A method for adaptive configuration of communication and sensing resources for a perception task

Abstract

The application discloses a kind of communication perception resource adaptive configuration methods for perception task, comprising: step S1: forming task demand vector;Step S2: forming state vector;Step S3: generating resource configuration vector;Step S4: calculating the perception utility of candidate resource configuration, resource cost and constraint violation degree corresponding;Step S5: based on task utility function and resource constraint solving current time resource configuration scheme;Step S6: resource configuration scheme is issued to each service node and executes;Step S7: feature extraction, target detection and multi-source fusion are carried out;Step S8: form task completion degree evaluation score;Step S9: according to task completion degree evaluation score, the adaptive update of communication perception resource is carried out.The application reduces resource occupation and system energy consumption while meeting the requirements of perception accuracy, delay and reliability.

Description

Technical Field

[0001] This invention relates to the field of integrated sensing and communication, and in particular to an adaptive configuration method for communication sensing resources for sensing tasks. Background Technology

[0002] Integrated sensing systems achieve simultaneous communication transmission and environmental perception within the same network by sharing spectrum, hardware, and signal processing capabilities. With the development of scenarios such as vehicle-to-everything (V2X), low-altitude monitoring, intelligent transportation, industrial inspection, and emergency communication, networks need to provide differentiated target detection, localization, tracking, and recognition capabilities for different sensing tasks. Different tasks have significantly different requirements in terms of sensing accuracy, refresh cycle, latency, reliability, and energy consumption. For example, target tracking tasks focus more on continuous estimation error and refresh frequency, category recognition tasks focus more on the integrity of multimodal features, while low-power scenarios focus more on resource consumption and energy management.

[0003] Existing resource allocation methods typically prioritize communication throughput, link rate, or overall power efficiency as optimization objectives, making it difficult to directly reflect the quality of sensing task completion. Furthermore, there are strong coupling relationships between spectrum, time slots, transmit power, beamwidth, sensing dwell time, sampling rate, edge computing resources, and sensor start / stop status; configuring only one type of resource can easily lead to insufficient sensing accuracy, resource redundancy, or increased system latency. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide an adaptive configuration method for communication sensing resources for sensing tasks. By constructing a joint mapping relationship between task requirements, network status and resource pool, the method achieves coordinated configuration of spectrum, power, beam, sensing dwell time, sampling rate, computing resources and sensor start / stop status, thereby reducing resource consumption and system energy consumption while meeting the requirements of sensing accuracy, latency and reliability.

[0005] The objective of this invention is achieved through the following technical solution: an adaptive configuration method for communication sensing resources oriented towards sensing tasks, which configures resources based on a communication sensing resource adaptive configuration system, wherein the communication sensing resource adaptive configuration system includes a central control node and multiple service nodes, wherein the service nodes are communication nodes, sensing devices, or computing nodes, and the method includes the following steps:

[0006] Step S1: The central control node determines the current sensing task and forms a task requirement vector;

[0007] Step S2: The central control node collects the status information of each service node and forms a status vector. ;

[0008] Step S3: The central control node represents communication resources, sensing resources, and computing resources as a unified resource allocation vector. ;

[0009] Step S4: The central control node determines the task requirement vector. System state vector and resource allocation vectors Calculate the perceived utility, resource cost, and constraint violation degree corresponding to the candidate resource configuration;

[0010] Step S5: The central control node solves the resource allocation scheme for the current moment based on the task utility function and resource constraints;

[0011] Step S6: The central control node distributes the resource configuration plan to each business node and controls each business node to configure according to the resource configuration plan;

[0012] Step S7: Each sensing device performs sensing tasks under the configured resource conditions, and performs feature extraction and target detection to obtain local target detection results or target state estimation results for each device. These results are then uploaded to the central control node for multi-source fusion to obtain the target state estimation results.

[0013] Step S8: The central control node forms a task completion evaluation score based on the target state estimation results, reference results, and resource consumption statistics.

[0014] Step S9: The central control node adaptively updates the communication sensing resources based on the task completion evaluation score.

