Spectrum sharing method, device and computing equipment based on graph convolutional neural network

Abstract

The present invention discloses a spectrum sharing method, device and computing equipment based on graph convolutional neural network, wherein the method comprises: step 1, obtaining current network status information, establishing an undirected graph model according to the network status information, and generating an environment matrix; step 2, processing user demand information and the environment matrix using a decision model to obtain a channel selection vector and an energy adaptation vector; step 3, judging whether a preset convergence condition is met through an objective function according to the channel selection vector, the energy adaptation vector and the environment matrix; if so, executing step 4; if not, executing step 5; step 4, allocating spectrum according to the current channel selection vector and the energy adaptation vector, and the method ends; step 5, updating the parameters of the decision model, and then jumping to execute step 2. The scheme solves the problem of imperfect graph structure when establishing a decision model in the prior art, and can conveniently obtain a better spectrum access scheme.

Description

Technical Field

[0001] The present invention relates to the field of spectrum sharing technology, and in particular to a spectrum sharing method, device and computing equipment based on graph convolutional neural network. Background Art

[0002] Spectrum resources are an important carrier of wireless services and play an increasingly important role. Traditional spectrum management is that the spectrum management agency issues use licenses to authorized users and enters the corresponding frequency usage information into the spectrum management database to ensure that the frequencies used by users do not conflict with each other. With the rapid development of 5G and the Internet of Things, a large number of frequency-using devices are constantly connected, and the need for sharing and dynamic allocation of wireless spectrum has become particularly urgent. The existing technology mainly solves these problems through the following methods: a cognitive radio network spectrum resource allocation method based on Q learning, a spectrum resource allocation method based on deep learning, and a spectrum strategy construction method based on graph convolutional neural network. However, these methods have defects such as imperfect graph structure when establishing a decision model and slow convergence of the strategy model based on the convolutional neural network (CNN) model. Summary of the invention

[0003] In view of the above problems, the present invention is proposed to provide a spectrum sharing method, apparatus and computing device based on graph convolutional neural network that overcome the above problems or at least partially solve the above problems.

[0004] According to one aspect of the present invention, a spectrum sharing method based on a graph convolutional neural network is provided, the method comprising:

[0005] Step 1, obtaining current network status information, establishing an undirected graph model according to the network status information, and generating an environment matrix using the undirected graph model; wherein the nodes in the undirected graph model include: base station nodes and device nodes;

[0006] Step 2: Use the decision model to process the user demand information and the environment matrix to obtain a channel selection vector and an energy adaptation vector;

[0007] Step 3: According to the channel selection vector, the energy adaptation vector and the environment matrix, determine whether the preset convergence condition is met through the objective function; if so, execute step 4; if not, execute step 5;

[0008] Step 4, spectrum allocation is performed according to the current channel selection vector and energy adaptation vector, and the method ends;

[0009] Step 5: Update the parameters of the decision model, and then jump to step 2.

[0010] Furthermore, the edges between the base station nodes and the device nodes represent signal links, and the edges between the device nodes represent interference links;

[0011] The base station nodes include: the main network base station nodes and the base station nodes within the coverage of the indoor equipment; the equipment nodes include: the main user equipment nodes and the secondary user equipment nodes.

[0012] Furthermore, the node information of the base station node includes: location information, movement direction, movement speed, and current channel occupancy vector;

[0013] The node information of the device node includes: location information, movement direction, movement speed and current channel usage vector.

[0014] Furthermore, generating the environment matrix using the undirected graph model further includes:

[0015] Analyze the undirected graph model to obtain the user distance matrix and user resource occupancy matrix;

[0016] Generate an environment matrix based on the user distance matrix and the user resource occupancy matrix.

[0017] Furthermore, the decision model is established using a graph convolutional neural network; the decision model includes: multiple graph convolutional layers, a single hidden layer feedforward neural network and a softmax layer.

[0018] Furthermore, updating the parameters of the decision model further includes:

[0019] The gradient descent algorithm is used to update the parameters of multiple graph convolutional layers in the decision model, and the approximate solution algorithm is used to update the parameters of the single hidden layer feedforward neural network in the decision model.

[0020] Furthermore, the preset convergence condition includes: the sum of data rates of all secondary user equipment nodes is a maximum value.

