Low-altitude intelligent system oriented to general sensing service and deployment method of low-altitude intelligent system

Abstract

The invention discloses a low-altitude intelligent system for a communication service and a deployment method thereof, and relates to the field of unmanned aerial vehicles. The system introduces an unmanned aerial vehicle as a low-altitude edge reasoning node, and completes receiving and reasoning processing of sensing data at a position close to communication equipment; the reasoning process is converted from far-end centralized processing to low-altitude edge nearby processing; according to the system, on the basis of integrated perception and communication, namely perception data acquired by ISAC equipment, the motion state of a target is rapidly recognized; the system operation process mainly comprises an ISAC stage and a calculation stage. The system can effectively reduce the data transmission time delay before reasoning and improve the reasoning response speed, thereby meeting the application requirement of the general sensing service on the real-time performance.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicles (UAVs), and more particularly to a low-altitude intelligent system for sensing services and its deployment method. Background Technology

[0002] In recent years, thanks to the continuous development of wireless communication and intelligent sensing technologies, integrated sensing and communication technologies have received widespread attention and application in scenarios such as environmental perception and target detection. ISAC, through shared spectrum resources and hardware architecture, achieves deep coupling between sensing and communication functions, enabling the system to efficiently complete wireless transmission of sensing data while acquiring environmental and target information. This improves the system's ability to perceive environmental conditions and enhances information interaction efficiency, providing crucial technical support for the application of low-altitude intelligent systems in smart cities, traffic monitoring, and other scenarios. However, limited by factors such as device size, power consumption, and hardware cost, existing sensing devices typically possess only limited computing power, making it difficult to complete complex cognitive reasoning and decision-making tasks locally. This can easily lead to insufficient reasoning capabilities and significant response latency. Therefore, sensing services often require transmitting sensing data back to base stations or remote data centers for centralized processing and reasoning analysis. However, in remote areas, disaster sites, or application scenarios with inadequate communication infrastructure, relying on fixed base stations or remote data centers for centralized reasoning still has significant limitations. Because the sensing data needs to be transmitted back over long distances via air-to-ground links, problems such as limited communication bandwidth, increased transmission latency, and insufficient link stability are likely to occur, making it difficult to meet the requirements of sensing services for low latency and high reliability, and to some extent limiting the overall inference capability of the system.

[0003] With the rapid development of the low-altitude economy, drones, with their advantages of flexible deployment and rapid response, possess the potential to support rapid inference in sensing services. By deploying drone platforms equipped with edge computing resources at low altitudes, inference tasks can be completed close to the sensing data source, effectively shortening data transmission distance, reducing communication latency, and improving the system's inference throughput. However, several challenges remain in drone-assisted inference scenarios. First, the drone's spatial location significantly impacts the quality of the air-to-ground communication link, directly determining the upload rate and transmission latency of the sensing data. Second, under constraints such as communication power, computing resources, and latency, limited communication performance further restricts the execution efficiency of inference tasks. Furthermore, the close coupling between drone location, sensing resources, and computing resources makes the system optimization problem non-convex, making it difficult to obtain optimal inference performance through simple rules or static strategies.

[0004] To address the aforementioned challenges, existing research has proposed various specific solutions for UAV-assisted edge inference scenarios. Some studies focus on UAV-assisted collaborative edge inference scenarios, where the UAV acts as an aerial edge server, aggregating multi-sensor features through aerial computation and completing inference tasks. This work uses discriminant gain as a task-oriented inference accuracy metric, jointly optimizing the UAV flight trajectory and sensor power allocation, and considering the importance of feature dimensions, thereby improving collaborative inference performance under communication-constrained conditions. Other research addresses the collaborative optimization of inference and perception functions under resource-constrained conditions for edge applications where UAVs simultaneously perform AI inference and target tracking. This work jointly optimizes the UAV trajectory and beamforming to minimize target tracking errors, achieving a balance between inference and perception performance. Still others propose an adaptive optimization method based on reinforcement learning for UAV-assisted edge inference scenarios in maritime environments. This work aims to minimize the weighted sum of energy consumption and inference latency, jointly optimizing neural network segmentation points, UAV transmission power, and the selection of collaborative edge servers to achieve a synergistic improvement in inference performance and energy efficiency in complex environments.