[0015] The beneficial effects of this invention are as follows: This invention focuses on the requirements of sensing tasks, incorporating communication resources, sensing resources, and computing resources into a unified configuration framework. It can dynamically adjust spectrum, power, beam, sampling rate, sensing dwell time, computing resources, and sensor start / stop status based on target uncertainty, channel quality, and equipment load, thereby improving resource utilization efficiency and ensuring sensing task performance. Simultaneously, this invention, through a feedback update mechanism and a safety-baseline configuration strategy, continuously optimizes resource configuration parameters during long-term operation, reducing sensing errors, communication overhead, computing overhead, and energy consumption, and ensuring the minimum performance requirements of critical sensing tasks under abnormal conditions. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0018] The method of the present invention is based on a communication sensing resource adaptive configuration system, which includes a central control node and multiple service nodes, wherein the service nodes are communication nodes, sensing devices or computing nodes.

[0019] The central control node includes base stations, roadside units, edge servers, fusion processors, or cloud control nodes, which are used to perform state modeling, task utility evaluation, resource allocation decisions, result fusion, and feedback updates.

[0020] Communication nodes can be base stations, access points, communication transceivers, or relay nodes, used to provide communication resources such as bandwidth, frequency bands, time slots, transmission power, and beams;

[0021] Sensing devices can be radar devices, camera devices, wireless sensing nodes, lidar, vehicle terminals, drones, or industrial sensing nodes, used to perform sensing data acquisition, feature extraction, target detection, or target estimation.

[0022] Computing nodes can be edge computing nodes, in-vehicle computing units, fusion processors, or cloud servers, providing edge computing power, caching resources, and cloud collaboration capabilities. The system adaptively coordinates communication, sensing, and computing resources based on the needs of the sensing task and the system state, reducing system overhead while meeting requirements for sensing accuracy, latency, and reliability.

[0023] like Figure 1 As shown, an adaptive configuration method for communication sensing resources for sensing tasks includes the following steps:

[0024] Step S1: The central control node determines the current sensing task based on the upper-layer business, application scenario, or scheduling instructions, and forms a task requirement vector;

[0025] The task requirement vector can be represented as: ,in Indicates the task type. Indicates the sensory region. Indicates the accuracy requirement. Indicates the maximum delay. Indicates reliability requirements, Indicates the refresh cycle. This indicates energy consumption constraints.

[0026] Step S2: The central control node collects the status information of each service node and forms a status vector. ;

[0027] The central control node collects the communication status, perception status, and computing status of each service node to form a state vector. :

[0028]

[0029] in, These are vectors corresponding to the communication state, the sensing state, and the computation state, respectively.

[0030] Specifically, the vector corresponding to the communication state It can include one or more of the following: channel gain, received signal-to-noise ratio, interference level, queue length, packet loss rate, transmission delay, spectrum occupancy, or link availability, to reflect whether the communication link can support the data transmission requirements of the current sensing task; the vector corresponding to the sensing state. This can include one or more of the following: target uncertainty, sensor confidence, historical sensing error, target motion speed, target occlusion status, current observation quality, or enabled sensor status, used to reflect the sensing device's observation capability of the current target or sensing area; calculate the vector corresponding to the status. It can include one or more of the following: computing load, available computing power, cache usage, task waiting queue, remaining energy, or cloud-edge collaboration status, to reflect the support capability of local, edge, or cloud computing resources for subsequent feature extraction, target detection, target estimation, or multi-source fusion tasks.

[0031] Furthermore, in some embodiments of this application, for a business node with only a single function, each node has a value for only its corresponding state vector, while the other state components are set to zero or null values; in other embodiments of this application, some nodes have communication, sensing or computing functions at the same time, and their corresponding multiple state components can all have values.

[0032] Step S3: The central control node represents communication resources, sensing resources, and computing resources as a unified resource allocation vector. ;

[0033]

[0034] in, These are vectors corresponding to communication resources, sensing resources, and computing resources, respectively.

[0035] Communication resources include bandwidth, frequency band, time slot, transmit power, and beam; sensing resources include sampling rate, sensing dwell time, and sensor start / stop status; computing resources include edge computing power, cache resources, and cloud collaboration capabilities.

