A method for sensing and integration network resource management and control oriented to edge inference

Abstract

The application discloses a method for managing and controlling network resources of a sensing-integrated network for edge inference, and comprises the following steps: based on the sensing-integrated network, a multi-user multi-target ISAC system is constructed by using frequency division multiple access technology; the influence of the transmission power of a base station, the communication rate of a user, the sensing signal-to-noise ratio of a sensing target, the edge example of an edge server and the memory requirement of the edge server on a target sensing task in the multi-user multi-target ISAC system is analyzed, and a maximum sensing target access quantity problem model is constructed; a double-layer iterative resource management and control method is used to solve the maximum sensing target access quantity problem model, so that the maximum sensing target access quantity and the determination of the sensing target are realized. The application can effectively coordinate the resource allocation between the communication and sensing tasks, and meet the edge inference requirement of the sensing task of the sensing-integrated network.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically, proposes a method for managing and controlling network resources in a sensory-integrated manner for edge reasoning. Background Technology

[0002] With the rapid evolution of mobile communication technology, sixth-generation wireless networks are considered a major driver for many emerging applications, such as smart manufacturing, smart homes, and autonomous driving. These scenarios not only require higher communication rates and wider connection scales, but also place higher demands on capabilities such as spatial perception, positioning accuracy, and environmental understanding. To meet the complex needs of these scenarios, Integrated Sensing and Communication (ISAC) technology has emerged, attracting widespread research interest.

[0003] With the widespread application of ISAC in multi-user, multi-target scenarios, the scale and processing complexity of the sensed data generated by the system are growing exponentially. Traditional sensed signal processing and optimization methods are gradually becoming insufficient to meet the demands for real-time performance, intelligence, and scalability. To overcome these limitations, Artificial Intelligence (AI) has begun to be introduced to further improve ISAC systems, enhancing their perception and resource scheduling capabilities in complex environments. AI can assist in environmental semantic understanding, intelligent beamforming design, resource scheduling optimization, and cross-modal task processing, enabling ISAC systems to dynamically adapt, autonomously optimize, and collaborate across scenarios, thereby meeting the intelligent requirements of future networks. While the introduction of AI has improved the intelligent processing capabilities of ISAC, it has also significantly increased the system's computational and communication burden. Terminal devices, limited by computing power, energy consumption, and storage, struggle to handle complete model inference; while uploading all sensed data to the cloud can provide powerful computing power, it faces risks of privacy leaks and latency issues caused by long-distance transmission, similarly failing to meet real-time inference requirements. Therefore, edge reasoning is gradually emerging as a new paradigm of intelligent services. Edge reasoning can provide a more reasonable division of labor among devices, networks and computing resources, enabling the edge side to undertake complex reasoning, while the terminal device only needs to perform lightweight preprocessing or feature extraction, thereby significantly alleviating its computing and storage burden.

[0004] However, in multi-target scenarios, edge nodes have limited computing power and storage resources, and the system performance cannot meet the requirements of inference tasks that process all targets simultaneously. To make full use of limited computing and storage resources, the system's processing efficiency and resource utilization can be improved by sorting and scheduling the target tasks based on their requirements, data volume, and priorities. Furthermore, in multi-user, multi-target scenarios, different targets have different spatial locations, signal strengths, and task requirements. Beamforming optimization can achieve coordinated scheduling of communication resources and sensing tasks, improving the overall system performance under limited resource conditions. Summary of the Invention

[0005] In response to the aforementioned technical problems, this invention provides a method for managing and controlling integrated sensory networks for edge inference, comprising the following steps:

[0006] S1: Based on the integrated sensing network, a multi-user, multi-target ISAC system is constructed using frequency division multiple access technology; S2: In a multi-user, multi-target ISAC system, the impact of base station transmit power, user communication rate, sensing signal-to-noise ratio of the sensing target, edge computing power of the edge server, and memory requirements of the edge server on the target sensing task is analyzed, and a model for maximizing the number of sensing targets accessed is constructed. S3: A two-layer iterative resource management method is adopted to solve the problem model of maximizing the number of sensing targets connected, thereby maximizing the number of sensing targets connected and determining the sensing targets.

