A 94ghz radar baseband signal solving method and system

Abstract

The present application belongs to the technical field of radar signal processing, and provides a 94Ghz radar baseband signal solving method and system, which comprises clock signal synchronization, signal receiving and preprocessing, analog-to-digital conversion, multi-stage parallel signal processing, parameter dynamic adjustment, multi-target solving and resource dynamic allocation. The present application introduces a FPGA multi-stage parallel processing architecture, improves the multi-target detection and tracking capability of the system in complex scenes, realizes dynamic optimization of signal processing parameters according to real-time environment through a deep reinforcement learning algorithm, ensures the accurate detection capability of the system in different scenes, enhances the real-time tracking and solving capability of the system for multi-target through a graph neural network multi-target cooperative processing algorithm, and realizes dynamic resource allocation through an energy perception algorithm, thereby greatly reducing the power consumption of the system while ensuring high-speed processing.

Description

A method and system for calculating 94GHz radar baseband signals Technical Field

[0001] This invention relates to the field of radar signal processing technology, and in particular to a method and system for calculating 94GHz radar baseband signals. Background Technology

[0002] With the widespread application of millimeter-wave radar technology in fields such as intelligent transportation and autonomous driving, the demand for radar signal processing in complex environments is increasing. Millimeter-wave radar systems face several technical challenges when processing signals in multi-target tracking and dynamic environments.

[0003] In complex scenarios, radar signals are easily reflected and scattered by environmental obstacles, resulting in multipath effects. These interferences lead to signal attenuation, affecting target detection accuracy and reducing system stability. Simultaneously, noise interference also increases system errors. Existing radar systems, due to limited computing power, often struggle to process multiple targets simultaneously in multi-target detection and high-speed moving target tracking. Sequential processing architectures limit the real-time performance of systems in complex scenarios, especially in applications requiring rapid response, such as autonomous driving, where existing technologies exhibit lag and are difficult to handle efficiently. Furthermore, existing millimeter-wave radar systems lack adaptive capabilities, typically relying on fixed signal processing flows and unable to dynamically adjust to environmental changes. This fixed processing method limits the system's flexibility in different scenarios and cannot meet the diverse needs of complex environments. In summary, existing technologies have shortcomings in signal attenuation, real-time multi-target detection, high-speed target tracking, and adaptive capabilities. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the present invention aims to provide a 94GHz radar baseband signal processing method and system. By introducing an FPGA multi-level parallel processing architecture, the system's multi-target detection and tracking capabilities in complex scenarios are improved. Deep reinforcement learning algorithms are used to dynamically optimize signal processing parameters based on the real-time environment, ensuring accurate detection capabilities in different scenarios. A graph neural network multi-target collaborative processing algorithm enhances the system's real-time tracking and processing capabilities for multiple targets. An energy sensing algorithm enables dynamic resource allocation, significantly reducing system power consumption while ensuring high-speed processing.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for calculating 94GHz radar baseband signals includes:

[0007] The system clock and synchronization module provides synchronous clock signals for the RF front-end module, analog-to-digital converter, FPGA processing module, deep learning optimization module, target detection and solving module, and data transmission and control module.

[0008] The radio frequency front-end module is used to receive a 94 GHz millimeter wave signal, and the 94 GHz millimeter wave signal is mixed, amplified, and denoised to obtain an intermediate frequency signal.

[0009] The intermediate frequency signal is converted into a digital signal using the analog-to-digital converter.

[0010] The FPGA processing module is used to perform noise filtering, normalization and multi-level parallel signal processing on the digital signal to obtain a multi-target difference comparison heatmap.

[0011] The deep reinforcement learning algorithm of the deep learning optimization module is used to dynamically adjust the FFT window, filter bandwidth, and gain control parameters of the FPGA processing module.

[0012] The multi-target collaborative processing algorithm based on graph neural networks in the target detection and solution module is used to detect and solve the distance, velocity, and angle of multiple targets according to the multi-target difference comparison heatmap, so as to obtain a multi-target solution map; the multi-target solution map includes: distance difference, velocity difference, and angle difference;

[0013] The computational resources of the FPGA processing module, the deep learning optimization module, and the target detection and solving module are dynamically allocated using the importance function and resource allocation optimization function of the energy sensing algorithm in the data transmission and control module, and the multi-target solving graph is uploaded to the target host computer.

[0014] Preferably, the construction process of the FPGA processing module includes:

[0015] The Doppler frequency shift features, echo intensity features, and noise features of the pre-acquired intermediate frequency signal are extracted using the state-action value function of the deep reinforcement learning algorithm to obtain a feature dataset; the calculation formula of the state-action value function is: ;in, Let r be the value of taking action a in state s; r be the immediate reward; and γ be the discount factor. This represents the state after action 'a' is performed; Representing state The following action; Representing state Choose the action with the highest value. Expected value;

[0016] The feature dataset is dimensionality reduced to obtain the training dataset;

[0017] Based on the Fast Fourier Transform and multi-objective collaborative processing algorithm, the FPGA processing module is obtained by iterative processing using a preset loss function on the training dataset; the calculation formula of the loss function is: Where L represents the total loss; and These are the true distance and predicted distance of the i-th target, respectively; and λ represents the actual velocity and predicted velocity of the j-th target, respectively; λ is the weighting factor. The number of targets used for distance calculation; The number of targets used for velocity calculation.