[0036] Similarly, in some embodiments of this application, each service node is a service node with only a single function, and only the corresponding resource component has a value, while the other resource components are set to zero or take a null value; in other embodiments of this application, some service nodes have communication, sensing or computing functions at the same time, and their corresponding multiple resource components can all have values.

[0037] Step S4: The central control node determines the task requirement vector. System state vector and resource allocation vectors Calculate the perceived utility corresponding to the candidate resource configuration. Resource costs and degree of violation of constraints .

[0038] Specifically, the central control node configures candidate resources. Combined with task requirement vector The task type, accuracy requirements, maximum latency, reliability requirements, and energy consumption constraints, as well as the state vector. The system considers the communication, sensing, and computational states within a given context to predict the sensing error, detection probability, fusion confidence, task latency, task reliability, energy consumption, and resource utilization under a given candidate resource configuration. This prediction can be achieved using historical statistical models, lookup table models, simulation models, machine learning prediction models, or a combination thereof.

[0039] Perceptual utility can be determined in conjunction with target estimation error, target detection probability, fusion confidence, or task completion rate. For example, for localization, velocity measurement, distance measurement, or angle measurement tasks, perceptual utility is negatively correlated with target state estimation error; for target detection or recognition tasks, perceptual utility is positively correlated with detection probability, recognition accuracy, or classification confidence. The specific expression is as follows:

[0040]

[0041] in, The benefit term representing the perceived error can be derived from the prediction of the perceived error. Maximum perceptual error allowed by the task Calculated, for example , This represents the benefit of target detection, which can be any one of the following: predicted detection probability, recognition accuracy, or classification confidence. The confidence level benefit term can be obtained by weighted summation of sensor confidence and link reliability. The delay benefit term can be derived from the predicted task delay. With maximum allowable delay Calculated, for example , The reliability benefit can be represented by any one of the following: predicted task reliability, packet loss rate, link availability, or retransmission success rate. , , , , The utility weight is determined by the task type;

[0042] Resource cost measures the extent to which candidate resource configurations consume communication, sensing, computing, and energy resources. It can be calculated by a weighted sum of various resource overheads, such as communication resource overhead, sensing resource overhead, and computing resource overhead, and its expression is:

[0043]

[0044] in, to For cost weighting coefficients, The communication resource overhead is represented by a normalized or weighted sum of the bandwidth, frequency band, time slot, transmit power, and beam occupancy in the candidate resource configuration. The sensing resource overhead is represented by the sampling rate, sensing dwell time, number of activated sensors, or sensing beam occupancy, obtained through normalization or weighted summation. This represents the computational resource overhead, obtained by normalizing or weighting the overhead from edge computing power, cache resources, and cloud collaboration. This represents the energy resource consumption, which is obtained by normalizing or weighting the energy consumption of communication transmission, sensing and sampling, and computing and processing. The change in the current candidate resource configuration compared to the previous time step is used to reflect the reconfiguration overhead caused by frequency band switching, beam switching, sensor start-up and shutdown, or computation migration.

[0045] The constraint violation degree is used to measure whether the candidate resource configuration meets the accuracy, latency, reliability, energy consumption, and resource capacity constraints of the current sensing task. When the candidate resource configuration satisfies all constraints, the constraint violation degree is zero; when any constraint is not satisfied, the constraint violation degree is positive, and the more severe the violation, the larger the value. Its expression is:

[0046]

