Semantic perception multi-unmanned aerial vehicle CoMP joint space deployment and resource optimization method

Abstract

The invention relates to a semantic perception multi-unmanned aerial vehicle CoMP joint space deployment and resource optimization method, which belongs to the technical field of semantic communication and unmanned aerial vehicle cooperative communication, and comprises the following steps: S1, establishing a system model integrating a multi-unmanned aerial vehicle cooperative communication CoMP physical layer and a task-oriented semantic communication layer; s2, an optimization problem mathematical model combining unmanned aerial vehicle space deployment, cooperative beam forming and semantic compression ratio SCR selection is established; s3, converting a mixed integer nonlinear programming MINLP problem into a sub-problem suitable for solving; and S4, designing a hybrid alternating learning optimization HALO algorithm, and obtaining effective solutions for optimizing variable unmanned aerial vehicle deployment, cooperative beam forming and semantic compression ratio selection.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of semantic communication and unmanned aerial vehicle cooperative communication, and relates to a semantic-aware multi-unmanned aerial vehicle CoMP joint spatial deployment and resource optimization method. BACKGROUND

[0002] Semantic communication represents a new paradigm for the sixth generation of wireless networks, prioritizing meaning-centric transmission. Unmanned aerial vehicles (UAVs) provide deployment flexibility and line-of-sight connectivity for coverage. The integration of these technologies presents challenges in the optimization of UAV deployment, wireless resources, and semantic content fidelity. There is a gap in current research. On the one hand, semantic communication research reports efficiency improvements, but usually assumes static network topology, which excludes the possibility of using mobility to enhance performance. On the other hand, UAV network optimization research focuses on indicators such as throughput, energy efficiency, and Quality of Experience (QoE). These methods are semantic-independent, meaning that resource allocation is performed without data semantic content information. Therefore, a framework is needed to connect the semantic layer with the physical layer that manages UAV deployment and cooperative transmission. For multi-UAV systems, this cooperative transmission is formally implemented through Coordinated Multipoint (CoMP) technology, which requires joint beamforming design of the entire fleet.

[0003] In recent years, there has been an increasing amount of research on semantic-driven wireless network transmission optimization, including wireless resource optimization methods for text transmission, maximizing semantic rate to improve task performance, and introducing successful transmission probability and semantic compression ratio (SCR) to trade off communication resources and task performance. However, these numerical optimization-based methods have high computational complexity and limited adaptability in dynamic wireless environments. To this end, researchers have begun to explore Deep Reinforcement Learning (DRL) techniques to achieve adaptive transmission and resource optimization. Despite this, there are two major limitations in existing research: one is that modeling mainly focuses on text, image, or video transmission, failing to address the unique challenges of semantic communication; the other is that algorithms mainly rely on numerical optimization or DRL, failing to effectively combine the advantages of both to further improve performance. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a semantic-aware multi-unmanned aerial vehicle CoMP joint spatial deployment and resource optimization method.

[0005] To achieve the above purpose, the present application provides the following technical solutions: A semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method includes the following steps: S1: Establish a system model that integrates the physical layer of Coordinated Multipoint (CoMP) communication with a task-oriented semantic communication layer; S2: Establish a mathematical model for the optimization problem of joint UAV spatial deployment, cooperative beamforming, and semantic compression ratio (SCR) selection; S3: Transform the Mixed-integer Nonlinear Programming (MINLP) problem into a subproblem suitable for solving; S4: Design a Hybrid Alternating Learning and Optimization (HALO) algorithm to obtain efficient solutions for optimizing variables such as UAV spatial deployment, cooperative beamforming, and semantic compression ratio selection.

[0006] Furthermore, in step S1, a downlink multi-UAV cooperative semantic communication system is established, including... One drone, A ground user and semantic model, all drones used This indicates that all ground users use This indicates that the system operates on a discrete-time series, with the index being... This allows for dynamic adaptation of drone location and resource allocation strategies; the established models include a semantic communication model, a drone deployment and mobility model, a cooperative transport model, and a quality of experience (QoE) model.