[0018] Preferably, the noise filtering process employs an autocorrelation function and a low-pass filter; the formula for calculating the autocorrelation function is: The calculation formula for the low-pass filter is as follows: ;

[0019] in, The autocorrelation function is mentioned above. Here, t is the low-pass filter; t is the integration variable; The target filtered signal; It is a time constant; The signal frequency; This is the filter cutoff frequency.

[0020] Preferably, the calculation function for the normalization process is: ;in, For normalized signals; For target normalization signal; The mean of the signal; This represents the standard deviation of the signal.

[0021] Preferably, the calculation formula for the multi-objective solution graph includes:

[0022] , as well as ;

[0023] in, It represents the distance difference; For speed difference; For the angle difference; The speed of light; The signal scan period; This is the frequency difference; Λ represents the radar signal bandwidth; Λ represents the radar wavelength. For Doppler frequency shift; Phase difference; This refers to the spacing between antenna elements.

[0024] Preferably, the importance function is: The resource allocation optimization function is:

[0025] ;

[0026] in, The importance weight of the i-th objective; This is the optimal resource allocation scheme; and These are the distance and velocity characteristics of the i-th target, respectively; and α and β are the distance and velocity characteristics of the j-th target, respectively; α and β are both weighting factors. Let be the utility function for the i-th objective; This represents finding the optimal solution for the distance difference.

[0027] Preferably, a 94GHz radar target detection and resolution module includes:

[0028] The receiving unit is used to receive a 94GHz millimeter wave signal and to perform frequency mixing, amplification, and noise reduction on the 94GHz millimeter wave signal to obtain an intermediate frequency signal.

[0029] A conversion unit is used to convert the intermediate frequency signal into the digital signal;

[0030] The parallel processing unit is used to perform noise filtering, normalization and multi-level parallel signal processing on the digital signal to obtain a multi-target difference comparison heatmap.

[0031] The calculation unit is used to detect and calculate the distance, velocity and angle of multiple targets based on the multi-target difference comparison heat map to obtain a multi-target calculation map;

[0032] An optimization unit is used to dynamically adjust the FFT window, filter bandwidth, and gain control parameters of the parallel processing unit.

[0033] The control unit is used to dynamically adjust the computing resources of the parallel processing unit, the solution unit, and the optimization unit using the importance function and the resource allocation optimization function, and upload the multi-objective solution graph to the target host computer.

[0034] The synchronization unit is used to provide a synchronization clock signal to the receiving unit, the conversion unit, the parallel processing unit, the calculation unit, the optimization unit, and the control unit.

[0035] The present invention discloses the following technical effects:

[0036] This invention provides a 94GHz radar baseband signal processing method and system. By introducing an FPGA multi-level parallel processing architecture, it solves the shortcomings of poor multi-target detection performance in existing technologies, realizing multi-target detection and tracking functions in complex scenarios. Through deep reinforcement learning algorithms, it addresses the problem of poor generalization and reliability caused by fixed parameters of the FPGA processing module, enabling dynamic optimization of signal processing parameters based on the real-time environment. Through graph neural network multi-target collaborative processing algorithms, it solves the problem of poor processing capability of conventional technologies, realizing real-time tracking and processing of multiple targets. Through energy sensing algorithms, it solves the problems of low efficiency and high power consumption caused by fixed computing power allocation in conventional technologies, realizing dynamic resource allocation. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 is a schematic diagram of the 94GHz radar baseband signal processing flow provided in an embodiment of the present invention;

[0039] Figure 2 is a schematic diagram of the overall architecture of the 94GHz millimeter-wave radar baseband signal processing system based on FPGA technology provided in an embodiment of the present invention.

[0040] Figure 3 is a heatmap showing the multi-target difference comparison provided in an embodiment of the present invention;

[0041] Figure 4 is a schematic diagram of the parallel signal processing flow provided in an embodiment of the present invention;

[0042] Figure 5 is a flowchart of the signal processing optimization provided in an embodiment of the present invention;

[0043] Figure 6 is a flowchart of multi-target collaborative processing provided in an embodiment of the present invention;

[0044] Figure 7 is a radar chart of target correlation analysis provided by an embodiment of the present invention;

[0045] Figure 8 is a schematic diagram of dynamic resource allocation provided in an embodiment of the present invention;

[0046] Figure 9 is a target importance ranking diagram provided by an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The purpose of this invention is to provide a 94GHz radar baseband signal processing method and system. By introducing an FPGA multi-level parallel processing architecture, the system's multi-target detection and tracking capabilities in complex scenarios are improved. Deep reinforcement learning algorithms are used to dynamically optimize signal processing parameters based on the real-time environment, ensuring accurate detection capabilities in different scenarios. A graph neural network multi-target collaborative processing algorithm enhances the system's real-time tracking and processing capabilities for multiple targets. An energy sensing algorithm is used to achieve dynamic resource allocation, significantly reducing system power consumption while ensuring high-speed processing.