[0047] in, This indicates a violation of the perception accuracy constraint, based on the predicted perception error. Does it exceed the maximum permissible perception error? Confirmed, if the value exceeds the limit, take 1; if it does not exceed the limit, take 0. This indicates a violation of the delay constraint, based on the predicted task delay. Does it exceed the maximum allowable delay? Confirmed, if the value exceeds the limit, take 1; if it does not exceed the limit, take 0. This indicates a violation of reliability constraints, based on reliability. Is it below reliability requirements? Determined: 1 for values ​​below 1, 0 for values ​​not below 1. Indicates energy consumption constraint violations and predicts energy consumption. Does it exceed energy consumption constraints? Confirmed, if the value exceeds the limit, take 1; if it does not exceed the limit, take 0. This indicates a violation of resource capacity constraints. It is determined based on whether the bandwidth, power, time slot, beam, sampling rate, sensing dwell time, computing resources, or cache resources in the candidate resource configuration exceed the system's available limit. If any one of these parameters exceeds the available limit, it is set to 1; if none of the parameters exceed the available limit, it is set to 0. to To constrain the penalty weights, , , and From the task requirement vector Give or by The accuracy requirements, maximum latency, reliability requirements, and energy consumption constraints are calculated from the data; the upper limits of each resource are derived from the state vector. And the configurable resource pool is determined.

[0048] Step S5: The central control node solves the resource allocation scheme for the current moment based on the task utility function and resource constraints;

[0049] The system solves the resource allocation scheme for the current moment based on the task utility function and resource constraints.

[0050] The task utility function is derived from the task demand vector. And the determination of the type of perception task:

[0051] For high-precision sensing tasks, the task utility function emphasizes target estimation error or detection accuracy; for low-latency tasks, the task utility function emphasizes task completion latency.

[0052] For low-power tasks, the task utility function places greater emphasis on energy consumption and resource usage.

[0053] Therefore, the task utility function can be pre-set by the upper-layer business requirements, or it can be trained and adjusted based on historical operating data, simulation data, or online feedback results; resource constraints are jointly determined by task requirements and available system resources, including at least total bandwidth constraints, frequency band or time slot constraints, transmit power constraints, beam or antenna resource constraints, sampling rate or sensing dwell time constraints, edge computing resource constraints, energy consumption budget constraints, maximum latency constraints, and minimum reliability constraints.

[0054] Before solving for the resource allocation scheme, the central control node determines the task requirement vector. State vector And generate a set of candidate resource configurations from the configurable resource pool. The system determines the initial resource configuration and initial weight parameters. The initial resource configuration can be generated based on the minimum accuracy, maximum latency, minimum reliability of the current task, and available system resources through preset rules. The initial weight parameters include utility weight, resource cost weight, and constraint penalty weight, which can be determined by task type, task priority, historical running data, or system default rules.

[0055] Resource allocation schemes refer to the optimal or suboptimal resource allocation obtained by solving optimization problems. , i.e. decision variables In the candidate resource allocation set The value of is taken from . Specifically, the resource allocation scheme at the current moment is obtained by solving the following optimization objective:

[0056]

[0057] in, This is a set of candidate resource configurations formed based on currently available communication resources, sensing resources, and computing resources. The perceived utility function obtained in step S4. The resource cost function obtained in step S4, The constraint violation degree function obtained in step S4, and These are the weight parameters corresponding to resource costs and constraint violation penalties, respectively.

[0058] The candidate configurations in the above optimization objectives can be obtained through rule search, heuristic search, convex optimization, dynamic programming, reinforcement learning, deep neural networks, or combinations thereof. For high-precision sensing tasks, the central control node increases the weight of the sensing error benefit term or the detection accuracy related weight; for low-latency tasks, it increases the weight of the latency benefit term and the latency constraint related weight; for high-reliability tasks, it increases the weight of the reliability benefit term and the reliability constraint related weight; and for low-power tasks, it increases the weight of the energy cost and resource cost related weight.

[0059] Step S6: The central control node distributes the resource configuration plan to each business node and controls each business node to configure according to the resource configuration plan;

[0060] In cases of insufficient resources or degraded link quality, priority should be given to ensuring the minimum accuracy and latency constraints of high-priority sensing tasks.

[0061] Step S7: Each sensing device performs sensing tasks under the configured resource conditions, and performs feature extraction and target detection to obtain local target detection results or target state estimation results for each device. These results are then uploaded to the central control node for multi-source fusion to obtain the target state estimation results.

