Ground-Based Synthetic Aperture Radar Data Processing for ...

Ground-Based Synthetic Aperture 

Radar Data Processing for Deformation 

Measurement 

Andreas Jungner 

Master’s of Science Thesis in Geodesy No. 3116 

TRITA-GIT EX 09-11 

Division of Geodesy 

Royal Institute of Technology (KTH) 

100 44 Stockholm, Sweden 

May 2009


ISSN 1653-5227 

ISRN KTH/GIT/EX–09/011-SE

Abstract 

This thesis describes a first hands-on experience working with a Ground-Based Synthetic 

Aperture Radar (GB-SAR) at the Institute of Geomatics in Castelldefels (Barcelona, 

Spain), used to exploit radar interferometry usually employed on space borne platforms. 

We describe the key concepts of a GB-SAR as well as the data processing procedure to 

obtain deformation measurements. A large part of the thesis work have been devoted to 

development of GB-SAR processing tools such as coherence and interferogram generation, 

automating the co-registration process, geocoding of GB-SAR data and the adaption of 

existing satellite SAR tools to GB-SAR data. Finally a series of field campaigns have 

been conducted to test the instrument in different environments to collect data necessary 

to develop GB-SAR processing tools as well as to discover capabilities and limitations of 

the instrument. 

The key outcome of the field campaigns is that high coherence necessary to conduct 

interferometric measurements can be obtained with a long temporal baseline. Several 

factors that affect the result are discussed, such as the reflectivity of the observed scene, 

the image co-registration and the illuminating geometry. 

iii

Sammanfattning 

Det här examensarbetet bygger på erfarenheter av arbete med en mark-baserad syntetisk 

apertur radar (GB-SAR) vid Geomatiska Institutet i Castelldefels (Barcelona, Spanien). 

SAR tekniken tillåter radar interferometri som är en vanligt förekommande teknik både på 

satellit och flygburna platformar. Det här arbetet beskriver instrumentets tekniska egenskaper 

samt behandlingen av data för att uppmäta deformationer. En stor del av arbetet 

har ägnats åt utveckling av GB-SAR data applikationer som koherens och interferogram 

beräkning, automatisering av bild matchning med skript, geokodning av GB-SAR data 

samt anpassning av befintliga SAR program till GB-SAR data. Slutligen har mätningar 

gjorts i fält för att samla in data nödvändiga för GB-SAR applikations utvecklingen samt 

få erfarenhet av instrumentets egenskaper och begränsningar. 

Huvudresultatet av fältmätningarna är att hög koherens nödvändig för interferometriska 

mätningar går att uppnå med relativ lång tid mellan mätepokerna. Flera faktorer som 

påverkar resultatet diskuteras, som det observerade områdets reflektivitet, radar bild 

matchningen och den illuminerande geometrin. 

v

Acknowledgments 

I want to thank my tutor Dr. Michele Crosetto at the Institute of Geomatics for good 

advice and for the opportunity the conduct my thesis at the institute. I also wish to thank 

my tutor at KTH, Dr. Milan Horemuž. Furthermore I want to acknowledge Dr. Bruno 

Crippa (Politecnico di Milano) for teaching me the co-registration technique used in this 

work. I also wish to acknowledge Virtual City Modeling Lab (UPC) for the surface model 

of Sagrada Família. Last but not least I thank Oriol Monserrat for daily help and advice 

on all kinds of topics including Catalan culture. 

vii

viii

Contents 

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v 

Sammanfattning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii 

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix 

1. Introduction 1 

1.1. Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 

1.2. Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 

2. GB-SAR Concepts 3 

2.1. Imaging Radar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 

2.1.1. Range Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 

2.1.2. Cross-range Resolution . . . . . . . . . . . . . . . . . . . . . . . . . 6 

2.2. Radar Interferometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 

2.2.1. Interferogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 

2.2.2. Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 

3. The GB-SAR System 11 

3.1. The Sensor Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 

3.2. Linear Scanner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 

3.3. Control Unit and Power Supply . . . . . . . . . . . . . . . . . . . . . . . . 13 

4. GB-SAR Data Processing 15 

4.1. Processing Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 

4.2. Co-Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 

4.3. Geocoding of GB-SAR Data . . . . . . . . . . . . . . . . . . . . . . . . . . 21 

4.3.1. Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 

4.3.2. Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 

5. Field Campaigns 23 

5.1. Port of Barcelona . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 

5.2. L’Eixample, Barcelona . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 

5.3. Castelldefels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 

5.4. Sagrada Família, Barcelona . . . . . . . . . . . . . . . . . . . . . . . . . . 35 

5.5. Montserrat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 

6. Conclusions 43 

References 45 

ix

Contents 

A. Appendix 47 

A.1. Co-registration Parameter File . . . . . . . . . . . . . . . . . . . . . . . . . 47 

