Azimuth angle estimation system and method

Abstract

The invention provides an azimuth angle estimation system and method, and aims to solve the problems of low angle resolution, large calculation amount and poor adaptability to a sparse array in the prior art. The system adopts a hierarchical search strategy and combines FFT and DBF technologies to realize high-precision azimuth angle estimation. Rough estimation of a large-view-field target is carried out through FFT, and then fine adjustment is carried out on a target area by using DBF, so that the processing efficiency is improved while the precision is ensured. According to the system, the processing flow is divided into a plurality of functional units through modular design, parallel processing of a plurality of targets is achieved through a Dataflow mechanism, and the processing throughput and the real-time performance are greatly improved. Besides, for a non-equal-interval antenna array, an array interpolation and zero padding mechanism is introduced, high-precision angle estimation is ensured, and hardware complexity is reduced. Generally speaking, the precision and the real-time performance of azimuth angle estimation are remarkably improved, and the method is suitable for high-precision target detection in the fields of automatic driving, intelligent transportation and the like.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of radar detection, and particularly relates to an azimuth angle estimation system and method. BACKGROUND

[0002] With the rapid development of automatic driving, intelligent transportation systems and other intelligent perception technologies, 4D imaging millimeter wave radars as key sensors are widely used in high-precision target detection and tracking. It can provide real-time target distance, speed, azimuth angle and time information, providing vital environmental perception support for automatic driving systems. However, the traditional azimuth angle estimation methods in the prior art still face many technical challenges when processing 4D imaging radar data, which restricts the overall performance and application effect of the system.

[0003] Currently, most millimeter wave radar systems rely on Fast Fourier Transform (FFT) for azimuth angle estimation. Although this method can provide approximate azimuth angle information of the target, it still has some obvious limitations. First, the angle resolution of the FFT method is highly dependent on the choice of FFT points. Although increasing the number of points can improve the angle accuracy, it will also significantly increase the computational burden, thereby affecting the real-time performance and processing capacity of the system. Second, as the number of targets increases, the amount of calculation required by the traditional method will also increase significantly, and the system will be difficult to meet the demand for real-time processing of a large number of targets in high-speed scenarios. In addition, existing methods usually assume that the radar antenna array is arranged at equal intervals, but in actual applications, the radar array often adopts sparse configuration. This irregular antenna arrangement will cause the accuracy and stability of the traditional azimuth angle estimation method to be unable to be guaranteed.

[0004] Therefore, how to improve the accuracy of angle estimation while reducing the amount of calculation and improving the real-time processing capacity of the system, especially in complex environments, has become the main challenge faced by current technology. In order to solve these problems, it is urgent to develop a new azimuth angle estimation system and method to realize higher precision, lower delay and higher concurrent processing capacity of azimuth angle estimation, and meet the high real-time requirements of automatic driving and intelligent transportation systems. SUMMARY

[0005] The purpose of the present application is to propose a high-precision, low-delay azimuth estimation method and device, aiming to solve the problems of insufficient azimuth estimation precision, high calculation delay and limited processing capacity for a large number of targets in existing radar systems. By adopting a hierarchical search structure and an efficient pipeline architecture, the present application aims to achieve high-precision azimuth estimation and complete real-time estimation of thousands of targets within milliseconds, meeting the needs of applications such as autonomous driving and intelligent transportation. At the same time, the system can also support sparse array antennas, reduce hardware costs, and provide additional information such as signal-to-noise ratio and peak energy to enhance the robustness and target data processing capacity of the system.

[0006] The present application provides an azimuth estimation system, which comprises: a storage unit for storing the echo signals corresponding to the detected targets received by the antenna array of the radar into DDR; a data extraction unit for extracting the RDFFT data of the target from the echo signals stored in the DDR; a preprocessing unit for preprocessing the RDFFT data; a frequency domain analysis unit for converting the time domain signals in the RDFFT data into frequency domain signals by FFT according to the preprocessed RDFFT data, generating the power spectrum of the target, and extracting the frequency domain features of the target from the power spectrum, performing a first azimuth search on the frequency domain features in the power spectrum to identify a preset number of reference peak points; a first DBF unit for performing a second azimuth search according to the preset number of reference peak points to obtain low-precision azimuth angle information of the target; a second DBF unit for performing a third azimuth search according to the low-precision azimuth angle information to obtain high-precision azimuth angle information of the target; an output unit for mapping the high-precision azimuth angle information to a real coordinate system to generate and output the final azimuth angle information of the target.

[0007] Further, the system further comprises: a parallel control unit for executing the preset steps in parallel according to the number of echo signals using the Dataflow mechanism.

[0008] Further, the parallel control unit comprises: a signal analysis module for analyzing the number of echo signals; The scheduling decision module is configured to determine, when the number of echo signals is greater than 1, to adopt an adaptive scheduling algorithm to dynamically adjust the execution order and parallelism of each processing unit in the azimuth angle estimation system, except the storage unit, the output unit and the parallel control unit; The feedback adjustment module is configured to dynamically adjust the scheduling strategy according to the real-time execution of each processing unit, except the storage unit, the output unit and the parallel control unit. The parallel execution module is configured to perform the preset steps in parallel according to the scheduling strategy.

[0009] Further, the data extraction unit is specifically configured to calculate the storage offset address of the target in the DDR to locate the storage position of the RDFFT data of the target in the DDR by processing the target echo signal through the CFAR algorithm, and extract the RDFFT data of the corresponding target from the storage position.

