Unmanned aerial vehicle assisted low latency semantic transmission method based on orthogonal model segmentation multiple access

Abstract

The application discloses a kind of unmanned aerial vehicle assisted low latency semantic transmission methods based on orthogonal model segmentation multiple access, comprising: S1, based on the physical connection relationship between unmanned aerial vehicle and the signal transmission mechanism of user to be served and constructs communication system;S2, based on the constructed communication system, with all user maximum transmission delay minimization as target, constructs communication system optimization problem model with four constraints;S3, the optimization problem model with constraint condition is solved, and the optimization scheme of all user maximum transmission delay minimization is obtained.The application constructs the alternate iteration logic of resource allocation and trajectory evolution mutual nesting, combines differential convex programming and continuous convex approximation technology, realizes the deep decoupling of highly nonlinear coupling variable, so as to minimize optimization of the worst user transmission delay of system under the premise of guaranteeing the reliability of each user heterogeneous semantic decoding.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and specifically to a UAV-assisted low-latency semantic transmission method based on orthogonal model segmented multiple access. Background Technology

[0002] As research into mobile communication systems deepens, the communication paradigm is shifting from traditional symbol-level accurate recovery to task-oriented semantic-level effective understanding. Semantic communication, by extracting feature representations of information, can convey core intent with extremely low data volume, demonstrating enormous potential in bandwidth-constrained scenarios such as low-altitude intelligent networks and mobile edge computing.

[0003] In unmanned aerial vehicle (UAV)-assisted networks, UAVs, due to their high mobility and on-demand deployment capabilities, are often used as aerial mobile base stations to provide communication and computing services to ground users. However, while semantic communication reduces transmission load, it also introduces additional semantic extraction computational overhead to the sending end. Therefore, how to achieve coordinated optimization of computational latency and transmission latency under the constraints of UAV computing power, transmission power, and limited wireless resources has become a key issue in improving system performance.

[0004] In multi-user shared transmission scenarios, traditional multiple access technologies primarily suppress interference by separating physical resources in the time, frequency, or power domains. In recent years, Model Division Multiple Access (MDMA) has been proposed to further improve spectral efficiency by mining shared semantic features among different users. However, existing MDMA methods have significant limitations. MDMA typically requires strong semantic correlation or modal consistency among users, and as the number of users increases or task differences widen, the shareable semantic features decrease significantly, limiting its performance in complex low-altitude networks.

[0005] To address these challenges, Orthogonal Model Division Multiple Access (ORMA) was introduced. It leverages the orthogonality between different semantic models to achieve co-channel transmission, overcoming the dependence on shared features. However, in practical applications, the performance of ORMA systems is deeply affected by multiple coupled factors. Multiple users transmitting on the same physical resources lead to severe co-channel interference, and the interference level is closely related to the UAV's spatial trajectory. While compressing semantic information can shorten latency, it significantly increases the signal-to-interference-plus-noise ratio (SINR) threshold requirement at the receiver. Significant nonlinear coupling exists between UAV flight trajectory, transmit power allocation, onboard computing power scheduling, and semantic signal length, making it difficult for existing technologies to directly obtain the globally optimal solution.

[0006] Therefore, designing a transmission scheme that can adaptively perceive the physical environment and mission requirements and achieve joint optimization of trajectory and multi-dimensional resources for UAV multi-user systems based on orthogonal model segmentation multiple access is a technical challenge that urgently needs to be solved in the field of low-latency semantic communication. Summary of the Invention

[0007] The present invention aims to at least partially solve one of the technical problems existing in the related art.

[0008] The purpose of this invention is to provide a low-latency semantic transmission method for UAV-assisted multi-user communication based on orthogonal model segmented multiple access, which solves the problem of minimizing maximum latency optimization caused by strong nonlinear coupling between UAV trajectory, transmission power, onboard computing power and semantic compression rate.

[0009] To achieve the above objectives, this invention provides a UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access, comprising the following steps:

[0010] S1. Construct a communication system based on the physical connection relationship and signal transmission mechanism between the drone and the user to be served;

[0011] S2. Based on the constructed communication system, with the goal of minimizing the maximum transmission delay for all users, a communication system optimization problem model is constructed. The communication system optimization problem model is subject to four constraints: UAV trajectory, downlink power allocation, semantic signal length, and airborne computing frequency allocation.

