A method, network device and system for scheduling common sense resources based on dynamic channel sensing

Abstract

The application relates to the field of new-generation wireless communication technology and is a kind of sensing resource scheduling method based on dynamic channel sensing, network equipment and system. In order to solve the problem that the existing reactive scheduling has lag and resource mismatch in a dynamic scene, a forward-looking scheduling method is adopted. The method constructs and updates a sensing digital twin model by continuously collecting historical channel and position data, uses an intelligent prediction system to deduce the digital twin model, and predicts the channel quality and target state at a future time. According to the prediction result, the scheduler makes cross-period sensing resource pre-allocation and optimization decision. The application improves the scheduling mechanism from instant decision based on the current state to forward-looking planning based on future state prediction, improves the timeliness and accuracy of resource scheduling, and can guarantee the reliability of communication while enhancing the sensing performance of the moving target.

Description

Technical Field

[0001] This invention relates to the field of next-generation wireless communication technology, and in particular to a method, network device and system for scheduling communication resources based on dynamic channel awareness. Background Technology

[0002] As a key technology of 6G, integrated sensing and communication can simultaneously achieve both communication and environmental perception (such as radar detection) functions using the same set of hardware and wireless signals, thereby greatly improving spectrum efficiency and reducing hardware costs and system power consumption. In dynamic scenarios, such as vehicle-to-everything (V2X) or drone swarm management, the core task is to efficiently schedule limited wireless resources (such as time, frequency, power, and beam) to ensure high-reliability, low-latency communication while achieving high-precision tracking of high-speed moving targets. Currently, most mainstream resource scheduling methods adopt a reactive mechanism, which makes decisions for the next scheduling cycle in real time based on the current channel state information and target state estimation. This mechanism achieves good results in quasi-static scenarios.

[0003] As application scenarios become increasingly dynamic, the inherent limitations of traditional reactive scheduling mechanisms are becoming increasingly apparent, particularly in densely populated urban areas and highways where numerous high-speed moving targets with uncertain trajectories exist. Due to inherent processing delays throughout the entire link from sensing, reporting, decision-making to execution, the actual state of the wireless channel and the target may have significantly changed by the time the scheduling decision takes effect. This "lag" in decision-making leads to a "mismatch" between resource allocation and rapidly changing actual needs, especially in high-speed moving scenarios. This can cause communication links to be interrupted due to sudden obstructions or the loss of detected targets due to failure to update beam pointing in a timely manner. Therefore, designing a forward-looking resource scheduling mechanism that can predict and proactively adapt to future changes in channel and target states to overcome the inherent delays and resource mismatches of reactive decision-making, thereby improving the timeliness and accuracy of scheduling, is a key challenge currently facing this field. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, to address the issues of lag and resource mismatch in existing reactive scheduling in dynamic scenarios, this invention provides a dynamic channel-aware sensing resource scheduling method, network device, and system.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for scheduling sensing resources based on dynamic channel awareness, which includes the following steps:

[0008] S1. Construct a probabilistic digital twin model: Construct and maintain in real time a probabilistic digital twin model that can simultaneously encompass the uncertainty of future channel state evolution and multiple possibilities of dynamic target trajectory in the form of a probability distribution.

[0009] S2. Generate candidate scheduling schemes: Based on the current service requirements of the network and around conflicting optimization goals, the range covers different strategy tendencies from pursuing the highest performance to pursuing the lowest energy consumption, generating candidate scheduling schemes with different design ideas. Each scheme is a pre-planned allocation plan for synesthetic resources for future scheduling cycles.

[0010] S3. Simulation and Performance Evaluation: Each candidate scheduling scheme is placed in the virtual future constructed by the probabilistic digital twin model. The performance of each scheme under various possibilities is tested by performing simulation. A detailed performance analysis report including expected benefits and potential risks is generated for each scheme.

[0011] S4. Final decision based on risk perception: Based on the performance analysis report, the candidate scheduling schemes are comprehensively evaluated and ranked using the utility function of risk perception, and the scheme with the best utility is selected as the final decision and executed.

[0012] As a preferred embodiment of the sensing resource scheduling method based on dynamic channel awareness described in this invention, the following steps are taken to construct and maintain in real time a probabilistic digital twin model that simultaneously encompasses the uncertainty of future channel state evolution and multiple possibilities of dynamic target motion trajectories in the form of a probability distribution:

[0013] The system comprises two parts: The first part is the tracking and prediction of dynamic targets. It employs a kinematic modeling framework and tracks targets through an iterative prediction-correction loop. In each loop, the kinematic modeling framework first makes a preliminary prediction based on the position at the previous moment, and then uses newly received sensing echoes to correct the preliminary prediction, continuously updating the target's state probability distribution and simultaneously outputting a covariance matrix that measures the reliability of the preliminary prediction. The second part is the modeling and prediction of wireless channels. It uses multimodal statistical methods to fit the complex time-varying characteristics of the channel. Based on measured channel state information, it continuously adjusts the parameters and weights of the internal probability components through an iterative optimization algorithm to form a comprehensive probability prediction model, which is used to deduce the overall evolution trend of the future channel state.

