A multi-modal communication scheduling method in an assisted driving scenario

Abstract

This invention belongs to the field of regulation system technology, specifically relating to a multimodal communication scheduling method in assisted driving scenarios. The method includes: constructing a spatial wave dynamics modeling module to predict signal blind spots and multipath interference peak areas within a future time window using wave equations; constructing a digital twin pre-simulation module to simulate communication load and generate a pre-scheduling timetable through a forward-running digital twin; and performing dynamic channel binding and arbitration, whereby physical vehicles actively initiate cross-modal channel binding according to the timetable, and the arbitrator dynamically adjusts the reporting frequency and compression rate of sensor data. This invention ensures zero-interruption transmission of critical commands in blind spots or areas of strong interference, enhancing the robustness and safety of the assisted driving system.

Description

Technical Field

[0001] This invention belongs to the field of regulation system technology, specifically relating to a multimodal communication scheduling method in assisted driving scenarios. Background Technology

[0002] With the evolution of intelligent connected vehicles and driver assistance technologies, multimodal communication systems have become the infrastructure for ensuring driving safety and collaborative perception. By integrating vehicle-to-infrastructure (V2I) networks, onboard sensor data, and environmental perception information, these systems provide real-time and massive data support for autonomous driving decisions. In complex urban traffic environments, the collaborative scheduling capabilities between heterogeneous networks determine the robustness and reliability of driver assistance systems.

[0003] Multimodal communication scheduling technology dynamically manages the bandwidth, latency, and packet loss rate of different communication links to ensure deterministic transmission of critical driving commands and environmental perception data in variable channel environments. Because assisted driving scenarios involve high-speed movement and complex physical space changes, communication systems need to efficiently integrate multimodal resources to cope with the rapidly changing channel quality fluctuations during vehicle operation, establishing an efficient mapping mechanism between the physical and protocol layers.

[0004] Traditional communication scheduling strategies typically employ a reactive allocation mechanism, relying solely on currently reported channel state indicators for delayed adjustments. This results in scheduling decisions failing to match the sudden physical environmental changes that vehicles traverse. Furthermore, existing technologies lack in-depth modeling of the electromagnetic characteristics of physical space, making it difficult to overcome the random interference of wireless channel fading. This leaves the scheduling process in a state of reactive compensation, leading to communication interruptions or loss of critical data in highly dynamic scenarios. Traditional methods cannot achieve a deep correlation between sensor load and predicted channel state, making it difficult to pre-schedule for signal blind spots within future time windows. Consequently, the overall resource allocation of the system lacks foresight and accuracy, failing to meet the zero-interruption transmission requirements of driving scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a multimodal communication scheduling method for assisted driving scenarios, which can effectively solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a multimodal communication scheduling method in assisted driving scenarios, comprising the following specific steps: Step 1: Construct a space wave dynamics modeling module. Using the wave equation in structural mechanics, model the physical environment along the vehicle's driving path as rigid obstacles and absorbing boundaries that affect the propagation of electromagnetic waves. By solving the wave equation, predict the signal blind zone and multipath interference peak area that the vehicle will encounter in the future within a preset time window. Step 2: Construct a digital twin pre-simulation module. Based on the prediction results of the signal blind zone and the multipath interference peak zone, construct a digital twin in the virtual space that runs in parallel with the physical vehicle. The digital twin shares sensor data with the physical vehicle and its operation sequence is ahead of the physical vehicle. Based on the predicted channel state, it pre-simulates the communication load and packet loss risk of the multimodal sensor and generates a pre-scheduled schedule. Step 3: Perform dynamic channel binding and arbitration. The physical vehicle actively initiates cross-modal channel binding before entering the signal blind zone according to the pre-scheduled schedule. At the same time, the arbitrator dynamically adjusts the reporting frequency and compression rate of sensor data according to the predicted multipath interference peak area.

[0007] Preferably, in step 1, during the process of modeling the physical environment using the wave equation in structural mechanics, the physical environment includes buildings, vegetation, and large moving targets. The modeling process defines the buildings and the large moving targets as rigid obstacles with preset reflection coefficients, and the vegetation as absorbing boundaries with preset attenuation coefficients. By establishing a spatial grid in the Cartesian coordinate system, the propagation process of electromagnetic waves is described as a scalar field fluctuation that evolves over time. Drawing on the logic of displacement continuity and stress balance at the interface of elastic waves in structural mechanics, and analogous to the medium interface conditions applied to the scalar field of electromagnetic waves, the continuous conservation of field quantity and flux is ensured, and the field strength distribution law of electromagnetic waves in complex spatial structures is calculated.

[0008] Preferably, the process of solving the wave equation in step 1 involves discretizing the geographical area, dividing the continuous physical space into multiple finite element units, and introducing a time-domain finite difference algorithm to numerically simulate the wave characteristics of electromagnetic waves. The numerical simulation process considers the multipath reflection effect, diffraction effect, and scattering effect on the vehicle's driving path. By integrating the vehicle displacement vector within a future preset time window, the received signal strength indication of the vehicle at different geographical coordinate points is determined, thereby identifying areas where the received signal strength is lower than a preset threshold as signal blind zones.

[0009] Preferably, the prediction of the multipath interference peak region in step 1 is achieved by analyzing the phase superposition relationship of electromagnetic waves propagating along different paths at the receiving end. When the coherent waves of multiple reflection paths are superimposed in phase at a spatial location, the spatial location is determined to be the interference peak region. The interference peak region corresponds to the region of severe fluctuation of Rayleigh fading or Rice fading of the signal. By establishing a mapping matrix between spatial location and fading depth, physical layer prediction support is provided for subsequent scheduling strategies.

[0010] Preferably, the digital twin constructed in step 2 operates at a speed in virtual space that is consistently faster than the actual speed of the physical vehicle. The speed lead is determined by a preset time lead. The digital twin integrates a communication protocol stack model, a perception system model, and a decision logic model that are completely identical to those of the physical vehicle. By receiving real-time data on the current position, speed, acceleration, and steering angle reported by the physical vehicle, the digital twin performs advanced simulations on the virtual environment map to obtain the radio frequency environment parameters that the physical vehicle will experience in the future within a preset time period.

[0011] Preferably, step 2 involves pre-simulating the communication load and packet loss risk of multimodal sensors. Specifically, this includes classifying the output data streams of vehicle-mounted cameras, vehicle-mounted millimeter-wave radar, vehicle-mounted lidar, and vehicle-road cooperative communication terminals. Based on the real-time generation rate of each sensor's data and the preset quality of service requirements, combined with the predicted channel state parameters, the data transmission delay and packet loss probability under channel bandwidth limitations are calculated. The channel state parameters include the predicted bandwidth capacity, signal-to-noise ratio, and frequency-selective fading characteristics.

[0012] Preferably, the pre-scheduling schedule in step 2 includes a series of instruction sets arranged in chronological order. Each instruction corresponds to a preset timestamp. The instruction set defines the activation status, transmission power, modulation and coding scheme, and data service type of each communication mode at different time nodes. The pre-scheduling schedule is distributed to the communication control unit of the physical vehicle through a preset synchronization mechanism to ensure that the physical vehicle can accurately execute the corresponding communication behavior changes when crossing spatial locations.

