CN104483670B - SAR (synthetic aperture radar) echo simulation method based on GPU (ground power unit) - Google Patents
SAR (synthetic aperture radar) echo simulation method based on GPU (ground power unit) Download PDFInfo
- Publication number
- CN104483670B CN104483670B CN201410718693.8A CN201410718693A CN104483670B CN 104483670 B CN104483670 B CN 104483670B CN 201410718693 A CN201410718693 A CN 201410718693A CN 104483670 B CN104483670 B CN 104483670B
- Authority
- CN
- China
- Prior art keywords
- kernel function
- echo data
- orientation time
- scattering point
- sar
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/40—Means for monitoring or calibrating
- G01S7/4052—Means for monitoring or calibrating by simulation of echoes
Abstract
The invention belongs to the technical field of radar signal processing, and particularly relates to an SAR (synthetic aperture radar) echo simulation method based on GPU (ground power unit). The SAR echo simulation method based on the GPU comprises the following steps of 1, setting working parameters of an SAR at the GPU; 2, establishing a coordinate system at the GPU by using the center of an SAR observing scene as a reference point, setting the position of each array element of the SAR in each azimuth time, and setting the coordinate and scattering coefficient of each scattering point in the SAR observing scene; 3, copying all data of a CPU (central processing unit) into a whole memory of the GPU; 4, using three Kernel functions to obtain the echo data of the SAR observing scene in the first azimuth time at the GPU, namely l=1, 2 until Na; 5, enabling l to be selected from 1, 2 until Na, and repeating the step 4, so as to obtain the echo data of the SAR observing scene in all bearing times.
Description
Technical field
The invention belongs to Radar Signal Processing Technology field, particularly to the synthetic aperture radar echo simulation based on gpu
Method, the present invention can be used for the echo simulation that airborne, missile-borne platform sar (synthetic aperture radar) is imaged.
Background technology
The research of sar imaging method needs the sar original echoed signals of specified conditions, and these data often cannot pass through radar
Carrier actual measurement obtains, such as missile-borne sar, therefore can obtain the required original echoed signals of imaging by emulation and have great meaning
Justice.
Echo simulation needs to carry out echo calculating to each scattering point of simulated scenario, and amount of calculation is huge, when echo generates
Between long.The Time-Domain algorithm operation efficiency of analogue echo is low, is not suitable for the simulation of large scene Area Objects echo data.
Content of the invention
It is an object of the invention to proposing synthetic aperture radar echo simulation method based on gpu, the present invention is using being based on
The frequency domain algorithm of concentric circular, further increases the performance of echo simulation, shortens simulation time.
For realizing above-mentioned technical purpose, the present invention adopts the following technical scheme that and is achieved.
Comprised the following steps based on the synthetic aperture radar echo simulation method of gpu:
Step 1, at cpu end, arranges the running parameter of synthetic aperture radar；The running parameter bag of described synthetic aperture radar
Include synthetic aperture radar orientation sampling number na；
Step 2, at cpu end, arranges sar observation scene, to set up coordinate system for reference point in the center that sar observes scene,
Setting each array element of sar, in the position of each orientation time, arranges sar and observes the coordinate of each scattering point and scattering system in scene
Number；
Step 3, all data duplications that cpu is arranged in step 1 and step 2 are in the global memory of gpu；
Step 4, at gpu end, using the 1st kernel function, the 2nd kernel function and the 3rd kernel function call
Go out the echo data that l-th orientation time sar observes scene, l=1,2 ..., na；
Step 5, makes l take 1,2 ..., n successivelya, and repeated execution of steps 4, draw all orientation time sar observation scenes
Echo data.
The invention has the benefit that invention applies multiple optimisation strategy, fully having excavated Area Objects echo simulation
High concurrency, ensure precision while, drastically reduce the area the generation time of echo.