A.2. Co-registration Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 

x

List of Figures 

2.1. Amplitude image of a GB-SAR acquisition . . . . . . . . . . . . . . . . . . 4 

2.2. Range measurement illuminating several targets . . . . . . . . . . . . . . . 5 

2.3. The relationship between the quadrature components . . . . . . . . . . . . 6 

2.4. Range measurements acquired along a rail . . . . . . . . . . . . . . . . . . 7 

2.5. Interferometric measurements principle . . . . . . . . . . . . . . . . . . . . 8 

2.6. Projected displacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 

2.7. Interferogram and Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . 10 

3.1. The IBIS-L instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 

3.2. Vertical antenna radiation pattern . . . . . . . . . . . . . . . . . . . . . . . 12 

4.1. A simplification of a SAR image formation . . . . . . . . . . . . . . . . . . 15 

4.2. Close up of amplitudes of well focused and badly focused data . . . . . . . 16 

4.3. Comparison of wrapped and unwrapped phase. . . . . . . . . . . . . . . . . 18 

4.4. Linear term correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 

4.5. Correlation windows positions . . . . . . . . . . . . . . . . . . . . . . . . . 20 

5.1. Sensor view, Port of Barcelona . . . . . . . . . . . . . . . . . . . . . . . . . 24 

5.2. Amplitude image, Port of Barcelona . . . . . . . . . . . . . . . . . . . . . . 24 

5.3. Loss of Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 

5.4. Thermal Signal to Noise Ratio . . . . . . . . . . . . . . . . . . . . . . . . . 25 

5.5. View of L’Eixample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 

5.6. Amplitude image of L’Eixample . . . . . . . . . . . . . . . . . . . . . . . . 27 

5.7. The importance of co-registration. . . . . . . . . . . . . . . . . . . . . . . . 28 

5.8. Sensor view, Castelldefels . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 

5.9. Acquisitions made with normal transmitted power level . . . . . . . . . . . 31 

5.10. Acquisitions made with increased transmitted antenna power level . . . . . 32 

5.11. Acquisitions made with single calibration configuration . . . . . . . . . . . 33 

5.12. Geocoding of Castelldefels . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 

5.13. Sagrada Família . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 

5.14. Sagrada Família Pointcloud from Terrestial Laser Scanner. . . . . . . . . . 36 

5.15. Amplitude image of Sagrada Família . . . . . . . . . . . . . . . . . . . . . 37 

5.16. Geocoding of Sagrada Família . . . . . . . . . . . . . . . . . . . . . . . . . 37 

5.17. Sensor view, Montserrat . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 

5.18. Montserrat geocoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 

5.19. Montserrat processing steps . . . . . . . . . . . . . . . . . . . . . . . . . . 41 

5.20. Montserrat processing steps (continued) . . . . . . . . . . . . . . . . . . . 42 

5.21. A Time Series of a pixel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 

xi

xii

1. Introduction 

This thesis was written at the Institute of Geomatics in Castelldefels (Barcelona, Spain) 

during July 2008 - April 2009 as a M.Sc. thesis in Engineering Geomatics at the Royal 

Institute of Technology (KTH). 

The Institute of Geomatics (http://www.ideg.cat/) is a public consortium made up of 

the Autonomous Government of Catalonia (Ministry of Town and Country Planning and 

Public Works, the Ministry of Innovation, Universities and Enterprise) and the Technical 

University of Catalonia created by Decree Law 256/1997 of the Autonomous Government 

of Catalonia on September 30, 1997. As a research centre, the Institute’s aim is promotion 

and development of Geomatics through applied research and teaching for the benefit of 

society. 

The Institute of Geomatics acquired a first commercial version of a Ground-Based Synthetic 

Aperture Radar (GB-SAR) instrument in July 2008. The opportunity to conduct 

my thesis work at the very beginning of the arrival of a new instrument made me gracefully 

choose the Institute of Geomatics for my thesis work. 

A Ground-Based Synthetic Aperture Radar (GB-SAR) is an active microwave acquisition 

sensor that provides its own illumination and measures the reflected signal. This 

makes data acquisition possible day and night independently of natural light. Synthetic 

Aperture Radar (SAR) is today a relatively mature technique implemented on numerous 

satellites and aircraft. In recent years the SAR technique have been implemented on 

ground-based platforms with the advantages of being able to illuminate an area of interest 

from an optimal angle and the possibility to acquire images at any time in comparison to 

available satellite systems. In case of fast deformations this is a decisive factor. 

Furthermore in contrast to other terrestrial deformation measurement instruments the 

GB-SAR covers a continuous surface up to approximately 1 km 2 from a single measurement 

position. GB-SAR has been used for landslide monitoring, glacier monitoring, 

avalanche prediction, volcano front monitoring, dams monitoring and subsidence monitoring 

(Noferini, 2004). 

1.1. Objectives 

The aim of this thesis work is to gain experience of a new ground-based deformation 

measurement instrument using a technique adopted from space borne platforms. A large 

part of this work consists of developing processing tools for the GB-SAR data such as coherence 

and interferogram generation, automating the co-registration process, geocoding 

of GB-SAR data and the adaption of existing satellite SAR tools to GB-SAR data. 

1

1. Introduction 

1.2. Outline 

The thesis in divided into four main parts. The first part describes general GB-SAR 

concepts and the second part describes the particular instrument used. The third part 

outlines the data processing and describes two processing steps more in detail. In the 

last part the processing tools are used to interpret real world environment data collected 

during five field campaigns to discover capabilities and limitations of the instrument. 

2

2. GB-SAR Concepts 

This Chapter describes the general concepts of a Ground-Based Synthetic Aperture Radar 

valid for the GB-SAR used in this project. The GB-SAR is fundamentally based on three 

different techniques: 

∙ Stepped Frequency Continuous Wave (SF-CW), a frequency modulation technique 

that makes it possible to resolve resolution in range. 

∙ Synthetic Aperture Radar technique (SAR), that makes it possible to resolve the 

cross-range resolution. 

∙ Interferometric Technique (InSAR), exploits the coherent phase of the received 

echoes. 

The first two techniques which make it possible to create two dimensional radar images are 

covered in the first section, while the last technique which exploits the phase to measure 

deformations, is covered in the second section. 

2.1. Imaging Radar 

Radar is an acronym for Range Detection And Ranging and refers both to a technique and 

an instrument. A radar works by transmitting short pulses of electro-magnetic energy, 

which are propagated at speed of light and reflected by the terrain surface creating return 

echoes that are collected by the receiving antenna. Measuring the time delay of the 

two-way propagation of the echo determines the range R by the equation 

t = 2 ⋅ R 

c 

(2.1) 

where c is the speed of light. The ability to determine range by measuring the time for 

the radar signal to propagate to the target and back is probably the distinguishing and 

most characteristic of a conventional radar (Skolnik, 1990). 

A special class of radars is imaging radars that are capable of not only measuring ranges 

but also creating images. To accomplish this, resolution has to be resolved in both the inrange 

and the cross-range directions. A requirement to fully exploit the image information 

is that the radar is coherent, which means that both the phase and the amplitude of the 

returned echo are stored for the image processing. 

The transmitted pulses are of conical shape with an elliptic base that illuminates an 

angular area producing images of polar geometry. See Figure 2.1. The following two 

subsections will cover how resolution is resolved. 

3


Figure 2.1.: Amplitude image of a GB-SAR acquisition of a mountain side. Red areas 

represent high backscatter echoes which mean that a large portion of the 

transmitted energy was reflected back. The type and geometry of the surface 

also affect the strength of the reflected backscatter. 

2.1.1. Range Resolution 

Resolution is defined as the minimal distance at which two distinct scatters of the same 

brightness can be uniquely discerned as a separate signal (Hanssen, 2001). To distinguish 

between objects at different distances a short pulse is used. The shorter the pulse, the 

better the resolution (Skolnik, 1990). 

Both a short pulse as well as a good Signal to Noise Ratio (SNR) that requires high 

peak power are desired. However, the shorter the pulse, the lower the transmitted energy 

since the energy the instrument can emit in a finite timespan is limited. The effect of 

a short pulse is obtained with a long pulse using a frequency modulated waveform that 

increases the spectral bandwidth of the pulse. When filtered with the transmitted signal, 

the returning waveform produces a compressed pulse whose duration is approximately the 

reciprocal of the spectral width of the modulated pulse τ ≈ 1 (Skolnik, 1990; Hanssen, 

B 

2001). This is referred to as pulse compression and defines range resolution ΔR as a 

function of bandwidth by 

ΔR = cτ 2 ≈ 

c 

2B 

(2.2) 

where c is the speed of light, τ is compressed pulse duration and B is bandwidth. Each 

discrete distance defined by ΔR is referred to as a range bin. See Figure 2.2. The 

range measurements are one dimensional which makes it impossible to distinguish between 

objects located in the same range bin. Multiple targets in the same range bin return a 

cumulative response. 

