A Resource Allocation Method for UAV-Assisted D2D-Cellular Hybrid Networks Based on Two-Layer Structure Entropy Minimization
By introducing a resource allocation method that minimizes entropy in a two-layer structure, the topology of the UAV-assisted D2D-cellular hybrid network is optimized, solving the problems of network disorder and interference propagation, and improving the stability and efficiency of the system.
Patent Information
- Application Number
- CN202611187101.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-06
- Publication Date
- 2026-11-10
AI Technical Summary
In existing UAV-assisted D2D-cellular hybrid networks, the disordered network topology, uncontrollable interference propagation, frequent network reconfiguration, and lack of quantification tools lead to insufficient system robustness and high operating costs.
Based on the resource allocation method of minimizing two-layer structural entropy, a cumulative structural entropy minimization optimization model is established by constructing a weighted communication structure graph and defining two-layer structural entropy. The model is then transformed into a linear upper bound approximation form using a first-order Taylor expansion. Finally, a hierarchical alternating iterative solution algorithm is used to optimize the network topology order and communication performance.
It achieves coordinated optimization of network topology order and communication performance, reduces interference cascading propagation and network reconfiguration frequency, reduces signaling overhead, and improves system stability and operating efficiency.
Smart Images

Figure CN123075490A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication and network optimization technology, specifically to a method for allocating resources in a UAV-assisted device-to-device (D2D)-cellular hybrid network based on minimizing entropy in a two-layer structure. This method is applicable to spectrum resource management and network topology control in 5G / 6G high-density, multi-connection wireless communication network scenarios. Background Technology
[0002] With the rapid development of mobile communication technology, data traffic in cellular networks is growing exponentially, leading to increasingly scarce spectrum resources. Device-to-device (D2D) communication, as a key technology that can improve spectrum utilization efficiency, reduce communication latency, and alleviate base station load, is widely used in cellular communication networks. Meanwhile, unmanned aerial vehicles (UAVs), with their advantages of flexible deployment and good air-to-ground line-of-sight link conditions, are being introduced into wireless communication networks to enhance coverage, increase system throughput, or improve link quality. In UAV-assisted D2D-cellular hybrid networks, UAVs can act as airborne relay nodes, providing multi-hop relay services to users outside the direct D2D communication range, thereby extending network coverage and improving communication reliability.
[0003] In UAV-assisted D2D-cellular hybrid networks, communication mode selection, spectrum resource reuse, and transmit power control are the core issues in resource allocation. Extensive research has been conducted on this topic. For example, Feng Daquan et al. studied the joint access control, mode selection, power control, and resource allocation optimization problem in a scenario where UAVs and cellular networks coexist, aiming to maximize the overall network throughput, and employed convex optimization and convex-concave process (CCCP) iterative algorithms for the solution. Zeng Xiaowan et al., focusing on UAV-assisted D2D communication networks, jointly optimized spectrum reuse and power control strategies with the goal of maximizing network security. Furthermore, other studies have modeled the resource allocation problem in UAV-assisted D2D communication as a mixed-integer nonlinear programming problem and solved it through subproblem decomposition and iterative algorithms.
[0004] However, the above-mentioned existing methods all have the following shortcomings: First, they ignore the ordered nature of network topology. Existing resource allocation methods primarily construct optimization objectives around traditional physical layer performance metrics such as maximizing throughput, maximizing energy efficiency, minimizing latency, or maximizing security rate. These methods only focus on link-level signal-to-interference-plus-noise ratio (SIR) and data rate during the optimization process, without imposing any constraints or guidance on the overall topology of communication connections. In high-density, multi-connection hybrid network scenarios, even if throughput or energy efficiency metrics reach optimal levels, communication connections may still exhibit a highly dispersed and disordered state, making the network extremely sensitive to changes in interference, node movement, and load fluctuations, resulting in insufficient system robustness.
[0005] Second, interference propagation is uncontrollable. In D2D link reuse of cellular spectrum resources, adjustments to the transmit power or mode switching of one link can affect multiple other links through interference coupling. Existing throughput-oriented optimization methods often excessively increase transmit power or select more interference-prone reuse modes to improve the rate of local links, causing interference to "cascade" through the topology and ultimately degrade the overall network performance. In UAV-assisted scenarios, the line-of-sight transmission characteristics of air-to-ground links further amplify the propagation range of interference, making this problem even more prominent.
[0006] Third, frequent network reconfigurations and high operating costs. In dynamic environments, existing methods lack consideration for topology stability. Whenever channel conditions or user locations change slightly, resource allocation schemes may be drastically adjusted, leading to frequent network reconfigurations and increasing signaling overhead and system operating costs. This deficiency is particularly evident in scenarios where UAV movement causes dynamic changes in air-to-ground links.
[0007] Fourth, there is a lack of systematic quantitative tools for network topology. In recent years, some researchers have attempted to introduce the concept of graph entropy into wireless communication networks to quantify the uncertainty, complexity, and potential information content of network topology. However, existing graph entropy methods are mainly aimed at static topology analysis of non-spatial or scale-free networks and have not yet been applied to real-time resource allocation and topology collaborative optimization in UAV-assisted D2D-cellular hybrid networks. Wireless communication networks are essentially spatial networks, and the connectivity between nodes depends on spatial distance. Existing graph entropy methods fail to effectively combine the reliability of communication links, interference coupling, and other wireless propagation characteristics, making them difficult to directly guide resource allocation decisions in dynamic networks.
[0008] Therefore, how to quantitatively measure and effectively regulate the orderliness of the network topology while ensuring the basic service quality for cellular and D2D users, thereby improving the stability and controllability of the system from a structural perspective, has become an urgent technical problem to be solved. Summary of the Invention
[0009] The purpose of this invention is to address the aforementioned problems by providing a UAV-assisted D2D-cellular hybrid network resource allocation method based on minimizing two-layer structural entropy. This method constructs a topology metric model and a cumulative structural entropy minimization optimization equation for the communication system based on structural information theory. While ensuring QoS constraints for both cellular and D2D users, it introduces structural entropy to quantify the degree of orderliness in the network topology. A multi-dimensional resource joint optimization framework centered on minimizing cumulative structural entropy is constructed, and a first-order Taylor expansion transforms the nonlinear structural entropy objective function into a linear upper-bound approximation. Through a hierarchical alternating iterative solution algorithm, it achieves coordinated optimization of communication performance and structural stability.
[0010] The technical solution of the present invention is as follows: A UAV-assisted D2D-cellular hybrid network resource allocation method based on minimizing two-layer structure entropy includes the following steps: Construct a wireless communication system: Establish a wireless communication system consisting of a cellular base station, a UAV relay node, multiple cellular users, and multiple pairs of D2D users; the system includes three types of channel links: downlink from the base station to the cellular user, direct link from the D2D transmitter to the receiver, and UAV-assisted mode link; Construct a weighted communication structure graph and define a two-layer structural entropy: calculate the received signal-to-interference-plus-noise ratio of each link, assign weights to communication edges based on link reliability, construct a weighted communication structure graph, define node weighting degree, and define a two-layer structural entropy that characterizes the network topological ordering based on node weighting degree; Constructing a joint optimization problem model: With minimizing the cumulative two-layer structural entropy over the entire time window as the optimization objective, a joint optimization problem model is constructed that includes spectrum allocation constraints, transmit power constraints, and quality of service constraints; Approximation and decomposition of the optimization problem: By utilizing the concavity of the logarithmic function, the entropy of the two-layer structure is transformed into a linear upper bound approximation of the node weighting degree by performing a first-order Taylor expansion at the current operating point. The original optimization problem is decomposed into three sub-problems: communication mode and spectrum reuse optimization, power allocation optimization, and weighted communication graph reconstruction. Iterative solution and output: By iterating alternately among the three subproblems, the system structure entropy is monotonically reduced to the convergence point, and the optimal communication mode selection, spectrum reuse status and transmit power of each node are output.
[0011] By introducing two-layer structural entropy as a quantitative measure of network topological orderliness and minimizing it as the optimization objective, the abstract technical requirement of making network topology more ordered is transformed into a concrete mathematical optimization problem. Simultaneously, by linearizing the nonlinear structural entropy objective through first-order Taylor expansion, a solvable linear form is achieved, making the previously computationally infeasible mixed-integer nonlinear programming problem solvable. The iterative solution framework of three subproblems ensures the monotonic convergence of structural entropy throughout the optimization process, fundamentally solving the technical problems of systemic instability, interference cascading propagation, and frequent network reconfiguration caused by topological disorder in high-density hybrid networks.
[0012] Furthermore, the two-layer structural entropy is defined as follows: network nodes are divided into four functional substructures according to their functions: base station substructure, UAV substructure, cellular user substructure, and D2D user substructure. The two-layer structural entropy consists of two items: inter-substructure entropy and intra-substructure entropy. Inter-substructure entropy quantifies the uncertainty of communication traffic distribution between functional substructures, while intra-substructure entropy quantifies the load balancing degree of nodes within each functional substructure.
