A task-driven dynamic allocation method for time-frequency resources

Abstract

This invention discloses a task-driven dynamic allocation method for time-frequency resources, comprising: constructing a hierarchical clustered network; wherein the hierarchical clustered network consists of M subnets, each subnet containing N nodes: 1 network management node, K gateway nodes, and N-1-K ordinary nodes; each node has a unique identity identifier; data is exchanged between subnets through gateway nodes; the network management node is used for time-frequency resource allocation; the current node type is determined based on the node identity identifier; when the current node is determined to be a network management node, the dynamic allocation module for time-frequency resources is invoked to allocate time-frequency resources. The method described in this invention constructs a new dynamic network resource allocation scheme for future combat scenarios, resolving the contradiction between differentiated service requirements in a highly dynamic, self-organizing battlefield environment and the limited multi-dimensional resources such as communication, computing, and sensing in military networks, achieving adaptive matching between multi-dimensional network resources and various service requirements.

Description

Technical Field

[0001] This invention belongs to the field of wireless network and communication technology, and particularly relates to a task-driven dynamic allocation method for time-frequency resources. Background Technology

[0002] In the future integrated and information-based communication scenario, networks will possess highly dynamic, highly adversarial, and self-organizing characteristics. In this scenario, weapons and platforms will no longer be independent combat entities; various platforms and battlefield elements will need to form an organic whole through communication networks to jointly execute and complete various types of tasks.

[0003] In a certain communication scenario, to ensure manageable and controllable transmission processes, especially transmission latency, foreign countries pioneered the use of TDMA to support multi-user network access and data transmission in wireless channels. A node's acquisition of network transmission resources depends on the transmission time slots it can be allocated during transmission; optimized configuration of network transmission resources is equivalent to dynamic optimization of transmission time slot allocation. In a certain military tactical data link Link11, the network adopts a master-slave mode, with nodes using a polling method to obtain available transmission time slots, resulting in low efficiency in transmission resource utilization. While Link16 allows for flexible multi-hop networking between nodes, the allocation of time slot resources between nodes often uses a relatively fixed allocation method, making it difficult to adaptively and flexibly adjust according to changes in service requirements.

[0004] To meet the QoS service requirements of differentiated transmission services, it is essential to dynamically, optimally, and on-demand configure and manage time slot resources. Currently, time slot allocation algorithms can be broadly categorized into three types: fixed time slot allocation algorithms, dynamic time slot allocation algorithms, and hybrid time slot allocation algorithms combining fixed and dynamic methods. Among these, based on the implementation method, TDMA protocols based on dynamic allocation algorithms can be further divided into centralized and distributed types; distributed dynamic TDMA protocols can also be further classified into topology-dependent and topology-transparent types based on whether topology information is required during time slot allocation.

[0005] Chenxi Zhu and M. Scott Corson published the FPRP protocol for TDMA ad hoc networks, which simultaneously solves the channel access and broadcast scheduling problems for nodes. Nodes guarantee conflict-free broadcast time slot allocation through a contentious five-step reservation mechanism. Rhee et al. proposed the Distributed Random Time Slot Scheduling Algorithm (DRAND), which employs a completely random coin-tossing time slot reservation algorithm to limit the probability of each node issuing a reservation request in each reserved time slot, thereby reducing the probability to some extent that nodes within a two-hop neighbor range simultaneously issue requests in the same reserved time slot.

[0006] In highly dynamic, self-organizing battlefield environments, there are numerous contradictions between the differentiated service requirements and the limited multidimensional resources (communication, computing, sensing, etc.) in military networks. How to achieve adaptive matching between these multidimensional network resources and diverse service requirements is one of the problems that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide a task-driven dynamic allocation method for time and frequency resources. It constructs a new dynamic allocation scheme for network resources in future combat scenarios, aiming to solve the contradiction between the differentiated service requirements in a highly dynamic and self-organized battlefield environment and the limited multi-dimensional resources such as communication, computing, and sensing in military networks. It achieves adaptive matching between multi-dimensional network resources and various service requirements, and has the advantages of rapid adjustment and efficient transmission.