[0062] Feature Extraction: In step S7, each sensing device performs sensing tasks according to the resource configuration scheme issued in step S6: For radar devices, radar echoes can be collected based on the configured sampling rate, sensing dwell time, transmit power, and beam direction, and target features such as distance, speed, and angle can be extracted through range-Doppler processing, beamforming, time-frequency analysis, deep neural networks, or combinations thereof; For vision devices, images or video frames can be collected based on the configured frame rate and computing resources, and target position, category, and appearance features can be extracted through convolutional neural networks, visual Transformers, or target detection networks; For wireless sensing nodes, channel state information can be obtained according to the configured frequency band, time slot, and power, and channel features related to target movement, occlusion, or position changes can be extracted.

[0063] Target detection: The system can obtain local target detection results or target state estimation results for each device through threshold detection, constant false alarm rate detection, classification network, regression network, tracking filter or detection-tracking joint network.

[0064] Multi-source fusion: Time alignment, coordinate transformation and confidence assessment are performed on the detection results, target state estimation results or intermediate features uploaded by different sensing devices, and a unified fusion sensing result is obtained by using weighted fusion, attention fusion, Kalman filter fusion, Bayesian fusion, trajectory-level fusion or feature-level fusion.

[0065] Specifically, the target state estimation result is calculated based on the local estimation results of each sensing device or the fused multi-source features, and can be expressed as:

[0066]

[0067] in, This represents the sensing features extracted by the k-th sensing device; This represents the detection confidence or estimated confidence of the k-th sensing device. This represents the link status corresponding to the k-th sensing device. Represents a multi-source fusion function. This represents the target state estimation function. Represents network parameters, The target state estimation result includes one or more of the following: target position, velocity, angle, distance, category, or trajectory.

[0068] The target state estimation function can be obtained using a neural network and through pre-training: first, several sets of... As sample features, the system collects the corresponding ground truth values ​​of the target state (one or more of the following: target position, velocity, angle, distance, category, or trajectory; the specific parameters selected must be consistent with the parameters included in the target state estimation result). The sample features are then input into the target state estimation function, which outputs the target state estimation result. A loss function (such as MSE) is calculated using this result and the ground truth values ​​of the target state. Finally, backpropagation or gradient descent is used to adjust the network parameters. After updating and training, the pre-trained target state estimation function is obtained, which can be directly used for target state estimation.

[0069] Step S8: The central control node forms a task completion evaluation score based on the target state estimation results, reference results, and resource consumption statistics.

[0070] The reference results can be real annotations, manually annotated results, historical trajectory prediction results, prior model outputs, subsequent observation results, or pseudo-reference results obtained from consistency checks of multiple devices. Specifically, for target localization, velocity measurement, ranging, or angle measurement tasks, the system can calculate the error between the target state estimation result and the reference result; for target detection tasks, the system can calculate the detection probability, false negative rate, false alarm rate, or intersection-union ratio; for target recognition tasks, the system can calculate the classification accuracy or classification confidence. Simultaneously, the system statistically analyzes transmission overhead, computational overhead, energy consumption, task latency, spectrum occupancy, and reliability indicators under the current resource configuration.

[0071] Calculate the task completion evaluation results , represented as:

[0072]

[0073] in, This represents the perception error, obtained from the difference between the target state estimation result and the reference result. The task latency is represented by the time difference between the task trigger time, the resource allocation time, the perception processing completion time, and the result output time. Task reliability can be expressed using one of the following metrics: link success rate, packet loss rate, retransmission count, fusion confidence score, or task success rate. These represent the overhead of communication resources, sensing resources, computing resources, and energy resources, respectively. Indicates spectrum occupancy. This indicates the degree of constraint violation, which is recalculated based on the actual measurement results using the constraint violation calculation method in step S4.

[0074] For multiple tasks, calculate the task completion evaluation results separately. Central control node The scores are normalized and then weighted (based on task type) and summed to obtain the task completion score. .

[0075] Task completion rating This is used to determine whether the current resource configuration meets the task requirements and serves as the feedback basis for the update in step S9.

[0076] Step S9: The central control node adaptively updates the communication sensing resources based on the task completion evaluation score.

[0077] Specifically, the central control node will use the task completion score obtained in step S8. As a feedback signal, and in conjunction with the initial resource configuration and current resource configuration in step S5 The weight parameters in the optimization objective are updated in a closed loop.

