A Spectral Collision Time Ratio Prediction Method Based on GC-2LSTM-NN Model

Abstract

This invention belongs to the field of cognitive radio, specifically relating to a spectrum conflict time ratio prediction method based on a GC-2LSTM-NN model. Spatiotemporal data of secondary transmitters are obtained from an asynchronous cognitive radio network. Historical information of the secondary transmitters in the past ξ time slots is extracted from the perception matrix of time slot t and input into the GC-2LSTM-NN model. The GC-2LSTM-NN model includes a graph convolutional layer and a two-layer LSTM. The two-layer LSTM includes N individual LSTMs, a swarm LSTM, and a linear layer. The final output is the predicted conflict time ratio vector for H time slots of time slot t. The GC-2LSTM-NN model is trained using a backpropagation algorithm. The trained GC-2LSTM-NN model is then used to predict the conflict time ratio vector for the H time slots of each time slot. This invention improves the accuracy of conflict time ratio prediction.

Description

Technical Field

[0001] This invention belongs to the field of cognitive radio, particularly the field of spectrum prediction technology in cognitive radio, and specifically relates to a method for predicting the time ratio of spectrum collisions based on the GC-2LSTM-NN (Graph Convolution and Two-tier Long Short-Term Memory Neural Network) model. Background Technology

[0002] To address the shortage of spectrum resources, cognitive radio networks employ dynamic spectrum access technology, allowing unlicensed wireless communication devices to share idle spectrum resources of licensed devices under specific conditions. Simultaneously, to address the challenge of insufficient power supply for wireless communication devices, energy capture technology enables these devices to obtain energy from the environment or dedicated power sources. In asynchronous energy capture cognitive radio networks, the primary user authorizes the primary transmitter to access a certain bandwidth of spectrum resources, enjoying priority access to perform the primary user's data transmission tasks. Secondary users, through spectrum sensing, spectrum decision-making, spectrum sharing, and spectrum switching, enable secondary transmitters to opportunistically access the primary user's licensed spectrum using dynamic access and capture energy from the primary user's radio frequency signals.

[0003] Because the spectrum state of asynchronous energy-capture cognitive radio networks can change at any time, and the severity of transmission conflicts differs between primary and secondary users, it is necessary to not only monitor spectrum state changes but also understand the impact of transmission conflicts on primary user transmissions. The conflict time ratio can reflect both spectrum state and the impact of transmission conflicts between primary and secondary users. Therefore, the conflict time ratio becomes a key reference for subsequent scheduling tasks regarding the impact of spectrum state and transmission conflict severity.

[0004] Predicting spectrum status has become a critical issue in wireless spectrum management and cognitive radio networks. Spectrum status prediction technology can effectively help the fusion center predict spectrum status, providing crucial support for subsequent scheduling tasks. In asynchronous energy capture cognitive radio networks, the secondary receiver, acting as the fusion center, records and utilizes a deep learning model to fuse historical information from the secondary transmitter. This information includes multi-dimensional spatiotemporal variation features composed of the secondary transmitter's actions (such as energy capture or spectrum sensing), local sensing results, and local energy status, to predict future spectrum conflict time ratios. Accurate prediction of spectrum status facilitates the correct implementation of subsequent dynamic scheduling strategies, thereby improving spectrum utilization. Therefore, a deep learning model capable of effectively learning and fusing the aforementioned multi-dimensional spatiotemporal variation features is essential for the fusion center to predict spectrum status.

[0005] However, learning and fusing the multidimensional spatiotemporal variation features in the historical information of secondary transmitters is highly complex, making accurate prediction of the spectrum conflict time ratio a challenge. Currently, traditional deep learning models, such as those combining Convolutional Neural Networks (CNNs) and LSTMs, are mostly used only to predict spectrum-sensing data such as the cumulative intensity of signals received by secondary transmitters. Furthermore, CNNs are typically used to process Euclidean spatial data with regular shapes, such as images, while LSTMs are used to handle correlations between time series; neither can describe the spatial correlation between secondary transmitters. Graph data structures can well describe the spatial correlation of secondary transmitters, and with the rapid development of graph neural networks in recent years, graph convolutional neural networks can effectively extract the spatial correlation of graph data and have been widely applied to prediction tasks of non-Euclidean spatial graph data such as knowledge graphs and network traffic. However, graph convolutional neural networks cannot effectively extract the temporal correlation between historical information of secondary transmitters. Therefore, there is an urgent need for a deep learning model that can effectively fuse the multidimensional features of graph data in terms of both space and time series, and can predict the spectrum conflict time ratio as accurately as possible, providing more accurate spectrum state information for subsequent scheduling tasks. Summary of the Invention

[0006] The purpose of this invention is to address the problems that convolutional neural network-LSTM fusion models cannot describe the spatial correlation between secondary transmitters, graph convolutional neural networks cannot effectively extract the temporal correlation between historical information of secondary transmitters, and it is difficult to accurately predict the spectrum conflict time ratio. Therefore, this invention proposes a spectrum conflict time ratio prediction method based on the GC-2LSTM-NN model.