[0008] To address the aforementioned technical problems, this invention discloses a task-driven dynamic allocation method for time-frequency resources, comprising:

[0009] Construct a hierarchical clustered network; wherein the hierarchical clustered network consists of M subnets, each subnet containing N nodes: 1 network management node, K gateway nodes and N-1-K ordinary nodes; each node has a unique identity; the subnets exchange data through the gateway nodes; the network management node is used for time and frequency resource allocation; M≥1, K≥1, N≥3;

[0010] Determine the current node type based on the node's identity identifier;

[0011] When the current node is determined to be a network management node, the time and frequency resource dynamic allocation module is invoked to allocate time and frequency resources.

[0012] In the above task-driven dynamic allocation method for time and frequency resources, the dynamic allocation module for time and frequency resources includes: an LSTM prediction module, a time slot resource allocation module, a synaptic computing resource management module, and a frequency hopping pattern allocation module.

[0013] In the above task-driven dynamic allocation method for time-frequency resources, when the current node is determined to be a network management node, the dynamic allocation module for time-frequency resources is invoked to allocate time-frequency resources, including:

[0014] When the current node is determined to be a network management node, the sensor computing resource management module is invoked to collect the local link traffic characteristics of other nodes as historical service traffic; among them, the historical service traffic characteristics are the characteristics of services that have been transmitted in the past.

[0015] The LSTM prediction module is invoked to predict the local link traffic characteristics in the next period based on the local link traffic characteristics of other nodes collected by the synaptic computing resource management module, and the prediction results are obtained.

[0016] The time slot resource allocation module is invoked to allocate time slot resources within the subnet based on the prediction results output by the LSTM prediction module and the link construction requirement information issued by the network layer. The link construction requirement information issued by the network layer is the characteristics of the services to be transmitted in the next cycle, i.e., future service traffic.

[0017] The frequency hopping pattern allocation module is invoked to allocate frequency hopping patterns within the subnet based on the spectrum interference results of all ordinary nodes in the entire network obtained from the network management node.

[0018] In the above task-driven dynamic allocation method for time-frequency resources, both historical and future service traffic include: source node, destination node, service priority, and service volume; the format is: <source node, destination node, service priority, service volume>.

[0019] In the above-mentioned task-driven dynamic allocation method for time and frequency resources, the flight process consists of multiple task phases, each of which is considered as a cycle; the task phases include: navigation phase, collision avoidance phase, enemy attack phase, and detection phase.

[0020] In the above task-driven dynamic allocation method for time-frequency resources, the LSTM prediction module, when predicting the local link traffic characteristics for the next period, includes:

[0021] The historical traffic of link e covered by slice s at time t is represented as {traffic e,s (t)};for {traffic es Normalize (t)} to obtain the normalized result {traffic es (t)} tr :

[0022]

[0023]

[0024] Where t = 1, 2, ..., h, h represents the time length of the current period; It means {traffic e,s The average value of (t)};

[0025] Then, time window T window The normalized result of future traffic within the length is represented as {traffic e,s (j)} tr ,have:

[0026] {traffic e,s (j)} tr={traffic e,s (j), traffic e,s (j+T window ),...,traffic e,s (j+D·T window )} tr

[0027] Where j = 1, 2, ..., T window ;

[0028] {traffic e,s (j)} tr The first D steps are used as the input values ​​for the LSTM neural network, and the (D+1)th step is used as the comparison value for the minimum mean square error, dividing the data into T... window Yes, we obtained the training data;

[0029] The LSTM neural network is trained based on the training data to obtain the prediction model;

[0030] traffic e,s The data from step D after (j) is used as input to the prediction model, and the prediction model outputs the predicted value traffic′. e,s (j+(D+1)·T window ); Take time window T window The predicted values ​​within the length are used to obtain the predicted sequence {traffic′}. e,s (1+(D+1)·T window ),...,traffic′ e,s (T window +(D+1)·T window )};

[0031] The predicted sequence is denormalized to obtain the predicted business traffic for the next period.

[0032] Repeat the above steps to obtain the predicted service traffic for all links in the network in the next cycle.

[0033] In the above task-driven dynamic allocation method for time-frequency resources, the loss function Loss of the LSTM neural network is:

[0034]

[0035] Among them, o j This represents the output value of the j-th pair of training data in the LSTM neural network.

