This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

Synthetic Aperture Radar

Interferometry

Utilizing radar principles

TASUKU TANAKA, I NYOMAN SUDI PARWATA, AND PUTU EDI YASTIKA

O ne of the most important applications of synthetic ap-

erture radar interferometry (SARIF) is making a geo-
metrical plot of observed points on the ground surface. For
SCOPE
The recent use of SARIF applications is on the rise [1]. There
are many open source or commercial application software
that purpose, we derive the two points’ distances using a programs available for SARIF processing, such as Gamma
radar principle. Mathematically, this problem is a two- SAR, Generic Mapping Tools Synthetic Aperture Radar,
unknown-variables problem. To solve, it we need only two SARscape, and Interferometric SAR (InSAR) Scientific
equations. Due to limitations of the measured distances by Computing Environment. They are easy to operate and give
radar, we require one more equation. The availability of the good results; however, the mathematical background of the
data is carefully investigated. Then, we obtain a very clear SARIF processor algorithm is not explained in detail. There
expression for the parallel and perpendicular components are many variables involved in SARIF processing, some of
of the distance. The perpendicular component is expressed which are given from the SAR sensor specification, while
by the difference of the measured phase differences with an others are measured by the SAR. We must clearly distin-
extremely large amplifier. We discuss the conversion of the guish all of the variables involved and make evident the
measured phase differences, the plot of which is an inter- mathematical background of SARIF calculations. These
ferogram, to distance. We present a numerical example that problems have been discussed in previous publications,
verifies the likelihood of obtaining the difference of the e.g., several textbooks [2]–[4] and papers [5], [6]; again,
phase differences. We also show the phase differences and however, for the authors, the mathematical background of
calculate the perpendicular component for the Advanced the SARIF calculations in these references remains nonper-
Land Observing Satellite–Phased Array-Type L-Band Syn- suasive.
thetic Aperture Radar (ALOS-PALSAR). The plot of ground The purpose of this article is to derive the two points’
points is very noisy; therefore, we must improve our calcu- distances on the ground using a radar principle and visibly
lation by reconsidering the coregistration calculation. reveal the characteristics of SARIF. During the process of
mathematical formulation, we express geometry and phys-
Digital Object Identifier 10.1109/MGRS.2019.2942952
ics in equations with known and unknown variables. The
Date of publication: 21 February 2020 number of equations must be equal to that of the unknown

2 0274-6638/20©2020IEEE IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE MONTH 2020

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at 19:25:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

variables (solutions), and the infinite approximation is ap-

z
plied in many places. In this article, we make the mathe- S (xs, zs)
matical derivation process as transparent as possible. M (xm, zm)

DISTANCE MEASUREMENT BY RADAR ρm 2

ρs 1 ρs 2
ρm 1
SYNTHETIC APERTURE RADAR
INTERFEROMETRY GEOMETRY
We can identify one point on the ground surface by con-
ventional radar measurements. Suppose we have two radar P2
P1 (x2, z2)
satellites, i.e., primary (M) and secondary (S). The radar
(x1, z1)
measurement geometry is depicted in Figure 1. To find
the distance of two points on the ground, we need four x
distances, i.e., t m1, t m2, t s1, and t s2. By solving the four ra-
FIGURE 1. The radar measurement geometry.
dar equations, we obtain the coordinates of points P1 and
P2: x 1, z 1, x 2, and z 2 . We can then calculate the point dis-
tance by the four coordinates. For this purpose, we need
the distance values for all t m1, t m2, t s1, and t s2. However, x
we cannot obtain all four distance values directly from the S (Bx, Bz)
radar measurement, so we have to find a new method, e.g.,
SARIF.
M
We introduced one coordinate system with the center at
M, the z-axis along M–P1, and the x-axis perpendicular to ρm 2
the z-axis. Point S is expressed as (B x, B z). And we introduce ρs 1
new variables: the parallel and perpendicular components ρs 2
ρm 1
of the P1–P2-distance, Tl < and Tl =. The SARIF geometry is
shown in Figure 2.
The four distances are expressed as