[0012] S3. Solve the optimization problem model with set constraints to obtain the optimization scheme that minimizes the maximum transmission delay for all users.

[0013] Preferably, the communication system includes an aerial base station drone and The ground user, the first The three-dimensional coordinates of a user are represented as follows: The drone in The three-dimensional coordinates of each time slot are: ,in, This indicates the fixed flight altitude of the drone, where T represents...

[0014] Preferably, the communication system optimization problem model is expressed as follows:

[0015]

[0016] in, , , and These represent the sets of optimization variables for UAV trajectory, downlink power allocation, semantic signal length, and onboard computing frequency allocation, respectively. Indicates user In the time slot The semantic signal length, The symbol transmission rate.

[0017] Preferably, the constraints of the communication system optimization problem model include:

[0018] User successfully decoded the constraint:

[0019]

[0020] in, For users In the time slot Signal-to-interference-to-noise ratio, This is the noise tolerance threshold for semantic recovery;

[0021] UAV transmit power constraints:

[0022]

[0023]

[0024] in, For the drone's transmission power, This represents the maximum transmission power of the drone.

[0025] UAV delay constraints within a single time slot:

[0026]

[0027] in, These represent semantic encoding time and transmission time, respectively. The maximum time of a time slot;

[0028] Constraints on the computing frequency of drones:

[0029]

[0030]

[0031] in, Transmitting user data for drones Calculate frequency, This represents the maximum computing frequency of the drone;

[0032] Unmanned aerial vehicle (UAV) flight constraints:

[0033]

[0034] in, For drones The location coordinates of the time slot This represents the maximum speed of the drone.

[0035] Semantic signal length constraints:

[0036]

[0037] in, The semantic signal length of the drone. This represents the minimum semantic signal length for the drone.

[0038] As a preferred option, step S3 specifically involves:

[0039] S31. Based on the decoupling of the algorithm framework of alternating optimization, in view of the strong nonconvexity of the original problem, the optimization variables are split into two mutually iterative sub-problem blocks: joint resource allocation and UAV trajectory planning;

[0040] S32. Based on the current fixed trajectory of the UAV, introduce the DC planning mechanism to solve the joint resource allocation scheme;

[0041] S33. After implementing the fixed resource strategy, introduce non-negative slack variables. The objective is to refactor to maximize the SINR margin for all time slots and users in the system;

[0042] S34. SINR margin-based precise optimization based on binary search method;

[0043] S35. Check the objective function of adjacent rounds. The convergence status, if it meets the set tolerance. If the maximum number of iterations is reached, the algorithm stops and outputs the optimal UAV coordinates, optimal transmission power, onboard computing power allocation, and optimal semantic compression length after trade-offs for each time slot.

[0044] As a preferred option, step S31 specifically involves:

[0045] The original highly coupled nonconvex problem is decomposed into two sub-problems: joint resource allocation. , and and drone trajectory planning ;

[0046] Set outer alternating index By optimizing resources under a fixed trajectory and evolving the trajectory under fixed resources, a closed-loop mechanism is used to decouple multidimensional nonlinear coupled variables.

[0047] Preferably, step S32 is as follows:

[0048] Fix current trajectory Calculate the channel power gain constant. ,:

[0049] ;

[0050] Using the DC programming approach, the signal-to-interference-plus-noise ratio (SIR) expression is logarithmically transformed, converting multiplication and division relationships into addition and subtraction relationships; at the local power point... Linear lower bound for constructing interference terms This allows for the efficient solution of originally non-convex fractional constraints by reconstructing them into standard convex constraints.

[0051] As a preferred option, step S33 specifically includes:

[0052] After implementing a fixed resource strategy, nonnegative slack variables are introduced. The objective is to refactor to maximize the SINR margin for all time slots and users in the system;

[0053] To address the deeply nested range term in the power gain of the air-to-ground channel, a set of relaxation variables is introduced. , representing the squared distance of the target signal, the squared distance of the interference signal, and the upper limit of the total interference power, respectively; the non-convex interference distance constraint is transformed into a linear lower bound constraint using a first-order Taylor expansion.

