A wireless ad hoc network power control method, medium and electronic device
By constructing a deep expansion network architecture and residual graph convolutional neural network based on WMMSE algorithm, the problems of high complexity and poor interpretability of power control algorithms in wireless ad hoc network are solved, and fast convergence and high-performance wireless ad hoc network are achieved. Network power control.
Patent Information
- Application Number
- CN202411116725.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-15
AI Technical Summary
The existing wireless ad hoc network power control algorithm has high complexity and long iteration time in large-scale and highly dynamic network scenarios. The deep learning-based methods have poor interpretability and large training sample requirements, making it difficult to perform and interpretability. strike a balance.
Build a deep expansion network architecture based on WMMSE algorithm, adopt deep expansion technology and residual graph convolution neural network, update weight information and transmit power through multi-layer perceptron, train deep expansion networks using unsupervised methods, optimize transmit power strategies To realize wireless ad hoc network power control.
Algorithm convergence with fewer iterations is achieved, algorithm convergence is accelerated, algorithm convergence is maintained, good interpretability and performance is enhanced, algorithm generalization ability and scalability are enhanced, and large-scale and highly dynamic wireless ad hoc networking scenarios are adapted.
Smart Images

Figure CN118647075B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communication technology, and in particular to a wireless ad hoc network power control method, medium and electronic equipment. Background Art
[0002] With the development of new generation communication technology, people's demand for communication networks has increased, while wireless network resources are limited. The contradiction between limited resources and high resource demand in wireless networks is particularly prominent. At the same time, wireless networks will develop in the direction of large-scale, dense and highly dynamic in the future, which further increases the difficulty of resource allocation. Among them, the density and high dynamics of wireless ad hoc network models will cause serious multi-user interference, which is one of the main factors limiting the system performance of wireless ad hoc network models. Reasonable power control in wireless ad hoc network models is crucial to controlling multi-user interference and improving communication system performance. Therefore, it is an important issue to efficiently allocate and manage scarce power resources to reduce interference and improve wireless network system performance.
[0003] At present, there are two main types of power control algorithms: traditional power control algorithms and power control algorithms based on deep learning. Traditional power control algorithms are generally based on convex optimization theory, such as the weighted minimum mean square error (WMMSE) algorithm and the successive convex approximation (SCA) algorithm. They have detailed mathematical models and can achieve good performance in small-scale and static network scenarios. However, such algorithms are usually highly complex and have long iteration times, making them difficult to adapt to large-scale and highly dynamic network scenarios. Due to its efficient real-time computing capabilities, deep learning has been applied in different fields of wireless communication systems. Therefore, many studies have used deep learning for power control. Compared with traditional algorithms, this method improves the convergence speed while reducing system latency and complexity. Existing deep learning-based methods usually have problems such as poor interpretability and large training sample requirements, making it difficult to strike a balance between performance and interpretability.
[0004] In summary, there is an urgent need for a wireless ad hoc network power control method, medium and electronic device to solve the problems in the prior art. Summary of the invention
[0005] The present invention aims to provide a wireless ad hoc network power control method, medium and electronic equipment, and the specific technical solution is as follows:
[0006] A wireless ad hoc network power control method comprises the following steps:
[0007] Construct a wireless ad hoc network model consisting of a transmitter and a receiver, calculate the received signal and signal-to-noise ratio received by each receiver in the wireless ad hoc network model, and the communication rate of each transmitter in the wireless ad hoc network model;
[0008] Constructing an optimization objective function for optimizing the transmission power of each transmitter;
[0009] Constructing a deep unfolding network architecture for solving the optimization objective function, the deep unfolding network architecture comprising a plurality of unfolding layers, each unfolding layer comprising a utility information updating module, a weight information updating module and a transmission power updating module;
[0010] The deep unfolded network architecture is trained in an unsupervised manner, and the loss function is a negative value of the optimization target mean;
[0011] The trained deep unfolded network architecture is called to calculate the transmission power strategy of each transmitter to realize power control of wireless ad hoc networks.