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Figure 1 is a schematic diagram of the 94GHz radar baseband signal processing flow provided in an embodiment of the present invention. As shown in Figure 1, the present invention provides a 94GHz radar baseband signal processing method, including:

[0051] Step 100: Use the system clock and synchronization module to provide a synchronization clock signal for the RF front-end module, analog-to-digital converter, FPGA processing module, deep learning optimization module, target detection and solving module, and data transmission and control module;

[0052] Step 200: Receive a 94GHz millimeter-wave signal using the radio frequency front-end module, and perform frequency mixing, amplification, and noise reduction on the 94GHz millimeter-wave signal to obtain an intermediate frequency signal;

[0053] Step 300: Use the analog-to-digital converter to convert the intermediate frequency signal into a digital signal;

[0054] Step 400: The FPGA processing module is used to perform noise filtering, normalization and multi-level parallel signal processing on the digital signal to obtain a multi-target difference comparison heatmap.

[0055] Step 500: Dynamically adjust the FFT window, filter bandwidth, and gain control parameters of the FPGA processing module using the deep reinforcement learning algorithm of the deep learning optimization module;

[0056] Step 600: Using the multi-target collaborative processing algorithm based on graph neural network in the target detection and solution module, the distance, velocity, and angle of the multiple targets are detected and solved according to the multi-target difference comparison heatmap to obtain a multi-target solution map; the multi-target solution map includes: distance difference, velocity difference, and angle difference;

[0057] Step 700: Dynamically allocate computing resources for the FPGA processing module, the deep learning optimization module, and the target detection and solving module using the importance function and resource allocation optimization function of the energy sensing algorithm in the data transmission and control module, and upload the multi-target solving graph to the target host computer (uploaded in UDP packet form).

[0058] Specifically, the construction process of the FPGA processing module includes:

[0059] The Doppler frequency shift features, echo intensity features, and noise features of the pre-acquired intermediate frequency signal are extracted using the state-action value function of the deep reinforcement learning algorithm to obtain a feature dataset; the calculation formula of the state-action value function is: ;in, Let r be the value of taking action a in state s; r be the immediate reward; and γ be the discount factor. This represents the state after action 'a' is performed; Representing state The following action; Representing state Choose the action with the highest value. Expected value;

[0060] The feature dataset is dimensionality reduced to obtain the training dataset;

[0061] Based on the Fast Fourier Transform and multi-objective collaborative processing algorithm, the FPGA processing module is obtained by iterative processing using a preset loss function on the training dataset; the calculation formula of the loss function is: Where L represents the total loss; and These are the true distance and predicted distance of the i-th target, respectively; and λ represents the actual velocity and predicted velocity of the j-th target, respectively; λ is the weighting factor. The number of targets used for distance calculation; The number of targets used for velocity calculation.

[0062] Specifically, the state-action value function Q(s,a): Q(s,a) represents the expected value of taking action a in the current state s. It measures the value of taking an action through accumulated rewards to optimize the signal processing strategy. Immediate reward r: r is the direct reward obtained after taking an action in the current state, reflecting the immediate effect of the action and guiding the learning direction of the deep reinforcement learning model. Discount factor γ: γ is a value between 0 and 1 used to discount future rewards. It determines the model's emphasis on long-term gains. A larger γ value indicates that the model pays more attention to future accumulated gains, suitable for complex signal processing scenarios requiring long-term adjustments. Future state s′: s′ represents the state after the current action a. The model optimizes the current policy by predicting subsequent states. Future action a′: a′ is the possible action to be taken in the future state s′. The model optimizes the decision of the current action by predicting the value of future actions. Maximizing maxQ(s′,a′): This term represents the expected value of choosing the action a′ with the highest value in the future state s′, used to update the action value of the current state. It guides the model to choose the optimal action at each step to increase cumulative rewards.

[0063] Furthermore, the total loss L: The total loss L is the target value for model training; minimizing this value improves the model's prediction accuracy. It combines the errors of both distance and velocity metrics to achieve the high accuracy requirements for multi-objective solutions. The true distance R... i and predicted distance It is the true distance to the i-th target. This is the distance to the i-th target predicted by the model. The squared difference between the two is... Used to calculate the error in distance prediction. Actual velocity V j and prediction speed V j It is the actual speed of the j-th target. This is the velocity of the j-th target predicted by the model. The squared difference between the two is... The distance and velocity error is used to calculate the velocity prediction error. The weighting factor λ is a weighting factor used to adjust the proportion of distance and velocity error terms in the total loss. By adjusting λ, the model's focus on distance and velocity prediction errors can be changed to adapt to different detection accuracy requirements. The number of targets N and M: N represents the number of targets to be detected (related to distance calculation), and M represents the number of targets to be detected (related to velocity calculation). They are used to determine the summation range of each type of error in the loss function.