4

2.1. Imaging Radar 

Range bin 

Target 

Antenna direction 

Echo intensity 

Range [m] 

Figure 2.2.: Range measurement of radar geometry illuminating several targets. Targets 

in the same range bin return a cumulative response. It is not possible to 

distinguish between different targets in the same range bin. 

Stepped Frequency Continuous Wave 

A large bandwidth is obtained using a Stepped Frequency Continuous Wave (SF-CW) 

frequency modulation technique. 

SF-CW is commonly used in close range applications since it permits modulated pulse 

duration to be longer than the two-way propagation delay of the signal. Frequency is increased 

in discrete steps through instrument bandwidth dwelling on each frequency step 

long enough for the transmitted signal to return. The SF-CW technique consists of synthesis 

and transmission of a burst of N monochromatic pulses equally and incrementally 

spaced in frequency (with fixed frequency step of Δf) within a bandwidth B (Bernardini 

et al., 2007a,b), where 

B = Δf(N − 1) (2.3) 

For each frequency step both the orthogonal In-phase (I) and Quadrature (Q) complex 

components of the returned echo are stored representing the frequency response of the 

N pulses. The data is then reconstructed in time domain using an Inverse Discrete 

Fourier Transform (IDFT) (Bernardini et al., 2007a). From the time domain quadrature 

components the amplitude and phase is obtained by means of the magnitude and the 

argument of the complex parts, respectively: 

A = √ I 2 + Q 2 (2.4) 

( ) Q 

φ = tan 

(2.5) 

I 

where A is the amplitude and φ is the phase. See Figure 2.3. 

5


Q 

A 

φ 

I 

Figure 2.3.: The relationship between the complex In-phase (I) and Quadrature (Q) components, 

amplitude (A) and phase (φ). The amplitude is the magnitude of the 

complex components marked with a thick line and the phase is the argument. 

2.1.2. Cross-range Resolution 

Using only one range measurement it is not possible to distinguish between different 

objects located at the same distance as illustrated in Figure 2.2. But combining all 

coherent range acquisitions acquired observing the same scene slightly offset creating a 

synthetic long antenna makes it possible to focus the acquisitions into two dimensional 

images exploiting the Synthetic Aperture Radar technique. 

This is accomplished by displacing the sensor along a rail parallel to the illuminated 

scene observing the same scene from slightly different angles. All offset range measurements 

acquired are focused into a single image with origin at the center of the baseline. 

See Figure 2.4. 

Analogously to pulse compression in range, the resolution in azimuth is obtained by 

compressing the range measurements in the cross-range direction (Hanssen, 2001). To 

obtain resolution in range a long pulse is compressed. In the cross-range direction a 

long acquisition time acquiring multiple range measurements is compressed into a single 

image considered captured at a same instant of time. Cross-range is defined as an angular 

resolution 

Δφ = λ 

(2.6) 

2L 

where λ is wavelength and L is synthesized antenna length. Note that the term 1/2 

accounts for the fact the that the combined acquisitions were not acquired at the same 

instant of time. Since cross-range resolution is defined as an angle the pixel size increases 

linearly in the cross range direction by distance. Moving objects in the illumined scene 

may cause focusing distortions since the SAR technique is based on observing the same 

scene from slightly different angles during a small timespan. 

2.2. Radar Interferometry 

Using coherent data acquired at different viewpoints or instants of time makes interferometry 

possible. The interferometric technique is based on measuring relative range 

6


x 0 

x i 

x n 

x 

Figure 2.4.: The sensor moves along a baseline from x 0 to x n acquiring range measurements 

of the same scene. This permits focusing the range measurements into 

a single image using Synthetic Aperture Radar (SAR) technique. 

differences by comparing the phase components of two images, denoted as master and 

slave. The phase difference of each pixel is calculated by the argument of the pointwise 

multiplicated complex master image M containing the quadrature components I and Q 

with the corresponding conjugate of the slave image S ∗ . The deformation length d for 

each pixel is then obtained by 

d = − λ 

4π arg (M ⋅ S∗ ) = − λ 

4π Δφ M−S (2.7) 

where λ is the wavelength, arg represent the argument function of complex numbers and 

Δφ M−S is the phase difference of each pixel between the compared image pair, denoted 

master and slave. See Figure 2.5. 

Data to be compared must be acquired at exactly the same position. For repeated 

measurements the system must be carefully repositioned and the images co-registered 

in post-processing using the amplitude information of the images to be compared to be 

perfectly superimposed. This is a very important requisite and will be further discussed 

in Section 4.2. 

Since the images to be compared are acquired at different instants of time the atmospheric 

conditions must be considered. The variation of the diffraction index, due 

to temperature, humidity and pressure, causes a variation of the wavefront propagation 

velocity which may introduce an error in the range measurements. 

Introducing the atmospheric contributions and ambiguity errors, the phase differences 

are summarized in Equation 2.8: 

Δφ M−S = Δφ(R) + Δφ atm + Δφ n + Δφ noise (2.8) 

where Δφ M−S is the measured phase difference, Δφ(R) is true phase difference, Δφ atm is 

the atmospheric contribution, Δφ n is the phase ambiguity and Δφ noise is noise. 

All measurements are along the radial direction, i.e. in the Line Of Sight (LOS) of the 

antennas. This must be considered when positioning the sensor in relation to the area to 

7


Master (t 0 ) 

T x 

Rx 

atm1 

Slave (t 1 ) 

d 

T x 

Rx 

atm2 

Figure 2.5.: Interferometric measurements of acquisitions acquired at the same position 

at two different instants of time ΔT = t 1 − t 0 . The phase components of 

the master acquisition under the influence of atmosphere atm1 and the slave 

acquisition under the influence of atm2 is compared to measure deformation 

d. 

be illuminated. In Figure 2.6 the geometrical relations between measured displacement 

and effective point displacement d are obtained with geometrical uniformity, 

d = d p 

R 

h 

(2.9) 

where d is effective point displacement, d p is projected point displacement, R is range and 

h is height. 

2.2.1. Interferogram 

To determine the range variations using SAR interferometry, two radar images of the 

same surface area are acquired and differenced in phase, forming a radar interferogram. 