[0012] Preferably, the received signal received by each receiver in the wireless ad hoc network model , expressed as:
[0013] ;
[0014] in, Indicates The transmission signal of a transmitter, For the The communication channel between the transmitter-receiver pairs, For transmitter and receiver The interference channel between is the channel Gaussian white noise; represents the number of transmitter-receiver pairs in the wireless ad hoc network model;
[0015] Receiver signal-to-noise ratio According to the received signal of the receiver, it is calculated:
[0016] ;
[0017] in, For the The transmission power of each transmitter, For the The transmission power of each transmitter, is the variance of the Gaussian distribution;
[0018] Communication rate of each transmitter According to the signal-to-noise ratio of the receiver, it is calculated as:
[0019] .
[0020] Preferably, the optimization objective of the optimization objective function is to maximize the weighted sum rate, the constraint condition is the maximum transmit power constraint, and the expression of the optimization objective function is as follows:
[0021] ;
[0022] in, is the maximum transmit power, is the weighting coefficient, indicating that the transmitter priority.
[0023] Preferably, a deep expansion network architecture based on the WMMSE algorithm is constructed to solve the optimization objective function, and the deep expansion network architecture is based on the channel information of the wireless ad hoc network model. And the initial transmission power of each transmitter in the wireless ad hoc network model As input, the goal is to maximize the weighted sum communication rate. The optimal power allocation for each transmitter is finally obtained through iterative optimization. Transmitter No. Utility information in the layer Update module and weight information Update module and transmit power The calculation formula for the update module is as follows:
[0024] ;
[0025] Among them, the utility information , weight information and transmit power The upper right corner of represents the first layer, represents the first layer; represents the residual graph convolutional neural network; Indicates channel information; Represents the weight information node feature; Indicates the information used to update the weight The training parameters of Indicates the transmit power node characteristics; Indicates the information used to update the weight The training parameters of .
[0026] Preferably, the deep expansion network architecture includes multiple expansion layers, and the expansion layers are composed of a utility information update module, a weight information update module and a transmission power update module. The utility information update module adopts the expression of utility information in the weighted minimum mean square error algorithm, and the weight information and transmission power update modules adopt residual graph convolutional neural network for learning.
[0027] Preferably, the weight information is learned using a residual graph convolutional neural network, specifically including:
[0028] The weight information node features include the transmission power of the previous layer and the utility information calculated by the layer, and the weight information node features and the channel information are input into the residual graph convolutional neural network to update the weight information;
[0029] The transmit power is learned using a residual graph convolutional neural network, specifically including:
[0030] The transmit power node feature includes the transmit power of the previous layer, the weight information calculated by the layer, and the weighting coefficient. The transmit power node feature and the channel information are input into the residual graph convolutional neural network to update the transmit power.
[0031] Preferably, the residual graph convolutional neural network The expression of the layer is as follows:
[0032] ;
[0033] in, Represents the residual graph convolutional neural network The output of the layer; Indicates Output of layer residual graph convolutional neural network; Represents the channel information of the wireless ad hoc network model, for The diagonal matrix of represents the interference channel information of the wireless ad hoc network model; represents the network training parameters for training channel information; represents a network training parameter for training interference channel information; ReLU activation function.
[0034] In the last layer of the residual graph convolutional neural network, the output feature information of each layer of the residual graph convolutional neural network is aggregated and input into the multi-layer perceptron to obtain the output information. The expression of the last layer in the residual graph convolutional neural network is as follows:
[0035] ;
[0036] in, Represents node features; Represents the output of the first layer of the residual graph convolutional neural network; Represents the output of the second layer of the residual graph convolutional neural network; Represents the residual graph convolutional neural network The output of the layer; represents the multilayer perceptron used; Indicates the concatenation operation on the feature matrix.
[0037] Preferably, the loss function is expressed as follows:
[0038] ;
[0039] in, represents the loss function; represents the mean function.
[0040] In addition, the present invention also includes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the wireless ad hoc network power control method as described above are implemented.
[0041] In addition, the present invention also includes an electronic device, including a memory and a processor;
[0042] The memory is used to store a computer program executable on the processor;
[0043] The processor is used to implement the steps of the above-mentioned wireless ad hoc network power control method when executing the computer program.
