Real-time interference avoidance method for vehicle-mounted wireless communication module based on edge computing

Abstract

The present application relates to the field of vehicle-mounted communication technology, and more particularly to a real-time interference avoidance method for a vehicle-mounted wireless communication module based on edge computing, which captures burst pulse interference in the vehicle networking and cellular frequency bands through edge spectrum sensing, extracts time domain duty cycle and spectrum mask features to form a dynamic interference feature set, classifies interference sources at multiple levels using a lightweight convolutional neural network, calculates confidence levels in combination with signal entropy values to generate interference situation data, constructs a risk heat map by fusing vehicle position, road topology and surrounding node state, drives the edge engine to dynamically redistribute spectrum resources, and outputs channel switching instructions; the instructions control the power and modulation coding scheme of the radio frequency front end, and synchronously collect link quality feedback; by comparing the feedback value with the safety threshold in real time, the standby link is seamlessly switched through the trigger of the redundancy mechanism, and logs are uploaded after encryption, ensuring high reliability and security of communication.

Description

Technical Field

[0001] This invention relates to the field of vehicle communication technology, and in particular to a real-time interference avoidance method for vehicle wireless communication modules based on edge computing. Background Technology

[0002] With the rapid development of intelligent connected vehicles and vehicle-to-everything (V2X) systems, in-vehicle wireless communication modules play a crucial role in the real-time exchange of key information such as location, speed, road conditions, and control commands in the V2X environment. However, the 5.9 GHz dedicated V2X frequency band (DSRC / C-V2X) and cellular communication frequency bands (such as LTE / 5G NR) that in-vehicle communication systems rely on are highly susceptible to non-cooperative interference sources in the open and complex electromagnetic environment of roads, such as radar signals, high-voltage power equipment, illegal transmission devices, and co-channel interference between densely packed vehicles. Such interference is characterized by its suddenness, strong pulse characteristics, and uncontrollable sources, which can easily lead to communication link interruptions, increased bit error rates, and even communication security incidents, seriously affecting the reliability of autonomous driving decisions and traffic safety.

[0003] In recent years, the rise of edge computing technology has provided new ideas for real-time interference handling in vehicular communication systems. By decentralizing spectrum sensing, interference identification, and decision-making control capabilities to vehicular edge computing nodes, rapid perception and closed-loop response to interference events can be achieved locally. However, how to efficiently extract interference characteristics, intelligently assess the interference situation, and generate dynamic avoidance strategies by combining vehicle operating status and road environment information in resource-constrained vehicular environments remains a current technical challenge. Especially when the main link continues to deteriorate, existing methods often lack reliable redundant link switching mechanisms and security auditing capabilities, making it difficult to guarantee communication continuity and event traceability.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art.

[0005] Therefore, those skilled in the art are dedicated to developing a real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing. Summary of the Invention

[0006] The main objective of this invention is to provide a real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing. This method aims to solve the technical problems of existing vehicle-mounted wireless communication systems, which are unable to achieve real-time perception, accurate identification, and rapid response to sudden interference in complex electromagnetic environments, and lack dynamic interference avoidance and link redundancy guarantee mechanisms that integrate edge intelligence and multi-source information, resulting in insufficient communication reliability and security.

[0007] To achieve the above objectives, the present invention provides a real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing, the method comprising:

[0008] Throughout the entire working cycle of the vehicle wireless communication module, edge spectrum sensing is performed on the dedicated frequency band for vehicle networking and the frequency band for cellular communication. The burst pulse signals generated by non-cooperative interference sources are captured in real time, and the time-domain duty cycle characteristics and spectrum mask boundaries of the interference signals are identified to form a dynamic interference feature set.

[0009] The dynamic interference feature set is input into the vehicle-mounted edge computing node. The interference source type is classified in multiple levels through a lightweight convolutional neural network model. The interference intensity confidence is calculated based on the signal entropy value to generate edge interference situation assessment data.

[0010] Based on edge interference situation assessment data, the system integrates real-time vehicle location coordinates, road topology semantic map and communication status of surrounding nodes to construct a communication link risk heat map, which drives the edge decision engine to generate a dynamic spectrum resource reallocation strategy and outputs a channel switching instruction sequence.

[0011] The channel switching command sequence is applied to the radio frequency front-end controller of the wireless communication module to perform adaptive adjustment of transmission power and switching of modulation and coding scheme, and simultaneously collect the link quality feedback value sequence after the avoidance operation.

[0012] The link quality feedback value sequence is compared with the preset communication security threshold in real time. When the continuous feedback value is lower than the threshold, the link redundancy activation mechanism is triggered, the multi-mode communication protocol stack is called to perform seamless switching between primary and backup links, and the communication log encryption upload protocol is started.

[0013] Optionally, the step of inputting the dynamic interference feature set into the vehicle-mounted edge computing node, performing multi-level classification of interference source types through a lightweight convolutional neural network model, and calculating the interference intensity confidence based on the signal entropy value to generate edge interference situation assessment data includes:

[0014] The dynamic interference feature set is input into the edge computing unit responsible for spectral feature preprocessing, and time-frequency feature normalization processing is performed to generate a standardized feature vector.

[0015] The standardized feature vectors are input into a pre-deployed lightweight convolutional neural network model, and the local temporal patterns of the interference signal are extracted through the convolutional layers to generate an interference type probability distribution matrix.

[0016] Based on the probability distribution matrix of interference types and combined with the spectral disorder index output by the signal entropy calculation module, the interference intensity is weighted and fused with confidence to generate marginal interference situation assessment data that includes interference type labels, intensity confidence, and duration prediction.

[0017] Optionally, the step of constructing a communication link risk heatmap based on edge interference situation assessment data, integrating real-time vehicle location coordinates, road topology semantic map, and communication status of surrounding nodes, drives the edge decision engine to generate a dynamic spectrum resource reallocation strategy, and outputs a channel switching instruction sequence, including:

[0018] It receives interference situation assessment data and simultaneously acquires the real-time vehicle position coordinates output by the high-precision positioning module and the road topology semantic map provided by the vehicle map engine.

[0019] The road topology semantic graph is parsed into node-edge structured data, and combined with the status information broadcast by surrounding vehicle communication nodes, a road network communication risk assessment model is constructed.

[0020] Based on the road network communication risk assessment model, the channel availability index and interference propagation attenuation coefficient of each communication link are calculated to generate a communication link risk heat map with spatial resolution up to the meter level.

[0021] Based on the communication link risk heatmap, the optimal spectrum resource reallocation path is selected through the heuristic search algorithm of the edge decision engine, and a channel handover command sequence containing the target frequency band, power adjustment amount and handover timing is output.

[0022] Optionally, the step of applying the channel switching command sequence to the radio frequency front-end controller of the wireless communication module to perform adaptive transmission power adjustment and modulation and coding scheme switching, and simultaneously collecting the link quality feedback value sequence after the avoidance operation, includes:

[0023] The channel switching command sequence is transmitted to the digital signal processing unit of the RF front-end controller to parse the target frequency band parameters and configure the phase-locked loop frequency synthesizer.

[0024] Based on power regulation commands, the bias voltage of the power amplifier is dynamically adjusted to achieve closed-loop adaptive regulation of the transmitted power.

[0025] Based on the modulation and coding scheme switching instruction, the modulation order and coding rate of the physical layer protocol stack are updated in real time to generate an optimized communication waveform;

[0026] During waveform transmission, the signal-to-noise ratio, bit error rate, and throughput parameters are synchronously collected through the channel state information feedback unit of the receiving link to form a link quality feedback value sequence.

[0027] Optionally, the step of performing edge-based real-time comparison of the link quality feedback value sequence with a preset communication security threshold, triggering a link redundancy activation mechanism when consecutive feedback values ​​are below the threshold, calling the multi-mode communication protocol stack to perform seamless switching between primary and backup links, and initiating an encrypted communication log upload protocol includes:

[0028] A real-time comparison engine is deployed within the vehicle-mounted edge computing node to perform sliding window analysis on the link quality feedback value sequence and preset multi-level communication security thresholds.

[0029] When a continuous feedback value is detected to be lower than the first-level safety threshold, a link degradation warning signal is generated; when a continuous feedback value is lower than the second-level safety threshold, a link redundancy activation mechanism is triggered.

[0030] Based on the link redundancy activation mechanism, the switching control module of the multi-mode communication protocol stack is called to migrate the traffic of the main communication link to the pre-negotiated backup link and execute the handshake protocol to verify the integrity of the switching.

[0031] After the switch is completed, the log encryption unit is activated to encrypt the interference event data using AES-256 and upload it to the roadside edge server through a low duty cycle transmission channel.

[0032] Optionally, the step of calculating the channel availability index and interference propagation attenuation coefficient of each communication link based on the road network communication risk assessment model, and generating a communication link risk heat map with spatial resolution up to the meter level, includes:

[0033] Extract lane curvature, obstacle density, and signal masking area parameters from the road topology semantic map to construct a spatial propagation loss feature vector;

[0034] By combining the channel occupancy rate in the communication status of surrounding nodes with historical interference records, the probability value of co-channel interference between links is calculated;

[0035] The spatial propagation loss feature vector and the co-channel interference probability value are input into a dynamic weight fusion unit to generate a channel availability index.

[0036] Based on the channel availability index and the intensity confidence in the interference situation assessment data, the interference propagation attenuation coefficient is derived through a Gaussian process regression model, generating a communication link risk heatmap with a spatial resolution of meters.

[0037] Optionally, the step of deriving the interference propagation attenuation coefficient through a Gaussian process regression model and generating a communication link risk heatmap with a spatial resolution of meters includes:

[0038] Using the channel availability index as an input variable, the covariance function kernel of the Gaussian process regression model is constructed;

[0039] Based on historical interference propagation sample data, the hyperparameters of the kernel function are optimized to make the model output match the measured attenuation curve.

[0040] In the gridded coordinate system of the road topology semantic graph, the interference propagation attenuation coefficient is calculated for each grid node to generate the spatial distribution matrix of the attenuation coefficient;

[0041] The spatial distribution matrix of the attenuation coefficient is subjected to bilinear interpolation to output a communication link risk heatmap with a spatial resolution of meters.

[0042] Optionally, during the AES-256 encryption process, the log encryption unit integrates a key derivation mechanism based on the vehicle's dynamic identity identifier, which generates a unique key seed by reading the real-time identity certificate output by the vehicle security chip and the high-precision positioning timestamp.

[0043] Based on the key seed, a lightweight key update protocol is used to periodically refresh the session key, and a cyclic redundancy check code and a message authentication code are embedded in the encrypted data packet;

[0044] When uploading through a low duty cycle transmission channel, the forward error correction coding rate is dynamically adjusted based on the link quality feedback value. When a sudden increase in the channel bit error rate is detected, it automatically switches to a high redundancy transmission mode.

[0045] Furthermore, to achieve the above objectives, the present invention also provides a real-time interference avoidance device for an in-vehicle wireless communication module based on edge computing. The device includes: a memory, a processor, and an edge computing-based real-time interference avoidance program for an in-vehicle wireless communication module stored in the memory and executable on the processor. The edge computing-based real-time interference avoidance program for an in-vehicle wireless communication module is configured to implement the steps of the edge computing-based real-time interference avoidance method for an in-vehicle wireless communication module as described above.

[0046] Furthermore, to achieve the above objectives, the present invention also provides a medium storing a real-time interference avoidance program for an edge computing-based vehicle wireless communication module. When the edge computing-based vehicle wireless communication module real-time interference avoidance program is executed by a processor, it implements the steps of the real-time interference avoidance method for an edge computing-based vehicle wireless communication module as described above.