The complex interferogram is defined as (Kampes and Usai, 1999) 

I = M ⋅ S ∗ (2.10) 

where I is the interferogram, M is the Master image, S is the Slave image and {⋅} ∗ is the 

complex conjugate. This is equal to the argument of the phase differences 

( ) 

QΔ 

Δφ M−S = arctan 

(2.11) 

I Δ 

The phase is periodic within [−π, π] which implies that if the phase exceeds this range the 

phase jumps a cycle. This phenomena is easily seen in a visualized phase and is referred 

to as fringes. See Figure 2.7(a). 

8


h 

R 

α 

d 

α 

d p 

Figure 2.6.: The illuminated area marked with a thick line is measured from a height h 

measuring a projected displacement d p . The fraction between range R and 

height h determines the effective point displacement d 

The rand value of the phase domain also defines the maximum non-ambiguous measurable 

phase between adjacent pixels as ± λ using ±π in Equation 2.7. It should be 

4 

emphasized that as long as the phase gradient is below this value the absolute phase 

value can be reconstructed with phase unwrapping. This topic will be further discussed 


2.2.2. Coherence 

SAR interferometry only works under coherent conditions where the received waveforms 

correlate in the compared SAR image pair. Coherence describes the correlation between 

them and is a vital measurement of where the phase is exploitable. It should be emphasized 

that coherence serves as quality measurement both of the acquisitions as well as 

the data processing. Data with high coherence is the only data of interest. The complex 

coherence between two images is defined as 

γ c = 

E {M ⋅ S ∗ } 

√ 

E {M ⋅ M 

∗ 

} ⋅ E {S ⋅ S ∗ } 

(2.12) 

where E {⋅} is statistical expectation. The coherence is defined by ∣γ c ∣, and its estimator 

as (Kampes and Usai, 1999) 

∑ 1 n 

ˆγ = 

n i=0 M iSi 

∗ √ 

∑ 

∣ 1 n 

n i=0 M ∑ 

iMi ∗ 1 n 

n i=0 S (2.13) 

iSi 

∗ ∣ 

The coherence resides in the range [0, 1] with high values being coherent. Coherence 

decorrelates with time due to changes in the observed scene. The material and shape of 

the scene highly affect the coherence. Vegetation has low coherence due to its entropic 

nature while solid materials such as rocks and structures maintain high coherence for a 

longer period of time. See Figure 2.7(b). 

9


(a) Interferogram 

(b) Coherence 

Figure 2.7.: An interferogram with fringes to the left. In this particular case the fringes 

are due to phase variations provoked by combining measurements acquired 

from positions with 10 cm height separation. To the right is the corresponding 

coherence image of the same scene. Bright pixels are coherent. 

10

3. The GB-SAR System 

The GB-SAR used in this project is a digital stepped-frequency continuous wave, high 

stability coherent interferometric radar system called IBIS-L (acronym for Image By Interferometric 

Survey - L) manufactured by Ingegneria Dei Sistemi (IDS) S.p.A. 

(http://www.idsgeoradar.com/). The system consists of four main components: 

∙ Sensor module, containing the radar head and antennas. 

∙ Linear scanner, consisting of a 2.5 meter long rail and motor used to displace the 

sensor module parallel to the observed scene to acquire multiple images of the same 

scene slightly offset exploiting the SAR technique. 

∙ Control unit, PC with software to control the radar system. 

∙ Power supply, containing two serial connected 12 V car batteries, fuses and serves 

as a hub for PC connections and external power sources. 

Figure 3.1.: The IBIS-L system. The yellow sensor module is mounted on top of the linear 

scanner acquiring multiple range measurements from left to right, exploiting 

the SAR technique. To the right is the power supply and control unit. 

11

3. The GB-SAR System 

3.1. The Sensor Module 

The IBIS-L is working on Ku-band at 17.1 GHz with a maximum bandwidth of 300 MHz, 

which gives a range resolution of 0.5 m using Equation 2.2: 

ΔR = 

c 

2B = 0.5m 

The system uses two identical antennas, one for transmitting and one for receiving. The 

standard antennas are characterized by a maximum gain of 20 dBi 1 . The amplitude 

characteristics of the antenna main lobe at -3 dB, which is defined as the angular area 

within which antenna gain is more than 50% of the maximum gain (−3dB = 10⋅log 10 (0.5)) 

are 17 ∘ in the horizontal plane and 15 ∘ in the vertical plane. This has to be considered 

when pointing the sensor at the scene of interest. See Figure 3.2. It is possible to change 

to antennas with other radiation pattern characteristics. However increasing the aperture 

the transmitted energy needs to be increased. 

Figure 3.2.: Vertical antenna radiation pattern for the 20 dB gain antenna 

3.2. Linear Scanner 

The linear scanner serves as the platform to acquire the range measurements along a 

straight track to create a synthesized antenna. In comparison to space borne SAR the 

rail represents a small part of the trajectory along the orbit. 

The rail weighs 54 kg and is 2.5 meter long of which 2 m is available for the synthesized 

antenna length. See Figure 3.1. Cross-range resolution is a function of synthesized 

antenna length and wavelength which defines the maximum cross-range resolution using 

Equation 2.6: 

Δφ = λ = 4.4 mrad 

2L 

1 dB(isotropic) - the forward gain of an antenna compared with the hypothetical isotropic antenna, which 

uniformly distributes energy in all directions 

12

3.3. Control Unit and Power Supply 

This means that a pixel is 4.4 m long in the cross-range direction at 1000 m distance. Using 

a smaller baseline gives as coarser cross-range resolution but a faster acquisition. The 

offset step along the rail for each measurement is 5 mm which means 401 measurements 

are acquired when using the maximum 2 meter synthesized antenna length. 

3.3. Control Unit and Power Supply 

The system is controlled from a laptop with controlling software. It is possible to control 

the system remotely. The controlling software is used to set resolution in range and crossrange, 

maximum range, transmitted power and number of acquisitions etc. It creates a 

preview image that re-focuses for every range measurement made along the synthesized 

antenna and provides a simple in-situ deformation visualization tool. 

The power supply is powered by two serial connected 12V car batteries that have a 

capacity of approximately 24 hours. It is possible to use power from an external solar 

panel. 