[0010] Further, the preprocessing unit comprises: The window function processing module is configured to multiply the time domain signal of the echo signal by a window function value to obtain the data processed by the window function, before preprocessing the RDFFT data, by applying the window function to the echo signal data. The antenna phase calibration module is configured to perform phase compensation on the data processed by the window function according to a preset antenna calibration coefficient. The data interpolation and zero padding module is configured to rearrange the data compensated by the phase calibration, and perform interpolation or zero padding operation on the rearranged data, according to the actual geometric distribution of the antenna array of the radar. The pitch angle compensation module is configured to correct the data after interpolation or zero padding by taking the total angle value as a compensation factor, to obtain the preprocessed RDFFT data, according to the geometric position of the antenna array of the radar and the echo signal.

[0011] In a second aspect, the application provides an estimation method based on an azimuth angle estimation system, which comprises the following steps: S1, storing the echo signal corresponding to the detected target received by the antenna array of the radar into the DDR; S2, extracting the RDFFT data of the target from the echo signal stored in the DDR; S3, preprocessing the RDFFT data; S4, converting time domain signals in the RDFFT data into frequency domain signals by FFT according to the preprocessed RDFFT data to generate a power spectrum of the target, and extracting frequency domain features of the target from the power spectrum, and performing a first azimuth angle search on the frequency domain features in the power spectrum to identify a preset number of reference peak points; S5, performing a second azimuth angle search according to the preset number of reference peak points to obtain low-precision azimuth angle information of the target; S6, performing a third azimuth angle search according to the low-precision azimuth angle information to obtain high-precision azimuth angle information of the target; S7, mapping the high-precision azimuth angle information to a real coordinate system to generate and output final azimuth angle information of the target; S8, executing the steps S1-S8 in parallel cycles by using a Dataflow mechanism according to the number of echo signals.

[0012] Further, the step S3 comprises: S31, before preprocessing the RDFFT data, applying a window function to the echo signal data, multiplying time domain signals of the echo signals by window function values to obtain data processed by the window function; S32, performing phase compensation on the data processed by the window function according to a preset antenna calibration coefficient; S33, rearranging the data compensated by the phase calibration according to an actual geometric distribution of an antenna array of the radar, and performing interpolation or zero padding operation on the rearranged data; S34, calculating a total angle value according to a geometric position of the antenna array and the received signals, and correcting the data after interpolation and zero padding by taking the total angle value as a compensation factor to obtain the preprocessed RDFFT data.

[0013] Further, the step S4 comprises: S41, calculating a power spectrum of the target by FFT according to the preprocessed RDFFT data; S42, performing normalization processing on the power spectrum and removing possible noise interference; S43, in the power spectrum after the normalization processing, applying a peak detection algorithm to find a plurality of peak points with the largest amplitude; S44, outputting angle positions and amplitude information of the plurality of peak points as starting points for subsequent refined azimuth angle estimation.

[0014] Further, the step S6 comprises: S61, performing fine azimuth search on the low-precision azimuth angle information using a high-precision algorithm to narrow the search range; S62, further adjusting the search step and precision based on the narrowed search range to obtain the high-precision azimuth angle information.

[0015] Further, the step S8 comprises: S81, analyzing the number of echo signals; S82, when the number of echo signals is greater than 1, using an adaptive scheduling algorithm to dynamically adjust the execution order and parallelism of each processing unit other than the storage unit, the output unit and the parallel control unit; S83, dynamically adjusting the scheduling strategy according to the real-time execution of each processing unit other than the storage unit, the output unit and the parallel control unit; S84, performing the steps S1-S8 in parallel according to the scheduling strategy and the Dataflow mechanism.

[0016] The advantages of the present application mainly lie in high precision, low delay and high concurrent processing capability. By introducing hierarchical search and modular pipeline architecture, the present application can process thousands of targets in real time while ensuring 0.1 degree angle precision, and greatly reduce the calculation delay. In addition, the system uses sparse array technology, which can not only maintain high angle estimation accuracy, but also reduce hardware complexity and cost, and has strong flexibility and adaptability. The system can also output rich estimation information, including SNR and peak energy, further enhancing the data quality control, filtering and fusion capability. Overall, the present application not only improves the precision and real-time performance of azimuth angle estimation, but also improves the processing efficiency of the system in high target density scenarios, and adapts to complex application environments such as automatic driving and intelligent transportation. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0018] Figure 1 is a functional module schematic diagram of an azimuth angle estimation system provided by an embodiment of the present application; Figure 2 Figure 2 is a functional module schematic diagram of a preprocessing unit of an azimuth angle estimation system provided by an embodiment of the present application; Figure 3This is a schematic diagram of the functional modules of a parallel control unit of an azimuth estimation system provided in one embodiment of this application; Figure 4 This is a schematic diagram of the Dataflow mechanism of an azimuth estimation system provided in one embodiment of this application; Figure 5 This is a flowchart of an azimuth estimation method based on a radar azimuth estimation system, provided in one embodiment of this application. Detailed Implementation

[0019] The specific embodiments of this application will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of this application, and not all of them. Based on the description of this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0020] Example 1 One embodiment of this application provides an azimuth estimation system. Please refer to [link to relevant documentation]. Figures 1 to 4 The system 100 includes: a storage unit 20, a data extraction unit 30, a preprocessing unit 40, a frequency domain analysis unit 50, a first DBF unit 60, a second DBF unit 70, and an output unit 80. And parallel control unit 90.