[0036] In the above task-driven dynamic allocation method for time-frequency resources, the time slot resource allocation module, when allocating time slot resources within the subnet, includes: filling in the micro-time slots required for inter-node service transmission within a frame based on the prediction results output by the LSTM prediction module and the link construction requirement information.

[0037] In the above task-driven dynamic allocation method for time-frequency resources, the frequency hopping pattern allocation module, when performing frequency hopping pattern allocation within a subnet, includes:

[0038] According to the 9th level irreducible polynomial x 9 +x 4 +1 generates an m-sequence f_p, and mod(F_num) is performed on the m-sequence f_p; where mod means taking the remainder, and F_num means the total number of frequency points;

[0039] Narrowband determination is performed on the frequency points used by N nodes at any given time.

[0040] If |f_(p+1)-f_p|<15, then f_(p+1)=[f_(p+1)+30]mod(500), which gives the frequency hopping pattern of N nodes in the subnet; where p=1,2,…,N.

[0041] The above task-driven dynamic allocation method for time-frequency resources also includes: calling the synesthetic computing resource management module to perform resource management.

[0042] The node service transmission demand is sorted from smallest to largest, and the nodes are divided into four levels of busyness according to the sorting size: the first level is the first N / 4 nodes, the second level is the nodes from N / 4 to N / 2 nodes, the third level is the nodes from N / 2 to 3N / 4 nodes, and the fourth level is the nodes from 3N / 4 to N nodes.

[0043] Initialization matrix G = adjacency matrix [x p,q ], where x p,q =1 indicates that node p is connected to node q, x p,q =0 indicates that node p and node q are not connected; q≠p, q=1,2,…,N;

[0044] Initialize all nodes, let x p =0, n p =0; where x p =1 indicates that node i is selected as the sensing node, x p =0 indicates that the node was not selected; n p This indicates that sensing node i will be separated by n. p Each frame transmits sensing data once;

[0045] Perform the following operations on the nodes of the first, second, third, and fourth tiers in sequence:

[0046] a) For nodes within the same gear range, based on the degree D i Sort the nodes from largest to smallest, prioritizing the node with the largest degree.

[0047] b) If the current node is a regular node, select the current node as the sensing node to collect characteristics of sudden business traffic, i.e., update the current node's x. i =1; then update matrix G: set the row and column of the current node to 0;

[0048] c) Determine if all elements in G are 0. If yes, end; otherwise, continue to perform operations a) to b) on other nodes in the same gear.

[0049] d) If all nodes in the same gear have been traversed, jump to the next gear and perform the same operation until all elements in G are 0; where, if the selected sensing node is a node in the first or second gear, the sensing data transmission frequency is once every 2 frames, that is, update the node's n. p =2; If the selected sensing node is within the third or fourth tier, the sensing data transmission frequency is once every 4 frames, that is, the n of the updated node is updated. p =4;

[0050] f) Output all selected sensing nodes and the transmission frequency of the sensing data of each selected sensing node, i.e., output x. p and n p .

[0051] The present invention has the following advantages:

[0052] (1) This invention discloses a task-driven dynamic allocation method for time and frequency resources, proposes a clustered hierarchical network topology architecture, uses network management nodes to manage the entire network within a cluster, and communicates between clusters through network management nodes, thereby realizing a large-scale self-organizing network for aircraft.

[0053] (2) This invention discloses a task-driven dynamic allocation method for time and frequency resources. It adopts an elastic network protocol system, and the transmission side is divided into physical layer, link layer and network layer. It can dynamically allocate time and frequency resources according to task needs, realize intelligent adaptation to tasks, and at the same time reduce communication latency and improve communication efficiency. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the architecture of a hierarchical clustering network in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of an overall time-frequency dynamic allocation scheme in an embodiment of the present invention;

[0056] Figure 3This is a schematic diagram of an LSTM neural network structure in an embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of a time slot structure in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments disclosed in the present invention will be described in further detail below with reference to the accompanying drawings.

[0059] To address the task-driven adjustment of time-frequency resources in large-scale wireless networking scenarios, this invention proposes a task-driven dynamic allocation method for time-frequency resources, comprising:

[0060] Step 1: Construct a hierarchical and clustered network.