[0044] The application of the technical solution of the present invention has the following beneficial effects:
[0045] The present invention proposes a wireless ad hoc network power control method, medium and electronic device. The method of the present invention utilizes deep expansion technology to perform deep expansion on the traditional WMMSE algorithm, and constructs a deep expansion architecture based on the WMMSE algorithm. Compared with the traditional algorithm, it achieves better performance with fewer iterations and accelerates the convergence of the algorithm. The deep expansion architecture follows the iterative structure of the WMMSE algorithm to make full use of the model domain knowledge and provide good interpretability. In the deep expansion architecture, the present invention retains the original utility information update structure in the WMMSE algorithm, and designs a residual graph convolutional neural network to design the weight information and transmission power in the algorithm architecture, so as to maintain excellent performance and computational efficiency while making full use of domain knowledge.
[0046] In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0048] Figure 1 is a flowchart of the steps of the wireless ad hoc network power control method in a preferred embodiment of the present invention;
[0049] Figure 2 It is a schematic diagram of the wireless ad hoc network model scenario;
[0050] Figure 3 It is a deep expanded network architecture diagram of the present invention. DETAILED DESCRIPTION
[0051] The wireless ad hoc network model is large-scale and equipment-intensive, which leads to serious interference between devices and reduced communication system performance. At the same time, the traditional power control algorithm is highly complex and takes a long time to iterate, and cannot adapt to equipment-intensive and highly dynamic scenarios. The power control algorithm based on deep learning lacks interpretability, requires a large number of training samples, and has poor generalization and scalability. Therefore, the present invention discloses a wireless ad hoc network power control method based on deep expansion and graph neural network.
[0052] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0053] Glossary:
[0054] Transmitter-receiver pair: A single-antenna communication connection is established between the transmitter and the receiver in the wireless ad hoc network model. A single-antenna communication connection consisting of a transmitter and a receiver is called a transmitter-receiver pair.
[0055] Residual Graph Convolutional Neural Network (RGCN): A graph convolutional neural network with a residual structure.
[0056] Utility Information : Utility information quantifies the impact of interference channels and noise on the receiver in the wireless ad hoc network model, and is used to represent the service quality or performance of the receiver in the wireless ad hoc network model.
[0057] Weight information : The weight information is associated with the utility information and the transmission power. During the algorithm calculation process, adjustments are made according to the values of the utility information and the transmission power, so that the algorithm can converge faster.
[0058] Transmit power : The power transmitted by each transmitter in the wireless ad hoc network model.
[0059] The present invention designs a wireless ad hoc network power control algorithm based on deep expansion and graph neural network, and uses deep expansion and graph neural network to construct a deep expansion network architecture based on the weighted minimum mean square error (WMMSE) algorithm. Specifically, the present invention uses the iterative structure of the WMMSE algorithm and respectively performs the WMMSE algorithm utility information , weight information and transmit power The update module is designed as follows: First, the utility information in the WMMSE algorithm is used in the deep expansion network architecture. The updated structure of is embedded into the algorithm architecture as model domain knowledge to provide good interpretability. In addition, the utility information As weight information and transmit power The input retains its update structure to ensure the weight information and transmit power Secondly, a residual graph convolutional neural network (RGCN) is designed to update the weight information in the WMMSE algorithm. and transmit power Learn and replace weight information and transmit power The update structure is used to accelerate the convergence of the algorithm and provide better generalization and scalability. By developing the architecture according to the designed algorithm, the channel information between the transmitter and the receiver, including the transmission channel and interference channel information, and the initial transmission power of the transmitter are used as inputs. The optimization goal is to maximize the weighted sum rate. The unsupervised method is used for centralized training and distributed deployment. The optimal power strategy of each transmitter is calculated on each transmitter according to the trained algorithm architecture, and finally the signal is transmitted according to the optimal power strategy.
