CN-121996389-A - Multi-mode radar signal quick implementation method based on GPU multi-stream processing

Abstract

The invention discloses a multi-mode radar signal rapid implementation method based on GPU multi-stream processing, which is suitable for real-time signal processing of three radar modes of active, passive and active-passive mixing, and radar echo data is processed in parallel through a GPU multi-stream processing mechanism by depending on GPU parallel computing capability and cuFFT special acceleration libraries. The CPU host end builds a circular queue and realizes the non-interference parallel of the in-and-out queue by adopting multithreading, after the UDP data receiving, the frame checking and the mode classification are completed, the effective data is copied to the GPU video memory through the PCIE bus, the GPU end accelerates the core algorithm of each mode through the corresponding stream parallel, the active radar pulse compression, the moving target detection, the passive radar clutter cancellation, the distance Doppler processing and the like are respectively realized, the mixed mode synchronously starts the two types of stream processing, and finally the detection result is returned to the CPU and is displayed in real time through the UI interface. The method effectively improves the processing efficiency of the multi-mode radar signal, solves the problem of multi-mode data interference and processing delay, and meets the real-time processing requirement.

Inventors

• LUO YANGJING
• FENG BOHAN
• WANG HUI
• WU KEXIANG
• DENG ZHONGHUI
• Lu Yonghan

Assignees

• 贺州学院

Dates

Publication Date

20260508

Application Date

20260327

Claims (4)