t 2m1 = t m2 1, (1) P2
t 2
m2 = ^t m1 + Tl < h2 + Tl 2=, (2) ∆l

t 2s1 = ^t m1 - B zh2 + B 2x , (3) P1 ∆l⊥

∆l
t 2s2 = ^t m1 + Tl < - B zh2 + ^Tl = - B x h2. (4) z

There are only two unknown variables, i.e., Tl < and Tl =. FIGURE 2. The SARIF geometry.
Equation (1) is a trivial equation, and we can ignore it;
but we need t m1 as a known variable. The value of t m1
should be given outside the radar measurement. Equation ^ t 2s2 - t 2s1 h - ^ t 2m2 - t 2m1 h = - 2B z Tl < - 2B x Tl =. (8)
(3) does not include Tl < and Tl =, so it too may be ignored.
However, we need (3) for the SARIF derivation. From (2), The left-hand side of (8) is further approximated as
we find
^t 2s2 - t 2s1 h - ^t m
2
2 - t m1 h = ^ t s2 - t s1 h^ t s2 + t s1 h - ^ t m2 - t m1 h
2

^ t m2 - t m1 h^ t m2 + t m1 h = 2t m1 Tl < + Tl + Tl =2 . (5)
2
< # ^t m2 + t m1 h + 2t m1"^t s2 - t s1 h
- ^t m2 - t m1 h,

Because the second and third terms on the right-hand side = 2t m1 dTt, (9)
are far smaller than the first term, they can be ignored.
Thus, we obtain Tl <: where dTt = (t s2 - t s1) - (t m2 - t m1). The vertical compo-
nent Tl = is achieved as
t 2m2 - t 2m1
Tl < = 2t m1 + t m2 - t m1. (6) t m1 B
Tl = = - B dTt - B z Tl <. (10)
x x
Subtracting (3) from (4), we achieve
Equation (10) shows the SARIF characteristic very clearly.
t 2s2 - t 2s1 = 2t m1 Tl < - 2B z Tl < - 2B x Tl = + Tl 2< + Tl 2=. (7) The vertical component of the distance of the two points on
the ground Tl = can be expressed as the difference of the dif-
Subtracting (5) from (7), we obtain ferences of the distances from points M and S to the two points

MONTH 2020 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE 3

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at 19:25:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

on the ground, enhanced by t m1 /B x, which is extremely large. critical calculation for SARIF. We used the coregistered sec-
In the mathematical derivation previously discussed in this ondary data, which are obtained from SARscape 5.3 soft-
section, we use three equations: (2), (3), and (4). We use only ware, not the original data. In this case, the normal condi-
(3) and (4) as their difference. So the new (11) is introduced as tion of the distance from the SAR to scatter tl is equal to
the distance from the SAR to the center of pixel t. Thus, we
t 2s2 - t 2s1 = ^t m1 + Tl < - B zh2 + ^Tl = - B x h2 - 6^t m1 - B zh2 + B 2x @. obtain the following:
(11)
^ t s2 - t s1 h - ^ t m2 - t m1 h = 4mr 6^} s2 - } s1 h - ^ } m2 - } m1 h@.
The problem is reduced to two unknown variables; i.e., Tl < (17)
and Tl =, with two equations, (2) and (11), respectively.
In some cases, (B z /B x) Tl < can be ignored because of B x & B z. The difference of the phases differences can be changed
to another difference of the phase differences:
RADAR MEASUREMENT
A radar measures distance tl, or, in other words, the dis- ^} s2 - } s1 h - ^} m2 - } m1h = ^} s2 - } m2h - ^} s1 - } m1h. (18)
tance between the SAR and the scatter position in a pixel
is expressed in the form of a complex number equation of The map of (} sn - } mn) for (n = 1, N) is called an interfero-
the phase }: gram. Although (} sn - } mn) is not exact, the difference of
the phase differences from point to point provides the exact
exp 6- j}@ = exp ;- j2r E, (12)
2tl
value from which we can calculate distance Tl =.
m

where m is the wavelength of the radar and j is the imagi- DISTANCE CALCULATION
nary unit. We cannot calculate tl from } using (12) be- The surface distance was calculated for Tl < by (6), and Tl =
cause the range of } is limited from -r to r. With SARIF, by (10). Taking into account the discussion in the previous
the distance from M to the first point on surface t m1 is giv- section, we obtain the final form for calculation:
en outside the radar measurement.
c
Nor can we the radar measurement to calculate the dif- Tl < = 2Sr , (19)
ference of the sequentially measured two points’ distance
t m1 m B
on surface tl2 - tl1 because of the limited range of the cal- Tl = = - B 4r dT} - B z Tl <. (20)
x x
culated phases. Instead, we obtain it from the SAR charac-
teristic, i.e., the range-sampling rate (Sr), as In (19) and (20), we needed Sr, t m1, B x, and B z, which were
provided from outside of the radar measurement. We must
tli + 1 - tli = 2Sr ^i = 1, 2, f h, (13)
c
calculate dT} or ^} s2 - } m2h - ^} s1 - } m1h from the radar

where c is the light velocity. Sr is a fundamental radar char- measurement; this calculation is the essential part of the
acteristic and is given as the radar specification. We use this SARIF calculation.
equation for the primary data:
c NUMERICAL CALCULATION
t m2 - t m1 = 2Sr . (14)