[0005] While the aforementioned studies have explored UAV-assisted edge inference systems in specific application scenarios, certain limitations remain. On the one hand, existing work primarily focuses on optimizing single indicators such as inference accuracy, perception performance, or energy efficiency, making it difficult to improve overall inference capabilities at the system level. On the other hand, the perception, communication, and computation processes lack collaborative modeling and unified optimization, failing to consider the coupling relationships among the three within a unified framework.

[0006] To address this, this invention proposes a low-altitude intelligent system for sensing services, comprising an ISAC (Intelligent Information Capability) phase and an edge computing phase. In the ISAC phase, ground-based ISAC devices transmit wireless signals to the target and receive echoes to complete the sensing task, while simultaneously transmitting the acquired sensing signals to a UAV (Unmanned Aerial Vehicle). In the edge computing phase, the UAV samples, denoises, and converts the received sensing signals into a spectrogram, which is then input into a deep neural network model for inference, thereby identifying the target's motion state. Furthermore, this invention proposes a flexible UAV deployment method that optimizes UAV positioning to improve the overall inference throughput of the system. Summary of the Invention

[0007] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is how to meet the real-time application requirements of sensing services.

[0008] To achieve the above objectives, this invention provides a low-altitude intelligent system for sensing services. The system is characterized by introducing a UAV as a low-altitude edge inference node, enabling the reception and inference processing of sensing data near the sensing device, thus transforming the inference process from centralized processing at a remote location to near-field processing at the low-altitude edge. Based on integrated sensing and communication (ISAC) devices, the system rapidly identifies the motion state of targets. The system operation mainly includes an ISAC phase and a computation phase.

[0009] Furthermore, in the ISAC phase, the ISAC device transmits a wireless signal to the target to complete the detection task, and simultaneously transmits the received echo signal to the UAV, which is only responsible for receiving it; the ISAC phase adopts a periodic frame structure based on radar pulse repetition interval (PRI); the duration of each PRI cycle is... It is divided into a sensing phase and a communication phase; each of the ISAC devices collects... Each sample consists of M frequency modulated continuous wave (FMCW) signals.

[0010] Furthermore, in the perception phase, the system transmits FMCW signals and receives target echoes. Each FMCW signal is linearly frequency-scanned in time to extract the target's motion features. In the communication phase, the echo signals are converted from analog to digital to generate complex samples, which are then serialized into a bit stream and transmitted to the UAV in real time via a communication link.

[0011] Furthermore, during the computation phase, after the UAV receives the signal from the ISAC device, it first samples and denoises it, and then converts it into a spectrogram through a short-time Fourier transform. Subsequently, the generated spectrogram is input into a deep neural network model deployed locally on the UAV for inference. The model can automatically extract the motion features of the target and complete the classification, thereby achieving accurate identification of the target's motion state.

[0012] Furthermore, the goal of the low-altitude intelligent system is to maximize the overall inference throughput, and the problem is modeled as follows:

[0013] C1 ensures that the communication rate meets the data upload requirements for each frame, ensuring the transmission of sensed data; C2 ensures that the total energy consumption of the ISAC device does not exceed its energy limit; C3 ensures that the total time to complete sensing, communication and inference does not exceed the maximum allowable delay. From constraint C2, we can obtain

[0014] From constraint C3, we can obtain

[0015] Therefore, to simultaneously satisfy both energy and latency constraints, the number of feasible inference samples for each device is:

[0016] Based on this, the optimization problem can be written as:

[0017] This problem can be further transformed into:

[0018] In this refactoring problem, the device set is based on and The size relationship is divided into two mutually exclusive subsets: Equipment with stricter energy constraints; Devices with stricter latency constraints.

[0019] A method for deploying a low-altitude intelligent system for sensing services is characterized by maximizing the number of inference samples that the system can support under resource constraints. At the same time, in order to reduce the computational complexity introduced by enumerating device constraint types, a device constraint pre-classification strategy for UAV deployment is further proposed.

[0020] Furthermore, the method includes the following steps: Step 1: Calculate the upper bound constant of the delay constraint for all devices. Based on its latency constraints and computing resources, calculate the maximum number of inference samples that can be supported under the latency constraints. ; Step 2: Equipment pre-classification. Based on the limitation relationship between each device and the number of feasible inference samples under energy constraints and time delay constraints, ISAC devices are divided into energy-constrained devices and time delay-constrained devices. Step 3: Enumerate flexible devices in groups. For a set of flexible devices... An enumeration strategy is adopted to consider these devices in and All possible allocations; since some devices have been pre-classified as energy-limited and must belong to Any allocation of these devices to All possible solutions were ignored, thus effectively narrowing the search space; therefore, only one solution needs to be evaluated. There are far fewer grouping schemes than a complete enumeration. kind; Step 4: Optimal Solution Selection. Finally, select the group that maximizes the overall system objective value and output the corresponding UAV positions. Total target value and the optimal number of inference samples for each device .