[0064] Preferably, the noise filtering process employs an autocorrelation function and a low-pass filter; the formula for calculating the autocorrelation function is: The calculation formula for the low-pass filter is as follows: ;

[0065] in, The autocorrelation function is mentioned above. Here, t is the low-pass filter; t is the integration variable; The target filtered signal; It is a time constant; The signal frequency; This is the filter cutoff frequency. Adjustment is made... It can filter out high-frequency noise and obtain a purer echo signal.

[0066] Specifically, the calculation function for the normalization process is: ;in, For normalized signals; For target normalization signal; The mean of the signal; This represents the standard deviation of the signal.

[0067] Preferably, the calculation formula for the multi-objective solution graph includes:

[0068] , as well as ;

[0069] in, It represents the distance difference; For speed difference; For the angle difference; The speed of light; The signal scan period; This is the frequency difference, which is the frequency difference between the radar transmitted signal and the echo signal; This refers to the radar signal bandwidth. For radar wavelength, defined as ,in The center frequency of the radar transmission (for the 94GHz band, approximately...) ); The Doppler frequency shift is obtained by measuring the frequency offset of the echo signal. Phase difference; The distance between antenna elements is denoted by . The range difference formula calculates the target's distance by solving for the frequency difference between the transmitted and echo signals. This algorithm eliminates velocity ambiguity caused by the Doppler effect, ensuring accurate target velocity calculation by the radar system in high-speed moving scenarios. The angle difference formula calculates the relative angle between the target and the radar by measuring the phase difference of the echo signals received by adjacent antenna elements in the phased array antenna array, ensuring high-precision target azimuth calculation.

[0070] Specifically, the importance function is: The resource allocation optimization function is:

[0071] ;

[0072] in, The importance weight of the i-th objective; This is the optimal resource allocation scheme; and These are the distance and velocity characteristics of the i-th target, respectively; and These are the distance and velocity features of the j-th target, respectively; α and β are weighting factors used to adjust the influence of different features on importance. Let be the utility function for the i-th objective; This represents finding the optimal solution for the distance difference.

[0073] Furthermore, a 94GHz radar target detection and resolution module includes:

[0074] The receiving unit is used to receive a 94GHz millimeter wave signal and to perform frequency mixing, amplification, and noise reduction on the 94GHz millimeter wave signal to obtain an intermediate frequency signal.

[0075] A conversion unit is used to convert the intermediate frequency signal into the digital signal;

[0076] The parallel processing unit is used to perform noise filtering, normalization and multi-level parallel signal processing on the digital signal to obtain a multi-target difference comparison heatmap.

[0077] The calculation unit is used to detect and calculate the distance, velocity and angle of multiple targets based on the multi-target difference comparison heat map to obtain a multi-target calculation map;

[0078] An optimization unit is used to dynamically adjust the FFT window, filter bandwidth, and gain control parameters of the parallel processing unit.

[0079] The control unit is used to dynamically adjust the computing resources of the parallel processing unit, the solution unit, and the optimization unit using the importance function and the resource allocation optimization function, and upload the multi-objective solution graph to the target host computer.

[0080] The synchronization unit is used to provide a synchronization clock signal to the receiving unit, the conversion unit, the parallel processing unit, the calculation unit, the optimization unit, and the control unit.

[0081] Preferably, the method for generating a multi-objective solution graph includes:

[0082] Multiple target feature parameters are processed in pairs to obtain interactive information such as distance, velocity difference, and angle change between two targets.

[0083] Using the aforementioned multi-objective solution model, the mutual influence between the various objectives is calculated;

[0084] Based on the calculation results, a multi-objective solution graph is generated to optimize target tracking and signal processing strategies in real time.

[0085] Specifically, the Fast Fourier Transform (FFT) formula used in the signal processing is as follows:

[0086]

[0087] in: This represents the frequency spectrum of the echo signal in the frequency domain. This formula converts the time-domain signal into a frequency-domain signal, which is then used to further calculate the target's distance, velocity, and angle.

[0088] Optionally, the autocorrelation function is typically used for time-domain preprocessing of signals. It analyzes the correlation between the signal and noise and performs preliminary noise reduction in the early stages of signal processing. The low-pass filter (LPF), on the other hand, is primarily used in the frequency-domain processing stage, especially when it's necessary to filter out noise at specific frequencies. It's a method of precisely processing the signal by adjusting the frequency. In complex signal processing, noise sources can be very diverse. To obtain a clearer signal, multiple techniques are usually combined to gradually reduce noise interference. The autocorrelation function captures noise characteristics and performs preliminary removal in the time domain, while the low-pass filter further refines and removes high-frequency noise in the frequency domain. This combination ensures that the system optimizes signal quality at different levels.

[0089] Referring to Figure 2, the specific functions of each module are as follows:

[0090] RF Front-End Module: This module receives 94GHz millimeter-wave signals. Through mixing, amplification, and noise reduction, it converts the signals into intermediate frequency (IF) signals. This step provides a stable signal input source for subsequent digital processing in the system.