[0007] Furthermore, the multi-user, multi-objective ISAC system includes: Base station transmission signal model: used to construct the transmission signals for joint communication and sensing at base stations; User received signal model: used to construct the base station transmitted signal sent by the base station transmitted signal model for communication users to evaluate communication performance; Radar sensing model: used to obtain task information of each sensing target based on the echo signal of the base station transmitted signal after being reflected by the sensing target, which is based on the user received signal model; Sampling quantization model: used to quantize and sample the sensing echo signal sent by the radar sensing model to generate discrete data for each sensing target; Edge computing power model: used to receive the sensing data quantized by the sampling quantization model and calculate the resource requirements for edge computing power and storage.

[0008] Furthermore, the sampling quantization model is constructed as follows: Multi-user, multi-objective ISAC systems include C One communication user group and M The first goal, of which the second cThe user group includes K c A communication user, in order to describe the target m Whether it is served by BS, introduce binary variables ,when When, it indicates the target. m Being served by BS, making Access the target set of binary variables; exist Within, based on the sampling frequency Sampling was performed, and the number of samples was [number missing]. For the first m The sensed echo signal of each target, and the signal sampled at the base station receiving antenna, are represented as follows:

[0009] in, Indicates the first m The complex reflection coefficient of a target, including path loss and scattering characteristics. From BS to target m The time delay, It is the target response matrix of target m. Antenna spacing, and The number of transmit and receive antennas, and They are respectively Directional transmission and reception steering vectors. The mean is zero and the variance is... Complex Gaussian white noise; Using transmitted signals The received echo signal is subjected to matched filtering, i.e.:

[0010] in, , Representing user groups c The Middle k The communication beamforming vector of the first user and the first m Perceived beamforming vector of a target; The quantized sampled signal is:

[0011] Introducing a receiving filter The process of processing the vectorized echo signal is as follows:

[0012] The perceived signal-to-noise ratio is:

[0013] The real and imaginary parts of the sampled signal are respectively treated using q -bits quantization, then the quantized digits are... m The sensed echo signals of the target are:

[0014] in, yes q -bits quantizer, then the 1st m The total number of bits of data in the echo signal of each sensing target is: .

[0015] Furthermore, the expression for maximizing the number of sensing targets accessed is as follows:

[0016] Wherein, constraint C1 indicates that the base station's transmit power does not exceed its upper limit. ; Constraint C2 indicates the user group c Chinese users k The communication rate is not lower than its lower limit. ; Constraint C3 represents the objective. m The perceived signal-to-noise ratio is not lower than its lower limit. ; Constraints C4 and C5 represent the maximum computing and storage resources of the base station. and ; Constraint C6 indicates that the target access variable is a binary variable.

[0017] Furthermore, the process of using a two-layer iterative resource management method to solve the problem model of maximizing the number of sensing targets accessed, and to determine the sensing targets, includes the following steps: In the outer layer, the sensing targets are traversed in ascending order according to computing power and storage requirements. Using a greedy scheduling algorithm for edge computing and storage resources, the sensing targets that can be accessed in the set of sensing targets are initially determined. Based on a defined set of sensing targets, the inner layer uses a successive convex approximation algorithm to optimize the beamforming transmitted by the base station to communication users and sensing targets, thereby ensuring the service performance of communication users and sensing targets. Finally, the feasibility of the constraint is verified based on the inner optimization results to determine whether to accept the current objective. By combining the outer greedy construction, the inner convex optimization and the result verification decision, a feasible solution is obtained efficiently under the constraint of limited resources.