[0021] Furthermore, in step 1, the upper bound is related to the drone's position. This is irrelevant and is used for comparative analysis in the subsequent equipment constraint pre-classification process.

[0022] Furthermore, the upper bound of the delay constraint for each device is obtained. Then, step 2 analyzes the upper bound of the energy constraint. Within the feasible location space of the UAV and The relationship is used to determine the constraint type of each device; if the maximum value of a certain device is... Always less than or equal to If its performance is limited by energy constraints, the device is classified as an energy-constrained device and included in the set. For the remaining equipment, its and The relationships may vary depending on the location of the UAV; these devices are energy-constrained in some areas and time-delay-constrained in others; such devices constitute a flexible equipment ensemble. The specific classification will be determined in the subsequent group enumeration.

[0023] Furthermore, in step 3 of each enumeration scheme, Includes pre-determined energy-constrained equipment and the current plan to allocate Flexible equipment, and This includes flexible devices assigned to this set.

[0024] Furthermore, step 3 also includes: Step 3-1: After obtaining the candidate grouping schemes, for each candidate group, fix the set of energy-constrained devices. And establish corresponding sub-problems This subproblem only involves The objective function and constraints of the device are designed to optimize the UAV position. and the communication rates of these devices If a feasible solution exists, it is denoted as... Subproblems The specific form is as follows:

[0025] in, To facilitate the solution, auxiliary variables are introduced. In addition, two auxiliary variables are introduced. and This is used to linearize the nonconvex terms related to the distance from the device to the UAV; based on these substitutions, the original objective function and constraints can be equivalently rewritten in a more easily solvable form:

[0026] Non-convex constraints C4', C7, and C10 are represented as the difference between two convex functions, and are therefore inherently non-convex. Therefore, the Continuous Convex Approximation (SCA) method is adopted to transform the original non-convex constraints into convex forms by performing a first-order Taylor expansion on the second convex function in each constraint, so that the overall problem can be solved efficiently using standard convex optimization tools (such as CVX). Step 3-2: Obtain the optimal UAV position Afterwards, verification Do all devices in the system meet the following constraints:

[0027] If all constraints are satisfied, the current group is considered a feasible group, and the total objective value for that group is calculated; the total objective value Corresponding to and The sum of inference samples from all devices in the middle

[0028] The present invention has the following technical effects: (1) After adopting the proposed low-altitude intelligent system, in application scenarios where base station coverage is sparse or communication conditions are limited, the system can effectively reduce the data transmission delay before inference and improve the inference response speed, thereby meeting the real-time application requirements of the sensing service.

[0029] (2) The proposed flexible UAV deployment method can obtain the optimal UAV location and the corresponding maximum total number of system inference samples, thereby improving the overall inference capability of the system. At the same time, the introduction of the device constraint pre-classification strategy can significantly reduce the group enumeration scale while ensuring optimality, effectively improving the solution efficiency of UAV deployment optimization.

[0030] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0031] Figure 1 This is a system diagram of a preferred embodiment of the present invention; Figure 2 This is a flowchart of an algorithm according to a preferred embodiment of the present invention. Detailed Implementation

[0032] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0033] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0034] Figure 1 This invention demonstrates a low-altitude intelligent system for sensing services proposed in this paper. The invention considers UAV-assisted sensing scenarios and rapidly identifies the motion state (e.g., stationary or moving) of a target based on sensing data acquired by the ISAC device. The system operation mainly includes an ISAC phase and a computation phase.

[0035] The first phase is the ISAC phase. The ISAC device transmits a radio signal to the target to complete the detection task, and simultaneously transmits the received echo signal to the UAV, which is only responsible for receiving it. The ISAC phase uses a periodic frame structure based on the radar pulse repetition interval (PRI). The duration of each PRI cycle is... It is divided into a sensing phase and a communication phase. Each ISAC device collects... Each sample consists of M frequency-modulated continuous wave (FMCW) signals. During the sensing phase, the system transmits FMCW signals and receives target echoes. Each FMCW signal undergoes a linear frequency scan in time to extract the target's motion characteristics. During the communication phase, the echo signals are converted from analog to digital to generate complex samples, which are then serialized into a bit stream and transmitted to the UAV in real time via the communication link.