[0013] By naturally dividing the network into four functional entities—base stations, drones, cellular users, and D2D users—the entropy between substructures of the two-layer structure can quantify whether the communication traffic distribution among these four functional entities is coordinated and balanced. The entropy within a substructure can quantify whether the load on each node within the same functional entity is uniform. This hierarchical characterization makes the source of network topology disorder identifiable—operators can clearly know whether the disorder stems from imbalances between functional layers or uneven load distribution within a specific functional layer, thus providing fine-grained information for targeted regulation.
[0014] Furthermore, the entropy of the two-layer structure The calculation expression is:
[0015] in, For the first A functional substructure, substructure volume, The total volume of the weighted communication graph. The total number of substructures. For nodes The weighting degree.
[0016] By numerically representing structural entropy at two levels, previously unquantifiable topological characteristics are transformed into calculable, comparable, and optimizable technical parameters. Lower structural entropy values indicate a more ordered network, allowing system operators to directly assess the health of the current network topology based on this value.
[0017] Furthermore, the approximate formula for the linear upper bound of the entropy of the two-layer structure with respect to the node weighting degree is as follows: , in, For the first The weight coefficients, determined by the gradient of the structural entropy with respect to the node weighting degree in the next iteration, are calculated using the following expression: , in, The total volume of the weighted communication graph. For the first In the next iteration, the nodes induced by the current weighted communication graph The structural probability distribution, This is a constant term that is independent of the variable currently being optimized.
[0018] The calculation of the weighting coefficients enables the automatic identification of weak nodes in the current topology and prioritizes strengthening their communication connections, guiding the weighted distribution of network nodes towards a more balanced state. The optimization results of adjacent time slots thus possess inherent smoothness and temporal continuity, effectively reducing network reconfiguration frequency and signaling overhead. Simultaneously, the linear approximation form ensures monotonically converging algorithm, and the runtime meets the engineering requirements for real-time scheduling of dynamic networks.
[0019] Furthermore, the node weighting degree The calculation expression is: , in, For time slots The set of activated communication links, To activate the edge The reliability edge weight is calculated as follows: , in, Represents a probability operator. For nodes To the node The received signal-to-interference-plus-noise ratio, This is the preset signal-to-interference-plus-noise ratio (SIR) threshold.
[0020] By using link reliability (the probability of the signal-to-interference-plus-noise ratio reaching a threshold) as the edge weight, the node weighting accurately reflects the effective communication capacity maintained by that node. During the process of minimizing structural entropy, the system adjusts power allocation and mode selection to make the node weighting more balanced, eliminating extreme super-hub nodes in the network. The impact of performance fluctuations of a single link on the overall topology is dispersed and diluted, and the propagation range and intensity of cascading interference are effectively suppressed.
[0021] Furthermore, the expression for the objective function in the construction of the joint optimization problem model is: , in, The set of decision variables to be optimized includes communication mode selection variables, spectrum reuse variables, and transmit power variables of each node; For time slot index, This represents the total number of time slots. Quality of service constraints include cellular user quality of service constraints and D2D user quality of service constraints, whose expressions are as follows: , , in, and For cellular users and D2D In the time slot For resource blocks The allocation indicator variable, For cellular users, the downlink achievable speed For D2D users' effective rate, and These are the preset minimum rate requirement thresholds; The expression for the transmit power constraint is: , in, , and These represent the transmission power of the base station, D2D transmitter, and drone in the corresponding resource block or mode, respectively. , and This corresponds to the maximum transmit power limit.
[0022] By using an architecture that minimizes structural entropy and constrains QoS, system operators can achieve a flexible and continuously adjustable trade-off between total system throughput and topological orderliness by adjusting the tightness of QoS constraint thresholds, rather than a rigid "either / or" choice. During emergency communications, structural stability can be prioritized, while throughput performance can be taken into account during normal services.
[0023] Furthermore, the specific steps for monotonically decreasing the system's structural entropy to the convergence point through alternating iterations among the three sub-problems include: Based on the weighted communication structure graph of the current iteration point, calculate the structural probability distribution of each node and the weight coefficients in the linear upper bound approximation; Under the condition of fixed transmission power of each node, a greedy scheduling strategy is adopted based on weight coefficients to optimize communication mode selection and resource block allocation, so as to minimize the weighted sum of node weights. Under the conditions of fixed communication mode and spectrum reuse scheme, the optimal transmit power of each node is solved by the successive convex approximation method; Based on the optimized communication mode, spectrum reuse, and transmit power, the signal-to-interference-plus-noise ratio and reliability weight of each link are recalculated, the weighted communication graph is reconstructed, and the weighting degree and two-layer structure entropy of each node are updated. Determine whether the difference between the current two-layer structure entropy and the previous two-layer structure entropy is less than the preset convergence tolerance. If so, stop the iteration and output the result; otherwise, return to S5.1 to continue the iteration.
[0024] In the iterative steps, weight coefficient calculation, greedy scheduling, SCA power optimization, and graph reconstruction form a closed-loop feedback, ensuring that the objective function does not increase in each iteration, and that the results of adjacent iterations have temporal continuity. The ability to achieve stable convergence within ten iterations meets the real-time requirements of slot-level scheduling in 5G / 6G networks, and solves the technical deficiency of traditional methods in their inability to quickly and adaptively optimize in dynamic environments.
[0025] Furthermore, the greedy scheduling strategy is as follows: sort all nodes in descending order of the absolute value of their weight coefficients, iterate through all possible communication modes and all available resource block combinations for each node, select the combination that minimizes the weighted sum of the overall node weights as the optimal choice for that node, and correct resource block allocation conflicts after all nodes have completed their selections.
[0026] By sorting nodes according to the absolute value of their weight coefficients from largest to smallest and optimizing them sequentially, the computational complexity of the optimization is linearly related to the system size, ensuring that the algorithm can run in real time in large-scale networks. Background technology points out that existing methods experience a sharp increase in computational complexity in high-density scenarios, making them difficult to deploy in engineering. The linear complexity characteristic of this application allows it to maintain real-time scheduling capabilities even when the number of D2D pairs increases significantly.
[0027] Furthermore, the effective rates for both cellular and D2D users are calculated using the Shannon capacity formula, with the effective rate for D2D assisted mode multiplied by the half-duplex transmission structure coefficient. .
[0028] This ensures the accuracy of rate calculations, enabling service quality constraints to accurately reflect actual available rates rather than theoretical upper limits. In actual deployments, optimization results will not lead to unmet service quality requirements due to overestimation of rates.
[0029] Furthermore, the UAV relay node integrates a half-duplex decoding and forwarding module, which performs signal reception and forwarding in two equal-length transmission phases respectively; in the first transmission phase, the base station sends downlink signals to the cellular user, while the D2D transmitter sends data to the UAV or the corresponding D2D receiver; in the second transmission phase, the UAV forwards the data received in the first phase to the D2D receiver.
[0030] The resource consumption characteristics of UAV-assisted mode in the time dimension were clarified, and the rate loss caused by the half-duplex transmission structure was accurately incorporated into the optimization model, thereby ensuring the system feasibility of multiple access time slot allocation.
[0031] Compared with existing technologies, the advantages of this invention are: 1. This application introduces structural information theory into the resource allocation field of UAV-assisted D2D-cellular hybrid networks. It uses two layers of structural entropy to accurately numerically represent the degree of order in the network topology, transforming the abstract technical requirement of "making the network topology more ordered" into a concrete mathematical optimization objective. During the optimization process, the weighting coefficients of the structural entropy can automatically identify weak nodes in the current topology and prioritize strengthening their communication connections, guiding the weighted distribution of network nodes towards equilibrium. This fundamentally suppresses the risk of systemic instability caused by topological disorder in high-density hybrid networks.
[0032] 2. This application achieves a synergy between topological order optimization and communication performance assurance by minimizing the entropy of a two-layer structure while preserving user service quality requirements within the constraints. Simulation results show that, even with continuously increasing user density, this application consistently keeps topological disorder to a minimum, and the denser the network, the more pronounced the relative advantage becomes. Traditional methods, in pursuit of throughput, activate as many concurrent links as possible, leading to disordered topological expansion. In contrast, this application proactively suppresses redundant connections while satisfying service quality requirements, guiding the network towards a streamlined and ordered form.
[0033] To address the challenging problem of solving the constructed mixed-integer nonlinear programming problem, this application presents an efficient solution algorithm that combines first-order Taylor expansion linearization with hierarchical alternating iteration. This algorithm boasts a strict guarantee of monotonical convergence, achieving stable convergence within a finite number of iterations without the risk of oscillation or divergence. Its runtime meets the engineering requirements for real-time scheduling in dynamic networks, allowing operators to predict and optimize time costs based on the convergence curve, thus facilitating engineering scheduling and resource planning.