[0018] Furthermore, the process of optimizing the beamforming emitted by the base station for communication users and sensing targets based on the successive convex approximation algorithm is as follows: S31, Given a set of perception target service variables Next, solve the communication beamforming and sensing beamforming matrices. It is a problem of finding a feasible solution:

[0019] S32. First, transform constraint C2 to obtain the signal-to-interference-plus-noise ratio (SINR) form, i.e. ,make The lower limit of the signal-to-interference-plus-noise ratio is represented by: (Expanded, we get:)

[0020] Right now This function is a difference convex function. Wherein, Indicates BS and the c The first in the user group k Channel matrix between user equipment For noise; Then, let , Transforming constraint C3, we get:

[0021] Right now ; S33. The successive convex approximation method is used to iteratively transform the difference convex constraint into a convex function to solve the problem. The lower left bound is found through a first-order Taylor expansion, and the process is iteratively processed. For the successive convex approximation... o In the next iteration, consider the following lower bound:

[0022]

[0023] Substituting the above formula, we get the first... o The problem expression for the next iteration is: :

[0024] Therefore, the problem The non-convex constraint in the equation is approximated as a convex constraint, and the convex constraint is iterated. O This implementation uses the MOSEK solver from the General CVX Optimization Toolkit for solving.

[0025] Furthermore, the process of initially determining the sensing targets in the access set of sensing targets by utilizing the greedy scheduling algorithm of edge computing power and storage resources is as follows: In the process of gradually constructing the set of sensing target services, the optimization problem of the sensing target access set is modeled as a 0-1 integer programming problem:

[0026] Solve the problem Based on a greedy strategy, the goal is to maximize the number of base station service targets under resource constraints. The specific process is as follows: (1) Calculate the computing power requirement for each sensing target. and h occupancy requirements Sort by resource demand in ascending order; (2) Consider the objectives one by one in the sorted order: For the current objective m First, determine whether its computing power and storage requirements exceed the remaining resources of the edge node. If the resources are sufficient, add it to the perception target service set and perform inner-layer beamforming optimization.

[0027] This invention presents a method for managing resources in an integrated sensing network oriented towards edge inference. Under an integrated sensing network architecture, based on frequency division multiple access (FDMA) technology, it constructs a model encompassing base station transmitted signal models, user received signal models, radar sensing echo models, sampling and quantization models, and edge computing power. By analyzing the constraints of multiple factors on sensing task performance, such as user transmit power, user communication rate, sensing signal-to-noise ratio of the sensing target, edge computing instances of the edge server, and edge server memory requirements, it jointly optimizes communication and sensing beamforming with the set of sensing target accesses, establishing a problem to maximize the number of sensing target accesses. To further solve this mixed-integer nonlinear programming problem, a two-layer iterative resource management method is designed. The outer layer utilizes a greedy scheduling algorithm for edge computing and storage resources to optimize the set of sensing target accesses, while the inner layer utilizes a successive convex approximation algorithm to optimize communication and sensing beamforming, thereby achieving coordinated optimization of communication, sensing, and computing resources. Compared with existing technologies, this invention has the following advantages: This invention provides a method for managing network resources in a sensor-integrated network for edge inference. In wireless edge networks with limited computing and storage resources, this method enables multiple-input multiple-output (MIMO) base station systems equipped with large language models to efficiently and reliably complete multi-target perception and intelligent processing tasks while ensuring multi-user communication performance. This effectively improves the perception service capabilities and resource utilization efficiency of sensor-integrated computing systems.

[0028] Based on the above reasons, this invention can be widely applied in fields such as wireless communication. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a network scene diagram used in embodiments of the present invention; Figure 3 Location deployment diagram provided for embodiments of the present invention; Figure 4 The convergence graph of the successive convex approximation algorithm; Figure 5 A graph showing the relationship between the successful access rate and the number of targets under different sampling numbers; Figure 6 A cumulative distribution map of user communication rates under different numbers of targets; Figure 7 This is a graph showing the variation in average communication rate for each user group under different transmit power constraints. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] Figure 1 This is a flowchart of the method of the present invention.