[0060] like Figure 1 As shown, this embodiment discloses a wireless ad hoc network power control method, comprising the following steps:
[0061] S1: Construct a wireless ad hoc network model consisting of a transmitter and a receiver, calculate the received signal and signal-to-noise ratio received by each receiver in the wireless ad hoc network model, and the communication rate of each transmitter in the wireless ad hoc network model;
[0062] S2: construct an optimization objective function to optimize the transmission power of each transmitter;
[0063] S3: constructing a deep expansion network architecture for solving the optimization objective function, wherein the deep expansion network architecture includes a plurality of expansion layers, each expansion layer includes a utility information update module, a weight information update module and a transmission power update module;
[0064] S4: training the deep unfolded network architecture in an unsupervised manner, wherein the loss function uses the negative value of the optimization target mean;
[0065] S5: Call the trained deep unfolded network architecture to calculate the transmission power strategy of each transmitter to achieve wireless ad hoc network power control.
[0066] In the embodiment of the present invention, the constructed For large-scale, densely populated wireless ad hoc network models with single-antenna transmitter-receiver configurations, such as Figure 2 In the wireless ad hoc network model, each transmitter communicates with its corresponding receiver and is also interfered by other transmitters. When there are more devices in the wireless ad hoc network model, the interference is more serious.
[0067] In the embodiment of the present invention, Figure 3 As shown, the deep expansion network architecture includes multiple expansion layers, and each of the expansion layers is composed of a utility information update module, a weight information update module and a transmission power update module. In an embodiment of the present invention, the utility information is updated using the expression of the utility information in the original WMMSE algorithm, and the weight information and the transmission power are learned through a residual graph convolutional neural network. The residual graph convolutional neural network (RGCN) includes a backbone part and an output part. The backbone part of the network is used for feature extraction, and a graph convolutional neural network GCN is used and residual connections are introduced. The output part is used to predict the output and is composed of a multi-layer perceptron (MLP). In the last layer of the RGCN, the output feature information of each layer of the RGCN backbone is aggregated and output through the multi-layer perceptron MLP. The deep expansion network architecture is based on the channel information of the wireless ad hoc network model. And the initial transmission power of each transmitter in the wireless ad hoc network model As input, the goal is to maximize the weighted sum communication rate. Layer iterative optimization, ultimately obtaining the optimal power allocation for each transmitter In the In the first iteration, Transmit power of the layer And the current channel information As the The input of the layer is Output of the layer .
[0068] In the embodiment of the present invention, the received signal received by each receiver in the wireless ad hoc network model is , expressed as:
[0069] ;
[0070] in, Indicates The transmission signal of a transmitter, For the The communication channel between the transmitter-receiver pairs, For transmitter and receiver The interference channel between is the channel Gaussian white noise, which obeys Gaussian distribution; is the variance of the Gaussian distribution, and Also used to represent noise, represents the number of transmitter-receiver pairs in the wireless ad hoc network model;
[0071] Receiver signal-to-noise ratio According to the received signal of the receiver, it is calculated:
[0072] ;
[0073] in, For the The transmission power of each transmitter, For the The transmission power of each transmitter;
[0074] Communication rate of each transmitter According to the signal-to-noise ratio of the receiver, it is calculated as:
[0075] ;
[0076] in, For the The transmission power of each transmitter, For the The transmit power of a transmitter.
[0077] Preferably, the optimization objective of the optimization objective function is to maximize the weighted sum rate, the constraint condition is the maximum transmit power constraint, and the expression of the optimization objective function is as follows:
[0078] ;
[0079] in, is the maximum transmit power, is the weighting coefficient, indicating that the transmitter priority.
[0080] Preferably, a deep expansion network architecture based on the WMMSE algorithm is constructed to solve the optimization objective function, and the deep expansion network architecture is based on the channel information of the wireless ad hoc network model. And the initial transmission power of each transmitter in the wireless ad hoc network model As input, the goal is to maximize the weighted sum communication rate. Layer iterative optimization, ultimately obtaining the optimal power allocation for each transmitter , No. Transmitter No. Utility information in the layer Update module and weight information Update module and transmit power The calculation formula for the update module is as follows:
[0081] (a);
[0082] (b);
[0083] (c);
[0084] Among them, the utility information , weight information and transmit power The upper right corner of represents the first layer, represents the first Layer, lower right corner Indicates transmitters; represents the residual graph convolutional neural network; Indicates channel information; Represents the weight information node feature; Indicates the information used to update the weight The training parameters of Indicates the transmit power node characteristics; Indicates the information used to update the weight The training parameters of .