[0036] The second stage is the computation stage. After receiving the signal from the ISAC device, the UAV first samples and denoises it, then converts it into a spectrogram using a short-time Fourier transform. Subsequently, the generated spectrogram is input into a locally deployed deep neural network model on the UAV for inference. The model can automatically extract the target's motion features and perform classification, thereby achieving accurate identification of the target's motion state (such as standing, walking, running, etc.). The detailed flow of this computation process is as follows: Figure 1 As shown.

[0037] The system's goal is to maximize overall inference throughput. The problem is modeled as follows:

[0038] Specifically, C1 ensures that the communication rate meets the data upload requirements for each frame, guaranteeing the transmission of sensed data. C2 ensures that the total power consumption of the ISAC device does not exceed its power limit. C3 ensures that the total time to complete sensing, communication, and inference does not exceed the maximum allowable delay.

[0039] From constraint C2, we can obtain

[0040] From constraint C3, we can obtain

[0041] Therefore, to simultaneously satisfy both energy and latency constraints, the number of feasible inference samples for each device is:

[0042] Based on this, the optimization problem can be written as:

[0043] This problem can be further transformed into:

[0044] In this refactoring problem, the device set is based on and The size relationship is divided into two mutually exclusive subsets: Equipment with stricter energy constraints.

[0045] Devices with stricter latency constraints.

[0046] However, when the number of devices in the system is large, the total number of possible groups grows exponentially, that is, K devices have a total of There are several possible groupings. Solve for each grouping. Subproblems will result in extremely high computational overhead.

[0047] To reduce computational complexity, a device constraint pre-classification strategy for UAV deployment is proposed. See details below. Figure 2 The core idea of ​​this strategy is to narrow the search space through theoretical analysis and pre-classification while maintaining the integrity of the search, thereby significantly improving computational efficiency without sacrificing optimality. The steps of the proposed method include: Step 1: Calculate the upper bound constant of the delay constraint: for all devices Based on its latency constraints and computing resources, calculate the maximum number of inference samples that can be supported under the latency constraints. The upper bound is related to the drone's location. It is irrelevant and is used for comparative analysis in the subsequent equipment constraint pre-classification process; Step 2: Device Pre-classification: Based on the constraints on the number of feasible inference samples for each device under energy and time delay constraints, ISAC devices are classified into energy-constrained devices and time-delay-constrained devices. The upper bound of the time delay constraint for each device is obtained. Then, the upper bound of the energy constraint is analyzed. Within the feasible location space of the UAV and The relationship is used to determine the constraint type for each device. If the maximum value of a certain device is... Always less than or equal to If its performance is limited by energy constraints, the device is classified as an energy-constrained device and included in the set. For the remaining equipment, its and The relationship may vary depending on the location of the UAV. These devices are energy-constrained in some regions and time-delay-constrained in others. Such devices constitute a flexible equipment ensemble. The specific classification will be determined in the subsequent grouping enumeration. This pre-classification step significantly reduces the search space and computational burden in the subsequent grouping enumeration stage; Step 3: Flexible device grouping enumeration: For the flexible device set An enumeration strategy is adopted to consider these devices in and All possible allocations. Because some devices have been pre-classified as energy-limited and must belong to... Any allocation of these devices to All possible solutions were ignored, thus effectively narrowing the search space. Therefore, only one solution needs to be evaluated. There are far fewer grouping schemes than a complete enumeration. In each enumeration scheme, Includes pre-determined energy-constrained equipment and the current plan to allocate Flexible equipment, and This includes the flexible devices assigned to this set. The process generates all possible grouping schemes, paving the way for solving subproblems under fixed groupings. Explicit input was provided; Step 3-1: Fixed Grouping Subproblems Solution: After obtaining the candidate grouping schemes, for each candidate group, fix the set of energy-constrained devices. And establish corresponding sub-problems This subproblem only involves The objective function and constraints of the device are designed to optimize the UAV position. and the communication rates of these devices If a feasible solution exists, it is denoted as... Subproblems The specific form is as follows:

[0048] in, To facilitate the solution, auxiliary variables are introduced. In addition, two auxiliary variables are introduced. and This is used to linearize the nonconvex terms related to the distance from the device to the UAV. Based on these substitutions, the original objective function and constraints can be equivalently rewritten in a more easily solvable form:

[0049] Non-convex constraints C4', C7, and C10 are represented as the difference between two convex functions, and are therefore inherently non-convex. Therefore, we employ the Continuous Convex Approximation (SCA) method, which transforms the original non-convex constraints into convex forms by performing a first-order Taylor expansion on the second convex function in each constraint. This allows the overall problem to be solved efficiently using standard convex optimization tools (such as CVX).

[0050] Step 3-2: Feasibility Verification: Obtaining the Optimal UAV Location Afterwards, verification Do all devices in the system meet the following constraints:

[0051] If all constraints are satisfied, the current group is considered feasible, and the overall objective value for that group is calculated. Corresponding to and The sum of inference samples from all devices in the system.

[0052] Step 4: Optimal Solution Selection: Finally, select the group that maximizes the overall system objective value and output the corresponding UAV positions. Total target value and the optimal number of inference samples for each device .

[0053] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A low-altitude intelligent system for sensing services, characterized in that, This system introduces UAVs as low-altitude edge inference nodes to receive and infer sensing data near the sensing devices, transforming the inference process from centralized processing at a remote location to local processing at the low-altitude edge. Based on integrated sensing and communication (ISAC) devices, the system rapidly identifies the motion state of targets. The system operation mainly includes the ISAC phase and the computation phase.

2. The low-altitude intelligent system for sensing services as described in claim 1, characterized in that, In the ISAC phase, the ISAC device transmits a wireless signal to the target to complete the detection task, and simultaneously transmits the received echo signal to the UAV, which is only responsible for receiving it; the ISAC phase adopts a periodic frame structure based on radar pulse repetition interval (PRI); the duration of each PRI cycle is... It is divided into a sensing phase and a communication phase; each of the ISAC devices collects... Each sample consists of M frequency modulated continuous wave (FMCW) signals.

3. The low-altitude intelligent system for sensing services as described in claim 2, characterized in that, During the perception phase, the system transmits FMCW signals and receives target echoes. Each FMCW signal is linearly scanned in time to extract the target's motion features. During the communication phase, the echo signal undergoes analog-to-digital conversion to generate complex samples, which are then serialized into a bit stream and transmitted to the UAV in real time via the communication link.

4. The low-altitude intelligent system for sensing services as described in claim 3, characterized in that, During the computation phase, after receiving the signal from the ISAC device, the UAV first samples and denoises it, and then converts it into a spectrogram through a short-time Fourier transform. Subsequently, the generated spectrogram is input into a deep neural network model deployed locally on the UAV for inference. The model can automatically extract the target's motion features and complete the classification, thereby achieving accurate identification of the target's motion state.

5. The low-altitude intelligent system for sensing services as described in claim 4, characterized in that, The objective of the low-altitude intelligent system is to maximize the overall inference throughput, and the problem is modeled as follows: By jointly optimizing the communication rate of the UAV's position q and the ISAC device Maximize the number of inference samples for each ISAC device The sum; C1 represents the set of all ISAC devices in the system, where k is the device index; C1 ensures that the communication rate meets the data upload requirements for each frame, thus ensuring the transmission of sensing data. This indicates the duration of the communication phase within a single radar pulse repetition interval period. C2 represents the amount of sensing data generated by each radar pulse; C2 ensures the total energy consumption of the ISAC device. Not exceeding its maximum available energy budget C3 guarantees the total latency for completing sensing, communication, and inference. The maximum inference latency allowed by the k-th device shall not exceed ; From constraint C2, we can obtain Where M is the number of radar pulses required for a single inference sample. This refers to the duration of the sensing phase within a single radar pulse repetition interval period. , and These represent the sensing transmission power, receiving power, and communication transmission power of the k-th ISAC device, respectively. Indicates the location of the drone Given the number of possible inference samples determined by energy constraints; From constraint C3, we can obtain in, Indicates the duration of the radar pulse repetition interval (PRI). This represents the number of floating-point operations required to process a single perceptual sample. This represents the computing power allocated to the k-th ISAC device, i.e., the number of floating-point operations that can be performed per second; based on this, This represents the upper bound of the maximum number of inference samples that the k-th ISAC device can support, as determined by the latency constraint. Therefore, to simultaneously satisfy both energy and latency constraints, the number of feasible inference samples for each device is: Based on this, the optimization problem can be written as: in, B represents the equivalent communication channel gain parameter of the k-th ISAC device, and B represents the system bandwidth. This represents the distance between the k-th ISAC device and the drone; the problem can be further transformed into: In this refactoring problem, the device set is based on and The size relationship is divided into two mutually exclusive subsets: Equipment with stricter energy constraints; Devices with stricter latency constraints.