[0091] Analog-to-digital converter (ADC): Converts analog intermediate frequency (IF) signals into digital signals. The ADC ensures the digitization of analog signals from the RF front-end module, providing the foundation for subsequent precise calculations by the FPGA processing module. The ADC features a high sampling rate and low power consumption, supporting real-time data acquisition and processing.

[0092] FPGA processing module: This is the core processing unit of the system, responsible for performing multi-level parallel signal processing.

[0093] Target Detection and Resolution Module: This module executes target detection tasks using data received from the FPGA processing module. It performs real-time calculations of the distance, velocity, and angle of multiple targets, analyzing their trajectories. Referring to Figure 3, the target detection and resolution module can generate a multi-target difference comparison heatmap based on the differences in distance, velocity, and angle of different targets. This figure illustrates the differences between multiple targets across different parameters, providing an intuitive comparative reference for the system's multi-target resolution and aiding in identifying the differences between targets. The module embeds a real-time multi-target resolution process, enabling the analysis and processing of the mutual influence of multiple targets.

[0094] Deep Learning Optimization Module: This module dynamically adjusts key parameters such as FFT window, filter bandwidth, and gain control through the Deep Reinforcement Learning (DRL) algorithm.

[0095] Data Transmission and Control Module: This module manages data transmission and communication between modules throughout the system, ensuring the normal operation and efficient collaboration of all modules. It employs a high-speed data bus and low-latency communication protocols to guarantee the real-time performance and efficiency of the radar signal processing flow. The Data Transmission and Control Module also handles interface connections with external systems, enabling the integration and sharing of data from various sensors.

[0096] System Clock and Synchronization Module: This module provides a synchronized clock signal for the entire system, ensuring that all modules operate in coordination under a unified time base. Through a high-precision clock source and clock synchronization algorithm, this module achieves efficient collaboration across the entire radar system. It also supports time synchronization across multiple radar devices, making it suitable for multi-radar collaborative scenarios.

[0097] Referring to Figure 3, the target comparison heatmap shows the differences between multiple targets in terms of three characteristics: distance difference, velocity difference, and angle difference. The values ​​and color intensity in the heatmap reflect the magnitude of the differences between the targets; darker colors indicate greater differences, and lighter colors indicate smaller differences. The following is a detailed explanation of the content in the figure: Targets A, B, C, and D: Represent different targets detected by the system. In radar detection, the differences between these targets are quantified using the characteristic values ​​of distance, velocity, and angle. Distance difference, velocity difference, and angle difference: Represent the differences in distance, velocity, and angle between different targets, respectively. These features are used to assess and compare the relative positions and motion states of each target. Target pairs such as AB and AC: Each cell in the heatmap represents the difference between two targets in a certain characteristic. Specifically: Distance difference row: Represents the distance difference between each target pair. For example, the distance difference between target A and target B is 4, and the distance difference between target A and target C is 12. Velocity difference row: Represents the velocity difference between each target pair. For example, the velocity difference between target A and target B is 4, and the velocity difference between target A and target C is 2. Angle Difference Row: Represents the angle difference between each target pair. For example, the angle difference between target A and target C is 25°, and the angle difference between target A and target D is 15°. Color Explanation: The darker the color, the greater the difference and the larger the value; the lighter the color, the smaller the difference. The darker the cell, the more significant the difference in that feature between the corresponding target pair. For example, the darkest color, indicating a large angle difference, is 25° between target A and target C. Overall Analysis: Target A and C have the largest difference in angle difference (25°), followed by a large difference in range (12°), indicating that these two targets have significant differences in relative azimuth and position. The velocity differences between target pairs are generally small, with low values, indicating that their relative motion states are relatively similar. This graphical method facilitates quick observation of the feature differences between different targets and can help the radar system prioritize target pairs with significant differences during target calculation, thereby optimizing resource allocation and processing strategies.

[0098] Referring to Figure 4, the parallel signal processing flow in the FPGA processing module specifically includes the following steps:

[0099] S1: IF signal reception and preprocessing. The RF front-end module receives the 94GHz millimeter-wave signal and generates an IF signal. The analog signal is converted to a digital signal via an analog-to-digital converter (ADC) and then processed in parallel by the FPGA processing module.

[0100] S2: Fast Fourier Transform (FFT). The FPGA processing module first performs a Fast Fourier Transform (FFT) on the input digital signal, converting the time-domain signal into a frequency-domain signal. The size of the FFT window is dynamically adjusted by the deep learning optimization module to adapt to the signal characteristics and noise levels under different environments, ensuring the accuracy and efficiency of signal processing.

[0101] S3: Filtering and Gain Control. The signal after FFT conversion is filtered by a bandpass filter to remove noise and unwanted frequency bands. The system further enhances the signal strength through gain control to ensure the clarity of the target signal and the accuracy of the solution.