[0021] According to another aspect of the present invention, a spectrum sharing device based on a graph convolutional neural network is provided, the device comprising:

[0022] The environment processing module is adapted to obtain current network status information, establish an undirected graph model according to the network status information, and generate an environment matrix using the undirected graph model; wherein the nodes in the undirected graph model include: base station nodes and device nodes;

[0023] A decision processing module, adapted to process user demand information and an environment matrix using a decision model to obtain a channel selection vector and an energy adaptation vector;

[0024] A judgment module, adapted to judge whether a preset convergence condition is met through an objective function according to the channel selection vector, the energy adaptation vector and the environment matrix;

[0025] A spectrum allocation module, adapted to perform spectrum allocation according to the current channel selection vector and energy adaptation vector if the judgment module judges that the preset convergence condition is met;

[0026] The parameter updating module is adapted to update the parameters of the decision model and then trigger the decision processing module if the judgment module determines that the preset convergence condition is not met.

[0027] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0028] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned spectrum sharing method based on graph convolutional neural network.

[0029] According to yet another aspect of the present invention, a computer storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned spectrum sharing method based on graph convolutional neural network.

[0030] According to the technical solution provided by the present invention, a graph learning method is introduced into the decision model constructed by using GCN, which has stronger adaptability compared with the prior art; an undirected graph model is dynamically established based on network status information, and the decision model is used for processing, which effectively solves the problem of imperfect graph structure when establishing the decision model in the prior art; the undirected graph model is introduced into the reinforcement learning process based on the graph convolutional neural network, and the channel selection vector and the energy adaptation vector are output. According to the channel selection vector, the energy adaptation vector and the environment matrix, the objective function is used to make a judgment, so that a better spectrum access scheme can be obtained quickly and conveniently, the spectrum sharing function of the secondary user equipment is completed, and the spectrum utilization efficiency is improved; and the location information and other feature information are considered separately in the decision model, which ensures that the characteristics of the location information will not cause information loss in the feature extraction process due to the influence of other information, thereby improving the efficiency of spectrum sharing; in addition, in actual use, the characteristics of the input data can be conveniently added to obtain the optimal graph structure, thereby ensuring the scalability of the present solution.

[0031] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0033] Figure 1 A schematic diagram of the process of a spectrum sharing method based on a graph convolutional neural network according to an embodiment of the present invention is shown;

[0034] Figure 2 A schematic diagram of a network architecture is shown;

[0035] Figure 3 A schematic diagram of a decision model is shown;

[0036] Figure 4 A schematic diagram of the architecture of a spectrum sharing method based on a graph convolutional neural network is shown;

[0037] Figure 5 A structural block diagram of a spectrum sharing device based on a graph convolutional neural network according to an embodiment of the present invention is shown;

[0038] Figure 6 A schematic structural diagram of a computing device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0039] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0040] Figure 1 FIG. 4 is a flow chart of a spectrum sharing method based on a graph convolutional neural network according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0041] Step S1, obtaining current network status information, establishing an undirected graph model according to the network status information, and generating an environment matrix using the undirected graph model.

[0042] In the cognitive radio scenario, the current network status information can be obtained, and based on the network status information, each base station, each device, the signal link between the base station and the device, and the interference link between the devices included in the current network architecture can be identified. Figure 2The schematic diagram of the network architecture shown in FIG. 1 shows that the outermost hexagon represents the main network, and the small hexagons in the main network represent the indoor equipment coverage network established due to insufficient network resources. The main network is formed by the main network base station, and the indoor equipment coverage network is formed by the base station in the indoor equipment coverage. All devices in the indoor equipment coverage network are called secondary user devices. Figure 2 The user devices other than the secondary user devices in the middle rectangle are called primary user devices. In this embodiment, the primary network base station can be represented as PU, the base station in the coverage area of ​​the indoor equipment can be represented as SBS, the primary user devices can be represented as PUs, and the secondary user devices can be represented as SUs.