[0013] Preferably, the proactive initiation of cross-modal channel binding in step 3 refers to the system logic layer automatically redirecting the high-priority service flow originally carried in the first communication mode to the idle resource block of the second communication mode before the physical vehicle senses that it is about to enter the predicted signal blind zone. The first communication mode includes vehicle-to-everything communication based on cellular networks, and the second communication mode includes lateral communication links based on millimeter-wave radar sensing or dedicated short-range communication links, by establishing a multipath transmission control protocol at the logic link layer.

[0014] Preferably, in step 3, the arbitrator dynamically adjusts the reporting frequency and compression rate of sensor data according to a priority scheduling algorithm. When it is predicted that the vehicle is about to enter the multipath interference peak area and the channel quality is expected to drop to a preset critical value, the arbitrator performs an importance assessment on the sensor data. For perception data related to driving safety, the arbitrator maintains the original reporting frequency and reduces the transmission ratio of non-critical redundant information. For environmental description data that is not related to safety, the arbitrator performs high-rate spatial downsampling or time compression to reduce the occupation of limited bandwidth and ensure deterministic transmission of core control commands in harsh channel environments.

[0015] Preferably, the specific method for adjusting the compression ratio in step 3 is as follows: based on the predicted signal-to-noise ratio change trend, a target compression ratio is selected from multiple preset compression levels. There is a negative correlation between the compression level and the signal-to-noise ratio, that is, the lower the signal-to-noise ratio, the higher the compression ratio corresponding to the selected compression level. The compressed data packet is sent after passing through preset redundant error correction coding to improve the anti-interference capability in the interference area.

[0016] Preferably, the length of the future preset time window in step 1 is dynamically adjusted according to the vehicle's current driving speed. When the vehicle speed is higher than the preset speed threshold, the length of the time window is automatically extended to compensate for the increased rate of environmental change caused by high-speed movement. When the vehicle speed is lower than the preset speed threshold, the length of the time window is shortened to improve the utilization rate of computing resources and the accuracy of prediction.

[0017] Preferably, the shared data between the digital twin and the physical vehicle in step 2 also includes historical channel quality records. The digital twin uses a long short-term memory neural network model to extract features from the historical channel quality records and combines them with the current wave dynamics prediction results to perform weighted correction on the predicted channel state, thereby reducing the prediction bias caused by the simplification of physical environment modeling.

[0018] Preferably, the arbitration logic for cross-modal channel binding in step 3 also considers the power consumption indicators and cost of different communication modes. Under the premise of meeting the preset requirements for secure transmission, the arbitrator prioritizes the communication link combination with the lowest power consumption or the lowest cost. By constructing a multi-objective optimization function, it seeks the optimal balance between reliability, latency, power consumption and cost.

[0019] Preferably, the modeling of rigid obstacles in step 1 also includes setting the electromagnetic parameters of the object's surface material. Different dielectric constants and permeabilities are set for metal, concrete, and glass materials respectively. By introducing a loss term into the wave equation, the energy absorption process of electromagnetic waves when penetrating obstacles of different materials is simulated, thereby improving the accuracy of signal strength prediction under non-line-of-sight transmission paths.

[0020] Preferably, the pre-scheduling schedule generated in step 2 has a feedback correction mechanism during the execution of the physical vehicle. If the deviation between the channel quality actually perceived by the physical vehicle and the predicted value in the pre-scheduling schedule exceeds a preset deviation threshold, the physical vehicle will trigger a real-time interruption request. The digital twin will then re-perform wave dynamics simulation and pre-playback based on the measured data fed back by the current physical vehicle and issue the corrected pre-scheduling schedule.

[0021] Preferably, in the process of dynamically adjusting the reporting frequency in step 3, for the video stream captured by the vehicle-mounted camera, the arbitrator determines the frame extraction ratio by identifying the density of key targets in the image. When there are obstacles in the image that are within the collision risk area, a high frame rate transmission is forcibly maintained. When the image contains only a static background, the transmission frequency is significantly reduced, and the saved bandwidth resources are allocated to the raw point cloud data of the millimeter-wave radar.

[0022] Preferably, in step 1, the space wave dynamics modeling module periodically obtains the latest high-precision map data from the cloud server. The high-precision map data contains physical entity information with three-dimensional semantic labels, and the modeling module converts the semantic labels into boundary parameters in wave dynamics calculations.

[0023] Preferably, the process of simulating packet loss risk in step 2 adopts a Markov chain model. By dividing the channel state into good state and bad state, the probability matrix of state transition at different physical space coordinate points is calculated, and then the probability of continuous loss of data packets within a preset transmission period is derived, providing a decision basis for the arbitrator to decide whether to enable the redundant transmission mechanism.

[0024] Preferably, in step 3, the arbiter uses an adaptive bit rate algorithm when adjusting the compression rate. By monitoring the backlog of the physical layer transmission queue, it dynamically adjusts the output bit rate of the source encoder to ensure that the output bit rate is always lower than the predicted instantaneous channel capacity, thus avoiding data delay jitter caused by congestion at the transmitting end.

[0025] Preferably, the processing of the absorption boundary in step 1 employs a fully matched layer technique, in which a specific artificial absorption layer is set at the edge of the simulation area to eliminate electromagnetic wave reflection at the boundary of the computational area, ensuring that the prediction results only reflect the wave dynamics effects of real physical entities on the driving path.

[0026] Preferably, the digital twin in step 2 also has a fault simulation function. When generating the pre-scheduled schedule, it considers the extreme case of a certain communication mode experiencing hardware failure or malicious interference in advance, providing the physical vehicle with a fault-tolerant alternative scheduling scheme, and ensuring that the assisted driving system can still maintain basic safe communication when local communication fails.

[0027] Preferably, the channel binding process in step 3 involves the dynamic reorganization of physical layer subcarriers. By allocating orthogonal frequency division multiplexing symbols among multiple heterogeneous networks, the receiver of the physical vehicle reorders and verifies the integrity of data blocks from different channels according to a preset sequence number.

[0028] Preferably, in step 1, multi-path ray tracing calculations are performed on the vehicle's driving path to obtain the arrival angle and departure angle information of each path. The angle information is then introduced into the initial conditions of the wave equation to guide the physical vehicle to adjust the pointing direction of the antenna array before entering a specific area.

[0029] Preferably, the process of dynamically adjusting the reporting frequency in step 3 is also linked to the vehicle's dynamic state. When the vehicle is in an emergency braking or sharp turning state, the arbitrator ignores the channel prediction state, forces all sensors to be given the highest reporting priority, and counteracts potential signal fading by increasing the transmission power.

[0030] Preferably, the physical environment modeling in step 1 also considers the influence of atmospheric environmental factors. By introducing correction parameters for waveguide characteristics based on air humidity, rainfall intensity, and fog concentration, the attenuation coefficient in the wave equation is adjusted to adapt to communication scheduling needs under different weather conditions.

[0031] Preferably, the pre-scheduling timetable generated in step 2 is transmitted through an auxiliary channel in encrypted and compressed form. The physical vehicle is equipped with a dedicated hardware acceleration unit for real-time decompression and parsing of the timetable to ensure that the execution delay of the scheduling instructions is lower than the preset system tolerance limit.

