterations (Epoch) and the vertical axis representing the training loss. As can be seen from the graph, the training loss initially decreases during training and eventually converges to a stable level. This indicates that the minimum mean square error algorithm for graph neural network optimization has a certain learning capability.

[0241] Figure 2The vertical axis represents the total V2V capacity with different power allocation methods; the vertical axis represents the total V2V capacity in bps / Hz; the horizontal axis represents five different power allocation methods, from left to right: the proposed graph neural network-optimized weighted minimum mean square error (GNN) algorithm; the classic weighted minimum mean square error algorithm; the weighted minimum mean square error algorithm learned using DNN (Deep Neural Networks); in the Orth algorithm, the vehicle's transmit power is the maximum transmit power; in the Rand algorithm, the vehicle's transmit power is randomized. Figure 2 Simulation conditions: vehicle speed 70km / h, V2V link signal-to-interference-plus-noise ratio threshold γ0=2dB.

[0242] Clearly, this invention performs well, achieving a higher total transmission rate for V2V links under the same conditions, and improving upon the classic weighted least mean square error algorithm by 5%.

[0243] Table 2 Comparison of execution time for different power allocation methods.

[0244]

[0245] In the table above, Method 1 is the weighted least mean square error (WMMSE) algorithm optimized by a graph neural network; Method 2 is the classic WMMSE algorithm; Method 3 uses a DNN to learn WMMSE; Method 4 sets the maximum vehicle transmission power; and Method 5 sets the transmission power of random vehicles. As the table shows, the execution time of the graph neural network-optimized weighted least mean square error algorithm is much shorter than that of the classic WMMSE algorithm. This is because the learning of training parameters by the GNN can accelerate network convergence, thus reducing execution time.

[0246] Figure 3 This represents the relationship between the total transmission rate of a V2V link and the vehicle speed when the signal-to-interference-plus-noise ratio (SIN / N) threshold γ0 = 2 dB. Figure 3A comparison of different vehicle speeds and total V2V transmission rate (Sum V2V capacity VS Vehiclevelocity) is shown, with vehicle speed on the x-axis and total V2V transmission rate (Sum V2Vcapacity) on the y-axis, in bps / Hz. As vehicle speed increases, the total V2V link transmission rate shows a decreasing trend. This decreasing trend is due to the simulation model settings; higher speeds result in a sparser vehicle distribution, increasing the average vehicle spacing. This leads to lower received power for vehicles, and the limitation on maximum transmit power for each vehicle further reduces the power allocated to vehicle users, thus decreasing the total V2V link system transmission rate. Compared to other algorithms, this invention achieves a higher total V2V link transmission rate under varying vehicle speeds.

[0247] Figure 4 This is a graph showing the variation of the total V2V link transmission rate with vehicle speed using a graph neural network to optimize the WMMSE algorithm. Figure 5 To illustrate the statistical graph of the total transmission rate of a V2V link as a function of vehicle speed using a graph neural network to optimize the MLP algorithm, Figure 6 This is a statistical graph showing the variation of the total V2V link transmission rate with vehicle speed using the WMMSE algorithm. The graph compares the total V2V transmission rate (Sum V2V capacity VS Vehicle velocity) at different vehicle speeds, with the horizontal axis representing vehicle speed and the vertical axis representing the total V2V transmission rate (Sum V2V capacity), in bps / Hz. As vehicle speed increases, the total V2V link transmission rate consistently decreases, for the same reason mentioned above. In tests across multiple datasets, comparing the total V2V transmission rate, this invention still maintains the best performance, demonstrating its stability.

[0248] Figure 7 The graph compares the number of V2V pairs with the total V2V transmission rate (Sum V2V capacity VS number of V2V pairs). The horizontal axis represents the number of V2V pairs, and the vertical axis represents the total V2V transmission rate (Sum V2V capacity). The unit is bps / Hz. Figure 8 This graph compares the number of V2V pairs with the average V2V transmission rate (Average V2V capacity vs. number of V2V pairs). The horizontal axis represents the number of V2V pairs, and the vertical axis represents the average V2V transmission rate (Average V2V capacity). The unit is bps / Hz. Figure 7It can be seen that the total transmission rate of a V2V link increases with the number of V2V pairs; this is a clear phenomenon. As the number of V2V pairs increases, the total V2V transmission rate will also increase. However, due to... Figure 8 It is known that the average transmission rate of a V2V link decreases as the number of V2V pairs increases. This is mainly because the increase in V2V pairs leads to increased interference, which affects the average transmission rate. By comparing with other algorithms, it can be seen that the algorithm proposed in this invention achieves the optimal total and average V2V transmission rates when the number of V2V pairs increases, demonstrating its practicality and robustness in large-scale V2V link environments.