[0043] In step S1, an undirected graph model is dynamically established according to the network status information, and the undirected graph model can also be called an environment model. The nodes in the undirected graph model may include: base station nodes and device nodes, the edges between the base station nodes and the device nodes represent signal links, and the edges between the device nodes represent interference links. Specifically, in the process of establishing the undirected graph model, according to the current network status information, the identified base station is used as a base station node, the device is used as a device node, the edge between the base station node and the device node is drawn according to the signal link between the base station and the device, and the edge between the device nodes is drawn according to the interference link between the devices, thereby completing the establishment of the undirected graph model. Among them, the base station node may further include a main network base station node (i.e., a PU node) and a base station node (i.e., an SBS node) in the coverage range of the indoor distribution equipment, and the device node may further include a main user equipment node (PUs node) and a secondary user equipment node (SUs node). The nodes in the undirected graph model can be represented by PU nodes, SBS nodes, PUs nodes, and SUs nodes. The edges between PU nodes, SBS nodes and PUs nodes, SUs nodes represent signal links, and the weights of the edges can be represented by channel gains; the edges between PU nodes and SBS nodes represent interference links; and the edges between PUs nodes and SUs nodes represent interference links.

[0044] In the undirected graph model, the node information of the base station node may include: the location information of the base station node, the movement direction of the base station node, the movement speed of the base station node and the current channel occupancy vector, wherein the movement direction of the base station node is 0 and the movement speed of the base station node is 0; the node information of the device node includes: the location information of the device node, the movement direction of the device node, the movement speed of the device node and the current channel usage vector.

[0045] After the undirected graph model is established, the undirected graph model can be analyzed to obtain a user distance matrix and a user resource occupancy matrix. Then, based on the user distance matrix and the user resource occupancy matrix, an environment matrix is ​​generated. The environment matrix can be expressed by the following formula (1):

[0046] S = {D(t), R(t)} formula (1);

[0047] Among them, S represents the environment matrix, specifically the graph Laplacian matrix generated by the undirected graph model; D(t) represents the user distance matrix, which is used to reflect the position matrix of each node in the current environment; R(t) represents the user resource occupancy matrix, which is used to reflect the occupancy matrix of each channel resource in the current environment.

[0048] Step S2: Process the user demand information and the environment matrix using the decision model to obtain a channel selection vector and an energy adaptation vector.

[0049] When the SUs node proposes a spectrum sharing requirement, the user demand information and the environment matrix generated by the undirected graph model can be input into the decision model, and the user demand information and the environment matrix can be processed by the decision model to output the channel selection vector and the energy adaptation vector. Among them, the user demand information may include: the node identifier of the SUs node that proposes the spectrum sharing requirement, etc. The decision model can be established using a graph convolution (GCN) neural network, and specifically can be established using a GCN-ELM (Extreme Learning Machine) neural network. Among them, the decision model may include: multiple graph convolution layers, a single hidden layer feedforward neural network, and a softmax layer. ELM is a fast machine learning algorithm that can be applied to the learning of a single hidden layer feedforward neural network. It has the advantages of simple structure, fast learning speed, and good generalization performance.

[0050] Figure 3 A schematic diagram of a decision model is shown, such as Figure 3As shown, the decision model includes: two graph convolution layers, a single hidden layer feedforward neural network and a softmax layer, wherein the two graph convolution layers are respectively referred to as graph convolution layer 1 and graph convolution layer 2. After the environment matrix is ​​input into the decision model, the graph convolution layer 1 in the decision model first processes it, and then the output result of the graph convolution layer 1 is processed by linear rectification (ReLU) and then input into the graph convolution layer 2 for further processing, and the output result of the graph convolution layer 2 is processed by linear rectification and then input into the single hidden layer feedforward neural network for further processing. In this embodiment, the decision model adopts a two-layer GCN structure for feature extraction, and after feature extraction, a single hidden layer feedforward neural network is used for decision processing, and then the data result output by the single hidden layer feedforward neural network is processed by the softmax method to obtain a channel selection vector and an energy adaptation vector.

[0051] Specifically, the convolution process of the convolution layer can be calculated using the following formula (2):

[0052] H l =σ(SH l-1 W l ) formula (2);

[0053] Among them, H l represents the output of the lth convolutional layer; σ represents the activation function; S represents the graph Laplacian matrix (i.e., the environment matrix) generated by the undirected graph model; H l-1 represents the output of the l-1th convolutional layer; W l represents the parameter matrix connecting the l-1th convolutional layer and the lth convolutional layer. For S, the following formula (3) can be used for calculation:

[0054]

[0055] Among them, x i represents the input data of the i-th node, x j Represents the input data of the jth node.