[0102] S4: Target Detection and Parameter Solving. Further, the filtered signal is transmitted to the target detection and solving module to extract parameters such as target distance, velocity, and angle. Through a multi-target collaborative processing algorithm based on graph neural networks (GNNs), the system can analyze the correlation information of multiple targets, generate a multi-target solution graph, and achieve real-time target tracking and prediction.

[0103] Furthermore, beamforming: Beamforming is a signal processing method based on phased array technology, primarily applied in the angular channel. By adjusting the signal phase and amplitude of each antenna element, beamforming technology can enhance the signal strength in a specific direction while suppressing interference signals from other directions. In this system, this technology is used to improve the angular resolution of the target, enabling the system to more accurately locate the target's position. The results of beamforming can effectively improve the system's detection performance in multi-target environments. DOA (Direction of Arrival) Algorithm: The DOA algorithm is used to determine the direction of signal arrival. This system uses multi-channel data reception and, based on the signal phase difference of each channel and the beamforming results, utilizes the DOA algorithm to accurately calculate the angle information of the target signal. The core of this algorithm lies in calculating the phase difference received by multiple antennas, thereby inferring the incident angle of the signal. The DOA algorithm, used in conjunction with beamforming, further improves the accuracy of angle measurement. Doppler Analysis: Doppler analysis is used for signal processing in the velocity channel, primarily determining the relative velocity of the target by detecting the Doppler frequency shift. Doppler shift is a frequency shift caused by the relative motion of the target. The system extracts the frequency shift information through time-domain and frequency-domain processing techniques and calculates the target's velocity. The results of Doppler analysis are further processed by a velocity threshold filter to filter out target signals that meet the velocity detection requirements. Array calibration: Array calibration is used to ensure the accuracy of the phased array antenna. This process involves correcting the phase and amplitude differences of each element in the antenna array to ensure consistency in signal processing. Array calibration is a fundamental step in ensuring the accuracy of beamforming and the DOA algorithm, maintaining the reliability of signal processing even in the presence of hardware deviations or environmental interference. Kalman filtering: Kalman filtering is used in the target aggregation module to filter and track the fused signal from multiple targets. It is a recursive algorithm that combines historical states and current observations to predict the target's trajectory and filter out noise interference. The application of Kalman filtering ensures the stability of range, velocity, and angle outputs, enabling the system to achieve accurate tracking in complex multi-target scenarios.

[0104] Referring to Figure 5, the process of adaptively optimizing signal processing parameters using the Deep Reinforcement Learning (DRL) algorithm is as follows:

[0105] S1: The system acquires intermediate frequency signals in real time through the FPGA processing module and extracts key signal features, such as Doppler frequency shift and echo intensity. The system can effectively capture real-time dynamic changes in signals, improve the accuracy and response speed of signal processing, and ensure that signal features can be captured in a timely manner even in complex and changing environments.

[0106] S2: Establishing a historical database. The system utilizes a large amount of historical data to train the deep reinforcement learning model. Specifically, the model dynamically adjusts the FFT window, filter bandwidth, and gain control parameters by continuously learning from the historical changes in signal characteristics and external environmental information to optimize the signal processing flow. Simulated signal data for different scenarios are used during this training process to ensure the model has broad adaptability.

[0107] S3: Based on the output of the deep reinforcement learning model, the system adjusts the FFT window size, filter bandwidth, and gain parameters in real time. By updating these parameters in real time, the system can adaptively process signals in different scenarios, ensuring that the optimized signal processing effect reaches its best state. During target detection, the system can perform fine-tuning of parameters for different frequency bands and signal strengths to improve detection accuracy in complex environments.

[0108] S4: The system uses a feedback mechanism to monitor the signal processing effect in real time, returning the processing results to the deep reinforcement learning model for optimization. The model continuously adjusts its processing strategy based on the new feedback, further enhancing its adaptive optimization capabilities. In this way, the system ensures high precision and stability in signal processing, reduces errors, and improves the overall system response speed.

[0109] Referring to Figure 6, the operation process of the multi-objective collaborative processing algorithm based on graph neural networks (GNN) is as follows:

[0110] S1: The system acquires preliminary solution data for multiple targets from the FPGA processing module, including basic information such as target distance, velocity, and angle. This data is analyzed to generate a preliminary multi-target solution map. The system can extract more accurate correlation information based on the spatiotemporal dependencies between different targets. The model learns the dynamic relationships between multiple targets, captures their spatiotemporal dependencies, and predicts their future trajectories.

[0111] S2: The graph neural network analyzes the correlation information of multiple targets, predicts the future trajectories of multiple targets, and generates an optimized multi-target solution graph through model optimization. This optimized graph combines the correlation of multiple targets with historical trajectories, improving the accuracy of target detection and solution.

[0112] S3: Based on the optimized multi-target solution graph, the system achieves real-time tracking and motion prediction of multiple targets. This system ensures high-precision and low-latency tracking of multiple targets in complex environments, making it particularly suitable for multi-target tracking and behavior prediction in complex traffic scenarios. Referring to Figure 7, the system also generates a target correlation analysis graph through data-driven analysis. This graph illustrates the relationships between multiple targets, providing more accurate data support for predicting target trajectories.