1. 1. A multi-mode radar signal rapid implementation method based on GPU multi-stream processing is characterized in that a processing system is built based on a CPU+GPU heterogeneous computing architecture, wherein a CPU host end is configured with a multi-thread processing module, a circular queue module, a UDP data receiving module, a data frame checking module, a GPU stream control module and a PCIE data transmission module, a GPU equipment end comprises a multi-stream processing module, a cuFFT special acceleration library, an algorithm parallel processing module and a video memory storage module, and the processing system can be adapted to three modes of an active radar, a passive radar and an active and passive hybrid radar, and the method comprises the following steps: The CPU host end builds a circular queue according to a radar data frame format, adopts a multithreading technology to respectively distribute enqueuing operation and dequeuing operation of the queue to independent threads corresponding to different CPU cores for execution, realizes interference-free parallelism of enqueuing and dequeuing operations through the independent thread isolation, and improves the operation efficiency of the queue; Step 2, the CPU host computer side receives small data packets and high throughput UDP data efficiently through a multithreading technology, sends the received data into the circular queue in the step 1 to finish the enqueuing operation, and recovers the complete radar echo signal data frame through a checking mechanism of the circular queue; Step 3, when the cyclic queue executes dequeuing operation, firstly, byte-by-byte scanning is carried out on data in the queue, and after a preset frame head of the data is detected, data frame length verification and radar working mode verification are sequentially executed; The CPU host end realizes independent control of GPU multi-flow through a multi-thread mechanism, the operation tasks corresponding to the GPU flows are mutually decoupled, no computing resource is occupied and operation interference exists, after the CPU host end completes self-adaptive classification of multi-mode radar signals, effective data frames of active radar and passive radar are respectively distributed to corresponding data processing queues, and associated processing threads are awakened; Step 5, at the GPU equipment end, executing GPU parallel acceleration processing on a radar signal processing algorithm by using a GPU stream corresponding to a radar mode; Step 6, when the flow processing of the active radar echo data is carried out, pulse compression, moving target detection, moving target display and constant false alarm detection radar algorithm processing are realized in parallel through the GPU; Step 7, when the flow processing of the passive radar echo data is carried out, the clutter cancellation, the distance Doppler processing and the constant false alarm detection radar algorithm processing are realized in parallel through the GPU; and 8, performing GPU processing on the mixed signal data of the active radar and the passive radar, namely simultaneously controlling and starting the flow processing of the active radar in the step 6 and the flow processing of the passive radar in the step 7 through a CPU multithreading mechanism, completing parallel processing of corresponding radar signals, copying a processing result from a GPU video memory to a CPU memory, and displaying target information through a radar information processing platform interface terminal.
2. 2. The method for rapidly implementing multi-mode radar signals based on GPU multi-stream processing according to claim 1, wherein in the step 6, when the stream processing of the active radar echo data is performed, pulse compression, moving target detection, moving target display and constant false alarm detection radar algorithm processing are implemented in parallel through the GPU, and the specific implementation steps are as follows: 6-1) the CPU host end opens up a thread space according to the data dimension of a frame of echo signals, calls a pulse compression processing kernel function, and executes discrete convolution operation on the radar echo signals and the pulse compression coefficients stored in the GPU video memory to obtain a pulse compression result after GPU parallel processing ; 6-2) The CPU host uses m radar echo total numbers as rows, n sampling points of each echo as column opening thread space, and pulse compression results are obtained Caching pulse compression results in a two-dimensional matrix form for facilitating data cancellation And calling a moving target detection processing kernel function to compress the pulse result Performing subtraction and cancellation operation according to the rows to obtain a moving target display result processed by the GPU ; 6-3) CPU host end is the detection result of the moving target Distributing GPU video memory space, taking a pulse signal length as a sliding window size, and matching The moving target display result of each pulse pressure data in the same distance Doppler dimension Execution of (a) Point windowing FFT processing to obtain FFT results Then multithreading pairs through the GPU Performing matrix transposition, performing modular operation, and finally obtaining a moving target detection result processed by the GPU ; 6-4) Configuring parameters, GPU video memory space and thread space of constant false alarm detection at the CPU host end, starting the constant false alarm detection processing kernel function, and detecting the result of the moving target Parallel reduction summation according to rows; 6-5) adopting a sliding window to reduce and sum the unit to be detected, respectively extracting the reference unit data at the left end and the right end of the unit to obtain noise background signals through summation and average operation ; 6-6) Use of noise background signals Solving for decision threshold Comparison of And If the size of (a) Directly determining the position as a non-target detection point, and setting the detection result If (1) Determining the position as the target echo detection point and setting the detection result ; 6-7) The detection result Copying the video memory of the GPU to the CPU through the PCIE data transmission bus, and finally displaying the target through the radar information processing platform interface terminal to complete the real-time processing of the active radar signal.
3. 3. The method for rapidly implementing multimode radar signals based on GPU multi-stream processing according to claim 1, wherein in the step 7, when stream processing of passive radar echo data is performed, clutter cancellation, range-doppler processing and constant false alarm detection radar algorithm processing are implemented in parallel by the GPU, and specifically the implementation steps are as follows: 7-1) the CPU host end opens up a thread space by a frame of data dimension after the synchronization of the passive radar monitoring channel and the reference channel, adopts an ECA-B frequency domain clutter cancellation method to write and execute a processing kernel function, and detects channel signals stored in the GPU video memory And a reference channel signal Matrix multiplication is finished in batches through cuBLAS library functions, and clutter cancellation results after GPU processing are obtained ; 7-2) CPU host side The total number of the passive radar observation data segments is one row, and each data segment The number of sampling points opens up a thread space for a row, and the result of the cancellation of the clutter is obtained Buffer memory is in a two-dimensional matrix form for facilitating distance compression ; 7-3) CPU host allocation range Doppler processing Structure Then starting up the decimation filter kernel function to cancel clutter Filtering; 7-4) extracting the clutter cancellation result after filtering for each range bin with one observation data segment length as the sliding window size Data processing Point windowed FFT processing to obtain Doppler domain FFT result Then utilizing GPU to carry out multithreading Performing matrix transposition, performing parallel modulo square operation on the transposed result, compensating optional range variation correction phase, and finally obtaining range Doppler result ; 7-5) CPU sets the core parameters of the constant false alarm detection module, allocates the GPU video memory space and the thread space, then starts the constant false alarm detection processing kernel function, and obtains the range Doppler result According to two-dimensional window parallel reduction summation, avoiding the protection units around the unit to be measured through a sliding window, and carrying out summation average on the reference unit data around the reduced unit to be measured to obtain a background noise signal ; 7-6) Use of background noise signals Solving judgment threshold value by combining preset false alarm probability Comparison of And If the size of (a) Directly determining the position as a non-target detection point, and setting the detection result If (1) Determining the position as the target echo detection point and setting the detection result ; 7-7) The detection result Copying the video memory of the GPU to the CPU through the PCIE data transmission bus, and finally displaying the target through the radar information processing platform interface terminal to complete the real-time processing of the passive radar signal.
4. 4. The method for rapidly implementing multimode radar signals based on GPU multi-stream processing according to claim 3, wherein the CPU in step 7-5) sets the core parameters of the constant false alarm detection module as the reference unit number, the protection unit number and the false alarm probability.