[0056] In this embodiment, a GCN network consisting of two graph convolutional layers is used, and the output of the graph convolutional neural network can be calculated using the following formula (4):

[0057] GCN output =σ(Sσ(SXW 0 )W 1 ) formula (4);

[0058] Among them, GCN output represents the output result of the graph convolutional neural network; S represents the graph Laplacian matrix generated by the undirected graph model; X represents the behavior matrix or power matrix at the current moment; W 0Represents the parameter matrix between the input layer and the first convolutional layer; W 1 Represents the parameter matrix between the first convolution layer and the second convolution layer.

[0059] Furthermore, the output of the GCN-ELM structure can be calculated using the following formula (5):

[0060] Policy output =softmax(σ(GCN output W1+b)W2) Formula (5);

[0061] Among them, Policy output represents the output result of the GCN-ELM structure; W1 represents the ELM network (i.e. Figure 3 The single hidden layer feedforward neural network in (a) represents the parameter matrix from the input layer to the hidden layer; b represents the bias vector of the hidden layer neurons; W2 represents the parameter matrix from the hidden layer to the output layer in the ELM network.

[0062] Step S3, judging whether the preset convergence condition is met through the objective function according to the channel selection vector, the energy adaptation vector and the environment matrix; if so, executing step S4; if not, executing step S5.

[0063] Step S4: performing spectrum allocation according to the current channel selection vector and energy adaptation vector.

[0064] If it is determined in step S3 that the preset convergence condition is met, spectrum allocation is performed according to the current channel selection vector and energy adaptation vector, and the method ends. Spectrum allocation is performed according to the channel selection vector and energy adaptation vector that meet the preset convergence condition, which can obtain a better spectrum access solution, improve the spectrum utilization efficiency, and conveniently complete the secondary user equipment spectrum sharing function.

[0065] Step S5, updating the parameters of the decision model.

[0066] If it is determined in step S3 that the preset convergence condition is not met, the parameters of the decision model are updated, and then the process jumps to step S2 to continue to use the decision model to process the user demand information and the environment matrix to obtain a new channel selection vector and a new energy adaptation vector. In the process of updating the parameters of the decision model, the parameters of multiple graph convolutional layers in the decision model can be updated using a gradient descent algorithm based on the objective function, and the parameters of a single hidden layer feedforward neural network in the decision model can be updated using an approximate solution algorithm.

[0067] Figure 4 A schematic diagram of the architecture of a spectrum sharing method based on a graph convolutional neural network is shown, Figure 4As shown, when the SUs node proposes a spectrum sharing requirement, the user demand information and the environment matrix generated by the undirected graph model can be input into the decision model, and the user demand information and the environment matrix can be processed by the decision model to output the channel selection vector and the energy adaptation vector. Then, according to the channel selection vector, the energy adaptation vector and the environment matrix, the objective function is used to determine whether the preset convergence condition is met, wherein the objective function includes the feedback function r(t). The preset convergence condition may include: the sum of the data rates of all secondary user equipment nodes is the maximum value. If the preset convergence condition is not met, the network structure of the decision model is optimized based on the cost function composed of the feedback function, the channel selection vector and the energy adaptation vector, and the parameters of the decision model are updated. Through the reinforcement learning process, the preset convergence condition is finally met, and the spectrum access method with the maximum network rate is output, thereby effectively improving the spectrum utilization efficiency. Among them, those skilled in the art can also set the preset convergence condition according to actual needs, which is not limited here. Through the above processing method, a spectrum sharing solution that guarantees the maximum data rate can be easily obtained.

[0068] The feedback function r(t) can be calculated using the following formula (6):

[0069]

[0070] Where S represents the total number of PU nodes; V represents the total number of SBS nodes; C sv Represents the data rate of each SUs node. sv The following formula (7) can be used for calculation:

[0071]

[0072] Among them, C s,v [n] represents the data rate of the nth SUs node; is a binary indicator, indicating whether the nth channel is allocated and occupied; W represents the channel bandwidth; N represents the number of channels; ξ s,v [n] represents the signal-to-noise ratio of the nth channel. s,v [n] can be calculated using the following formula (8):

[0073]

[0074] Among them, h s,v [n] represents the channel gain between s and v; P s,v [n] represents the power of the interference link between s and v; σ 2 is Gaussian white noise; I s,v [n] represents the interference of all links. s,v[n] can be calculated using the following formula (9):

[0075]

[0076] Where U represents the number of PUs nodes; represents the channel gain between PUs node and SBS node; It represents the transmission power of the interference link between PUs node and SUs node in the same channel; represents the channel gain between the SBS node and the SUs node; It represents the transmission power of the interference link from other SUs nodes in the same channel under the coverage of the same SBS node.