[0014] As a preferred embodiment of the dynamic channel-aware sensing resource scheduling method described in this invention, the following steps are taken: Based on the current network service requirements and considering conflicting optimization objectives, ranging from pursuing maximum performance to pursuing minimum energy consumption, candidate scheduling schemes with different design ideas are generated. Each scheme is a sensing resource allocation plan formulated for future scheduling cycles. The specific steps are as follows:

[0015] The comprehensive service requirements of the network are decomposed into clear, independent, and quantifiable optimization objectives, including maximizing communication throughput, minimizing sensing and positioning errors, minimizing transmission latency, and minimizing system energy consumption. An intelligent agent with efficient scheduling capabilities is employed. This agent learns how to formulate efficient sensing scheduling schemes by repeatedly trying and evaluating different resource scheduling decisions in an offline simulation environment. Different policy instructions are issued to the intelligent agent, guiding it to focus its optimization efforts on the core objectives during decision-making. Under the same network conditions, candidate scheduling schemes that prioritize communication, sensing, latency, and energy saving are generated.

[0016] As a preferred embodiment of the dynamic channel-aware sensing resource scheduling method described in this invention, the candidate scheduling schemes are placed in the virtual future constructed by the probabilistic digital twin model, and their performance under various possibilities is tested by performing simulations. A detailed performance analysis report including expected benefits and potential risks is generated for each scheme. The specific steps are as follows:

[0017] For the candidate scheduling schemes to be evaluated, parallel simulation threads are started. In the initial stage of each simulation thread, random sampling is performed independently from the probability distribution defined by the probabilistic digital twin model to generate a future event trajectory for the simulation thread that includes specific channel evolution and target motion details. Within each simulation thread, according to the resource allocation rules of the candidate scheduling scheme and under the future event trajectory dedicated to the simulation thread, the complete synesthetic task process is simulated and executed, and the communication and sensing performance indicators corresponding to the future event trajectory are calculated and recorded.

[0018] As a preferred embodiment of the dynamic channel-aware sensing resource scheduling method of the present invention, the candidate scheduling schemes are comprehensively evaluated and ranked according to the performance analysis report using a risk-aware utility function, and the scheme with the best utility is selected as the final decision and executed. The specific steps are as follows:

[0019] The performance results of each candidate scheduling scheme obtained from multiple simulations are statistically processed into performance distribution results that reflect the expected values ​​and distribution patterns of each key performance indicator. A risk-aware utility function is used to quantitatively evaluate the performance distribution results of each candidate scheme. The design of the utility function comprehensively reflects the evaluation items of its average performance level and combines risk items used to evaluate performance in a few worst-case scenarios. All candidate scheduling schemes are ranked according to the calculated scores of the utility function, and the optimal scheme is selected as the final scheduling decision.

[0020] As a preferred embodiment of the sensing resource scheduling method based on dynamic channel awareness described in this invention, the specific steps for evaluating sensing performance are as follows:

[0021] In independent simulations, virtual echo signals are generated based on the resource configuration set by the candidate scheduling scheme to be evaluated and in combination with future event trajectories. Based on the virtual echo signals and known transmitted signal waveforms, a key information matrix is ​​calculated to measure the amount of information about the target state contained in the echo signals. The inverse of the key information matrix, i.e., the theoretical minimum error value that can be achieved, is used as the final indicator for evaluating the perception performance of the candidate scheduling scheme under simulation.

[0022] As a preferred embodiment of the dynamic channel-aware communication resource scheduling method described in this invention, the specific steps for evaluating communication performance are as follows:

[0023] In independent simulations, based on the resource allocation set by the candidate scheduling scheme to be evaluated and combined with the future channel evolution trajectory, the clarity index of the communication signal is continuously calculated within a standard time segment. Based on the dynamic changes of the clarity index, the complete transmission process of data packets is simulated. The process includes: a strategy to adjust the data packaging method in real time according to the signal clarity, and an intelligent retransmission mechanism to ensure accurate data delivery. The simulated transmission results of all data packets throughout the entire scheduling cycle are statistically analyzed to obtain communication performance indicators that reflect the quality of service.

[0024] Secondly, the present invention provides a sensing resource scheduling network device based on dynamic channel awareness, characterized in that it includes: a processor; and a memory, wherein the memory stores a computer program.

[0025] Thirdly, the present invention provides a sensing resource scheduling system based on dynamic channel awareness, characterized in that:

[0026] An environmental detection beam generation module is configured to generate and control the emission of a low-power active detection beam with a specific spatial and frequency structure during breaks in performing routine sensing tasks. The active detection beam is used to excite a specific multipath scattering response in the wireless environment.

[0027] The differential response analysis module is configured to: accurately synchronize and capture the differential echo signal excited by the active detection beam, which contains information about minute environmental changes, and extract key environmental features about changes in non-line-of-sight path structure and the presence of dynamic scatterers in potential sensing blind spots from these extremely weak differential signals through highly sensitive correlation analysis.

[0028] The feature-driven scheduling decision module is configured to: use the environmental features extracted by the differential response analysis module as key context information to guide the scheduling of synesthetic resources, and directly drive the scheduling strategy to make real-time adaptive adjustments in the direction of being able to actively utilize these newly discovered environmental features, thereby improving the overall adaptability and performance of the system to complex dynamic environments without increasing conventional sensing overhead.