It is a rather heavy and bulky instrument of which the radar head weights about 10 

kg, the baseline rail 54 kg and the power supply 89 kg. The system characteristics are 

summarized in Table 3.1: 

Table 3.1.: System Characteristics Summary 

Frequency band 

Antennas 

Wavelength 

Maximum range 

Spatial resolution 

Power supply 

Dimensions 

Weight 

Power consumption 

Processing unit 

Ku-band, 17.05-17.35 GHz 

Horn antennas (20 dB or 13.5 dB) 

1.8 cm 

4.0 km 

Max resolution; range: 0.5 m, cross-range: 4.4 mrad 

Batteries or 12VDC solar cells 

250 x 50 x 60cm (linear scanner) 

170 kg (with power supply) 

70 W 

PC Panasonic CF-19 with IBIS-L operational software 

13

4. GB-SAR Data Processing 

This Chapter describes the different processing steps of the GB-SAR data adopted in this 

project starting with the image acquisitions. The processing chain is outlined in the first 

section and two of the steps are described in detail in subsequent sections. 

4.1. Processing Chain 

∙ Data Collection This first fundamental step is to collect the data in the field which 

involves transportation and data acquisitioning. Since the instrument is heavy and 

bulky two persons are needed for the transportation and assembly of the instrument. 

About half an hour is needed for the assembly of the instrument. 

The reflectivity of the area of interest is a key factor for good measurements, especially 

at long ranges when the returning signal becomes weaker. Reflectivity is a 

function of material and observed geometry. Long range measurements also increase 

the acquisition time which increases the probability of focusing errors, especially in 

a noisy ambiance. However it is possible to shorten the acquisition time by using a 

single calibration configuration. When using the single calibration configuration the 

instrument makes only one phase calibration instead of one for each range measurement. 

This is sufficient if the instrument has reached its internal working temperature 

needed to maintain a stable phase acquired for accurate measurements. This 

is accomplished by warming up the instrument acquiring measurements for about 

half an hour or using the internal calibration capabilities of the instrument. 

∙ Focusing Raw SAR data is spread out through the image in azimuth and range 

direction. In the range direction the information is spread out due to the frequency 

modulated pulse (SF-CW) and in the azimuth direction it is spread due to the 

acquisition time used to observe the same scene slightly offset. The focusing process 

serves to collect the dispersed data into single output pixels. See Figure 4.1. 

point target raw data point target 

data 

acquisition 

data 

processing 

Figure 4.1.: A simplification of a SAR image formation. A point target has been measured 

from several azimuth positions along a linear scanner. The hyperbola represent 

range as a function of azimuth position. The point target is reconstructed 

with the SAR processing. 

15


Focusing errors occur if the linear scanner could not be kept stable during the 

acquisition. The system is very sensitive to this because of its weight and the 

inertia caused by the displacement of the sensor on top of the linear scanner. To 

avoid this the instrument needs to be fastened in a heavy base support. Two custom 

made 25 kg support concrete blocks were constructed to secure the instrument. An 

instrument rotation at millimeter level may produce measurements meters off target 

in far-range. When range measurements acquired from an instable platform are 

focused a global distortion is induced and the amplitudes look blurry or out of focus. 

Focusing errors also occurs due to moving objects in the illuminated scene. These 

movements cause local distortions in the images. The errors caused by instrument 

movement cannot be compensated for. See comparison of well focused and badly 

focused images in Figure 4.2. Significant distortions will also be shown in section 

5.1. 

(a) Well focused amplitude 

(b) Badly focused amplitude with global distortion 

Figure 4.2.: Close up of amplitudes of well focused and badly focused data induced by 

movement of the system during the acquisition. 

A target with very high reflectivity may also cause focusing errors if the received 

energy spreads out to adjacent pixels. This can however be compensated for by 

applying a windowing function such as a Kaiser or Hanning window during the 

focusing at the cost of loosing signal over the whole image. In summary the errors 

are a function of acquisition time and observed scene characteristics. 

∙ Co-Registration Co-registration is the process of aligning images acquired at the 

same scene exactly on top of each other to be able to exploit the phase information. 

This is done comparing the amplitudes of two images, referred to as Master and 

Slave, calculating the correlation of a large number of windows distributed over the 

images. The correlation windows are chosen visualizing the amplitudes and selecting 

high gain points evenly distributed ever the Region Of Interest. This will be further 

explained in Section 4.2. 

∙ Interferogram and Coherence Generation Using the resampled slave from the 

16

4.1. Processing Chain 

co-registration step, cumulative interferogram and coherence are calculated from 

the quadrature components. Cumulative in the sense that they all share the same 

master image so the history of each pixel can be evaluated, Δφ M−i . 

∙ Phase Unwrapping The principal observation is the two-dimensional relative 

phase from the interferogram which is the 2π-modulus of the unknown absolute 

phase. The inverse problem of unwrapping a phase from the interval [−π, π) is 

arguably one of the main difficulties in radar interferometry (Hanssen, 2001). The 

absolute (unwrapped) phase for a pixel i, j is equal to the relative (wrapped) phase 

and a 2π-modulus. 

Φ(i, j) = ψ(i, j) + 2πk(i, j) (4.1) 

where Φ(i, j) is unwrapped phase, ψ(i, j) is wrapped phase and k(i, j) is the number 

of 2π cycles. With the assumption that the phase difference between adjacent pixels 

always is smaller than ∣π∣ the following formula is valid for a neighboring pixel in 

the x direction and analogously for y: 

⎧ 

⎪⎨ Δψ x 

Φ x = Δψ x − 2π 

⎪⎩ 

Δψ x + 2π 

if ∣Δψ x ∣ ≤ π 

if Δψ x > π 

if Δψ x < −π 

(4.2) 

where Φ x is unwrapped phase and ψ x is wrapped phase. If the phase difference of 

adjacent pixels do not stay smaller than ∣π∣, there exists aliasing and the ambiguity 

of the 2π-modulus cannot be determined. Then a shorter temporal baseline ΔT 

should be used when forming the phase interferogram. The unwrapping in this work 

was done using a minimal cost flow (MCF) method solved as a global minimization 

problem as suggested by Costantini (1998). See Figures 4.3(a) - 4.3(d) of wrapped 

and unwrapped phase. 

∙ Atmospheric Correction A plane model was used to compensate for the atmospheric 

effects with the assumption that atmospheric phase contributions are mainly 

linear. A mask is used to select points from which plane parameters are to be estimated 

with a Least Squares Solution with observation equation. 

Φ = A + B ⋅ line + C ⋅ column (4.3) 

where Φ is phase and A, B, C are plane coefficients. See phase trend, estimated 

plane and detrended phase in Figure 4.4. 

∙ Deformation Time Series Since interferograms are comparisons of two phase 

measurements they are relative measurements. To be able to compare the phases 

they must be anchored at a point considered stable. The images are normalized by 

adjusting this point to zero. This is done masking the stable point and subtracting 

the median value of the mask from the image. After the images are normalized the 

stack of interferograms are referenced to the first image in the series by subtracting 

the first image from each image in the stack. This sets the first image to zero and 

the following interferograms as relative changes to the first one. 