[0047] The present invention has the following technical effects:

[0048] This invention provides a real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing. The method performs edge-based spectrum sensing throughout the entire operating cycle of the vehicle-mounted wireless communication module, enabling real-time capture of burst pulse signals generated by non-cooperative interference sources. It accurately extracts the time-domain duty cycle features and spectral mask boundaries to form a dynamic interference feature set, effectively improving the detection sensitivity and timeliness of transient interference in complex electromagnetic environments. A lightweight convolutional neural network model is used to perform multi-level classification of interference source types locally at the vehicle edge node. Combined with signal entropy values, interference intensity confidence is calculated, achieving low-latency, high-precision edge-based interference situation assessment. This avoids dependence on cloud computing and significantly enhances the system's intelligent decision-making capabilities in resource-constrained environments. By integrating real-time vehicle location, road topology semantic graph, and communication status of surrounding nodes, a communication link risk heatmap with meter-level spatial resolution is constructed, enabling the dynamic spectrum resource reallocation strategy to possess geographic meaning. The system incorporates semantic perception and contextual understanding capabilities, enabling a shift from passive response to proactive avoidance, significantly improving spectrum utilization and communication reliability. Channel switching commands are applied to the RF front-end controller, simultaneously executing adaptive transmission power adjustment and modulation-coding scheme (MCS) switching, forming a multi-parameter collaborative optimization mechanism. This mechanism dynamically adjusts physical layer transmission characteristics based on the interference environment, maximizing link quality. By comparing link quality feedback values ​​with communication security thresholds in real time, a link redundancy activation mechanism is automatically triggered when continuous degradation occurs. This calls the multi-mode communication protocol stack to achieve seamless switching between primary and backup links, avoiding communication interruptions and significantly improving the system's fault tolerance and service continuity in severe interference or link failure scenarios. An encrypted communication log upload protocol is implemented to ensure the integrity, confidentiality, and non-repudiation of key data related to interference events, avoidance operations, and link switching processes during transmission, providing reliable data support for safe operation, post-event analysis, and liability determination in the vehicle-to-everything (V2X) network. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating an embodiment of the real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing according to the present invention. Detailed Implementation

[0050] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0051] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing according to the present invention.

[0052] In one embodiment, the real-time interference avoidance method for the edge computing-based vehicle wireless communication module includes:

[0053] Step S100: During the entire working cycle of the vehicle wireless communication module, edge spectrum sensing is performed on the dedicated frequency band for vehicle networking and the frequency band for cellular communication. The burst pulse signals generated by non-cooperative interference sources are captured in real time, and the time-domain duty cycle characteristics and spectrum mask boundaries of the interference signals are identified to form a dynamic interference feature set.

[0054] The vehicle-mounted wireless communication module can be a hardware unit integrated into the vehicle for performing vehicle-to-everything (V2X) and cellular communication functions. It can be used to enable wireless information interaction between the vehicle and roadside units, other vehicles, and network infrastructure. In this embodiment, the vehicle-mounted wireless communication module can work collaboratively with a radio frequency transceiver, baseband processor, and protocol stack to complete the modulation, demodulation, and data transmission of wireless signals. The dedicated V2X frequency band can be a radio spectrum resource specifically designed for vehicle-to-infrastructure (V2I) communication, typically such as the 5.9 GHz DSRC / C-V2X band, which can be used for low-latency communication of critical safety messages between vehicles and between vehicles and the road. For example, the dedicated V2X frequency band may include, but is not limited to, one or more of the following: the DSRC band, the C-V2X PC5 interface band, and the ITS-G5 band.

[0055] Cellular communication frequency bands can be licensed or unlicensed spectrum used to support cellular network communications such as LTE / 5G NR. They can provide wide-area coverage, high-bandwidth data connectivity, and support remote control and large data backhaul. Furthermore, cellular communication frequency bands can include, but are not limited to, one or more of LTE Band 41, 5G NR n78, and Sub-6GHz bands. Non-cooperative interference sources can be external transmitters that do not participate in communication protocol negotiation, are uncontrollable, and may cause electromagnetic pollution to the target frequency band. They can be used as targets for interference detection and avoidance, as they threaten the stability of communication links. In an exemplary embodiment, non-cooperative interference sources can include, but are not limited to, one or more of radar systems, illegal broadcasting equipment, and high-voltage power equipment.

[0056] Sudden pulse signals can be interference signals that appear briefly in time, have concentrated energy, and possess strong transient characteristics. They can be used to characterize the main forms of non-cooperative interference and are key targets for spectrum sensing. For example, sudden pulse signals can include, but are not limited to, one or more of narrowband pulse interference, broadband frequency-hopping interference, and periodic sudden interference. The time-domain duty cycle characteristic can be a parameter describing the proportion of the interference signal in an active state within a unit time window. It can be used to quantify the persistence and activity level of the interference and assist in identifying the type of interference. In this embodiment, the time-domain duty cycle characteristic can be obtained by statistically analyzing the proportion of effective pulse duration after energy threshold detection of the sampled signal. The spectral mask boundary can be the energy distribution contour of the interference signal in the frequency domain and its effective bandwidth range. It can be used to define the frequency range occupied by the interference, providing a basis for channel avoidance. Furthermore, the spectral mask boundary can be obtained by extracting the frequency domain energy envelope based on Fast Fourier Transform (FFT) or filter bank analysis.

[0057] The dynamic interference feature set can be an updatable data set composed of multi-dimensional attributes of interference signals obtained through real-time sensing. It can be used as the input basis for interference classification and situation assessment, reflecting the current electromagnetic environment state. In one specific embodiment, the dynamic interference feature set can be generated by edge-based spectrum sensing and received by a lightweight convolutional neural network model for classification. Edge-based spectrum sensing of the vehicle-to-everything (V2X) dedicated frequency band and cellular communication frequency band throughout the entire operating cycle of the vehicle wireless communication module can be achieved by continuously scanning the designated frequency band throughout vehicle operation, with signal acquisition and preliminary analysis completed by local hardware. Furthermore, this operation can be implemented by using a software-defined radio (SDR) platform for wideband real-time sampling, or by using a dedicated spectrum sensing coprocessor to monitor multiple sub-frequency bands in parallel, thereby achieving full-time-domain coverage monitoring of interference and avoiding sensing blind spots.

[0058] Real-time capture of burst pulse signals generated by non-cooperative interference sources can be achieved by identifying short-duration high-energy signals using a high-sampling-rate ADC and energy detection algorithms. For example, this operation can be implemented through energy mutation detection based on a sliding window, or by using a matched filter bank to perform correlation detection on a known pulse template, thereby improving the ability to capture transient interference. Identifying the time-domain duty cycle characteristics and spectral mask boundaries of the interference signal can be achieved by performing time-frequency analysis on the captured signal to extract its activation ratio and frequency-domain energy distribution. In an exemplary embodiment, this operation can be achieved by using short-time Fourier transform (STFT) combined with time-frequency analysis, or by extracting multi-scale time-frequency features through wavelet transform, thereby forming structured features that can distinguish interference types. Forming a dynamic interference feature set can be achieved by organizing the extracted features into an updatable data structure by timestamps. Further, this operation can be achieved by storing the features as feature vectors in a circular buffer, or by constructing a feature database with time decay weights, thereby providing standardized input for subsequent classification.

[0059] Step S200: Input the dynamic interference feature set into the vehicle edge computing node, perform multi-level classification of interference source types through a lightweight convolutional neural network model, calculate the interference intensity confidence based on the signal entropy value, and generate edge interference situation assessment data.

[0060] The vehicle-mounted edge computing node can be an edge processing unit deployed locally in the vehicle, possessing limited computing power and storage capabilities. It can be used to perform local intelligent decision-making and real-time responses without relying on the cloud. In this embodiment, the vehicle-mounted edge computing node can be integrated into the vehicle-mounted computing platform, running a lightweight AI inference framework and communication middleware. The lightweight convolutional neural network model can be a deep learning model with compressed structure, simplified parameters, and suitable for deployment on resource-constrained devices. It can be used to achieve efficient multi-level classification of interference source types on the vehicle-mounted edge node. For example, the lightweight convolutional neural network model can include, but is not limited to, one or more of the following: MobileNet variants, ShuffleNet architecture, TinyCNN structure, etc.

[0061] The type of interference source can be a semantic classification of non-cooperative interference sources based on the interference generation mechanism or device category. This classification can guide the selection of subsequent avoidance strategies, such as whether it is predictable or periodic. In a specific embodiment, the type of interference source can include, but is not limited to, one or more of industrial equipment interference, military radar interference, and civilian illegal emission interference. The signal entropy value can be a statistical measure of the uncertainty of information in the time-frequency domain of the interference signal. It can be used to quantify the confidence level of the interference intensity and assist in judging the severity of the interference. Furthermore, the signal entropy value can be obtained by calculating the signal amplitude or power spectral density distribution based on the Shannon entropy formula. The interference intensity confidence level can be a reliable probability value representing the judgment of the current interference impact level. It can be used to improve the reliability of the interference situation assessment and avoid ineffective switching due to misjudgment. In an exemplary embodiment, the interference intensity confidence level can be calculated from the signal entropy value and used as part of the marginalized interference situation assessment data.

[0062] Edge interference situation assessment data can be a comprehensive assessment result generated locally at the vehicle edge node, including interference type, intensity, and confidence level. This data can be used to drive subsequent risk heatmap construction and spectrum redistribution strategy generation. Furthermore, the edge interference situation assessment data can be jointly output by a lightweight convolutional neural network model and signal entropy values. Inputting the dynamic interference feature set into the vehicle edge computing node can be achieved by transferring the feature data to the edge processing unit via an internal bus or shared memory. For example, this operation can be implemented by transferring feature tensors through an IPC mechanism or by directly writing to the AI ​​accelerator input buffer using DMA, thereby achieving a local closed loop for perception and decision-making. Multi-level classification of interference source types using a lightweight convolutional neural network model can be achieved by running a compressed CNN model at the edge node and outputting the probability distribution of interference categories. In a specific embodiment, this operation can be implemented by using an INT8 quantized model optimized with TensorRT for inference or by deploying a channel-pruned MobileNetV2 variant on the NPU, thereby achieving high-precision, low-latency interference semantic recognition.

[0063] Calculating interference strength confidence based on signal entropy can be achieved by calculating Shannon entropy from the amplitude or power spectrum of the interference signal and mapping it to a confidence score. Furthermore, this operation can be implemented by calculating entropy after histogram statistics of the time-domain envelope, or by calculating normalized entropy values ​​from the energy distribution of spectral sub-bands, thereby enhancing the robustness of interference assessment and reducing noise misjudgments. Generating marginalized interference situation assessment data can be achieved by fusing classification results with confidence scores to form a structured assessment output. For example, this operation can be implemented by generating a JSON-formatted situation report or by constructing a triplet data structure containing type, strength, and confidence scores, thus providing semantic input for risk modeling.

[0064] Step S300: Based on the edge interference situation assessment data, the real-time location coordinates of vehicles, road topology semantic map and communication status of surrounding nodes are integrated to construct a communication link risk heat map, drive the edge decision engine to generate a dynamic spectrum resource reallocation strategy, and output a channel switching instruction sequence.

[0065] The real-time vehicle location coordinates can be the vehicle's current geographic coordinates obtained through GNSS or other positioning systems, which can be used to provide geographic anchors for spatial interference risk modeling. In this embodiment, the real-time vehicle location coordinates can be obtained through the output of an onboard GNSS receiver or a fused inertial navigation system. The road topology semantic map can be a high-precision digital map containing semantic information such as lane lines, intersections, and speed limit zones, which can be used to provide road structure context and assist in understanding the spatial constraints of the communication environment. For example, the road topology semantic map can include, but is not limited to, one or more of the following: OpenDRIVE format maps, NDS standard maps, and high-precision semantic vector maps.

[0066] The communication status of surrounding nodes can be information on the communication link quality, load, and available channels of nearby vehicles or roadside units. This information can reflect the cooperative status of the local communication environment and be used for collaborative avoidance decisions. Furthermore, the communication status of surrounding nodes can be obtained through V2V / V2I broadcast messages or link probing protocols. The communication link risk heatmap can be a visualized data structure presenting the interference risk level of communication links in each area at meter-level spatial resolution. This can be used to enable geographic semantic awareness in spectrum resource allocation and support proactive avoidance. In an exemplary embodiment, the communication link risk heatmap can be generated by fusing edge interference situation assessment data, vehicle location, road topology, and surrounding status. The edge decision engine can be a rule- or model-driven policy generation module running on the vehicle's edge node, which can be used to output the optimal spectrum reallocation scheme based on the risk heatmap. In this embodiment, the edge decision engine can operate based on reinforcement learning strategies, rule bases, or hybrid decision mechanisms.