[0077] Furthermore, the alternating optimization strategy of the GCN-ELM model training method is optimized, and after the graph feature extraction, the parameters are optimized according to the objective functions required by different requirements. Among them, the objective function may also include: the objective function corresponding to the channel selection vector and the objective function corresponding to the energy adaptation vector. The objective function Loss1 corresponding to the channel selection vector and the objective function Loss2 corresponding to the energy adaptation vector can be expressed by the following formulas (10) and (11) respectively:

[0078]

[0079]

[0080] Among them, Z represents the total number of nodes; T represents the total number of moments; Represents the feedback function value of the behavior vector at time t; represents the feedback function value of the power vector at time t; t represents a certain moment; r a Represents the matrix composed of feedback function values ​​obtained based on the behavior matrix at all times and node calculations; r p It represents the matrix composed of the feedback function values ​​obtained based on the power matrix at all times and node calculations; L represents the Laplace matrix generated by the graph structure, that is, the environment matrix.

[0081] According to the spectrum sharing method based on graph convolutional neural network provided by this embodiment, a graph learning method is introduced into the decision model constructed by using GCN, which has stronger adaptability compared with the prior art; an undirected graph model is dynamically established according to network state information, and the decision model is used for processing, which effectively solves the problem of imperfect graph structure when establishing the decision model in the prior art; the undirected graph model is introduced into the reinforcement learning process based on the graph convolutional neural network, and the channel selection vector and the energy adaptation vector are output. According to the channel selection vector, the energy adaptation vector and the environment matrix, the objective function is used for judgment, so that a better spectrum access scheme can be obtained quickly and conveniently, the spectrum sharing function of the secondary user equipment is completed, and the spectrum utilization efficiency is improved; and the location information and other feature information are considered separately in the decision model, so that the features of the location information will not cause information loss in the feature extraction process due to the influence of other information, thereby improving the efficiency of spectrum sharing; in addition, in actual use, the features of the input data can be conveniently added to obtain the optimal graph structure, thereby ensuring the scalability of this scheme.

[0082] Figure 5 FIG. 4 shows a structural block diagram of a spectrum sharing device based on a graph convolutional neural network according to an embodiment of the present invention. Figure 5 As shown, the device includes: an environment processing module 501, a decision processing module 502, a judgment module 503, a spectrum allocation module 504 and a parameter updating module 505.

[0083] The environment processing module 501 is adapted to: obtain current network status information, establish an undirected graph model according to the network status information, and generate an environment matrix using the undirected graph model.

[0084] The nodes in the undirected graph model include: base station nodes and device nodes.

[0085] The decision processing module 502 is adapted to: process the user demand information and the environment matrix using the decision model to obtain a channel selection vector and an energy adaptation vector.

[0086] The judgment module 503 is adapted to judge whether a preset convergence condition is satisfied through an objective function according to the channel selection vector, the energy adaptation vector and the environment matrix.

[0087] The spectrum allocation module 504 is adapted to: if the determination module 503 determines that the preset convergence condition is met, perform spectrum allocation according to the current channel selection vector and energy adaptation vector.

[0088] The parameter updating module 505 is adapted to update the parameters of the decision model and then trigger the decision processing module 502 if the judgment module 503 judges that the preset convergence condition is not met.

[0089] Optionally, the edge between the base station node and the device node represents a signal link, and the edge between the device nodes represents an interference link; the base station nodes include: a main network base station node and a base station node within the coverage range of the indoor equipment; the device nodes include: a primary user device node and a secondary user device node.

[0090] Optionally, the node information of the base station node includes: location information, movement direction, movement speed and current channel occupancy vector; the node information of the device node includes: location information, movement direction, movement speed and current channel usage vector.

[0091] Optionally, the environment processing module 501 is further adapted to: analyze the undirected graph model to obtain a user distance matrix and a user resource occupancy matrix; and generate an environment matrix according to the user distance matrix and the user resource occupancy matrix.

[0092] Optionally, the decision model is established using a graph convolutional neural network; the decision model includes: multiple graph convolutional layers, a single hidden layer feedforward neural network and a softmax layer.