[0085] Preferably, the deep expansion network architecture includes multiple expansion layers, and the expansion layers are composed of a utility information update module, a weight information update module and a transmission power update module. The utility information update module adopts the expression of utility information in the weighted minimum mean square error algorithm, and the weight information update module and the transmission power update module adopt residual graph convolutional neural network for learning.
[0086] Preferably, the weight information updating module adopts residual graph convolutional neural network for learning. Taking a transmitter as an example, the specific calculation process of weight information is explained, which specifically includes:
[0087] The weight information node features include the transmission power of the previous layer and the utility information calculated by this layer. The weight information node features and channel information are input into the residual graph convolutional neural network to update the weight information. Specifically, the residual graph convolutional network RGCN is used to replace the weight information in the weighted minimum mean square error (WMMSE) algorithm. The calculation steps of weight information node feature Including The transmission power of the transmitter on the upper layer And the utility information calculated by this layer , and combine it with the channel information Input into RGCN for update to get updated weight information , the expression is shown in (b).
[0088] The transmit power update module adopts the residual graph convolutional neural network for learning. Taking a transmitter as an example, the specific calculation process of the transmission power is explained, which includes:
[0089] The transmit power node features include the transmit power of the previous layer, the weight information calculated by this layer, and the weighting coefficient. The transmit power node features and channel information are input into the residual graph convolutional neural network to update the transmit power. Specifically, the transmit power is updated using RGCN. To update, Transmit power node characteristics include the The transmission power of the transmitter on the upper layer , weight information of the previous layer and weighting coefficients , and combine it with the channel characteristic information Input into RGCN for update to get the updated transmit power , the expression is shown in (c).
[0090] In the embodiment of the present invention, the weight information is enhanced by RGCN and transmit power The update structure of is used to accelerate the convergence of weight information and transmit power. and transmit power The fast convergence of will also accelerate the convergence of the unfolded architecture, thereby achieving better performance with fewer iterations. The weight information and transmit power will be used to input the calculated feature information as node features into RGCN for training. This can increase the correlation between node features through multiple iterations of feature extraction and improve the performance of RGCN training.
[0091] Preferably, the residual graph convolutional neural network The layer expression is as follows:
[0092] ;
[0093] in, Represents the residual graph convolutional neural network The output of the layer; Represents the residual graph convolutional neural network The output of the layer; Represents the channel information of the wireless ad hoc network model, for The diagonal matrix of is represented as the communication channel between the transmitter and its corresponding receiver, which is also trained through a parameter Conduct training to highlight the importance of transmission channels over interference channels; represents the network training parameters for training channel information; represents the network training parameters for training the interference channel information, focusing on training the channel information and the interference channel information respectively; Represents the ReLU activation function, which is used to improve the nonlinear expression ability of the network.
[0094] In the last layer of RGCN, the output feature information of each layer of RGCN is aggregated and input into the multi-layer perceptron (MLP) to obtain the output information. This realizes the feature fusion within different neighborhoods of each node, enhances the expressiveness of the node, and can extract more effective information from the aggregated features. The formula of the last layer of RGCN is as follows:
[0095] ;
[0096] in, Represents node features; Represents the channel information of the wireless ad hoc network model; Represents the output of the first layer of the residual graph convolutional neural network; Represents the output of the second layer of the residual graph convolutional neural network; Represents the residual graph convolutional neural network The output of the layer; represents the multilayer perceptron used; Indicates the concatenation operation on the feature matrix.
[0097] Preferably, the loss function is expressed as follows:
[0098] ;
[0099] in, represents the loss function; represents the mean function.
[0100] The method of the present invention performs centralized training based on the constructed deep expansion network architecture with the above loss function as the target, and adopts distributed deployment in the execution stage. The optimal transmission power strategy of each transmitter is obtained by calculation based on the trained network architecture. Finally, the transmitter transmits the signal according to the optimal transmission power to maximize the communication rate of the entire wireless ad hoc network model.