Brief description
Fig. 1 is the flow chart of the synthetic aperture radar echo simulation method based on gpu of invention；
Fig. 2 a is the schematic diagram distributing thread grid in the present invention for the 1st kernel function；
Fig. 2 b is the schematic diagram distributing thread grid in the present invention for the 2nd kernel function；
Fig. 3 be the present invention in the 2nd kernel function each thread block corresponding data is carried out tree add up ask
The schematic diagram of sum；
Fig. 4 is the schematic diagram distributing thread grid in the present invention for the 3rd kernel function；
Fig. 5 a is the map of magnitudes of the scattering point two dimension echo data being drawn using the present invention in emulation experiment；
Fig. 5 b is the phase diagram of the scattering point two dimension echo data being drawn using the present invention in emulation experiment；
Fig. 6 a is Area Objects emulation original scene figure in emulation experiment；
Fig. 6 b is the Area Objects echo two dimensional amplitude figure being drawn using the present invention in emulation experiment；
Fig. 6 c is the Area Objects imaging results schematic diagram being drawn using the present invention in emulation experiment.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings:
With reference to Fig. 1, it is the flow chart of the synthetic aperture radar echo simulation method based on gpu of the present invention.Gpu should be based on
Synthetic aperture radar echo simulation method comprise the following steps:
Step 1, at cpu end, arranges the running parameter of synthetic aperture radar, the running parameter bag of described synthetic aperture radar
Include: synthetic aperture radar field angle, synthetic aperture radar distance are to sampling number (range cell number) nr, synthetic aperture radar side
Position is to sampling number na, synthetic aperture radar distance is to sample frequency fs, synthetic aperture radar distance is to sampling interval δ r, synthesis
Aperture radar distance is to oblique distance r of first sampling unit (first range cell)0, light velocity c, synthetic aperture radar transmitting letter
Number bandwidth b, the carrier wavelength lambda of synthetic aperture radar transmission signal.
Step 2, at cpu end, arranges sar observation scene, to set up coordinate system for reference point in the center that sar observes scene,
Each array element of sar is set in the position of each orientation time, the interval between scattering point in scene set, setting sar observation scene
In the coordinate of each scattering point and scattering coefficient.
Step 3, all data duplications that cpu is arranged in step 1 and step 2 are in the global memory of gpu；Now,
Need to be the data distribution global memory from cpu at gpu end in advance.
Step 4, at gpu end, using the 1st kernel function, the 2nd kernel function and the 3rd kernel function call
Go out the echo data that l-th orientation time sar observes scene, l=1,2 ..., na.
Show that l-th orientation time sar observes the process of the echo data of scene and include following sub-step:
(4.1) at gpu end, it is that the 1st kernel function distributes n1 thread block, Expression rounds up, and point_num represents l-th orientation time
Sar observes the number of all scattering points of scene kind, and n is the natural number setting.It (is each that each thread block comprises 256 threads
Size block_size of thread block is set to 256).So ensure that the total number of threads for the 1st kernel function distribution
Multiple for 256, unnecessary thread does not distribute task.Reference picture 2a, is to be that the 1st kernel function distributes thread in the present invention
The schematic diagram of grid；In Fig. 2 a, block (0,0), block (1,0), block (2,0) ... are expressed as the 1st kernel function and divide
The thread block joined, thread0 to thread255 represents 256 threads of each thread block.
Each thread of 1st kernel function draws the echo data of n scattering point of l-th orientation time, is calculating
During the echo data of l-th each scattering point of orientation time, calculate l-th orientation time corresponding scattering point place range cell
Echo data simultaneously carries out nrSecondary sinc interpolation, the echo data of corresponding scattering point place range cell is carried out kth time sinc and inserts
Data after value is designated as the kth time sinc interpolated data of corresponding scattering point, and k takes 1 to nr；The 1st sinc by corresponding scattering point
The n-th of the extremely corresponding scattering point of interpolated datarSecondary sinc interpolated data is combined as the echo data of corresponding scattering point.As can be seen that
The echo data of each scattering point is length is nrVector (include nrPoint data).
After each thread of the 1st kernel function draws the echo data of n scattering point of l-th orientation time,
Echo data to n scattering point of l-th orientation time is added up (the same distance list of n scattering point of l-th orientation time
The sinc interpolated data of unit is added up), draw the 1st range cell of the echo data of n scattering point of l-th orientation time
Accumulation result to the n-th of the echo data of n scattering point of l-th orientation timerThe accumulation result of individual range cell, by l
The accumulation result of the 1st range cell of the echo data of individual n scattering point of orientation time is to the n scattering of l-th orientation time
The n-th of the echo data of pointrThe accumulation result of individual range cell forms the echo data of n scattering point of l-th orientation time
Accumulation result.The accumulation result of the echo data of n scattering point of l-th orientation time is by nrThe vector of individual plural elements composition
(include nrPoint data),
In the embodiment of the present invention, according to preceding description, the number of the thread of the 1st kernel function use is Expression rounds up；The n scattering point that each thread of 1st kernel function draws
The accumulation result of echo data is n for lengthrVector (nrPoint data), nrRepresent synthetic aperture radar distance to sampling number.