[0021] Its functions and implementation will be described in detail below: Specifically, the storage unit 20 is used to store the echo signal received by the radar antenna array corresponding to the detected target into a double data rate synchronous dynamic random access memory (DDR).

[0022] Optionally, the main function of the storage unit 20 is to store the echo signal received by the radar antenna array into DDR. DDR is a high-efficiency memory that can transmit data simultaneously on both the rising and falling edges of the clock, thus its data transfer rate is twice that of traditional SDRAM. By storing the echo signal in DDR, the storage unit provides a basis for fast access to subsequent data retrieval and processing. By storing the echo signal in DDR, the system can quickly retrieve data when needed, ensuring real-time performance.

[0023] Specifically, the data extraction unit 30 is used to extract the range-Doppler Fast Fourier Transform (RDFFT) data of the target from the echo signal stored in the DDR.

[0024] Optionally, the main function of the data extraction unit 30 is to extract the target's RDFFT data from the echo signal stored in DDR. This unit uses the Constant False Alarm Rate (CFAR) algorithm to analyze the echo signal, identify potential target signals, and calculate the target's storage offset address in DDR. The CFAR algorithm, through statistical analysis of the echo signal, can effectively distinguish the target signal from background noise, thereby improving the accuracy of target detection. Through the above process, the system can efficiently locate and extract the required data, laying the foundation for subsequent data preprocessing and analysis.

[0025] Specifically, the preprocessing unit 40 is used to preprocess the RDFFT data.

[0026] In one embodiment, the preprocessing unit includes: a window function processing module 401, an antenna phase calibration module 402, a data interpolation and zero-padding module 403, and an elevation angle compensation module 404.

[0027] The window function processing module 401 is used to apply a window function to the echo signal data before performing RDFFT, multiplying the time-domain signal of the echo signal with the window function value to reduce the spectral leakage of the time-domain signal during the FFT process.

[0028] Optionally, the window function processing module 401 applies a window function to the echo signal data before performing RDFFT to reduce spectral leakage during the FFT process. The application of the window function can be achieved through the following formula:

[0029] Here, s[n] is the original signal, and w[n] is the window function. By multiplying the time-domain signal with the window function, the window function can effectively suppress the edge effects of the signal during FFT transformation, thereby improving the resolution of the spectrum.

[0030] The antenna phase calibration module 402 is used to perform phase compensation on the data processed by the window function according to the preset antenna calibration coefficient, so as to eliminate or reduce the phase difference between the data.

[0031] Optionally, the antenna phase calibration module 402 performs phase compensation on the data processed by the window function according to a preset antenna calibration coefficient. Due to errors in the manufacturing and installation process of radar antennas, initial phase differences may exist between different antennas. This module eliminates or reduces these phase differences through phase compensation, ensuring data consistency. Phase compensation can be expressed as:

[0032] in, It is the phase calibration coefficient, y[n] is the signal after phase calibration, j represents the imaginary unit, equal to .

[0033] The data interpolation and zero-padding module 403 is used to rearrange the phase-calibrated and compensated data according to the actual geometric distribution of the antenna array, and construct a dense virtual array through interpolation or zero-padding operations to improve the resolution of the data in the time-frequency domain of FFT.

[0034] Optionally, the data interpolation and zero-padding module 403 rearranges the phase-calibrated data according to the geometric distribution of the antenna array, and constructs a dense virtual array through interpolation or zero-padding operations. The purpose of interpolation and zero-padding operations is to increase the number of data points and improve the resolution of the FFT time-frequency domain. In this way, the system can better adapt to the problems caused by non-equidistant arrays and ensure that more accurate angle information can be obtained in subsequent processing.

[0035] The elevation angle compensation module 404 is used to calculate the total angle based on the geometric position of the antenna array and the received signal, and use the total angle as a compensation factor to correct the zero-padding data after interpolation, thereby correcting the elevation angle power spectrum shift caused by antenna arrangement error.

[0036] Optionally, the elevation angle compensation module 404 calculates the total angle based on the geometric position of the antenna array and the received signal, and corrects the data. The calculated total angle is used as a compensation factor to correct the elevation angle power spectrum shift caused by antenna arrangement errors, thereby improving the accuracy of angle estimation. The compensation formula can be expressed as:

[0037] in, It is the pitch angle compensation factor, z[n] is the signal after pitch angle compensation, j represents the imaginary unit, equal to .

[0038] Specifically, the frequency domain analysis unit 50 is used to convert the time domain signal in the preprocessed RDFFT data into a frequency domain signal through FFT, generate the power spectrum of the target and extract the frequency domain features of the target, perform a first azimuth search on the frequency domain features in the power spectrum, and identify a preset number of reference peak points.

[0039] Optionally, the main task of the frequency domain analysis unit 50 is to convert the preprocessed RDFFT data into a frequency domain signal using FFT, generate the target's power spectrum, and extract frequency domain features. This unit first performs an FFT transform on the time domain signal to obtain the frequency domain signal. The FFT formula can be expressed as:

[0040] Where X(f) is the frequency domain signal, x[n] is the time domain signal, N is the number of points in the FFT, and j represents the imaginary unit, equal to .

[0041] By using FFT transformation, the system can generate the target's power spectrum and perform an initial azimuth search within the power spectrum to identify a predetermined number of reference peak points. These reference peak points provide important reference information for subsequent fine-grained searches.

[0042] Specifically, the first digital beamforming (DBF) unit 60 is used to perform a second azimuth search based on the preset number of reference peak points to obtain low-precision azimuth information of the target.