[0007] Furthermore, the core of the semantic communication model is a deep neural network trained end-to-end on a large-scale shared knowledge base; the semantic communication model learns a direct mapping from source data to channel resilient symbols, unifying the tasks of semantic feature extraction and robust channel coding within a single architecture; it is expected to provide users with... The original data is denoted as The information is processed by a learned semantic encoder on the drone to extract its basic meaning; the level of information abstraction is controlled by the key parameter SCR, denoted as... Model SCR as a discrete variable; constrain SCR as... A set of predefined levels:

[0008] Time slot The SCR decision vectors of all users within the system are denoted as follows: .

[0009] Further, in the UAV deployment and mobility model, The mobility of a UAV is controlled by a set of operational constraints that ensure mission safety and physical feasibility, including: UAVs are restricted to operate within a designated horizontal service area ensuring they remain above the intended coverage area:

[0010] A minimum separation distance must be maintained between the horizontal positions of any two UAVs , which are:

[0011] The kinematic constraints of a UAV are modeled by limiting its maximum speed ; the displacement of any UAV within a time slot of duration is therefore constrained: .

[0012] Let denote the vector of all UAV positions.

[0013] Further, in the cooperative transmission model, all UAVs cooperate to transmit signals to users, each UAV equipped with a uniform linear array of antennas, and each ground user possessing a single antenna; in a time slot , the beamforming vector from the UAVs to the users is denoted by ; the transmission power of each UAV is limited by its maximum capability , such that:

[0014] The air-to-ground channel from the UAVs to the users is characterized by a large-scale path loss and a small-scale fading, which is modeled as:

[0015] where denotes the small-scale Rayleigh fading component, is the large-scale path loss; The path loss model is expressed as: ​

[0016] where is the channel power gain at the reference distance of one meter, is the path loss exponent; time the drone and the user between them in 3D Euclidean distance is computed as:

[0017] the user received signal to interference plus noise ratio, SINR, is a direct result of the cooperative transmission, given by:

[0018] where denotes the conjugate transpose, is the power of the additive white Gaussian noise; the SINR of each user must satisfy a minimum threshold , with the following constraints: .

[0019] Further, the QoE model uses semantic similarity to quantify the fidelity of conveying meaning, measuring the cosine similarity between high-dimensional feature embeddings of the original source data and the reconstructed data ; a Transformer-based architecture builds the semantic processing model, where an embedding function maps text or image data to a semantic vector space, and the semantic similarity is computed as:

[0020] The semantic similarity must be kept above a predefined threshold , i.e.:

[0021] With QoE as the main optimization objective, the QoE model must capture the quantity and quality of the transmitted semantic information; based on the established principle of modeling the logarithm of the perceived quality in multimedia streams, the QoE of the user in the time slot is formulated as:

[0022] where is the effective semantic rate, a measure of the useful information successfully delivered, where is the amount of semantic information under a unit of content, is the content length; parameter and is a scaling constant that maps this efficiency to a numerical satisfaction score.

[0023] Further, in the step S2, the objective of the optimization problem mathematical model is to maximize the total QoE of the network range by jointly optimizing the UAV positions and related resources; the resources include wireless layer variables and semantic layer variables; for each time slot , the optimization problem is transformed to optimize the UAV positions , beamforming vectors and SCR allocations ; the optimization problem is formulated as: .

[0024] Further, in the step S3, a logical function is used to model the relationship between the semantic similarity and SINR , SCR ; this function captures the characteristic S-shaped performance curve of the communication system, approximating the similarity is expressed as:

[0025] where , and are fitting parameters that characterize the model performance at each SCR level , which are determined by offline empirical measurements; By substituting the tractable approximation in the equation into the problem P1, the problem P2 is obtained: .