[0034] 3. Because this application aims to minimize structural entropy, guiding the network topology to continuously evolve towards an ordered and balanced form, rather than independently maximizing instantaneous throughput in each time slot, the optimization results of adjacent time slots naturally possess smoothness and temporal continuity. Under the same dynamic channel conditions, this application requires significantly fewer network reconstructions than traditional methods, resulting in a substantial reduction in signaling overhead and system operating costs. This effect is particularly pronounced in dynamic scenarios where UAV movement causes continuous changes in the air-to-ground link, implying lower control plane overhead, a more stable user experience, and longer equipment lifespan.
[0035] Meanwhile, by minimizing structural entropy to guide the orderly evolution of the topology, extreme "super hub nodes" no longer appear in the network. In interference environments, super hub nodes are often the main distribution centers of interference, and their performance fluctuations can spread to the entire network through multiple connections. This application eliminates extreme nodes, thus dispersing and diluting the impact of performance fluctuations of a single link on the overall topology, and effectively suppressing the propagation range and intensity of cascaded interference. In UAV-assisted scenarios, this problem is particularly prominent due to the line-of-sight transmission characteristics of air-to-ground links and the wider range of interference propagation, making the interference suppression effect of this application even more significant.
[0036] 4. This application uses the minimum user rate requirement as a constraint rather than abandoning throughput optimization. This allows system operators to achieve a continuously adjustable and flexible trade-off between total system throughput and topological orderliness by adjusting the tightness of the service quality constraint threshold, rather than a rigid "either / or" choice. This trade-off relationship presents a smooth curve shape without abrupt changes or discontinuities, allowing operators to flexibly configure it according to actual scenarios—prioritizing stability during emergency communications and considering throughput performance during normal services, thus possessing broad engineering applicability.
[0037] Furthermore, this application incorporates the UAV's transmit power and spatial location-related channel gain into a unified optimization system, enabling the UAV's spatial deployment freedom to be fully utilized for topology control. Without the need for manual preset flight altitude or reliance on traversal search, the joint optimization framework can automatically discover and converge to the optimal spatial deployment scheme, achieving collaborative optimization of UAV's three-dimensional spatial deployment and network topology orderliness.
[0038] 5. In simulation verification, this application also produced unexpected technical effects that those skilled in the art could not anticipate or derive from existing technologies. First, as network user density continues to increase, the topological disorder growth rate of this application is significantly lower than that of traditional methods and exhibits sublinear characteristics. The disorder cost brought about by network expansion is effectively controlled. This counterintuitive characteristic cannot be directly derived from the idea of introducing structural entropy optimization. Second, structural entropy exhibits a non-monotonic change with UAV altitude, first decreasing and then increasing, with an optimal altitude range. This complex coupling relationship cannot be predicted by simple engineering intuition, but this application can automatically converge to this optimal range without manual intervention, significantly reducing the complexity of engineering deployment. Third, the trade-off between structural entropy and throughput exhibits a continuous and smooth transition, without abrupt changes or segmentations, making the switching of system operating states more stable and controllable.
[0039] In summary, this application systematically solves the technical challenge of quantifying and controlling topological disorder in UAV-assisted D2D-cellular hybrid communication networks by establishing a two-layer structural entropy measurement model, defining the objective of minimizing cumulative structural entropy, and designing an efficient solution algorithm. It achieves synergistic optimization of communication performance assurance and network structure stability, demonstrating outstanding substantive features and significant progress. Attached Figure Description
[0040] Figure 1 This is a flowchart of a UAV-assisted D2D-cellular hybrid network resource allocation method based on minimizing the entropy of a two-layer structure.
[0041] Figure 2 This is a model diagram of the wireless communication system of this application.
[0042] Figure 3 This is a convergence curve showing the cumulative structural entropy of the system in this application as a function of the number of algorithm iterations.
[0043] Figure 4 This is a graph showing the relationship between the structural entropy of the system after convergence and the number of D2D pairs in this application.
[0044] Figure 5 This is a comparison chart showing the trend of the final structural entropy of the system in this application as a function of the UAV's flight altitude.
[0045] Figure 6 This is a graph showing the trade-off between the cumulative structural entropy and the total throughput of the system in this application. Detailed Implementation
[0046] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0047] The features and performance of the present invention will be further described in detail below with reference to embodiments.
[0048] Please see Figure 1-6 A resource allocation method for UAV-assisted D2D-cellular hybrid networks based on minimizing the entropy of a two-layer structure, such as... Figure 1 As shown, it includes the following steps: Step 1: Construct a wireless communication system.
[0049] A wireless communication system is established, consisting of a cellular base station, a UAV relay node, multiple cellular users, and multiple pairs of D2D users. The base station is configured to be located at the center of a single cell to provide downlink communication services to cellular users. The UAV relay node is configured for a single UAV scenario and integrates a half-duplex decoding and forwarding module to perform signal reception and forwarding in two equal-length transmission phases. The system consists of three types of channel links: a downlink from the base station to the cellular user, a direct link from the D2D transmitter to the receiver, and a UAV-assisted mode link supported by UAV relay. These three types of channel links reuse spectrum through resource blocks.
[0050] Step 2: Construct a weighted communication structure diagram and define the two-layer structural entropy.
[0051] Based on the constructed wireless communication system, receive signal models for various links are built, and the signal-to-interference-plus-noise ratio (SIR) of each link under different transmission phases is calculated. (Cellular users) In resource block The expression for calculating the received signal-to-interference-plus-noise ratio is: , in, For base stations in resource blocks On the transmit power, For base station to cellular users The channel gain magnitude, For additive white Gaussian noise power, For D2D pairs Reuse resource blocks Indicator variables, This refers to the transmit power of the D2D transmitter. The interference channel gain magnitude from the D2D transmitter to the cellular user. This is a time slot index.
[0052] Weights are assigned to communication edges based on link reliability to construct a weighted communication structure graph. Activation edges are then performed. The expression for calculating the reliability edge weight is: , in, Represents a probability operator. For nodes To the node The received signal-to-interference-plus-noise ratio, This is the preset signal-to-interference-plus-noise ratio (SIR) threshold.
[0053] node The weighted degree calculation expression is: , in, For time slots The set of activated communication links.
[0054] Based on node weighted degrees, a two-layer structural entropy formula characterizing the topological order of a network is defined using structural information theory. The expression for calculating the two-layer structural entropy is:
[0055] in, For a given function division The entropy of the two-layer structure below, Indicates the first A functional substructure, substructure volume, For weighted communication graph in time slots The total volume The total number of substructures (in this application) (corresponding to four functional substructures: base stations, drones, cellular users, and D2D users). For nodes The weighting degree.
[0056] The two-layer structural entropy is defined based on the natural division of network functions, classifying network nodes into four functional substructures: base station substructure, UAV substructure, cellular user substructure, and D2D user substructure. The first term of the two-layer structural entropy characterizes the uncertainty of communication traffic distribution among the functional substructures, while the second term characterizes the load balancing degree of nodes within each functional substructure. Together, they constitute a complete quantitative description of the network topological orderliness.
[0057] Step 3: Construct a joint optimization problem model with the objective of minimizing the cumulative entropy of the two-layer structure.
[0058] Based on the two-layer structure entropy formula, a joint optimization problem model is constructed with the goal of minimizing the cumulative two-layer structure entropy over the entire time window, which includes spectrum allocation constraints, transmit power constraints, and quality of service constraints.
[0059] The expression for the objective function to be optimized is: , in, The set of decision variables to be optimized includes communication mode selection variables, spectrum reuse variables, and transmit power variables of each node; For time slot index, This represents the total number of time slots.
[0060] Define a set of optimization variables, including communication mode selection variables, spectrum reuse variables, and transmit power variables for each node. Under the conditions of meeting the minimum data rate requirements of cellular users and D2D users and hardware power constraints, construct an objective function to guide the network topology towards a stable and ordered form.
[0061] The expression for the cellular user service quality constraint is: .
[0062] The expression for the D2D user service quality constraint is: .
[0063] The expressions for the transmit power constraints of each node are as follows: , in, and For cellular users and D2D In the time slot For resource blocks The assignment indicator variable; For cellular users In resource block Uplink and downlink achievable speeds For D2D pairs In resource block Effective rate; and These are the preset minimum rate requirement thresholds for cellular users and D2D users, respectively. , and These represent the transmission power of the base station, D2D transmitter, and drone in the corresponding resource block or mode, respectively. , and This corresponds to the maximum transmit power limit.
[0064] Step 4: Approximate and decompose the optimization problem.
[0065] Based on the nonlinear characteristics of the objective function, and utilizing the concavity of the logarithmic function, the entropy of the two-layer structure is transformed into a linear upper bound approximation of the node weighting degree by performing a first-order Taylor expansion at the current operating point.
[0066] The approximate formula for the linear upper bound of the entropy of a two-layer structure with respect to the node weighting degree is as follows: , in, For the first The weight coefficients, determined by the gradient of the structural entropy with respect to the node weighting degree in the next iteration, are calculated using the following expression: , in, The total volume of the weighted communication graph. For the first In the next iteration, the nodes induced by the current weighted communication graph The structural probability distribution, This is a constant term that is independent of the variable currently being optimized.
