A Fast SAR Raw Data Simulation Algorithm Based On

Data Correction Using Chirp Scaling Algorithm

Q.Gao1,Y.H. Xu1, T. Zeng1, Z.G. Ding1
1
Key Laboratory of Electronic and Information Technology in Satellite Navigation(Beijing Institute
of Technology), Ministry of Education, School of Information and Electronics, Beijing Institute of
Technology, Beijing, China.
[email protected],[email protected],[email protected], [email protected]
Keywords:SyntheticAperture Radar(SAR), raw datasimu-
lation, chirp scalingalgorithm (CSA), data correction 2 SAR echomodel
Abstract The SAR echo signal model is the fundamental of SAR
image process.
In this paper, a fast SAR raw echo simulationalgorithm Taking a point targetfor example; the SAR echo equa-
based on data correction using chirp scalingalgorithm is tion can be written as 2.1.
proposed.It simulates the new raw data thought correcting s0 (W ,K ) Ai wr (W  2 R(K ) / c) wa (K  Kc )
original raw datawhich has different range and veloci-
tybetween radar and target.Firstly, it corrects the range
u exp{ j 4S f 0 R(K ) / c} 2.1
and azimuth phase errorbetween the original data and new u exp{ jS K r (W  2 R(K ) / c) 2 }
data. Then,the range cell migration is modified using chirp Where
scaling algorithm. Finally, theeffect of theproposedmethod
isanalyzed.And the approach in this paper has a significant R(K ) R02  v 2W 2 2.2
meaning for the investigation of SAR echo simulation. In 2.1,Ai represents the backscattering coefficient of
1 Introduction point target, f 0 represents the carrier frequency, R(K )
represents the instantaneous distance between target and
The research on SAR echo simulation technology is the moving platform, R0 represents the distance between
veryvaluable for improving SAR systems and amending
SAR signalprocessing algorithms.With the development target and the moving platform in zero time, K r repre-
of SAR technology,theSARecho simulation algorithm has sents the range FM rate of chirp signal, W and K repre-
rapidly developed.
The existing simulationmethods can be roughly catego-
sents the fast time and theslow time, wr and wa repre-
rized into three groups.The first group works in the time sents the range envelope and the azimuth envelope.
domain [1]. These methods simulate the echo generation 2 R0
srd (W , fK ) Awr {W  }Wa ( fK  fKc )
process accurately, and it can create ahigh-precision raw cD( fK ,Vr )
data. However, these methodshave a low computational
4S f 0 R0 D( fK ,Vr ) 2.3
efficiency. Usually, it is used to simulate point targets. u exp{ j }
The second group operates in the two-dimensional fre- c
quency domain[2], and it uses 2D-FFT to achieve signal 2 R0
convolution,so it could significantly reduce the amount of
u exp{ jS K m [W  ]2 }
cD( fK ,Vr )
computation,and obtain a higheraccuracy.But thesemeth-
Where
ods require more memory space. The third group simu-
latesSAR raw data inhybrid domain [3][4], and it can also c 2 fK2
acquiregoodprecision, but it can’t avoid the time-domain D( fK ,Vr ) 1 2.4
integraloperation, and it still has a low computational effi-
4Vr2 f 02
ciency. Kr 2.5
Km
Usually, we need to simulate massivesimilar SAR raw cR0 fK2
data. They have the same pulse repetition frequency (PRF), 1  Kr
bandwidth, pulse width, sample rate, and the main differ- 2Vr2 f 03 D3 ( fK ,Vr )
enceis the range and velocity between radar and target. The SAR echo equation in Range-Doppler domain can
For example, somespaceborne SARdata is simulated in be written as 2.3
sameorbit atdifferent time. This paper proposes anew SAR In 2.3, D( fK ,Vr ) represents the migration parameter,
rawdata simulation algorithm based on data correctionus-
ing chirp scaling algorithm (DCA).The method corrects fK represents the azimuth frequency, and K m represents
the data in frequency domain, and has a high efficiency. the new range FM rate.