We can calculate the difference of differences of two THE CALCULATED DIFFERENCE

points dTt = ^tls2 - tls1h - ^tlm2 - tlm1 h . Because its corre- OF PHASE DIFFERENCES
sponding difference of phase differences dT} = ^} s2 - } s1 h - The key calculation in SARIF is the perpendicular compo-
^ } m2 - } m1 h is within the range of -r to r: nent of the distance on the ground, Tl =, based on the differ-
ence of the phase differences, dT}. To verify this equation,
^ tls2 - tls1 h - ^ tlm2 - tlm1 h = 4mr 6^} s2 - } s1 h - ^} m2 - } m1 h@. given Tl =, we numerically evaluated the expected differ-
(15) ence of the phase differences, Td}, as

4r ^
(The possibility of radar measurement of the difference of dT} = - B Tl + B z Tl < h. (21)
mt m1 x =
the distance differences is shown by the numerical example
in the following section.) It should be noted that the sec- We then obtained the following numerical results:
ondary data are coregistered to the primary data and that
Tl = Td} ^rd h
they no longer satisfy (13):
1m - 0.2603
c 5m - 0.301 ,
tls2 - tls1 ! 2Sr . (16)
10 m - 0.3518
50 m - 0.7583
As a result, (15) is not 0 in all or most all of the cases. From
this consideration, we recognize that the coregistration of where t m1 = 848.3 km, B x = 161.9 m, B z = 424 m, Tt = 9.4 m,
the secondary data to the primary data is the one essential, and m = 0.23 m, respectively, are used. Thus, we can

4 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE MONTH 2020

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at 19:25:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

3.5
2.5
1.5

Phase (rd)
0.5
–0.5
–1.5
–2.5
–3.5
600 650 700 750 800 850 900
Pixel Number

(a)
FIGURE 5. The phase difference for pixels along the first line of the
interferogram.

3
2

Phase (rd)
1
0
–1
(b)
–2
–3
FIGURE 3. The (a) primary amplitude and (b) secondary (coregis- 780 785 790 795 800 805 810 815 820 825
tered) amplitude images of ALOS-PALSAR. Pixel Number

FIGURE 6. The phase difference shown in Figure 5 when it is

enlarged for pixel numbers 782 to 820.

pixel data is expressed with 64 b, and its center is located at

40.4539 N, 141.5205 E.
From these data we obtain the interferogram by SARscape
FIGURE 4. An interferogram, obtained from SARscape 5.3. 5.3, as displayed in Figure 4. Here, the data in azimuth is
compressed from the image data by 1:4 (i.e., 250 pixels), and
assuredly measure the difference of distance differences the azimuth resolution becomes 13.2 m. The data ranges are
from the difference of phase differences. We calculated the from r to -r.
difference of the distance of two points on the ground, i.e., The plot of phase difference } sn - } mn of the interfero-
t m2 - t m1, with (14). Given a range-Sr of 16 MHz, we ob- gram’s first line (depicted in Figure 4) is shown in Figure 5.
tained Dl < = 9.4 m. This value is, of course, far larger than From this plot, we see that the measured difference of
the wavelength; therefore, we cannot measure the differ- phases is noisy; however, we can plainly recognize the ex-
ence of the distances by the radar measurement. pected feature of the 2-r cycle and the decreasing tendency
of phases. We calculated the averaged difference of phase
CALCULATION OF REAL SATELLITE DATA differences Td} and the averaged Tl = using these interfero-
The noise comes primarily from the assumption that the gram data. The total change-of-phase differences between
satellite-to-pixel-center distance is the satellite-to-scatter pixel 790 and 830 is -2r. The averaged difference of phase
distance. However, we used the satellite-to-scatter distance differences is 0.16 rd, and the averaged Tl = is 12.1 m; there-
not by itself, but in the difference of the neighboring two fore, the average values are very reasonable.
points, which, themselves, exist closely nearby. Conse- To examine this tendency further, we calculated the
quently, noise may be reduced. Secondarily, we considered phase difference } sn - } mn on one part of the line in Fig-
that there exist possibilities which may improve the coreg- ure 5; it is shown in Figure 6.
istration calculation. We numerically evaluated the differ-
ence of phase differences (interferogram). We processed SURFACE POINT PLOTS
the ALOS-PALSAR data acquired in Japan on 20 July and Based on the calculation of Tl < and Tl =, we plotted the
4 September 2007, which were provided by JAXA with the points on the ground one by one (see Figure 7):
application software SARscape 5.3 (Figure 3). The data sizes ◗◗ putting distance t m1 from M and arriving at point P1
are 1,000 rows and 1,000 lines. The pixel resolutions are 3.3 ◗◗ putting distance Tl < from P1 along M–P1 and arriving
and 14.6 m in azimuth and range, respectively. Each set of at point P1’