[0249] Figure 9 This is a statistical graph showing the change in the total transmission rate of the V2V link as a function of the maximum interruption probability in this invention. Figure 9 The total V2V capacity with different outrage probabilities is represented by the graph, with the horizontal axis representing the maximum outrage probability and the vertical axis representing the total V2V capacity, in bps / Hz. The total V2V capacity increases with increasing outrage probability. The minimum transmission power of the V2V link transmitter will decrease, allowing the transmitter to communicate at a lower transmission rate. This reduces the limitation on transmission power and also reduces interference, thus increasing the total V2V transmission.

[0250] Figure 10 This graph shows the statistical change in the total transmission rate of the V2V link as a function of the total interference threshold in this invention. It compares the total transmission rate of the V2V link under different interference constraints (Sum V2V capacity with different interference restraints). The horizontal axis represents the interference restraint (in dBm), and the vertical axis represents the total V2V transmission rate (Sum V2V capacity) (in bps / Hz). As the maximum interference threshold increases, the interference received during V2V communication also increases, thus reducing the total transmission rate.

[0251] Figure 11This graph illustrates the change in the total transmission rate of the V2V link as a function of the amount of data transmitted, comparing the total V2V capacity with different communication transmission rates. The horizontal axis represents the communication transmission rate (KB), and the vertical axis represents the total V2V transmission rate (bps / Hz). As the amount of data transmitted increases, the latency of the V2V link will increase, and the probability of communication interruption will also increase. Therefore, the number of V2V links that do not meet the latency requirements will increase, resulting in a decrease in the total transmission rate.

[0252] The various embodiments in this invention are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

Claims

1. A method for maximizing V2V link speed based on graph neural networks, characterized in that, Includes the following steps: Establish a vehicle network communication scenario; the vehicle network communication scenario includes a vehicle network model and a channel model. Establishing a vehicle network communication scenario includes the following steps: The vehicle network communication scenario is a two-way multi-lane model: a communication base station is deployed on one side of the road, covering vehicles within that lane's range, and there are [vehicles] in the lane. A vehicle traveling at an average speed V, in Among the vehicles, there are Communication is achieved through V2V links; A channel power gain model is established, in which the formula for calculating large-scale fading is: In the formula: Represented as V2V to the transmitter to the receiving end Large-scale channel fading during communication; This is the path fading coefficient; The shadow fading coefficient follows a log-normal distribution. For V2V link transmitter to the receiving end The distance; The path fading index; A small-scale fading channel model is modeled using a first-order Gaussian Markov process in each transmission slot: In the formula: This indicates small-scale channel fading; To predict channel gain, an exponential distribution with unit mean is used; The estimation error also follows an exponential distribution with a unit mean, similar to the predicted channel gain. They are mutually independent; coefficients Used to characterize the uncorrelation between two consecutive time slot channels; The channel gain of a V2V link is expressed as: Each V2V link shares its channel information with the others, and the input to the graph neural network constructed from this is the channel gain matrix of the V2V link. This includes the channel gain of the V2V link itself. Channel gain and neighboring nodes The channel gain of the neighboring node is the interference channel gain transmitted by other V2V links; Signal-to-interference-plus-noise ratio of V2V links The calculation formula is as follows: In the formula: Indicates the first Channel power gain from vehicle transmitter to vehicle receiver in a V2V link; Indicates the first The transmit power of each V2V link; This represents the power of additive white Gaussian noise; Indicates the sending end of other V2V links to the first Interference channel power gain at the receiver in a V2V link; Interference power; According to Shannon's formula, the first Transmission rate of a V2V link for: In the formula: Base station allocated in the first The bandwidth of a V2V link; For signal-to-interference-to-noise ratio; The optimization problem is to maximize the total transmission rate of the V2V link: The first constraint C1 is the transmission reliability requirement of the V2V link. The signal-to-interference-plus-noise ratio (SIR / NDR) threshold specified for the transmitter in a V2V link. This represents the maximum probability threshold for V2V link outage; The second constraint C2 is the communication latency requirement of the V2V link. Let the communication delay of the k-th V2V link be . It is the maximum communication transmission delay; The third constraint, C3, requires that interference from other V2V links be less than the interference threshold. Interference threshold; The fourth constraint, C4, is the maximum power limit for the V2V link transmitter. Let k be the transmit power of the k-th V2V link. This represents the maximum transmit power allowed at the V2V link transmitter. coefficient The expression is as follows: In the formula, It is a Bessel function of the first kind, order 0; For the maximum Doppler frequency shift, The carrier frequency; It's the vehicle speed; The speed of wave propagation; For periodic feedback delay; Construct a convolutional graph neural network; embed the graph neural network into the WMMSE algorithm model; the convolutional graph neural network consists of a two-layer convolutional network GCN, as shown below: In the formula: Z is the output variable of the first layer of the convolutional graph neural network; For the output variables of the convolutional graph neural network iteration; The input to the graph neural network is a matrix of all ones; This is the channel power gain matrix for the V2V link; for The diagonal value, that is ; The weight parameters are the set of weight parameters for each V2V link in a two-layer graph neural network. Shared within the vehicle-to-everything (V2X) system; The activation function is ReLU; Use the sigmoid activation function to ensure that the output is between 0 and 1. The loss function for a convolutional graph neural network is as follows: In the formula: there are K V2V links in the vehicle-to-everything (V2V) system; the mean value E represents the total transmission rate and mean value of the V2V links; Indicates the first Channel power gain from vehicle transmitter to vehicle receiver in a V2V link; Indicates the first The transmit power of each V2V link; This represents the power of additive white Gaussian noise; This indicates the sending end of other V2V links to the first... Interference channel power gain at the receiver in a V2V link; Interference power; The loss function is defined as the negative of the total transmission rate and mean of the V2V link, and stochastic gradient descent is used to optimize the loss. This combines the reliability and latency constraints of V2V link transmission with the WMMSE algorithm model optimized by graph neural networks; The WMMSE algorithm model optimized by graph neural network is trained centrally and implemented in a distributed manner on V2V links.