[0067] Dynamic spectrum resource reallocation strategies can be schemes that dynamically adjust the allocation of available channels based on the current interference situation and geographical context. This can be used to optimize spectrum utilization efficiency and reduce the probability of co-channel interference. For example, dynamic spectrum resource reallocation strategies may include, but are not limited to, one or more of the following: channel frequency hopping strategy, frequency band switching strategy, and power-frequency band joint allocation strategy. The channel switching command sequence can be an operation command stream generated by the edge decision engine to control the RF front-end to perform channel changes, and can be used to achieve real-time adjustment of physical layer communication parameters. Furthermore, the channel switching command sequence can act on the RF front-end controller and trigger power and MCS synchronization adjustments.

[0068] By integrating real-time vehicle location coordinates, road topology semantic maps, and communication status of surrounding nodes, multi-source heterogeneous data can be aligned and correlated within a unified geographic coordinate system. In one specific embodiment, this operation can be achieved by matching location with map features using spatial indexes (such as R-trees) or by fusing topology and communication status using graph neural networks, thereby constructing a communication environment model with context-aware capabilities. Constructing a communication link risk heatmap can be achieved by calculating and visualizing the comprehensive interference risk score for each area on a meter-level grid. For example, this operation can be achieved by generating a continuous heatmap based on kernel density estimation interpolation or by assigning risk levels grid-by-grid to a raster map, thereby enabling an explicit spatial expression of interference risk.

[0069] The edge decision engine generates a dynamic spectrum resource reallocation strategy, which can be achieved by selecting available channels and planning handover paths based on low-risk areas in a heatmap. Furthermore, this operation can be implemented by selecting the lowest-risk channel using a greedy algorithm or by using a Q-learning strategy to learn the optimal allocation action, thereby achieving proactive spectrum avoidance. The output channel handover command sequence can be a transformation of the strategy into a specific RF control command sequence. In an exemplary embodiment, this operation can be achieved by generating a command packet containing the target frequency and filter configuration, or by outputting a timestamped multi-step handover script, thus realizing the conversion from strategy to execution.

[0070] In step S400, the channel switching command sequence is applied to the radio frequency front-end controller of the wireless communication module to perform adaptive adjustment of transmission power and switching of modulation and coding scheme, and the link quality feedback value sequence after the avoidance operation is collected simultaneously.

[0071] The RF front-end controller can be a hardware control unit that manages the RF link parameters (such as frequency, gain, and filters) of the wireless communication module, and can be used to perform physical layer operations such as channel switching and power adjustment. In this embodiment, the RF front-end controller can operate by receiving control commands through register configuration or an SPI / I2C interface. Adaptive transmission power adjustment can be the behavior of dynamically adjusting the transmit power according to link conditions to maintain communication quality, and can be used to balance energy consumption and link robustness in interference environments. For example, adaptive transmission power adjustment can include, but is not limited to, one or more of open-loop power control, closed-loop feedback power control, and path loss-based prediction adjustment.

[0072] Modulation-coding scheme (MCS) switching can be a physical layer adaptation mechanism that dynamically selects the modulation order and coding rate based on channel quality. This can be used to reduce the MCS level to improve demodulation reliability when interference increases. Furthermore, MCS switching can include, but is not limited to, one or more of QPSK+1 / 2 code rate, 16QAM+3 / 4 code rate, and 64QAM+5 / 6 code rate. The link quality feedback value sequence can be a continuous sequence of indicators reflecting communication link performance, such as RSSI, SINR, and PER. This can be used to evaluate the effectiveness of circumvention operations and trigger redundancy mechanisms. In one specific embodiment, the link quality feedback value sequence can be obtained periodically through reporting from the baseband processor or MAC layer.

[0073] Applying the channel switching command sequence to the RF front-end controller of the wireless communication module can be achieved by sending commands through a hardware interface to trigger RF parameter reconfiguration. For example, this operation can be implemented by writing to the RF chip register via the SPI bus or by calling the RF driver API to perform an atomic switching operation, thereby completing the physical layer parameter adjustment. Performing adaptive transmission power adjustment and modulation / coding scheme switching can be achieved by simultaneously adjusting the transmit power and MCS level to match the new channel conditions. Further, this operation can be implemented by selecting the MCS and power combination based on a SINR lookup table or by dynamically calculating the optimal power based on a link budget model, thereby forming multi-parameter collaborative optimization to maximize link quality. Synchronously collecting the link quality feedback value sequence after the avoidance operation can be achieved by continuously monitoring link performance indicators after parameter adjustment. In an exemplary embodiment, this operation can be achieved by collecting SINR and PER every 10ms or by reporting quality data in real time through a PHY layer callback function, thereby verifying the effectiveness of the avoidance and providing input for the redundancy mechanism.

[0074] Step S500: The link quality feedback value sequence is compared with the preset communication security threshold in real time. When the continuous feedback value is lower than the threshold, the link redundancy activation mechanism is triggered, the multi-mode communication protocol stack is called to perform seamless switching between primary and backup links, and the communication log encryption upload protocol is started.

[0075] The preset communication security threshold can be a performance indicator threshold for determining whether a link is in a secure communication state, and can be used as a basis for triggering the link redundancy activation mechanism. For example, the preset communication security threshold may include, but is not limited to, one or more of the following: SINR security threshold, packet loss rate upper limit threshold, and latency tolerance threshold. The link redundancy activation mechanism can be fault-tolerant control logic that automatically activates a backup communication path when the primary link continuously deteriorates, and can be used to ensure communication service continuity and prevent interruptions. Furthermore, the link redundancy activation mechanism can be triggered by comparing the link quality feedback value sequence with the security threshold.

[0076] A multi-mode communication protocol stack can be a collection of protocol software supporting multiple wireless communication standards (such as C-V2X, LTE, and 5G NR), which can be used to achieve seamless switching and interoperability between different communication modes. In a specific embodiment, the multi-mode communication protocol stack may include, but is not limited to, one or more of the following: C-V2X+LTE dual-mode stack, 5G NR+DSRC heterogeneous stack, and Wi-Fi6+C-V2X converged stack. Seamless switching between primary and backup links can be a process of smoothly migrating from a primary link to a backup link without interrupting upper-layer application communication. It can be used to maintain service continuity and avoid the loss of control commands due to link failure. In this embodiment, seamless switching between primary and backup links can be achieved through protocol stack state synchronization, session persistence, and fast reconnection mechanisms.

[0077] A communication log encryption upload protocol can be a communication standard that encrypts and securely transmits records of interference events, evasion operations, and switching. It can be used to ensure the integrity, confidentiality, and non-repudiation of log data, and supports post-event auditing. For example, the communication log encryption upload protocol can include, but is not limited to, one or more of TLS-based log upload, SM4 national cryptographic encryption upload, and blockchain-based evidence-based log transmission. Edge-based real-time comparison of the link quality feedback value sequence with a preset communication security threshold can be performed locally, continuously comparing the feedback value with the threshold to determine if there is continuous degradation. Furthermore, this operation can be implemented by using a sliding window to count the number of consecutive values ​​below the threshold, or by using an exponentially weighted moving average (EWMA) for smoothing before comparison, thereby achieving low-latency fault detection.

[0078] When a continuous feedback value falls below a threshold, a link redundancy activation mechanism is triggered. This can be achieved by initiating a backup link initialization process after certain conditions are met. In an exemplary embodiment, this operation can be implemented by setting a counter that triggers the mechanism after three consecutive values ​​falling below the threshold, or by combining both duration and frequency conditions to prevent communication interruption. Calling the multi-mode communication protocol stack to perform seamless primary / backup link switching can involve activating the backup communication mode and migrating the session context. For example, this operation can achieve transparent IP layer switching through virtual network interface abstraction, or by utilizing the protocol stack's built-in fast reconnection and state synchronization mechanisms, thereby maintaining the continuity of upper-layer application communication. Initiating an encrypted communication log upload protocol can involve encrypting event logs and securely transmitting them to a remote server. Furthermore, this operation can use SM4 encryption and upload via HTTPS, or employ digital signatures + TLS channels to ensure non-repudiation and integrity, thereby guaranteeing the security and auditability of log data.

[0079] For example, in a high-density V2X communication scenario at an urban intersection, the real-time interference avoidance method of the vehicle wireless communication module based on edge computing in this embodiment can be used when multiple autonomous vehicles are communicating densely at complex intersections without traffic lights during morning and evening rush hours, while there are illegal 5.8GHz wireless local area network devices nearby that generate sudden pulse interference. The vehicle-mounted wireless communication module continuously performs edge spectrum sensing, captures the interference, and extracts its high duty cycle and temporal burst characteristics to form a dynamic interference feature set. A lightweight convolutional neural network model classifies it as "civilian illegal transmission interference" and determines it as high-intensity interference based on the high signal entropy value. The system integrates the vehicle's high-precision positioning, the intersection semantic map, and the communication status of neighboring vehicles to construct a heat map of the northwest quadrant of the intersection as a high-risk area. The edge decision engine generates a strategy to switch to the 5.905GHz low-interference sub-channel and outputs the command. The RF front-end synchronously reduces the transmit power and switches to QPSK+1 / 2MCS. After the switch, the link quality feedback shows that the SINR recovers, but then falls below the threshold three times in a row due to the appearance of new interference. The system automatically activates the 5GNR backup link and seamlessly switches through the multi-mode protocol stack. The entire process log is encrypted with SM4 and uploaded to the traffic management center for post-event accountability.

[0080] In one embodiment, a dynamic interference feature set is input into the vehicle-mounted edge computing node. A lightweight convolutional neural network model is used to perform multi-level classification of interference source types, and interference intensity confidence is calculated based on signal entropy values ​​to generate edge-based interference situation assessment data, including:

[0081] The dynamic interference feature set is input into the edge computing unit responsible for spectral feature preprocessing, and time-frequency feature normalization processing is performed to generate a standardized feature vector.

[0082] The edge computing unit for spectral feature preprocessing can be a submodule or computing unit within an onboard edge computing node dedicated to standardizing the original interference features. This can eliminate differences in feature dimensions and dynamic range under different interference scenarios, improving model input consistency. Furthermore, the edge computing unit for spectral feature preprocessing can implement a time-frequency feature normalization algorithm through software task scheduling or a hardware accelerator. Time-frequency feature normalization can be an operation that unifies the scale and standardizes the distribution of time-domain and frequency-domain features in a dynamic interference feature set, which can improve the generalization ability and classification stability of subsequent neural network models. In an exemplary embodiment, time-frequency feature normalization can use Z-score standardization (mean 0, variance 1) or Min-Max linear scaling to the [0, 1] interval. The standardized feature vector can be an interference feature representation with unified dimensions and numerical range formed after normalization, which can be used as a standardized input for a lightweight convolutional neural network model, improving inference accuracy. For example, the standardized feature vector is generated by time-frequency feature normalization and received by the lightweight convolutional neural network model.

[0083] Inputting the dynamic interference feature set into the edge computing unit responsible for spectral feature preprocessing can be achieved by routing the raw interference feature data to a dedicated preprocessing submodule. Furthermore, this operation can be distributed to the preprocessing thread via a task queue, or it can be directly written to the preprocessing unit cache using hardware DMA, thereby achieving pipelined division of labor between sensing and preprocessing and improving processing efficiency. Performing time-frequency feature normalization can involve scaling the input features according to preset statistical parameters. In a specific embodiment, this operation can be dynamically normalized by calculating the sliding window mean and standard deviation online, or statically normalized using global statistical parameters obtained from offline training, thereby eliminating feature shifts caused by sensor differences and environmental fluctuations. Generating a standardized feature vector can be achieved by outputting a normalized fixed-dimensional feature array. Furthermore, this operation can be stored in shared memory as a floating-point tensor, or quantized into INT8 format for NPU to accelerate inference, thus providing a uniformly formatted input for the neural network.

[0084] The standardized feature vectors are input into a pre-deployed lightweight convolutional neural network model, and the local temporal patterns of the interference signal are extracted through the convolutional layers to generate an interference type probability distribution matrix.