MONTH 2020 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE 5

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at 19:25:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

CONCLUSIONS
Satellite We derived the two points’ distance on the surface by the
radar principle in the form of its parallel and perpendicular
θ
components, i.e., Tl < and Tl =:

c
Tl < = 2Sr ,

Tl = = - B 4r 6^} s2 - } m2 h - ^} s1 - } m1 h@ - B z Tl <.
t m1 m B
x x

P1 There are several key factors for SARIF, such as:

∆l i P2 P3
∆l i P4 1) The radar-measured data are used only for the difference
∆l i
∆l⊥12 ∆l i P5 of distance differences, even though they are truncated
P1′ ∆l⊥23 ∆l⊥34
P2′ P3′ ∆l⊥45 to the limited range (-r, r).
P4′
2) Parallel component Tl < is calculated by the radar-Sr, not
FIGURE 7. The distance mapping. by the radar measurement.
3) The difference of phase differences emerged as a result
of coregistration, the essential, critical calculation for
TABLE 1. THE GROUND SURFACE PLOT. SARIF.
PIXEL NUMBER d}(rad) Td}(rad) Tl=(M)
4) As a mathematical problem of SARIF, we can set it up as
two unknown variables, Tl < and Tl =, with two equations.
790 1.6812 — —
Finally, we acknowledge that the plot of Tl = is still noisy.
791 0.7365 −0.9447 68.346
We therefore need further improvement of our calculation
792 1.0258 0.2893 −53.0865 by reconsidering the coregistration calculation.
793 0.7069 −0.319 6.7687
794 0.4605 −0.2463 −0.3795 ACKNOWLEDGMENTS
795 −.099 −0.5595 30.4404 We appreciate the important discussions with Yamaguchi
796 −0.1144 −0.0154 −23.1006 University Prof. Norikazu Shimizu, Prof. Kenji Imaoka,
and Prof. Fusanori Miura. The satellite data used in this ar-
797 0.2714 0.3859 −62.588
ticle were provided by JAXA. This work was funded by the
798 −0.2525 −0.5239 26.9396
Japan Aerospace Exploration Agency, the Ministry of Edu-
799 0.4938 0.7463 −98.0541
cation, Culture, Sports, Science and Technology of Japan,
800 −0.1453 −0.639 38.2648 and Yamaguchi University.
801 −0.6958 −0.5505 29.5577
802 −1.0845 −0.3887 13.6282 AUTHOR INFORMATION
803 −0.7182 0.3663 −60.6661 Tasuku Tanaka (tttanaka@yamaguchi-u.ac.jp) received his