[0032] Preferably, in step 3, the arbitrator uses a semantic compression algorithm based on deep learning to compress the sensor data, extracting key structural information and target attribute information in the perception scene. When the channel is extremely poor, only structured semantic vectors are transmitted instead of pixel arrays, so as to maintain the system's macroscopic perception capability of the surrounding environment with extremely low data volume.

[0033] Preferably, in step 1, by establishing a digital grid model of the physical space, the complex building outline is simplified into a combination of polyhedra, thereby reducing the computational load of geometric collision detection while ensuring the accuracy of wave dynamics calculation.

[0034] Preferably, in step 2, the simulation of communication load by the digital twin includes the calculation of network layer signaling overhead. It not only considers the transmission of service data, but also the access process delay and authentication process delay caused by frequent channel switching, ensuring that the pre-scheduled schedule can reflect the real end-to-end communication performance.

[0035] Preferably, after cross-modal channel binding in step 3, the arbitrator will perform real-time bandwidth estimation on the merged virtual link. If it is found that the total bandwidth after binding is still insufficient to support the minimum requirements of all current sensors, an active call drop strategy will be executed according to the service priority label to prioritize the flow of positioning data and control commands.

[0036] Preferably, the identification of the multipath interference peak region in step 1 also includes the prediction of Doppler frequency shift. By using the angle relationship between the vehicle speed vector and the wave propagation vector, the center frequency shift of the received signal is calculated, and the spectral dispersion phenomenon caused by high-speed movement is predicted, so as to provide preset parameters for the frequency compensation stage of the receiver.

[0037] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention transforms traditional reactive communication scheduling into preventative pre-scheduling by introducing spatial wave dynamics modeling, thus solving the scheduling failure problem caused by channel perception lag and ensuring communication stability in complex physical environments.

[0038] 2. This invention uses digital twins for advanced simulation, which can simulate communication load and packet loss risk before physical vehicles arrive at a specific area. The generated pre-schedule timetable provides the physical vehicles with spatially adaptable action guidelines, improving the throughput and reliability of the multimodal communication system.

[0039] 3. This invention adopts a dynamic channel binding and arbitration mechanism, which ensures the zero-interruption transmission of key assisted driving commands in signal blind spots or strong interference areas through cross-modal resource sharing and adaptive data compression. This reduces the risk of accidents caused by communication failures and enhances the robustness and safety assurance capabilities of the assisted driving system in extreme electromagnetic environments. Attached Figure Description

[0040] Figure 1 The above is a flowchart of the overall technical solution architecture proposed according to the present invention; Figure 2 This is a schematic diagram of the multi-level interaction relationship and data flow according to the present invention; Figure 3 This is a flowchart illustrating the logical flow framework for predicting signal blind zones and interference based on spatial wave dynamics modeling according to the present invention. Figure 4 The logical flow framework diagram for generating a pre-scheduled schedule based on digital twin advance simulation according to the present invention is shown below; Figure 5 This is a flowchart illustrating the logical process of performing dynamic channel binding and arbitration adjustment according to the present invention. Detailed Implementation

[0041] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 In this embodiment, a multimodal communication scheduling method for assisted driving scenarios is provided. This method introduces the physical field theory from structural mechanics into the field of wireless communication. In the above method, step 1 involves constructing a spatial wave dynamics modeling module. Using the wave equation from structural mechanics, the physical environment along the vehicle's driving path is modeled as rigid obstacles and absorbing boundaries that affect the propagation of electromagnetic waves. By solving the wave equation, the signal blind spots and multipath interference peak areas encountered by the vehicle within a preset time window in the future are predicted.

[0042] In step 1, during the modeling of the physical environment using the wave equation from structural mechanics, the physical environment encompasses buildings, vegetation, large moving targets, and various road facilities in scenarios such as urban roads, highways, and underground parking garages. During the modeling phase, the system acquires physical entity information with 3D semantic tags through a high-precision map interface. The modeling module defines buildings, large moving targets, and road medians with high reflectivity as rigid obstacles with preset reflection coefficients. For buildings, different electromagnetic parameters, including dielectric constant and permeability, are set according to their surface materials, such as concrete, marble, or all-glass curtain walls. These parameters affect the reflection term coefficients in the wave equation, determining the proportion of energy reflected when electromagnetic waves contact the obstacle surface.

[0043] For vegetated environments, such as roadside trees and green belts, the modeling process defines them as absorbing boundaries with preset attenuation coefficients. Vegetation exhibits frequency selectivity in its scattering and absorption of electromagnetic waves; therefore, the system introduces a loss term into the wave equation to simulate the energy dissipation that occurs when electromagnetic waves penetrate leaves and branches. By establishing a spatial grid in a Cartesian coordinate system, the system discretizes the entire driving space.

[0044] Specifically, the electromagnetic wave scalar field wave equation described in the space wave dynamics modeling module is expressed as: ; in, This represents the scalar field strength distribution function of electromagnetic waves in Cartesian coordinates. Represents a time variable. This indicates the speed at which electromagnetic waves propagate in the current medium. Represents the Laplace operator. This represents the energy attenuation coefficient caused by the material of vegetation or buildings. When introducing the finite-difference time-domain algorithm, the continuous space is discretized into a step size of... The grid, scalar field At spatial grid points and time step The iterative update formula at point is: ; in, Indicates the time step. This represents the predicted field strength value for the next time step. This represents the field strength value at the current time step. This represents the field strength values ​​of six adjacent grid points at the current time step. Through iteration of this formula, the three-dimensional field strength distribution data output by the module is transmitted to the blind zone identification submodule, where it is compared with a preset threshold to output the signal blind zone coordinates.

[0045] The propagation process of electromagnetic waves is described as a scalar field fluctuation evolving over time. Its core logic lies in transforming the intensity distribution of radio frequency signals into the evolution of a spatial pressure field. The boundary conditions of this scalar field at the interface of dissimilar media follow the principles of displacement continuity and stress balance, ensuring the conservation and transformation of wave energy when crossing the interface. Through iterative solutions to the wave equation, the system can calculate the field strength distribution of electromagnetic waves in complex spatial structures.

[0046] In solving the wave equation, the system involves fine discretization of a specific geographical region. The continuous physical space is divided into millions of finite element elements, typically employing hexahedral or tetrahedral structures to accommodate irregular building shapes. The system introduces a finite-difference time-domain algorithm to numerically simulate the wave characteristics of electromagnetic waves.

[0047] In this simulation, electric and magnetic field components are alternately arranged in a spatial grid and updated alternately with time steps. The numerical simulation process not only considers the multipath reflection effect along the vehicle's path but also deeply integrates the physical calculations of diffraction and scattering effects. For example, when electromagnetic waves encounter sharp corners of buildings, the generation process of secondary wave sources is calculated based on Huygens' principle to predict signal coverage in non-line-of-sight areas.

[0048] To achieve forward-looking predictions, the system simulates the precise trajectory of a vehicle by integrating the vehicle's displacement vector within a preset future time window. At each discrete coordinate point on this precise trajectory, the system calculates the corresponding received signal strength using a wave field. Subsequently, the system compares these predicted signal strengths with a preset communication threshold, identifies areas where the received signal strength is below the threshold, and marks these areas as signal blind spots.