[0043] Optionally, the main function of the first DBF unit 60 is to perform a second azimuth search based on a preset number of reference peak points to obtain low-precision azimuth information of the target. The DBF unit 60 generates a corresponding steering vector based on the extracted reference peak points for performing the low-precision azimuth search. The formula for calculating the steering vector is:

[0044] in, It's the azimuth. d is the signal wavelength, d is the antenna spacing, N is the number of antennas, and j represents the imaginary unit, equal to .

[0045] The DBF unit performs a dot product between the steering vector and the received signal to calculate the energy value at each candidate angle. The formula for calculating the energy value is:

[0046] Where y is the received signal and H represents the conjugate transpose.

[0047] By comparing the energy values ​​of different candidate angles, the DBF unit 60 determines the angle with the highest energy as a preliminary estimate of the low-precision azimuth angle.

[0048] Specifically, the second DBF unit 70 is used to perform a third azimuth search based on the low-precision azimuth information to obtain the high-precision azimuth information of the target.

[0049] Optionally, the main function of the second DBF unit 70 is to perform a third azimuth search based on the low-precision azimuth information obtained by the first DBF unit 60 to obtain high-precision azimuth information of the target. Based on the low-precision azimuth output by the first DBF unit 60, the second DBF unit 70 narrows the search range, focusing on a more detailed search around that angle. Within the narrowed search range, the second DBF unit 70 uses a smaller angle step size to improve angular resolution. The second DBF unit 70 calculates the energy values ​​of different candidate angles, identifies the peak point with the highest energy, and uses it as the final output of the high-precision azimuth. Finally, the high-precision azimuth information output by the second DBF unit 70 will be used for subsequent target tracking and data fusion.

[0050] Specifically, the output unit 80 is used to map high-precision azimuth information to the real coordinate system to generate and output the final azimuth information of the target.

[0051] Optionally, the output unit 80 maps high-precision azimuth information to a real coordinate system to generate and output the target's final azimuth information. Through coordinate transformation, the system can combine high-precision azimuth information with the actual scene, providing support for subsequent target tracking and data fusion. The output process can be achieved using the following formula:

[0052] in, It is the azimuth information mapped to the real coordinate system. It is the azimuth information in the local coordinate system. It is a coordinate transformation matrix.

[0053] Specifically, the parallel control unit 90 is used to dynamically schedule the execution order and parallelism of each processing unit of the system other than the storage unit 20, the output unit 80 and the parallel control unit 90, based on the characteristics of the echo signals when multiple target echo signals are received, using the Dataflow mechanism.

[0054] In one embodiment, the parallel control unit includes: a signal analysis module 901, a scheduling decision module 902, a feedback adjustment module 903, and a parallel execution module 904.

[0055] The signal analysis module 901 is used to analyze the number of echo signals.

[0056] Optionally, the main function of the signal analysis module 901 is to analyze the number of received echo signals. This module uses the CFAR algorithm to analyze the received echo signals and calculate the number of signals. The CFAR algorithm dynamically adjusts the detection threshold based on the statistical characteristics of background noise by setting a false alarm rate, thereby improving the accuracy of target detection.

[0057] It is understood that the detection threshold refers to a threshold used to distinguish target signals from background noise during signal processing and target detection. Specifically, when using the CFAR algorithm, the detection threshold is dynamically calculated based on the statistical characteristics of background noise. When the received signal strength exceeds this threshold, the system determines that it may contain a target signal. In the CFAR algorithm, the threshold is typically set based on the noise level of the surrounding area to ensure a certain false alarm rate (i.e., the probability of incorrectly identifying noise as a target signal) under different environmental conditions. By dynamically adjusting the detection threshold, the CFAR algorithm can adapt to environmental changes, enhance the accuracy of target detection, and reduce missed detections and false detections.

[0058] The scheduling decision module 902 is used to determine that when the number of echo signals is greater than 1, it uses an adaptive scheduling algorithm to dynamically adjust the execution order and parallelism of each processing unit except for the storage unit 20, the output unit 80 and the parallel control unit 90.

[0059] Optionally, the main task of the scheduling decision module 902 is to dynamically adjust the execution order and parallelism of each processing unit based on the output of the signal analysis module, when the number of echo signals is greater than 1. This module employs an adaptive scheduling algorithm to dynamically adjust the execution order of processing units according to the current system load and the number of targets. For example, when the number of targets is large, signal analysis, data extraction, and preprocessing units are prioritized to ensure timely data processing; while when the number of targets is small, the parallelism of the processing units can be reduced to save computing resources. Based on the analysis results, the scheduling decision module generates a scheduling strategy that indicates the execution order and parallelism of each processing unit to achieve efficient resource utilization.

[0060] The feedback adjustment module 903 is used to dynamically adjust the scheduling strategy based on the real-time execution status of each processing unit other than the storage unit 20, the output unit 80 and the parallel control unit 90.

[0061] Optionally, the feedback adjustment module 903 is used to dynamically adjust the scheduling strategy based on the real-time execution status of each processing unit. This module continuously monitors the execution status of each processing unit, including execution time and resource utilization, to ensure efficient system operation. Based on the monitoring results, the feedback adjustment module can adjust the scheduling strategy in real time. For example, if the execution time of a certain processing unit is too long, the system can choose to reduce its parallelism or adjust its execution order to improve overall processing efficiency. Through dynamic adjustment, the feedback adjustment module ensures efficient utilization of system resources and avoids performance degradation of the entire system due to the delay of a certain processing unit.