[0054] Preferably, step S34 is as follows:

[0055] Set SINR margin search interval In the given test value The trajectory subproblem is then transformed into a convex feasibility verification problem.

[0056] Continuously update the Taylor expansion points to reduce the approximation error. If a feasible solution exists below, then update the lower bound. Otherwise update the upper bound. ;

[0057] Through nested inner and outer layers, the final output is the optimal physical coordinate sequence that can overcome co-channel interference in multiple access segmentation by orthogonal models to the greatest extent. .

[0058] Beneficial Effects: This invention introduces an orthogonal model segmented multiple access mechanism, utilizing the orthogonality between different semantic models to achieve resource reuse across model dimensions. By constructing an alternating iterative logic where resource allocation and trajectory evolution are nested, and combining differential convex programming and continuous convex approximation techniques, deep decoupling of highly nonlinear coupled variables is achieved. This results in minimizing and maximizing the worst-case transmission latency of the system while ensuring the reliability of heterogeneous semantic decoding for each user. Attached Figure Description

[0059] Figure 1 The flowchart of a low-latency semantic transmission method for UAV-assisted multi-user communication based on orthogonal model segmented multiple access is provided in Example 1.

[0060] Figure 2 This is a model diagram of the UAV-assisted multi-user communication system based on orthogonal model segmentation multiple access in Example 1.

[0061] Figure 3 The graph shows the change of the maximum semantic signal length of the system with the number of iterations under different maximum transmit powers for the algorithm of this invention and the traditional algorithm.

[0062] Figure 4 The algorithm of this invention and the traditional algorithm under different maximum transmit power The graph shows how the extreme values ​​change with the number of iterations.

[0063] Figure 5 The graph shows the variation of the maximum semantic signal length of the system with the number of iterations under different system symbol transmission rates for the algorithm of this invention and the traditional algorithm.

[0064] Figure 6 The algorithm of this invention and the traditional algorithm under different system symbol transmission rates The graph shows how the extreme values ​​change with the number of iterations.

[0065] Figure 7 The image shows the flight trajectory of the UAV under different maximum power conditions using the algorithm of this invention.

[0066] Figure 8 This is a flight trajectory diagram of a drone under different maximum power conditions using a traditional algorithm.

[0067] Figure 9 The algorithm was developed to visualize the flight trajectory of a drone under different system symbol transmission rates.

[0068] Figure 10 The image shows the flight trajectory of the UAV under different system symbol transmission rates using a traditional algorithm. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0070] To overcome the shortcomings of existing technologies, this invention provides a collaborative optimization framework for UAV downlink semantic transmission based on alternating optimization, achieving joint optimization of UAV trajectory planning, transmit power allocation, semantic signal length, and onboard computing resources. The system introduces an orthogonal model segmented multiple access mechanism, utilizing the orthogonality between different semantic models to achieve resource reuse across model dimensions. By constructing an alternating iterative logic where resource allocation and trajectory evolution are nested, combined with differential convex programming and continuous convex approximation techniques, deep decoupling of highly nonlinear coupled variables is achieved. This ensures the reliability of heterogeneous semantic decoding for each user while minimizing and maximizing the worst-case transmission latency.

[0071] The following is combined with Figures 1-10 This invention describes a low-latency semantic transmission method for UAV-assisted multi-user communication based on orthogonal model segmented multiple access.

[0072] Example 1: This example provides a low-latency semantic transmission method for UAV-assisted multi-user communication based on orthogonal model segmented multiple access, such as... Figure 1 As shown, it includes the following steps:

[0073] S1. Construct a communication system based on the physical connection relationship and signal transmission mechanism between the drone and the user to be served;

[0074] Composed of a rotary-wing drone and It consists of several ground users. The drone, acting as an aerial mobile base station, provides downlink semantic transmission services to all users within its service area during its flight mission, such as... Figure 2As shown. To further improve the multi-user transmission efficiency under limited wireless resources, the system adopts an orthogonal model segmentation multiple access mechanism instead of the traditional frequency domain or power domain resource separation. Leveraging the orthogonality between different semantic models, the UAV can simultaneously transmit semantic signals to multiple users on the same time domain resource, thereby freeing up wireless resource load and providing an efficient implementation method.