After each thread of the 1st kernel function draws the accumulation result of echo data of n scattering point, it is stored in
In the global memory of gpu, therefore need to distribute n for the 1st kernel function in the global memory of gpur×point_num/n
Individual plurality of elements (each plurality of elements is used for storing complex data).
Obviously, it is that the thread grid that the 1st kernel function distributes is closely related with the size of n, in order to single thread
Distribute task as much as possible to improve the efficiency of the 1st kernel function and to consider the restriction of global memory, setting changes n to be made
The size of the accumulation result of the echo data of n scattering point that each thread of the 1st kernel function draws be limited in 1gb with
Under, the value of the n value of the maximum thus the found out n for setting.
(4.2) at gpu end, it is that the 2nd kernel function distributes n2 thread blockEach thread block for the 2nd kernel function distribution includes
256 threads.N2 thread block can be indicated with two-dimentional thread block grid, and the line number of two-dimentional thread block grid is nr, row
Number isReference picture 2b, is to divide for the 2nd kernel function in the present invention
The schematic diagram of distribution journey grid.In Fig. 2 b, block (0,0), block (1,0) ... block (0,1), block (1,1) ...,
block(0,nr),block(1,nr) ... it is represented as the 2nd kernel function and distribute n2 thread block；Thread0 is extremely
Thread255 represents 256 threads of each thread block.
256 × the n drawing for each thread block of the 1st kernel functionrIndividual plural elements (i.e. the 1st kernel
The accumulation result of the echo data of n scattering point of 256 l-th orientation time that each thread block of function draws), the 2nd
N is called in individual kernel functionrIndividual thread block；During l-th orientation drawing for each thread block of the 1st kernel function
Between k-th range cell of n scattering point 256 plural elements, in the 2nd corresponding n of kernel functionrMake in individual thread block
K-th of n scattering point of l-th orientation time each thread block of 1st kernel function being drawn with k-th thread block away from
Carry out the cumulative summation (parallel reduction summation) of tree from 256 plural elements of unit, draw l-th orientation time the 2nd
The corresponding n of individual kernel functionrThe cumulative summed result of k-th thread block in individual thread block, k takes 1 to nr.As can be seen that the 2nd
Individual kernel function draws the n2 cumulative summed result of l-th orientation time.
Specifically, the 2nd kernel function will draw to each thread block of the 1st kernel function simultaneously parallel
256×nrIndividual plural elements carry out the cumulative summation of tree, in the 2nd corresponding n of kernel functionrUse in individual thread block
256 × n that k-th thread block draws to the corresponding thread block of the 1st kernel functionrIndividual plural elements carry out tree and tire out
Plus the process of summation is: the kth of the n scattering point of l-th orientation time that the corresponding thread block of the 1st kernel function is drawn
The accumulation result of individual range cell reads shared drive from the global memory of gpu, then at (being made up of 256 plural elements)
In the shared drive of gpu, using the 2nd corresponding n of kernel functionrK-th thread block in individual thread block is to the 1st
The accumulation result tree of l-th orientation time, k-th range cell of n scattering point that the corresponding thread of kernel function draws
Shape structure adds up and sues for peace (parallel reduction summation).With reference to Fig. 3, it is each thread block pair of the 2nd kernel function in the present invention
Corresponding data carries out tree and adds up the schematic diagram of summation.Each thread of the corresponding thread block of the 2nd kernel function will
Corresponding plural elements read the shared drive of gpu from the global memory of gpu, after parallel reduction summation terminates, knot
First element that fruit is located in shared drive, is write accumulation result in the overall situation of gpu by No. 0 thread in corresponding thread block
Deposit.The parallel reduction summation of shared drive to be realized requires that the thread data number of each thread block is 2 integer power,
But last thread block in each range cell needs the block size that data to be processed is possibly less than in program in practice
256, but because, in the programming model of cuda single instrction multithreading, it also opens the shared drive of 256 plural elements simultaneously
And global memory can be read, therefore in order that the 2nd kernel function correctly executes, in thread receive data according to shared drive
Before, whether the sequence number of thread to be judged treats cumulative data amount check less than a range cell, reads if less than then this thread
To the relevant position of shared drive, otherwise to it, corresponding element in shared drive sets to 0 this thread global memory's respective element.