[0026] Further, in the step S4, the problem P2 is decomposed into its discrete and continuous domains using a hybrid alternating learning optimization algorithm, including: employing a DRL agent to manage the combinatorial selection of discrete SCRs, and using classical optimization methods to handle the continuous variables of UAV spatial deployment and beamforming; S41: Discrete SCR selection by PPO: The selection of the SCR vector is modeled as a sequential decision process, making it suitable for a DRL solution; a centralized DRL agent is deployed to sequentially determine the SCR for each user within a given time slot; the Markov decision process is defined as follows: State : the state for making decisions for user includes the complete channel state information matrix , and the SCRs allocated to the previous users; Action : Action is to select SCR level for user , i.e. , action space is discrete space with cardinality ; Reward: Instantaneous reward is QoE achieved for user after solving subsequent UAV spatial deployment and beamforming sub-problems; For DRL algorithm, actor-critic method PPO is adopted to optimize clipped surrogate objective function, actor network maps state to probability distribution over actions, critic network estimates state value function; S42: Joint spatial deployment and beamforming by alternating optimization: For given SCR vector determined by PPO agent, problem 2 is simplified to continuous optimization sub-problem over variables and ; A set of variables is iteratively optimized by alternating optimization method while keeping another set fixed until convergence; S421: Beamforming optimization with fixed deployment locations: In the case of fixed UAV locations , channel realizations are known; The sub-problem is to find optimal beamformers ; The minimum SINR required to satisfy semantic constraints is found by inverting the logical function:

[0027] The beamforming problem is to maximize total QoE under power constraints and effective SINR constraints , which is solved using convex optimization solvers; S422: Deployment location optimization with fixed beamforming: In the case of fixed beamformers , the objective function only depends on UAV locations , which is updated by gradient projection method; The gradient of objective function with respect to each UAV location is calculated , then the location is updated along the projected gradient direction:

[0028] where is the alternating optimization iteration index, is a positive step size, is a projection operator to enforce feasibility of constraints.

[0029] The present application has the beneficial effect that: the present application develops a multi-UAV collaborative semantic communication framework, solves the joint optimization problem of UAV spatial deployment, collaborative beamforming and semantic compression. We formulate this problem as a QoE maximization problem, and propose HALO, a hybrid algorithm that combines deep reinforcement learning and alternating optimization to solve the resulting MINLP problem. Simulation results show that compared with the baseline method without joint optimization, the proposed method achieves higher QoE, which shows that it is beneficial to adopt an overall resource management strategy in semantic-aware UAV networks. Future work can explore the impact of imperfect channel state information and the integration of reconfigurable intelligent surfaces.

[0030] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification, which is to be taken in conjunction with the accompanying drawings, wherein: BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to make the purposes, technical solutions and advantages of the present application clearer, the preferred detailed description of the present application will be combined with the drawings as follows, wherein: Figure 1 The downlink multi-UAV cooperative semantic communication system framework provided by the present application is shown in the figure; Figure 2 The convergence diagram of the PPO agent in the HALO algorithm in the present application is shown in the figure, which shows the average total QoE of each round under different values; Figure 3 The relationship diagram between the total QoE of all schemes in the present application and the number of users ; Figure 4 The relationship diagram between the total QoE of all schemes in the present application and the maximum transmission power ; Figure 5 The relationship diagram between the total QoE of all schemes in the present application and the number of antennas ; DETAILED DESCRIPTION

[0032] Following, the present application is described through specific examples, and other advantages and effects of the present application can be easily understood by those skilled in the art from the description. The present application can also be implemented or applied through other different specific embodiments, and various modifications or changes can be made to the details in the description based on different views and applications without departing from the spirit of the present application. It should be noted that the diagrams provided in the following examples only illustrate the basic concepts of the present application in a schematic manner, and the diagrams only show the components related to the present application without drawing the number, shape and size of the components in actual implementation. The shapes, numbers and proportions of the components in actual implementation can be changed arbitrarily, and the layout of the components can be more complex.