[0062] The parallel execution module 904 is used to execute the scheduling process in parallel according to the scheduling strategy.

[0063] Optionally, the parallel execution module 904 executes the scheduling process in parallel according to a scheduling strategy. This module initiates multiple processing units to work in parallel based on the strategy generated by the scheduling decision module, thereby improving processing throughput. The parallel execution module assigns tasks to each processing unit, ensuring that each unit works efficiently within its capacity, thus optimizing processing efficiency. The parallel execution module is responsible for coordinating the synchronization between the processing units, ensuring smooth data flow and timely feedback of processing results.

[0064] For example, the Dataflow mechanism of the azimuth estimation system can be summarized as follows: As can be understood, the Dataflow mechanism is a data-driven computing model that allows the system to dynamically schedule and execute tasks based on data availability. In this mechanism, processing units are connected via data flow, and the generation and consumption of data determine the execution order of the processing units. The units participating in this mechanism's flow include: storage unit 20, data extraction unit 30, preprocessing unit 40, frequency domain analysis unit 50, first DBF unit 60, second DBF unit 70, output unit 80, and parallel control unit 90. These units are connected in series to form a data-driven pipeline structure through a unified data interface protocol.

[0065] Optionally, the DDR of the storage unit 20 stores preset antenna array parameters, antenna calibration coefficients, total angle values, and data such as velocity-distance FFT and peak energy (PEAKMAP) obtained from the output unit, which can be called up at any time as a data source for other processing units.

[0066] Optionally, the signal analysis module in the parallel control unit 90 determines the number of targets to be processed based on the number of received echo signals; the scheduling decision module determines the execution order and parallelism of each processing unit based on the number of targets and the system status. If the number of targets is equal to 1, data is transmitted and processed in the normal order of storage unit 20, data extraction unit 30, preprocessing unit 40, frequency domain analysis unit 50, first DBF unit 60, second DBF unit 70, and output unit 80. If the number of targets is greater than 1, an adaptive scheduling algorithm is used to dynamically adjust the execution order and parallelism of each processing unit except for storage unit 20, output unit 80, and parallel control unit 90; the feedback adjustment module dynamically adjusts the scheduling strategy based on the real-time execution status to cope with changes in system load; the parallel execution module executes the tasks of each processing unit in parallel according to the scheduling strategy to ensure that multiple targets can be processed simultaneously.

[0067] We can understand that we consider the flow of a batch of data from storage unit 20 to output unit as a data lifecycle. Within this lifecycle, the parallel control unit 90 exists independently, providing dynamic scheduling services to each processing unit. For example... Figure 4 As shown, in any data lifecycle, the operations of multiple processing units are independent and performed in parallel.

[0068] The parallel control unit 90 dynamically allocates tasks to the data extraction unit 30, preprocessing unit 40, frequency domain analysis unit 50, first DBF unit 60, and second DBF unit 70 based on the received echo signals from multiple targets. Each processing unit receives an independent task, ensuring that data from multiple targets can be processed simultaneously. For example, the data extraction unit 30 may initiate multiple extraction tasks in parallel, with each task responsible for extracting the RDFFT data of one target.

[0069] Data extraction unit 30 extracts RDFFT data from storage unit 20 and immediately passes it to preprocessing unit 40 without waiting for subsequent units to complete their operations. This approach allows the system to prepare for the next target's data extraction task while extracting the RDFFT data for the current target.

[0070] The extracted RDFFT data is sent to the preprocessing unit 40. At this stage, the preprocessing unit 40 processes the RDFFT data of each target independently and passes the preprocessed data to the frequency domain analysis unit 50, without waiting for subsequent units to complete their operations. This allows the preprocessing unit 40 to process data from multiple targets simultaneously, improving processing efficiency.

[0071] The preprocessed data enters the frequency domain analysis unit 50 for FFT calculation and peak detection. When processing the RDFFT data of each target, the frequency domain analysis unit 50 can perform FFT calculations in parallel and pass the peak point information of each target to the first DBF unit 60 without waiting for the processing of other targets to be completed.

[0072] The first DBF unit 60 receives peak point information from the frequency domain analysis unit 50 and performs a low-precision azimuth search. The parallel control unit allocates multiple parallel low-precision search tasks to the first DBF unit. Each task independently processes the low-precision azimuth search of one target, ensuring that the search for multiple targets can be performed simultaneously.

[0073] The second DBF unit 70 receives low-precision azimuth information from the first DBF unit 60 and performs a high-precision azimuth search. Similarly, multiple high-precision search tasks can be executed in parallel to ensure that the high-precision azimuth information of all targets can be calculated simultaneously.

[0074] After completing all processing, the second DBF unit 70 transmits the high-precision azimuth information to the output unit 80. The output unit 80 packages this information with the relevant peak index map (PEAKMAP) and signal-to-noise ratio (SNR) data to form a complete output result, which is then stored in the DDR of the storage unit 20 for use by the unit in the next data cycle.

[0075] During the same IPEAK (Initial Peak Index, such as IPEAK1), the processing unit corresponding to each IPEAK is different due to the task allocation mechanism of the parallel control unit. This is because the signal characteristics and processing requirements of each target may differ. The system allocates a specific processing unit to each target through the parallel control unit 90 so that different targets can be processed independently at the same time. For example, when processing IPEAK1, the data extraction unit 30 may be extracting the RDFFT data of target A, while the preprocessing unit 40 is processing the RDFFT data of target B, and the frequency domain analysis unit 50 is performing FFT calculations on target C.