[0034] like Figure 1 As shown, this embodiment of the invention provides a method for managing and controlling integrated sensory network resources for edge inference, comprising the following steps: S1: Based on the integrated sensing network, a multi-user, multi-target ISAC system is constructed using frequency division multiple access technology; S2: In a multi-user, multi-target ISAC system, the impact of base station transmit power, user communication rate, sensing signal-to-noise ratio of the sensing target, edge computing power of the edge server, and memory requirements of the edge server on the target sensing task is analyzed, and a model for maximizing the number of sensing targets accessed is constructed. S3: A two-layer iterative resource management method is used to solve the problem model of maximizing the number of sensing targets connected, thereby maximizing the number of sensing targets connected and determining the sensing targets. The above steps are executed in the order of S1 / S2 / S3; Figure 2 This is a scenario diagram used in an embodiment of the present invention.

[0035] Consider a sensing-integrated system consisting of a MIMO base station, multiple communication user groups, and multiple sensing targets. The base station synchronously serves downlink communication users and sensing targets by transmitting sensing-integrated signals. The echo signals generated by the sensing targets are received by the base station and then sampled and quantized into an edge large-scale language model for task inference. However, due to limitations in computing power and storage resources, edge servers cannot process all sensing task requests simultaneously. Therefore, this paper proposes a resource scheduling method for edge inference tasks, enabling intelligent acceptance and scheduling of sensing tasks.

[0036] The multi-user, multi-objective ISAC system includes: Base station transmitted signal model: used to determine the signals transmitted by the base station; User signal receiving model: used to receive signals transmitted by the base station transmission signal model and to allocate signals to multiple users; Radar sensing model: used to receive signals transmitted by the base station signal transmission model, allocate multiple sensing targets, and generate sensing signal echoes; Sampling and quantization model: used to receive the sensing signal echo transmitted by the radar sensing model and perform sampling and quantization; Edge computing power model: used to receive the quantized sensing echo signal transmitted by the sampling and quantization model and to calculate the edge computing power.

[0037] The process of establishing a base station sensing transmission signal model using frequency division multiple access technology is as follows: A multi-user multi-target ISAC system includes C communication user areas and M sensing targets, where the first... c Each user area contains One communication user. Consider a duration. Within, frequency division multiple access (FDMA) technology is used to divide the total bandwidth evenly. Each of the three non-overlapping sub-bands is independently allocated to a group of communication users. and a perceived target ISAC group , For user groups c middle The data symbols of each user have beamforming vectors of... , For the first m The sensing signal of each target has a beamforming vector of... Assuming signal and They are independent of each other, therefore the base stations are in time t The transmitted signal is:

[0038] in, Let be a binary variable, representing the target m Whether it is served by a base station When it is active, it indicates that the system is being served by the base station and incurs computing and storage costs; otherwise... .make Access the target set of binary variables.

[0039] The user-received signal model is as follows: Assuming the frequency band is pre-allocated to the ISAC group, therefore the user group c The first in k The signal received by a user equipment in a sub-band can be represented as:

[0040] in, It is a base station on a sub-band and the first c The first in the user group k Channel matrix between user equipment This indicates that the mean is zero and the covariance is... White noise with a complex Gaussian distribution.

[0041] No. k The channel matrix for each user is a Ricean channel composed of complex line-of-sight link components and non-line-of-sight link components:

[0042] in, It is the first k Rice factor for each user.

[0043] No. kThe signal-to-interference-plus-noise ratio for each user is:

[0044] No. k Individual user communication and rate:

[0045] The radar perception model is as follows: Perceive target m The target response matrix is:

[0046] Transmit and receive steering vectors are and , and These represent the number of transmitting and receiving antennas, respectively. It is the carrier wavelength, and d Fixed as Half of This indicates the angle of the target relative to the base station.