[0049] Step 1 also predicts the multipath interference peak region. This process is achieved by analyzing the phase superposition relationship of electromagnetic waves propagating along different paths at the receiver. When coherent waves from multiple reflection paths superimpose in phase at a specific spatial location, the amplitude at the receiver will increase according to the principle of wave interference. When the phase difference approaches 180 degrees, deep fading occurs. The system determines these spatial locations where severe interference occurs as interference peak regions. These interference peak regions correspond to the regions of severe fluctuations in Rayleigh or Ricean fading. By establishing a mapping matrix between spatial location coordinates and fading depth, the system provides underlying physical prediction support for subsequent scheduling strategies.

[0050] As a preferred implementation, for predicting the multipath interference peak region, the system extracts the complex received signals from multiple reflection paths along the vehicle's travel path, the first... The complex baseband signal of each path is represented as: ; in, Indicates the first Complex signals of a path, Indicates the first The signal amplitude of the path, Indicates the first The phase offset introduced by spatial distance and medium along the path. Total received signal. For all Vector superposition of path signals: ; The system calculates the envelope amplitude of the total received signal. When the formula is satisfied At that time, the current location is determined to be in the interference peak area. Among them, This represents the reference value for the received signal amplitude in a line-of-sight transmission path when there is no multipath interference. This represents the preset in-phase superposition interference threshold factor, with a value range of [value range missing]. The determination result is stored as a spatial coordinate attribute in a mapping matrix and transmitted to the digital twin pre-simulation module as a prerequisite for channel state mutation.

[0051] Furthermore, the length of the preset future time window is not fixed but dynamically adjusted based on the vehicle's current speed. When the vehicle speed reported by the onboard sensors exceeds a preset speed threshold, the system automatically extends the time window to ensure that the prediction range covers the area of ​​environmental change that the vehicle rapidly approaches due to high-speed movement. When the vehicle speed is below the preset speed threshold, the system shortens the time window to reduce computational complexity and improve prediction accuracy.

[0052] The space wave dynamics modeling module periodically retrieves the latest high-precision map data from the cloud server. This data includes dynamically updated construction areas, traffic control information, and more. The modeling module then converts these semantic labels into boundary parameters for wave dynamics calculations in real time.

[0053] In the advanced implementation of step 1, the modeling of the physical environment also considers the influence of atmospheric environmental factors. The system introduces correction parameters for waveguide characteristics based on air humidity, rainfall intensity, and fog concentration. Rainfall causes electromagnetic wave scattering attenuation, and the system adjusts the attenuation coefficient in the wave equation according to the amount of rainfall to adapt to communication scheduling needs under different weather conditions. The identification of multipath interference peak regions also includes the prediction of Doppler frequency shift. The system uses the angle between the vehicle motion vector and the electromagnetic wave propagation vector to calculate the center frequency shift of the received signal. This prediction can anticipate the spectral dispersion phenomenon caused by high-speed vehicle movement, providing preset prior parameters for the frequency compensation stage at the receiver.

[0054] After completing the spatial wave dynamics modeling, step 2 is executed to construct a digital twin pre-simulation module. Based on the prediction results of the signal blind zone and the multipath interference peak zone, a digital twin that runs in parallel with the physical vehicle is constructed in virtual space. The digital twin shares sensor data with the physical vehicle and runs ahead of time. The communication load and packet loss risk of the multimodal sensors are simulated in advance according to the predicted channel state, and a pre-scheduled schedule is generated.

[0055] During implementation, the digital twin operates at a consistently faster speed in virtual space than the actual speed of the physical vehicle. This proactive operation is determined by a preset lead time, ensuring that the digital twin can detect impending communication crises that the physical vehicle may encounter. The digital twin integrates a communication protocol stack model completely identical to that of the physical vehicle, including all logic at the physical layer, media access control layer, and network layer.

[0056] The digital twin also integrates a perception system model and a decision-making logic model. By receiving real-time data on the current position, speed, acceleration, and steering angle reported by the physical vehicle, the digital twin performs advanced simulations on a virtual environment map to obtain precise radio frequency environmental parameters that the physical vehicle will experience within a preset time period in the future.

[0057] In step 2, pre-simulating the communication load and packet loss risk of multimodal sensors is a crucial step. The system performs refined classification of the output data streams from vehicle-mounted cameras, vehicle-mounted millimeter-wave radar, vehicle-mounted lidar, and vehicle-to-everything (V2X) communication terminals. The real-time generation rate of each sensor's data and the preset quality of service (QoS) requirements are used as input variables. Combining the channel state parameters predicted in step 1, the digital twin simulates the data transmission process under specific channel bandwidth constraints. The simulation process employs a Markov chain model, dividing the channel state into good and bad states and calculating the probability matrix of state transitions at different physical spatial coordinate points. Based on this probability transformation relationship, the system can derive the probability of continuous packet loss within a preset transmission period.

[0058] Specifically, the Markov chain model defines the channel state space as a discrete set. ,in Indicates a good channel condition. Indicates the channel difference state. At physical space coordinates... At that point, the state transition probability matrix Defined as: ; in, Indicates at coordinate point The probability that the channel remains in good condition. This represents the probability of transitioning from a good state to a bad state. This represents the probability of transitioning from a poor state to a good state. Let represent the probability of maintaining a poor state, and satisfy . and The parameters in this matrix are derived by mapping the proportion of fading depths exceeding a threshold as predicted by the aforementioned wave dynamics. Within a preset transmission period... The probability of consecutive packet loss within a time slot (i.e., the first time slot is normal, and all subsequent time slots are lost, or the first time slot enters a poor state and remains so). The derived formula is: ; in, This indicates the total number of time slots contained in a transmission cycle. This represents the probability of transitioning from a good state to a bad state in the first time slot. Indicates that in the following The probability of continuously locking in a poor state within a time slot. This probability of continuous loss. The output is sent to the arbitrator as a decision threshold input to determine whether to enable the redundant transmission mechanism.

[0059] The digital twin's simulation of communication load also includes in-depth calculations of network layer signaling overhead. The system considers not only the transmission load of the service data itself, but also the access process delay, authentication delay, and bandwidth resources occupied by retransmission requests caused by frequent channel switching. The digital twin utilizes a long short-term memory neural network model to extract features from historical channel quality records reported by physical vehicles. By learning from historical data, the neural network can identify periodic interference characteristics in specific areas and weightedly fuse these features with current wave dynamics prediction results, reducing prediction bias caused by simplification of physical environment modeling.