[0085] The convolutional layer can be a basic computational unit in a lightweight convolutional neural network model used to extract local spatiotemporal correlations. It can be used to capture local patterns of interference signals in the time-frequency plane, such as pulse periodicity and spectral jumps. In an exemplary embodiment, the convolutional layer can be one or more of the following, including but not limited to depthwise separable convolutional layers, dilated convolutional layers, and 1D temporal convolutional layers. The local temporal pattern can be the repetitive or structured time-frequency behavior of the interference signal within a short time window, which can be used as a key criterion for distinguishing different types of interference sources. For example, the local temporal pattern can include, but is not limited to, fixed-period pulse sequences, random burst clusters, and linear frequency modulation sweep patterns.

[0086] The interference type probability distribution matrix can be a structured representation of the probability of each interference category being assigned from the output of a lightweight convolutional neural network. It can be used to provide soft classification results and support subsequent confidence fusion and uncertainty analysis. Furthermore, the interference type probability distribution matrix is ​​generated by extracting features from convolutional layers and then passing them through fully connected layers or softmax, serving as one of the inputs for confidence-weighted fusion. Inputting the standardized feature vectors into a pre-deployed lightweight convolutional neural network model can be achieved by calling an already loaded model to perform forward inference. In one specific embodiment, this operation can be performed through optimized inference using the TensorRT engine, or by running a TinyML format model on an MCU, thereby initiating the interference type identification process.

[0087] Extracting local temporal patterns of interference signals through convolutional layers can be achieved by sliding convolutional kernels across a time-frequency feature map to activate local responses. Furthermore, this operation can use 1D convolutions to extract pulse interval features along the time axis, or it can employ 2D convolutions to simultaneously model joint time-frequency variations, thereby capturing the structured behavioral patterns unique to interference. Generating an interference type probability distribution matrix can be achieved by outputting the probabilities of each category through a softmax layer. In an exemplary embodiment, this operation can output a K-dimensional probability vector (K being the number of interference categories), and temperature scaling can be added to calibrate the probability distribution, thus providing multi-class soft-decision with confidence levels.

[0088] Based on the probability distribution matrix of interference types and combined with the spectral disorder index output by the signal entropy calculation module, the interference intensity is weighted and fused with confidence to generate marginal interference situation assessment data that includes interference type labels, intensity confidence, and duration prediction.

[0089] The signal entropy calculation module can be a dedicated edge computing unit for calculating the spectral disorder index of interfering signals. It can quantify the randomness and unpredictability of interference, aiding in the assessment of its severity. Furthermore, the signal entropy calculation module can perform statistical calculations on the spectral energy distribution based on Shannon entropy or spectral entropy algorithms. The spectral disorder index, output by the signal entropy calculation module, is a numerical measure reflecting the degree of spectral disorder. It can be used as an important basis for weighted confidence assessment of interference intensity; high disorder typically corresponds to low predictability interference. For example, the spectral disorder index and the interference type probability distribution matrix jointly participate in the confidence-weighted fusion.

[0090] Interference type labels can be semantic identifiers selected from the interference type probability distribution matrix, corresponding to the category with the highest probability. These labels can provide a clear identification of the interference source category for strategy selection. In one specific embodiment, interference type labels may include, but are not limited to, radio interference, illegal broadcasting equipment, and dense V2X co-channel interference. Duration prediction can be an estimate of the expected duration of the current interference event. This can be used to assist in deciding whether long-term avoidance or temporary tolerance is necessary, optimizing resource scheduling. Furthermore, duration prediction can be derived based on historical interference pattern matching or lightweight time-series prediction models (such as a simplified version of LSTM).

[0091] The spectral disorder index, based on the probability distribution matrix of interference types combined with the signal entropy calculation module, can correlate two types of information at the feature or decision level. In an exemplary embodiment, this operation can use spectral disorder as a weight to adjust the probability distribution, or it can construct a dual-input fusion network for joint learning, thereby achieving multi-source evidence fusion and improving assessment robustness. Confidence-weighted fusion of interference intensity can adjust the confidence of the interference intensity determination based on spectral disorder. Further, this operation can use a weighted average: confidence = α·max(probability) + (1−α)·(1−normalized entropy), or it can use logistic regression to fuse the two indices to generate the final confidence, thus avoiding misclassification of high-entropy (high-randomness) interference as weak interference. Generating marginalized interference situation assessment data, including interference type labels, intensity confidence, and duration predictions, can integrate classification, confidence, and prediction results to form a structured output. In one specific embodiment, the operation can be encapsulated as a JSON object containing type, confidence, and duration fields, or it can generate a Protobuf format message for downstream modules to parse, thereby providing semantically rich evaluation results that can directly drive decision-making.

[0092] For example, in a scenario of sudden radio interference at the exit of a highway tunnel, the real-time interference avoidance method of the vehicle-mounted wireless communication module based on edge computing in this embodiment can be used when a vehicle encounters strong pulse interference from a nearby speed measuring radar as it exits the tunnel. The dynamic interference feature set includes high-energy narrow pulses and time-varying frequency offsets. This feature set is fed into the edge computing unit for spectrum feature preprocessing, where Min-Max normalization is performed to generate standardized feature vectors. A lightweight CNN model identifies local temporal patterns with fixed repetition intervals through depthwise separable convolutional layers, outputting a radio interference probability of 0.92. Simultaneously, the signal entropy calculation module detects that the spectrum energy is highly concentrated and the spectrum disorder index is low (small entropy value). Based on this, the system performs weighted fusion of the interference intensity confidence to arrive at a high-intensity, high-confidence conclusion, and predicts that the interference will last for approximately 8 seconds (based on historical radar operating cycles). Finally, edge-based interference situation assessment data with a "radio interference" label, a confidence level of 0.95, and a duration of 8 seconds is generated, providing a highly reliable basis for subsequent switching to non-overlapping C-V2X sub-channels.

[0093] In one embodiment, based on edge interference situation assessment data, and by fusing real-time vehicle location coordinates, road topology semantic graphs, and communication status of surrounding nodes, a communication link risk heatmap is constructed. This heatmap drives the edge decision engine to generate a dynamic spectrum resource reallocation strategy and outputs a channel switching instruction sequence, including:

[0094] It receives interference situation assessment data and simultaneously acquires the real-time vehicle location coordinates output by the high-precision positioning module and the road topology semantic map provided by the vehicle map engine.

[0095] The high-precision positioning module can be an onboard positioning unit that provides vehicle position coordinates at the centimeter to meter level. It can be used to provide precise geographic anchors for communication risk modeling and support the construction of meter-level spatial resolution heatmaps. In an exemplary embodiment, the high-precision positioning module can output real-time position coordinates through multi-source sensor fusion algorithms (such as Kalman filtering). The onboard map engine can be a software module running on the onboard system for parsing and rendering high-precision semantic maps. It can provide structured road topology information and support scene semantic understanding. Furthermore, the onboard map engine can load OpenDRIVE or NDS format map data and provide APIs for upper-layer calls. Receiving interference situation assessment data and simultaneously acquiring the real-time vehicle position coordinates output by the high-precision positioning module and the road topology semantic map provided by the onboard map engine can be achieved by concurrently reading three types of heterogeneous data sources via an internal bus or API. For example, this operation can be achieved by using a timestamp synchronization mechanism to align multi-source data or by batch reading the latest status through a shared memory pool, thereby achieving spatiotemporal alignment between the electromagnetic situation and the geographic context.

[0096] The road topology semantic graph is parsed into node-edge structured data, and combined with the status information broadcast by surrounding vehicle communication nodes, a road network communication risk assessment model is constructed.

[0097] The node-edge structured data can be a data representation that abstracts the road topology semantic graph into vertices (intersections, key points) and connecting edges (lane segments) in graph theory, facilitating graph traversal, path planning, and risk propagation modeling. In a specific embodiment, the node-edge structured data can be one or more of the following, including but not limited to directed weighted graphs, hierarchical road network graphs, and spatiotemporal extended graphs. The road network communication risk assessment model can be a computational model that integrates interference situation, road structure, and neighboring vehicle status to quantify the reliability of communication links in each road segment. This model can be used to achieve a systematic assessment of spatial communication risks, transcending the limitations of single-point perception. Furthermore, the road network communication risk assessment model can be jointly constructed from node-edge structured data and the communication status of surrounding nodes, outputting a channel availability index and an interference propagation attenuation coefficient.

[0098] Parsing a road topology semantic graph into node-edge structured data can be achieved by calling a map engine interface to extract features such as intersections and lanes and constructing the graph structure. For example, this operation can be implemented by building an adjacency list to store the road network topology or by using a graph database (such as an embedded Neo4j) to manage node-edge relationships, thereby transforming the semantic map into a computable network model.

[0099] By combining the status information broadcast by surrounding vehicle communication nodes, a road network communication risk assessment model can be constructed. This can involve mapping information such as SINR, load, and frequency band occupancy reported by neighboring vehicles to the corresponding edges of the road network graph. In an exemplary embodiment, this operation can be achieved by extracting communication status based on the CAM / BSM fields in V2V messages or by using graph neural networks to aggregate risk features of neighboring nodes, thereby enabling collaborative risk modeling to overcome the limitations of single-vehicle perception.

[0100] Based on the road network communication risk assessment model, the channel availability index and interference propagation attenuation coefficient of each communication link are calculated to generate a communication link risk heat map with a spatial resolution of meters.

[0101] The channel availability index can be a frequency band availability score calculated by integrating factors such as interference intensity, adjacent channel leakage, and multipath effects. It can be used as a core input for heatmap generation, reflecting the communication feasibility of a specific location-frequency band combination. Furthermore, the channel availability index can be obtained by fusing and normalizing multidimensional indicators based on a weighted function.

[0102] The interference propagation attenuation coefficient can be a physical parameter describing the degree of attenuation of interference signals in space with distance and obstruction. It can be used to correct the spatial extrapolation of local interference intensity and improve the accuracy of risk prediction. In a specific embodiment, the interference propagation attenuation coefficient may include, but is not limited to, one or more of the following: free space path loss coefficient, building penetration loss factor, vegetation scattering attenuation factor, etc. Based on the road network communication risk assessment model, the channel availability index and interference propagation attenuation coefficient of each communication link can be calculated. This can be achieved by performing a multi-factor evaluation on each candidate frequency band at each road network edge. For example, this operation can be achieved by using a lookup table method combined with a pre-stored propagation model to quickly estimate attenuation or by running a simplified ray tracing simulation to calculate the impact of obstruction online, thereby generating fine-grained, physically reliable link quality predictions.

[0103] Generating a communication link risk heatmap with a spatial resolution down to the meter level can be achieved by interpolating risk values ​​calculated along the road network edges onto a continuous geographic grid. In one exemplary embodiment, this operation can be accomplished by generating a raster heatmap using inverse distance weighted (IDW) interpolation or by preserving the boundaries of high-risk road segments in vector form, thereby providing a visualized and queryable spatial risk distribution.

[0104] Based on the communication link risk heatmap, the optimal spectrum resource reallocation path is selected through the heuristic search algorithm of the edge decision engine, and a channel handover command sequence containing the target frequency band, power adjustment amount and handover timing is output.

[0105] The heuristic search algorithm can be a path planning method that quickly finds near-optimal solutions under limited computational resources, and can be used to efficiently screen spectrum reallocation paths under the constraints of communication risk heatmaps. Furthermore, the heuristic search algorithm can include, but is not limited to, one or more of the following: A* algorithm variants, greedy best-first search, and restricted depth-first search. The spectrum resource reallocation path can be an optimized handover sequence from the current channel to the target channel, including intermediate frequency hopping steps and resource selection, and can be used to guide the system to complete a smooth, low-collision spectrum migration. In a specific embodiment, the spectrum resource reallocation path can include, but is not limited to, one or more of the following: single-step direct handover path, multi-hop progressive avoidance path, and power-band joint adjustment path.

[0106] The target frequency band can be a new operating frequency band selected after evaluation to avoid interference. It can be used as a core parameter of the channel handover command to determine the physical layer communication frequency. For example, the target frequency band can include, but is not limited to, one or more of the following: vehicle-to-everything (V2X) communication control channel sub-band, 5G NR-U unlicensed frequency band, DSRC backup channel, etc. The power adjustment amount can be the transmit power increment or proportion that needs to be adjusted to match the new channel environment. It can be used to optimize the link budget for coordinated frequency band handover and avoid over-coverage or under-coverage. Further, the power adjustment amount can include, but is not limited to, one or more of the following: absolute power value (dBm), relative gain adjustment (dB), path loss compensation amount, etc. The handover timing can be a precise time arrangement or triggering condition for performing the channel handover operation. It can be used to ensure that the handover action is aligned with vehicle movement and communication cycle, reducing service interruption. In an exemplary embodiment, the handover timing can include, but is not limited to, one or more of the following: frame boundary aligned handover, event-driven instantaneous handover, pre-scheduled delayed handover, etc.