6. A method for deploying a low-altitude intelligent system for sensing services, characterized in that, This method maximizes the number of inference samples that the system can support under resource constraints. At the same time, in order to reduce the computational complexity introduced by enumerating device constraint types, a device constraint pre-classification strategy for UAV deployment is further proposed.

7. The method for deploying a low-altitude intelligent system for sensing services as described in claim 5, characterized in that, The method includes the following steps: Step 1: Calculate the upper bound constant of the delay constraint for all devices. Based on its latency constraints and computing resources, calculate the maximum number of inference samples that can be supported under the latency constraints. ; Step 2: Equipment pre-classification. Based on the limitation relationship between each device and the number of feasible inference samples under energy constraints and time delay constraints, ISAC devices are divided into energy-constrained devices and time delay-constrained devices. Step 3: Enumerate flexible devices in groups. For a set of flexible devices... An enumeration strategy is adopted to consider these devices in and All possible allocations; since some devices have been pre-classified as energy-limited and must belong to Any allocation of these devices to All possible solutions were ignored, thus effectively narrowing the search space; therefore, only one solution needs to be evaluated. There are far fewer grouping schemes than a complete enumeration. kind; Step 4: Optimal Solution Selection. Finally, select the group that maximizes the overall system objective value and output the corresponding UAV positions. Total target value and the optimal number of inference samples for each device .

8. The method for deploying a low-altitude intelligent system for sensing services as described in claim 6, characterized in that, In step 1, the upper bound is related to the position of the drone. This is irrelevant and is used for comparative analysis in the subsequent equipment constraint pre-classification process.

9. The method for deploying a low-altitude intelligent system for sensing services as described in claim 7, characterized in that, Obtain the upper bound of the delay constraint for each device. Then, step 2 analyzes the upper bound of the energy constraint. Within the feasible location space of the UAV and The relationship is used to determine the constraint type for each device; If the maximum of a certain device Always less than or equal to If its performance is limited by energy constraints, the device is classified as an energy-constrained device and included in the set. For the remaining equipment, its and The relationships may vary depending on the location of the UAV; these devices are energy-constrained in some areas and time-delay-constrained in others; such devices constitute a flexible equipment ensemble. The specific classification will be determined in the subsequent group enumeration.

10. The method for deploying a low-altitude intelligent system for sensing services as described in claim 8, characterized in that, In each enumeration scheme, step 3 is described in... Includes pre-determined energy-constrained equipment and the current plan to allocate Flexible equipment, and This includes the flexible devices assigned to this set; Step 3 also includes: Step 3-1: After obtaining the candidate grouping schemes, for each candidate group, fix the set of energy-constrained devices. And establish corresponding sub-problems This subproblem only involves The objective function and constraints of the device are designed to optimize the UAV position. and the communication rates of these devices If a feasible solution exists, it is denoted as... Subproblems The specific form is as follows: in, To facilitate the solution, auxiliary variables are introduced. In addition, two auxiliary variables are introduced. and This is used to linearize the nonconvex terms related to the distance from the device to the UAV; based on these substitutions, the original objective function and constraints can be equivalently rewritten in a more easily solvable form: Let C be the position of the k-th ISAC device; the non-convex constraints C4', C7, and C10 are represented as the difference between two convex functions, and are therefore inherently non-convex; therefore, the continuous convex approximation SCA method is adopted, and the original non-convex constraints are transformed into convex forms by performing a first-order Taylor expansion on the second convex function in each constraint, so that the overall problem can be solved efficiently using standard convex optimization tools. Step 3-2: Obtain the optimal UAV position Afterwards, verification Do all devices in the system meet the following constraints? If all constraints are satisfied, the current group is considered a feasible group, and the total objective value for that group is calculated; the total objective value Corresponding to and The sum of inference samples from all devices in the system.

Images