So, the 2nd kernel unit just can do any number of parallel reduction summation to the data of each range cell, and does not require
Data volume is 256 integral multiple.
(4.3) at gpu end, it is the 3rd kernel function distribution nrIndividual thread block, is the 3rd kernel function distribution
Each thread block includes 256 threads.With reference to Fig. 4, it is to be the signal that the 3rd kernel function distributes thread grid in the present invention
Figure；In Fig. 4, block (0,0), block (0,1) ... block (0, nr) it is expressed as the thread block that the 3rd kernel function distributes,
Thread0 to thread255 represents 256 threads of each thread block.
After calling the 2nd kernel function to realize partly the adding up of echo data, using the 3rd kernel function by the
The output result of 2 kernel functions is added up with shared drive again.In embodiments of the present invention, the 3rd kernel letter
K-th thread block of number reads k-th of the l-th orientation time that the 2nd kernel function draws from the global memory of gpu
The accumulation result (being made up of 256 plural elements) of range cell, l-th orientation then the 2nd kernel function being drawn
The cumulative summed result of k-th range cell of time carries out the cumulative summation (parallel reduction summation) of tree, draws correspondence
Cumulative summed result, k takes 1 to nr.3rd kernel function draws the n of l-th orientation timerIndividual cumulative summed result.Need
If it is noted that the 3rd kernel function treats that in same range cell cumulative data is more than 256, first by line
256 data in this range cell are write shared drive by journey block, and other all data of this range cell are followed with while afterwards
Ring is added on shared drive, so it is ensured that a thread block of the 3rd kernel function can realize a distance list
Metadata cumulative；If the 3rd kernel function treats that in a range cell cumulative data is less than 256, the 3rd
Kernel function, using adopting in a like fashion with the 2nd kernel function, is mended 0 to the unnecessary element of shared drive, then is entered
Row reduction is sued for peace.After the completion of 3rd kernel function execution, in nrN is produced on individual range cellrPoint echo.
In the embodiment of the present invention, the 1st kernel function, the 2nd kernel function and the 3rd kernel function are pressed respectively
Execute successively according to respective thread grid, complete Parallel Implementation on gpu and launch the process that pulsatile once receives echo, that is,
The superposition of all scattering point echoes in beam area.
Compare traditional analogue echoes algorithm, the present invention is when calculating the echo data of each scattering point it is only necessary to calculate
The echo of its place range cell simultaneously carries out multiple spot interpolation, greatly reduces amount of calculation.Draw l-th each scattering of orientation time
The process of the echo data of point includes following sub-step:
A whether () running parameter first according to synthetic aperture radar, judge corresponding scattering point in synthetic aperture radar
In beam area.Specifically, calculate the angle of strabismus of corresponding scattering point, if below the 1/2 of field angle, then correspond to scattering
Point is in the beam area of synthetic aperture radar；Otherwise, corresponding scattering point is not in the beam area of radar.
B () calculates instantaneous oblique distance r of corresponding scattering point in the beam area of synthetic aperture radar,
C (), with first range cell of synthetic aperture radar as reference unit, determines the beam area of synthetic aperture radar
Sequence number ind of interior corresponding scattering point place range cell,
D () determines interpolation length 2l+1 of sinc interpolation, it is higher that 2l+1 typically chooses the bigger interpolation precision of 40～100, l,
But the efficiency (speed) calculating each scattering point echo data also accordingly reduces
E () judges corresponding scattering point whether in interpolation sampling unit, if l+1≤ind≤nr- l, then illustrate corresponding dissipating
Exit point, in interpolation sampling unit, otherwise, illustrates corresponding scattering point not in interpolation sampling unit.As l+1≤ind≤nr- l,
Skip to sub-step (f).
F () represents the sequence number of the interpolation point in 2l+1 point interpolation with m, the span of m is-l≤m≤l.Draw corresponding dissipating
The echo data of m-th interpolation point of exit point, the echo data of m-th interpolation point of i-th scattering point is expressed as si(ind+
M), as-l≤m≤l and 1≤ind+m≤nrWhen, si(ind+m) it is:
Otherwise, si(ind+m) it is 0.σiRepresent the scattering coefficient of i-th scattering point, c represents the light velocity；R (i) represents i-th
The oblique distance of scattering point.
By si(ind+m) nrIndividual non-zero valued combinations become the echo data s of i-th scattering pointi.