[0061] In this embodiment, the network architecture of the hierarchical cluster network is as follows: Figure 1 As shown, the hierarchical clustered network consists of M subnets, each containing N nodes: 1 management node, K gateway nodes, and N-1-K ordinary nodes. Each node has a unique identifier. Data exchange between subnets occurs through the gateway nodes; the management node manages the resources of nodes within a subnet and allocates time-frequency resources to these nodes. Where M≥1, K≥1, and N≥3.

[0062] Step 2: Determine the current node type based on the node's identity identifier.

[0063] In this embodiment, the first step is to determine the type of the current node, i.e., whether it is a gateway node, a network management node, or a regular node. Different nodes are responsible for different functional modules; only network management nodes are responsible for time-frequency resource allocation within the subnet, and only gateway nodes can communicate with other subnet nodes.

[0064] Step 3: When the current node is determined to be a network management node, the time and frequency resource dynamic allocation module is invoked to allocate time and frequency resources.

[0065] In this embodiment, when the current node is determined to be a network management node, the time-frequency resource dynamic allocation module is invoked to allocate time-frequency resources, such as... Figure 2 As shown, the dynamic time-frequency resource allocation module includes: an LSTM prediction module, a time slot resource allocation module, a sensory computing resource management module, and a frequency hopping pattern allocation module. Among these, the LSTM prediction module, time slot resource allocation module, sensory computing resource management module, and frequency hopping pattern allocation module are located at the link layer, enabling interaction with the network layer and physical layer.

[0066] Since there are multiple mission phases during flight, such as navigation, collision avoidance, enemy attack, and detection phases, each mission phase is considered as a cycle (note that the duration of each cycle is not fixed and unpredictable). Based on the link construction requirements received for the next cycle, historical service traffic, and the sensing data transmission requirements provided by the sensing resource management module, the network management node needs to plan the time-frequency resource allocation for the next cycle within its subnet and generate a time-frequency resource allocation table for the service channels within the subnet.

[0067] The LSTM prediction module is used for point-to-point burst traffic prediction in the next cycle.

[0068] The time slot resource allocation module allocates time slot resources for the transmission channels within the subnet based on the link construction requirements for the next cycle and the point-to-point burst traffic predicted by the LSTM module.

[0069] The frequency hopping pattern allocation module collects the spectrum results sensed by each node, and the network management node reallocates the frequency hopping patterns to obtain the frequency hopping pattern set of all nodes in the subnet. Each node corresponds to a unique frequency hopping pattern, and the frequency hopping patterns are as orthogonal as possible to ensure maximum transmission efficiency.

[0070] The sensory resource management module is used to collect burst traffic characteristic data in the next task phase, serving as input for the LSTM prediction module in that phase. During the task adjustment phase, the network management node will, based on the point-to-point service transmission requirements of the next phase, network topology, available computing resources, and node location information, perform the following: 1) Selection of sensing nodes, i.e., which nodes in the subnet collect burst traffic characteristic data for the next phase and transmit it to the network management node; 2) Transmission frequency of sensing data, i.e., how many frames every sensing node needs to transmit the currently collected burst traffic characteristic data; 3) Sensing data computation indication, i.e., whether the node needs to preprocess the collected sensing data before transmitting it to the network management node. Furthermore, the sensory resource management module comprises two parts: a sensing module and a computing module. The sensing module determines the selection of sensing nodes and the data transmission frequency, while the computing module, based on the results of the sensing module, available computing resources, and node transmission capabilities, decides whether the sensing data needs local processing. Furthermore, the synaptic resource management module will collect and plan the characteristic data of burst traffic in the next task phase based on the point-to-point transmission requirements of the next phase, and generate the data transmission requirements for the next phase, which will be passed to the time slot resource allocation module. The time slot resource allocation module will then dynamically allocate time slot resources based on the transmission requirements of the next phase and the point-to-point burst traffic information predicted by the LSTM module, forming a time slot table for the new phase.

[0071] In addition, the process ends when it is determined that the current node is not a network management node.

[0072] In this embodiment, when the current node is determined to be a network management node, the specific process of calling the time-frequency resource dynamic allocation module to allocate time-frequency resources is as follows:

[0073] Sub-step 31: When the current node is determined to be a network management node, the sensory computing resource management module is invoked to collect the local link traffic characteristics of other nodes as historical service traffic. Historical service traffic characteristics refer to the characteristics of services transmitted in the past, including: source node, destination node, service priority, and service volume, etc., typically in the format: <source node, destination node, service priority, service volume>.