[0029] Fourthly, the present invention provides a preferred scheme for a sensing resource scheduling system based on dynamic channel awareness, wherein: the environmental detection beam generation module and the differential response analysis module, the specific steps of which are as follows:

[0030] A pair of baseband signals with mutually orthogonal coding sequences applied to their phases are generated, one signal serving as a reference signal and the other as a perturbation signal; the baseband signals are processed using beamforming weights to form two spatially overlapping probe beams for transmission; the probe beams are controlled to transmit at a power level lower than the background noise level and during the scheduling gaps of the sensing task to achieve low-interference active detection of the environment.

[0031] The received mixed echo signal is correlated with the local orthogonal coding sequences used to generate the reference beam and the perturbation beam, respectively, to separate two independent echo signals. The two separated echo signals are then subjected to coherent integration processing, in which the symmetrical echo component generated by static environmental reflection is suppressed, while the asymmetrical echo component caused by dynamic environmental changes forms a detectable energy leakage. By detecting and analyzing the characteristics of the energy leakage, key environmental feature information about weak dynamic events in the environment is extracted.

[0032] The beneficial effects of this invention are:

[0033] This invention achieves its goals by employing a forward-looking scheduling framework. This framework continuously collects historical data to construct and maintain in real-time a probabilistic digital twin model that describes the uncertainties of future channel evolution and the various possibilities of target movement in a probabilistic form. Based on this model, multiple simulations are performed to evaluate the performance distribution of different candidate scheduling schemes under various future possibilities. A risk-aware utility function is used to assess both average expected returns and potential risks in a few worst-case scenarios, thereby selecting the most effective scheme for decision-making. This invention elevates the scheduling mechanism from immediate decision-making based on the current state to forward-looking planning based on future state prediction, improving the timeliness and accuracy of resource scheduling and enhancing the perception performance of high-speed moving targets while ensuring the quality of communication services. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is an overall flowchart of the synesthetic resource scheduling method of the present invention;

[0036] Figure 2 Flowchart for constructing probabilistic digital twin models;

[0037] Figure 3 Generate flowcharts for candidate scheduling schemes;

[0038] Figure 4 A flowchart for performance evaluation of a single simulation;

[0039] Figure 5 This is a diagram of the network equipment structure.

[0040] Figure 6 This is a system structure diagram. Detailed Implementation

[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0042] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0043] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0044] Example 1

[0045] Reference Figures 1-4 This is the first embodiment of the present invention, which provides a method for scheduling sensing resources based on dynamic channel awareness, including the following steps:

[0046] S1. Construct and maintain in real time a probabilistic digital twin model that, in the form of a probability distribution, can simultaneously encompass the uncertainty of future channel state evolution and multiple possibilities of dynamic target trajectory. The specific operation steps are as follows:

[0047] The first part focuses on the tracking and prediction of dynamic targets. It employs a kinematic modeling framework, using an iterative prediction-correction loop to track targets. In each loop, the kinematic modeling framework first makes a preliminary prediction based on the position at the previous moment, then uses newly received sensing echoes to correct the preliminary prediction, continuously updating the target's state probability distribution and simultaneously outputting a covariance matrix that measures the reliability of the preliminary prediction. The second part focuses on the modeling and prediction of wireless channels. It uses multimodal statistical methods to fit the complex time-varying characteristics of the channel. Based on measured channel state information, iterative optimization algorithms continuously adjust the parameters and weights of the internal probability components to form a comprehensive probabilistic prediction model, which is used to deduce the overall evolution trend of the future channel state.

[0048] It should be noted that the first part focuses on the tracking and prediction of dynamic targets. This embodiment uses a state-space model based on the Extended Kalman Filter (EKF) as the kinematic modeling framework, which continuously tracks the target through an iterative "prediction-correction" cycle.

[0049] Prediction phase: in each cycle At the beginning, the model uses the previous time step Optimal state estimation (Including position, velocity, acceleration, etc.) Combined with state transition function This involves making a preliminary prediction of the target's current state. This state prediction process can be represented by the following formula:

[0050]

[0051] Wherein, the state vector It can be represented as , representing the target's position, velocity, and acceleration components on the two-dimensional plane, respectively. This is the control input at the current moment, which can be considered as the zero vector in this passive tracking scenario.

[0052] Correction phase: When a new sensing echo is received, the measurement value is extracted from it. The system then calculates the prediction residual between the measured value and the expected measured value derived from the predicted state. The EKF algorithm then uses this residual, along with the dynamically updated Kalman gain, to... The initial prediction is then finely revised to obtain a posterior state estimate that incorporates new observation information. This state update process can be represented by the following equation:

[0053]

[0054] in, Is The measurement vector at time, in this embodiment, is provided by a radar sensor, and its specific form is as follows: , and These represent the target's slant range and azimuth, respectively. A nonlinear observation function that maps the state vector to the measurement space; It is a weight matrix dynamically calculated based on prediction error and measurement noise, used to balance the reliability of predicted and measured values.