Above-mentioned sub-step (a) is all completed by the 1st kernel function to sub-step (f).1st kernel function is not only sentenced
Break each scattering point whether in the wave beam of synthetic aperture radar, calculates oblique distance and the echo data of scattering point, it is right to also achieve
Partly the adding up of echo data.Each thread will execute two-layer circulation: calculate and in global memory cumulative n scattering point time
Ripple, a scattering point needs to carry out 2l+1 point interpolation.
2nd kernel function, the 3rd kernel function obtain each range cell using parallel reduction summation method
Echo data, that is, the output result tree of the 1st kernel function is added up.Every in 2nd kernel function
Individual distance is added up to the echo data of sample point with multiple thread block, obtains multiple values.Every in 3rd kernel function
Individual distance is collected echo result for a value to sample point by a thread block, obtains nrIndividual distance returning to sample point
Ripple.After 1st kernel function, the 2nd kernel function and the 3rd kernel function execute successively, when obtaining an orientation
Between nrPoint echo.
The present invention is optimized to them with following multiple optimisation strategy.1st kernel function, the 2nd kernel letter
The thread block size of number and the 3rd kernel function is disposed as 256, can improve the occupancy of stream handle (sm) thread block
Rate, the shortest through testing program execution time when thread block size is 256.A thread due to the 1st kernel function will
Execution two-layer circulation, calculates more complicated, therefore optimizes the use to depositor first, to reduce unnecessary double counting and right
The read-write of global memory, improves the performance of the 1st kernel function.But the 1st single thread of kernel function needs simultaneously
Register variable is more to also limit the thread block number simultaneously executing in sm, so reusing the 1st kernel as far as possible
Register variable in function, reduces the use unnecessary to depositor.In order to improve the degree of parallelism of instruction in the program of gpu end,
The present invention reads real imaginary part with two independent instructions from global memory, effectively increases the performance of program.gpu
On complex operation not using Inline Function but real part and imaginary part are respectively calculated, program feature improves further.Will
After realizing the while loop unrolling of tree-shaped parallel reduction summation process, the 2nd kernel function and the 3rd kernel function
Execution time reduces further.
Step 5, by the 3rd n that kernel function drawsrIndividual cumulative summed result is designated as l-th orientation time sar observation
The echo data of scene, makes l take 1,2 ..., n successivelya, and repeated execution of steps 4, draw all orientation time sar observation fields
The echo data of scape, the echo data that all orientation time sar are observed scene writes in the global memory of gpu.All orientation
Time sar observes the echo data of scene and observes the echo data of scene to n-th by the 1st orientation time saraThe individual orientation time
Sar observes the echo data composition of scene.It can be seen that the global memory of gpu needs to distribute na×nrIndividual plurality of elements is storing
The echo data of all orientation time.
Wherein,It is apart from fast time, tmIt is the orientation slow time, rsIt is oblique distance during the inswept target of beam central line, r (tm；
rs) it is instantaneous oblique distance, wr() represents the window function of transmission signal, wa() represents orientation window function.Represent with
Apart from the fast timeOrientation slow time tmAnd oblique distance r during the inswept target of beam central linesCorresponding echo data.
Step 6, at gpu end, the echo data observing scene to all orientation time sar does distance to Fourier transformation
(fft), obtain the echo data apart from frequency domain all orientation time；Echo data apart from frequency domain all orientation time is multiplied by
Apart from frequency matching item, draw the echo data of all orientation time after frequency matching；Adjust the distance all after frequency matching
The echo data of orientation time makees distance to inverse Fourier transform (ifft), draws all orientation time after frequency matching
Time domain echo data, so far just can realize the generation of the echo data of large scene.Due to orientation same in global memory
The echo data of sample point is continuous, so when the echo data to all orientation time makees distance to Fourier transformation, right
In the echo data of all orientation time, the echo data of each orientation time is nrPoint distance is to Fourier transformation.Cuda is built-in
Fft built-in function can once realize the fft batch processing of multirow.
Distance realizes echo data and the multiplication apart from frequency matching item by calling gpu kernel function to after fft.Distance frequency
Domain occurrence is a distance to nrPoint vector, because amount of calculation is less, copies to gpu end after calculating on cpu, then by parallel
Each thread taken advantage of again to the element in occurrence to by each sampling unit of echo data and distance by distance.