[0067] Based on the linear upper bound approximation, the original mixed-integer nonlinear programming problem is decomposed into three interrelated subproblems: communication mode and spectrum reuse optimization, power allocation optimization, and weighted communication graph reconstruction. These subproblems form a closed-loop feedback loop through alternating iterations.
[0068] Step 5: Solve the problem using a hierarchical alternating iterative algorithm to output the optimal resource allocation scheme.
[0069] Under fixed power conditions, a greedy scheduling strategy is employed to optimize communication mode selection and resource block allocation. Under fixed mode, a successive convex approximation method is used to optimize the transmit power of each node. Based on the updated resource allocation results, link weights are recalculated and the weighted communication graph is reconstructed. Through alternating iterations among the three sub-problems, the system structural entropy monotonically decreases to the convergence point. The final expression for the iterative equation minimizing the node weight is: , The final output includes optimized communication mode selection variables, spectrum reuse indicator variables, transmit power of each node, and corresponding structural entropy indices.
[0070] This application can be widely used in high-density, multi-connectivity wireless network scenarios such as fifth-generation mobile communication systems (5G), sixth-generation mobile communication systems (6G), large-scale Internet of Things, drone swarm collaborative communication, emergency communication support, industrial Internet monitoring, and intelligent transportation systems.
[0071] The specific embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0072] Construction of wireless communication systems: First, we construct a UAV-assisted D2D-cellular hybrid wireless communication system.
[0073] like Figure 2 As shown, the wireless communication system includes the following four types of nodes: a cellular base station located at the center of the cell, used to provide downlink communication services to cellular users; a UAV relay node flying within a preset altitude range, with a half-duplex decoding and forwarding module integrated inside; multiple cellular users evenly distributed within the cell, obtaining communication services through the base station downlink; and multiple pairs of D2D users, each pair of D2D users including a D2D transmitter and a D2D receiver, with the distance between the D2D transmitter and receiver randomly distributed within a preset range.
[0074] Wireless communication systems include the following three types of channel links: a downlink from base station to cellular user, used by the base station to send downlink data to the cellular user; a direct link from D2D transmitter to receiver, used for direct communication between D2D user pairs; and a drone-assisted mode link, which includes two transmission phases—in the first phase, the D2D transmitter sends data to the drone, and in the second phase, the drone forwards data to the D2D receiver.
[0075] The wireless communication system adopts a half-duplex transmission structure with two equal-length time slots: In the first transmission phase, the base station sends downlink signals to cellular users, while the D2D transmitter sends data to the UAV (auxiliary mode) or the corresponding D2D receiver (direct mode); In the second transmission phase, the UAV forwards the data received in the first phase to the D2D receiver.
[0076] In this embodiment, the number of cellular users is set to 10, the number of D2D user pairs is set to 8, the drone flight altitude range is set to 80 meters to 120 meters, and the number of resource blocks is set to 10. It should be noted that the above parameter values are merely illustrative and this application is not limited thereto. Those skilled in the art can flexibly configure the above parameters according to actual application scenarios.
[0077] In other embodiments of this application, the number of cellular users can be set to 5 to 50 based on cell capacity requirements, the number of D2D user pairs can be set to 2 to 30 pairs based on network load, the drone flight altitude range can be set to 50 to 300 meters based on urban building height and airspace management regulations, and the number of resource blocks can be set to 6 to 100 based on system bandwidth configuration. The specific values of the above parameters should not be construed as limitations on this application.
[0078] To facilitate those skilled in the art in implementing this application, Table 1 provides the system parameter configuration of a specific embodiment of this application.
[0079] Table 1 System Parameter Configuration Table
[0080] Construction of weighted communication structure diagram and definition of two-layer structure entropy: (a) Calculation of signal-to-interference-plus-noise ratio: In the constructed wireless communication system, the received signal-to-interference-plus-noise ratio (SIR) of various links under different transmission phases is first calculated.
[0081] Taking cellular users as an example, when a cellular user receives a downlink signal from a base station on a resource block, its received signal-to-interference-plus-noise ratio (SINR) is determined by the following factors: the base station's transmit power on that resource block, the channel gain magnitude from the base station to the cellular user, the additive white Gaussian noise power, and the interference caused to the cellular user by all D2D links multiplexing the same resource block. In resource block The expression for calculating the received signal-to-interference-plus-noise ratio is: , in, For base stations in resource blocks On the transmit power, For base station to cellular users The channel gain magnitude, For additive white Gaussian noise power, For D2D pairs Reuse resource blocks Indicator variables ( Indicates reuse. (Indicates not to be reused) This refers to the transmit power of the D2D transmitter. The interference channel gain magnitude from the D2D transmitter to the cellular user. For time slot index, It is the set of all D2D pairs.
[0082] For D2D direct-connect links, the received signal-to-interference-plus-noise ratio (SNR) at the D2D receiver is jointly determined by the transmit power of the D2D transmitter, the channel gain magnitude from the D2D transmitter to the D2D receiver, the additive white Gaussian noise power, and interference from the base station and other D2D links sharing the same resource block. In D2D direct-connect mode, the D2D receiver... In resource block The expression for calculating the received signal-to-interference-plus-noise ratio is: , in, This represents the magnitude of the channel gain from the D2D transmitter to its corresponding receiver. The magnitude of the interference channel gain from the base station to the D2D receiver. This represents the magnitude of the interference channel gain from other D2D transmitters to this D2D receiver.
[0083] For UAV-assisted mode links, signal transmission is divided into two phases: In the first phase, the UAV receives signals from the D2D transmitter, and its received signal-to-interference-plus-noise ratio (SINR) is determined by the transmit power of the D2D transmitter, the channel gain magnitude from the D2D transmitter to the UAV, the additive white Gaussian noise (ADG) power, and other interfering links. In the second phase, the D2D receiver receives relayed signals from the UAV, and its received SINR is determined by the UAV's transmit power, the channel gain magnitude from the UAV to the D2D receiver, the ADG power, and other interfering links. In the first phase, the UAV is in the resource block... The expression for calculating the received signal-to-interference-plus-noise ratio is: , D2D receiver in the second phase In resource block The expression for calculating the received signal-to-interference-plus-noise ratio is: , in, The channel gain modulus from the D2D transmitter to the UAV. The magnitude of the interference channel gain from the base station to the drone. The channel gain modulus from the UAV to the D2D receiver. For drones in resource blocks The transmission power.
[0084] The channel gain described above takes into account both large-scale path loss and small-scale fading effects. For links between ground nodes, Rayleigh fading model is used to describe small-scale fading; for air-to-ground links involving UAVs, Rice fading model is used to describe small-scale fading to reflect the line-of-sight dominance characteristic of air-to-ground links.
[0085] In the specific implementation of this application, the channel gain from the base station to the cellular user is jointly determined by path loss and Rayleigh fading. The path loss adopts the 3GPP urban macrocell propagation model, with a path loss index of 3.5 and a shadow fading standard deviation of 8 dB. The channel gain from the D2D transmitter to the D2D receiver is jointly determined by path loss and Rayleigh fading. The path loss index is dynamically adjusted according to the D2D communication distance: 2.8 when the communication distance is less than 50 meters and 3.2 when the communication distance is greater than or equal to 50 meters. The channel gain from the D2D transmitter to the UAV is jointly determined by path loss and Ricean fading. The Ricean factor is determined based on the UAV's altitude and the ground environment: 15 dB when the UAV's altitude is greater than 100 meters and 10 dB when the UAV's altitude is between 80 and 100 meters, to accurately reflect the characteristics of line-of-sight transmission in air-to-ground links that vary with altitude.
[0086] In a preferred embodiment of this application, a line-of-sight (LAS) and non-LAS probability model in large-scale path loss is further considered. For air-to-ground links, the LAS transmission probability is related to the UAV altitude, horizontal distance, and environmental parameters (urban, suburban, or rural). The LAS probability model adopts the air-to-ground channel model defined in 3GPP TR 36.777, where the LAS probability in an urban environment decreases with increasing horizontal distance. By introducing the LAS / Non-LAS probability model, this application can more accurately characterize the stochastic characteristics of air-to-ground links and improve the accuracy of edge weight calculation in the weighted communication structure graph.
[0087] (II) Construction of the weighted communication structure diagram: Based on the signal-to-interference-plus-noise ratio (SIR) of the above-mentioned links, a weighted communication structure diagram is constructed.
[0088] Specifically, each communication node in the system (including base stations, drones, cellular users, and D2D transmitters and receivers) is treated as a vertex in a weighted communication structure graph. For any two vertices, if there is a potential communication link (including downlink from base station to cellular user, direct link from D2D transmitter to receiver, uplink from D2D transmitter to drone, downlink from drone to D2D receiver, etc.), an edge is established between the two vertices.