[0093] Optionally, the parameter updating module 505 is further adapted to: update the parameters of multiple graph convolutional layers in the decision model using a gradient descent algorithm, and update the parameters of a single hidden layer feedforward neural network in the decision model using an approximate solution algorithm.

[0094] Optionally, the preset convergence condition includes: the sum of data rates of all secondary user equipment nodes is a maximum value.

[0095] According to the spectrum sharing device based on graph convolutional neural network provided by this embodiment, a graph learning method is introduced into the decision model constructed by using GCN, which has stronger adaptability compared with the prior art; an undirected graph model is dynamically established according to network status information, and the decision model is used for processing, which effectively solves the problem of imperfect graph structure when establishing the decision model in the prior art; the undirected graph model is introduced into the reinforcement learning process based on the graph convolutional neural network, and the channel selection vector and the energy adaptation vector are output. According to the channel selection vector, the energy adaptation vector and the environment matrix, the objective function is used for judgment, so that a better spectrum access scheme can be obtained quickly and conveniently, the spectrum sharing function of the secondary user equipment is completed, and the spectrum utilization efficiency is improved; and the location information and other feature information are considered separately in the decision model, so that the features of the location information will not cause information loss in the feature extraction process due to the influence of other information, thereby improving the efficiency of spectrum sharing; in addition, in actual use, the features of the input data can be conveniently added to obtain the optimal graph structure, thereby ensuring the scalability of this scheme.

[0096] The present invention also provides a non-volatile computer storage medium, which stores at least one executable instruction, and the executable instruction can execute the spectrum sharing method based on graph convolutional neural network in any of the above method embodiments.

[0097] Figure 6 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0098] like Figure 6 As shown, the computing device may include: a processor (processor) 602 , a communications interface (Communications Interface) 604 , a memory (memory) 606 , and a communication bus 608 .

[0099] in:

[0100] The processor 602 , the communication interface 604 , and the memory 606 communicate with each other via a communication bus 608 .

[0101] The communication interface 604 is used to communicate with other devices such as clients or other servers.

[0102] The processor 602 is used to execute the program 610, and specifically can execute the relevant steps in the above-mentioned embodiment of the spectrum sharing method based on graph convolutional neural network.

[0103] Specifically, the program 610 may include program codes, which include computer operation instructions.

[0104] The processor 602 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0105] The memory 606 is used to store the program 610. The memory 606 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0106] Program 610 can be specifically used to enable processor 602 to execute the spectrum sharing method based on graph convolutional neural network in any of the above-mentioned method embodiments. The specific implementation of each step in program 610 can refer to the corresponding description in the corresponding steps and units in the above-mentioned spectrum sharing embodiment based on graph convolutional neural network, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the aforementioned method embodiment, which will not be repeated here.

[0107] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0108] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0109] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those expressly recited in each claim. Rather, as reflected in the claims, inventive aspects lie in less than all of the features of the individual embodiments previously disclosed. Therefore, the claims that follow the detailed description are hereby expressly incorporated into the detailed description, with each claim itself serving as a separate embodiment of the present invention.

[0110] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0111] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims, any one of the claimed embodiments may be used in any combination.

[0112] The various component embodiments of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0113] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

Claims

1. A spectrum sharing method based on graph convolutional neural network, characterized in that: The method comprises: Step S1, obtaining current network status information, establishing an undirected graph model according to the network status information, and generating an environment matrix using the undirected graph model; wherein the nodes in the undirected graph model include: base station nodes and device nodes; Step S2, using a decision model to process the user demand information and the environment matrix to obtain a channel selection vector and an energy adaptation vector; wherein the decision model includes: multiple graph convolution layers, a single hidden layer feedforward neural network and a softmax layer; the multiple graph convolution layers in the decision model sequentially extract features from the environment matrix, use a single hidden layer feedforward neural network for decision processing, and use a softmax layer to process the data results output by the single hidden layer feedforward neural network to obtain a channel selection vector and an energy adaptation vector; Step S3, judging whether a preset convergence condition is met through an objective function according to the channel selection vector, the energy adaptation vector and the environment matrix; if so, executing step S4; if not, executing step S5; wherein the objective function includes a feedback function r(t), an objective function Loss1 corresponding to the channel selection vector and an objective function Loss2 corresponding to the energy adaptation vector; Where S represents the total number of base station nodes in the main network; V represents the total number of base station nodes in the coverage area of ​​the indoor distribution equipment; C sv represents the data rate of each secondary user equipment node; Z represents the total number of nodes; T represents the total number of time; Represents the feedback function value of the behavior vector at time t; represents the feedback function value of the power vector at time t; t represents a certain time; the channel selection vector and r a Related, a Represents the matrix composed of the feedback function values ​​obtained based on the behavior matrix for all time and node calculations; the energy adaptation vector and r p Related, p represents the matrix composed of the feedback function values ​​obtained based on the power matrix at all times and node calculations; L represents the environment matrix; Policy output Represents the output result of the GCN-ELM structure; Step S4, performing spectrum allocation according to the current channel selection vector and energy adaptation vector, and the method ends; Step S5, updating the parameters of the decision model, and then jumping to step S2.