[0074] Sub-step 32: Call the LSTM prediction module to predict the local link traffic characteristics in the next period based on the local link traffic characteristics of other nodes collected by the synaptic computing resource management module, and obtain the prediction result.

[0075] In this embodiment, the prediction process of the LSTM prediction module is as follows:

[0076] First, data processing is performed. Data processing refers to the process of normalization to transform the input data (business traffic) into a series of values ​​acceptable to the LSTM neural network. The LSTM neural network model is as follows: Figure 3 As shown. The specific process of normalization is as follows:

[0077] The historical traffic of link e covered by slice s at time t is represented as {traffic e,s (t)};for {traffic e,s Normalize (t)} to obtain the normalized result {traffic e,s (t)} tr :

[0078]

[0079]

[0080] Where t = 1, 2, ..., h, h represents the time length of the current period; It means {traffic e,s The average value of (t)}.

[0081] Furthermore, due to the time window T window Future business traffic within the length and its D·T window The normalized result of the future traffic of link e covered by slice s can be expressed as {traffic}, which is related to the previous D-step traffic within the length. e,s (j)} trThen we have:

[0082] {traffic e,s (j)} tr ={traffic e,s (j), traffic e,s (j+T window ),...,traffic e,s (j+D·T window )} tr

[0083] Where j = 1, 2, ..., T window .

[0084] Furthermore, {traffic e,s (j)} tr The first D steps are used as the input values ​​for the LSTM neural network, and the (D+1)th step is used as the comparison value for the minimum mean square error. This divides the data into T... window Yes, this type of data together constitutes a batch of training data.

[0085] Based on the training data, an LSTM neural network can be trained to obtain a prediction model. The loss function of the LSTM neural network is:

[0086]

[0087] Among them, o j This represents the output value of the j-th pair of training data in the LSTM neural network.

[0088] traffic e,s The data from step D after (j) is used as input to the prediction model, and the prediction model outputs the predicted value traffic′. e,s (j+(D+1)·T window ); Take time window T window The predicted values ​​within the length are used to obtain the predicted sequence {traffic′}. e,s (1+(D+1)·T window ),...,traffic′ e,s (T window +(D+1)·T window )}.

[0089] Furthermore, the predicted sequence is denormalized to obtain the predicted business traffic for the next period.

[0090] Finally, by repeating the above steps, the predicted service traffic for all links in the network in the next cycle can be obtained.

[0091] Sub-step 33: Call the time slot resource allocation module to allocate time slot resources within the subnet based on the prediction results output by the LSTM prediction module and the link construction requirement information issued by the network layer.

[0092] In this embodiment, the link construction requirement information issued by the network layer is the characteristics of the service to be transmitted in the next cycle, i.e., the future service traffic, including information such as source node, destination node, service priority and service volume, and the format is generally: <source node, destination node, service priority, service volume>.

[0093] Preferably, the specific time slot resource allocation process of the time slot resource allocation module is as follows: 1) Based on the x output by the synesthetic computing resource management module p and n p The local link traffic characteristics of the corresponding node are sent to the network management node; the prediction results output by the LSTM prediction module and the link construction requirement information are input into the time slot resource allocation module, according to... Figure 4 The time slot structure shown fills the micro-time slots required for inter-node service transmission within a frame. The specific number of micro-time slots in a frame is determined by the specific usage.

[0094] Sub-step 34: Call the frequency hopping pattern allocation module to perform frequency hopping pattern allocation within the sub-network based on the spectrum interference results of all ordinary nodes in the entire network obtained from the network management node summary.

[0095] In this embodiment, the specific frequency hopping pattern allocation process of the frequency hopping pattern allocation module is as follows:

[0096] According to the 9th level irreducible polynomial x 9 +x 4 +1 generates an m-sequence f_p, and then mod(F_num) is performed on the m-sequence f_p; where mod represents the remainder, and F_num represents the total number of frequency points. Here, p = 1, 2, ..., N.