[0089] For each edge, its weight is defined as a reliability metric for the link, i.e., the probability that the received signal-to-interference-plus-noise ratio (SINNR) of the link reaches a preset threshold. Activate edge. The expression for calculating the reliability edge weight is: , in, Represents a probability operator. For nodes To the node The received signal-to-interference-plus-noise ratio, This is a preset signal-to-interference-plus-noise ratio (SIR) threshold. This reliability weight reflects the ability of the communication link to maintain effective communication in a dynamic wireless environment—the higher the SIR and the more stable the link, the greater its reliability weight; conversely, the lower the SIR or the less stable the link, the smaller its reliability weight.
[0090] Based on the above definition of edge weights, for each vertex, its weighted degree is defined as the sum of the weights of all edges connected to that vertex. Node The weighted degree calculation expression is: , in, For time slots The set of activated communication links. The weighted degree of a vertex reflects the importance of that node in the network topology—the higher the weighted degree, the more effective and higher-quality communication connections it has with surrounding nodes, and the more important its role in the network.
[0091] Based on the weighted degree of each vertex, a probability distribution induced by the network structure and a node probability distribution can be constructed. This distribution reflects the steady-state probability of visiting each vertex during a random walk on the network, where This represents the total volume of the weighted communication graph.
[0092] In this application, the link reliability weights are calculated as follows: For each link, sample values of its received signal-to-interference-plus-noise ratio (SINR) are statistically analyzed within multiple preset time slots, and the frequency at which the sample values exceed a preset threshold is used as an estimate of the link reliability weight. When the channel is a fast fading channel, the reliability weight is approximated using the ergodic capacity formula; when the channel is a quasi-static fading channel, the reliability weight is calculated using the complement function of the outage probability.
[0093] The value of the preset threshold is determined according to the modulation and coding scheme. In the preferred embodiment of this application, when using Quadrature Phase Shift Keying (QPSK) modulation and 1 / 2 code rate channel coding, the preset threshold is 3dB; when using 16 Quadrature Amplitude Modulation (16QAM) and 3 / 4 code rate channel coding, the preset threshold is 8dB; and when using 64 Quadrature Amplitude Modulation (64QAM) and 5 / 6 code rate channel coding, the preset threshold is 15dB. Those skilled in the art can flexibly set the value of the preset threshold according to the actual modulation and coding scheme used.
[0094] Node weighting has a clear physical meaning: in a weighted communication structure graph, a node's weighting is equal to the sum of the reliability weights of all active links of that node, reflecting the total effective communication capacity maintained by that node. When a node's weighting is large, it indicates that the node has established highly reliable communication connections with multiple neighboring nodes, placing it at the hub position in the network; when a node's weighting is small, it indicates that the node has few communication connections or low reliability, placing it at the edge position in the network.
[0095] (III) Definition of entropy in a two-layer structure: Based on the above weighted communication structure diagram, this application defines a two-layer structural entropy to characterize the topological order of the network.
[0096] First, based on the natural division of network functions, all nodes in the network are divided into four functional substructures. Base station substructure It includes base station nodes and drone substructures. Includes drone nodes; cellular user substructure It includes all cellular user nodes; D2D user substructure It includes all D2D transmitter and receiver nodes.
[0097] For each type of functional substructure, its volume is defined as the sum of the weighted degrees of all nodes within that substructure: The total volume of the entire weighted communication graph is defined as the sum of the weighted degrees of all nodes: .
[0098] The entropy of a two-layer structure consists of two terms, and its calculation expression is as follows:
[0099] in, The total number of substructures (in this application) ).
[0100] The first term is the entropy between substructures, which quantifies the uncertainty of communication traffic distribution among functional substructures—the more uneven the distribution of communication traffic among the four types of functional substructures, the larger the entropy value of this term; the second term is the entropy within a substructure, which quantifies the load balancing degree of nodes within each functional substructure—the more uneven the weighting distribution of nodes within a substructure, the larger the entropy value of this term.
[0101] The smaller the entropy of the two-layer structure, the higher the degree of order in the network topology—this is reflected in both the coordinated and balanced distribution of communication traffic among the functional substructures and the uniform and orderly distribution of load among the nodes within each substructure.
[0102] In this application, the minimum entropy between substructures is zero (when all communication traffic is concentrated in one substructure), and the maximum value is... (When communication traffic is When the substructures are evenly distributed, among which The total number of substructures in this application The smaller the entropy between substructures, the more concentrated the communication connections between the four types of functional entities are, indicating a clearer division of network functions.
[0103] The minimum entropy within a substructure is zero (when the communication load of all nodes within the substructure is completely concentrated on one node), and the maximum value is... (When the communication load within the substructure is evenly distributed among all nodes), where For the first The number of nodes within a substructure. The smaller the entropy within a substructure, the more uneven the load distribution among the nodes within that functional entity.
[0104] The sum of the entropies of the two-layer structure ranges from zero to... It should be noted that the optimization objective of this application is to minimize rather than maximize the entropy of the two-layer structure. Therefore, the optimization process guides the network to evolve from a high-entropy disordered state to a low-entropy ordered state.
[0105] The calculation method of the above two-layer structural entropy is closely related to the definition of the four types of functional substructures. It is precisely because of the essential differences in communication functions among base stations, drones, cellular users, and D2D users that structural entropy based on functional partitioning can reflect the structural orderliness of the network topology. If nodes are arbitrarily grouped or randomly partitioned, the calculated structural entropy will lose its physical meaning.
[0106] Construction of the joint optimization problem model: Based on the above definition of two-layer structural entropy, this application constructs a joint optimization problem model with the objective of minimizing the cumulative two-layer structural entropy.
[0107] (a) Definition of optimization objective: The optimization objective is to minimize the sum of the cumulative entropy of the two-layer structure across all time slots within the entire time window.
[0108] Specifically, in each time slot, based on the current communication mode selection, spectrum resource reuse, and node transmit power configuration, the weighted communication structure diagram and its corresponding two-layer structure entropy for that time slot are calculated. The two-layer structure entropies of all time slots are summed to obtain the cumulative two-layer structure entropy for the entire time window.
[0109] The expression for the objective function to be optimized is: , The goal of optimization is to select the optimal communication mode, spectrum reuse, and power allocation strategy to minimize the cumulative entropy of the two-layer structure.
[0110] (ii) Definition of optimization variables: The optimization variables involved in this application include the following three categories: The first category is communication mode selection variables. Used to indicate each D2D user to In each time slot Select direct communication mode ( ) or UAV-assisted communication mode ( ).
[0111] The second category is spectrum reuse variables, including indicator variables for base stations allocating resource blocks to cellular users. ( Indicates cellular user In the time slot Allocated resource blocks ), and indicator variables for D2D link multiplexing cellular resource blocks. ( Indicates D2D pairs In the time slot Reuse resource blocks Each cellular user is allocated one or more resource blocks per time slot for downlink communication; each D2D link may choose to multiplex certain cellular resource blocks for communication in each time slot.
[0112] The third category is the transmit power variables of each node, including the transmit power of the base station on each resource block. Transmit power of each D2D transmitter and the launch power of drones on each resource block. .
[0113] In summary, the set of decision variables Represented as: .
[0114] (III) Definition of Constraints: The optimization problem needs to satisfy the following constraints: First, spectrum allocation constraints. Each resource block can only be allocated to one cellular user in the same time slot, ensuring that there are no conflicts in resource block allocation among cellular users: .
[0115] Second, transmit power constraints. The transmit power of each node must be within the preset minimum and maximum power ranges and must not exceed the hardware capability limits. , , , Third, service quality constraints. The average data rate for each cellular user throughout the entire time window must not be lower than the preset minimum rate requirement; the average data rate for each D2D user throughout the entire time window must not be lower than the preset minimum rate requirement. , , This constraint ensures that this application optimizes topological ordering without sacrificing the user's basic communication service quality.
[0116] Fourth, communication mode consistency constraints. Each D2D user can only select one communication mode (direct connection mode or UAV-assisted mode) per time slot, and cannot select two modes simultaneously: .
[0117] In a preferred embodiment of this application, the spectrum allocation constraint further includes: each cellular user is allocated a maximum of two resource blocks in the same time slot, i.e. This is to control the degree of dispersion of base station transmit power and avoid the decrease in signal-to-interference-plus-noise ratio of each link due to the power being dispersed across too many resource blocks.
[0118] The transmit power constraint further includes: the transmit power of the base station on each resource block can be continuously adjustable (analog domain power control) or discretely adjustable (digital domain power control). In a preferred embodiment of this application, the transmit power of each node adopts a continuously adjustable method, with a power adjustment step size of less than or equal to 1dB, to obtain a more refined topology control capability.