2. The method for maximizing V2V link speed based on graph neural networks according to claim 1, characterized in that, The latency analysis of V2V links includes the following steps: Delay constraint analysis: Latency of the k-th V2V link The calculation formula is: In the formula, This represents the number of bits required to transmit the message in the k-th V2V link communication. For the first The transmission rate of each V2V link; V2V links need to meet latency requirements for communication, otherwise communication will not be possible. The latency of the k-th V2V link. The following constraints must be met: In the formula, This indicates the maximum transmission delay threshold for a V2V link. Represented as the transmission delay of the k-th V2V link Less than the maximum transmission delay threshold The probability, This is the maximum interruption probability threshold; Substituting and simplifying, we get: in, Base station allocated in the first The bandwidth of a V2V link; Simplifying the above equation, we get: The above formula yields the signal-to-interference-plus-noise ratio (SIR) threshold allowed for V2V communication. To ensure V2V communication quality, the probability of interruption during communication must be less than [a certain value]. If the above reliability and latency requirements are not met, the V2V link will experience a communication interruption. V2V link interruption is defined as: when the signal-to-interference-plus-noise ratio (SIN / INR) of the V2V link falls below a certain threshold. If a message cannot be effectively demodulated during communication, the communication is interrupted; if the communication latency of the V2V link exceeds the latency threshold... If the interference is interrupted, the interference on other V2V links will be ensured to be less than the interference threshold. Otherwise, excessive interference will affect the communication process; To ensure the reliability and latency of V2V links, a signal-to-interference-plus-noise ratio (SINNR) threshold is defined for the k-th V2V link. Size satisfies signal-to-interference-plus-noise ratio and Between, it is represented as:

3. The method for maximizing V2V link speed based on graph neural networks according to claim 2, characterized in that, The reliability analysis of V2V links includes the following steps: Series Theorem: Let Independent random variables that follow an exponential distribution have a mean that satisfies ,but In the formula: For the first i A random variable that follows an exponential distribution; In the series theorem, the first... i random variables The mean; Represented as the first i The reciprocal of the mean of a random variable It is the reciprocal of the mean of the first random variable; Substituting the signal-to-interference-plus-noise ratio (SIR) condition and the reliability requirement of the V2V link into the series theorem, we obtain: Further simplification: Lower bound of V2V link transmit power for: From the above expression, we get The constraints are integrated and transformed, that is: In the formula: This represents the maximum transmit power allowed at the V2V link transmitter. Based on nonlinear parameters The WMMSE algorithm model optimized by the graph neural network is guaranteed to have output parameters of varying magnitudes at each layer. Within the specified range, the reliability constraints of the V2V link are met.