After the time domain echo data of all orientation time after drawing apart from frequency matching, it is institute after frequency matching
Have the time domain echo data of orientation time allocation space on the internal memory at cpu end, will after frequency matching all orientation time
Global memory from gpu for the time domain echo data copy to cpu end, write file preserves.
The effect of the present invention can be further illustrated by following emulation experiment.
In emulation experiment, echo simulation is carried out using the following two kinds method, method one is the present invention, method 2 is only sharp
Carry out echo simulation with cpu.The software platform being related in emulation, cpu, gpu are as shown in the table:
| Software platform | Visual studio 2010, cuda toolkit v5.5 |
| cpu | intel x5650 |
| gpu | tesla c2070 |
Reference picture 5a, is the map of magnitudes of the scattering point two dimension echo data being drawn using the present invention in emulation experiment.
Reference picture 5b, is the phase diagram of the scattering point two dimension echo data being drawn using the present invention in emulation experiment.Fig. 5 a
In Fig. 5 b, transverse axis represents the sequence number in orientation for the sampling unit, the longitudinal axis represent sampling unit distance to sequence number, from 5a and
Fig. 5 b finds out, linear FM signal is positive frequency sweep, and phase of echo shows as two-dimentional hyp form, and practical situation is coincide.
Respectively the scene comprising 512*512 and 2000*3000 scattering point is emulated, angle of strabismus is 50 degree, reference
Fig. 6 a, is Area Objects emulation original scene figure in emulation experiment, reference picture 6b, is the face being drawn using the present invention in emulation experiment
Target echo two dimensional amplitude figure, reference picture 6c, is the Area Objects imaging results schematic diagram being drawn using the present invention in true experiment.Figure
In 6b, transverse axis represent sampling unit orientation sequence number, the longitudinal axis represent sampling unit distance to sequence number.From Fig. 6 c
Find out, the Area Objects imaging results of the present invention ensure that good focusing effect.Simulation time and speed-up ratio are as shown in the table
| Scene scatters are counted | Cpu version takes (s) | This law takes (s) | Speed-up ratio |
| 512*512 | 55701 | 138 | 404 |
| 2000*3000 | 1127424 | 1599 | 705 |
In sum, realization on cpu for the present invention is one three layers circulation, and outermost loop number is sampled for orientation
Points, the inside two-layer circulates the echo for calculating the generation of pulsatile once all scattering points, subsequently does distance to echo and takes advantage of to fft
With apart from frequency matching item be again ifft become back to the time domain, the step for carry out in outermost loop.The present invention is by cpu version
This inside two-layer circulation, that is, this process of echo that calculating radar emission pulsatile once receives is real parallel on gpu
Existing, with large-scale thread, the echo that each scattering point produces is realized reduction summation, obtain the echo of this orientation time simultaneously
Write global memory.The circulation of the outermost orientation sampled point kernel function to call gpu end is executed on cpu, obtains
The echo data of all orientation time.
Obviously, those skilled in the art can carry out the various changes and modification essence without deviating from the present invention to the present invention
God and scope.So, if these modifications of the present invention and modification belong to the scope of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to comprise these changes and modification.
Claims (2)
1. the synthetic aperture radar echo simulation method based on gpu is it is characterised in that comprise the following steps:
Step 1, at cpu end, arranges the running parameter of synthetic aperture radar；The running parameter of described synthetic aperture radar includes closing
Become aperture radar bearing to sampling number na；
Step 2, at cpu end, arranges sar observation scene, to set up coordinate system for reference point in the center that sar observes scene, setting
Each array element of sar, in the position of each orientation time, arranges sar and observes the coordinate of each scattering point and scattering coefficient in scene；
Step 3, all data duplications that cpu is arranged in step 1 and step 2 are in the global memory of gpu；
Step 4, at gpu end, draws l using the 1st kernel function, the 2nd kernel function and the 3rd kernel function
The echo data of individual orientation time sar observation scene, l=1,2 ..., na；
The running parameter of described synthetic aperture radar also includes synthetic aperture radar distance to sampling number nr；In step 4, obtain
The process going out the echo data that l-th orientation time sar observes scene includes following sub-step:
(4.1) at gpu end, it is that the 1st kernel function distributes n1 thread block, Expression rounds up, and point_num represents the number of l-th orientation time sar observation all scattering point of scene kind, and n is to set
Fixed natural number, is that each thread block of the 1st kernel function distribution comprises 256 threads；
Draw the echo data of n scattering point of l-th orientation time using each thread of the 1st kernel function, calculating
During the echo data of l-th each scattering point of orientation time, calculate l-th orientation time corresponding scattering point place range cell
Echo data simultaneously carries out sinc interpolation, and the echo data of corresponding scattering point place range cell is carried out after kth time sin c interpolation
Data be designated as the kth time sinc interpolated data of corresponding scattering point, k takes 1 to nr；The 1st sinc interpolation by corresponding scattering point
The n-th of the extremely corresponding scattering point of datarSecondary sinc interpolated data is combined as the echo data of corresponding scattering point；
After each thread of the 1st kernel function draws the echo data of n scattering point of l-th orientation time, to l
The echo data of individual n scattering point of orientation time is added up, and draws the echo data of n scattering point of l-th orientation time
The accumulation result of the 1st range cell is to the n-th of the echo data of n scattering point of l-th orientation timerIndividual range cell tired
Plus result, by the accumulation result of the 1st range cell of the echo data of n scattering point of l-th orientation time to l-th orientation
The n-th of the echo data of n scattering point of timerThe accumulation result of individual range cell forms n scattering point of l-th orientation time
The accumulation result of echo data；During drawing the accumulation result of echo data of n scattering point of l-th orientation time,
Each thread block of 1st kernel function draws 256 × nrIndividual plural elements；
(4.2) at gpu end, it is that the 2nd kernel function distributes n2 thread block
Each thread block for the 2nd kernel function distribution includes 256 threads；
256 × the n drawing for each thread block of the 1st kernel functionrIndividual plural elements, in the 2nd kernel function
Call nrIndividual thread block；The n scattering point kth of l-th orientation time drawing for each thread block of the 1st kernel function
256 plural elements of individual range cell, in the 2nd corresponding n of kernel functionrUsing k-th thread block in individual thread block
256 of l-th orientation time, k-th range cell of n scattering point that each thread block of 1st kernel function is drawn
Plural elements carry out parallel reduction summation, draw the 2nd corresponding n of kernel function of l-th orientation timerIn individual thread block
The cumulative summed result of k thread block；
(4.3) at gpu end, it is the 3rd kernel function distribution nrIndividual thread block, is each line of the 3rd kernel function distribution
Journey block includes 256 threads；
Call the 2nd kernel function realize echo data partly add up after, using the 3rd kernel function by the 2nd
The output result of kernel function is added up with shared drive again；K-th thread block of the 3rd kernel function is from gpu
Global memory in read the accumulation result of k-th range cell of l-th orientation time that the 2nd kernel function draws,
Then the cumulative summed result of k-th range cell of the l-th orientation time that the 2nd kernel function draws is carried out parallel
Reduction is sued for peace, and draws corresponding cumulative summed result；The n of the l-th orientation time that the 3rd kernel function drawsrIndividual cumulative
Summed result is the echo data of l-th orientation time sar observation scene；
Step 5, makes l take 1,2 ..., n successivelya, and repeated execution of steps 4, show that all orientation time sar observe returning of scene
Wave number evidence.
2. the synthetic aperture radar echo simulation method based on gpu as claimed in claim 1 it is characterised in that step 5 it
Afterwards, at gpu end, the echo data observing scene to all orientation time sar does distance to Fourier transformation, obtains apart from frequency domain
The echo data of all orientation time；Echo data apart from frequency domain all orientation time is multiplied by apart from frequency matching item, obtains
Go out the echo data of all orientation time after frequency matching；Adjust the distance the echo data of all orientation time after frequency matching
Make distance to inverse Fourier transform, draw the time domain echo data of all orientation time after frequency matching.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410718693.8A CN104483670B (en) | 2014-12-01 | 2014-12-01 | SAR (synthetic aperture radar) echo simulation method based on GPU (ground power unit) |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410718693.8A CN104483670B (en) | 2014-12-01 | 2014-12-01 | SAR (synthetic aperture radar) echo simulation method based on GPU (ground power unit) |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104483670A CN104483670A (en) | 2015-04-01 |