[0119] In the Quality of Service (QoS) constraints, the data rate for cellular users is calculated using the Shannon capacity formula, which states that the rate equals the resource block bandwidth multiplied by the logarithm to base 2—where the argument of the logarithm is 1 plus the signal-to-interference-plus-noise ratio (SIR) of the cellular user on the corresponding resource block. , The effective rate of the D2D direct connection mode is also calculated using the Shannon capacity formula: , The effective rate of D2D assisted mode needs to take into account the coefficient effect of the half-duplex transmission structure—since UAV decoding and forwarding requires the transmission time of two equal-length phases, its effective rate is equal to half the phase rate. , The signal-to-interference-plus-noise ratio (SIR) in the above rate calculations is determined by the received signal model of each link, where the interference term includes signals from all other links that reuse the same resource block. This interference coupling relationship is one of the root causes of the high complexity of the optimization problem in this application—the rates of each link are not independent of each other, but rather form a complex mutual constraint relationship through the spectrum reuse structure.
[0120] Approximation and decomposition of optimization problems: The aforementioned joint optimization problem is a mixed-integer nonlinear programming problem—the optimization variables include both integer-type mode selection and spectrum reuse variables, as well as continuous-type power variables; the objective function (two-layer structure entropy) is a nonlinear function of node weighting; and the constraints include non-convex rate constraints. Directly solving this type of problem is computationally infeasible.
[0121] To solve this technical problem, this application adopts a solution strategy that combines linear approximation with hierarchical decomposition.
[0122] (a) Linearized approximation of the structural entropy objective: First, the objective function of the two-layer structure entropy is approximated using linearization. The expression for the two-layer structure entropy contains logarithmic and fractional terms, making it a nonlinear function. This application utilizes the concavity of the logarithmic function, performing a first-order Taylor expansion on the two-layer structure entropy expression at the current operating point, transforming it into a linear upper bound approximation with respect to the node weighting degree: , In this linear approximation, the weighted degree of each node is multiplied by a weighting coefficient. The weighting coefficient is determined by the structural probability distribution of the current iteration point, and its calculation expression is: , in For the first In the next iteration, the nodes induced by the current weighted communication graph The structural probability distribution. When a node is in a "marginal" or "weak" position in the current topology (i.e., The smaller the weighting factor, the larger the absolute value of its corresponding weighting coefficient, meaning that increasing the weighting factor at this node has a higher marginal benefit in reducing structural entropy.
[0123] This linear approximation has an important mathematical property: it constitutes an upper bound on the entropy of the original two-layer structure. Therefore, minimizing this linear approximation guarantees a monotonically decreasing entropy of the original two-layer structure—this is the theoretical basis for the algorithm's convergence.
[0124] Through the above linearization process, the originally complex problem of minimizing the entropy of nonlinear structures is transformed into a node-weighted degree-weighted sum minimization problem that can be solved efficiently, which greatly reduces the solution complexity of the optimization problem.
[0125] In this application, the first-order Taylor expansion of the two-layer structure entropy with respect to the node weighting degree holds under the following mathematical conditions: the weighting degree of each node is positive (the weighting degree of an isolated node is zero, but in this application, all nodes have at least one valid communication link, so the weighting degree is greater than zero), the volume of each functional substructure is greater than zero, and the total volume of the weighted communication graph is greater than zero. These conditions are naturally satisfied in practical communication systems.
[0126] The linearization error of the first-order Taylor expansion is determined by the higher-order remainder terms. In a preferred embodiment of this application, in order to ensure the accuracy of the linearization approximation, a backtracking search method is used to determine the iteration step size in each iteration—that is, when the linearization approximation point deviates far from the current working point, a damping step size strategy is used to limit the maximum change in the node weighting degree, ensuring that the linearization approximation is effective within the confidence region.
[0127] Weighting coefficient The absolute value of the weight coefficient reflects the priority of the corresponding node in minimizing the current structural entropy: the larger the absolute value of the weight coefficient, the more significant the impact of the weight change on the structural entropy, and the communication connection of the node should be adjusted first in the optimization. This priority ranking provides a physical basis for the greedy scheduling in subsequent subproblems.
[0128] (II) Task decomposition of joint optimization problems: Based on the above linear upper bound approximation, this application decomposes the original mixed-integer nonlinear programming problem into three interrelated subproblems: Sub-problem 1: Communication mode and spectrum reuse optimization. Given a fixed transmit power at each node, optimize the communication mode selection variables and spectrum reuse variables based on the weighting coefficients in the current linear approximation: , The physical meaning of this subproblem is: under the premise of fixed power level, by adjusting "who communicates with whom and which resource block is used for communication", the connection structure of the topology is changed, so that the weighting degree of the nodes is concentrated on the nodes with larger weight coefficients, thereby reducing the structural entropy.
[0129] Sub-problem 2: Power allocation optimization. Under the conditions of fixed communication mode and spectrum reuse scheme, optimize the transmit power of each node: , The physical meaning of this subproblem is: under the premise of fixed communication connection, by adjusting the transmission power of each link to change the signal-to-interference-plus-noise ratio and reliability weight of the link, the node weighting degree can be adjusted, thereby further reducing the structural entropy.
[0130] Subproblem 3: Weighted Communication Graph Reconstruction. Based on the optimization results of subproblems 1 and 2, the signal-to-interference-plus-noise ratio (SIR) and reliability weights of each link are recalculated, the weighted communication graph is reconstructed, and the weighting degree of each node and the two-layer structural entropy are updated. The physical meaning of this subproblem is to "map" the optimization results of the first two steps back to the topology space, forming a benchmark for a new round of iterations.
[0131] The three sub-problems form a closed-loop feedback loop of "connection relationship adjustment → link quality optimization → topology state update": the output of sub-problem one is the input of sub-problem two, the output of sub-problem two determines the reconstructed graph of sub-problem three, and the new topology of sub-problem three determines the weight coefficient of sub-problem one in the next round. The three sub-problems are functionally mutually supportive and have an interactive relationship.
[0132] Layered alternating iterative solution: Based on the decomposition of the above three sub-problems, this application adopts a hierarchical alternating iterative algorithm for solving the problem.
[0133] (a) Algorithm initialization: At the start of the algorithm, initial values for each optimization variable are set, including the initial communication mode selection for each D2D user pair, the initial spectrum reuse scheme for each resource block, and the initial transmit power for each node. Based on the above initial configuration, the initial weighted communication structure diagram and the initial two-layer structure entropy are calculated. .
[0134] (II) Iterative process: Each iteration includes the following four steps: Step 1: Calculate the weighting coefficients. Based on the weighted communication structure graph of the current iteration point, calculate the structural probability distribution of each node. Then, the weight coefficients of each node in the linear approximation are calculated. : , Step 2: Optimize Communication Mode and Spectrum Reuse. With a fixed transmit power for each node, based on the weighting coefficients calculated in Step 1, a greedy scheduling strategy is used to optimize the selection of communication mode and resource block allocation. Specifically, the optimal communication mode and resource block reuse scheme are selected for each node in descending order of weighting coefficients, ensuring the weighted sum of the node weights... minimize.
[0135] Step 3: Optimize transmit power. Under the conditions of fixed communication mode and spectrum reuse scheme, the optimal transmit power of each node is solved using the successive convex approximation method. The successive convex approximation method transforms the original non-convex problem into a series of convex optimization subproblems by applying a convex approximation to the non-convex constraints at the current operating point, and solves them iteratively until convergence.
[0136] Step 4: Reconstruct the weighted communication graph and update the structural entropy. Based on the optimization results of Steps 2 and 3, recalculate the signal-to-interference-plus-noise ratio and reliability weights of each link, reconstruct the weighted communication graph, and update the weighting degree of each node and the two-layer structural entropy. .
[0137] (III) Convergence Determination and Result Output: After each iteration, determine whether the difference between the entropy of the current two-layer structure and the entropy of the two-layer structure in the previous iteration is less than the preset convergence tolerance. : , If the above conditions are met, the algorithm is considered to have converged and the iteration stops; otherwise, the next iteration continues.
[0138] Since the linear approximation constitutes the upper bound of the original structure entropy, and the solution of the three subproblems in each iteration ensures that the objective function does not increase, the entire algorithm has monotonically convergent properties and will inevitably converge to a local optimum within a finite number of iterations.
[0139] To facilitate implementation of this application by those skilled in the art, the complete flow of the hierarchical alternating iterative algorithm of this application is given below in pseudocode form.
[0140] Algorithm 1: A layered alternating iterative algorithm based on minimizing the entropy of a two-layer structure.
[0141]
[0142] In a preferred embodiment of this application, the specific process of the greedy scheduling strategy is as follows: First, sort all nodes in descending order of the absolute value of their weight coefficients; then, perform the following operations on each node in sequence—traverse all possible communication modes and all available resource block combinations, evaluate the change in weight degree of the node and its neighboring nodes under each combination, and select the combination that maximizes the overall weight degree. The smallest combination is selected as the optimal choice for that node; finally, after all nodes have made the above selections, any potential resource block allocation conflicts are checked and corrected to obtain the final communication mode and spectrum reuse scheme.
[0143] The computational complexity of the greedy scheduling strategy is O(n log n). ,in For the number of D2D users, For the number of resource blocks, The number of selectable communication modes for each D2D user pair is 2 in this application, namely direct connection mode and drone-assisted mode. This computational complexity is linearly related to the system size, ensuring that the algorithm can run in real time in large-scale networks.