[0078] During the initial run, the system uses the initial resource configuration and initial weight parameters determined in step S5; in subsequent time slots, the system updates the resource configuration priority and weight parameters based on the task completion evaluation results of the previous time slot.

[0079] For the For each type of resource, the system calculates its marginal contribution based on the changes in task completion and resource overhead before and after the addition of that resource.

[0080]

[0081] in, This indicates the task completion score under the current resource configuration. Indicates reducing or removing the first The task completion score obtained after classifying resources can be predicted through offline simulation, historical sample playback, short-term trial configuration, or resource requirement estimation models. Indicates the first Overhead of class resources To avoid constants with a denominator of zero. If Less than the preset contribution threshold This indicates that the marginal contribution of this type of resource to the current task is low, and the system will reduce the allocation ratio or priority of this type of resource in subsequent configurations. For example, when increasing the sampling rate does not significantly reduce the perception error, the marginal contribution of perception resources is low, and the system will reduce the priority of the sampling rate; when increasing bandwidth does not significantly improve link reliability or perception results, the marginal contribution of communication resources is low, and the system will reduce the bandwidth or power allocation ratio; when increasing computing resources provides limited improvement in task latency or estimation error, the marginal contribution of computing resources is low, and the system will reduce the occupancy of edge computing resources.

[0082] To determine whether a certain type of constraint is continuously violated, the central control node can use a sliding window method. Specifically, the first... The degree of violation of class constraints is represented as In length of Calculate the average violation level within the sliding window ,like Greater than the preset threshold ,or continuous The number of time slots is greater than the threshold. If a constraint is continuously violated, the central control node will determine that the constraint is being violated. When the perception accuracy constraint is continuously violated, the central control node will increase the priority of sampling rate, perception dwell time, beam resources, or multi-modal sensor activation. When the latency constraint is continuously violated, the configuration priority of low-latency links, bandwidth, edge computing resources, or cloud-edge collaborative resources will be increased. When the reliability constraint is continuously violated, the priority of transmit power, low-interference frequency bands, redundant links, or retransmission resources will be increased. When the energy consumption constraint is continuously violated, the priority of high-power configurations will be reduced, and the priority of low-power transmission, low sampling rate, or low-complexity computing modes will be increased.

[0083] Furthermore, the central control node can update the weight parameters in the optimization objective of step S5 based on the degree of constraint violation. The weight parameters in the optimization objective include... and These correspond to resource costs and penalty items for constraint violations, respectively.

[0084] Meanwhile, in step S4 These correspond to violations of accuracy, latency, reliability, energy consumption, and resource capacity constraints, respectively.

[0085] Specifically, the constraint penalty weights can be updated as follows:

[0086]

[0087] in, Indicates the first The penalty weight corresponding to the remainder violation item. Indicates the update step size. This indicates that the priority is limited to a preset range. These represent the upper and lower limits of the weights, respectively. If a certain type of constraint is continuously violated, the corresponding penalty weight is increased, making subsequent optimization processes more inclined to satisfy that type of constraint; if a certain type of constraint is satisfied for a long time and the resource cost is too high, the corresponding penalty weight can be reduced or the corresponding resource cost weight can be increased.

[0088] In addition to the above schemes, the resource cost weight can be updated according to task completion and resource overhead: when communication resource overhead is too high and task completion meets the requirements, the weight should be increased. When the perception resource overhead is too high and the perception error is already below the threshold, increase... When computational resource overhead is too high and latency requirements are met, increase... When energy consumption exceeds energy consumption constraints, increase... When frequent resource switching leads to system instability, improve... When any constraint is continuously violated, increase The system can also... As training samples, the resource demand estimation model or policy network is incrementally trained, allowing subsequent resource allocation strategies to gradually adapt to the current scenario. Through this method, the initial resource allocation, utility function weights, resource cost weights, constraint penalty weights, marginal contribution, and task completion evaluation results can form a closed-loop correspondence, thereby achieving adaptive updates to the resource allocation strategy.