Authorized licensed use limited to: ShanghaiTech University. Downloaded on November 26,2024 at 07:44:15 UTC from IEEE Xplore. Restrictions apply.
 From the 2.3,we can see that the difference between 4S f 0 R01 D( fK ,Vr1 )
originaldata and new data is the migration parameter S rdc 1 (W , fK )=S rd 1 (W , fK ) u exp{ j }
c
D( fK ,Vr ) , and it influences the range envelope, azimuth 4S f 0 R02 D( fK ,Vr 2 )
u exp{ j }
phase,and range phase. The aim of the newsimulation c
methodis correcting the migration parameter D( fK ,Vr ) . 2 R01
u exp{- jS K m [W  ]2 } 3.1
cD( fK ,Vr1 )
3 Fast SAR raw datasimulationalgorithm 2 R02
u exp{ jS K m [W  ]2 }
cD( fK ,Vr 2 )
3.1 Algorithmflow chart 2 R01
Awr {W  }Wa ( fK  fKc )
The Whole simulation process mainly includes four cD( fK , Vr1 )
times Fourier Transformationand three times complex
4S f 0 R02 D( fK ,Vr 2 )
multiplication.The simulation process is shown as fig 2 u exp{ j }
c
Old SAR
2 R02
Raw Echo u exp{ jS K m [W  ]2 }
cD( fK ,Vr 2 )
Azimuth Fourier In 3.1, R01 and R02 representthe minimum range be-
Transform
tween radar and target in the originalecho data and new
Differential
Phase
RCM echo data, and V01 , V02 represent the image velocity.
Correct
factor Correct
factor 3.3 Range migration error correction
Range Fourier Transform Because of the different ranges between radar and tar-
get, the range cell migration trajectory is also different
Bulk
RCM between originaldata and new data. Usually the maximum
Correct range cell migration error is more than one resolution
factor cell,so it must be compensated. The error equation can be
2D Fourier Transform
written as 3.2.
'Rerror ( R1  R10 )  ( R2  R20 )
New SAR ( R 2 +(V K ) 2  R ) 3.2
01 r1 01
Raw Echo
 ( R +(Vr 2K )  R02 )
2
02
2

Fig 1algorithm flow chart The equation3.2 can be expanded with Tay-
Firstly, the phase error of theoriginalecho datais cor- lor’sformula,and ignores the higher order terms.The er-
rected in the Range-Doppler domain. Then, the range cell rorequationcan be approximatelywritten as 3.3.
migration (RCM) is corrected using the chirp scaling algo- Vr22K 2 Vr22K 2
rithm. The RCM error correction is divided into two steps.
'Rerror |  3.3
2 R01 2 R02
The differential RCM error is corrected in Range-Doppler
From 3.3, It can be seenthat the error is related to the
domain, and the bulk RCM error is corrected in the two-
range, platform velocity and azimuth time. The scene
dimensional frequency domain. Finally, it transforms the
width is approximately equal to the productof velocity and
data fromtwo dimensional frequency domains to time do-
azimuth time.The table 1 shows the maximumRCM er-
main.
rorof spaceborne SAR in different ranges and scene width.
3.2 Phase error correction Table 1 the maximum RCM error
R01(Km) R02(Km) Width (Km) Error(m)
The phase is corrected in the Range-Doppler domain,
42000 43000 500 17.23
andit includes azimuth phase error and range phase error.
42000 43000 200 2.76
The error is different ineachrange cell and azimuth, so the
azimuth phase factor and range phase factor arecalculated 42000 43000 100 0.69
intwo dimensions. The phase error factor can be written as 600 620 100 65.85
3.1. 1000 1020 100 24.09
2000 2020 100 6.09
When the maximum RCM error is more than half of
the resolution cell, it must be corrected. Inthiscase, the
range migration error is corrected using the chirp scaling
algorithm (CSA) [5][6]. According to the CSA, RCM can
be dividedinto two parts, the bulk range cell migration
RCM bulk ( fK ) and the differential range cell migration
RCM diff ( R0 , fK ) .