[0007] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0008] A method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model includes:

[0009] Construct an asynchronous cognitive radio network, which includes a main network and a secondary network. The main network contains a primary user and a licensed channel, and the secondary network includes N single-antenna secondary transmitters and N antenna secondary receivers. The transmission phase of the secondary network is divided into H equal-length time slices.

[0010] Spatiotemporal data of the secondary transmitter is obtained from an asynchronous cognitive radio network, the spatiotemporal data including an adjacency matrix with the secondary transmitter as a node. And the sensing matrix consisting of historical information from the secondary transmitter received by the secondary receiver within a predetermined time period;

[0011] A GC-2LSTM-NN model is constructed. Historical information of the secondary transmitters in the past ξ time slots is extracted from the perception matrix of time slot t to obtain the extracted perception matrix. The true collision time ratio vector y' of the H time slices contained in time slot t is calculated as the label of the extracted perception matrix. Based on the extracted perception matrix and the adjacency matrix... Construct an independent undirected graph for each time slot

[0012] Given ξ undirected graphs The graph convolutional layer in the GC-2LSTM-NN model is fed into the graph to compute ξ spatially augmented graph feature matrices.

[0013] Based on ξ spatial enhancement map feature matrices Obtain the spatiotemporal characteristic matrix f st Extract the spatiotemporal feature matrix f st The time-series characteristics of N individual entities corresponding to each sub-transmitter in the time dimension;

[0014] The temporal features of N individuals are input into the first layer of the N individual LSTM in the two-layer LSTM, and the N enhanced individual temporal feature results are output.

[0015] The N enhanced individual temporal feature results are stacked on the temporal feature dimension to obtain global temporal features. The global temporal features are then input into the second layer of the population LSTM in the two-layer LSTM to obtain the enhanced global temporal feature results.

[0016] The enhanced global temporal feature results are flattened into a one-dimensional feature vector and input into a linear layer in a two-layer LSTM for fusion. The Sigmoid function is combined to output the predicted conflict time ratio vector y of the H time slices in the t-th time slot.

[0017] The mean squared error between the actual conflict time ratio vector y' and the predicted conflict time ratio vector y is calculated as the loss function, and the GC-2LSTM-NN model is trained using the backpropagation algorithm.

[0018] The trained GC-2LSTM-NN model is used to predict the conflict time ratio vector of H time slices in each time slot.

[0019] Furthermore, the adjacency matrix The adjacency matrix is ​​calculated based on the physical two-dimensional coordinates and radial basis functions of the secondary transmitter. matrix components The calculation formula is as follows:

[0020]

[0021] Where, di Let d represent the physical two-dimensional coordinates of transmitter i, i∈[1,N], and d j Let ρ represent the physical two-dimensional coordinates of the transmitter j, j∈[1,N], and ρ represent the preset hyperparameters.

[0022] Furthermore, the step of extracting the historical information of the sub-transmitters from the past ξ time slots in the sensing matrix of the t-th time slot to obtain the extracted sensing matrix includes:

[0023] Each time slot includes a control phase and a transmission phase. The control phase is divided into a sensing sub-phase and a reporting sub-phase.

[0024] In the sensing sub-phase, each sub-transmitter independently performs sensing, letting y i (k) represents the kth sample of the main network signal received by the secondary transmitter i, expressed by the following formula:

[0025]

[0026] Among them, w i (k) represents the noise of the k-th sample of the main network signal received by the secondary transmitter i, v i (k) represents the k-th sample of the main network signal received by the secondary transmitter i, which is the main transmitter signal. and These represent two scenarios: one where the main network is not using spectrum resources and the other where it is using spectrum resources.

[0027] The specific sensing data is the cumulative received signal strength of the secondary transmitter, expressed by the formula:

[0028]

[0029] Among them, R i The cumulative received signal strength of transmitter i is represented by K, and K represents the number of samples of the main network signal received by transmitter i.

[0030] During the reporting sub-phase, each sub-transmitter reports the cumulative received signal strength and remaining energy to the sub-receiver. If the remaining energy of a sub-transmitter is less than the preset energy, it remains silent during the reporting sub-phase.