[0107] Based on the communication link risk heatmap, the optimal spectrum resource reallocation path is selected through a heuristic search algorithm of the edge decision engine. This can be achieved by searching for low-risk, high-availability band switching sequences under heatmap constraints. For example, this operation can be implemented by running an A* search with channel availability as the edge weight or by using a greedy strategy to select locally optimal frequency bands hop-by-hop, thus enabling forward-looking, low-overhead spectrum scheduling. The output is a channel switching instruction sequence containing the target frequency band, power adjustment amount, and switching timing. This can be achieved by encapsulating the search results into structured control commands. In a specific embodiment, this operation can be implemented by generating JSON format instructions containing frequency band, power, and timing fields, or by outputting binary control flow for the underlying driver to parse, thereby providing a complete and executable multi-parameter adjustment scheme for the RF front-end.

[0108] For example, in a scenario with dense V2X communication under an urban overpass, the real-time interference avoidance method of the vehicle-mounted wireless communication module based on edge computing in this embodiment can be: when a vehicle is driving under an overpass, it encounters co-channel interference generated by dense vehicle-to-everything (V2X) communication above. The system receives edge interference situation assessment data (determined to be high-intensity V2X interference), and simultaneously obtains the current location (30 meters under the bridge) output by the high-precision positioning module and the elevated-ground dual-layer road network topology provided by the vehicle map engine. The map is parsed into a graph structure containing upper and lower layer nodes and vertical connecting edges. Combining the channel occupancy rate (85% in the specified frequency band) and SINR (average 12dB) of the surrounding vehicle broadcasts, a road network communication risk assessment model is constructed. The model calculates that the channel availability index of the specified frequency band in the area under the bridge is only 0.3, and the interference propagation attenuation coefficient is abnormally low due to reflection from the metal bridge surface (weak attenuation). The system generates a meter-level heat map, showing that the entire section under the bridge is a red high-risk area. The edge decision engine uses the A* variant algorithm to select the target frequency band (availability 0.82) under the guidance of the heat map, and calculates that the transmit power needs to be increased by 2dB to compensate for diffraction loss. The switching timing is set to the start time of the next V2X broadcast frame. Finally, the system outputs a channel switching command sequence containing this triplet to achieve active avoidance.

[0109] In one embodiment, the channel switching command sequence is applied to the radio frequency front-end controller of the wireless communication module to perform adaptive transmission power adjustment and modulation and coding scheme switching, and simultaneously collects the link quality feedback value sequence after the avoidance operation, including:

[0110] The channel switching command sequence is transmitted to the digital signal processing unit of the RF front-end controller to parse the target frequency band parameters and configure the phase-locked loop frequency synthesizer.

[0111] The digital signal processing unit (DSP) can be an embedded processing module integrated into the RF front-end controller, responsible for parsing control commands and driving RF hardware configuration. It can be used to achieve precise mapping from channel switching commands to hardware register configurations. In an exemplary embodiment, the DSP can run a command parsing algorithm through firmware to output control words such as frequency and gain to the RF submodule. The target frequency band parameters can be the new operating frequency and related RF configuration parameters specified in the channel switching command, which can be used as input for reconfiguration of the phase-locked loop frequency synthesizer. For example, the target frequency band parameters can include, but are not limited to, one or more of the following: center frequency value, channel bandwidth identifier, and frequency offset tolerance parameters.

[0112] A phase-locked loop (PLL) frequency synthesizer can be a frequency synthesis circuit used to generate stable local oscillation signals to support RF transceiver operations. It can be used to achieve fast and accurate frequency switching, supporting dynamic spectrum access. Furthermore, the PLL frequency synthesizer can receive digital control words and adjust the output target frequency of the voltage-controlled oscillator (VCO). The digital signal processing unit (DSP) that transmits the channel switching command sequence to the RF front-end controller can write the commands into the DSP control register via an internal high-speed bus (such as AXI or SPI). Furthermore, this operation can be achieved by using DMA to batch transfer multiple commands or by triggering the command loading process via an interrupt, thereby establishing a control path between upper-level decision-making and lower-level RF hardware.

[0113] Parsing the target frequency band parameters and configuring the phase-locked loop (PLL) frequency synthesizer can be achieved by extracting frequency values ​​using a DSP, converting them into control words such as the division ratio and charge pump current, and then writing them into the PLL register. In one specific embodiment, this operation can be achieved by using a pre-calibrated lookup table to accelerate frequency switching or by enabling a fast lock-on mode to shorten the PLL settling time, thereby enabling millisecond-level frequency switching and supporting dynamic spectrum access.

[0114] Based on power regulation commands, the bias voltage of the power amplifier is dynamically adjusted to achieve closed-loop adaptive regulation of the transmitted power.

[0115] The power adjustment command can be a control parameter issued by the upper-layer decision engine to adjust the transmit power, guiding the power amplifier to achieve a balance between energy efficiency and link quality. For example, the power adjustment command may include, but is not limited to, one or more of the following: target output power (dBm), relative gain adjustment step size, and maximum power limit threshold. The power amplifier can be a radio frequency power device that amplifies the baseband signal to the required transmit power level, determining the actual radiated energy intensity and affecting communication distance and interference levels. In an exemplary embodiment, the power amplifier can control its gain and output power by adjusting the bias voltage or supply current. The bias voltage can be a DC control voltage applied to the gate or base of the power amplifier, directly regulating the amplifier's operating point and output power, and is a key physical quantity for achieving adaptive power adjustment. Furthermore, the bias voltage may include, but is not limited to, one or more of the following: static bias voltage, dynamically adjustable bias, and temperature-compensated bias.

[0116] Based on power adjustment commands, the bias voltage of the power amplifier is dynamically adjusted. This can be achieved by the DSP controlling the DAC output voltage to the PA bias pin. In one specific embodiment, this operation can be implemented by using a digital variable gain amplifier (DVGA) instead of analog bias adjustment or by combining a temperature sensor for bias compensation, thereby enabling fine-grained, low-power adjustment of the transmit power. Closed-loop adaptive adjustment of the transmit power can be achieved by dynamically correcting the power command based on the SINR or ACK / NACK feedback from the receiver. Furthermore, this operation can be achieved by pre-setting the initial power based on an open-loop path loss model and then fine-tuning it in the closed loop, or by using a proportional-integral (PI) controller to stabilize the power output, thereby minimizing the transmit power while meeting QoS requirements.

[0117] Based on the modulation and coding scheme switching instruction, the modulation order and coding rate of the physical layer protocol stack are updated in real time to generate an optimized communication waveform.

[0118] The modulation and coding scheme switching instruction can be an operation command that instructs the physical layer to change modulation and coding parameters, and can be used to drive the communication waveform to adapt to the current channel conditions. For example, the modulation and coding scheme switching instruction can include, but is not limited to, one or more of the following: MCS index number, modulation type + code rate combination, adaptive MCS selection flag, etc. The physical layer protocol stack can be a software / firmware module that implements the physical layer functions of wireless communication (such as modulation, coding, and synchronization), and can be used to generate communication waveforms conforming to the new parameters in real time according to the instructions. In an exemplary embodiment, the physical layer protocol stack can update the modulator and encoder parameters by configuring registers or calling APIs.

[0119] The modulation order can be the number of bits carried by each symbol in the modulation scheme, reflecting spectral efficiency. It can be used to reduce the order to improve demodulation robustness when channel quality deteriorates. Furthermore, the modulation order can include, but is not limited to, one or more of BPSK (1 bit / symbol), QPSK (2 bits / symbol), and 16-QAM (4 bits / symbol). The coding rate can be the ratio of information bits to total transmitted bits in the forward error correction code. It can be used to provide stronger error correction capabilities at a low coding rate, suitable for high-interference environments. For example, the coding rate can include, but is not limited to, one or more of 1 / 2 code rate, 3 / 4 code rate, and 5 / 6 code rate. The communication waveform can be the actual transmitted signal time-domain waveform formed after modulation and coding. It can be used to carry information and propagate in the wireless channel, and its parameters directly affect link performance. In a specific embodiment, the communication waveform can be generated by the physical layer protocol stack based on the modulation order and coding rate.

[0120] According to the modulation and coding scheme switching instruction, the modulation order and coding rate of the physical layer protocol stack are updated in real time. This can be achieved by modifying the modulator configuration and FEC encoder parameters in the baseband processor. Furthermore, this operation can be implemented by loading parameters from a pre-stored MCS configuration table or dynamically recompiling the LDPC code generation matrix, thereby adapting the communication waveform to the current channel conditions. Generating the optimized communication waveform can be achieved by the baseband chip performing IFFT (OFDM) or pulse shaping (single carrier) according to the new parameters. In one specific embodiment, this operation can be achieved by enabling peak clipping (CFR) to reduce PAPR to accommodate PA nonlinearity, or by inserting additional pilots to improve channel estimation accuracy, thereby outputting a physical layer signal with enhanced anti-interference capabilities.

[0121] During waveform transmission, the signal-to-noise ratio, bit error rate, and throughput parameters are synchronously collected through the channel state information feedback unit of the receiving link to form a link quality feedback value sequence.

[0122] The channel state information feedback unit of the receiving link can be a monitoring module located at the receiving end, used to estimate and report channel quality indicators in real time, and can be used to provide the original data source for the link quality feedback value sequence. In an exemplary embodiment, the channel state information feedback unit of the receiving link can calculate parameters such as SINR and BER based on pilot signals or decoding results and transmit them back through the control channel. The signal-to-noise ratio (SNR) can be the ratio of received signal power to noise power, measuring the basic channel quality and can be used as one of the core indicators for evaluating interference suppression effectiveness. For example, the SNR can include, but is not limited to, one or more of SINR (signal-to-interference-plus-noise ratio), SNR (single signal-to-noise ratio), and RSRP (reference signal received power). The bit error rate (BER) can be the ratio of the number of decoded erroneous bits at the receiving end to the total number of transmitted bits, and can be used to directly reflect communication reliability and is sensitive to sudden interference. Further, the BER can include, but is not limited to, one or more of BER (bit error rate), PER (packet error rate), and FEC (featured error rate). The throughput parameter can be the amount of effective data successfully transmitted per unit time, and can be used to comprehensively reflect the actual service capability of the link under the current MCS and interference. In one specific embodiment, the throughput parameter may include, but is not limited to, one or more of the following: application layer throughput, MAC layer effective throughput, and PHY layer theoretical throughput.

[0123] During waveform transmission, signal-to-noise ratio (SNR), bit error rate (BER), and throughput parameters are synchronously collected through the channel state information feedback unit of the receiving link. This can be achieved by the receiver calculating and buffering CSI metrics in each frame or subframe period. Furthermore, this operation can be implemented by using a sliding window to smooth instantaneous fluctuations or by transmitting CSI summaries only on critical control channels, thus providing highly timely link performance observations. Forming a link quality feedback value sequence can be achieved by organizing the collected multidimensional metrics into a structured data stream by timestamps. In a specific embodiment, this operation can be implemented by storing the most recent N samples in a circular buffer, or by appending timestamps and frequency band IDs to form metadata, thereby providing quantified input for upper-layer redundancy triggering and policy verification.

[0124] For example, in a scenario where there is sudden co-channel interference at the entrance of a highway service area, the real-time interference avoidance method of the vehicle-mounted wireless communication module based on edge computing in this embodiment can be as follows: when a vehicle enters the service area ramp and encounters co-channel interference generated by dense DSRC equipment, the upper-layer decision engine issues a channel switching command sequence: target frequency band 5.905GHz, power increase of 1.5dB, and MCS reduced to QPSK+1 / 2. This command is transmitted to the digital signal processing unit of the RF front-end controller. After DSP parsing, the phase-locked loop frequency synthesizer is configured to switch to the 5.905GHz center frequency point; at the same time, the DAC output corresponding bias voltage increases the power amplifier gain; the physical layer protocol stack synchronously reduces the modulation order from 16-QAM to QPSK and the coding rate from 3 / 4 to 1 / 2, generating a communication waveform with enhanced robustness; during the transmission of the new waveform, the channel state information feedback unit of the receiving link collects SINR (rising back to 18dB) and bit error rate (reduced to 10⁻) every 10ms. 4 The data collection and throughput (stable at 2Mbps) form a sequence of link quality feedback values. This sequence is subsequently used to verify the effectiveness of the circumvention and to avoid triggering redundancy mechanisms due to continuous compliance, thus achieving efficient adaptive communication.