17


(a) Wrapped phase 

(b) Unwrapped phase 

(c) Wrapped phase 

(d) Unwrapped phase 

Figure 4.3.: Comparison of wrapped and unwrapped phase. 

∙ Geocoding and Visualization This is a fundamental step for interpreting the 

data consisting of projection and coordinate transformation. This will be covered 


4.2. Co-Registration 

Co-registration is the process of aligning images acquired at the same scene exactly on 

top of each other to be able to exploit the phase information. This is done comparing 

the amplitudes of two images, referred to as Master and Slave, calculating the correlation 

of a large number of windows distributed over the images. The correlation windows 

are chosen visualizing the amplitudes choosing high gain points evenly distributed ever 

the Region Of Interest. This can be semi-automated using a visualizing tool like ENVI. 

The co-registration was done using a free InSAR processor named DORIS (Delft Objectoriented 

Radar Interferometric Software). This program is intended for satellite SAR 

18

4.2. Co-Registration 

(a) Phase with linear term 

(b) Estimated plane 

(c) Detrended phase 

Figure 4.4.: Linear term correction using 51889 observations to estimate three plane parameters. 

19


image processing and requires hundreds of input parameters lines containing information 

about the master and slave image and the processing steps. Pseudo satellite parameters 

such as orbit information have to be included in the parameter files for the InSAR processor 

used to accept the GB-SAR data. An example of slave image parameters is attached 

in Appendix A.1. The generation of input parameters was handled using shell scripts. 

The co-registration process is outlined in the following steps. 

∙ Correlation windows position selection. High gain amplitudes are chosen with good 

distribution. Typically more than 10000 windows are chosen and stored in an ASCII 

file. See Figure 4.5. 

Figure 4.5.: Correlation windows positions marked in with red. 

∙ Input parameter files generation. Four input parameter files are created; One each 

for the master and slave image, one for the co-registration parameters and one for 

the resampling. 

∙ Co-registration. The offset vectors to align the slave image to the master are computed 

with sub pixel accuracy for the correlation window positions selected in the 

master. The offset between master and slave is estimated by computing the correlation 

of the amplitude images for shifts at pixel level. Next, in a local neighborhood of 

the maximum (correlation at pixel level) these correlations are harmonically oversampled 

to find the maximum at sub pixel level. Based on the estimated offsets 

computed and correlation threshold selected, the transformation parameters are 

estimated by means of a Least Squares Solution (Kampes and Usai, 1999). 

∙ Resampling. Using the transformation parameters estimated in the previous step, 

the slave image is resampled to the master by an affine transformation. This step 

can be quite time consuming. Using Generic Mapping Tools (GMT) fitting statistic 

20

4.3. Geocoding of GB-SAR Data 

plots are obtained. Figures of offset vectors obtained from resampling with DORIS 

will be presented in Section 5.3. 

{ 

x′ = a 0 + a 1 x + a 2 y 

(4.4) 

y′ = b 0 + b 1 x + b 2 y 

where a 0 and b 0 are translations and a 1 , a 2 , b 1 , b 2 are rotations. 

4.3. Geocoding of GB-SAR Data 

This section describes the geocoding procedure which consists of assigning ground referenced 

coordinates to GB-SAR data. The procedure is solved as an inverse problem by 

projecting the surface model geometry into radar polar geometry. 

Required data for geocoding is a surface model and known sensor position and rotation 

between surface model and radar local reference system. Sensor position and rotation are 

determined by identifying known points in the acquired images. Corner reflectors that 

return high radar backscatters are preferably used. Identifying enough points allows over 

determination using a Least Squares Estimate. 

In this project surface models from a Digital Elevation Model (DEM) of Catalonia and 

a Terrestrial Laser Scanner point cloud (TLS) are used. 

4.3.1. Projection 

All points in surface model are projected into radar geometry (line and column) with 

respect to radar position and rotation between surface model and radar local reference 

system. For each pixel of the surface model the range R and azimuth angle θ is calculated 

with respect to radar position by 

R = √ (E − e 0 ) 2 + (N − n 0 ) 2 + (Z − z 0 ) 2 (4.5) 

( ) E − e0 

θ = arctan 

(4.6) 

N − n 0 

where θ is azimuth angle with respect to radar antenna pointing direction and E, N, Z is 

easting, northing and orthometric height of the surface model pixel. The radar position 

e 0 , n 0 , z 0 is expressed in equal coordinates as the used surface model. These coordinates 

are obtained from identified corner reflectors in acquired radar image or with an external 

measurement such as GPS or Laser Scanner. Finally the line and column of acquired 

radar image is obtained by 

line = R ΔR 

(4.7) 

column = θ − θ 0 

Δθ 

+ central column (4.8) 

where ΔR is used radar range resolution and Δθ is used cross-range resolution. The 

antenna pointing direction is expressed with θ 0 . The central column of the acquired radar 

image is added to obtain positive column values. 

21


In the case of large binary Digital Elevation Model (DEM) it is reasonable to cut out the 

scene of interest to minimize the calculation time. All surface model points assigned to the 

same radar pixel are interpolated to a discrete position. Multiple surface model points 

within the same pixel are interpolated using bilinear interpolation or inverse distance 

weighted (IDW) interpolation in case of insufficient interpolation data. The result is 

theoretical radar coverage. Filtering of data to obtain true coverage is done applying 

a mask visualizing only data that are actually seen by the sensor or areas of interest, 

e.g. containing deformation. The mask can be based on a high coherence threshold or 

high gain amplitudes. Overall it is a very useful tool for GBSAR measurements allowing 

prediction of measurements a priori given a surface model of the area of interest. 

4.3.2. Transformations 

Once data is projected and interpolated it is transformed into desired reference system. 

In this project surface model data was given in Universe Transverse Mercator (UTM) 

meridian zone 31N with European 1950 datum. GBSAR measurements were transformed 

to WGS84 geodetic coordinates to visualize data with Google Earth TM . The transformation 

UTM (N,E,H) with ED50 datum to WGS84 (φ, λ, h) is done in five steps using a 

geoide model of Catalonia for the second step. 

1. UTM ED50 (N,E,H) → ED50 (φ, λ, H) 

2. ED50 (φ, λ, H) → ED50 (φ, λ, h) 

3. ED50 (φ, λ, h) → ED50 (X,Y,Z) 

4. ED50 (X,Y,Z) → WGS84 (X,Y,Z) 

5. WGS84 (X,Y,Z) → WGS84 (φ, λ, h) 

where N is northing, E is easting, H is orthometric height, φ is latitude, λ is longitude 

and h is ellipsoidal height. 

22

5. Field Campaigns 

Five field campaigns have been made to test the instrument in real world environments. 