[0033] It should be noted that the diagrams provided in the following examples only illustrate the basic concepts of the present application in a schematic manner, and the diagrams only show the components related to the present application without drawing the number, shape and size of the components in actual implementation. The shapes, numbers and proportions of the components in actual implementation can be changed arbitrarily, and the layout of the components can be more complex.

[0034] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application, however, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details, and in other embodiments, the known structures and devices are shown in the form of block diagrams rather than in the form of details to avoid making the embodiments of the present application difficult to understand.

[0035] Embodiment 1: The present application provides a semantic-aware multi-UAV CoMP joint spatial deployment and resource optimization method, comprising the following steps: S1: establishing a system model integrating a Coordinated Multipoint (CoMP) physical layer and a task-oriented semantic communication layer; S2: establishing an optimization problem mathematical model of joint UAV spatial deployment, cooperative beamforming and semantic compression ratio (SCR) selection; S3: converting the mixed-integer nonlinear programming (MINLP) problem into a sub-problem suitable for solving; S4: designing a hybrid alternating learning and optimization (HALO) algorithm to obtain effective solutions of the optimization variables of UAV spatial deployment, cooperative beamforming and semantic compression ratio selection.

[0036] A downlink multi-UAV cooperative semantic communication system in step S1 is as followsFigure 1 The system is designed to provide on-demand, high-quality multimedia services to a group of ground users by exploiting the principles of semantic communication. The system operates on a sequence of discrete time slots, with denoting the dynamic adaptation of the UAV position and resource allocation strategy. The network consists of two main categories of agents: a swarm of cooperating quadcopters, denoted by and a group of ground users, denoted by Each UAV is equipped with a designated multimedia content and a powerful pre-trained artificial intelligence (AI) model designed for end-to-end semantic transmission. This enables the UAVs to act as intelligent aerial base stations, transcending simple data relaying and instead transmitting only semantically relevant information to their respective users. The environment is modeled within a three-dimensional Cartesian coordinate system, where the ground users have fixed positions, denoted by At each time slot , the UAVs dynamically adjust their horizontal positions, denoted by , while maintaining a constant altitude . The foundation of the physical layer design is CoMP transmission, under which the entire swarm of UAVs operates as a distributed multi-antenna transmitter. This cooperative strategy is crucial for generating high-quality channels that support reliable semantic information delivery, facilitating coherent signal combination and complex interference management across the network. The specific steps are as follows: B1. Establishing a semantic communication model The core of the semantic framework is a deep neural network trained end-to-end on a large-scale shared knowledge base. This model learns a direct mapping from the source data to channel-resilient symbols, effectively unifying the tasks of semantic feature extraction and robust channel coding within a single architecture. The raw data intended for the user , denoted by , is processed by a learned semantic encoder on the UAV to extract its underlying meaning.

[0037] The level of information abstraction is controlled by a key parameter SCR, denoted by . Lower values correspond to more detailed representations, thus having higher potential semantic fidelity, but at the cost of increased channel resource consumption. The SCR is constrained to a set of predefined levels:

[0038] The SCR decision vector for all users within a time slot t is denoted by .

[0039] B2. Drone Deployment and Mobility Model The effectiveness of drone-assisted networks is fundamentally determined by the strategic deployment of their aerial platforms. The mobility of a drone is controlled by a set of operational constraints that ensure mission safety and physical feasibility.

[0040] First, drones are restricted to designated horizontal service areas. Internally, ensure they remain above the intended coverage area:

[0041] Secondly, to ensure safe operation and prevent mid-air collisions, a minimum horizontal separation distance must be maintained between any two drones. Therefore, we have:

[0042] Finally, the kinematic limitations of the drone are achieved by limiting its maximum speed. To model. Any drone during the duration The displacement within the time slot is therefore constrained:

[0043] In summary, the above constraints define the feasible deployment space for drone swarms. A vector representing the positions of all drones.