[0060] In one specific embodiment, the Long Short-Term Memory (LSTM) neural network model includes an input layer, two hidden layers, and an output layer. Based on the historical channel quality records reported by physical vehicles, past... The signal-to-noise ratio sequence of each time window is used as the input tensor. ,in Indicates the first Each time window includes signal-to-noise ratio, bandwidth capacity, and latency. 3D feature vectors. LSTM hidden layers at time steps The Gate of Oblivion Input gate Output gate and cell state The update formula is as follows: ; ; ; ; ; ; in, This represents the Sigmoid activation function. This represents the hyperbolic tangent activation function. Represents the Hadama product. , , , These represent the weight matrices for each gating control. , , , This represents the corresponding bias vector. This indicates the hidden state of the previous time step. The last time step... Hidden state After mapping through a fully connected layer, the result is used as the historical feature extraction result. The current prediction results are combined with the output of the space wave dynamics modeling module. The final predicted channel state is obtained through a weighted correction formula. : ; in, The confidence weight of the wave dynamics prediction result is represented, and its value range is [value range missing]. The refreshness is dynamically adjusted based on the current update time of the high-precision map. The data is passed to the pre-scheduled timetable generation submodule. During training, this LSTM network uses mean squared error as the loss function. ,in This represents the total number of training samples. Indicates the first The true channel state of each sample is labeled, and the weight matrix is ​​updated through backpropagation using the Adam optimizer.

[0061] Based on the simulation results above, the digital twin generates the pre-scheduled timetable. This pre-scheduled timetable contains a series of instructions arranged chronologically, each instruction corresponding to a high-precision preset timestamp. The instruction set defines the activation status of various communication modes of the physical vehicle at different time points, the transmit power adjustment parameters, the selection of modulation and coding schemes (MCS), and the specific data service types carried. The pre-scheduled timetable is transmitted to the physical vehicle in encrypted and compressed form via an auxiliary channel. The physical vehicle is equipped with a dedicated hardware acceleration unit for real-time decompression and parsing of the timetable, ensuring that the execution latency of the scheduling instructions is below the system's set tolerance limit.

[0062] During the execution of step 2, a feedback correction mechanism is also included. If the deviation between the channel quality perceived by the physical vehicle during actual driving and the predicted value in the pre-scheduled timetable exceeds a preset deviation threshold, the physical vehicle will immediately trigger a real-time interruption request. After receiving the measured data, the digital twin will quickly re-perform wave dynamics simulation and pre-playback, and issue a corrected pre-scheduled timetable to achieve closed-loop adaptive adjustment. The digital twin also has a fault simulation function. When generating the scheduling scheme, it pre-considers extreme cases such as hardware failure or external malicious electromagnetic interference in a certain communication mode, providing the physical vehicle with fault-tolerant alternative scheduling schemes.

[0063] Step 3 is executed, performing dynamic channel binding and arbitration. The physical vehicle proactively initiates cross-modal channel binding before entering the signal blind zone, according to the pre-scheduled timetable. Simultaneously, the arbitrator dynamically adjusts the reporting frequency and compression rate of sensor data based on the predicted multipath interference peak area.

[0064] The proactive initiation of cross-modal channel binding in step 3 refers to the system logic layer automatically reallocating resources before the physical vehicle senses it is about to enter a predicted signal blind zone. High-priority service flows originally carried in the first communication mode, such as collision warning information and remote control commands, will be redirected to the idle resource blocks of the second communication mode. The second communication mode includes lateral communication links based on millimeter-wave radar sensing, dedicated short-range communication links (DSRC), or visible light communication links between vehicles.

[0065] By establishing a multipath transmission control protocol at the logical link layer, the system can achieve seamless switching and load sharing between different physical media. This binding process involves the dynamic reconfiguration of physical layer subcarriers, enabling parallel transmission of data blocks across networks by allocating orthogonal frequency division multiplexing symbols among multiple heterogeneous networks. The receiving end of the physical vehicle reorders and verifies the integrity of data blocks from different channels according to preset sequence numbers, ensuring data integrity and order.

[0066] Meanwhile, the arbitrator plays a key decision-making role in step 3. The arbitrator dynamically adjusts the sensor data reporting frequency and compression rate, strictly adhering to a priority scheduling algorithm. When it is predicted that a vehicle is about to enter a multipath interference peak region and the channel quality is expected to degrade to a preset critical value, the arbitrator performs a real-time importance assessment of all sensor data.

[0067] For perception data related to driving safety, such as emergency braking trigger signals and obstacle distance warnings, the arbitrator maintains its original high reporting frequency and compresses the data volume by reducing the proportion of non-critical redundant information transmitted in the data packets. For non-safety-related environmental description data, such as roadside landscape videos and non-core diagnostic data, the arbitrator performs high-rate spatial downsampling or temporal compression, for example, downsampling high-definition video streams into low-resolution images, or even performing frame extraction.

[0068] For video streams captured by vehicle-mounted cameras, the arbitrator's frame extraction strategy is deeply coupled with the image content. By identifying the density of key targets in the image, the system forces a high frame rate transmission when obstacles within the collision risk zone are present; when the image contains only static background, the transmission frequency is reduced. The saved bandwidth resources are reallocated to raw point cloud data from millimeter-wave radar or key frames from lidar, which are extremely sensitive to the communication environment.

[0069] The arbitrator also employs a deep learning-based semantic compression algorithm for compressing sensor data. In situations with extremely poor channel conditions and insufficient available bandwidth to transmit the pixel array, the system extracts only key structural and target attribute information from the perceived scene, encoding this information into semantic vectors with extremely low data volume for transmission. The receiving end reconstructs the scene in the digital twin space based on these semantic vectors, maintaining the system's macroscopic perception capability of the surrounding environment under extremely narrowband conditions.

[0070] Specifically, the deep learning-based semantic compression algorithm employs a convolutional variational autoencoder architecture. This architecture includes an encoder network and a decoder network, with the encoder network consisting of four convolutional layers and two fully connected layers. For the input perceptual scene image tensor... (in and (representing the image's height and width in pixels, respectively). After multi-layer feature extraction and downsampling by the encoder, a low-dimensional semantic vector is output. The calculation process is as follows: ; in, This represents the nonlinear mapping function of the encoder. This represents the network weight parameters of the encoder. The dimension of the semantic vector is represented by a value. or This semantic vector The bounding box coordinate offsets, class probability distributions, and depth gradient structure information of obstacles are explicitly encoded. During training, the reconstruction loss function of this convolutional variational autoencoder... Defined as: ; in, This represents the reconstruction function of the decoder. This represents the weight parameters of the decoder. Indicates the training batch size. The squared L2 norm is used to calculate pixel-level reconstruction error. The regularization coefficient representing the KL divergence. This indicates the calculation of KL divergence. and Let represent the mean and variance of the Gaussian distribution in the latent variable space, respectively. When the prediction channel is in extremely poor condition, the physical vehicle will only transmit the encoded semantic vector. Transmitted through a physical layer queue, the receiving digital twin uses a pre-trained decoder to reconstruct the structured environment topology based on the vector, maintaining the system's macroscopic perception of the surrounding environment with extremely low data volume.

[0071] The arbiter also employs an adaptive bit rate algorithm when adjusting the compression ratio. By monitoring the backlog in the physical layer transmission queue, it dynamically adjusts the output bit rate of the source encoder to ensure that the output bit rate is always lower than the predicted instantaneous channel capacity, thus avoiding data delay jitter caused by network congestion at the transmitting end. Furthermore, the choice of compression ratio is negatively correlated with the predicted signal-to-noise ratio (SNR) trend. The lower the SNR, the higher the compression ratio corresponding to the selected compression level, and the compressed data packets undergo stronger redundant error correction coding to improve anti-interference capabilities in interference areas.