4. The method for maximizing V2V link speed based on graph neural networks according to claim 1, characterized in that, Embedding a graph neural network into the WMMSE algorithm model includes the following steps: The WMMSE algorithm model applied to V2V links is constructed as follows: In the formula: For vehicles i Input variables; Indicates vehicle i intermediate variables; For vehicles i Output variables, For vehicles j Output variables; Let be the mean square error value of vehicle i; For vehicles in V2V links i Channel power gain; This represents the power of additive white Gaussian noise; For vehicles j The channel power gain transmitted to vehicle i; C1~C4 represent different constraints, where: , These are the rated minimum transmit power and rated maximum transmit power of the V2V link, respectively; For signal-to-interference-to-noise ratio, The signal-to-interference-plus-noise ratio (SIR / NDR) threshold specified for the transmitter in a V2V link. This represents the maximum probability threshold for V2V link outage; Let the communication delay of the k-th V2V link be . It is the maximum communication transmission delay; This is the total interference threshold for other V2V links; When input variables intermediate variables and output variables When both are optimal, we get: That is, the optimal transmit power of the i-th V2V is the output variable value. The minimum mean square error problem is solved using the block coordinate descent method. It is proven that when two of the three variables (input parameter, output parameter, and intermediate parameter) are fixed, the objective is convex in each variable. This transforms the maximum rate problem in the V2V link into the minimum mean square error problem. This represents the optimal transmit power for the i-th V2V link.

5. The method for maximizing V2V link speed based on graph neural networks according to claim 4, characterized in that, The WMMSE algorithm model uses the square root of the maximum transmit power of the V2V link. For input, after In the k-th iteration, the optimal V2V link transmit power is obtained. The specific process of the layer is as follows: (1) Input variables: In the formula: Represented as the k-th V2V link in the k-th V2V link Input variables for the next iteration; It is represented as the channel gain coefficient of the k-th V2V link; Represented as the k-th V2V link The output variables of the next iteration; This represents the power of additive white Gaussian noise; Indicating other V2V links, then Indicates other V2V links The output variables of the next iteration; (2) Intermediate variables: In the formula: Represented as the k-th V2V link in the k-th V2V link Intermediate variables in the next iteration; and parameters These are trainable parameters; For input variables; It is represented as the channel gain coefficient of the k-th V2V link; This is represented as the output variable of the previous iteration in the k-th V2V link; (3) Output variables: In the formula: Represented as the k-th V2V link in the k-th V2V link The output variables of the next iteration; It is a nonlinear parameter; This is expressed as the channel gain coefficient of the k-th V2V link. The interference channel gain of other V2V links on the k-th V2V link; Represented as the k-th V2V link in the k-th V2V link Input variables for the next iteration This indicates that it is the first in other V2V links. Input variables for the next iteration; For the k-th V2V link, the first... Intermediate variables in the next iteration. For other V2V links Intermediate variables in the next iteration; In the three expressions above, the input variables depend entirely on the channel state of the V2V link receiver, while the parameters in the intermediate variables... and parameters Trainable parameters; nonlinear parameters in the output variables Ensure that the output parameters of each layer are within Within the specified range, the constraints are guaranteed to be satisfied.

6. A method for maximizing V2V link speed based on a graph neural network according to any one of claims 1 to 5, characterized in that, The WMMSE algorithm model, implemented in a distributed manner on V2V links, includes: both the transmitter and receiver of the V2V link possess channel information for all V2V links; the receiver calculates the input variables. and intermediate variables The information is sent to the sending end; the sending end calculates the output variable. The optimal transmit power for the V2V link is obtained based on the vehicle index information. This ensures that the total transmission rate of the V2V link is maximized while meeting the reliability and latency requirements of the V2V link.

Images