[0097] For any N nodes at any given time, narrowband judgment is performed on the frequency points used. If |f_(p+1)-f_p|<15, then f_(p+1)=[f_(p+1)+30]mod(500), and the frequency hopping pattern of N nodes in the subnet is obtained.

[0098] In this embodiment, resource management can also be performed by calling the synesthetic computing resource management module:

[0099] a) Sort the node service transmission demand from smallest to largest, and divide the node busyness into 4 levels according to the sort size: the first level is the first N / 4 nodes (including N / 4), the second level is the nodes from N / 4 to N / 2 (including N / 2), the third level is the nodes from N / 2 to 3N / 4 (including 3N / 4), and the fourth level is the nodes from 3N / 4 to N (including N).

[0100] b) Initialize matrix G = adjacency matrix [x p,q ]. Where, x p,q =1 indicates that node p is connected to node q, x p,q =0 indicates that node p and node q are not connected; q≠p, q=1,2,…,N.

[0101] c) Initialize all nodes, let x p =0, n p = 0. Where, x p =1 indicates that node i is selected as the sensing node, x p =0 indicates that the node was not selected; n p This indicates that sensing node i will be separated by n. p Each frame transmits sensing data once.

[0102] d) Perform the following operations on the nodes of the first, second, third, and fourth gears in sequence:

[0103] d1) For nodes within the same gear range, based on the degree D i Sort the nodes from largest to smallest, prioritizing the node with the largest degree.

[0104] d2) If the current node is a normal node, select the current node as the sensing node to collect the characteristics of sudden business traffic, that is, update the x of the current node. i =1; then update matrix G: set the row and column of the current node to 0.

[0105] d3) Determine if all elements in G are 0. If yes, end the operation; otherwise, continue to perform d1) to d2) operations on other nodes in the same gear.

[0106] d4) If all nodes in the same gear have been traversed, jump to the next gear and perform the same operation until all elements in G are 0. If the selected sensing node is within the first or second gear, the sensing data transmission frequency is once every two frames, i.e., updating the node's n... p =2; If the selected sensing node is within the third or fourth tier, the sensing data transmission frequency is once every 4 frames, that is, the n of the updated node is updated. p =4.

[0107] d5) Output all selected sensing nodes and the transmission frequency of the sensing data of each selected sensing node, i.e., output x. p and n p .

[0108] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention by utilizing the methods and techniques disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.

[0109] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A task-driven dynamic allocation method for time-frequency resources, characterized in that, include: Construct a hierarchical clustered network; wherein the hierarchical clustered network consists of M subnets, each subnet containing N nodes: 1 network management node, K gateway nodes and N-1-K ordinary nodes; each node has a unique identity; the subnets exchange data through the gateway nodes; the network management node is used for time and frequency resource allocation; M≥1, K≥1, N≥3; Based on the node's identity identifier, determine the current node type; when the current node is determined to be a network management node, call the time-frequency resource dynamic allocation module to allocate time-frequency resources; the time-frequency resource dynamic allocation module includes: LSTM prediction module, time slot resource allocation module, inductive computing resource management module, and frequency hopping pattern allocation module; When the current node is determined to be a network management node, the time-frequency resource dynamic allocation module is invoked to allocate time-frequency resources. This includes: when the current node is determined to be a network management node, the inductive computing resource management module is invoked to collect the local link traffic characteristics of other nodes as historical service traffic. The historical service traffic characteristics are the characteristics of services that have been transmitted in the past; the LSTM prediction module is invoked to predict the local link traffic characteristics in the next period based on the local link traffic characteristics of other nodes collected by the inductive computing resource management module, and the prediction result is obtained; the time slot resource allocation module is invoked to allocate time slot resources within the subnet based on the prediction result output by the LSTM prediction module and the link construction requirement information issued by the network layer. The link construction requirement information issued by the network layer is the characteristics of services that will be transmitted in the next period, i.e., future service traffic; and the frequency hopping pattern allocation module is invoked to allocate frequency hopping patterns within the subnet based on the spectrum interference results of all ordinary nodes in the entire network obtained by the network management node.

2. The task-driven dynamic allocation method for time-frequency resources according to claim 1, characterized in that, Both historical and future business traffic include: source node, destination node, business priority, and business volume; the format is: <source node, destination node, business priority, business volume>.