[0144] In a preferred embodiment of this application, the specific process of the successive convex approximation method is as follows: First, a first-order Taylor expansion is performed on the non-convex rate constraint at the current operating point to obtain its concave lower bound approximation; then, the original non-convex power optimization problem is transformed into a convex optimization problem, and the interior point method or gradient projection method is used to solve the convex optimization subproblem; next, the solution result is used as the new operating point, and the above process is repeated until the change in the power variable is less than a preset threshold. The successive convex approximation method guarantees the generation of a feasible solution in each iteration and converges to the stationary point of the original problem within a finite number of iterations.
[0145] Convergence tolerance The preset value is determined based on actual application requirements. In a preferred embodiment of this application, the convergence tolerance of the normalized structural entropy is set to... That is, when the change in structural entropy between two adjacent iterations is less than The algorithm is considered convergent at a certain time. For applications requiring higher accuracy, the convergence tolerance can be set to [value missing]. For applications requiring faster convergence speeds, the convergence tolerance can be relaxed to [a certain value]. .
[0146] In the specific implementation of this application, the convergence performance of the above iterative algorithm was verified by Monte Carlo simulation on the MATLAB simulation platform. The simulation results show that, under most random channel implementations, the algorithm of this application converges within 10 iterations, and the corresponding computation time is on the order of tens of milliseconds (on a standard PC platform), which meets the real-time requirements of slot-level scheduling in 5G / 6G networks.
[0147] Specific application examples and simulation verification: To help those skilled in the art better understand and implement this application, a specific application example is provided below.
[0148] In this example, consider a circular cell with a radius of 500 meters, with the base station located at the cell center. Ten cellular users and eight D2D user pairs are evenly distributed within the cell. A single drone flies at an altitude of 80 to 120 meters directly above the cell center. The system has 10 resource blocks, each with a bandwidth of 180 kHz. The system time slot length is 0.1 seconds, and a complete scheduling cycle contains 200 time slots.
[0149] The maximum transmit power for each node is set as follows: 30dBm per resource block for base stations, 23dBm per D2D transmitter, and 23dBm for drones. The minimum data rate requirement for cellular users is set to 0.5Mbps, and the minimum data rate requirement for D2D users is set to 0.3Mbps. The reliability signal-to-interference-plus-noise ratio (SNR) threshold is set to 5dB.
[0150] Under the above parameter configuration, the method of this application shall be implemented according to the following steps: First, based on the location of each node and channel conditions, the channel gain of all potential links in each time slot is calculated. For ground links, path loss (path loss exponent of 3.5) and Rayleigh small-scale fading are considered together; for air-to-ground links, path loss (path loss exponent of 2.2) and Rice small-scale fading (Rice factor of 10dB) are considered together.
[0151] Secondly, an initial communication mode selection and spectrum reuse scheme is constructed. In this example, the initial scheme can employ random allocation or a simple distance-based heuristic allocation—for example, when the distance between the D2D transmitter and receiver is less than 50 meters, the direct connection mode is preferred; when the distance is greater than 50 meters, the drone-assisted mode is preferred. Resource block allocation uses a round-robin method to ensure that each cellular user is allocated at least one resource block.
[0152] Then, iterative optimization is performed according to the aforementioned hierarchical alternating iterative algorithm (Algorithm 1). In each time slot, four steps are executed sequentially: weight coefficient calculation, communication mode and spectrum reuse optimization, power allocation optimization, and weighted communication graph reconstruction, until the structural entropy of that time slot converges. The convergence tolerance is set to... .
[0153] Within a complete scheduling cycle (200 time slots), the above optimization process is executed slot by slot to minimize the cumulative entropy of the two-layer structure over the entire time window. Between adjacent time slots, the optimization result of the previous time slot is used as the initial value for the next time slot iteration to accelerate the convergence speed and maintain the temporal continuity of the optimization results between adjacent time slots.
[0154] In the above implementation, the flight trajectory of the UAV between adjacent time slots is also constrained by the maximum horizontal flight speed: The maximum horizontal flight speed constraint in this application is embodied in the inequality constraint of the continuity of the UAV's horizontal position—the horizontal displacement of the UAV in adjacent time slots cannot exceed the maximum horizontal speed multiplied by the time slot length. Since this constraint only involves the UAV's position variable, while the optimization variables in this application are communication mode and resource allocation, and the UAV's position participates in the optimization of each time slot as a known parameter, this constraint is handled independently time-slot by time slot within the time window, without increasing the complexity of the joint optimization problem.
[0155] To verify the technical effectiveness of this application, the applicant conducted a Monte Carlo simulation using MATLAB, averaging the results of 500 random node locations and channel fading. The simulation parameters are shown in Table 1.
[0156] Figure 3The figure shows the convergence curve of the cumulative structural entropy of the system as a function of the number of algorithm iterations in the method of this application. The horizontal axis represents the number of iterations, and the vertical axis represents the normalized cumulative two-layer structural entropy (normalized with the initial structural entropy as 1). It is clear from the figure that the structural entropy of the algorithm in this application exhibits a monotonically decreasing characteristic from the initial value, with the fastest decrease (approximately 40%) in the second iteration, a slower decrease in the fourth iteration, a stable state after the sixth iteration, and essentially reaching the final convergence value in the eighth iteration. The entire convergence process is completed within 10 iterations. This verifies that the algorithm of this application has good convergence performance and low computational complexity, meeting the engineering requirements of real-time scheduling in dynamic networks. In contrast, the figure also shows the divergent oscillation curve of the structural entropy without the alternating iterative structural entropy update mechanism, further demonstrating the superiority of the convergence mechanism of the algorithm in this application.
[0157] Figure 4 The figure illustrates the relationship between the structural entropy after system convergence and the number of D2D pairs. The horizontal axis represents the number of D2D user pairs (from 4 to 16 pairs), and the vertical axis represents the structural entropy after convergence. The proposed method is compared with the traditional throughput maximization strategy, with a "random resource allocation" curve added as a lower bound reference. The results show that as the number of D2D pairs increases from 4 to 16, the structural entropy of all three methods increases, but the structural entropy of the proposed method remains the lowest. Specifically, when the number of D2D pairs is 4, the structural entropy of the proposed method is about 12% lower than that of the throughput maximization strategy; when the number of D2D pairs is 10, the structural entropy of the proposed method is about 25% lower; and when the number of D2D pairs is 16, the structural entropy of the proposed method is about 35% lower. As user density increases, the relative advantage of the proposed method continues to expand, and the structural entropy growth rate is significantly lower than that of the traditional method. This comparative result fully demonstrates the significant advantage of the proposed method in suppressing topological disorder expansion in large-scale, high-density networks.
[0158] Figure 5The figure illustrates the trend of the system's final structural entropy as a function of UAV flight altitude. The horizontal axis represents the UAV's flight altitude (from 60 meters to 140 meters), and the vertical axis represents the converged structural entropy. The results show that the structural entropy exhibits a non-monotonic trend with UAV flight altitude—when the flight altitude increases from 60 meters to 80 meters, the structural entropy gradually increases from 0.52 to 0.55 (due to severe ground obstruction at low altitudes); when the flight altitude increases from 80 meters to 100 meters, the structural entropy significantly decreases from 0.55 to 0.43 (due to improved line-of-sight conditions and reduced path loss); when the flight altitude increases from 100 meters to 110 meters, the structural entropy reaches its lowest value of approximately 0.41; when the flight altitude continues to increase from 110 meters to 140 meters, the structural entropy gradually recovers from 0.41 to 0.48 (due to increased path loss and expanded interference range). This indicates the existence of an optimal altitude range (approximately 100 to 110 meters) where the structural entropy is minimized. This proves that the present application can effectively utilize the spatial degrees of freedom of UAV to control the network topology, and can automatically converge to the optimal altitude range through the joint optimization framework without the need for manual preset flight altitude.
[0159] Figure 6 The graph illustrates the trade-off between cumulative structural entropy and total system throughput. The horizontal axis represents total system throughput (Mbps), and the vertical axis represents cumulative structural entropy. The curves were obtained by adjusting the service quality constraint thresholds (incrementing the minimum rate requirement for cellular users from 0.2 Mbps to 1.0 Mbps). The results show that when the rate requirement threshold is low (0.2 Mbps to 0.4 Mbps), structural entropy decreases slightly with increasing throughput—because under lower service quality requirements, the optimization algorithm has more degrees of freedom to manipulate the topology, and structural entropy decreases with increasing available degrees of freedom. When the rate requirement threshold is at a medium level (0.4 Mbps to 0.7 Mbps), structural entropy increases slowly with increasing throughput, exhibiting an approximately linear positive correlation—for every approximately 0.1 Mbps increase in throughput, structural entropy increases by approximately 0.02. When the rate requirement threshold is high (0.7 Mbps to 1.0 Mbps), structural entropy increases sharply with increasing throughput—because to meet high service quality requirements, the system is forced to activate more links and increase transmission power, leading to a rapid increase in topology disorder. Overall, the method proposed in this application can achieve a continuous and smooth trade-off between structural entropy and throughput, rather than a rigid "either / or" choice. System operators can configure the optimization direction according to actual needs—for emergency communication scenarios, the left side of the curve can be selected to prioritize structural stability, while for ordinary business scenarios, the middle part of the curve can be selected to balance throughput and stability.