[0047] Assuming that during this ISAC process, all point targets are either stationary or moving slowly, the base station receives the first... m The echo signal of each sensed target is:

[0048] in, Indicates the first m The complex reflection coefficient of a sensing target, including path loss and scattering characteristics. From BS to target m The time delay, The mean is zero and the variance is... The complex Gaussian white noise. Assume that multiple time delays can be perfectly compensated by synchronization.

[0049] The sampling quantization model is as follows: exist The internal system uses a sampling frequency of Sampling was performed, and the number of samples was [number missing]. Ignoring the mismatch in the steering vector, for the first... m The sensed echo signal of each target, and the signal sampled at the base station receiving antenna, are represented as follows:

[0050] To improve the output signal-to-noise ratio, the transmitted signal is used. The received echo signal is subjected to matched filtering, i.e.:

[0051] in, .

[0052] The quantized sampled signal is

[0053] Introducing a receiving filter The vectorized echo signal is then processed.

[0054] The perceived signal-to-noise ratio is

[0055] The real and imaginary parts of the sampled signal are respectively treated using q -bits quantization, then the quantized digits are... m The sensed echo signals of the target are:

[0056] in, yes q -bits quantizer. Then the... m The total number of bits of data in the echo signal of each sensing target is:

[0057] The edge computing power model is as follows: The quantized sensing echo signals are grouped by sample, each... q Each bit sample is grouped into a single token, so at this point, the first bit sample... m The length of the input sequence of the target sensing echo signal is:

[0058] Deploying a generative large language model on an edge device, consisting of stacked transformer decoding layers, including an initialization phase and an autoregressive phase, with each transformer decoding layer having the same structure, therefore... The computational requirements of the transformer decoding layer are

[0059] in, This indicates the computational requirements in the initial stage. This indicates the computational requirements during the autoregressive phase. d m This represents the hidden dimension of the transformer. n h This indicates that attention should be paid to the number of heads. dh The hidden dimension represents a single attention head. d f This represents the hidden dimension of the feedforward neural network. This indicates the length of the output token.

[0060] The Transformer decoding layer reserves a key-value storage component for reuse in the generation of subsequent tokens; therefore, a key-value cache is generated during batch processing. m The sensed echo signals of each target are stored in the KV cache of the edge server.

[0061] Furthermore, the process for maximizing the number of serviceable sensing targets is as follows: The perceived signal-to-noise ratio (SNR) is used to evaluate perception performance. Since the perceived SNR is difficult to optimize directly, the lower bound of the perceived SNR is derived using Jensen's inequality:

[0062] in, , .

[0063] The Chinson inequality serves as a constraint in optimization problems. If the lower bound of SNR satisfies the threshold, then SNR must also satisfy it, thus ensuring the performance of perception. Given limited base station computing and storage resources, the expression for maximizing the number of sensing targets by jointly optimizing communication and sensing beamforming and the set of target access variables is as follows:

[0064] Wherein, constraint C1 indicates that the BS transmit power does not exceed its upper limit. Constraint C2 represents the user group. c Chinese users k The communication rate is not lower than the lower limit Constraint C3 represents the objective. m Lower limit of perceived signal-to-noise ratio Constraints C4 and C5 represent the maximum computing and storage resources of the base station; constraint C6 indicates that the target access variable is a binary variable.

[0065] Furthermore, the process of using a two-layer iterative resource management method to solve the problem model of maximizing the number of sensing targets accessed, and to determine the sensing targets, includes the following steps: In the outer layer, the sensing targets are considered one by one in a certain order. When considering each sensing target, the edge computing and storage resource greedy scheduling algorithm is used to initially determine the sensing targets that will be connected to the set of sensing targets. Based on the defined sensing target, the successive convex approximation algorithm is used in the inner layer to optimize the beamforming emitted by the base station for the communication user and the sensing target, thereby determining the performance of the communication user and the sensing target.