The main goal of the field campaigns was to collect data necessary to develop GB-SAR 

processing tools as well as discover limitations of the instrument and gain experience. 

5.1. Port of Barcelona 

The sensor was positioned at an altitude of 178 meters on the hill of Montjuïc beside the 

port admitting a good illuminating geometry of part of the port pointing the antennas 

downwards. Instrument settings and acquisition data are summarized in Table 5.1. 

Table 5.1.: Settings: Port of Barcelona 

Height 

178 m 

Max range 

3000 m 

Range resolution 

1 m 

Cross-range resolution 5.1 mrad 

Image Dimensions nx: 341, ny: 3000 

Acquisition time 15’ 

Acquisitions 16 

Epochs 2 

Four positions labeled a-d are marked in Figures 5.1 and 5.2 displaying the sensor view 

and the corresponding amplitude image for easier interpretation of the image. 

Note the distortion rings marked with b. These focusing distortions are due to the 

sideways movement of the cruise ships. The movement results in range measurements 

with different distances to the same objects. When combined during the focusing the 

distortions occur because the distance to the ships varied. The ideal scenario is completely 

stationary. The port is however a noisy environment where this will be difficult to achieve. 

A solution may be to measure in low season with little or no cruise ships. It may be better 

to measure at night to capture a more placid scene. This would also be advantageous for 

the atmospheric error contributions. 

There are known subsidence deformations in a newly constructed pier at about 1800 

meters distance. The distance is however ten times the altitude of the sensor position 

which only permits a 10% projected measurement of the point deformation. The port of 

Barcelona covers a land area of more than 800 ha, which makes it difficult to reach all 

areas of interest with reasonable height to distance ratio. 

Note the substantial loss of coherence at point a in Figure 5.3 in two consecutive 

acquisitions although it is a large structure clearly visible in the amplitude image. An 

explanation is found in Figure 5.4 showing a low Signal to Noise Ratio (SNR). In fact 

23


position a is outside of the -10dB SNR area receiving less than 10% of the transmitted 

energy. 

a b c 

d 

Figure 5.1.: Sensor view, Port of Barcelona. Note the locations marked a − d marked for 

easier interpretation of sequent radar images. 

b 

a 

c 

d 

Figure 5.2.: Amplitude image, Port of Barcelona. The letters a − d corresponds to the 

positions marked in Figure 5.1. 

24

5.1. Port of Barcelona 

a 

a 

c 

c 

(a) Amplitude 

(b) Coherence 

Figure 5.3.: Loss of Coherence at point a. 

b 

a 

c 

d 

Figure 5.4.: Thermal SNR, -3dB antenna beamwidth (blue line) and -10dB antenna 

beamwidth (red line) 

25


5.2. L’Eixample, Barcelona 

The sensor was positioned at the hill of Turò de la Rovira with a view of district L’Eixample 

in central Barcelona. The scene is a completely urbanized area with high reflectivity. See 

Figure 5.5. These acquisitions served to test the long range capabilities of the instrument 

and to test the adopted co-registration technique intended for satellite SAR images in an 

advantageous environment. To evaluate the co-registration one of the acquisitions was 

acquired rotating the linear scanner a few centimeters to provoke a slightly different pointing 

direction of the antennas. Instrument settings and acquisition data are summarized 

in Table 5.2. 

Table 5.2.: Settings L’Eixample 

Height 

260 m 

Max range 

4000 m 


1 m 





Epochs 1 

The temple of Sagrada Família is marked in Figures 5.5 and 5.6 for easier interpretation. 

Note the absence of received backscatter outside the indicated -10dB beam width area. 

This measurement confirm the importance of the antenna beam width in Figure 5.6 as 

previous seen in the acquisitions from the port of Barcelona. 

Note the vast difference of coherence and phase created from co-registered images versus 

non co-registered images in Figures 5.7(a)-5.7(d). Without co-registering the images 

contain only noise and the information is lost. In this particular case the coherence 

drops significantly after three kilometers, indicating that it is problematic to measure 

long distances. See Figure 5.7(b). Note the fringes in the phase in Figure 5.7(d) that 

are provoked by the rotation of the linear scanner. The rotation induces the slave image 

to have slightly different ranges to the same targets with respect to the master acquired 

before the rotation. 

26

5.2. L’Eixample, Barcelona 

a 

Figure 5.5.: View of L’Eixample from the sensor position at Turò de la Rovira. 

Sagrada Família indicated with an a. 

Note 

a 

Figure 5.6.: Amplitude image of L’Eixample. The width of the -10dB antenna lobes are 

marked with red dashed lines. Note Sagrada Família marked with an a. 

27


(a) Coherence calculated without 

co-registration of input data. 

(b) Phase calculated without coregistration 

of input data. 

a 

a 

(c) Coherence from co-registrered 

data. 

(d) Phase from co-registrered data. 

Figure 5.7.: Note the fringes in the phase. In this particular case the fringes are due to 

the rotation of the sensor. Note Sagrada Família marked with an a. 

28

5.3. Castelldefels 


In contrast to previous acquisition positions, the system was positioned at low altitude 

pointing the antennas slightly upwards. The scene consists of mountain area, the village of 

Castelldefels and a 16 th century castle. See Figure 5.8. The illuminated area is rather hilly 

making it suitable for geocoding of radar data using a Digital Elevation Model (DEM). The 

scene consists of a large vegetated area, which is difficult to capture due to entropic nature. 

To minimize this problem attempts were made to make faster acquisitions using single 

calibration configuration. Some acquisitions were also acquired with increased transmitted 

power. Forty acquisitions were made at seven epochs to compare coherence obtained from 

images with different temporal baselines. Instrument settings and acquisition data are 

summarized in Table 5.3. 

Table 5.3.: Settings Castelldefels 

Height 

7 m 

Max range 

4000 m 


1 m 



Acquisition time 

23’ (7’ single calibration) 


Epochs 7 

Four position are marked in Figure 5.8 for easier interpretation. Note the mountain ridges 

at a and c mainly covered with vegetation. At point b is the castle and the village of the 

Castelldefels at d. Three different acquisition configurations are compared and some 

observations are presented below. The default setting is compared to increased antenna 

power and single calibration configuration allowing a faster acquisition. 

∙ Normal antenna power. Coherence and amplitude with offset vectors obtained from 

the resampling for the standard antenna power are shown in Figures 5.9(a) - 5.9(d). 

The first image pair with a temporal baseline of 24 minutes shows good results while 

second image pair with 48 minutes shows poor results. There is a shift obtained by 

the fitting in far range. 

∙ Increased antenna power. Figures 5.10(a) - 5.10(d). This image pair shows opposite 

results compared to normal transmitted antenna power. The longer temporal 

baseline show higher coherence. 