| CN104483670B true CN104483670B (en) | 2017-02-01 |
Family
ID=52758235
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410718693.8A Expired - Fee Related CN104483670B (en) | 2014-12-01 | 2014-12-01 | SAR (synthetic aperture radar) echo simulation method based on GPU (ground power unit) |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104483670B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105759266B (en) * | 2016-02-29 | 2017-11-21 | 西安电子科技大学 | A kind of SAR radar return acquisition methods based on graphics processing unit |
| CN107229051B (en) * | 2017-05-26 | 2020-05-01 | 西安电子科技大学 | GPU-based video SAR echo simulation parallel implementation method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101526614A (en) * | 2009-04-03 | 2009-09-09 | 北京理工大学 | SAR echo rapid simulation method based on sub-aperture and equivalent scatterer |
| CN103630891A (en) * | 2013-12-03 | 2014-03-12 | 西安电子科技大学 | Method for estimating incoming wave directions of targets in radar based on external illuminators by aid of GPU (graphics processing unit) |
-
2014
- 2014-12-01 CN CN201410718693.8A patent/CN104483670B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101526614A (en) * | 2009-04-03 | 2009-09-09 | 北京理工大学 | SAR echo rapid simulation method based on sub-aperture and equivalent scatterer |
| CN103630891A (en) * | 2013-12-03 | 2014-03-12 | 西安电子科技大学 | Method for estimating incoming wave directions of targets in radar based on external illuminators by aid of GPU (graphics processing unit) |
Non-Patent Citations (4)
| Title |
|---|
| Evalution of angular interpolation kernels in fast back-projection SAR processing;Frolind et al.;《IEEE Proceedings-Radar,Sonar and Navigation》;20060630(第3期);第243-249页 * |
| GPU环境下几种并行图像处理算法的设计与实现;孙东;《中国优秀硕士学位论文全文数据库_信息科技辑》;20140615(第6期);I138-705 * |
| 基于CPU的SAR回波信号模拟的研究与实现;李博;《中国优秀硕士学位论文全文数据库_信息科技辑》;20141115(第11期);正文第46页第4段、第51页-第52页第5.2.2节，图4.4、图5.12 * |
| 基于CUDA的SAR仿真系统设计;喻勤;《中国优秀硕士学位论文全文数据库_信息科技辑》;20110315(第3期);I136-1581 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104483670A (en) | 2015-04-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104459666B (en) | Missile-borne SAR echo simulation and imaging method based on LabVIEW | |
| CN104459693A (en) | Missile-borne SAR forward-squint imaging method based on GPU | |
| CN105911532B (en) | Synthetic aperture radar echo Parallel Simulation method based on depth collaboration | |
| CN104833974B (en) | The SAR Imaging fasts rear orientation projection method of compression is composed based on image | |
| CN111383741B (en) | Method, device and equipment for establishing medical imaging model and storage medium | |
| CN104007419B (en) | About residence time and the radar time resource combined distributing method of heavily visiting interval | |
| CN108008389B (en) | GPU-based rapid frequency domain back projection three-dimensional imaging method | |
| CN108957450A (en) | A kind of millimetre-wave radar GPU real time three-dimensional imaging method | |
| CN104766280A (en) | Quality map phase unwrapping method based on heap sort | |
| CN109683147A (en) | Real-time Generation, device and the electronic equipment of random pulse stream signal | |
| Liu et al. | Parallel processing of massive remote sensing images in a GPU architecture | |
| CN105137402B (en) | Airborne foresight scanning radar parallel processing method based on GPU | |
| CN106199524B (en) | Far field Broadband RCS data acquisition and the compression method of denoising are tracked based on base | |
| CN109343060A (en) | ISAR imaging method and system based on deep learning time frequency analysis | |
| CN113985408B (en) | Inverse synthetic aperture radar imaging method combining gate unit and transfer learning | |
| CN104483670B (en) | SAR (synthetic aperture radar) echo simulation method based on GPU (ground power unit) | |
| CN109409442A (en) | Convolutional neural networks model selection method in transfer learning | |
| CN109991602A (en) | ISAR image resolution enhancement method based on depth residual error network | |
| CN107589421A (en) | A kind of array Forward-looking SAR imaging method | |
| CN106802409A (en) | External illuminators-based radar real-time signal-processing method based on multi -CPU treatment | |
| CN106646664A (en) | GPU-based echo simulation method and system of microwave of human body | |
| CN109343059A (en) | A kind of Nonlinear Orthogonal FM signal generation method and device | |
| CN105891826A (en) | Airborne radar fast maximum posteriori imaging method | |
| CN105068071B (en) | A kind of fast imaging method based on backprojection operator | |
| Zhang et al. | GPU-based parallel back projection algorithm for the translational variant BiSAR imaging |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170201 Termination date: 20171201 |