[0066] Furthermore, the process of optimizing the beamforming emitted by the base station for communication users and sensing targets based on the successive convex approximation algorithm is as follows: S31, Given a set of target service variables Next, solve the communication beamforming and sensing beamforming matrices. It is a problem of finding a feasible solution:

[0067] Constraints S32, C2, and C3 are non-convex constraints, which prevents the problem from being solved directly. First, constraint C2 is transformed to obtain... ,make We can obtain:

[0068] Right now , is a difference convex function.

[0069] Then, let , Transforming constraint C3, we can obtain:

[0070] Right now .

[0071] S33. The successive convex approximation method is used to iteratively transform the difference convex constraint into a convex function to solve the problem. The lower left bound is found through a first-order Taylor expansion, and the process is iteratively processed. For the successive convex approximation... o In the next iteration, consider the following lower bound:

[0072]

[0073] Substituting the above formula, we get the first... o The problem expression for the next iteration is: :

[0074] Therefore, the problem The non-convex constraints in the model are approximated as convex constraints, which are implemented in O iterations using the MOSEK solver from the general CVX optimization toolbox. To describe the convergence performance of the multi-user, multi-objective ISAC system, in each iteration of the convex approximation, we use the Frobenius norm of all communication and sensing beamforming updates as the convergence criterion, i.e., if the Frobenius norm of the updates is greater than or equal to the norm of the convex approximation... o The overall change in the beamforming set obtained in the next iteration compared to the previous iteration satisfies:

[0075] It is then assumed that the successive convex approximation iteration has converged, and the inner loop is terminated early.

[0076] The process of initially determining the sensing targets in the access set of sensing targets using edge computing and a greedy scheduling algorithm for storage resources is as follows: In the process of gradually constructing the set of perception target services, the optimization problem of the set of perception target services can be modeled as a 0-1 integer programming problem:

[0077] Because the scheduling of sensing tasks and beamforming are interdependent, the problem... It is difficult to solve directly, so in order to solve the problem effectively A greedy strategy is proposed, aiming to maximize the number of service targets under resource constraints. The specific process is as follows: (1) Calculate the computation and storage requirements for each target. and Sort by resource demand in ascending order; (2) Consider the objectives one by one in the sorted order: For the current objective m First, it determines whether the computing power and storage requirements of the edge node exceed the remaining resources. If the resources are sufficient, it adds the edge node to the perception target service set and calls the inner layer algorithm to optimize beamforming.

[0078] Simulation conditions In the simulation, we consider a MIMO base station with 10 transmit and 10 receive antennas each, serving multiple single-antenna users and targets, with a total transmit power of... Ten user groups and five targets are set up, randomly distributed within a range of [60, 500] from the base station. Assume the Ricean factor of the Ricean channel is... Furthermore, the noise variances of communication and sensing are the same, i.e. And the lower limit of the communication rate for each user is set to The target perception SNR lower limit is set to The number of samples is The convergence threshold of beamforming is 3-bit quantization is used, and the quantization threshold follows a centrosymmetric structure, i.e. .

[0079] Simulation content and result analysis Simulation 1: Deployment map of user and target locations.

[0080] like Figure 3 As shown, a two-dimensional location distribution map of communication users and sensing targets within the system is presented. Different colors in the map are used to distinguish service objects in different sub-frequency bands. This distribution map intuitively reflects the random spatial distribution relationship of communication and sensing objects.

[0081] Simulation 2: Convergence of beamforming optimization process based on successive convex approximation algorithm.

[0082] like Figure 4 As shown, after 5 iterations, the successive convex approximation algorithm... The slight reduction in performance validates the fast convergence and effectiveness of the proposed algorithm. Simulation 3: Comparative analysis of the relationship between target success rate and target number under different sampling numbers.