[0044] B3. Collaborative Transmission Model Under the CoMP framework, all drone collaboration Each user transmits signals. Each drone is equipped with... A uniform linear array of antennas, with each ground user having a single antenna. (In time slots) From drones To users The transmitted beamforming vector is denoted as Each drone The transmission power is limited by its maximum capacity The restrictions make:

[0045] From drones To users A2G channel Characterized by large-scale path loss and small-scale fading. This channel is modeled as ,in This represents the small-scale Rayleigh fading component. is the large-scale path loss, which fundamentally depends on the distance between the UAV and the user. The widely adopted path loss model is given by where is the channel power gain at the reference distance of one meter, is the path loss exponent. The time when the UAV and the user are separated by the 3D Euclidean distance is calculated as .

[0046] The SINR received by the user is a direct consequence of the cooperative transmission and is given by

[0047] where denotes the conjugate transpose, is the power of the additive white Gaussian noise. To ensure a baseline for reliable decoding, the SINR of each user must satisfy a minimum threshold . Therefore, there is the following constraint:

[0048] B4. QoE model In semantic communication, the ultimate performance indicator is not the raw data rate, but the fidelity of conveying the meaning. This fidelity is quantified using semantic similarity which measures the cosine similarity between the high-dimensional feature embeddings of the original source data and the reconstructed data . For a concrete implementation, consider an AI model for semantic processing to be a Transformer-based architecture, where the embedding function maps textual or image data to a semantic vector space. The similarity is then computed as:

[0049] To guarantee a minimum level of meaningful communication, this semantic similarity must remain above a predefined threshold , i.e.,

[0050] Adopt QoE as the main optimization objective, as it directly reflects user satisfaction. The QoE model must capture both the quantity and quality of transmitted semantic information. Based on the established principle of modeling the logarithm of perceived quality in multimedia streams, the QoE of the user in a time slot is formulated as:

[0051] Here​ The effective semantic rate, which can be interpreted as a measure of useful information successfully delivered, is given by where is the amount of semantic information under the unit of content, is the content length. The parameters and are scaling constants that map this effective rate to a numerical satisfaction score.

[0052] The goal in step S2 is to maximize the total QoE of the network range by jointly optimizing the UAV positions and related resources, which includes the following steps: The resources include the wireless layer variables (cooperative beamformers) and the semantic layer variables (SCRs). For each time slot This translates into optimizing the UAV deployment locations the beamforming vectors and the SCR allocations The problem is formally stated as

[0053] Problem 1 is a MINLP problem. The key challenge to solve this problem comes from the semantic similarity function This function lacks a closed-form analytical expression with respect to the optimization variables , making direct optimization impossible. The values of are outputs of a complex, nonlinear AI model, which acts as a black box within the optimization framework.

[0054] To make problem 1 tractable, we approximate the performance of the semantic communication system using standard methods from the literature. We model the semantic similarity as a function of its key input driving factors (i.e., SINR and SCR ). This function effectively captures the characteristic sigmoid performance curve of the communication system, where the accuracy saturates at high SINR. Specifically, the approximated similarity can be expressed as:

[0055] where , and are fitting parameters that characterize the performance of the given AI model at each SCR level These parameters are determined through offline empirical measurements.

[0056] In step S3, the MINLP problem is converted into a suitable subproblem. Its content includes the following steps: By substituting the tractable approximation in the equation into Problem 1, we obtain an approximate but still challenging optimization problem, namely Problem 2, which is an approximation of Problem 1:

[0057] Despite this restatement, question 2 remains a difficult MINLP due to three fundamental issues: (i) continuous variables and Strong coupling within non-convex SINR and QoE expressions; (ii) Discrete SCR variable selection (iii) The nonconvexity of collision avoidance constraints. Exhaustive search of discrete variables leads to exponential complexity, making it impractical for real-time applications. Such complex structures require the development of sophisticated and computationally efficient optimization algorithms.