[0076] This design ensures that the system can achieve high concurrency and real-time processing within milliseconds. When multiple target PEAKs are input in parallel, the parallel control unit schedules them to continue executing the data transmission process sequentially. Multiple data streams can be advanced in parallel to processing units at different data lifecycles. In this way, the resources of the processing units are fully utilized, and the parallelism of the system is significantly expanded, thereby achieving efficient and real-time multi-target data processing.

[0077] Example 2 One embodiment of this application provides an azimuth estimation method. Please refer to [link to relevant documentation]. Figure 5 This method is based on the azimuth estimation system described in Example 1, and includes the following steps: S1 stores the echo signal received by the radar's antenna array corresponding to the detected target into DDR.

[0078] S2, extract the target's RDFFT data from the echo signal stored in the DDR.

[0079] S3, preprocess the RDFFT data.

[0080] In one embodiment, step S3 includes: S31, Before preprocessing the RDFFT data, a window function is applied to the echo signal data, multiplying the time-domain signal of the echo signal by the window function value to obtain the data after window function processing.

[0081] Optionally, this step uses a window function to window the echo signal data, multiplying the time-domain signal by the window function value to calculate the signal after window function processing. :

[0082] S32, perform phase compensation on the data processed by the window function according to the preset antenna calibration coefficient.

[0083] Optionally, phase compensation is performed on the data processed by the window function according to a preset antenna calibration coefficient, and the phase-calibrated signal is calculated. :

[0084] S33, based on the actual geometric distribution of the radar's antenna array, rearrange the phase-calibrated and compensated data, and perform interpolation or zero-padding operations on the rearranged data.

[0085] Optionally, the phase-calibrated data can be rearranged according to the geometric distribution of the antenna array. A dense virtual array can be constructed through interpolation or zero-padding to increase the number of data points and ensure more accurate angle information is obtained in subsequent processing.

[0086] S34. The total angle value is calculated based on the geometric position of the antenna array and the received signal. The total angle value is used as a compensation factor to correct the interpolated zero-padded data to obtain the preprocessed RDFFT data.

[0087] Optionally, the total angle is calculated based on the geometric position of the antenna array and the received signal, and used as a compensation factor to correct the zero-padded data after interpolation, thus calculating the signal after elevation angle compensation.

[0088]

[0089] S4. Based on the preprocessed RDFFT data, the time-domain signal in the RDFFT data is converted into a frequency-domain signal by FFT to generate the power spectrum of the target. The frequency-domain features of the target are extracted from the power spectrum. The frequency-domain features are searched for a first azimuth angle in the power spectrum to identify a preset number of reference peak points.

[0090] In one embodiment, step S4 includes: S41, Based on the preprocessed RDFFT data, calculate the power spectrum of the target using FFT.

[0091] Optionally, the preprocessed time-domain signal can be transformed using the FFT algorithm to obtain the frequency-domain signal. :

[0092] After obtaining the frequency domain signal, the power spectrum of the target is generated.

[0093] S42, normalize the power spectrum and remove any possible noise interference.

[0094] Optionally, the calculated power spectrum can be normalized to ensure comparisons under different conditions. Then, potential noise interference can be removed to ensure the accuracy of subsequent analyses.

[0095] S43, In the normalized power spectrum, a peak detection algorithm is applied to find several peak points with the largest amplitude.

[0096] Optionally, a peak detection algorithm can be applied to the normalized power spectrum to find several peak points with the largest amplitudes. The angular positions and amplitude information of these peak points are then output to provide an initial reference for subsequent refined azimuth estimation.

[0097] S44, output the angular position and amplitude information of the several peak points as the starting point for subsequent refinement of azimuth angle estimation.

[0098] Optionally, the angle position and amplitude information of the identified peak points are organized and output to the subsequent processing unit.

[0099] S5, perform a second azimuth search based on the preset number of reference peak points to obtain the low-precision azimuth information of the target.

[0100] S6. Perform a third azimuth search based on the low-precision azimuth information to obtain the high-precision azimuth information of the target.

[0101] In one embodiment, step S6 includes: S61, a high-precision algorithm is used to perform a fine azimuth search on the low-precision azimuth information to narrow the search range.

[0102] Optionally, a high-precision algorithm can be used to analyze the low-precision azimuth information. Based on the existing low-precision information, the search range can be narrowed to improve the efficiency and accuracy of subsequent searches.

[0103] S62, based on the narrowed search range, further adjust the search step size and precision to obtain the high-precision azimuth information.

[0104] S7, map the high-precision azimuth information to the real coordinate system to generate and output the final azimuth information of the target.

[0105] S8. Based on the number of echo signals, the S1-S8 steps are executed in parallel and cyclically using the Dataflow mechanism.

[0106] In one embodiment, step S8 includes: S81, Analyze the number of echo signals; Optionally, a quantity analysis is performed on the received echo signals to determine the number of targets that the current system needs to process. The number of signals is calculated by analyzing the received signals using the CFAR algorithm.

[0107] S82, when it is determined that the number of echo signals is greater than 1, an adaptive scheduling algorithm is used to dynamically adjust the execution order and parallelism of each processing unit except the storage unit 20, the output unit 80 and the parallel control unit 90; Optionally, if the number of echo signals is greater than 1, an adaptive scheduling algorithm is activated. The execution order and parallelism of each processing unit, except for storage unit 20, output unit 80, and parallel control unit 90, are dynamically adjusted to optimize processing efficiency.