[0083] like Figure 5 As shown, with the same number of samples Under the condition, as the number of targets increases M As the number of targets increases, the overall success rate of target access decreases. This indicates that when the number of targets to be served increases, the system's power and beam resources need to be allocated among more targets, making it more difficult to meet access constraints, thus leading to a decrease in the proportion of targets that can be successfully accessed. With the same number of targets... M Down, The larger the sample size, the lower the success rate. This indicates that increasing the sample size leads to higher edge computing and storage requirements. Due to limited system computing power and storage resources, the number of targets that can simultaneously meet resource constraints decreases.

[0084] Simulation 4: Comparative analysis of the cumulative distribution of user communication rates under different target numbers.

[0085] along with M As the value increases, the cumulative distribution curve of communication rate shifts to the left under the same cumulative probability, indicating that the overall distribution of user rate is shifting towards the lower rate range. M Increasing the threshold from 1 to 5 significantly squeezed communication resources, leading to a substantial drop in user speed. M When increasing from 5 to 10, the additional performance degradation caused by further increasing the number of perceived targets is very limited.

[0086] Simulation 5: Average communication rate variation curves for each user group under different transmit power constraints.

[0087] Figure 6 A cumulative distribution map of user communication rates under different numbers of targets; like Figure 7 As shown, when both the number of users and the target number are 5, with The increase in power budget leads to an overall increase in the average rate across all user groups. This is because the increased power budget allows for greater effective transmit power allocation in beamforming, thereby improving communication rates and resulting in rate gain. However, the differences in rate levels and growth rates among different user groups indicate variations in channel conditions and interference environments, leading to inconsistent communication performance between groups.

Claims

1. A method for managing and controlling synesthetic network resources for edge reasoning, characterized in that, Includes the following steps: S1: Based on the integrated sensing network, a multi-user, multi-target ISAC system is constructed using frequency division multiple access technology; S2: In a multi-user, multi-target ISAC system, the impact of base station transmit power, user communication rate, sensing signal-to-noise ratio of the sensing target, edge computing power of the edge server, and memory requirements of the edge server on the target sensing task is analyzed, and a model for maximizing the number of sensing targets accessed is constructed. S3: A two-layer iterative resource management method is adopted to solve the problem model of maximizing the number of sensing targets connected, thereby maximizing the number of sensing targets connected and determining the sensing targets; The process of using a two-layer iterative resource management method to solve the problem model of maximizing the number of sensing targets connected, and to determine the sensing targets, includes the following steps: In the outer layer, the sensing targets are traversed in ascending order of computing power and storage requirements. Using a greedy scheduling algorithm for edge computing and storage resources, the sensing targets that can be accessed in the set of sensing targets are initially determined. Based on a defined set of sensing targets, the inner layer uses a successive convex approximation algorithm to optimize the beamforming of the base station for communication users and sensing targets, thereby ensuring the service performance of communication users and sensing targets. Finally, the feasibility of the constraint is verified based on the inner optimization results to determine whether to accept the current objective. By combining the outer greedy construction, the inner convex optimization and the result verification decision, a feasible solution is obtained efficiently under the constraint of limited resources.

2. The method for managing and controlling integrated sensory network resources for edge reasoning according to claim 1, characterized in that, The multi-user, multi-objective ISAC system includes: Base station transmission signal model: used to construct the transmission signals for joint communication and sensing at base stations; User received signal model: used to construct the base station transmitted signal sent by the base station transmitted signal model for communication users to evaluate communication performance; Radar sensing model: used to obtain task information of each sensing target based on the echo signal of the base station transmitted signal after being reflected by the sensing target, which is sent by the user received signal model; Sampling quantization model: used to quantize and sample the sensing echo signal sent by the radar sensing model to generate discrete data for each sensing target; Edge computing power model: used to receive the sensing data quantized by the sampling quantization model and calculate the resource requirements for edge computing power and storage.