[0125] In one embodiment, the link quality feedback value sequence is compared with a preset communication security threshold in real time at the edge. When the continuous feedback value is lower than the threshold, a link redundancy activation mechanism is triggered, the multi-mode communication protocol stack is invoked to perform seamless switching between primary and backup links, and the communication log encrypted upload protocol is started, including:

[0126] A real-time comparison engine is deployed within the vehicle-mounted edge computing node to perform sliding window analysis on the link quality feedback value sequence and preset multi-level communication security thresholds.

[0127] The real-time comparison engine can be a lightweight judgment module deployed in an onboard edge computing node for continuous analysis of the relationship between link quality and security thresholds. It can be used to achieve low-latency, highly robust link status monitoring and tiered response triggering. In an exemplary embodiment, the real-time comparison engine can perform continuous statistical analysis of the feedback sequence using a sliding window algorithm. The multi-level communication security thresholds can be multiple link quality critical values ​​divided according to risk levels, used to distinguish between warning and handover actions, and can be used to support tiered response strategies, avoiding overreaction or response lag. For example, the multi-level communication security thresholds can include, but are not limited to, one or more of the following: a first-level security threshold (warning threshold), a second-level security threshold (handover threshold), and a third-level security threshold (emergency shutdown threshold). Sliding window analysis can be a method of continuously sampling and statistically judging link quality feedback values ​​within a fixed-length time window, which can be used to filter instantaneous fluctuations and identify continuous degradation trends. Furthermore, sliding window analysis can employ methods such as counting N consecutive values ​​below a threshold based on a counter or comparing results after smoothing with an exponentially weighted moving average (EWMA).

[0128] Deploying a real-time comparison engine within the vehicle-mounted edge computing node allows for sliding window analysis of the link quality feedback value sequence against preset multi-level communication security thresholds. This can be achieved by continuously running windowed comparison logic locally and outputting status flags. Furthermore, this operation can be implemented by using a circular buffer to maintain the most recent 10 samples for majority decision-making or by smoothing them with an FIR filter before comparing them with the threshold, thus enabling jitter-resistant link quality trend assessment.

[0129] When the continuous feedback value is detected to be lower than the first-level safety threshold, a link degradation warning signal is generated; when the continuous feedback value is lower than the second-level safety threshold, the link redundancy activation mechanism is triggered.

[0130] The primary security threshold can be a critical value for link quality that triggers a link degradation warning but does not initiate a switchover. It can be used to provide early risk alerts for logging or policy preloading. In one specific embodiment, the primary security threshold may include, but is not limited to, one or more of the following: SINR = 15dB, PER = 10⁻², and throughput decrease of 30%. The secondary security threshold can be a critical value for link quality that triggers a link redundancy activation mechanism. It can be used as a basis for primary / backup switchover decisions to ensure communication continuity. For example, the secondary security threshold may include, but is not limited to, one or more of the following: SINR = 8dB, PER = 10⁻¹, and throughput below the minimum QoS requirement.

[0131] Link degradation early warning signals can be non-disruptive alarm flags generated when link quality continuously falls below a primary threshold. These signals can be used to preload backup link resources or notify upper-layer applications to degrade. In one exemplary embodiment, the link degradation early warning signal can be generated by a real-time comparison engine, without triggering a switchover but logging the data. When a continuous feedback value is detected to be below the primary safety threshold, a link degradation early warning signal is generated, which can involve setting a warning flag and recording a timestamp after the condition is met. Furthermore, this operation can be implemented by triggering log pre-caching or notifying the application layer to enter degradation mode, thus providing early risk warnings without interrupting communication. When a continuous feedback value falls below the secondary safety threshold, a link redundancy activation mechanism is triggered, which can involve initiating a redundancy switchover process after meeting more stringent conditions. Furthermore, this operation can be triggered by three consecutive values ​​below the threshold or by combining a duration (e.g., >100ms) for dual judgment, thereby ensuring the continuity of communication services.

[0132] Based on the link redundancy activation mechanism, the switching control module of the multi-mode communication protocol stack is called to migrate the traffic of the main communication link to the pre-negotiated backup link and execute the handshake protocol to verify the integrity of the switching.

[0133] The handover control module can be a sub-component in the multi-mode communication protocol stack responsible for coordinating the state transition and traffic switching of the primary and backup links. It can be used to achieve seamless, low-latency communication mode switching. In one specific embodiment, the handover control module can maintain the primary and backup link session context and perform state synchronization and traffic redirection. The primary communication link can be the currently active wireless connection carrying the main V2X or cellular services, and can be used to provide the main data transmission channel. For example, the primary communication link can include, but is not limited to, one or more of the following: C-V2X PC5 direct link, 5G NR-Uu cellular link, DSRC basic service set, etc.

[0134] The pre-negotiated backup link can be a redundant communication path that has been established and maintained in a synchronized state during the normal operation of the primary link. It can be used to ensure a seamless transition without the need for re-establishing the link during handover. In an exemplary embodiment, the pre-negotiated backup link can include, but is not limited to, one or more of the following: a 5G NSA backup link, an LTE-V2X auxiliary link, and a Wi-Fi 6 vehicle-to-everything (V2X) cooperative channel. The handshake protocol can be a lightweight interactive process used during primary / backup handover to verify the readiness state of the backup link and data consistency. It can be used to confirm the integrity of the handover and prevent data loss or duplication. Furthermore, the handshake protocol can include, but is not limited to, one or more of the following: a three-step confirmation handshake, a sequence number synchronization protocol, and ACK / NACK fast verification.

[0135] Switching integrity can be a quality indicator measuring service continuity and data consistency during primary / standby switching, and can be used as a criterion for determining whether a switch is successful. In a specific embodiment, switching integrity can be determined by the status code or sequence number matching result returned by the handshake protocol. Based on the link redundancy activation mechanism, the switching control module of the multi-mode communication protocol stack is invoked to migrate traffic from the primary communication link to the pre-negotiated standby link. This can be achieved by calling the switching control module via API to perform traffic redirection. Furthermore, this operation can be implemented by using virtual network interface abstraction to achieve transparent IP layer switching or by completing session migration through the internal state machine of the protocol stack, thereby achieving seamless service migration. The handshake protocol is executed to verify switching integrity, which can be achieved by the standby link sending an acknowledgment message and the primary control module verifying sequence number consistency. Furthermore, this operation can be achieved by exchanging the sequence number of the last successfully received packet or by performing a short-term bidirectional ping to verify connectivity, thereby ensuring data integrity and no data loss after the switch.

[0136] After the switch is completed, the log encryption unit is activated to encrypt the interference event data using AES-256 and upload it to the roadside edge server through a low duty cycle transmission channel.

[0137] The log encryption unit can be a dedicated hardware or software module in the vehicle system for performing high-strength encryption on communication event logs, ensuring log security during transmission and storage. In one specific embodiment, the log encryption unit can integrate an AES-256 encryption engine to support secure key management. AES-256 encryption can be an advanced encryption standard symmetric encryption algorithm using a 256-bit key, providing high-security log protection and meeting vehicle network security compliance requirements. For example, AES-256 encryption can include, but is not limited to, one or more of the following: CTR mode AES-256, GCM certified encryption mode, and alternatives to the national standard SM4 (optional).

[0138] The low duty cycle transmission channel can be a dedicated channel that occupies minimal main communication bandwidth and is used to transmit small data packets such as control or logs. This can be used to prevent encrypted log uploads from interfering with main service communication. In an exemplary embodiment, the low duty cycle transmission channel may include, but is not limited to, one or more of the following: V2X reserved signaling sub-channel, 5GURLLC low-load control channel, and periodic idle frame insertion channel. The roadside edge server can be an edge node deployed on the roadside with local computing and storage capabilities. It can be used to receive and store encrypted logs, supporting near-end auditing and fast backtracking. Furthermore, the roadside edge server can receive vehicle-uploaded data via V2I or cellular backhaul interfaces.

[0139] After the handover is complete, the log encryption unit is activated to encrypt the interference event data using AES-256. This can be achieved by calling an encryption API, inputting the original log, and outputting ciphertext. Furthermore, this operation can be accelerated using a Hardware Security Module (HSM) or by adding a digital signature to achieve non-repudiation, thus ensuring the confidentiality and integrity of the event data. The data is then uploaded to the roadside edge server via a low duty cycle transmission channel. This can be achieved by encapsulating the encrypted log into small packets and sending them on a dedicated low-priority channel. Furthermore, this operation can be achieved by piggybacking the upload using reserved fields in V2X broadcast messages or by sending it during a 5G inactive DRX cycle, thereby avoiding the occupation of main communication resources and achieving covert and reliable uploading.

[0140] Taking the example of a sudden strong interference at the exit of a city tunnel causing continuous link degradation, the real-time interference avoidance method of the vehicle-mounted wireless communication module based on edge computing in this embodiment can be as follows: when the vehicle exits the tunnel, if the C-V2X main link SINR is lower than 8dB (secondary threshold) for 5 consecutive times due to dense radar interference, the real-time comparison engine triggers the link redundancy activation mechanism; the switching control module immediately calls the multi-mode communication protocol stack to migrate the service flow to the pre-activated 5GNR backup link and executes the three-step handshake protocol to confirm the sequence number synchronization; after the switching is completed, the log encryption unit automatically extracts the interference characteristics, avoidance actions and switching timing data, and encrypts them using AES-256-GCM mode; the encrypted log is uploaded to the roadside edge server deployed at the tunnel exit through the low duty cycle signaling sub-channel reserved by V2X without affecting the main service; the whole process takes less than 50ms, the upper-layer autonomous driving application is not aware of the interruption, and the responsibility can be traced by decrypting the log through the server afterward.

[0141] In one embodiment, based on a road network communication risk assessment model, the channel availability index and interference propagation attenuation coefficient of each communication link are calculated to generate a communication link risk heatmap with a spatial resolution of meters, including:

[0142] Lane curvature, obstacle density, and signal obstruction area parameters are extracted from the road topology semantic map to construct a spatial propagation loss feature vector.

[0143] Lane curvature can be a quantitative parameter describing the geometry of a road curve, reflecting the degree of path curvature. It can be used to assess the occlusion effect of curves on direct signals, affecting line-of-sight propagation. In an exemplary embodiment, lane curvature can include, but is not limited to, one or more of the following: local radius of curvature, cumulative steering angle, and rate of change of curvature. Obstacle density can be the number or coverage ratio of static or dynamic obstructions (such as buildings or large vehicles) per unit length of road. It can be used to characterize non-line-of-sight (NLOS) propagation probability, affecting multipath and penetration loss. For example, obstacle density can include building projection density, large vehicle distribution density, vegetation occlusion index, etc. Signal occlusion area parameters can be semantic labels or geometric boundaries that identify areas with severe electromagnetic signal attenuation, such as tunnels, underpasses, and elevated bridges. They can be used to directly mark communication blind spots as hard constraints for high-risk areas. Further, signal occlusion area parameters can include tunnel entrance / exit coordinates, elevated bridge projection polygons, and markers of densely packed metal structures.

[0144] Extracting lane curvature, obstacle density, and signal occlusion area parameters from the road topology semantic map can be achieved by calling a map engine API to parse high-precision map elements and calculate geometric / semantic features. In one specific embodiment, this operation can be achieved by calculating curvature using OpenDRIVE's road / laneSection data or by statistically analyzing the percentage of obstacle pixels per unit area using a rasterized map, thereby transforming the static road environment into a calculable propagation loss indicator. Constructing a spatial propagation loss feature vector can be achieved by combining the three types of extracted parameters into a fixed-dimensional vector in a preset order. Furthermore, this operation can be achieved by normalizing and concatenating them into a 10-dimensional feature vector, or by adding timestamps to support dynamic environment updates, thus forming a standardized environment input for subsequent fusion.