[0055] Uncertainty Quantification: During the above process, EKF simultaneously outputs a covariance matrix. This matrix mathematically describes the degree of uncertainty in the current state estimate, and its size and shape intuitively reflect the reliability of the prediction. For example, a larger covariance matrix means that the current prediction has a larger error range, and the system has less certainty about the target location.

[0056] The second part focuses on modeling and predicting wireless channels. This embodiment uses a Gaussian mixture model (GMM) as a multimodal statistical method to fit the complex time-varying characteristics of the channel. Wireless channels may exhibit different statistical characteristics at different times.

[0057] Modeling process: The GMM model consists of multiple weighted Gaussian probability components, each of which can be regarded as representing a typical channel state. Based on the continuously measured Channel State Information (CSI) values ​​over a period of time, the model continuously adjusts the mean, variance, and weight of each Gaussian component through iterative optimization algorithms such as Expectation-Maximization (EM).

[0058] Predictive Applications: After training, this GMM model forms a comprehensive probabilistic prediction model. When it is necessary to predict the overall evolution trend of future channel states, the system can sample from this weighted composite probability distribution to generate a large number of future evolution trajectories that conform to the statistical characteristics of real channels.

[0059] S2. Based on the current network service requirements and considering conflicting optimization goals, ranging from pursuing maximum performance to minimizing energy consumption, candidate scheduling schemes with different design approaches are generated. Each scheme is a contingency plan for the allocation of synesthetic resources for future scheduling cycles. The specific operation steps are as follows:

[0060] The comprehensive service requirements of the network are decomposed into clear, independent, and quantifiable optimization objectives, including maximizing communication throughput, minimizing sensing and positioning errors, minimizing transmission latency, and minimizing system energy consumption. An intelligent agent with efficient scheduling capabilities is adopted. The agent learns how to formulate efficient sensing scheduling schemes by repeatedly trying and evaluating different resource scheduling decisions in an offline simulation environment. Different policy instructions are issued to the agent to guide it to focus on the core objectives when making decisions. Under the same network conditions, candidate scheduling schemes that reflect different priorities such as communication, sensing, latency, and energy saving are generated.

[0061] It should be noted that, based on the current network service requirements (e.g., a user needs high-bandwidth video transmission while simultaneously requiring high-precision perception of the surrounding environment), candidate scheduling schemes with different design approaches are generated around conflicting optimization objectives. In this embodiment, the specific implementation is as follows:

[0062] The first step is to decompose the network's overall service requirements into clear, independent, and quantifiable core optimization objectives. This embodiment defines four core objectives: maximizing downlink communication throughput, minimizing the root mean square error (RMSE) of target positioning, minimizing end-to-end transmission latency of data packets, and minimizing the total transmit power of the base station. These four objectives inherently conflict; for example, improving sensing accuracy typically requires allocating more power and bandwidth, which conflicts with the objectives of maximizing communication throughput and minimizing energy consumption.

[0063] The second step is to train an intelligent agent with decision-making capabilities.

[0064] Training Environment: The agent's training takes place entirely in an offline simulation environment. At the heart of this environment is the probabilistic digital twin model built in Phase S1. The agent, acting as the decision-maker, interacts with this virtual world that simulates the future.

[0065] The learning process is a training process involving extensive interactive trial and error. The agent observes the current system state (including channel information, target location, etc.) and makes resource scheduling decisions. Based on this decision, the digital twin model deduces possible outcomes and feeds back a reward signal to the agent. This reward signal is a scalar value, the magnitude of which depends on the extent to which the current decision achieves the preset optimization objective.

[0066] Learning Objective: The learning objective of the agent is not to maximize the immediate single-step reward, but to maximize the overall reward value that balances long-term cumulative gains and policy exploration. The reward function design not only includes a weighted evaluation of the four core optimization objectives mentioned above, but also introduces a policy entropy term as an exploration incentive. This entropy term encourages the agent to try more diverse and untried decisions in the early stages of training, avoiding premature convergence to a suboptimal policy. Through repeated cycles of "simulated decision-deduction-reward," the agent gradually learns how to weigh multiple conflicting objectives in complex dynamic environments, thus mastering the intrinsic strategy for efficiently formulating synesthetic resource scheduling schemes. Its learning process aims to maximize the expected long-term cumulative reward. The objective can be defined by the following formula:

[0067]

[0068] In this embodiment, the instant reward function The design is a weighted summation of multiple core optimization objectives, and its specific form can be expressed as:

[0069]

[0070] in:

[0071] This represents the expected value for all possible trajectories. The discount factor is a number between 0 and 1 used to weigh the importance of future rewards; in this embodiment, it can be 0.99. It is in state Execute scheduling actions The instant reward received afterward.

[0072] , , , These are the normalized quantized values ​​of communication throughput, sensing error, transmission delay, and system energy consumption.

[0073] , , , These are the corresponding weight coefficients, which can be configured according to balance requirements, for example, as a vector [0.4, 0.3, 0.2, 0.1].