[0113] Referring to Figure 8, the dynamic optimization allocation of FPGA resources is achieved through an energy-sensing algorithm, specifically including the following steps:

[0114] S1: Load Monitoring and Signal Acquisition. The system monitors the current load status in real time, detecting the computational pressure and signal processing requirements of the FPGA module. By monitoring the FPGA load, the system accurately assesses the current resource utilization.

[0115] S2: Execution of the energy-sensing algorithm. The system executes the energy-sensing algorithm based on real-time load data to assess the energy consumption and computational requirements of each processing module. Through energy consumption assessment, the system predicts the energy consumption of each processing module under the current load and allocates resources according to actual needs.

[0116] S3: Resource Optimization Allocation. Based on the evaluation results of the energy-aware algorithm, the system optimizes the resource allocation of the FPGA processing module. By reducing the resource consumption of low-priority tasks, the system ensures that the resource requirements of high-priority tasks are fully met, especially ensuring that the system can still efficiently execute important tasks under high load. Referring to Figure 9, an importance ranking chart of the targets is generated based on the system's data processing results. This chart analyzes the importance of multiple targets, shows their priorities, and provides a reference for resource allocation and decision-making.

[0117] S4: Energy Consumption Optimization Feedback. The system continuously adjusts its resource allocation strategy through a feedback mechanism to ensure sustained improvement in energy consumption optimization. By optimizing system energy consumption in real time, the system can significantly reduce energy consumption while meeting computing needs, improving overall energy efficiency. This is suitable for energy-constrained application scenarios and solves the problem of uneven energy distribution at the same time.

[0118] Furthermore, the energy detection module monitors the power, signal strength, and spectrum occupancy of the input signal. This module analyzes the signal characteristics so that the subsequent resource allocation module can appropriately allocate signals of different frequency bands and power levels. The resource allocation module dynamically allocates resources based on the signal characteristics output by the energy detection module. This module combines the power and strength characteristics of the signal to optimize spectrum allocation, ensuring the rational use of system resources. Spectrum allocation is one of the key functions of the resource allocation module. It dynamically allocates different frequency resources based on the spectrum occupancy information obtained from energy detection, ensuring that the bandwidth requirements of the signal are met during transmission and reducing signal interference and spectrum overlap. Power control is used to adjust the power of signal transmission, ensuring a balance between signal coverage and power consumption. Power control can adaptively adjust the transmission power according to the target distance and signal environment, reducing power consumption while maintaining signal transmission quality. Time slot control is another key function in the resource allocation module. Time slot control avoids multiple signals transmitting at the same time by allocating different time intervals (time slots) to each signal, thereby reducing signal interference. This time-slot scheduling mechanism is particularly important in multi-target detection and multi-signal processing environments, ensuring the orderly timing of signal processing and improving the system's real-time performance and stability. The final output signal is optimized in terms of spectrum, power, and time slots. Through these controls and allocations, the system ensures that the signal is transmitted in its optimal state, effectively coping with complex signal environments and achieving stable signal output.

[0119] Preferably, the allocation of resources is handled by the dynamic resource allocation module. Target detection and resolution module: In radar systems, the target detection and resolution module is a computationally intensive component. The dynamic resource allocation module needs to ensure that this module receives sufficient resources under high computational complexity, especially in multi-target detection scenarios where computational pressure is high. FPGA processing module: As the core processing unit, the FPGA can dynamically allocate resources to adjust resource usage at different signal processing stages under high load, such as filtering, gain control, and FFT processing. Deep learning optimization module: This module uses the DRL algorithm to adaptively optimize signal processing parameters. Dynamic resource allocation ensures that the deep learning optimization module can prioritize the allocation of sufficient FPGA computing power when it requires significant computational resources to support real-time optimization. Feedback mechanism: Dynamic resource allocation not only adjusts resources based on target importance ranking but also provides feedback based on the system's real-time load status. By monitoring the resource usage of each module in real time (such as FPGA load), the feedback path feeds this data back to the resource allocation module, optimizing subsequent resource scheduling. The system can dynamically adjust the signal processing flow through the feedback mechanism to cope with different scenario changes.

[0120] The beneficial effects of this invention are as follows:

[0121] This invention improves the system's multi-target detection and tracking capabilities in complex scenarios by introducing a multi-level parallel processing architecture using FPGA; it achieves dynamic optimization of signal processing parameters based on the real-time environment through deep reinforcement learning algorithms, ensuring the system's accurate detection capabilities in different scenarios; it enhances the system's real-time tracking and calculation capabilities for multiple targets through a graph neural network multi-target collaborative processing algorithm; and it achieves dynamic resource allocation through an energy sensing algorithm, significantly reducing the system's power consumption while ensuring high-speed processing.