[0101] In addition, this embodiment also discloses a computer device, including a memory and a processor;
[0102] The memory is used to store a computer program executable on the processor;
[0103] The processor is used to implement the steps of the above-mentioned wireless ad hoc network power control method when executing the computer program.
[0104] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the computer device.
[0105] The computer device may be a computing device such as a mobile phone, a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may include, but is not limited to, a processor and a memory. For example, the computer device may also include an input / output device, a network access device, a bus, etc.
[0106] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer device, and uses various interfaces and lines to connect various parts of the entire computer device.
[0107] The memory can be used to store the computer program and / or module, and the processor implements the computer program by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.
[0108] Wherein, if the module / unit integrated in the computer device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.
[0109] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned wireless ad hoc network power control method are implemented.
[0110] Compared with the prior art, the present invention discloses a wireless ad hoc network power control method, medium and electronic device. The method of the present invention utilizes deep expansion technology to perform deep expansion on the traditional WMMSE algorithm, and constructs a deep expansion network architecture based on the WMMSE algorithm. Compared with the traditional algorithm, it uses fewer iterations to achieve better performance and accelerate the convergence of the algorithm. The deep expansion network architecture follows the iterative structure of the WMMSE algorithm to make full use of the model domain knowledge and provide good interpretability. In the deep expansion network architecture, the method of the present invention retains the update structure of the utility information in the WMMSE algorithm, which quantifies the impact of channel interference and noise on the receiver, and adds it as domain knowledge to the expansion architecture, which can provide good interpretability. At the same time, the utility information is used as the input of weight information and transmission power, and its update structure is retained, which can ensure the update of weight information and transmission power, thereby ensuring the performance of the algorithm. At the same time, the method of the present invention designs a residual graph convolutional neural network to design the weight information and transmission power in the deep expansion network architecture, so as to enhance the update structure of weight information and transmission power and accelerate the convergence of the algorithm. And the deep expansion network architecture extracts node features multiple times through multiple iterations to increase the correlation between node features. In addition, the residual graph convolutional neural network introduces residual connections, which can aggregate node features of different layers to obtain more effective node representation, thereby enhancing the generalization and scalability of the algorithm architecture. The proposed deep unfolding network architecture is subject to distributed deployment, reduces algorithm calculation delay and communication overhead, and can meet the low latency requirements of the communication system; at the same time, it has good scalability for large-scale wireless ad hoc network models.
[0111] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A wireless ad hoc network power control method, It is characterized in that The steps include: Construct a wireless ad hoc network model consisting of a transmitter and a receiver, calculate the received signal and signal-to-noise ratio received by each receiver in the wireless ad hoc network model, and the communication rate of each transmitter in the wireless ad hoc network model; Constructing an optimization objective function for optimizing the transmission power of each transmitter; Constructing a deep unfolding network architecture for solving the optimization objective function, the deep unfolding network architecture comprising a plurality of unfolding layers, each unfolding layer comprising a utility information updating module, a weight information updating module and a transmission power updating module; The deep unfolded network architecture is trained in an unsupervised manner, and the loss function is a negative value of the optimization target mean; The trained deep unfolding network architecture is used to calculate the transmission power strategy of each transmitter to achieve power control of wireless ad hoc networks. The deep expansion network architecture is based on the channel information H of the wireless ad hoc network model and the initial transmission power of each transmitter in the wireless ad hoc network model. As input, the optimal power allocation for each transmitter is finally obtained through iterative optimization of k = {1, ..., K} layers with the goal of maximizing the weighted sum communication rate. The calculation formulas of the utility information u update module, weight information w update module and transmission power v update module in the k-th layer of the i-th transmitter are as follows: Among them, the upper right corner subscript (k) of the utility information u, weight information w, and transmit power v represents the kth layer of the deep unfolding network architecture, and (k-1) represents the k-1th layer of the deep unfolding network architecture; ψ represents