[0031] Define the energy indicator for energy capture during the transmission phase as follows:

[0032]

[0033] Where, μ i (t) represents the energy indicator for energy capture during the transmission phase of transmitter i in time slot t. This represents the remaining energy of transmitter i at the end of the control phase in the t-th time slot. This represents the remaining energy of transmitter i at the start of the control phase in the (t+1)th time slot, when μ i When (t) = 1, it indicates that transmitter i was charged during the transmission phase of the t-th time slot. i When (t) = -1, it indicates that transmitter i is inactive during the transmission phase of the t-th time slot. i When (t) = 0, it means that transmitter i transmitted an update packet during the transmission phase of the t-th time slot;

[0034] Let λ i (t) indicates that in the t-th time slot, the transmission indicator corresponding to the update packet sent by transmitter i is:

[0035]

[0036] Where, when λ i (t) = 1 indicates that the secondary transmitter i successfully transmitted the update packet to the secondary receiver in the t-th time slot without a collision. i (t) = -1 indicates that the update packet transmission failed, when λ i (t) = 0 indicates other cases;

[0037] The extracted perception matrix S(t) is expressed by the following formula:

[0038]

[0039] Among them, s i (t)=(μ i (t-1),λ i (t-1),R i (t) represents the historical information of the sub-transmitter i provided in time slot t, where i∈[1,N], μ i (t-1) represents the energy indicator for energy capture during the transmission phase of transmitter i in the (t-1)th time slot, λ i (t-1) represents the transmission indicator corresponding to the update packet sent by transmitter i in the (t-1)th time slot, R i (t) represents the cumulative received signal strength of transmitter i in the t-th time slot. Where N represents the number of individual transmitters, ξ represents the time dimension, and |s i (t)| represents the dimension of the specific content of the historical information provided by the sub-transmitter i in time slot t.

[0040] Furthermore, the calculation of the actual conflict time ratio vector y' of the H time slices contained in the t-th time slot includes:

[0041] Regarding the situation The actual conflict time ratio of the h-th time slice in the t-th time slot Given by the following formula:

[0042]

[0043] Where, τ t τ represents the duration of the transmission phase. I (t) represents the time the main user has been idle, x represents the duration of the main user's idle or busy state, and z represents the differential variable.

[0044] Regarding the situation The actual conflict time ratio of the h-th time slice in the t-th time slot Given by the following formula:

[0045]

[0046] in, as well as τ B (t) represents the time the main user has been continuously busy;

[0047] The actual conflict time ratios of H time slices are combined to obtain the actual conflict time ratio vector y' of the t-th time slot.

[0048] Furthermore, the step of using the extracted perception matrix and adjacency matrix... Construct an independent undirected graph for each time slot This includes: Let the shape of all components in the same column of the extracted perception matrix be N×|s i The graphical signal feature matrix X of (t)| t The feature matrix X of the graph signal t Combined with adjacency matrix Model as an undirected graph in spatial dimensions

[0049] Furthermore, the ξ undirected graphs The graph convolutional layer in the GC-2LSTM-NN model is fed into the graph to compute ξ spatially augmented graph feature matrices. The calculation formula is as follows:

[0050]

[0051] Where ReLU represents the activation function, and its calculation expression is: The adjacency matrix representing the self-connection relationships of the nodes is... I represents the identity matrix. express The angle matrix, diagonal components of the matrix express The matrix components, W GCN These are the learnable weights for the graph convolutional layer.

[0052] Furthermore, the step based on ξ spatial enhancement map feature matrices... Obtain the spatiotemporal characteristic matrix f st ,include:

[0053] ξ spatial enhancement map feature matrices Sort and stack them according to the time dimension to obtain the spatiotemporal feature matrix f st The formula is as follows:

[0054]

[0055] Where Concatenate represents a stacking operation, F E This indicates the number of enhanced spatial features, representing the feature dimension.

[0056] Compared with existing technologies, the significant advantages of this invention are as follows: The GC-2LSTM-NN model utilizes graph convolutional layers (GCN) to enhance the representation of spatial features and employs a two-layer LSTM to learn time-series features from individual to global perspectives, thereby improving the learning and fusion capabilities of multidimensional spatiotemporal variation features in non-Euclidean spatial data. The GC-2LSTM-NN model also improves the accuracy of conflict time ratio prediction, providing reliable information for subsequent scheduling tasks. Attached Figure Description

[0057] Figure 1 This is a flowchart of the overall structure of the present invention;

[0058] Figure 2 This is a diagram of the spatial feature fusion structure of the GC-2LSTM-NN model of this invention;

[0059] Figure 3 This is a diagram of the temporal feature fusion structure of the GC-2LSTM-NN model of this invention;

[0060] Figure 4 This is a training loss evaluation diagram of the GC-2LSTM-NN model and the comparison model of this invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0062] This invention provides a method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model, such as... Figure 1 As shown, the specific method is as follows:

[0063] (1) Construct an asynchronous cognitive radio network and create a dataset. The constructed asynchronous cognitive radio network includes a main network and a secondary network. The main network contains a primary user and a licensed channel. The secondary network includes N single-antenna secondary transmitters and one N-antenna secondary receiver. The secondary transmitters are equipped with energy harvesting modules and RF link modules, and the secondary receivers are equipped with stable power supplies. There is no synchronous time-division communication protocol between the main network and the secondary network. The secondary network follows a time-division multiple access protocol, namely the H-TDMA protocol. Each time slot includes a control phase and a transmission phase. The transmission phase of the secondary network is divided into H equal-length time slices. The idle state of the licensed channel of the primary user follows a two-parameter Weibull distribution with a scale parameter α and a shape parameter λ. Its idle duration is expressed as T. I The busy state of the channel follows a two-parameter Weibull distribution with a scale parameter β and the same shape parameter λ as the idle state, and its busy duration is denoted as T. B T I and T B The expectations can be expressed as follows: The secondary network dynamically accesses the authorized channel of the primary network to transmit update packets.

[0064] Furthermore, a control phase model is presented, which is divided into a sensing sub-phase and a reporting sub-phase: In the sensing sub-phase, each sub-transmitter independently performs sensing, letting y i (k) represents the k-th sample of the main network signal received by the secondary transmitter i, which can be expressed as the following binary hypothesis test, where... and We will assume two scenarios: one where the main network is not using spectrum resources and the other where it is using spectrum resources.

[0065]

[0066] Among them, w i (k) represents the noise of the k-th sample of the main network signal received by the secondary transmitter i, v i (k) represents the primary transmitter signal of the kth sample of the primary network signal received by secondary transmitter i. The specific sensing data is the cumulative received signal strength of the secondary transmitter, expressed as:

[0067]

[0068] Among them, R i denoted by , where represents the cumulative received signal strength of transmitter i, and K represents the number of samples of the main network signal received by transmitter i.

[0069] In the reporting sub-phase, each secondary transmitter first reports the accumulated received signal strength and remaining energy to the secondary receiver, which acts as the fusion center. Specifically, this is done by making... This represents the remaining energy of transmitter i at the start of the control phase in the t-th time slot. This represents the remaining energy of the secondary transmitter i at the end of the control phase in the t-th time slot. If the secondary transmitter does not have enough energy to report its accumulated received signal strength and remaining energy, it remains silent during the reporting sub-phase.

[0070] Furthermore, the process of the secondary receiver generating the sensing matrix S(t) is given. After receiving reports from all secondary transmitters, the secondary receiver performs information fusion: First, the energy indicator for energy capture during the transmission phase is defined as:

[0071]

[0072] in, This represents the remaining energy of transmitter i at the start of the control phase in the (t+1)th time slot, when μ i When (t) = 1, it indicates that transmitter i was charged during the transmission phase of the t-th time slot. i When (t) = -1, it indicates that transmitter i is inactive during the transmission phase of the t-th time slot. i When (t) = 0, it indicates that transmitter i transmitted an update packet during the transmission phase of the t-th time slot.

[0073] Secondly, the transmission results of past update packets record some historical information about the past state of the primary user network, let λ i (t) indicates that in the t-th time slot, the transmission indicator corresponding to the update packet sent by transmitter i is:

[0074]

[0075] Where, when λ i (t) = 1 indicates that the secondary transmitter i successfully transmitted the update packet to the secondary receiver in the t-th time slot without a collision. i (t) = -1 indicates that the update packet transmission failed, when λ i (t) = 0 indicates other cases. The busy state of the channel can be determined from whether the transmission is successful or not. It should be noted that from the perspective of the secondary receiver, it knows the transmission indicators of all secondary transmitters. Therefore, the secondary transmitter only needs to report its cumulative received signal strength and remaining energy to the secondary receiver in the reporting sub-phase.

[0076] In this embodiment of the invention, the sensing matrix S(t) constructed for time slot t consists of the cumulative received signal strength, energy indicator, and transmission indicator of the past ξ time slots, such as:

[0077]

[0078] Among them, the components s of the perception matrix S(t) i (t)=(μ i (t-1),λ i (t-1),R i (t) corresponds to the historical information of transmitter i in the t-th time slot. μ i (t-1) represents the energy indicator for energy capture during the transmission phase of transmitter i in the (t-1)th time slot, λ i (t-1) represents the transmission indicator corresponding to the update packet sent by transmitter i in the (t-1)th time slot, R i (t) represents the cumulative received signal strength of transmitter i in the t-th time slot. Where N represents the individual dimension of the transmitter, ξ represents the time dimension, and |s i (t)| represents the dimension of the specific content of the historical information provided by the secondary transmitter i in time slot t. The perception matrix S(t) constructed in time slot t reflects the spatial and temporal characteristics of the state of the primary user network.