[0075] The communication system includes: a UAV-user transmission channel model; UAV downlink transmission based on orthogonal model segmented multiple access; and UAV semantic extraction and semantic transmission constraints.

[0076] In the UAV-user transmission channel model, the first The three-dimensional coordinates of a user are represented as follows: The drone in The three-dimensional coordinates of each time slot are: ,in, Indicates the fixed flight altitude of the drone; drone and user In the time slot The Euclidean distance is: Drones and users The horizontal distance between them is Drones relative to users The angle of elevation can be expressed as: Since the drone is subject to a maximum speed limit during flight, the trajectory changes between adjacent time slots must satisfy motion constraints: ,in, This represents the maximum flight speed of the drone. Furthermore, to ensure mission continuity and executability, the initial and final positions of the drone need to be constrained according to the specific scenario. The air-to-ground propagation link between the drone and the ground user is also considered. Due to building obstructions in urban environments, the air-to-ground link is typically affected by both line-of-sight (LoS) and non-line-of-sight (NLoS) propagation. Using A2G channel modeling methods, the user... In the time slot The LoS probability model is as follows: ,in, and Here, is a constant parameter related to the ground environment, used to characterize the impact of building density and height distribution on the propagation environment. Correspondingly, the NLoS probability is: The additional loss factors in the LoS and NLoS states are respectively and ,user In the time slot The average path loss can be expressed as: ,in, For carrier frequency, To represent the speed of light, and to simplify subsequent optimization derivations, we define an equivalent environmental attenuation coefficient: Including the free space loss component, the total path loss It can be simplified to a more intuitive form: Furthermore, the channel power gain is defined as follows: .

[0077] UAV downlink transmission based on orthogonal model segmented multiple access includes: in the... In each time slot, the UAV uses an orthogonal model segmented multiple access (SMI) mechanism to simultaneously transmit semantic signals to multiple users. Let the time slots allocated to users be... The transmission power is Send to user The normalized semantic signal is ,satisfy Then the drone will be in the time slot. The downlink superimposed transmission signal can be represented as: Accordingly, users In the time slot The received signal is: ,in, It is additive white Gaussian noise. In orthogonal model segmentation multiple access, the semantic signals corresponding to different users are generated by different semantic models. The semantic signals from other users appear as interference terms for the target user. Therefore, the user In the time slot SINR can be expressed as: .

[0078] Unmanned aerial vehicle (UAV) semantic extraction and semantic transmission constraints include: Indicates the CPU cycle for sending users The workload of semantic extraction. ,in , These are task-related constants. This represents the length of the semantic signal. The additional latency caused by local semantic extraction is represented as: The symbol transmission rate is Then the user The transmission delay can be written as: ,user In the time slot The total service latency condition is: ScE is introduced to measure the degree of semantic compression, which is defined as: ,in, Let be the minimum semantic signal length of the UAV, and let the noise tolerance threshold for semantic recovery be . Simplifying it, its noise tolerance threshold function is: ,in, and These are empirically fitted parameters that are highly correlated with specific intelligent tasks and the semantic codec network used. To ensure successful semantic transmission, users... In the time slot The following constraints must be met for successful decoding: .

[0079] S2. Based on the constructed communication system, with the goal of minimizing the maximum transmission delay for all users, a communication system optimization problem model is constructed. The communication system optimization problem model is subject to four constraints: UAV trajectory, downlink power allocation, semantic signal length, and airborne computing frequency allocation.

[0080] The communication system optimization problem model is represented as follows:

[0081]

[0082] in, , , and These represent the sets of optimization variables for UAV trajectory, downlink power allocation, semantic signal length, and onboard computing frequency allocation, respectively. Represented as user In the time slot The semantic signal length, The symbol transmission rate.