[0089] In the embodiments of this application, the configuration method further includes step S10: security constraint and backup configuration mechanism: when a channel interruption, equipment abnormality, insufficient energy, rapid target maneuvering, or perception error exceeding a threshold is detected, the system triggers a backup configuration strategy, temporarily increasing critical link resources, increasing the sampling rate, increasing the perception dwell time, or enabling redundant sensors; when the system recovers and stabilizes, it switches back to adaptive optimization configuration.

[0090] Furthermore, the strategy learning and online updating of this allocation method are explained. Specifically, during the offline training phase, simulation data or historical operational data under different tasks, channels, target motion states, and resource constraints can be used to train the resource allocation strategy model. The input of this model is a task demand vector and a system state vector, and the output is a resource allocation vector or a resource allocation probability distribution. During the online operation phase, the system fine-tunes the strategy model or updates the threshold based on real-time feedback. When feedback results indicate that a certain type of resource contributes little to the task, the system reduces the allocation ratio of that type of resource; when feedback results indicate that the sensing error continues to increase or the target uncertainty rises, the system increases the sensing dwell time, sampling rate, beam resources, or computing resources.

[0091] The foregoing description illustrates and describes a preferred embodiment of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for adaptive configuration of communication sensing resources for sensing tasks, wherein resource configuration is performed based on a communication sensing resource adaptive configuration system, characterized in that: The adaptive configuration system for communication sensing resources includes a central control node and multiple service nodes, wherein the service nodes are communication nodes, sensing devices, or computing nodes. The method includes the following steps: Step S1: The central control node determines the current sensing task and forms a task requirement vector; Step S2: The central control node collects the status information of each service node and forms a status vector. ; Step S3: The central control node represents communication resources, sensing resources, and computing resources as a unified resource allocation vector. ; Step S4: The central control node determines the task requirement vector. System state vector and resource allocation vectors Calculate the perceived utility, resource cost, and constraint violation degree corresponding to the candidate resource configuration; Step S5: The central control node solves the resource allocation scheme for the current moment based on the task utility function and resource constraints; Step S6: The central control node distributes the resource configuration plan to each business node and controls each business node to configure according to the resource configuration plan; Step S7: Each sensing device performs sensing tasks under the configured resource conditions, and performs feature extraction and target detection to obtain local target detection results or target state estimation results for each device. These results are then uploaded to the central control node for multi-source fusion to obtain the target state estimation results. Step S8: The central control node forms a task completion evaluation score based on the target state estimation results, reference results, and resource consumption statistics. Step S9: The central control node adaptively updates the communication sensing resources based on the task completion evaluation score.

2. The adaptive configuration method for communication sensing resources for sensing tasks according to claim 1, characterized in that: The task requirement vector is represented as follows: ； in Indicates the task type. Indicates the sensory region. Indicates the accuracy requirement. Indicates the maximum delay. Indicates reliability requirements, Indicates the refresh cycle. This indicates energy consumption constraints.

3. The method of claim 1, wherein the method is a method of communication and sensing resource adaptive configuration for a perception task. The communication resources include bandwidth, frequency band, time slot, transmit power, and beam; the sensing resources include sampling rate, sensing dwell time, and sensor start / stop status; the computing resources include edge computing power, cache resources, and cloud collaboration capabilities.