[0079] Furthermore, a specific calculation method for the actual conflict time ratio is given. The actual conflict time ratio for each time slice is divided into two cases based on the main user's activity status at the end of the control phase: and At the end of the control phase in the t-th time slot, for the case... τ I (t) represents the duration the primary user has been idle, for the following situation. τ B (t) represents the time the main user has been continuously busy.

[0080] Regarding the situation The actual conflict time ratio of the h-th time slice in the t-th time slot Given by the following formula:

[0081]

[0082] Where, τ t This indicates the duration of the transmission phase, where x represents the duration of the primary user's idle or busy state. as well as

[0083] Regarding the situation The actual conflict time ratio of the h-th time slice in the t-th time slot Given by the following formula:

[0084]

[0085] in, as well as The actual conflict time ratio of each time slice can be calculated using the above formula.

[0086] Furthermore, the network model constructed in this embodiment is used for simulation modeling and the perception matrix S(t) is generated and saved as the input of the constructed dataset to the GC-2LSTM-NN model. The numerical integration is performed according to the above time-slice conflict time ratio formula, and the actual conflict time ratio vector y' of the H time slices contained in the t-th time slot is saved as the label of the constructed dataset.

[0087] (2) The GC-2LSTM-NN model learns and infers from the input data. The specific structure of the GC-2LSTM-NN model is as follows: Figure 1 As shown. First, the GC-2LSTM-NN model splits the input data's N×F×ξ perceptual matrix according to the time dimension, obtaining ξ N×F matrices as the graph signal feature matrix for the t-th time slot. Where F represents the number of specific historical information features of the sub-transmitter in each time slot. An N×N adjacency matrix is ​​calculated based on the physical two-dimensional coordinates of the sub-transmitter and the radial basis function. Specifically, its matrix components The calculation is as follows:

[0088]

[0089] Where, d i Let d represent the physical two-dimensional coordinates of transmitter i. j Let represent the physical two-dimensional coordinates of the secondary transmitter j, and ρ be a preset hyperparameter. The purpose of this adjacency matrix calculation method is to fully utilize the prior knowledge that secondary transmitters with close physical distances have more similar physical wireless communication environments. It should be noted that this invention only calculates the adjacency matrix once. Based on the adjacency matrix... With ξ graphical signal feature matrices X t Construct ξ undirected graphs using conventional mathematical steps for graph modeling. Where V and E represent the nodes formed by the secondary transmitter and can be represented by the adjacency matrix, respectively. The feature matrix X of the graph signal t The described edge relationships.

[0090] Next, consider ξ undirected graphs. The input is fed into the graph convolutional layer (denoted as GCN) of the GC-2LSTM-NN model for computation to obtain ξ spatially augmented graph feature matrices. It is important to note that the GC-2LSTM-NN model uses spectral graph convolution as the computation method for the GCN layer. Specifically, the computation expression for the graph convolution layer in the GC-2LSTM-NN model is as follows:

[0091]

[0092] Where ReLU represents the activation function, and its calculation expression is: The adjacency matrix that includes self-connection relationships between nodes, i.e. I represents the identity matrix. express The angle matrix, diagonal components of the matrix in for The matrix components, W GCN These are the learnable weights for the GCN layer. The specific process of the spatial feature fusion part is as follows: Figure 2 As shown.

[0093] Furthermore, the internal structure of LSTM is explained. LSTM consists of three basic components: the input gate, the forget gate, and the output gate. The expressions for each gate and their explanations are as follows:

[0094] 1. The input gate controls how the cell state is updated in order to store new information; it is represented as I. t The Sigmoid activation function is used to generate an output in the range [0,1] to control the weights assigned to the input data. The expression for the input gate is as follows:

[0095] I t =σ(W I ·[h t-1 ,x t ]+b I )

[0096] Where σ() represents the Sigmoid activation function, h t-1 Let x represent the hidden state of time slot t-1. t This represents the input data for time slot t. W I and b I These correspond to the learnable weights and biases of the input gate, respectively.

[0097] 2. The forgetting gate controls the retention or discarding of information in a cell's state; it is represented by F. t Using the Sigmoid activation function, the expression for the forget gate is as follows:

[0098] Ft =σ(W F ·[h t-1 ,x t ]+b F )

[0099] Among them, W F and b F These represent the learnable weights and biases of the forget gate, respectively.