[0145] By combining the channel occupancy rate in the communication status of surrounding nodes with historical interference records, the probability value of co-channel interference between links is calculated.

[0146] Channel occupancy rate can be the proportion of time spent or the proportion of resource blocks occupied by surrounding nodes on a specific frequency band. It reflects the intensity of co-channel competition and is a key factor in calculating the interference probability. In an exemplary embodiment, channel occupancy rate can be obtained through channel status reports in V2X broadcast messages or MAC layer monitoring. Historical interference records can be locally stored information on the time, location, type, and intensity of past interference events, which can be used to statistically infer the interference recurrence probability of a specific road segment. Furthermore, historical interference records are maintained by a local cache module before the communication log encryption upload protocol. The co-channel interference probability value can be the likelihood of a link suffering co-channel interference calculated based on current channel occupancy and historical data. It can be used to quantify the risk of cooperative interference and supplement local perception blind spots. In a specific embodiment, the co-channel interference probability value can be calculated through Bayesian estimation or sliding window frequency statistics.

[0147] By combining channel occupancy rates from surrounding node communication status with historical interference records, the probability of co-channel interference between links can be calculated. This can be achieved by fusing real-time occupancy data with historical frequencies and using a statistical model to estimate the probability of interference occurrence. For example, this operation can be implemented by modeling the arrival rate of interference events using a Poisson process or by setting a probability threshold based on the average occupancy rate within a sliding window, thereby capturing spatiotemporally correlated cooperative interference patterns.

[0148] The spatial propagation loss feature vector and the co-channel interference probability value are input into a dynamic weight fusion unit to generate a channel availability index.

[0149] The spatial propagation loss feature vector can be a multi-dimensional input vector for modeling path loss, composed of lane curvature, obstacle density, and occlusion region parameters. This vector can provide prior environmental knowledge for channel availability calculation. Furthermore, the spatial propagation loss feature vector is constructed by extracting parameters from the road topology semantic graph and serves as one of the inputs to the dynamic weight fusion unit. The dynamic weight fusion unit can be a lightweight fusion module that adaptively adjusts the weights of each input feature according to the context. It can be used to achieve non-linear weighting of propagation loss and interference probability, improving the accuracy of the availability index. In a specific embodiment, the weights of the dynamic weight fusion unit can be dynamically updated by vehicle speed, interference type, or time; for example, at high speeds, the propagation model is given more weight.

[0150] The spatial propagation loss feature vector and the probability value of co-channel interference are input into a dynamic weighted fusion processor. This can be achieved by adjusting the weights according to the current context (e.g., vehicle speed > 60 km / h) and then performing a weighted sum or a nonlinear combination. Furthermore, this operation can be implemented by automatically learning the weights using an attention mechanism or by selecting a preset weighted combination based on the scenario category using a lookup table, thereby generating an availability assessment jointly driven by the environment and interference. A channel availability index is generated, which can be a continuous value between 0 and 1, with higher values ​​indicating more reliable links. For example, this operation can be achieved by compressing the fusion result to [0, 1] using a sigmoid function, or by adding a confidence interval to represent the assessment uncertainty, thereby providing a basic availability score for the heatmap.

[0151] Based on the channel availability index and the intensity confidence in the interference situation assessment data, the interference propagation attenuation coefficient is derived through a Gaussian process regression model, generating a communication link risk heatmap with a spatial resolution of meters.

[0152] The intensity confidence level can be a reliable probability of judging the interference intensity from marginal interference situation assessment data. It can be used as a spatial observation prior for Gaussian process regression to guide interference field interpolation. Furthermore, the intensity confidence level and the channel availability index are jointly input into the Gaussian process regression model. The Gaussian process regression model can be a probabilistic regression model based on a Bayesian nonparametric method, suitable for small-sample spatial interpolation, and can be used to derive the interference propagation attenuation coefficient in continuous space based on sparse interference sampling points. In an exemplary embodiment, the Gaussian process regression model uses the interference intensity confidence level at known locations as the observation value and performs smooth extrapolation using a radial basis function (RBF) kernel.

[0153] Based on the channel availability index and the intensity confidence in the interference situation assessment data, the interference propagation attenuation coefficient is derived using a Gaussian process regression model. This can be achieved by using the interference intensity confidence at the vehicle's current location as the observation point and using GPR to predict the attenuation trend in the vicinity. Furthermore, this operation can reduce computational complexity by using sparse GPR or by fixing hyperparameters to adapt to the real-time requirements of the vehicle, thus enabling continuous spatial modeling of the interference field and overcoming the limitations of discrete sampling. Generating a communication link risk heatmap with a spatial resolution down to the meter level can be achieved by fusing the channel availability index and the interference attenuation coefficient and mapping it to a geographic grid. In a specific embodiment, this operation can be achieved by storing the comprehensive risk score in a 1m×1m grid or by using color coding to distinguish between occlusion-dominated and interference-dominated risks, thereby providing high-precision, semantically rich risk visualization.

[0154] For example, in a high-density communication scenario at a crossroads in an urban commercial area, the real-time interference avoidance method for the vehicle-mounted wireless communication module based on edge computing in this embodiment can be as follows: When a vehicle approaches the crossroads, the road topology semantic map shows that the lane curvature in this area is low (for straight traffic), but the obstacle density is high (surrounding shops and bus stops), and there is a signal obstruction area (underground passage entrance). The system extracts parameters to construct a spatial propagation loss feature vector; at the same time, it receives C-V2X channel occupancy data from five surrounding vehicles, which is 78%, and queries local historical records to find that high-frequency radar interference occurs at this intersection every day from 17:00 to 19:00. Based on this, the co-channel interference probability is calculated to be 0.65. The dynamic weighted fusion unit assigns a higher weight to the interference probability based on the current vehicle speed (30km / h), generating a channel availability index of 0.42. The Gaussian process regression model uses the current vehicle position interference intensity confidence level of 0.9 as the observation point and derives that the interference decreases exponentially with distance, with an attenuation coefficient of 0.12 / m. Finally, a meter-level heat map is generated, clearly showing that the center of the intersection is a red high-risk zone (mainly due to co-channel interference), while the underground entrance is a dark red blind zone (mainly due to occlusion), and the eastbound straight lane is a green low-risk zone. The edge decision engine selects the eastbound sub-channel and switches in advance accordingly to achieve proactive avoidance.

[0155] In one embodiment, the interference propagation attenuation coefficient is derived using a Gaussian process regression model to generate a communication link risk heatmap with spatial resolution down to the meter level, including:

[0156] Using the channel availability index as an input variable, the covariance function kernel of the Gaussian process regression model is constructed.

[0157] The covariance function kernel can be a function used in a Gaussian process regression model to describe the output correlation between any two points, and can be used to encode the spatial continuity and attenuation characteristics of interference propagation. In an exemplary embodiment, the covariance function kernel can be one or more of the following, including but not limited to the squared exponential kernel (RBF kernel), the Matérn kernel, and the periodic kernel. Furthermore, constructing the covariance function kernel of the Gaussian process regression model using the channel availability index as an input variable can be achieved by using the location corresponding to the channel availability index as input, selecting the kernel function form, and initializing the hyperparameters. For example, this operation can be achieved by using a squared exponential kernel or a Matérn3 / 2 kernel to enhance the modeling capability for non-smooth attenuation, thereby establishing a probabilistic mapping relationship between interference intensity and spatial location.

[0158] Based on historical interference propagation sample data, the hyperparameters of the kernel function are optimized to make the model output match the measured attenuation curve.

[0159] Historical interference propagation sample data can be a training dataset consisting of interference intensity, location, and environmental context recorded by the vehicle during its past driving. This dataset can be used to calibrate the hyperparameters of the GPR model, improving its adaptability to the local electromagnetic environment. In one specific embodiment, historical interference propagation sample data can be extracted from locally stored interference event logs, using geographically tagged intensity-distance samples. The hyperparameters of the kernel function can be key parameters controlling the shape of the covariance function, such as length scale and signal variance, which can be used to determine the spatial range and severity of interference impact. Furthermore, the hyperparameters of the kernel function can include, but are not limited to, length scale (controlling the relevant distance), signal variance (controlling the output amplitude), and noise variance (controlling observation uncertainty). The measured attenuation curve can be an empirical curve showing the relationship between interference intensity and propagation distance obtained through actual measurements. This curve can be used as a target for model optimization to ensure that the GPR output conforms to physical laws. In an exemplary embodiment, the measured attenuation curve is obtained by fitting the SINR or RSSI from the link quality feedback value sequence with the location data. Furthermore, based on historical interference propagation sample data, the hyperparameters of the kernel function can be optimized to make the model output match the measured attenuation curve. This can be achieved by adjusting parameters such as the length scale through maximum likelihood estimation, minimizing the error between prediction and measurement. For example, this operation can be achieved by using the L-BFGS optimizer to solve for hyperparameters or by using k-fold cross-validation to prevent overfitting, thereby improving the generalization accuracy of the GPR model to the local electromagnetic environment.

[0160] In the gridded coordinate system of the road topology semantic graph, the interference propagation attenuation coefficient is calculated for each grid node, generating an attenuation coefficient spatial distribution matrix.

[0161] The gridded coordinate system can be a two-dimensional spatial indexing system that divides the road topology semantic map coverage area into regular meter-level grids, providing a structured spatial sampling framework for calculating interference attenuation coefficients. In one specific embodiment, the gridded coordinate system generates local grids with the vehicle's current position as the center and a step size of 1m × 1m. Each grid node can be a discrete spatial sampling point in the gridded coordinate system, possessing unique (x, y) coordinates, which can be used as input locations for GPR inference to calculate local attenuation coefficients. For example, grid nodes can include, but are not limited to, intersection center nodes, lane centerline nodes, and shading zone boundary nodes. The attenuation coefficient spatial distribution matrix can be a two-dimensional numerical matrix storing the interference propagation attenuation coefficients corresponding to each grid node, serving as an intermediate discrete representation for heatmap generation. Further, the attenuation coefficient spatial distribution matrix is ​​generated point-by-point on the grid nodes by a Gaussian process regression model.

[0162] Furthermore, in the gridded coordinate system of the road topology semantic graph, the interference propagation attenuation coefficient is calculated for each grid node. This can be achieved by performing a GPR posterior prediction on each grid node and outputting the mean attenuation coefficient for that point. Exemplarily, this operation can be implemented by computing the predicted values ​​of multiple grid nodes in parallel or by imposing an attenuation upper bound constraint on regions far from the observation point, thereby generating a structured spatial attenuation field representation. Generating the spatial distribution matrix of the attenuation coefficients can be achieved by organizing the attenuation coefficients of each grid node into a two-dimensional array in row and column order. In an exemplary embodiment, this operation can be implemented by storing the prediction uncertainty as a floating-point matrix in shared memory or by adding a confidence matrix, thereby forming an interpolable and visualized intermediate data structure.

[0163] The spatial distribution matrix of the attenuation coefficient is subjected to bilinear interpolation to output a communication link risk heatmap with a spatial resolution of meters.

[0164] Bilinear interpolation can be a mathematical method for smoothing upsampling of a continuous field on a two-dimensional grid by weighted averaging of four neighboring points. This can improve the visual continuity and spatial resolution of heatmaps and eliminate the grid-based staircase effect. For example, bilinear interpolation can be implemented by performing fast interpolation in GPU texture units or by using lookup tables to accelerate weight calculation. Further, bilinear interpolation of the attenuation coefficient spatial distribution matrix can be performed by inserting new points on the original grid and calculating their values ​​through a four-neighbor weighted average. In a specific embodiment, this operation can be achieved by interpolating to a 0.5m resolution to meet meter-level requirements or by combining road boundary masks to avoid interpolating invalid areas, thereby generating a visually continuous, jagged-free, high-resolution risk field. Outputting a communication link risk heatmap with meter-level spatial resolution can be achieved by mapping the interpolated attenuation coefficients to colors or risk levels and rendering them. For example, this operation can be achieved by outputting a PNG format heatmap for the visualization module to call or by carrying risk values ​​in GeoJSON format for the edge decision engine to query, thereby providing a high-precision spatial risk view that can be used for decision-making.