[0072] In one specific embodiment, the adaptive bit rate algorithm dynamically adjusts the output bit rate of the source encoder by constructing a feedback control model based on queue depth. Within the time window... Internally, the backlog in the physical layer send queue The calculation formula is: ; in, This indicates the queue backlog in the previous time window. This indicates the current data generation rate of the sensor. This represents the predicted instantaneous channel capacity issued by the digital twin. Indicates the scheduling time interval. The target output bit rate of the source encoder in the next time window. The adjustment formula is: ; in, This represents the channel capacity utilization factor, with a value range of [value range missing]. To allow for anti-shake margin, This represents the queue backlog penalty coefficient, with a value range of [value missing]. When the backlog When it increases, This decrease forces the source encoder to increase the compression rate, avoiding data delay jitter caused by congestion at the transmitter. The target output bit rate... It is directly transmitted as a control command to the source encoder module.

[0073] The cross-modal channel binding arbitration logic in step 3 also incorporates considerations of power consumption and cost. Under the premise of meeting secure transmission requirements, the arbitrator seeks the optimal balance between reliability, latency, power consumption, and cost by constructing a multi-objective optimization function. For example, when the vehicle's battery is low, the system will prioritize low-power short-range communication modes over high-power cellular transmitters. After cross-modal channel binding is completed, the arbitrator continuously estimates the real-time bandwidth of the merged virtual link. If it is found that even with all available channels bound, the total bandwidth still cannot support the minimum security requirements, a proactive call-drop strategy is executed according to the service priority label, prioritizing the flow of location data and control commands, and temporarily suspending non-core voice calls or entertainment streaming services.

[0074] The process of dynamically adjusting the reporting frequency in step 3 is also linked to the vehicle's dynamic state. When the onboard inertial measurement unit (IMU) detects that the vehicle is in a state of emergency braking, extreme steering, or triggering the electronic stability control system (ESC), the arbitrator will ignore the channel prediction state, force all sensors to be given the highest reporting priority, and counteract potential signal fading by increasing the transmission power and increasing the number of redundant retransmissions, so as to ensure the absolute delivery of safety commands.

[0075] In this embodiment, step 1 further employs a fully matched layer technique for processing the absorbing boundary. During the numerical simulation of the wave equation, to prevent false reflections of electromagnetic waves at the artificial edges of the simulation area, the system sets up a specific artificial absorbing layer at the edge of the area. By introducing anisotropic conductivity distribution, electromagnetic waves are completely absorbed upon entering this fully matched layer without generating echoes. This ensures that the prediction results accurately reflect the wave dynamics effects of real physical entities along the driving path. Simultaneously, the modeling module simplifies complex building outlines into polyhedral combinations by establishing a digital grid model of the physical space. This geometric simplification technique reduces the computational load of geometric collision detection while maintaining the accuracy of wave dynamics calculations, enabling the complex prediction module to run in real-time on an in-vehicle embedded computing platform.

[0076] In practical applications, step 1 further involves performing multi-path ray tracing calculations on the vehicle's travel path to obtain the arrival and departure angles for each path. The system incorporates this spatial angle information into the initial conditions of the wave equation, enabling accurate prediction of the beamforming effect of the directional antenna. This prediction guides physical vehicles to pre-adjust the antenna array's beam direction before entering specific obstructed areas, utilizing the reflection paths of buildings to achieve signal diffraction transmission and creating virtual channels in signal blind zones.

[0077] Example 2: Based on Example 1, this example provides an enhanced scheduling scheme based on roadside unit (RSU) collaboration. In step 1, this scheme not only utilizes onboard computing power for wave dynamics modeling but also integrates local high-frequency electromagnetic environment sampling data from roadside units.

[0078] In step 1, the roadside unit (LSU) acts as a fixed spatial anchor point, using its deployed high-performance radio frequency sensing array to measure and upload electromagnetic scattering fingerprints within a 100-meter radius in real time. This fingerprint data is sent to the vehicle-mounted spatial wave dynamics modeling module to correct the initial field strength distribution in the wave equation. Because the LSU is positioned at a high vantage point, it provides more direct observation of the wave shadow regions caused by large moving targets. The system improves the prediction accuracy for instantaneous signal blind spots caused by dynamic obstacles by performing spatial convolution operations between the real-time wave shadow features sensed by the LSU and the static model of the vehicle-mounted high-precision map.

[0079] In step 2, the digital twin pre-simulation module adopts a distributed architecture in this embodiment. Part of the digital twin's logic runs on edge computing nodes (MECs), while another part runs on the vehicle-mounted computing platform. The edge computing nodes are responsible for handling large-scale wave dynamics evolution calculations, utilizing server-level powerful computing resources to complete the electromagnetic environment evolution simulation of a complex street within the next 10 seconds in a very short time. The vehicle-mounted platform is responsible for synchronizing the real-time vehicle dynamics status to the edge platform.

[0080] This architecture reduces the power consumption burden on the vehicle platform. The generated pre-scheduled timetable is sent from the edge computing node to the physical vehicle via the ultra-low latency Uu interface. The timetable synchronization mechanism adopts a unified timing pulse based on the Global Positioning System (GPS) or Beidou system to ensure that the absolute synchronization error of the timestamp is within the microsecond level.

[0081] In step 3, the cross-modal channel-bonded arbitrator further introduces roadside perspective functionality. When the vehicle-mounted camera's perception fails due to obstruction by a large vehicle ahead, the arbitrator adjusts its own sensor's data stream and actively requests the roadside unit's viewpoint data stream. The channel-bonding logic specifically allocates high-bandwidth slices for receiving visual feature point clouds forwarded by the roadside unit. Before entering the predicted signal blind zone, the arbitrator pre-establishes a directional beam connection with the roadside unit at the physical layer according to a pre-scheduled timetable. This connection uses pre-configured random access resource blocks, avoiding the latency jitter caused by traditional contentious access.

[0082] To address the multipath interference peak region, the arbitrator in this embodiment employs coherent cooperative transmission technology. When a vehicle is predicted to enter a strong interference zone, the arbitrator instructs the physical vehicle to simultaneously transmit the same critical control command using multiple distributed antennas deployed on its roof and front, employing Space-Time Block Coding (STBC). This method achieves diversity gain at the physical layer, offsetting the multipath destructive fading predicted by the space wave dynamics model.

[0083] Example 3: This example focuses on the specific implementation details of the method in extremely confined environments such as underground multi-story parking garages and tunnels. In these scenarios, the propagation of electromagnetic waves exhibits waveguide effects, making space wave dynamics modeling particularly important.

[0084] In step 1, for the tunnel scenario, the spatial wave dynamics modeling module models the tunnel walls as reflecting boundaries with specific roughness. Based on the material of the tunnel walls—either shotcrete or ceramic tile—the system sets the corresponding electromagnetic scattering tensor in the wave equation. Due to the narrow space within the tunnel, the reciprocating reflection of electromagnetic waves between the tunnel walls creates a complex standing wave field. By solving the wave equation containing higher-order modes, the modeling module accurately predicts the periodic strength variation nodes of the signal field within the tunnel. This prediction resolution can reach the centimeter level, far exceeding that of traditional statistical channel models.