[0100] 3. The output gate controls the output of the cell state and the update of the hidden state. The output gate is represented as O. t The Sigmoid activation function is used to generate output values ​​in the range [0,1], and the Tanh activation function is used to determine the values ​​to be propagated to the current cell state in the (t+1)th time slot. The expressions for the output gate controlling the output, cell state, and hidden state are as follows:

[0101] O t =σ(W O ·[h t-1 ,x t ]+b O )

[0102] C t =f t ⊙C t-1 +I t ⊙Tanh(W C ·[h t-1 ,x t ]+b C )

[0103] h t =O t ⊙Tanh(C t )

[0104] Among them, W O and W c b represents the learnable weights of the output gate and the cell state, respectively. O and b C represents the bias of the output gate and the cell state, respectively, and ⊙ represents element-wise multiplication.

[0105] Furthermore, the proposed dual-layer LSTM structure is explained below. The dual-layer LSTM structure is as follows: The first layer consists of N individual LSTMs, namely LSTM-1, LSTM-2, ..., LSTM-N, which learn and enhance the independent temporal features of each transmitter. The second layer consists of a swarm LSTM and a linear layer. The swarm LSTM stacks the temporal features learned and enhanced by the individual LSTMs along the feature dimension, then learns and enhances the global temporal features, and finally passes them to the linear layer for feature fusion to output a predicted conflict time ratio vector. The linear layer is a single-layer perceptron. It is important to note that the N individual LSTMs in the GC-2LSTM-NN model do not share weights.

[0106] Furthermore, the specific process of learning and fusing temporal features of data using the dual-layer LSTM proposed in this invention is explained. The specific process of learning and fusing temporal features of data using the dual-layer LSTM is as follows: [The process involves processing the obtained ξ spatial augmentation map feature matrices...] Stacking along the time dimension yields the spatiotemporal feature matrix f. st The stacking operation (represented by Concatenate) refers to concatenating multiple matrices along a specified dimension to form a new multidimensional matrix. In the current step, this involves the spatiotemporal feature matrix. The ξ eigenvectors in the i-th row along the time dimension are stacked in order, i.e. Where F E This indicates the number of enhanced spatial features, representing the feature dimension.

[0107] Next, extract the aforementioned spatiotemporal feature matrix f. st The N individual temporal characteristics of the same transmitter in the time dimension are as follows: The temporal features of N individuals are input into their respective individual LSTMs; after the N individual LSTMs learn and enhance the temporal features, a total of N enhanced individual temporal feature results are output, i.e. Among them, F T This represents the enhanced temporal feature dimension.

[0108] Next, the proposed GC-2LSTM-NN model stacks the temporal feature results of N augmented individuals along the temporal feature dimension. The overall feature after stacking is the temporal feature F of the N sub-transmitters. T Global temporal features composed of joint components The * operator represents scalar multiplication. The input is fed into a population LSTM (represented as CooperativeLSTM), where it learns and is augmented to obtain the final augmented global temporal feature result. Right now Finally, the GC-2LSTM-NN model will Perform the flattening operation, that is, order Flattened into a one-dimensional feature vector, it is input into a linear layer for fusion, and finally combined with the Sigmoid function to output the predicted conflict time ratio vector y of the H time slices of the predicted time slot t, that is:

[0109] Where Flatten is the flattening operation, Linear is the linear layer, and σ is the Sigmoid function, which can be specifically expressed as: The time feature fusion part is as follows Figure 3 As shown.

[0110] (3) The proposed GC-2LSTM-NN model is trained using the backpropagation algorithm. In this embodiment of the invention, the mean square error (MSE) is used as the loss function to calculate the loss. Specifically, the predicted conflict time ratio vector y obtained by the GC-2LSTM-NN model for a certain time slot of the input perception matrix S(t) is compared with the actual conflict time ratio vector y' corresponding to the perception matrix S(t) in the dataset using the mean square error calculation to obtain the training loss (denoted as Loss), which is expressed as follows:

[0111]

[0112] Among them, y h y' represents the predicted conflict time ratio for the h-th time slice. h This represents the actual conflict time ratio of the h-th time slice.

[0113] Subsequently, this embodiment of the invention utilizes the backpropagation algorithm to differentiate the parameters in the GC-2LSTM-NN model and executes the gradient descent algorithm to update the existing parameters in the GC-2LSTM-NN model in order to minimize the loss, where the parameters are represented by θ. The specific parameter update formula in the gradient descent algorithm is: Where α is the preset initial hyperparameter learning rate.