[0165] In one embodiment, the log encryption unit integrates a key derivation mechanism based on the vehicle dynamic identity during the AES-256 encryption process, and generates a unique key seed by reading the real-time identity certificate output by the vehicle security chip and the high-precision positioning timestamp.

[0166] The vehicle dynamic identity identifier can be a spatiotemporally variable identity credential generated by the onboard security chip based on the vehicle's unique hardware characteristics and real-time status. It can serve as a high-entropy input source for key derivation, preventing the static identity from being impersonated or replayed. The onboard security chip can be a hardware root of trust module compliant with national or international security standards, used to securely store keys and perform encryption operations, providing an environment for generating unclonable identity certificates and secure keys. Furthermore, the onboard security chip can protect sensitive data through hardware isolation and tamper-proof design, outputting a signed real-time identity certificate. The real-time identity certificate can be a digital certificate issued by the onboard security chip at the current moment, containing the vehicle's public key and identity statement, and can be used as a trusted identity basis for key seed generation. The high-precision positioning timestamp can be a nanosecond-level time stamp output by the high-precision positioning module, synchronized with geographic coordinates, which can introduce time dimension uniqueness and enhance the unpredictability of the key seed.

[0167] Based on the key seed, a lightweight key update protocol is used to periodically refresh the session key, and a cyclic redundancy check code and a message authentication code are embedded in the encrypted data packet;

[0168] The lightweight key update protocol can be a periodic refresh mechanism for session keys suitable for resource-constrained in-vehicle environments. It can be used to achieve forward security and limit the impact of a single key leak. The session key can be a symmetric key used for single or short-term log encryption, derived from a unique key seed. It can be used to actually perform AES-256 encryption operations and has a controlled lifecycle. The encrypted data packet can be a complete transmission unit containing encrypted log content and secure additional fields. It can be used to carry trusted audit information and support end-to-end security verification. The Cyclic Redundancy Check (CRC) code can be a non-encrypted checksum used to detect bit errors during data transmission. It can provide basic transmission integrity guarantees and assist FEC error correction. The Message Authentication Code (MAC) can be a short tag generated based on the key. It is used to verify the authenticity of the message source and the integrity of the content. It can be used to prevent log tampering or forgery and achieve non-repudiation.

[0169] When uploading through a low duty cycle transmission channel, the forward error correction coding rate is dynamically adjusted based on the link quality feedback value. When a sudden increase in the channel bit error rate is detected, it automatically switches to a high redundancy transmission mode.

[0170] The forward error correction coding rate, which is the ratio of redundant parity bits to information bits in FEC coding, determines the error correction capability and effective throughput. It can be used to improve transmission reliability by increasing redundancy when the channel deteriorates. A high-redundancy transmission mode can be a transmission configuration that uses a high proportion of FEC coding, sacrificing rate for strong error correction capability. It can be used to ensure that critical logs can still be reliably delivered under sudden interference.

[0171] Furthermore, to achieve the above objectives, the present invention also provides a real-time interference avoidance device for an in-vehicle wireless communication module based on edge computing. The device includes: a memory, a processor, and an edge computing-based real-time interference avoidance program for an in-vehicle wireless communication module stored in the memory and executable on the processor. The edge computing-based real-time interference avoidance program for an in-vehicle wireless communication module is configured to implement the steps of the edge computing-based real-time interference avoidance method for an in-vehicle wireless communication module as described above.

[0172] Furthermore, to achieve the above objectives, the present invention also provides a medium storing a real-time interference avoidance program for an edge computing-based vehicle wireless communication module. When the edge computing-based vehicle wireless communication module real-time interference avoidance program is executed by a processor, it implements the steps of the real-time interference avoidance method for an edge computing-based vehicle wireless communication module as described above.

[0173] Other embodiments or specific implementations of the real-time interference avoidance device for vehicle-mounted wireless communication modules based on edge computing described in this invention can be referred to the above-described method embodiments, and will not be repeated here.

[0174] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A real-time interference avoidance method for an onboard wireless communication module based on edge computing, characterized in that, The method includes: Throughout the entire working cycle of the vehicle-mounted wireless communication module, edge spectrum sensing is performed on the dedicated frequency band for vehicle networking and the frequency band for cellular communication. The burst pulse signals generated by non-cooperative interference sources are captured in real time, and the time-domain duty cycle characteristics and spectrum mask boundaries of the interference signals are identified to form a dynamic interference feature set. The dynamic interference feature set is input into the vehicle-mounted edge computing node. The interference source type is classified in multiple levels through a lightweight convolutional neural network model. The interference intensity confidence is calculated based on the signal entropy value to generate edge interference situation assessment data. Based on edge interference situation assessment data, the system integrates real-time vehicle location coordinates, road topology semantic map and communication status of surrounding nodes to construct a communication link risk heat map, which drives the edge decision engine to generate a dynamic spectrum resource reallocation strategy and outputs a channel switching instruction sequence. The channel switching command sequence is applied to the radio frequency front-end controller of the wireless communication module to perform adaptive adjustment of transmission power and switching of modulation and coding scheme, and simultaneously collect the link quality feedback value sequence after the avoidance operation. The link quality feedback value sequence is compared with the preset communication security threshold in real time. When the continuous feedback value is lower than the threshold, the link redundancy activation mechanism is triggered, the multi-mode communication protocol stack is called to perform seamless switching between primary and backup links, and the communication log encryption upload protocol is started.

2. The real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing as described in claim 1, characterized in that, The process involves inputting a dynamic interference feature set into the vehicle-mounted edge computing node, performing multi-level classification of interference source types using a lightweight convolutional neural network model, and calculating interference intensity confidence based on signal entropy values ​​to generate edge interference situation assessment data, including: The dynamic interference feature set is input into the edge computing unit responsible for spectral feature preprocessing, and time-frequency feature normalization processing is performed to generate a standardized feature vector. The standardized feature vectors are input into a pre-deployed lightweight convolutional neural network model, and the local temporal patterns of the interference signal are extracted through the convolutional layers to generate an interference type probability distribution matrix. Based on the probability distribution matrix of interference types and combined with the spectral disorder index output by the signal entropy calculation module, the interference intensity is weighted and fused with confidence to generate marginal interference situation assessment data that includes interference type labels, intensity confidence, and duration prediction.

3. The real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing as described in claim 2, characterized in that, The data, based on edge interference situation assessment, integrates real-time vehicle location coordinates, road topology semantic graphs, and communication status of surrounding nodes to construct a communication link risk heatmap. This heatmap drives the edge decision engine to generate a dynamic spectrum resource reallocation strategy and outputs a channel switching instruction sequence, including: It receives interference situation assessment data and simultaneously acquires the real-time vehicle position coordinates output by the high-precision positioning module and the road topology semantic map provided by the vehicle map engine. The road topology semantic graph is parsed into node-edge structured data, and combined with the status information broadcast by surrounding vehicle communication nodes, a road network communication risk assessment model is constructed. Based on the road network communication risk assessment model, the channel availability index and interference propagation attenuation coefficient of each communication link are calculated to generate a communication link risk heat map with spatial resolution up to the meter level. Based on the communication link risk heatmap, the optimal spectrum resource reallocation path is selected through the heuristic search algorithm of the edge decision engine, and a channel handover command sequence containing the target frequency band, power adjustment amount and handover timing is output.

4. The real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing as described in claim 3, characterized in that, The process of applying the channel switching command sequence to the radio frequency front-end controller of the wireless communication module to perform adaptive transmission power adjustment and modulation and coding scheme switching, and simultaneously acquiring the link quality feedback value sequence after the avoidance operation, includes: The channel switching command sequence is transmitted to the digital signal processing unit of the RF front-end controller to parse the target frequency band parameters and configure the phase-locked loop frequency synthesizer. Based on power regulation commands, the bias voltage of the power amplifier is dynamically adjusted to achieve closed-loop adaptive regulation of the transmitted power. Based on the modulation and coding scheme switching instruction, the modulation order and coding rate of the physical layer protocol stack are updated in real time to generate an optimized communication waveform; During waveform transmission, the signal-to-noise ratio, bit error rate, and throughput parameters are synchronously collected through the channel state information feedback unit of the receiving link to form a link quality feedback value sequence.

5. The real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing as described in claim 4, characterized in that, The step involves edge-comparing the link quality feedback value sequence with a preset communication security threshold in real time. When consecutive feedback values ​​fall below the threshold, a link redundancy activation mechanism is triggered, invoking the multi-mode communication protocol stack to perform seamless primary / backup link switching, and initiating an encrypted communication log upload protocol. This includes: A real-time comparison engine is deployed within the vehicle-mounted edge computing node to perform sliding window analysis on the link quality feedback value sequence and preset multi-level communication security thresholds. When the continuous feedback value is detected to be lower than the first-level safety threshold, a link degradation warning signal is generated; when the continuous feedback value is lower than the second-level safety threshold, the link redundancy activation mechanism is triggered. Based on the link redundancy activation mechanism, the switching control module of the multi-mode communication protocol stack is called to migrate the traffic of the main communication link to the pre-negotiated backup link and execute the handshake protocol to verify the integrity of the switching. After the switch is completed, the log encryption unit is activated to encrypt the interference event data using AES-256 and upload it to the roadside edge server through a low duty cycle transmission channel.

6. The real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing as described in claim 5, characterized in that, The road network communication risk assessment model calculates the channel availability index and interference propagation attenuation coefficient for each communication link, generating a communication link risk heatmap with a spatial resolution of meters, including: Extract lane curvature, obstacle density, and signal masking area parameters from the road topology semantic map to construct a spatial propagation loss feature vector; By combining the channel occupancy rate in the communication status of surrounding nodes with historical interference records, the probability value of co-channel interference between links is calculated; The spatial propagation loss feature vector and the co-channel interference probability value are input into a dynamic weight fusion unit to generate a channel availability index. Based on the channel availability index and the intensity confidence in the interference situation assessment data, the interference propagation attenuation coefficient is derived through a Gaussian process regression model, generating a communication link risk heatmap with a spatial resolution of meters.

7. The real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing as described in claim 6, characterized in that, The process of deriving the interference propagation attenuation coefficient using a Gaussian process regression model and generating a communication link risk heatmap with a spatial resolution of meters includes: Using the channel availability index as an input variable, the covariance function kernel of the Gaussian process regression model is constructed; Based on historical interference propagation sample data, the hyperparameters of the kernel function are optimized to make the model output match the measured attenuation curve. In the gridded coordinate system of the road topology semantic graph, the interference propagation attenuation coefficient is calculated for each grid node to generate the spatial distribution matrix of the attenuation coefficient; The spatial distribution matrix of the attenuation coefficient is subjected to bilinear interpolation to output a communication link risk heatmap with a spatial resolution of meters.

8. The real-time interference avoidance method for vehicle-mounted wireless communication modules based on edge computing as described in claim 5, characterized in that, During the AES-256 encryption process, the log encryption unit integrates a key derivation mechanism based on the vehicle's dynamic identity identifier, which generates a unique key seed by reading the real-time identity certificate output by the vehicle security chip and the high-precision positioning timestamp. Based on the key seed, a lightweight key update protocol is used to periodically refresh the session key, and a cyclic redundancy check code and a message authentication code are embedded in the encrypted data packet; When uploading through a low duty cycle transmission channel, the forward error correction coding rate is dynamically adjusted based on the link quality feedback value. When a sudden increase in the channel bit error rate is detected, it automatically switches to a high redundancy transmission mode.

9. A real-time interference avoidance device for an onboard wireless communication module based on edge computing, characterized in that, The device includes: a memory, a processor, and an edge computing-based real-time interference avoidance program for an in-vehicle wireless communication module stored in the memory and executable on the processor, wherein the edge computing-based real-time interference avoidance program for an in-vehicle wireless communication module is configured to implement the steps of the edge computing-based real-time interference avoidance method for an in-vehicle wireless communication module as described in any one of claims 1 to 8.

10. A medium, characterized in that, The medium stores a real-time interference avoidance program for an vehicular wireless communication module based on edge computing. When the processor executes the real-time interference avoidance program for an vehicular wireless communication module based on edge computing, it implements the steps of the real-time interference avoidance method for an vehicular wireless communication module based on edge computing as described in any one of claims 1 to 8.

Images