[0074] The third step involves issuing different policy instructions to the trained agent, guiding it to generate diverse solutions. The system will periodically modify the weights in the reward function corresponding to each core objective. For example, when a "communication-first" solution is needed, the system will maximize the weight of the "maximizing communication throughput" term; when a "energy-saving-first" solution is needed, the system will maximize the weight of the "minimizing system energy consumption" penalty term. In this way, the agent is guided to focus its optimization efforts on each of the aforementioned core objectives during decision-making. Consequently, under the same network conditions, candidate scheduling solutions that embody different design philosophies, such as communication-first, perception-first, latency-first, and energy-saving-first, are generated. Each solution is a complete sensory resource allocation plan formulated for a future scheduling cycle (e.g., 100 milliseconds).

[0075] S3. Place the candidate scheduling schemes in a virtual future constructed by a probabilistic digital twin model, and test their performance under various possibilities by performing simulations. Generate a detailed performance analysis report for each scheme, including expected benefits and potential risks. The specific steps are as follows:

[0076] For the candidate scheduling schemes to be evaluated, parallel simulation threads are started. In the initial stage of each simulation thread, random sampling is performed independently from the probability distribution defined by the probabilistic digital twin model to generate future event trajectories containing specific channel evolution and target motion details for the simulation thread. Within each simulation thread, according to the resource allocation rules of the candidate scheduling scheme, and under the future event trajectory dedicated to the simulation thread, the complete synesthetic task process is simulated and executed, and the communication and sensing performance indicators corresponding to the future event trajectory are calculated and recorded.

[0077] It should be noted that candidate scheduling schemes with different styles are generated first, followed by a comprehensive robustness evaluation. In this embodiment, this step is specifically implemented as follows: each candidate scheduling scheme is placed in the virtual future of the probabilistic digital twin model constructed in step S1, and its performance under various possibilities is comprehensively tested by performing massive parallel simulations.

[0078] The first step is to implement parallel simulation: For the candidate scheduling scheme to be evaluated, the system will start (for example, 1000) parallel simulation threads.

[0079] At the initial stage of each simulation thread, the system independently samples randomly from the probability distribution defined by the probabilistic digital twin model. Each thread receives a unique scenario instance, i.e., a future event trajectory. This trajectory specifies in detail how the channel quality will fluctuate and how the dynamic target will move specifically (e.g., whether it accelerates straight ahead or decelerates to turn right) during the following scheduling cycle.

[0080] Within each simulation thread, the system strictly adheres to the resource allocation rules of the current candidate scheduling scheme (e.g., time-frequency resources and power levels allocated to communication and sensing) to simulate the complete synesthetic task process within the thread's dedicated scenario instance. During this process, the system calculates and records in detail the communication and sensing performance metrics achievable by the scheme under that specific trajectory.

[0081] The second step is the evaluation method for perception performance: In each independent simulation, in order to efficiently evaluate perception performance, this embodiment adopts an evaluation method based on information theory.

[0082] Based on the resource allocation (such as transmission power, waveform duration, etc.) allocated to the sensing task by the candidate scheme, and combined with the future event trajectory specific to this simulation (especially the target's position and reflection characteristics), a virtual echo signal containing target reflection information and environmental noise was generated.

[0083] The system does not perform computational estimation algorithms; instead, it calculates the target state parameters based on the virtual echo signal and the known transmitted signal waveform. The key information matrix (such as distance and speed) is the Fisher Information Matrix (FIM). This matrix measures the maximum amount of information about the target's true state that can be obtained from the current noisy echo signal. The effective information. The Fisher information matrix is ​​defined as:

[0084]

[0085] in, It is the likelihood function, which describes the likelihood of the true state being... At that time, the measured value was observed. The probability of this. In this embodiment, it can be assumed that the measurement noise follows a Gaussian distribution, i.e. As a Mean, noise variance Let be the probability density function of the normal distribution of variance.

[0086] By inverting the key information matrix, the theoretical minimum error value, also known as the Cramer-Rao lower bound (CRLB), is obtained. The smaller this value, the higher the potential perception performance of the scheme in this simulation. The theoretical minimum error value is used as the final indicator for evaluating perception performance. The relationship between the Cramer-Rao lower bound (CRLB) and the FIM is expressed by the following formula:

[0087]

[0088] in, Represents the parameter Any unbiased estimator The variance (i.e., mean squared error) is given by the formula. This formula shows that the minimum estimation error is determined by the inverse of the Fisher information matrix. This embodiment directly uses this theoretical minimum error value as the final metric for evaluating perception performance.

[0089] The third step is the evaluation method for communication performance: in each independent simulation, the evaluation of communication performance simulates the transmission process of the real physical layer.

[0090] Based on the resource allocation of the candidate schemes to the communication task, and combined with the future channel evolution trajectory specific to this simulation, the system will continuously calculate the clarity index of the communication link, namely the instantaneous signal-to-interference-plus-noise ratio, within a series of extremely small time segments.

[0091] Based on this dynamically changing resolution metric curve over time, the system simulates the complete transmission process of each data packet. This simulation process is highly detailed and includes:

[0092] The most suitable data packaging method (i.e., modulation and coding strategy, MCS) is selected in real time based on the current signal clarity. For example, when the signal is good, a higher-order modulation (such as 256-QAM) is used to increase the data rate, while when the signal is poor, a more robust lower-order modulation (such as QPSK) is switched.