[0122] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0123] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for calculating 94GHz radar baseband signals, characterized in that, include: The system clock and synchronization module provides a synchronous clock signal for the RF front-end module, analog-to-digital converter, FPGA processing module, deep learning optimization module, target detection and solution module, and data transmission and control module. The RF front-end module receives a 94GHz millimeter-wave signal and performs mixing, amplification, and noise reduction to obtain an intermediate frequency (IF) signal. The IF signal is converted into a digital signal by the analog-to-digital converter. The FPGA processing module performs noise filtering, normalization, and multi-level parallel signal processing on the digital signal to obtain a multi-target difference comparison heatmap. The deep reinforcement learning algorithm of the deep learning optimization module dynamically adjusts the FFT window, filter bandwidth, and gain control parameters of the FPGA processing module. The multi-target collaborative processing algorithm based on graph neural networks in the target detection and solution module detects and solves the distance, velocity, and angle of multiple targets based on the multi-target difference comparison heatmap, obtaining a multi-target solution map. The multi-target solution map includes distance difference, velocity difference, and angle difference. The computational resources of the FPGA processing module, the deep learning optimization module, and the target detection and solving module are dynamically allocated using the importance function and resource allocation optimization function of the energy sensing algorithm in the data transmission and control module, and the multi-target solving graph is uploaded to the target host computer.

2. The method for calculating 94GHz radar baseband signals according to claim 1, characterized in that, The construction process of the FPGA processing module includes: extracting the Doppler frequency shift features, echo intensity features, and noise features of the pre-acquired intermediate frequency signal using the state-action value function of the deep reinforcement learning algorithm to obtain a feature dataset; the calculation formula of the state-action value function is: ;in, Let r be the value of taking action a in state s; r be the immediate reward; and γ be the discount factor. This represents the state after action 'a' is performed; Representing state The following action; Representing state Choose the action with the highest value. The expected value; the feature dataset is dimensionality reduced to obtain a training dataset; based on the Fast Fourier Transform and multi-objective collaborative processing algorithm, the training dataset is iterated using a preset loss function to obtain the iterated FPGA processing module; the calculation formula of the loss function is: Where L represents the total loss; and These are the true distance and predicted distance of the i-th target, respectively; and λ represents the actual velocity and predicted velocity of the j-th target, respectively; λ is the weighting factor. The number of targets used for distance calculation; The number of targets used for velocity calculation.

3. The method for calculating 94GHz radar baseband signals according to claim 1, characterized in that, The noise filtering process employs an autocorrelation function and a low-pass filter; the formula for calculating the autocorrelation function is as follows: The calculation formula for the low-pass filter is as follows: ;in, The autocorrelation function is mentioned above. Here, t is the low-pass filter; t is the integration variable; The target filtered signal; It is a time constant; The signal frequency; This is the filter cutoff frequency.

4. The method for calculating 94GHz radar baseband signals according to claim 1, characterized in that, The calculation function for the normalization process is: ;in, For normalized signals; For target normalization signal; The mean of the signal; This represents the standard deviation of the signal.

5. The method for calculating 94GHz radar baseband signals according to claim 2, characterized in that, The calculation formula for the multi-objective solution graph includes: 、 as well as ;in, It represents the distance difference; For speed difference; For the angle difference; The speed of light; The signal scan period; This is the frequency difference; Λ represents the radar signal bandwidth; Λ represents the radar wavelength. For Doppler frequency shift; Phase difference; This refers to the spacing between antenna elements.

6. The method for calculating 94GHz radar baseband signals according to claim 5, characterized in that, The importance function is: The resource allocation optimization function is: ;in, The importance weight of the i-th objective; This is the optimal resource allocation scheme; and These are the distance and velocity characteristics of the i-th target, respectively; and α and β are the distance and velocity characteristics of the j-th target, respectively; α and β are both weighting factors. Let be the utility function for the i-th objective; This represents finding the optimal solution for the distance difference.

7. A 94GHz radar target detection and resolution module, characterized in that, The method for calculating a 94GHz radar baseband signal according to claim 1 includes the following system: a receiving unit for receiving a 94GHz millimeter-wave signal and performing mixing, amplification, and noise reduction on the 94GHz millimeter-wave signal to obtain an intermediate frequency (IF) signal; a conversion unit for converting the IF signal into a digital signal; a parallel processing unit for performing noise filtering, normalization, and multi-level parallel signal processing on the digital signal to obtain a multi-target difference comparison heatmap; and a calculation unit for calculating the distance, velocity, and angle of multiple targets based on the multi-target difference comparison heatmap. The system performs line detection and calculation to obtain a multi-objective solution graph; an optimization unit is used to dynamically adjust the FFT window, filter bandwidth, and gain control parameters of the parallel processing unit; a control unit is used to dynamically adjust the computing resources of the parallel processing unit, the calculation unit, and the optimization unit using the importance function and the resource allocation optimization function, and upload the multi-objective solution graph to the target host computer; a synchronization unit is used to provide a synchronization clock signal to the receiving unit, the conversion unit, the parallel processing unit, the calculation unit, the optimization unit, and the control unit.

Images