[0058] Step S4 uses a hybrid alternating learning optimization algorithm to decompose problem P2 into its discrete and continuous domains, including: using a DRL agent to manage the combinatorial selection of discrete SCRs, and using classical optimization methods to handle the continuous variables of UAV spatial deployment and beamforming.

[0059] The detailed process of the HALO algorithm, including the specific steps, is as follows: 1) Initialize the executor network and commentator network .

[0060] 2) Within each training set, observe the initial state. Initialize an empty buffer .

[0061] 3) For each user The agent, according to the policy Select SCR ;based on Call the algorithm of the subproblem to calculate the optimal Calculate rewards ; Observe the next state And in the buffer middle record tuple .

[0062] 4) Use the PPO algorithm and buffers Update network parameters and The above process iterates until training converges or the maximum number of rounds is reached.

[0063] This algorithm decomposes the problem into its discrete and continuous domains. Specifically, a DRL agent is used to manage the combinatorial selection of discrete SCRs, while classical optimization methods are used to handle the continuous variables of UAV spatial deployment and beamforming.

[0064] SCR vector The selection process is modeled as a sequential decision-making process, making it suitable for DRL solutions. A centralized DRL agent is deployed to sequentially determine the SCR for each user within a given time slot. The MDP process is defined as follows: Status: For user The decision-making process includes the complete CSI matrix. and allocated to the former Each user's SCR. This provides the agent with comprehensive context for decision-making.

[0065] Actions: Actions are for the user Select SCR level, i.e. The action space is a cardinality of 1. The discrete space.

[0066] Rewards: Immediate rewards are given to users after the subsequent space deployment and beamforming issues are resolved. This implements QoE, which directly connects the agent's actions to the overall system goal.

[0067] For the DRL algorithm, PPO is used, an actor-critic method known for its stability and data efficiency. (Actor Network) The probability distribution that maps states to actions, and the commentator network Estimate the state-value function. Optimize and prune the agent's objective function using PPO to ensure stable policy updates.

[0068] Given Post-joint optimization The alternating optimization algorithm has the following specific steps: 1) Input SCR variable Position at the previous moment ,initialization .

[0069] 2) Alternate optimization of the main loop: fixed Solving for optimal beamforming using SOCP ;fixed Update location via GPM .

[0070] 3) Continue until the algorithm converges, then output the optimized position. and beamforming variables .

[0071] like Figure 2 As shown, this invention presents the relationship between different discount factors based on Algorithm 1 and the average total QoE. The curves indicate that for all tested discount factors, the agent successfully learned a stable and effective policy, with the total QoE converging after approximately 2500 rounds. Notably, the convergence performance remained consistent across different γ values. This result confirms the stability and effectiveness of the DRL-based method for complex discrete resource allocation tasks.

[0072] like Figure 3 As shown, this invention provides the total QoE and number of users for all solutions. The relationship between them. As can be seen from the figure, due to increased interference between users, the total QoE of all schemes decreases with... Increases and decreases. However, HALO consistently outperforms all benchmarks. The performance gap highlights HALO's ability to intelligently manage resources in interference-constrained scenarios by jointly optimizing UAV spatial deployment, beamforming, and semantic compression to find the optimal trade-offs.

[0073] like Figure 4 As shown, this invention provides the total QoE and maximum transmission power for all schemes. The relationship between the two strategies is shown in the figure. As can be seen, all strategies benefit from increased power, but HALO maintains its superiority throughout. This is because HALO simultaneously utilizes enhanced SINR to optimize beamforming and semantic compression, resulting in a greater QoE gain than a power-only strategy.

[0074] like Figure 5 As shown, this invention provides the total QoE and the number of antennas for all schemes. The relationship between these factors is evident in the figure. The increased spatial degrees of freedom provide superior beamforming gain, allowing for more effective interference mitigation. HALO most effectively leverages this advantage, translating enhanced physical layer capabilities into maximum QoE gain. The consistent and widening performance gap across all these scenarios highlights the synergistic advantages of the overall design, demonstrating that joint optimization of semantic, communication, and mobility resources is crucial to unlocking the full potential of next-generation UAV networks.