Description

Multi-mode radar signal quick implementation method based on GPU multi-stream processing Technical Field The invention belongs to the technical field of radar signal processing, relates to real-time signal processing of three radar modes of active, passive and active-passive mixing, and particularly relates to a multi-mode radar signal rapid implementation method based on GPU multi-stream processing. Background The radar is used as core equipment in the fields of modern national defense, aerospace and civil monitoring, and the real-time performance and high efficiency of signal processing directly determine the target detection precision and response speed of a radar system. Along with the continuous development of radar technology, multimode working scenes such as active radars, passive radars and active and passive hybrid radars are increasingly common, various radar echo data have the characteristics of high throughput and high complexity, and strict requirements are put on the parallel computing capability of a signal processing architecture. At present, the traditional radar signal processing mostly adopts a CPU single-core or multi-core architecture, is limited by inherent defects of CPU serial calculation, is difficult to deal with the parallel processing requirement of multi-mode radar massive echo data, has the problems of high processing delay, low efficiency, insufficient resource utilization rate and the like, and especially has the problems of data interference of different radar modes when the radar system works in a mixed mode, and seriously influences the real-time target detection and recognition performance of the radar system. In order to solve the above problems, heterogeneous computing architecture is gradually applied to the field of radar signal processing, wherein cpu+gpu heterogeneous architecture is an important direction for improving processing efficiency by virtue of the powerful parallel computing capability of GPU. However, in the existing radar signal processing method based on the CPU and the GPU, most of the radar signal processing methods do not fully utilize a GPU multi-stream processing mechanism, processing tasks of all radar modes are mutually coupled, the problems of computing resource preemption, data transmission delay and the like are easy to occur, and an efficient data caching and checking mechanism is lacking, so that synchronous and rapid processing of multi-mode radar signals is difficult to realize. Meanwhile, in the prior art, most of signal processing algorithms aiming at active, passive and hybrid radars are designed independently, and a unified parallel acceleration scheme is lacked, so that the processing efficiency and the system compatibility are further reduced. Thus in order to improve the real-time processing and display capabilities of the radar under multimode signal echoes. According to the method, the data difference of radar echoes under different modes is utilized, the CPU multithreading technology is used for controlling the data of each circular queue, and meanwhile, the GPU multithreading processing mechanism and the efficient parallel computing capacity are adopted, so that the problems of processing delay, data interference, low resource utilization rate and the like are reduced, and meanwhile, the effective detection and real-time display of targets on the multimode radar signals are realized. Disclosure of Invention The invention aims to overcome the defects of processing delay and data interference in three radar modes of active, passive and active-passive mixed mode, and provides a multi-mode radar signal rapid implementation method based on GPU multi-stream processing. The method effectively improves the processing efficiency of the multi-mode radar signal, solves the problem of multi-mode data interference and processing delay, and meets the real-time processing requirement. The technical scheme for realizing the aim of the invention is as follows: A multi-mode radar signal rapid implementation method based on GPU multi-stream processing is disclosed, wherein a processing system is built based on a CPU+GPU heterogeneous computing architecture, a multi-thread processing module, a circular queue module, a UDP data receiving module, a data frame checking module, a GPU stream control module and a PCIE data transmission module are configured at a CPU host end, a GPU equipment end comprises a multi-stream processing module, a cuFFT special acceleration library, an algorithm parallel processing module and a video memory storage module, the processing system can be adapted to three modes of an active radar, a passive radar and an active and passive hybrid radar, and the method comprises the following steps: The CPU host end builds a circular queue according to a radar data frame format, adopts a multithreading technology to respectively distribute enqueuing operation and dequeuing operation of the queue to independent threads corresponding