[0093] An intelligent retransmission mechanism is adopted to ensure accurate data delivery. If a transmission fails, the receiving end can save the received error information and merge it with the subsequent retransmitted data for decoding to improve the success rate.

[0094] Through the above simulation, the system can accurately determine the final transmission result (success, failure, or still in transmission) of each data packet and its end-to-end delay.

[0095] The fourth step is to generate a performance analysis report: After all simulation threads have finished executing, the performance of each candidate scheduling scheme under different scenario instances is summarized, and a detailed performance analysis report is finally generated. This report reveals the expected benefits (e.g., average throughput, average perceived error) and potential risks (e.g., how high the latency will spike in the 5% worst case; or what the maximum possible perceived error is) of the scheme in the form of statistical distribution.

[0096] S4. Based on the performance analysis report, a risk-aware utility function is used to comprehensively evaluate and rank the candidate scheduling schemes, and the scheme with the best utility is selected as the final decision and executed. The specific operation steps are as follows:

[0097] The performance results of each candidate scheduling scheme obtained from multiple simulations are statistically processed into performance distribution results that reflect the expected values ​​and distribution patterns of each key performance indicator. A risk-aware utility function is used to quantitatively evaluate the performance distribution results of each candidate scheme. The design of the utility function comprehensively reflects the evaluation items of its average performance level, and combines risk items used to evaluate performance in a few worst-case scenarios. All candidate scheduling schemes are ranked according to the calculated scores of the utility function, and the optimal scheme is selected as the final scheduling decision.

[0098] It should be noted that, based on the performance analysis report generated in step S3, the system will use a risk perception assessment model to finally score and rank all candidate solutions, and select the optimal solution for execution. This "risk perception assessment model" is specifically implemented through a risk perception utility function. Unlike traditional decision-making methods that only focus on average performance, this utility function design integrates considerations from two dimensions:

[0099] The first dimension is an evaluation item that reflects its average performance level, namely the expected value of each key performance indicator (e.g., average throughput, average latency, average perceived error). This evaluation item represents the "average benefit" that the solution can typically bring.

[0100] The second dimension is a risk item used to assess its performance in a few worst-case scenarios. This embodiment uses "Conditional Value at Risk" (VaR). ( ) as the specific implementation of the risk item.

[0101] The utility function calculates a total score for each candidate solution, which is a weighted sum of its "average benefit" and "risk" components. The weights can be adjusted based on the reliability requirements of the current business. For example, for safety-critical applications such as autonomous driving, the "risk" component will have a very high weight. Ultimately, the solution that achieves the optimal balance between risk and benefit and has the highest utility function score will be determined as the final scheduling decision and executed by the network equipment. The defining formula is:

[0102]

[0103] in,

[0104] This refers to the candidate scheduling schemes to be evaluated.

[0105] This is the expected benefit of the scheme, such as the average communication throughput obtained in all simulations.

[0106] This is the conditional value-at-risk (VAT) component of the plan, which measures its cost / risk ratio. (For example, end-to-end latency) in the worst case Average performance under % of conditions.

[0107] This refers to the risk level, for example, 5%.

[0108] and These are configurable weighting coefficients used to balance the pursuit of returns and the avoidance of risks. For example, for ordinary internet access services, a weighting coefficient of 100% can be used. =0.8, =0.2; for autonomous driving safety services, then it can be taken as 0.2. =0.1, =0.9.

[0109] Conditional Value at Risk The definition is as follows, which calculates random variables. Expected conditions after (e.g., delay) exceeds its Value at Risk (VaR):

[0110]

[0111] in, yes In the distribution, located in (1- The value at the quantile.

[0112] Example 2

[0113] Embodiment 2 of the present invention provides a sensing resource scheduling network device based on dynamic channel awareness, referring to... Figure 5 The network device 100 is physically manifested as an intelligent sensing base station and a roadside unit. Its characteristic is that the device includes a processor 110 and a memory 120.

[0114] In terms of hardware configuration, the network device 100 includes a general computing and radio frequency architecture. Its core components are a processor 110, a memory 120, and a radio frequency front-end (communication interface 130) and an antenna array 140 for transmitting and receiving wireless signals.

[0115] Processor 110 is the core of the device's operation. It is a combination of a central processing unit and a digital signal processor, and is responsible for executing complex algorithms and logic control.

[0116] The memory 120 is a combination of non-volatile storage media (such as flash memory) and volatile storage media (such as RAM) for storing the operating system, application programs, and computer program instructions for implementing the method of the present invention.

[0117] In terms of functional implementation, the computer program stored in memory 120, when loaded and executed by processor 110, fully implements the forward-looking synesthetic resource scheduling method based on "prediction-evaluation-decision" as described in Embodiment 1. Specifically, processor 110, by executing the corresponding program modules, undertakes the following responsibilities:

[0118] Building and maintaining probabilistic digital twin models: The processor continuously processes the sensed echoes from the antenna array and the channel state information from the terminal to run algorithms such as extended Kalman filtering and Gaussian mixture models, and updates the probabilistic predictions of future channel and target states in real time.

[0119] Generate and evaluate candidate scheduling schemes: The processor runs pre-trained deep reinforcement learning agents to generate diverse candidate schemes, initiates parallel simulations, and performs a comprehensive robustness evaluation of these schemes.