[0083] The constraints of the optimization problem include:

[0084] Constraint 1: ;

[0085] Constraint 2: ;

[0086] Constraint 3: ;

[0087] Constraint 4: ;

[0088] Constraint 5: ;

[0089] Constraint 6: ;

[0090] Constraint 7: ;

[0091] Constraint 8: ;

[0092] Among them, constraint 1 is the user's successful decoding constraint; constraints 2 and 6 are UAV transmit power constraints; constraint 3 is the UAV delay constraint within a single time slot; constraints 4 and 7 are constraints on the UAV's calculation frequency; constraint 5 is the UAV flight constraint; and constraint 8 is the semantic signal length constraint. For users In the time slot Signal-to-interference-plus-noise ratio (SINR) For the drone's transmission power, These represent semantic encoding time and transmission time, respectively. Transmitting user data for drones Calculate frequency, For drones The location coordinates of the time slot The semantic signal length of the drone. The maximum time of a time slot, , These represent the maximum transmit power and computing frequency of the drone, respectively. This represents the maximum speed at which the drone can move.

[0093] S3. Solve the optimization problem model with set constraints to obtain the optimization scheme that minimizes the maximum transmission delay for all users.

[0094] Step S31: Perform algorithm framework decoupling based on alternating optimization. In view of the strong nonconvexity of the original problem, the optimization variables are split into two mutually iterative subproblem blocks.

[0095] The original highly coupled nonconvex problem is decomposed into two sub-problems: one is joint resource allocation. , and One is drone trajectory planning. .

[0096] Set outer alternating index By optimizing resources under a fixed trajectory and evolving the trajectory under fixed resources, a closed-loop mechanism is used to decouple multidimensional nonlinear coupled variables.

[0097] Step S32: Joint resource allocation optimization based on DC planning.

[0098] Fix current trajectory Calculate the channel power gain constant. , in This eliminates the nonlinear terms introduced by spatial location.

[0099] Using the DC planning concept, the signal-to-interference-plus-noise ratio (SIR) expression is logarithmically transformed, converting multiplication and division relationships into addition and subtraction relationships. At the local power point... Linear lower bound for constructing interference terms This allows for the efficient solution of originally non-convex fractional constraints by reconstructing them into standard convex constraints.

[0100] Step S33: Perform trajectory optimization modeling based on spatial slack variables.

[0101] After implementing a fixed resource strategy, nonnegative slack variables are introduced. The objective is to refactor to maximize the SINR margin for all time slots and users in the system.

[0102] To address the deeply nested range term in the power gain of the air-to-ground channel, a set of relaxation variables is introduced. , representing the squared distance of the target signal, the squared distance of the interfering signal, and the upper limit of the total interference power, respectively. A first-order Taylor expansion is used to transform the non-convex interference distance constraint into a linear lower bound constraint.

[0103] Step S34: SINR margin precise optimization based on binary search method;

[0104] Set SINR margin search interval At the given test value The trajectory subproblem is then transformed into a convex feasibility verification problem.

[0105] Continuously update the Taylor expansion points to reduce the approximation error. If a feasible solution exists below, then update the lower bound. Otherwise update the upper bound. .

[0106] Through nested inner and outer layers, the final output is the optimal physical coordinate sequence that can overcome co-channel interference in multiple access segmentation by orthogonal models to the greatest extent. .

[0107] Step S35: Check the objective function of adjacent rounds The convergence status, if it meets the set tolerance. The algorithm stops if the maximum number of iterations is reached. The optimal UAV coordinates, optimal transmit power, onboard computing power allocation, and optimal semantic compression length after trade-offs are determined for each time slot.

[0108] To verify the effectiveness of this embodiment, different experimental environments were used to verify the effectiveness of the invention:

[0109] like Figure 3 As shown, the joint optimization algorithm can achieve stable convergence within approximately 13 iterations under different transmit power constraints. Furthermore, by deeply co-scheduling the UAV trajectory space maneuverability and multi-dimensional resource allocation, it significantly reduces the maximum semantic signal length of the system compared to traditional strategies. This strongly demonstrates the significant advantages of the joint scheduling scheme in suppressing co-frequency interference in orthogonal model segmentation multiple access and eliminating the performance bottleneck of the worst user.