3. The task-driven dynamic allocation method for time-frequency resources according to claim 1, characterized in that, The flight process consists of multiple mission phases, each of which can be considered as a cycle. These mission phases include: navigation phase, collision avoidance phase, enemy attack phase, and detection phase.

4. The task-driven dynamic allocation method for time-frequency resources according to claim 1, characterized in that, When the LSTM prediction module predicts the local link traffic characteristics for the next period, it includes: Will Time slice Covered links Historical business traffic is represented as ;right Normalization is performed to obtain the normalized result. : in, , Indicates the duration of the current period; express The average value; Then, time window The normalized result of future business traffic within the length is expressed as: ,have: in, ; Will The former The first step is used as the input value for the LSTM neural network. The step is used as the comparison value with the minimum mean square error to divide the data into... Yes, we obtained the training data; The LSTM neural network is trained based on the training data to obtain the prediction model; Will After Step data is used as input to the prediction model, and the prediction model outputs predicted values. ; Retrieve time window The predicted values ​​within the length are used to obtain the predicted sequence. ; The predicted sequence is denormalized to obtain the predicted business traffic for the next period. Repeat the above steps to obtain the predicted service traffic for all links in the network in the next cycle.

5. The task-driven dynamic allocation method for time-frequency resources according to claim 4, characterized in that, Loss function of LSTM neural network for: in, In the LSTM neural network, the first... Output values ​​for the training data.

6. The task-driven dynamic allocation method for time-frequency resources according to claim 1, characterized in that, When allocating time slot resources within a subnet, the time slot resource allocation module includes: filling in the micro-time slots required for inter-node service transmission within a frame based on the prediction results output by the LSTM prediction module and the link construction requirement information.

7. The task-driven dynamic allocation method for time-frequency resources according to claim 1, characterized in that, The frequency hopping pattern allocation module, when performing frequency hopping pattern allocation within a subnet, includes: According to the 9th level irreducible polynomial x 9 +x 4 +1 generates an m-sequence f_p, and then performs... Operation; where mod represents the remainder, and F_num represents the total number of frequency points; Narrowband determination is performed on the frequency points used by N nodes at any given time. If |f_(p+1)-f_p|<15, then f_(p+1)=[f_(p+1)+30] This yields the frequency hopping pattern of N nodes within the subnet; where p = 1, 2, ..., N.

8. The task-driven dynamic allocation method for time-frequency resources according to claim 1, characterized in that, Also includes: Call the synesthetic computing resource management module to perform resource management: The node service transmission demand is sorted from smallest to largest, and the nodes are divided into four levels of busyness according to the sorting size: the first level is the first N / 4 nodes, the second level is the nodes from N / 4 to N / 2 nodes, the third level is the nodes from N / 2 to 3N / 4 nodes, and the fourth level is the nodes from 3N / 4 to N nodes. Initialization matrix G = adjacency matrix [x p,q ], where x p,q =1 indicates that node p is connected to node q, x p,q =0 indicates that node p and node q are not connected; q≠p, q=1,2,…,N; Initialize all nodes, let x p =0, n p =0; where x p =1 indicates that node i is selected as the sensing node, x p =0 indicates that the node was not selected; n p This indicates that sensing node i will be separated by n. p Each frame transmits sensing data once; Perform the following operations on the nodes of the first, second, third, and fourth tiers in sequence: a) For nodes within the same gear range, based on the degree D i Sort the nodes from largest to smallest, prioritizing the node with the largest degree. b) If the current node is a regular node, select the current node as the sensing node to collect characteristics of sudden business traffic, i.e., update the x of the current node. i =1; then update matrix G: set the row and column of the current node to 0; c) Determine if all elements in G are 0. If yes, end the operation; otherwise, continue with operations a) to b) for other nodes in the same tier. d) If all nodes in the same gear have been traversed, jump to the next gear and perform the same operation until all elements in G are 0; where, if the selected sensing node is a node in the first or second gear, the sensing data transmission frequency is once every 2 frames, that is, update the node's n. p =2; If the selected sensing node is within the third or fourth tier, the sensing data transmission frequency is once every 4 frames, that is, the n of the updated node is updated. p =4; f) Output all selected sensing nodes and the transmission frequency of the sensing data from each selected sensing node, i.e., output x. p and n p .

Images