Authorized licensed use limited to: ShanghaiTech University. Downloaded on November 26,2024 at 07:44:15 UTC from IEEE Xplore. Restrictions apply.
 R0 R0 3.4 N real 3 N a _ map N r _ map  N a _ fft  5 N r _ fft N a _ fft
RCM total ( R0 , fK ) 
D( fK ,Vr ) D( fKref ,Vr ) N cmp 3 2 N a _ fft N r _ fft  N r _ fft N a _ fft 3.13
Rref R0 (log 2 N r _ fft  log 2 N a _ fft )
RCM bulk ( fK )  3.5
D( fK ,Vrref ) D( fKref ,Vrref ) z Data correct algorithm
RCM diff ( R0 , fK ) RCM tital ( R0 , fK ) N real 2(3N a _ raw N r _ raw  2 N a _ raw N r _ raw
3.6
 RCM bulk ( fK )  4 N a _ raw N r _ raw  3N a _ raw N r _ raw )
Firstly, the differential RCM error is corrected in the N cmp 3N a _ raw N r _ raw 3.14
Range-Doppler domain. The chirp scalingequation is
shown as 3.7.  N a _ raw N r _ raw log 2 N r _ raw
D( fKref ,Vrref )
H diff (W , fK ) exp( jS K m [  1]W 2 ) 3.7  N r _ raw N a _ raw log 2 N a _ raw
D( fK ,Vrref )
Inequation 3.11-3.14, N a _ raw represents the number of
The compensationequation is shown as 3.8.
Srdcc 1 (W , fK ) Srdc 1 (W , fK )( H diff 1  H diff 2 ) 3.8 the range line, N r _ raw represents the number ofthe sam-
Secondly,the bulk RCM error in two-dimensional fre- pling points for each line, N r _ map and N a _ map represent
quency domain is corrected. The chirp scaling equation is
shown as 3.9. the number of the targets in range and azimuth, N r _ fft and

4S 1 1 N a _ fft represent the number of Fourier Transformation in

H bulk ( fW , fK ) exp( j [  ]Rref fW ) 3.9
c D( fK ,Vrref ) D( fKref ,Vrref ) range and azimuth, N real represents the number of the
The compensationequation is shown as 3.10. real multiplication, N cmp represents the number of the
Srdcc1 ( fW , fK ) Srdc 1 ( fW , fK )( Hbulk1  Hbulk 2 ) 3.10 complex multiplication.
10
x 10 Time Domain
Finally, the range cell migration is corrected, and the 6
new data can be generated by transforming the 5
Srdccc1 ( fW , fK ) into two-dimensional time domain.
Computation

3.4 Computationcomplexity 3

The computation complexity of the original SAR raw 2

echo simulation algorithmsis related to the target quantity, 1

0 1 2 3 4 5 6 7 8 9 10
but thedata correction algorithm(DCA)is only relatedto Target Quantity 5
x 10
raw data dimension.The paper [7] analysesthe computa-
tion complexityof the originalalgorithm.In the data correc- Fig 2 the computation of time domain
tion algorithmthere are mainly four times Fourier Trans-
formation and three times complex multiplication. The x 10
14
Range Frequency Domain
computation complexity of different algorithm is showed 12

10
as follows.
8
z Time domain algorithm
Computation

2 f sW  N a _ raw N map (1  N r _ map )

6
N real1 4

1 2

N cmp1 N r _ fft log 2 N r _ fft (1  2 N a _ raw ) 3.11 0

2 0 10 20 30 40 50
Target Quantity
60 70 80 90 100

 N a _ raw ( N r _ fft  N r _ map N a _ map ) Fig 3the computation of range frequencydomain

z Range frequency domain coherent algorithm x 10
11
2D FFT

2 f sW  N a _ raw N map (1  N r _ map )

1.4281
N real 2 1.4281

 2 N r _ fft N r _ map N a _ raw 1.4281

Computation

1.4281

1 3.12
N r _ fft log 2 N r _ fft (1  2 N a _ raw )
1.4281
N cmp 2
2 1.4281

1.4281
 N a _ raw N r _ fft (1  N r _ map N a _ map ) 0 1 2 3 4 5
Target Quantity
6 7 8 9 10
5
x 10

z Two-dimensionalfrequency domainalgorithm Fig 4the computation of 2D FFT

Authorized licensed use limited to: ShanghaiTech University. Downloaded on November 26,2024 at 07:44:15 UTC from IEEE Xplore. Restrictions apply.
 last,this paper compares the data generated by original-
9
x 10 Data Correction
9.3952
method and the datagenerated by the new method in Fig
9.3952 8). The relation coefficient of echodata is more than 0.9,
Computation

and the image correction coefficient, mean and variance

9.3952
are nearly the same. From the two pictures we can see that
9.3952 the data correction algorithmhasa higher accuracy.