[0085] In step 2, the digital twin introduces a more complex frequency-selective fading model to simulate packet loss risk, taking into account the unique environment within the tunnel. Due to the severe multipath delay spread within the tunnel, the digital twin specifically simulates inter-symbol interference (ISI) in an orthogonal frequency division multiplexing (OFDM) system during the pre-simulation. Simulation results show that at certain locations, the packet loss rate increases because the multipath delay exceeds the guard interval. Based on this, the pre-scheduling timetable will, in advance, instruct the communication module to double the subcarrier spacing or increase the length of the cyclic prefix before these locations arrive, sacrificing some spectral efficiency for improved communication reliability within the tunnel.

[0086] In step 3, the dynamic channel binding and arbitrator implements a special redundant link splitting strategy for the tunnel environment. Inside the tunnel, cellular signals are typically provided via leaky cables or distributed antenna systems. The arbitrator identifies signal overlap areas within the tunnel based on a pre-scheduled timetable. Before entering an overlap area, the system proactively initiates channel binding with two adjacent leaky cable segments to achieve dual-connectivity transmission. When adjusting the sensor compression rate, the arbitrator specifically identifies the impact of dim tunnel lighting on camera image quality. If excessive video noise is detected, the arbitrator instructs the activation of auxiliary transmission from the vehicle-mounted infrared sensor and performs semantic fusion compression with the video stream. This compression method extracts only the outlines of pedestrians and vehicles from the infrared thermal image, reducing the amount of data required to be transmitted in the unstable signal tunnel environment.

[0087] In low-speed reversing or automatic parking scenarios such as underground parking garages, vehicle movement is extremely slow. In this case, the future preset time window in step 1 is shortened, but the calculated grid density increases to the millimeter level. This high-precision wave dynamics prediction can identify minute signal shadows caused by large concrete pillars within the garage. The arbitrator dynamically adjusts the data reporting ratio of the onboard ultrasonic radar and camera based on the predicted distribution of these tiny shadows. When the rear of the vehicle is about to enter a shadow area, causing a decrease in V2X positioning accuracy, the arbitrator forces compensation using local ranging data from the millimeter-wave radar, ensuring that the core obstacle avoidance commands during automatic parking are not compromised due to a momentary communication interruption.

[0088] In all embodiments, step 1, the identification of multipath interference peak regions, also includes the prediction of Doppler frequency shift compensation. In enclosed spaces such as tunnels, the Doppler spectrum exhibits non-stationary characteristics due to the relative velocity changes between the vehicle and the wall reflection point. The space wave dynamics modeling module calculates the instantaneous frequency shift value at each future moment by solving the time-varying wave equation. The pre-scheduling timetable generated in step 2 sends these frequency shift values ​​as preset parameters of the physical layer. When a physical vehicle actually enters this area, its RF front-end phase-locked loop can directly load these preset parameters, achieving rapid acquisition and accurate tracking of the carrier frequency, thus solving the spectral dispersion problem in high-speed moving scenarios.

[0089] For semantic compression of sensor data, the arbitrator in step 3 incorporates a lightweight deep neural network encoder. When dynamically adjusting the compression ratio, this encoder maps a frame of input 3D laser point cloud data into a low-dimensional manifold space. This low-dimensional vector (i.e., the semantic vector) contains the topological relationships of obstacles around the vehicle. Even within predicted extreme signal blind zones, where the instantaneous throughput of the cellular network drops to only a few kilobits per second, the arbitrator can still use this semantic vector to transmit key environmental topological features to the cloud-based dispatch center or surrounding vehicles. This wave-dynamic prediction-based semantic switching enables the assisted driving system to maintain logically consistent global perception even under extremely poor communication conditions.

[0090] All the above embodiments together constitute a closed-loop dynamic scheduling system. Through deep cross-layer linkage between the physical layer (wave dynamics), network layer (digital twin pre-simulation), and service layer (dynamic arbitration), this method completely breaks the lag chain of perception-feedback in traditional communication scheduling. The implementation details of each step are designed to transform the uncertain wireless propagation environment into a deterministic, computable, and pre-simulated physical evolution process, providing absolutely safe and zero-interruption communication guarantees for assisted driving.

[0091] In implementing the above scheme, the vehicle's internal communication control unit (CCU) serves as the core actuator. It is responsible for receiving the pre-scheduled timetable and translating it into specific low-level register configuration instructions. For example, during the execution of step 3, when the pre-scheduled timetable indicates that channel binding will be performed at timestamp T1, the CCU will pre-start the radio frequency transceiver link of the second communication mode within a preset millisecond lead time before that moment arrives and complete the relevant signaling configuration for carrier aggregation. This high-precision spatiotemporal coordination ensures that the upper application layer is completely unaware of the underlying physical switching process during communication mode switching, truly achieving a seamless connectivity experience.

[0092] The space wave dynamics modeling module also considers the impact of vehicle scattering on the field strength. During modeling, the vehicle is defined as a metal cavity structure with a complex profile, and the antenna's mounting position is set as different excitation points of radiation sources in the wave equation. The system calculates the antenna gain distribution in different directions by calculating the shielding effect of the vehicle's metal body panels on electromagnetic waves. When the vehicle makes a sharp turn, the directivity of the antenna gain changes. The modeling module integrates this dynamic change in antenna pattern into channel prediction, and the arbitrator adjusts its antenna selection strategy accordingly, always choosing the antenna combination with the best propagation path for communication under the current environment.

[0093] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A multimodal communication scheduling method for assisted driving scenarios, characterized in that, Includes the following steps: Step 1: Construct a space wave dynamics modeling module. Using the wave equation in structural mechanics, model the physical environment along the vehicle's driving path as rigid obstacles and absorbing boundaries that affect the propagation of electromagnetic waves. By solving the wave equation, predict the signal blind zone and multipath interference peak area that the vehicle will encounter in the future within a preset time window. Step 2: Construct a digital twin pre-simulation module. Based on the prediction results of the signal blind zone and the multipath interference peak zone, construct a digital twin in the virtual space that runs in parallel with the physical vehicle. The digital twin shares sensor data with the physical vehicle and its operation sequence is ahead of the physical vehicle. Based on the predicted channel state, it pre-simulates the communication load and packet loss risk of the multimodal sensor and generates a pre-scheduled schedule. Step 3: Perform dynamic channel binding and arbitration. The physical vehicle actively initiates cross-modal channel binding before entering the signal blind zone according to the pre-scheduled schedule. At the same time, the arbitrator dynamically adjusts the reporting frequency and compression rate of sensor data according to the predicted multipath interference peak area.

2. The multimodal communication scheduling method in an assisted driving scenario according to claim 1, characterized in that, In the process of modeling the physical environment using the wave equation in structural mechanics, the physical environment includes buildings, vegetation, and large moving targets. The modeling process defines the buildings and the large moving targets as rigid obstacles with a preset reflection coefficient, and the vegetation as an absorbing boundary with a preset attenuation coefficient. By establishing a spatial grid in the Cartesian coordinate system, the propagation process of electromagnetic waves is described as a scalar field fluctuation that evolves over time. Drawing on the boundary treatment logic of displacement continuity and stress balance in structural mechanics for dealing with elastic wave interface problems, and applying it analogously to the medium interface conditions of electromagnetic wave scalar fields, the continuous conservation of field quantity and flux is ensured, and the field strength distribution law of electromagnetic waves in complex spatial structures is calculated. For the building, different dielectric constants and magnetic permeabilities are set according to its surface material, including concrete, marble and glass curtain wall. By introducing a loss term into the wave equation, the energy absorption process of electromagnetic waves when penetrating obstacles of different materials and reflecting on the surface of the obstacles is simulated.