804 −0.6725 0.0457 −29.1137

B.E. and M.E. degrees in precise mechanical engineering
from the Graduate School of Engineering, the University
805 −0.8496 −0.1771 −7.1873
of Tokyo, Japan, in 1971 and his Ph.D. degree from the
806 −0.8348 0.0148 −26.0726
Graduate School of Science, University of Tokyo, in 2003.
807 −0.6525 0.1824 −42.564 Since January 2006, he has been a professor at Yamaguchi
808 −0.3277 0.3248 −56.5811 University, Japan. He has worked at the National Space De-
809 −1.3633 −1.0357 77.2955 velopment Agency of Japan (NASDA) (currently the Japan
810 −1.4988 −0.1355 −11.2844 Aerospace Exploration Agency). During his career at NAS-
DA, he was assigned to several strategic positions, such as
NASDA representative for the International Space Station
◗◗ putting distance Tl = from P1’, perpendicular to M–P1, in 1990, deputy director of the Earth Observation Center
and arriving at P2 in 1993, and director of the Earth Observation Research
◗◗ calculating t m2 by t m2 = [(t m1 + Tl <) 2 + Tl =2 ] 1/2. Center in 1995. In 1998, he was a NASDA research fellow,
Repeating this process further to P3 and beyond, we ob- working at NASA’s Goddard Space Flight Center, Greenbelt,
tained the surface map schematically depicted in Figure 7 Maryland. He received the NASA Special Award for Interna-
Applying this calculation to the data in the interferogram tional Cooperation in 1992; NASDA Special Award for 30
plot of Figure 6 in the previous section, we obtained the years contribution to the Japanese space program in 2001;
ground plot, as shown by Table 1. Yamaguchi University Award for contributions to the joint
Here, t m1 = 848.3 km, Sr = 16 MHz, B x = 161.9 m, and graduate school, in cooperation with Udayana University,
B z = 424 m. In Table 1, we see that the surface points are Denpasar, Bali, Indonesia, in 2016; and a special award
not well calculated because of the noisy data. from Udayana University in 2017.

6 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE MONTH 2020

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at 19:25:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

I Nyoman Sudi Parwata (sudi_jbc@yahoo.com) received he received his doctorate from the Graduate School of Sci-
his bachelor’s degree in physics from the Faculty of Math- ence and Engineering at Yamaguchi University, where he is
ematics and Natural Science, Udayana University, Den- currently a postdoctoral fellow. In 2017, he was awarded the
pasar, Bali, Indonesia, in 2012. From 2012 to 2014, he was Outstanding Oral Presentation Award at the Young Schol-
enrolled in the postgraduate double degree program jointly ar’s Symposium on Rock Mechanics for his paper on ap-
with Udayana University and Yamaguchi University, Ja- plications of synthetic aperture radar (SAR) interferometry
pan, during which time he received his M.Eng. degree from (SARIF) in geotechnical engineering. His research interests
the Department of Mechanical Engineering, Yamaguchi include satellite remote sensing, particularly SAR image
University, and his M.Sc. degree in environmental science processing and its applications.
from Udayana University in 2014. He is currently pursuing
his Ph.D. degree with the Graduate School of Science and REFERENCES
Engineering at Yamaguchi University. He is a research as- [1] K. Ouchi, “Recent trend and advance of synthetic aperture radar
sistant with the Center for Research and Application of Sat- with selected topics,” Remote Sens., vol. 5, no. 2, 706–807, 2013.
ellite Remote Sensing, Yamaguchi University. His research [2] R. F. Hanssen, Radar Interferometry. Norwell, MA: Kluwer, 2001.
interests include interferometric SAR (InSAR) for digital [3] G. Franceschett and R. Lanari, Radar Processing. Boca Raton, FL:
elevation model generation, differential InSAR (DInSAR) CRC, 1999.
for surface deformation, and DInSAR time-series analysis [4] A. Ferretti, A. Monti-Guarnieri, C. Prati, F. Rocca, and D. Mas-
(small baseline subset and persistent scatterer-InSAR). son, InSAR Principle: Guidelines for SAR Interferometry and Inter-
Putu Edi Yastika (eyastika@yamaguchi-u.ac.jp) received pretation, TM-19. Noordwijk, The Netherlands: ESA Publications,
his bachelor’s degree in physics from the Faculty of Math- 2007.
ematics and Natural Science, Udayana University, Den- [5] M. A. Richards, “Beginner’s guide to interferometric SAR con-
pasar, Bali, Indonesia, in 2012. From 2013 to 2015, he was cepts and signal processing,” IEEE Aerosp. Electron. Syst. Mag., vol.
enrolled in the postgraduate double degree program jointly 22, no. 9, 2007, pp. 5–29.
with Udayana University and Yamaguchi University, Ja- [6] P. A. Rosen et al., “Synthetic aperture radar interferometry,” Proc.
pan, during which time he received his M.Sc. and M.Eng. IEEE, vol. 88, no. 3, 2000, pp. 333–382.
degrees in environmental remote sensing in 2015. In 2019, GRS

MONTH 2020 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE 7

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at 19:25:11 UTC from IEEE Xplore. Restrictions apply.