[0120] Final decision based on risk perception: The processor uses the utility function of risk perception to score and rank the evaluation results of all candidate solutions, selects the solution with the best utility, and generates specific resource scheduling instructions.

[0121] Therefore, network device 100 itself constitutes a complete functional entity capable of autonomously executing forward-looking and intelligent resource scheduling. It possesses the ability to perform deduction and optimization decisions based on internal models, and can adaptively respond to high-speed and dynamic wireless environments without the need for complex external coordination.

[0122] Example 3

[0123] Reference Figure 6 This is a third embodiment of the present invention, which provides a sensing resource scheduling system based on dynamic channel awareness, comprising:

[0124] The environmental detection beam generation module is configured to generate and control the emission of low-power active detection beams with specific spatial and frequency structures during the intervals of performing routine sensing tasks. These beams are not used for data transmission or target sensing, but are specifically designed to excite specific multipath scattering responses in the wireless environment.

[0125] The differential response analysis module is configured to: accurately synchronize and capture differential echo signals excited by the active probe beam that contain information about minute environmental changes, and extract key environmental features about changes in non-line-of-sight path structure and the presence of dynamic scatterers in potential sensing blind spots from these extremely weak differential signals through highly sensitive correlation analysis.

[0126] The feature-driven scheduling decision module is configured to use the environmental features extracted by the differential response analysis module as key contextual information to guide the scheduling of synesthetic resources. This directly drives the scheduling strategy to make real-time adaptive adjustments in the direction of actively utilizing or avoiding these newly discovered environmental features, thereby improving the overall adaptability and performance of the system to complex dynamic environments without increasing conventional sensing overhead.

[0127] In summary, this invention achieves its goals by employing a forward-looking scheduling framework. This framework continuously collects historical data to construct and maintain in real-time a probabilistic digital twin model that describes the uncertainties of future channel evolution and the various possibilities of target movement in a probabilistic form. Based on this model, multiple simulations are performed to evaluate the performance distribution of different candidate scheduling schemes under various future possibilities. A risk-aware utility function is used to assess both the average expected return and the potential risks in a few worst-case scenarios, thereby selecting the most effective scheme for decision-making. This invention elevates the scheduling mechanism from immediate decision-making based on the current state to forward-looking planning based on future state prediction, improving the timeliness and accuracy of resource scheduling and enhancing the perception performance of high-speed moving targets while ensuring the quality of communication services.

[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for scheduling sensing resources based on dynamic channel awareness, characterized in that, Includes the following steps: S1. Construct a probabilistic digital twin model: Construct and maintain in real time a probabilistic digital twin model that is in the form of a probability distribution and simultaneously includes the uncertainty of future channel state evolution and multiple possibilities of dynamic target motion trajectory. S2. Generate candidate scheduling schemes: Based on the current service requirements of the network and around conflicting optimization goals, the range covers different strategy tendencies from pursuing the highest performance to pursuing the lowest energy consumption, generating candidate scheduling schemes with different design ideas. Each scheme is a pre-planned allocation plan for synesthetic resources for future scheduling cycles. S3. Simulation and Performance Evaluation: The candidate scheduling schemes are placed in the virtual future constructed by the probabilistic digital twin model. The performance of the schemes under various possibilities is tested by performing simulations. A detailed performance analysis report including expected benefits and potential risks is generated for each scheme. S4. Final decision based on risk perception: Based on the performance analysis report, the candidate scheduling schemes are comprehensively evaluated and ranked using the utility function of risk perception, and the scheme with the best utility is selected as the final decision and executed.

2. The sensing resource scheduling method based on dynamic channel awareness as described in claim 1, characterized in that, The construction and real-time maintenance of a probabilistic digital twin model includes the following two parts: The first part focuses on the tracking and prediction of dynamic targets. A kinematic modeling framework is employed, using an iterative prediction-correction loop to track the target. In each loop, the kinematic modeling framework first makes a preliminary prediction based on the previous position, then corrects the preliminary prediction using newly received sensor echoes, continuously updating the target's state probability distribution and simultaneously outputting a covariance matrix that measures the reliability of the preliminary prediction. The second part focuses on the modeling and prediction of wireless channels. A multimodal statistical method is used to fit the complex time-varying characteristics of the channel. Based on measured channel state information, an iterative optimization algorithm continuously adjusts the parameters and weights of the internal probability components to form a comprehensive probabilistic prediction model, used to deduce the overall evolution trend of the future channel state.

3. The sensing resource scheduling method based on dynamic channel awareness as described in claim 1, characterized in that, The specific implementation of generating candidate scheduling schemes includes: The comprehensive service requirements of the network are decomposed into clear, independent, and quantifiable optimization objectives, including maximizing communication throughput, minimizing sensing and positioning errors, minimizing transmission latency, and minimizing system energy consumption. An intelligent agent with efficient scheduling capabilities is employed. This agent learns how to formulate efficient sensing scheduling schemes by repeatedly trying and evaluating different resource scheduling decisions in an offline simulation environment. Different policy instructions are issued to the intelligent agent, guiding it to focus its optimization efforts on the core objectives during decision-making. Under the same network conditions, candidate scheduling schemes that prioritize communication, sensing, latency, and energy saving are generated.