4. The adaptive configuration method for communication sensing resources for sensing tasks according to claim 1, characterized in that: In step S4, the central control node configures the candidate resources. Combined with task requirement vector The task type, accuracy requirements, maximum latency, reliability requirements, and energy consumption constraints, as well as the state vector. The system considers the communication, sensing, and computational states in the data to predict the sensing error, detection probability, fusion confidence, task latency, task reliability, energy consumption, and resource usage under the candidate resource configuration. The prediction is achieved through historical statistical models, lookup table models, simulation models, machine learning prediction models, or combinations thereof. Computational perceived utility : ； in, The benefit term for perceived error is determined by predicting the perceived error. Maximum perceptual error allowed by the task Calculations show that , This represents the benefit of target detection, which can be any one of the following: predicted detection probability, recognition accuracy, or classification confidence. The fusion confidence gain term is obtained by weighted summation of sensor confidence and link reliability. This represents the latency benefit term, derived from the predicted task latency. With maximum allowable delay Calculations show that , The reliability benefit can be represented by any one of the following: predicted task reliability, packet loss rate, link availability, or retransmission success rate. , , , , Utility weight; computational resource cost : ； in, to For cost weighting coefficients, The communication resource overhead is represented by a normalized or weighted sum of the bandwidth, frequency band, time slot, transmit power, and beam occupancy in the candidate resource configuration. The sensing resource overhead is represented by the sampling rate, sensing dwell time, number of activated sensors, or sensing beam occupancy, obtained through normalization or weighted summation. This represents the computational resource overhead, obtained by normalizing or weighting the overhead from edge computing power, cache resources, and cloud collaboration. This represents the energy resource consumption, which is obtained by normalizing or weighting the energy consumption of communication transmission, sensing and sampling, and computing and processing. The change range between the current candidate resource configuration and the resource configuration at the previous time step is used; Computing a degree of constraint violation : ； in, This indicates a violation of the perception accuracy constraint; if the predicted perception error... Exceeding the maximum permissible perception error , =1, otherwise =0; This indicates a violation of the delay constraint if the task delay is predicted. Exceeding the maximum allowable delay , Take 1, otherwise Set to 0; This indicates a violation of reliability constraints, if reliability Below reliability requirements Sure, Take 1, otherwise Set to 0; This indicates a violation of energy consumption constraints, if energy consumption is predicted. Exceeding energy consumption constraints , Take 1, otherwise Set to 0; This indicates a violation of resource capacity constraints. It is determined based on whether the bandwidth, power, time slot, beam, sampling rate, sensing dwell time, computing resources, or cache resources in the candidate resource configuration exceed the system's available limit. If any parameter in the candidate resource configuration exceeds the available limit... If the candidate resource configuration parameters do not exceed the available limit, then... Take 0, to To constrain the penalty weights.

5. The adaptive configuration method for communication sensing resources for sensing tasks according to claim 4, characterized in that: Step S5 includes: Before solving for the resource allocation scheme, the central control node determines the task requirement vector. State vector And generate a set of candidate resource configurations from the configurable resource pool. And determine the initial resource configuration and initial weight parameters; The resource allocation scheme for the current moment is obtained by solving the following optimization objective: ； in, and These are the weight parameters corresponding to resource costs and constraint violation penalties, respectively; The resource allocation scheme refers to the optimal or suboptimal resource allocation obtained by solving an optimization problem, denoted as... , i.e. decision variables In the candidate resource allocation set The optimal or second-best value in the range.

6. The adaptive configuration method for communication sensing resources for sensing tasks according to claim 5, characterized in that: Step S8 includes: Calculate the task completion evaluation results , represented as: ； in, This represents the perception error, obtained from the difference between the target state estimation result and the reference result. The task latency is represented by the time difference between the task trigger time, the resource allocation time, the perception processing completion time, and the result output time. Task reliability can be expressed using one of the following metrics: link success rate, packet loss rate, retransmission count, fusion confidence score, or task success rate. These represent the overhead of communication resources, sensing resources, computing resources, and energy resources, respectively. Indicates spectrum occupancy. Indicates the degree of constraint violation; For multiple tasks, calculate the task completion evaluation results separately. Central control node The data is normalized, and then weighted and summed according to task scoring criteria to obtain the task completion score. ; Task completion rating This is used to determine whether the current resource configuration meets the task requirements and serves as the feedback basis for the update in step S9.

7. The adaptive configuration method for communication sensing resources for sensing tasks according to claim 1, characterized in that: The configuration method further includes step S10: security constraint and backup configuration mechanism: when a channel interruption, equipment abnormality, insufficient energy, rapid target maneuvering, or perception error exceeding the threshold is detected, a backup configuration strategy is triggered to temporarily increase critical link resources, increase sampling rate, increase perception dwell time, or enable redundant sensors; when the system recovers and stabilizes, it switches back to adaptive optimization configuration.

Images