[0108] S83, dynamically adjust the scheduling strategy based on the real-time execution status of each processing unit other than the storage unit 20, the output unit 80 and the parallel control unit 90; Optionally, the real-time execution status of each processing unit can be monitored, and scheduling strategies can be dynamically adjusted to ensure that the system can still operate efficiently under high load.

[0109] S84, execute steps S1-S8 in parallel loop according to the scheduling strategy and Dataflow mechanism.

[0110] Optionally, this process ensures that multiple targets can be processed simultaneously, thereby improving the overall processing capacity and real-time performance of the system.

[0111] Specifically, the parallel control unit 90 schedules each processing unit in real time according to the scheduling strategy and the Dataflow mechanism: At the beginning of the entire processing flow, storage unit 20, acting as the main data repository, is responsible for storing echo signal data from various targets. During this stage, the parallel control unit 90 monitors newly received target echo signals and, based on the characteristics of these signals, determines the number and type of data extraction tasks to initiate. Specifically, storage unit 20 prepares the corresponding RDFFT data according to the instructions of the parallel control unit for extraction and processing in subsequent steps.

[0112] The data extraction unit extracts RDFFT data from the storage unit 20. At this time, the parallel control unit 90 initiates multiple parallel extraction tasks for the data extraction unit 30 based on the echo signals from the multiple targets currently received. Each task independently extracts the RDFFT data of one target, ensuring that the signals from multiple targets can be processed simultaneously.

[0113] The extracted RDFFT data is then sent to the preprocessing unit 40. During this stage, the parallel control unit 90 initiates multiple parallel preprocessing extraction tasks for the preprocessing unit 40. Each task independently processes the RDFFT data of one target within the preprocessing unit. Because these processes are independent of each other, the preprocessing unit 40 can preprocess data from multiple targets simultaneously.

[0114] The preprocessed data is fed into the frequency domain analysis unit 50. During this stage, the parallel control unit 90 initiates multiple parallel frequency domain analysis tasks for the frequency domain analysis unit 50. Each task independently processes the RDFFT data of one target within the frequency domain analysis unit, performs FFT calculations on it, generates a power spectrum, detects peak points in the power spectrum, and identifies a preset number of reference peak points. Because the frequency domain analysis of each target is independent, multiple FFT calculations can be performed in parallel, thereby accelerating the processing speed.

[0115] The first DBF unit 60 performs a low-precision azimuth search based on the reference peak point of each target. During this stage, the parallel control unit initiates multiple parallel low-precision search tasks for the first DBF unit 60. Each task independently processes the low-precision azimuth search of one target within the first DBF unit 60, ensuring that searches for multiple targets can be performed simultaneously.

[0116] The second DBF unit 70 receives low-precision azimuth information from the first DBF unit 60 and performs a high-precision azimuth search. During this stage, the parallel control unit initiates multiple parallel high-precision search tasks for the second DBF unit 70. Each task independently processes the high-precision search of one target within the second DBF unit 70, ensuring that the high-precision azimuth information of all targets can be calculated simultaneously.

[0117] The second DBF unit 70 transmits the high-precision azimuth information obtained from the refined search to the output unit 80. The output unit 80 packages this information with the relevant PEAKMAP and SNR data to form a complete set of output data. After outputting the results, the system clears the state of the current processing unit to prepare for the next data cycle.

[0118] Finally, the system will automatically return to the data extraction step (S1) and repeat all processing steps from S1 to S8. During this process, the storage unit 20 will prepare new RDFFT data according to the instructions of the parallel control unit 90, ensuring efficient processing of new target data in each data cycle. The parallel control unit 90 will dynamically adjust the parallel processing strategy to ensure that the system can continuously and efficiently respond to newly received target echo signals, achieving real-time multi-target tracking and positioning. Through this cyclical mechanism, the system can complete high-precision azimuth angle estimation of multiple targets within milliseconds, meeting the needs of modern autonomous driving and intelligent transportation systems.

[0119] This application addresses several technical challenges faced by existing millimeter-wave radar azimuth estimation methods when processing 4D imaging radar data, including low angular resolution, high computational cost, and insufficient real-time processing capability in high-target-density scenes. Traditional methods rely on FFT for azimuth estimation, which provides approximate azimuth information, but the computational burden increases significantly with the number of targets, making it difficult for the system to meet the requirements of real-time performance and high accuracy. By introducing a hierarchical search strategy and a modular pipeline architecture, this application effectively improves the accuracy and real-time performance of azimuth estimation, ensuring efficient processing of a large number of targets in complex application environments.

[0120] This application significantly improves the accuracy and real-time performance of azimuth estimation by introducing a hierarchical search strategy and a modular pipeline architecture. The system combines FFT and digital beamforming (DBF) techniques to achieve both coarse and fine target estimation, ensuring real-time processing of tens of thousands of targets while maintaining an angular accuracy of 0.1 degrees. Furthermore, the use of a Dataflow mechanism and sparse array technology not only improves processing efficiency and reduces computational latency but also enhances the system's flexibility and adaptability, thereby meeting the real-time processing requirements of high-target-density scenarios.