4. The sensing resource scheduling method based on dynamic channel awareness as described in claim 1, characterized in that, The specific implementation of parallel execution of multiple simulations includes: For the candidate scheduling schemes to be evaluated, parallel simulation threads are started. In the initial stage of each simulation thread, random sampling is performed independently from the probability distribution defined by the probabilistic digital twin model to generate a future event trajectory for the simulation thread that includes specific channel evolution and target motion details. Within each simulation thread, according to the resource allocation rules of the candidate scheduling scheme and under the future event trajectory dedicated to the simulation thread, the complete synesthetic task process is simulated and executed, and the communication and sensing performance indicators corresponding to the future event trajectory are calculated and recorded.

5. The sensing resource scheduling method based on dynamic channel awareness as described in claim 1, characterized in that, The specific implementation of comprehensive evaluation and ranking based on the utility function of risk perception includes: The performance results of each candidate scheduling scheme obtained from multiple simulations are statistically processed into performance distribution results that reflect the expected values ​​and distribution patterns of each key performance indicator. A risk-aware utility function is used to quantitatively evaluate the performance distribution results of each candidate scheme. The design of the utility function comprehensively reflects the evaluation items of its average performance level and combines risk items used to evaluate performance in a few worst-case scenarios. All candidate scheduling schemes are ranked according to the calculated scores of the utility function, and the optimal scheme is selected as the final scheduling decision.

6. The sensing resource scheduling method based on dynamic channel awareness as described in claim 4, characterized in that, The specific implementation for evaluating perceived performance includes: In independent simulations, virtual echo signals are generated based on the resource configuration set by the candidate scheduling scheme to be evaluated and in combination with future event trajectories. Based on the virtual echo signals and known transmitted signal waveforms, a key information matrix is ​​calculated to measure the amount of information about the target state contained in the echo signals. The inverse of the key information matrix, i.e., the theoretical minimum error value that can be achieved, is used as the final indicator for evaluating the perception performance of the candidate scheduling scheme under simulation.

7. The sensing resource scheduling method based on dynamic channel awareness as described in claim 6, characterized in that, The specific implementation for evaluating communication performance includes: In independent simulations, based on the resource allocation set by the candidate scheduling scheme to be evaluated and combined with the future channel evolution trajectory, the clarity index of the communication signal is continuously calculated within a standard time segment. Based on the dynamic changes of the clarity index, the complete transmission process of data packets is simulated. The process includes: a strategy to adjust the data packaging method in real time according to the signal clarity, and an intelligent retransmission mechanism to ensure accurate data delivery. The simulated transmission results of all data packets throughout the entire scheduling cycle are statistically analyzed to obtain communication performance indicators that reflect the quality of service.

8. A sensory resource scheduling network device based on dynamic channel awareness, characterized in that, include: processor; And a memory storing a computer program, wherein the computer program is configured to, when executed by the processor, implement the method as described in any one of claims 1 to 7.

9. A sensory resource scheduling system based on dynamic channel awareness, comprising a sensory resource scheduling method based on dynamic channel awareness as described in any one of claims 1 to 7, characterized in that, The system includes: An environmental detection beam generation module is configured to generate and control the emission of a low-power active detection beam with a specific spatial and frequency structure during breaks in performing routine sensing tasks. The active detection beam is used to excite a specific multipath scattering response in the wireless environment. The differential response analysis module is configured to: accurately synchronize and capture the differential echo signal excited by the active detection beam, which contains information about minute environmental changes, and extract key environmental features about changes in non-line-of-sight path structure and the presence of dynamic scatterers in potential sensing blind spots from these extremely weak differential signals through highly sensitive correlation analysis. The feature-driven scheduling decision module is configured to: use the environmental features extracted by the differential response analysis module as key context information to guide the scheduling of synesthetic resources, and directly drive the scheduling strategy to make real-time adaptive adjustments in the direction of being able to actively utilize these newly discovered environmental features, thereby improving the overall adaptability and performance of the system to complex dynamic environments without increasing conventional sensing overhead.

10. The sensing resource scheduling system based on dynamic channel awareness as described in claim 9, characterized in that, The specific implementation of the environmental detection beam generation module and the differential response analysis module includes: A pair of baseband signals with mutually orthogonal coding sequences applied to their phases are generated, one signal serving as a reference signal and the other as a perturbation signal; the baseband signals are processed using beamforming weights to form two spatially overlapping probe beams for transmission; the probe beams are controlled to transmit at a power level lower than the background noise level and during the scheduling gaps of the sensing task to achieve low-interference active detection of the environment. The received mixed echo signal is correlated with the local orthogonal coding sequences used to generate the reference beam and the perturbation beam, respectively, to separate two independent echo signals. The two separated echo signals are then subjected to coherent integration processing, in which the symmetrical echo component generated by static environmental reflection is suppressed, while the asymmetrical echo component caused by dynamic environmental changes forms a detectable energy leakage. By detecting and analyzing the characteristics of the energy leakage, key environmental feature information about weak dynamic events in the environment is extracted.

Images