[0075] This invention develops a multi-UAV cooperative semantic communication framework that solves the joint optimization problem of UAV spatial deployment, cooperative beamforming, and semantic compression. We formulate this problem as a QoE maximization problem and propose HALO, a hybrid algorithm that combines deep reinforcement learning and alternating optimization to solve the resulting MINLP problem. Simulation results show that our proposed method achieves a higher QoE compared to baseline methods without joint optimization, demonstrating the benefit of adopting a holistic resource management strategy in semantically aware UAV networks. Future work could explore the impact of imperfect channel state information and the integration of reconfigurable smart surfaces.

[0076] Example 2: An electronic device, comprising a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method described in Embodiment 1 when executing the computer program.

[0077] Example 3: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0078] Example 4: A computer program product includes a computer program that, when executed by a processor, implements the method described in Example 1.

[0079] In the above embodiments, the reference to "this embodiment" in the specification indicates that a specific feature, structure, or characteristic described in connection with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple appearances of "this embodiment" do not necessarily all refer to the same embodiment.

[0080] In the above embodiments, although the invention has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory structures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed. The embodiments of the invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0081] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0082] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0083] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0084] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0085] This invention can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0086] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method, characterized in that: Includes the following steps: S1: Establish a system model that integrates the CoMP physical layer for multi-UAV cooperative communication with a task-oriented semantic communication layer; S2: Establish a mathematical model for the optimization problem of joint UAV spatial deployment, cooperative beamforming, and semantic compression ratio (SCR) selection; S3: Transform the mixed-integer nonlinear programming MINLP problem into a subproblem suitable for solving; S4: Design a hybrid alternating learning optimization HALO algorithm to obtain efficient solutions for optimizing variables such as UAV deployment, cooperative beamforming, and semantic compression ratio selection.

2. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: In step S1, a downlink multi-UAV cooperative semantic communication system is established, including... One drone, A ground user and semantic model, all drones used This indicates that all ground users use This indicates that the system operates on a discrete-time series, with the index being... This allows for dynamic adaptation of drone location and resource allocation strategies; The established models include a semantic communication model, a drone deployment and mobility model, a cooperative transport model, and a quality of experience (QoE) model.

3. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: The core of the semantic communication model is a deep neural network trained end-to-end on a large-scale shared knowledge base; the semantic communication model learns a direct mapping from source data to channel resilient symbols, unifying the tasks of semantic feature extraction and robust channel coding within a single architecture; it is expected to provide users with... The original data is denoted as The basic meaning is extracted by a learned semantic encoder on the drone. The level of information abstraction is controlled by the key parameter SCR, denoted as... Model SCR as a discrete variable; constrain SCR as... A set of predefined levels: Time slot The SCR decision vectors of all users within the system are denoted as follows: .

4. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: In the aforementioned drone deployment and mobility model, The mobility of a drone is controlled by a set of operational constraints that ensure mission safety and physical feasibility, including: Drones are restricted to designated horizontal service areas. Internally, ensure they remain above the intended coverage area: Minimum separation distance must be maintained between any two drones in horizontal position. ,have: The kinematic limitations of drones are achieved by limiting their maximum speed. To model; any drone during the duration The displacement within the time slot is therefore constrained: 。 Final use A vector representing the positions of all drones.

5. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: In the cooperative transmission model, all drone collaboration Each user transmits signals, and each drone is equipped with... A uniform linear array of antennas, with each ground user having a single antenna; in time slots From drones To users The transmitted beamforming vector is denoted as Each drone The transmission power is limited by its maximum capacity The restrictions make: From drones To users air-to-ground channel Characterized by large-scale path loss and small-scale fading, this channel is modeled as follows: in This represents the small-scale Rayleigh fading component. It is a large-scale path loss; The path loss model is expressed as: in It is the channel power gain at a reference distance of one meter. It is the path loss index; time drones and users 3D Euclidean distance between The calculation is as follows: user The signal-to-interference-plus-noise ratio (SINR) is a direct result of cooperative transmission and is given by the following formula: in This indicates the conjugate transpose. It is the power of additive white Gaussian noise; Each user's SINR must meet a minimum threshold. The following constraints apply: 。 6. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: The QoE model uses semantic similarity. Quantify the fidelity of the conveyed meaning, and measure the original source data. and reconstructing data Cosine similarity between high-dimensional feature embeddings; A semantic processing model is built based on the Transformer architecture, in which embedding functions Text or image data is mapped to a semantic vector space, and semantic similarity is calculated as follows: Semantic similarity must be maintained at a predefined threshold. Above, that is: Using QoE as the primary optimization objective, the QoE model must capture both the quantity and quality of transmitted semantic information; based on the established principle of logarithmic modeling of perceived quality in multimedia streams, time slots are... Chinese users The QoE formula is: in Effective semantic rate is a measure of the useful information successfully delivered, in which... It represents the amount of semantic information per unit of content. This is the content length; parameter and It is a scaling constant that maps this efficiency to numerical satisfaction scores.

7. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: In step S2, the objective of the optimization problem mathematical model is to maximize the total QoE of the network range by jointly optimizing the UAV's location and related resources; resources include radio layer variables and semantic layer variables; for each time slot... This translates into optimizing the drone's positioning. Beamforming vector and SCR allocation The optimization problem is described as follows: 。 8. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: In step S3, semantic similarity is modeled using a logistic function. With SINR SCR The relationship between them; This function captures the characteristic S-shaped performance curve of a communication system, approximating its similarity. Expressed as: in , and The representation model is at each SCR level. The performance fitting parameters were determined through offline empirical measurements; By substituting the tractable approximation in the equation into problem P1, we obtain problem P2: 。 9. The semantically aware multi-UAV CoMP joint spatial deployment and resource optimization method according to claim 1, characterized in that: In step S4, the problem P2 is decomposed into its discrete and continuous domains using a hybrid alternating learning optimization algorithm, including: using a DRL agent to manage the combined selection of discrete SCRs, and using classical optimization methods to handle the continuous variables of UAV spatial deployment and beamforming. S41: Discrete SCR selection via PPO: SCR vector The selection is modeled as a sequential decision process to suit the DRL solution; a centralized DRL agent is deployed to sequentially determine the SCR for each user within a given time slot; the Markov decision process is defined as follows: state For users The decision-making process includes a complete channel state information matrix. and allocated to the former SCR for each user; action Actions are for the user Select the SCR level, i.e. The action space is based on a base of 1. Discrete space; Rewards: Immediate rewards are given to users after the issues of subsequent drone space deployment and beamforming are resolved. Implemented QoE; For the DRL algorithm, the actor-critic method PPO is used to optimize the pruning agent objective function, and the actor network is employed. The probability distribution that maps states to actions, commentator network Estimate the state value function; S42: Joint space deployment and beamforming through alternating optimization: For a given SCR vector determined by the PPO agent Problem 2 can be simplified to variables. and The continuous optimization subproblem is solved by iteratively optimizing one set of variables using an alternating optimization method while keeping another set fixed until convergence. S421: Beamforming optimization for fixed deployment locations: in fixed UAV locations Given that the channel implementation is known, the subproblem is finding the optimal beamformer. The minimum SINR required to satisfy the semantic constraints is found by reversing the logic function: Beamforming is a problem under power constraints and effective SINR constraints. To maximize the total QoE, a convex optimization solver is used to solve this beamforming problem. S422: Fixed beamforming deployment location optimization: in fixed beamformers In this case, the objective function depends only on the drone's position. Updated using gradient projection; calculate the gradient of the objective function with respect to each UAV position. Then update the position along the projection gradient direction: in It is an alternating optimization of the iterative index. It is the marching length. It is a projection operator that enforces the feasibility of constraints.

Images