2. The method according to claim 1, characterized in that The edges between the base station nodes and the device nodes represent signal links, and the edges between the device nodes represent interference links; The base station nodes include: a main network base station node and a base station node within the coverage of indoor equipment; the equipment nodes include: a main user equipment node and a secondary user equipment node.

3. The method according to claim 1, characterized in that The node information of the base station node includes: location information, movement direction, movement speed and current channel occupancy vector; The node information of the device node includes: location information, movement direction, movement speed and current channel usage vector.

4. The method according to claim 1, characterized in that: The generating of the environment matrix by using the undirected graph model further comprises: Analyze the undirected graph model to obtain a user distance matrix and a user resource occupancy matrix; An environment matrix is ​​generated according to the user distance matrix and the user resource occupancy matrix.

5. The method according to claim 1, characterized in that: The decision model is established using a graph convolutional neural network.

6. The method according to claim 5, characterized in that The updating of the parameters of the decision model further comprises: The parameters of the multiple graph convolutional layers in the decision model are updated using a gradient descent algorithm, and the parameters of the single hidden layer feedforward neural network in the decision model are updated using an approximate solution algorithm.

7. The method according to any one of claims 1 to 6, characterized in that: The preset convergence condition includes: the sum of data rates of all secondary user equipment nodes is a maximum value.

8. A spectrum sharing device based on graph convolutional neural network, characterized in that: The device comprises: An environment processing module is adapted to obtain current network status information, establish an undirected graph model according to the network status information, and generate an environment matrix using the undirected graph model; wherein the nodes in the undirected graph model include: base station nodes and device nodes; A decision processing module is adapted to process the user demand information and the environment matrix using a decision model to obtain a channel selection vector and an energy adaptation vector; wherein the decision model includes: multiple graph convolution layers, a single hidden layer feedforward neural network and a softmax layer; the multiple graph convolution layers in the decision model sequentially extract features from the environment matrix, use a single hidden layer feedforward neural network for decision processing, and use a softmax layer to process the data results output by the single hidden layer feedforward neural network to obtain a channel selection vector and an energy adaptation vector; A judgment module, adapted to judge whether a preset convergence condition is met through an objective function according to the channel selection vector, the energy adaptation vector and the environment matrix; wherein the objective function includes a feedback function r(t), an objective function Loss1 corresponding to the channel selection vector and an objective function Loss2 corresponding to the energy adaptation vector; Where S represents the total number of base station nodes in the main network; V represents the total number of base station nodes in the coverage area of ​​the indoor distribution equipment; C sv represents the data rate of each secondary user equipment node; Z represents the total number of nodes; T represents the total number of time; Represents the feedback function value of the behavior vector at time t; represents the feedback function value of the power vector at time t; t represents a certain time; the channel selection vector and r a Related, a Represents the matrix composed of the feedback function values ​​obtained based on the behavior matrix for all time and node calculations; the energy adaptation vector and r p Related, p represents the matrix composed of the feedback function values ​​obtained based on the power matrix at all times and node calculations; L represents the environment matrix; Policy output Represents the output result of the GCN-ELM structure; a spectrum allocation module, adapted to perform spectrum allocation according to the current channel selection vector and energy adaptation vector if the determination module determines that the preset convergence condition is met; The parameter updating module is adapted to update the parameters of the decision model and then trigger the decision processing module if the judgment module determines that the preset convergence condition is not met.

9. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the spectrum sharing method based on graph convolutional neural network as described in any one of claims 1-7.

10. A computer storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the spectrum sharing method based on a graph convolutional neural network as described in any one of claims 1 to 7.

Images