9.3952
0 2 4 6 8 10 5 Conclusions
Target Quantity 5
x 10

Fig 5the computation ofData Correction SAR echo data simulation is an important tool for de-
signing SAR systems and amending SAR signalpro-
The computation complexities of different algorithms cessing algorithms..This paper introduces a SAR raw-
areshown in Fig3-Fig6. From the Fig it can beseen that the datasimulationbased on data correctionusing chirp scaling
computation complexity of data correction algorithm re- algorithm, which greatly reduces the computationalcom-
ducessignificantly. ( Nr _ raw Na _ raw 4096 , plexity.With a higher accuracy .It is also found that the
N Nreal  4 Ncmp ) improved echo simulation method has a better fidelity
than traditional methods.The improved echo simulation
has been verified by simulation, and can be widely applied
4 Simulation result in SAR simulation.

In order to verify the efficiency of the data correction Acknowledgement

algorithm, an example for SAR raw data simulation is-
presented.The simulation parameteris shown in table 2.
This work was supported by 111 Project of China under
Table 2simulation parameter Grant B14010.
Parameter Original New
bandwidth(MHZ) 40 40
References
pulse width(us) 55 55
[1] Mori, A., and De Vita, F.,“A time-domain raw signal
sample rate(MHZ) 48 48
simulator forinterferometric SAR,” IEEE Transac-
orbit height 755 855
tions on Geoscience and RemoteSensing, 42(9),
velocity 7125 7024 pp.1811-1817, 2004.
PRF 1242 1242 [2] Franceschetti, G., and Migliaccio, M.,and Riccio, D.,
Scene(KmhKm) 21h25 21h25 andSchirinzi, G., “SARAS:A SAR raw signal simu-
lator,” IEEE Transactions on Geoscience andRemote
Sensing, 30(1), pp.110–123, 1992.
[3] Franceschetti, G., and Iodice, A., andPerna, S., and
Riccio, D.,“SAR sensortrajectory deviations: Fourier
domain formulation and extended scenesimulation
of raw signal,”IEEE Transactions on Geoscience an-
dRemote Sensing, 44(9),pp.2323–2334,2006.
[4] Vandewal, M., andSpeck, R., and Suess, H., “Effi-
cient SAR raw datageneration including low squint
angles and platform instabilities,”IEEE Geoscience
and Remote Sensing Letters, 5(1),pp.26-30,2008.
[5] Runge, H., and Bamler. R.,“ANovel High Precision
Fig 6originaldata image Fig 7 new data image SAR Focusing Algorithm Based On Chirp Scaling,”
In Proc. Int. Geoscience and Remote Sensing Symp.,
IGARSS’92,pp372-275,Clear Lake,TX,May 1992
. Original New
[6] Raney, R.K., and Runge, H., “Precision SAR pro-
Echo Correlation
---- 0.9064 cessing using chirpscaling” [J]. IEEE Transactions
Coefficient
on Geoscience and Remote Sensing, 32(4),pp. 786 –
Image
79,1994.
Correlation ---- 0.9991
[7] WeiLi-deng., and Li, Shao-en., and Wu, Yi-
Coefficient
rong.,Xiang Mao-sheng,“performance comparison of
ImageMean 9.9087 9.9193 algorithm for SAR raw signal generation,” Journal
ImageVariance 125.7202 125.9857 of Electronics & Information Technology,vol.27
Fig 8 the correlation coefficient, mean and variance No.2 Feb.2005

The originaldata simulation result is showed in fig7,

and new the data simulation result is showed in fig8.At

Authorized licensed use limited to: ShanghaiTech University. Downloaded on November 26,2024 at 07:44:15 UTC from IEEE Xplore. Restrictions apply.