3. The method for managing and controlling synesthetic network resources for edge reasoning according to claim 2, characterized in that, The sampling quantization model is constructed as follows: Multi-user, multi-objective ISAC systems include C One communication user group and M The first goal, of which the second c The user group includes K c A communication user, in order to describe the target m Whether it is served by BS, introduce binary variables ,when When, it indicates the target. m Being served by BS, making Access the target set of binary variables; exist Within, based on the sampling frequency Sampling was performed, and the number of samples was [number missing]. For the first m The sensed echo signal of each target, and the signal sampled at the base station receiving antenna, are represented as follows: in, This represents the complex reflection coefficient of the m-th target, which includes path loss and scattering characteristics. It is transmitting a signal; It is the target response matrix of target m. Antenna spacing, and The number of transmit and receive antennas, and They are respectively Directional transmission and reception steering vectors, The mean is zero and the variance is... Complex Gaussian white noise; Using transmitted signals The received echo signal is subjected to matched filtering, i.e.: in, , Representing user groups c The Middle k The communication beamforming vector of the first user and the first m Perceived beamforming vector of a target; The quantized sampled signal is: Introducing a receiving filter The process of processing the vectorized echo signal is as follows: The perceived signal-to-noise ratio is: The real and imaginary parts of the sampled signal are respectively treated using q -bits quantization, then the quantized digits are... m The sensed echo signals of the target are: in, yes q -bits quantizer, then the 1st m The total number of bits of data in the echo signal of each sensing target is: 。 4. The method for managing and controlling synesthetic network resources for edge reasoning according to claim 3, characterized in that: The expression for constructing the maximum number of target access points for perception is as follows: Wherein, constraint C1 indicates that the base station's transmit power does not exceed its upper limit. ; Constraint C2 indicates the user group c Chinese users k The communication rate is not lower than its lower limit. ; Constraint C3 represents the objective. m The perceived signal-to-noise ratio is not lower than its lower limit. ; Constraints C4 and C5 represent the maximum computing and storage resources of the base station. and ; Constraint C6 indicates that the target access variable is a binary variable. The computing power requirement of target m. : Cache requirements for target m.

5. A method for managing and controlling synesthetic network resources for edge reasoning according to claim 4, characterized in that, The process of optimizing the beamforming emitted by the base station for communication users and sensing targets based on the successive convex approximation algorithm is as follows: S31, Given a set of perception target service variables Next, solve the communication beamforming and sensing beamforming matrices. It is a problem of finding a feasible solution: S32. First, transform constraint C2 to obtain the signal-to-interference-plus-noise ratio (SINR) form, i.e. ,make The lower limit of the signal-to-interference-plus-noise ratio is represented by: (Expanded, we get:) Right now This function is a difference convex function, where, Indicates BS and the c The first in the user group k Channel matrix between user equipment For noise; Then, let , Transforming constraint C3, we get: Right now ; S33. The successive convex approximation method is used to iteratively transform the difference convex constraint into a convex function to solve the problem. The lower left bound is found through a first-order Taylor expansion, and the process is iteratively processed. For the successive convex approximation... o In the next iteration, consider the following lower bound: Substituting the above formula, we get the first... o The problem expression for the next iteration is: : Therefore, the problem The non-convex constraint in the equation is approximated as a convex constraint, and the convex constraint is iterated. O This implementation uses the MOSEK solver from the General CVX Optimization Toolkit for solving.

6. The method for managing and controlling synesthetic network resources for edge reasoning according to claim 4, characterized in that, The process of initially determining the sensing targets to be included in the set of sensing targets using edge computing and a greedy scheduling algorithm for storage resources is as follows: In the process of gradually constructing the set of sensing target services, the optimization problem of the sensing target access set is modeled as a 0-1 integer programming problem: Solve the problem Based on a greedy strategy, the goal is to maximize the number of base station service targets under resource constraints. The specific process is as follows: (1) Calculate the computing power requirement for each sensing target. and h occupancy requirements Sort by resource demand in ascending order; (2) Consider the objectives one by one in the sorted order: For the current objective m First, determine whether its computing power and storage requirements exceed the remaining resources of the edge node. If the resources are sufficient, add it to the perception target service set and perform inner beamforming optimization.

Images