[0110] like Figure 4 As shown, the proposed algorithm can achieve SINR margin within 10 to 15 iterations under different power budgets. The algorithm exhibits rapid and stable convergence, demonstrating strong robustness. In performance comparisons, the proposed algorithm achieves superior performance in both low-power (0.1W) and high-power (6.0W) scenarios. The extreme values, approximately 0.0021 and 0.0045 respectively, are significantly better than the traditional strategy, which is approximately 0.012 and 0.026 respectively. This strongly demonstrates that by deeply co-scheduling the UAV's spatial trajectory and downlink transmission power, the system can effectively avoid co-frequency interference in orthogonal model segmentation multiple access and optimize line-of-sight links, thereby extracting more robust transmission for multi-user semantic networks.

[0111] like Figure 5 As shown, the proposed algorithm exhibits superior convergence performance and anti-interference capability compared to traditional methods at different symbol transmission rates. A significant physical logic is that, with increasing rate... By leveraging the reduced transmission time, the system proactively relaxed the semantic compression ratio, thereby increasing the maximum semantic signal length. The value subsequently increased from 460 to approximately 620, which strongly demonstrates that the algorithm can keenly perceive changes in the physical layer and achieve the optimal adaptive dynamic trade-off between latency constraints, computational pressure, and semantic richness.

[0112] like Figure 6 As shown, the proposed algorithm can achieve a better SINR margin than the traditional strategy within 15 iterations under different symbol transmission rates. Convergence; its profound physical logic lies in the symbol rate. The improvement effectively reduces the semantic compression requirement by releasing latency margin, thereby significantly expanding the physical feasible domain of UAV trajectory optimization. This enables the system to completely suppress co-channel interference under the orthogonal model segmented multiple access mechanism through better spatial station positioning, ultimately achieving a dual leap in anti-interference capability and transmission reliability.

[0113] like Figure 7-8As shown, the baseline strategy exhibits rigid and completely consistent trajectories under different power constraints, reflecting its disconnect between resource allocation and physical motion. The proposed algorithm, however, demonstrates superior spatial adaptive optimization capabilities: in extremely low-power (0.1W) scenarios, the UAV deviates significantly from the shortest straight path and approaches the user group at an extreme close, with the trajectory peak reaching approximately 350m. This reduction in physical distance compensates for the severe power deficit by gaining extremely high channel gain. As the transmit power budget increases to 1.2W and even 6.0W, the system intelligently lowers and flattens the flight trajectory, converging towards the initial shortest straight line to optimize range. This dynamic trajectory evolution process powerfully demonstrates that the algorithm can keenly perceive real-time changes in system resource endowment, achieving a highly flexible dynamic trade-off between gain through detours and range reduction through shortcuts. This ensures that the multi-user semantic network achieves optimal cooperative transmission performance under various stringent constraints.

[0114] Figure 9 Figure 10 shows the flight trajectories of UAVs with different system symbol transmission rates under the joint optimization algorithm of this invention and the flight trajectories of UAVs with different system symbol transmission rates under the traditional algorithm. The baseline strategy maintains an unchanged trajectory under different transmission rates, exposing its lack of linkage between physical layer rate and spatial maneuverability; while the proposed algorithm exhibits significant parameter sensitivity and spatial adaptability, even at low transmission rates. At a speed of 25000, due to the increased decoding threshold caused by semantic compression, the drone approaches the user by taking a significant detour, with the highest point of its trajectory reaching 220m to obtain high channel gain. However, at high speeds... When the value is 80000, the trajectory is actively reduced and flattened by utilizing the ample time delay margin, and the peak value is reduced to about 170m to shorten the unnecessary flight distance, thereby achieving the optimal coordinated scheduling of communication resources and trajectory space in the multi-user semantic communication network.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access, characterized in that, Includes the following steps: S1. Construct a communication system based on the physical connection relationship and signal transmission mechanism between the drone and the user to be served; S2. Based on the constructed communication system, with the goal of minimizing the maximum transmission delay for all users, a communication system optimization problem model is constructed. The communication system optimization problem model is subject to four constraints: UAV trajectory, downlink power allocation, semantic signal length, and airborne computing frequency allocation. S3. Solve the optimization problem model with set constraints to obtain the optimization scheme that minimizes the maximum transmission delay for all users.

2. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access according to claim 1, characterized in that, The communication system includes an aerial base station, a drone, and The ground user, the first The three-dimensional coordinates of a user are represented as follows: The drone in The three-dimensional coordinates of each time slot are: ,in, This indicates the fixed flight altitude of the drone.

3. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access according to claim 2, characterized in that, The optimization problem model of the communication system is expressed as follows: ； in, , , and These represent the sets of optimization variables for UAV trajectory, downlink power allocation, semantic signal length, and onboard computing frequency allocation, respectively. Indicates user In the time slot The semantic signal length, The symbol transmission rate.

4. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access according to claim 3, characterized in that, The constraints of the communication system optimization problem model include: User successfully decoded the constraint: ； in, For users In the time slot Signal-to-interference-to-noise ratio, This is the noise tolerance threshold for semantic recovery; UAV transmit power constraints: ； ； in, For the drone's transmission power, This represents the maximum transmission power of the drone. UAV delay constraints within a single time slot: ； in, These represent semantic encoding time and transmission time, respectively. The maximum time of a time slot; Constraints on the computing frequency of drones: ； ； in, Transmitting user data for drones Calculate frequency, This represents the maximum computing frequency of the drone; Unmanned aerial vehicle (UAV) flight constraints: ； in, For drones The location coordinates of the time slot This represents the maximum speed of the drone. Semantic signal length constraints: ； in, The semantic signal length of the drone. This represents the minimum semantic signal length for the drone.

5. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access according to claim 4, characterized in that, Step S3 is as follows: S31. Based on the decoupling of the algorithm framework of alternating optimization, in view of the strong nonconvexity of the original problem, the optimization variables are split into two mutually iterative sub-problem blocks: joint resource allocation and UAV trajectory planning; S32. Based on the current fixed trajectory of the UAV, introduce the DC planning mechanism to solve the joint resource allocation scheme; S33. After implementing the fixed resource strategy, introduce non-negative slack variables. The objective is to refactor to maximize the SINR margin for all time slots and users in the system; S34. SINR margin-based precise optimization based on binary search method; S35. Check the objective function of adjacent rounds. The convergence status, if it meets the set tolerance. If the maximum number of iterations is reached, the algorithm stops and outputs the optimal UAV coordinates, optimal transmission power, onboard computing power allocation, and optimal semantic compression length after trade-offs for each time slot.

6. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access according to claim 5, characterized in that, Step S31 is as follows: The original highly coupled nonconvex problem is decomposed into two sub-problems: joint resource allocation. , and and drone trajectory planning ; Set outer alternating index By optimizing resources under a fixed trajectory and evolving the trajectory under fixed resources, a closed-loop mechanism is used to decouple multidimensional nonlinear coupled variables.

7. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access according to claim 6, characterized in that, Step S32 is as follows: Fix current trajectory Calculate the channel power gain constant. : ； Using the DC programming approach, the signal-to-interference-plus-noise ratio (SIR) expression is logarithmically transformed, converting multiplication and division relationships into addition and subtraction relationships; at the local power point... Linear lower bound for constructing interference terms This allows for the efficient solution of originally non-convex fractional constraints by reconstructing them into standard convex constraints.

8. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmentation multiple access according to claim 7, characterized in that, Step S33 is as follows: After implementing a fixed resource strategy, nonnegative slack variables are introduced. The objective is to refactor to maximize the SINR margin for all time slots and users in the system; To address the deeply nested range term in the power gain of the air-to-ground channel, a set of relaxation variables is introduced. , representing the squared distance of the target signal, the squared distance of the interfering signal, and the upper limit of the total interference power, respectively; The non-convex disturbance distance constraint is transformed into a linear lower bound constraint using a first-order Taylor expansion.

9. The UAV-assisted low-latency semantic transmission method based on orthogonal model segmented multiple access according to claim 8, characterized in that, Step S34 is as follows: Set SINR margin search interval In the given test value The trajectory subproblem is then transformed into a convex feasibility verification problem. Continuously update the Taylor expansion points to reduce the approximation error. If a feasible solution exists below, then update the lower bound. Otherwise update the upper bound. ; Through nested inner and outer layers, the final output is the optimal physical coordinate sequence that can overcome co-channel interference in multiple access segmentation by orthogonal models to the greatest extent. .

Images