[0160] Based on the above simulation results, this application systematically solves the technical problem of the inability to quantify and control topological disorder in UAV-assisted D2D-cellular hybrid communication networks by establishing a two-layer structural entropy measurement model, defining the objective of minimizing cumulative structural entropy, and designing a layered alternating iterative solution algorithm. This achieves synergistic optimization of communication performance assurance and network structure stability.
[0161] In other embodiments of this application, the base station can be a single base station or multiple base stations in a multi-base station cooperative scenario. When the system contains multiple base stations, each base station serves cellular users within its coverage area. Resource allocation and interference coordination among the base stations can be achieved through an extended form of the method in this application—simply by extending the base station substructure into a set of multiple base station substructures, and correspondingly extending the first term of the two-layer structural entropy to the entropy between multiple substructures.
[0162] In other embodiments of this application, the drone can be a single drone or a drone swarm. When the system contains multiple drones, each drone can serve different D2D user pairs, or they can collaboratively serve the same D2D user pair. The method of this application in multi-drone scenarios is extended by expanding the drone substructure into a set of multiple drone substructures, and correspondingly adding the entropy term between the drone substructures to the first term of the entropy of the two-layer structure, thereby achieving topological ordering optimization in multi-drone scenarios.
[0163] In other embodiments of this application, the half-duplex decoding and forwarding transmission structure can be replaced with a full-duplex transmission structure, or with an amplification and forwarding transmission structure. When a full-duplex transmission structure is used, it is not necessary to distinguish between the two transmission phases, and the effective rate of the D2D assisted mode can be increased by about 100%. The calculation method of the structure entropy is adjusted accordingly based on the calculation method of the link weight in the full-duplex scenario.
[0164] In other embodiments of this application, the resource block can adopt an orthogonal frequency division multiple access (OFDMA) resource block structure or a non-orthogonal multiple access (NOMA) resource block structure. When a non-orthogonal multiple access structure is adopted, multiple users can be multiplexed on the same resource block, and the mathematical expression of the spectrum reuse variables needs to be adjusted accordingly, but the basic framework and optimization algorithm of this application are still applicable.
[0165] In other embodiments of this application, the minimum rate requirement in the quality of service constraints can be set differently according to different service types—for example, the minimum rate requirement for voice services is 0.1 Mbps, the minimum rate requirement for video services is 1 Mbps, and the minimum rate requirement for data services is 0.5 Mbps. This application supports differentiated quality of service assurance for different types of users by setting different rate thresholds in the constraints.
[0166] In other embodiments of this application, the preset signal-to-interference-plus-noise ratio (SINR) threshold can be adaptively adjusted according to the modulation and coding scheme of the communication link. When adaptive modulation and coding technology is used, the modulation and coding scheme of the link changes with the SINR, and the reliability threshold... Accordingly, it is adjusted to the minimum signal-to-interference-plus-noise ratio requirement corresponding to the current modulation and coding scheme.
[0167] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.
Claims
1. A UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization, characterized in that, Includes the following steps: Construct a wireless communication system: Establish a wireless communication system consisting of a cellular base station, a UAV relay node, multiple cellular users, and multiple pairs of D2D users; the system includes three types of channel links: downlink from the base station to the cellular user, direct link from the D2D transmitter to the receiver, and UAV-assisted mode link; Construct a weighted communication structure graph and define a two-layer structural entropy: calculate the received signal-to-interference-plus-noise ratio of each link, assign weights to communication edges based on link reliability, construct a weighted communication structure graph, define node weighting degree, and define a two-layer structural entropy that characterizes the network topological ordering based on node weighting degree; Constructing a joint optimization problem model: With minimizing the cumulative two-layer structural entropy over the entire time window as the optimization objective, a joint optimization problem model is constructed that includes spectrum allocation constraints, transmit power constraints, and quality of service constraints; Approximation and decomposition of the optimization problem: By utilizing the concavity of the logarithmic function, the entropy of the two-layer structure is transformed into a linear upper bound approximation of the node weighting degree by performing a first-order Taylor expansion at the current operating point. The original optimization problem is decomposed into three sub-problems: communication mode and spectrum reuse optimization, power allocation optimization, and weighted communication graph reconstruction. Iterative solution and output: By iterating alternately among the three subproblems, the system structure entropy is monotonically reduced to the convergence point, and the optimal communication mode selection, spectrum reuse status and transmit power of each node are output.
2. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 1, characterized in that, The two-layer structure entropy is defined as follows: network nodes are divided into four functional substructures according to their functions: base station substructure, UAV substructure, cellular user substructure, and D2D user substructure. The two-layer structure entropy consists of two items: inter-substructure entropy and intra-substructure entropy. Inter-substructure entropy quantifies the uncertainty of communication traffic distribution between functional substructures, while intra-substructure entropy quantifies the load balancing degree of nodes within each functional substructure.
3. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 2, characterized in that, The entropy of the two-layer structure The calculation expression is: in, For the first A functional substructure, substructure volume, The total volume of the weighted communication graph. The total number of substructures. For nodes The weighting degree.
4. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 1, characterized in that, The linear upper bound approximation formula for the entropy of the two-layer structure with respect to the node weighting degree is as follows: , in, For the first The weight coefficients, determined by the gradient of the structural entropy with respect to the node weighting degree in the next iteration, are calculated using the following expression: , in, The total volume of the weighted communication graph. For the first In the next iteration, the nodes induced by the current weighted communication graph The structural probability distribution, This is a constant term that is independent of the variable currently being optimized.
5. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 1, characterized in that, The node weighting The calculation expression is: , in, For time slots The set of activated communication links, To activate the edge The reliability edge weight is calculated as follows: , in, Represents a probability operator. For nodes To the node The received signal-to-interference-plus-noise ratio, This is the preset signal-to-interference-plus-noise ratio (SIR) threshold.
6. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 1, characterized in that, The expression for the objective function in the joint optimization problem model is as follows: , in, The set of decision variables to be optimized includes communication mode selection variables, spectrum reuse variables, and transmit power variables of each node; For time slot index, This represents the total number of time slots. Quality of service constraints include cellular user quality of service constraints and D2D user quality of service constraints, whose expressions are as follows: , , in, and For cellular users and D2D In the time slot For resource blocks The allocation indicator variable, For cellular users, the downlink achievable speed For D2D users' effective rate, and These are the preset minimum rate requirement thresholds; The expression for the transmit power constraint is: , in, , and These represent the transmission power of the base station, D2D transmitter, and drone in the corresponding resource block or mode, respectively. , and This corresponds to the maximum transmit power limit.
7. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 1, characterized in that, The specific steps for monotonically decreasing the system's structural entropy to the convergence point through alternating iterations among the three sub-problems include: Based on the weighted communication structure graph of the current iteration point, calculate the structural probability distribution of each node and the weight coefficients in the linear upper bound approximation; Under the condition of fixed transmission power of each node, a greedy scheduling strategy is adopted based on weight coefficients to optimize communication mode selection and resource block allocation, so as to minimize the weighted sum of node weights. Under the conditions of fixed communication mode and spectrum reuse scheme, the optimal transmit power of each node is solved by the successive convex approximation method; Based on the optimized communication mode, spectrum reuse, and transmit power, the signal-to-interference-plus-noise ratio and reliability weight of each link are recalculated, the weighted communication graph is reconstructed, and the weighting degree and two-layer structure entropy of each node are updated. Determine whether the difference between the current two-layer structure entropy and the previous two-layer structure entropy is less than the preset convergence tolerance. If so, stop the iteration and output the result; otherwise, return to S5.1 to continue the iteration.
8. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 7, characterized in that, The greedy scheduling strategy is as follows: sort all nodes in descending order of the absolute value of their weight coefficients, iterate through all possible communication modes and all available resource block combinations for each node, select the combination that minimizes the weighted sum of the overall node weights as the optimal choice for that node, and correct resource block allocation conflicts after all nodes have made their selections.
9. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 1, characterized in that, The effective rates for both cellular and D2D users are calculated using the Shannon capacity formula, with the effective rate for D2D assisted mode multiplied by the half-duplex transmission structure coefficient. .
10. The UAV-assisted D2D-cellular hybrid network resource allocation method based on two-layer structure entropy minimization according to claim 1, characterized in that, The UAV relay node integrates a half-duplex decoding and forwarding module, which performs signal reception and forwarding in two equal-length transmission phases respectively. In the first transmission phase, the base station sends downlink signals to cellular users, while the D2D transmitter sends data to the UAV or the corresponding D2D receiver. In the second transmission phase, the UAV forwards the data received in the first phase to the D2D receiver.