3. The multimodal communication scheduling method in an assisted driving scenario according to claim 2, characterized in that, The process of solving the wave equation includes: discretizing the geographical region, dividing the continuous physical space into multiple finite element units, and introducing a time-domain finite difference algorithm to numerically simulate the wave characteristics of electromagnetic waves. In the numerical simulation, the electric field components and magnetic field components are arranged alternately in the spatial grid and are updated alternately with the time step. The numerical simulation process takes into account the multipath reflection effect, diffraction effect and scattering effect on the vehicle's travel path. By integrating the vehicle displacement vector within a future preset time window, the movement of the vehicle on a future preset trajectory is simulated, and the received signal strength indication of the vehicle at different geographical coordinate points is determined. Then, areas where the received signal strength is lower than a preset threshold are identified as signal blind spots.

4. The multimodal communication scheduling method in an assisted driving scenario according to claim 3, characterized in that, The prediction process for the multipath interference peak region includes: analyzing the phase superposition relationship of electromagnetic waves propagating along different paths at the receiving end; when coherent waves from multiple reflection paths are superimposed in phase at a spatial location, causing an increase in the amplitude at the receiving end, the spatial location is determined to be the interference peak region. The interference peak region corresponds to the region of severe fluctuations in Rayleigh fading or Rice fading of the signal; By establishing a mapping matrix between spatial location and fading depth, physical layer prediction support is provided for subsequent scheduling strategies. The length of the future preset time window is dynamically adjusted according to the vehicle's current driving speed. When the vehicle speed is higher than the preset speed threshold, the length of the time window is automatically extended to compensate for the increase in the rate of environmental change. When the vehicle speed is lower than the preset speed threshold, the length of the time window is shortened to improve the accuracy of the prediction.

5. The multimodal communication scheduling method in an assisted driving scenario according to claim 4, characterized in that, The digital twin operates at a speed that is always faster than the actual speed of the physical vehicle in the virtual space, and the amount of the speed lead is determined by a preset time advance. The digital twin integrates a communication protocol stack model, a perception system model, and a decision logic model that are consistent with those of the physical vehicle. By receiving real-time data on the current position, speed, acceleration, and steering angle reported by the physical vehicle, the digital twin performs advanced simulations in a virtual environment map to obtain the radio frequency environment parameters experienced by the physical vehicle within a preset time period in the future. The space wave dynamics modeling module periodically retrieves high-precision map data containing three-dimensional semantic labels from the cloud server and converts the semantic labels into boundary parameters in wave dynamics calculations.

6. The multimodal communication scheduling method in an assisted driving scenario according to claim 5, characterized in that, The pre-simulation of communication load and packet loss risk of multimodal sensors specifically includes classifying the output data streams of vehicle-mounted cameras, vehicle-mounted millimeter-wave radars, vehicle-mounted lidars, and vehicle-road cooperative communication terminals; Based on the real-time generation rate of each sensor data and the preset quality of service requirements, combined with the predicted channel state parameters, the data transmission delay and packet loss probability under channel bandwidth constraints are calculated. The channel state parameters include the predicted bandwidth capacity, signal-to-noise ratio, and frequency-selective fading characteristics. The process of simulating packet loss risk uses a Markov chain model. By dividing the channel state into good and bad states, the probability matrix of state transition at different physical space coordinates is calculated, and then the probability of continuous packet loss within a preset transmission period is derived.

7. A multimodal communication scheduling method for assisted driving scenarios according to claim 6, characterized in that, The digital twin uses a long short-term memory neural network model to extract features from the historical channel quality records reported by physical vehicles, and combines the current wave dynamics prediction results to perform weighted correction on the predicted channel state. The pre-scheduled timetable contains a series of instructions arranged in chronological order, with each instruction corresponding to a preset timestamp; The instruction set defines the activation status, transmission power, modulation and coding scheme, and data service type of each communication mode at different time points; The pre-scheduled schedule has a feedback correction mechanism during the execution of the physical vehicle. When the deviation between the channel quality actually perceived by the physical vehicle and the predicted value in the pre-scheduled schedule exceeds a preset deviation threshold, the physical vehicle triggers a real-time interruption request, and the digital twin re-deduces and issues the corrected pre-scheduled schedule.

8. A multimodal communication scheduling method for assisted driving scenarios according to claim 7, characterized in that, The aforementioned proactive cross-modal channel binding refers to the system logic layer redirecting the high-priority service flow originally carried in the first communication mode to the idle resource block of the second communication mode before the physical vehicle enters the predicted signal blind zone. The first communication mode is vehicle-to-everything (V2X) communication based on cellular networks, and the second communication mode includes lateral communication links based on millimeter-wave radar sensing, short-range communication links, or visible light communication links between vehicles. By establishing a multipath transmission control protocol at the logical link layer, seamless switching and load sharing between different physical media can be achieved; The channel binding process involves the dynamic reconfiguration of physical layer subcarriers by allocating orthogonal frequency division multiplexing symbols among multiple heterogeneous networks.

9. A multimodal communication scheduling method for assisted driving scenarios according to claim 8, characterized in that, The arbitrator dynamically adjusts the reporting frequency and compression rate of sensor data according to a priority scheduling algorithm. When it is predicted that a vehicle is entering a multipath interference peak region and the channel quality degrades to a preset critical value, the arbitrator performs an importance assessment on the sensor data. For perception data related to driving safety, the arbitrator maintains the original reporting frequency and reduces the proportion of transmission of non-critical redundant information; For environmental description data that is not related to safety, the arbitrator performs high-rate spatial downsampling or temporal compression; For video streams captured by vehicle-mounted cameras, the arbitrator determines the frame extraction ratio by identifying the density of key targets in the image, and maintains high frame rate transmission when there are obstacles in the image that are within the collision risk area. Reduce the transmission frequency when the image contains only a static background.

10. A multimodal communication scheduling method for assisted driving scenarios according to claim 9, characterized in that, The arbitrator uses a deep learning-based semantic compression algorithm to compress sensor data, extracting key structural information and target attribute information in the perception scene, and transmitting structured semantic vectors when the channel is poor. When adjusting the compression rate, an adaptive bit rate algorithm is used. By monitoring the backlog of the physical layer transmission queue, the output bit rate of the source encoder is dynamically adjusted to ensure that the output bit rate is lower than the predicted instantaneous channel capacity. The selection of the compression ratio is negatively correlated with the predicted signal-to-noise ratio change trend; that is, the lower the signal-to-noise ratio, the higher the selected compression ratio. The process of dynamically adjusting the reporting frequency is linked to the vehicle's dynamic state. When the vehicle is detected to be in an emergency braking or sharp turning state, the arbitrator forces all sensors to be given the highest reporting priority and counteracts signal fading by increasing the transmission power.

Images