[0121] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims. It should be understood that this disclosure is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. An azimuth estimation system, characterized in that, include: The storage unit is used to store the echo signal corresponding to the detected target received by the radar's antenna array into the DDR. A data extraction unit is used to extract the RDFFT data of the target from the echo signal stored in the DDR; A preprocessing unit is used to preprocess the RDFFT data; The frequency domain analysis unit is used to convert the time domain signal in the preprocessed RDFFT data into a frequency domain signal through FFT to generate the power spectrum of the target, extract the frequency domain features of the target from the power spectrum, and perform a first azimuth search on the frequency domain features in the power spectrum to identify a preset number of reference peak points. The first DBF unit is used to perform a second azimuth search based on the preset number of reference peak points to obtain low-precision azimuth information of the target. The second DBF unit is used to perform a third azimuth search based on the low-precision azimuth information to obtain the high-precision azimuth information of the target. The output unit is used to map the high-precision azimuth information to the real coordinate system to generate and output the final azimuth information of the target.

2. The azimuth estimation system as described in claim 1, characterized in that, The system also includes: A parallel control unit is used to execute preset steps in parallel and cyclically based on the number of echo signals using a Dataflow mechanism.

3. The azimuth estimation system as described in claim 2, characterized in that, The parallel control unit includes: The signal analysis module is used to analyze the number of echo signals; The scheduling decision module is used to determine, when the number of echo signals is greater than 1, to dynamically adjust the execution order and parallelism of the other processing units in the azimuth estimation system, excluding the storage unit, the output unit and the parallel control unit, using an adaptive scheduling algorithm. The feedback adjustment module is used to dynamically adjust the scheduling strategy based on the real-time execution status of each processing unit other than the storage unit, the output unit and the parallel control unit. The parallel execution module is used to execute preset steps in parallel and cyclically according to the scheduling strategy.

4. The azimuth estimation system as described in claim 1, characterized in that, The data extraction unit is specifically used to process the target echo signal using the CFAR algorithm, calculate the storage offset address of the target in the DDR to locate the storage location of the target's RDFFT data in the DDR, and extract the corresponding target's RDFFT data from the storage location.

5. The azimuth estimation system as described in claim 1, characterized in that, The preprocessing unit includes: The window function processing module is used to apply a window function to the echo signal data before preprocessing the RDFFT data, multiplying the time-domain signal of the echo signal with the window function value to obtain the data after window function processing. The antenna phase calibration module is used to perform phase compensation on the data processed by the window function according to the preset antenna calibration coefficients. The data interpolation and zero-padding module is used to rearrange the phase-calibrated and compensated data according to the actual geometric distribution of the radar's antenna array, and to perform interpolation or zero-padding operations on the rearranged data. The elevation angle compensation module is used to calculate the total angle value based on the geometric position of the radar's antenna array and the echo signal, and to use the total angle value as a compensation factor to correct the interpolated or zero-padded data to obtain the preprocessed RDFFT data.

6. An azimuth estimation method, implemented based on the azimuth estimation system of claim 1, characterized in that, Includes the following steps: S1 stores the echo signal received by the radar's antenna array corresponding to the detected target into DDR; S2, extract the RDFFT data of the target from the echo signal stored in the DDR; S3, preprocess the RDFFT data; S4. Based on the preprocessed RDFFT data, the time-domain signal in the RDFFT data is converted into a frequency-domain signal by FFT to generate the power spectrum of the target, and the frequency-domain features of the target are extracted from the power spectrum. The frequency-domain features are searched for a first azimuth angle in the power spectrum to identify a preset number of reference peak points. S5, perform a second azimuth angle search based on the preset number of reference peak points to obtain the low-precision azimuth angle information of the target; S6, perform a third azimuth search based on the low-precision azimuth information to obtain the high-precision azimuth information of the target; S7, map the high-precision azimuth information to the real coordinate system to generate and output the final azimuth information of the target; S8. Based on the number of echo signals, the S1-S8 steps are executed in parallel and cyclically using the Dataflow mechanism.

7. The azimuth estimation method as described in claim 6, characterized in that, Step S3 includes: S31, Before preprocessing the RDFFT data, a window function is applied to the echo signal data, and the time-domain signal of the echo signal is multiplied by the window function value to obtain the data after the window function processing; S32, perform phase compensation on the data processed by the window function according to the preset antenna calibration coefficient; S33, based on the actual geometric distribution of the radar's antenna array, rearrange the phase-calibrated and compensated data, and perform interpolation or zero-padding operations on the rearranged data; S34. The total angle value is calculated based on the geometric position of the antenna array and the received signal. The total angle value is used as a compensation factor to correct the interpolated zero-padded data to obtain the preprocessed RDFFT data.

8. The azimuth estimation method as described in claim 6, characterized in that, Step S4 includes: S41, Calculate the power spectrum of the target using FFT based on the preprocessed RDFFT data; S42, normalize the power spectrum and remove any possible noise interference; S43, In the normalized power spectrum, a peak detection algorithm is applied to find several peak points with the largest amplitude; S44, output the angular position and amplitude information of the several peak points as the starting point for subsequent refinement of azimuth angle estimation.

9. The azimuth estimation method as described in claim 6, characterized in that, Step S6 includes: S61, Use a high-precision algorithm to perform a fine azimuth search on the low-precision azimuth information to narrow the search range; S62, based on the narrowed search range, further adjust the search step size and precision to obtain the high-precision azimuth information.

10. The estimation method as described in claim 6, characterized in that, Step S8 includes: S81, Analyze the number of echo signals; S82, when the number of echo signals is greater than 1, an adaptive scheduling algorithm is used to dynamically adjust the execution order and parallelism of each processing unit except the storage unit, the output unit and the parallel control unit; S83, dynamically adjust the scheduling strategy based on the real-time execution status of each processing unit other than the storage unit, the output unit and the parallel control unit; S84, execute steps S1-S8 in parallel loop according to the scheduling strategy and Dataflow mechanism.

Images