[0114] Furthermore, this embodiment of the invention employs Root Mean Square Propagation (RMSProp) to dynamically adjust the learning rate. In each iteration, the RMSProp algorithm dynamically adjusts the learning rate based on historical gradient information. Its purpose is to ensure that for parameters with larger gradients, the learning rate is smaller, and for parameters with smaller gradients, the learning rate is larger. RMSProp allows the GC-2LSTM-NN model to better adapt to the update requirements of different parameters during training. Specifically, RMSProp uses an exponentially weighted average to calculate the moving average of the squared gradient for each parameter's gradient. Calculate the moving average of the gradient Here, β is a decay factor used to control the weight of the influence of historical gradients on the current gradient, and its value is generally in the range [0,1]. Next, the learning rate is adjusted. The training parameter is used, where ε is a small constant to avoid division by zero. The parameters are then updated using the adjusted learning rate.

[0115] Furthermore, four models—CNN, LSTM, GCN, and CNN-LSTM—were applied to the task described in this embodiment of the invention, maintaining the same simulation environment parameter settings. They were then evaluated and compared with the GC-2LSTM-NN model proposed in this invention. Figure 4 As shown, the proposed GC-2LSTM-NN model has a lower mean squared error, which improves prediction accuracy and is more advantageous than the other models mentioned above.

[0116] The above embodiments may appear in whole or in part as a computer program product, which includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated.

[0117] This invention focuses on providing a novel end-to-end deep learning model. The proposed GC-2LSTM-NN utilizes GCN to enhance the representation of spatial features and employs a two-layer LSTM to learn time-series features from individual to global perspectives, improving the learning and fusion capabilities of multi-dimensional spatiotemporal variation features in non-Euclidean spatial data. The GC-2LSTM-NN model improves the accuracy of conflict time ratio prediction, providing reliable information for subsequent scheduling tasks. Furthermore, the GC-2LSTM-NN model offers new ideas and methods for further research in the field of spectrum state prediction technology within cognitive radio.

[0118] The embodiments described above are merely illustrative of one or more implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the appended claims.

Claims

1. A method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model, characterized in that, The spectral collision time ratio prediction method based on the GC-2LSTM-NN model includes: Construct an asynchronous cognitive radio network, which includes a main network and a secondary network. The main network contains a primary user and a licensed channel, and the secondary network includes N single-antenna secondary transmitters and N antenna secondary receivers. The transmission phase of the secondary network is divided into H equal-length time slices. Spatiotemporal data of the secondary transmitter is obtained from an asynchronous cognitive radio network, the spatiotemporal data including an adjacency matrix with the secondary transmitter as a node. And the sensing matrix consisting of historical information from the secondary transmitter received by the secondary receiver within a predetermined time period; A GC-2LSTM-NN model is constructed. Historical information of the secondary transmitters in the past ξ time slots is extracted from the perception matrix of time slot t to obtain the extracted perception matrix. The true collision time ratio vector y' of the H time slices contained in time slot t is calculated as the label of the extracted perception matrix. Based on the extracted perception matrix and the adjacency matrix... Construct an independent undirected graph for each time slot Given ξ undirected graphs The graph convolutional layer in the GC-2LSTM-NN model is fed into the graph to compute ξ spatially augmented graph feature matrices. Based on ξ spatial enhancement map feature matrices Obtain the spatiotemporal characteristic matrix f st Extract the spatiotemporal feature matrix f st The time-series characteristics of N individual entities corresponding to each sub-transmitter in the time dimension; The temporal features of N individuals are input into the first layer of the N individual LSTM in the two-layer LSTM, and the N enhanced individual temporal feature results are output. The N enhanced individual temporal feature results are stacked on the temporal feature dimension to obtain global temporal features. The global temporal features are then input into the second layer of the population LSTM in the two-layer LSTM to obtain the enhanced global temporal feature results. The enhanced global temporal feature results are flattened into a one-dimensional feature vector and input into a linear layer in a two-layer LSTM for fusion. The Sigmoid function is combined to output the predicted conflict time ratio vector y of the H time slices in the t-th time slot. The mean squared error between the actual conflict time ratio vector y' and the predicted conflict time ratio vector y is calculated as the loss function, and the GC-2LSTM-NN model is trained using the backpropagation algorithm. The trained GC-2LSTM-NN model is used to predict the conflict time ratio vector of H time slices in each time slot.

2. The method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model according to claim 1, characterized in that, The adjacency matrix The adjacency matrix is ​​calculated based on the physical two-dimensional coordinates and radial basis functions of the secondary transmitter. matrix components The calculation formula is as follows: Where, d i Let d represent the physical two-dimensional coordinates of transmitter i, i∈[1,N], and d j Let ρ represent the physical two-dimensional coordinates of the transmitter j, j∈[1,N], and ρ represent the preset hyperparameters.