the residual graph convolutional neural network; H represents the channel information; Represents the weight information node feature; θ ω Represents the training parameters used to update the weight information w; represents the transmission power node characteristics; θ v represents the training parameters used to update the weight information v; h ii is the communication channel between the i-th transmitter-receiver pair, h ij is the interference channel between transmitter j and receiver i; σ 2 represents the variance of the Gaussian distribution; The deep unfolding network architecture includes multiple unfolding layers, and the unfolding layer is composed of a utility information update module, a weight information update module and a transmission power update module. The utility information update module adopts the expression of utility information in the weighted minimum mean square error algorithm, and the weight information and transmission power update modules adopt residual graph convolutional neural network for learning; The weight information is learned using a residual graph convolutional neural network, specifically including: The weight information node features include the transmission power of the previous layer and the utility information calculated by the current layer. The weight information node features and the channel information are input into the residual graph convolutional neural network to update the weight information. The transmit power is learned using a residual graph convolutional neural network, specifically including: The transmit power node features include the transmit power of the previous layer, the weight information calculated by the current layer, and the weighting coefficient. The transmit power node features and channel information are input into the residual graph convolutional neural network to update the transmit power. The expression of the lth layer of the residual graph convolutional neural network is as follows: G (l) =α(diag(H)G (l-1) i (l) +HG (l-1) i ′(l) )+G (l-1) ; Among them, G (l) Represents the output of the lth layer of the residual graph convolutional neural network; G (l-1) represents the output of the l-1th layer residual graph convolutional neural network; H represents the channel information of the wireless ad hoc network model, diag(H) is the diagonal matrix of H, which represents the interference channel information of the wireless ad hoc network model; θ represents the network training parameters for training the channel information, θ (l) represents the network training parameters for training channel information in the l-th layer residual graph convolutional neural network; θ′ represents the network training parameters for training interference channel information, θ′ (l) represents the network training parameters for training interference channel information in the l-th layer residual graph convolutional neural network; α represents the ReLU activation function; In the last layer of the residual graph convolutional neural network, the output feature information of each layer of the residual graph convolutional neural network is aggregated and input into the multi-layer perceptron to obtain the output information. The expression of the last layer in the residual graph convolutional neural network is as follows: ψ(H,Q;θ)=MLP(G (1) ||G (2) ||...||G (L) ); Among them, Q represents the node feature; G (1) Represents the output of the first layer of the residual graph convolutional neural network; G (2) Represents the output of the second layer of the residual graph convolutional neural network; G (L) Represents the output of the Lth layer of the residual graph convolutional neural network; MLP represents the multi-layer perceptron used; || represents the concatenation operation on the feature matrix; The expression of the loss function is as follows: Among them, L(θ) represents the loss function; E represents the mean function.
2. The wireless ad hoc network power control method according to claim 1, It is characterized in that The received signal y received by each receiver in the wireless ad hoc network model i , expressed as: Among them, x i represents the transmission signal of the i-th transmitter, n i is the channel Gaussian white noise; M represents the logarithm of transmitter-receiver pairs in the wireless ad hoc network model; Receiver signal-to-noise ratio SINR i According to the received signal of the receiver, it is calculated: Among them, v i is the transmission power of the ith transmitter, v j is the transmission power of the jth transmitter; The communication rate of each transmitter is c i According to the signal-to-noise ratio of the receiver, it is calculated as: Among them, v i is the transmission power of the ith transmitter, v j is the transmission power of the jth transmitter.
3. The wireless ad hoc network power control method according to claim 2, It is characterized in that The optimization objective of the optimization objective function is to maximize the weighted sum rate, and the constraint condition is the maximum transmit power constraint. The expression of the optimization objective function is as follows: Among them, v max is the maximum transmission power, λ i is the weighting coefficient, indicating the priority of transmitter i.
4. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the wireless ad hoc network power control method according to any one of claims 1 to 3 are implemented.
5. An electronic device, It is characterized in that including memory and processor; The memory is used to store a computer program executable on the processor; The processor is used to implement the steps of the wireless ad hoc network power control method according to any one of claims 1 to 3 when executing the computer program.