3. The method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model according to claim 1, characterized in that, The process of extracting historical information of the sub-transmitters from the past ξ time slots in the sensing matrix of the t-th time slot to obtain the extracted sensing matrix includes: Each time slot includes a control phase and a transmission phase. The control phase is divided into a sensing sub-phase and a reporting sub-phase. In the sensing sub-phase, each sub-transmitter independently performs sensing, letting y i (k) represents the kth sample of the main network signal received by the secondary transmitter i, expressed by the following formula: Among them, w i (k) represents the noise of the k-th sample of the main network signal received by the secondary transmitter i, v i (k) represents the k-th sample of the main network signal received by the secondary transmitter i, which is the main transmitter signal. and These represent two scenarios: one where the main network is not using spectrum resources and the other where it is using spectrum resources. The specific sensing data is the cumulative received signal strength of the secondary transmitter, expressed by the formula: Among them, R i The cumulative received signal strength of transmitter i is represented by K, and K represents the number of samples of the main network signal received by transmitter i. During the reporting sub-phase, each sub-transmitter reports the cumulative received signal strength and remaining energy to the sub-receiver. If the remaining energy of a sub-transmitter is less than the preset energy, it remains silent during the reporting sub-phase. Define the energy indicator for energy capture during the transmission phase as follows: Where, μ i (t) represents the energy indicator for energy capture during the transmission phase of transmitter i in time slot t. This represents the remaining energy of transmitter i at the end of the control phase in the t-th time slot. This represents the remaining energy of transmitter i at the start of the control phase in the (t+1)th time slot, when μ i When (t) = 1, it indicates that transmitter i was charged during the transmission phase of the t-th time slot. i When (t) = -1, it indicates that transmitter i is inactive during the transmission phase of the t-th time slot. i When (t) = 0, it means that transmitter i transmitted an update packet during the transmission phase of the t-th time slot; Let λ i (t) indicates that in the t-th time slot, the transmission indicator corresponding to the update packet sent by transmitter i is: Where, when λ i (t) = 1 indicates that the secondary transmitter i successfully transmitted the update packet to the secondary receiver in the t-th time slot without a collision. i (t) = -1 indicates that the update packet transmission failed, when λ i (t) = 0 indicates other cases; The extracted perception matrix S(t) is expressed by the following formula: Among them, s i (t)=(μ i (t-1),λ i (t-1),R i (t) represents the historical information of the sub-transmitter i provided in time slot t, where i∈[1,N], μ i (t-1) represents the energy indicator for energy capture during the transmission phase of transmitter i in the (t-1)th time slot, λ i (t-1) represents the transmission indicator corresponding to the update packet sent by transmitter i in the (t-1)th time slot, R i (t) represents the cumulative received signal strength of transmitter i in the t-th time slot. Where N represents the number of individual transmitters, ξ represents the time dimension, and |s i (t)| represents the dimension of the specific content of the historical information provided by the sub-transmitter i in time slot t.

4. The method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model according to claim 3, characterized in that, The calculation of the true conflict time ratio vector y' of the H time slices contained in the t-th time slot includes: Regarding the situation The actual conflict time ratio of the h-th time slice in the t-th time slot Given by the following formula: Where, τ t τ represents the duration of the transmission phase. I (t) represents the time the main user has been idle, x represents the duration of the main user's idle or busy state, and z represents the differential variable. Regarding the situation The actual conflict time ratio of the h-th time slice in the t-th time slot Given by the following formula: in, as well as τ B (t) represents the time the main user has been continuously busy; The actual conflict time ratios of H time slices are combined to obtain the actual conflict time ratio vector y' of the t-th time slot.

5. The method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model according to claim 3, characterized in that, The extracted perception matrix and adjacency matrix are used as a basis for further processing. Construct an independent undirected graph for each time slot This includes: Let the shape of all components in the same column of the extracted perception matrix be N×|s i The graphical signal feature matrix X of (t)| t The feature matrix X of the graph signal t Combined with adjacency matrix Model as an undirected graph in spatial dimensions 6. The method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model according to claim 5, characterized in that, The ξ undirected graphs The graph convolutional layer in the GC-2LSTM-NN model is fed into the graph to compute ξ spatially augmented graph feature matrices. The calculation formula is as follows: Where ReLU represents the activation function, and its calculation expression is: The adjacency matrix representing the self-connection relationships of the nodes is... I represents the identity matrix. express The angle matrix, diagonal components of the matrix express The matrix components, W GCN These are the learnable weights for the graph convolutional layer.

7. The method for predicting the spectral collision time ratio based on the GC-2LSTM-NN model according to claim 1, characterized in that, The feature matrices based on ξ spatial enhancement maps Obtain the spatiotemporal characteristic matrix f st ,include: ξ spatial enhancement map feature matrices Sort and stack them according to the time dimension to obtain the spatiotemporal feature matrix f st The formula is as follows: Where Concatenate represents a stacking operation, F E This indicates the number of enhanced spatial features, representing the feature dimension.

Images