∙ Single calibration configuration. Figures 5.11(a) - 5.11(b). The fast acquisitions 

show by far the best results. Note that the single calibration leaves an artifact 

along the centerline of the images as seen in the amplitude image. 

All data are calculated from consecutive acquisitions using identical co-registration parameters. 

The co-registration is the decisive factor. Where correlating points are found the 

coherence is high. The scene displays a difficult geometry since there is little information 

in the upper part of the image to co-register the images. The concentration of correlation 

29


points in near range admits a high degree of rotation which ruins the coherence. The far 

range area is also very vegetated making the faster acquisition more suitable. 

The Castelldefels test site served well for geocoding of data since the illuminated scene 

corresponds well to the available Digital Elevation Model (DEM) in contrast to urban areas. 

A 30x30 meter resolution DEM was resampled to 5x5 meter using bicubic convolution 

interpolation. See Figure 5.12. 

a 

c 

d 

b 

Figure 5.8.: Sensor view, Castelldefels. Note the mountain ridges at a and c mainly covered 

with vegetation. At point b is a castle, and at point d the village of 

Castelldefels. 

30


Observations_distribution+offsets 

4000 

3500 

3000 

2500 

Range 

2000 

1500 

1000 

500 

100 200 300 400 

2009 Apr 14 12:42:28 

Azimuth 

(a) Offset vectors, ΔT 24’ (b) Coherence, ΔT 24’ 


4000 

3500 

3000 

2500 

Range 

2000 

1500 

1000 

500 

100 200 300 400 

2009 Apr 14 13:41:52 

Azimuth 

(c) Offset vectors, ΔT 48’ (d) Coherence, ΔT 48’ 

Figure 5.9.: Acquisitions made with normal transmitted power level. Offset vectors obtained 

with correlation windows threshold 0.8 using different temporal baselines. 

31



4000 

3500 

3000 

2500 

Range 

2000 

1500 

1000 

500 

100 200 300 400 

2009 Apr 14 12:03:54 

Azimuth 

(a) Offset vectors, ΔT 24’ (b) Coherence, ΔT 24’ 


4000 

3500 

3000 

2500 

Range 

2000 

1500 

1000 

500 

100 200 300 400 

2009 Apr 14 12:27:18 

Azimuth 

(c) Offset vectors, ΔT 48’ (d) Coherence, ΔT 48’ 

Figure 5.10.: Acquisitions made with increased transmitted antenna power level. Offset 

vectors obtained with correlation windows threshold 0.8 using different 

temporal baselines. 

32



4000 

3500 

c 

3000 

2500 

a 

Range 

2000 

1500 

d 

1000 

b 

500 

100 200 300 400 

2009 Apr 14 13:53:13 

Azimuth 

(a) Offset vectors, ΔT 21’ 

(b) Coherence, ΔT 21’ 

Figure 5.11.: Acquisitions made with single calibration configuration. Offset vectors obtained 

with correlation windows threshold 0.8. Fitting statistics of this coregistration 

is attached in Appendix A.2. The letters corresponds to to the 

locations marked in Figure 5.8. 

33


c 

a 

d 

b 

Figure 5.12.: Geocoding of Castelldefels visualized with Google Earth TM . 

corresponds to to the locations marked in Figure 5.8. 

The letters 

34

5.4. Sagrada Família, Barcelona 


This campaign served primarily as a geocoding test of an object with a complicated 

geometry and was an attempt to integrate data from different sensors. The integration of 

sensors will not be discussed here, see Marambio et al. (2009). The ”El Nacimiento” façade 

of the Sagrada Família temple in Barcelona was chosen because of its complexity. In this 

case the surface model was made with a RIEGL LMS Z420i Laser Scanner equipped with 

a calibrated Nikon D100 6 Mega Pixel digital camera. The laser scanner measurements 

were conducted by the Virtual City Modeling Lab, Technical University of Catalonia 

(UPC). Instrument settings and acquisition data are summarized in Table 5.4. 

Table 5.4.: Settings Sagrada Família 

Height 

7 m 

Max range 

400 m 


0.5 m 





Epochs 1 

In Figure 5.15 an amplitude image show the façade of Sagrada Família at a distance of 

120-160 meters. At this distance the pixel size is about 50×50 centimeters at the bottom 

om the building growing in the cross-range direction to about 50×70 centimeters at the top 

of the building. This measurement show the importance of geocoding for interpretation 

of data. 

The point cloud consisting of more 5.3 million points were projected into polar radar 

geometry. The pointcloud is shown in Figure 5.14. Multiple points assigned to equal 

pixels were interpolated resulting in a reduction of the point cloud to 6047 radar pixels. 

These pixels are visualized with Google Earth TM in Figure 5.16. 

35


Figure 5.13.: Sagrada Família 

Figure 5.14.: Sagrada Família Pointcloud from Terrestial Laser Scanner. 

36


Figure 5.15.: SAR amplitude image of the façade of Sagrada Família at a range of 120– 

150 meters. Note the focusing distortions from moving objects during the 

acquisition. In this case there were moving construction cranes. 

Figure 5.16.: Geocoding of Sagrada Família. Colors represent orthometric height, but 

could be visualized with respect to arbitrary information, e.g. deformation 

if such data were available. 

37


5.5. Montserrat 

The Montserrat mountain is situated about 50 kilometers northeast from Barcelona. It 

is considered one of the most important mountains in Catalonia because of its thousand 

years pilgrimage history and its monastery situated at 720 meters height. The mountain 

reaches 1236 meters above sea level and consists of conglomerate sedimentary rocks. The 

mountain has a history of rockfall caused by a combination of the conglomerate characteristics 

and its steep geometry. A recent rockfall in December 2008 detaching more than 

100 m 3 of rocks, closed the only road to the monastery, immuring more than 200 cars 

and damaged a rail road. Instrument settings and acquisition data are summarized in 

Table 5.5. 

Table 5.5.: Settings Montserrat 

Height 

218 m 

Max range 

2001 m 


0.5 m 





Epochs 3 

The sensor was placed at 218 m altitude with antennas pointed upwards at the rock fall 

zone. A small concrete base was casted as a stable support permitting repositioning of 

the instrument. All acquired images were co-registered in post-processing. 

See sensor view from instrument position in Figure 5.17, with four positions marked 

for comparison with amplitude image in Figure 5.19(a). The zone of interest is situated 

at about 1200 m range, marked with a. The afflicted road and railway are marked with 

c and d respectively. 

Using a maximum spatial resolution of 0.5 m in range and 4.4 mrad in the cross range 

direction a pixel dimension in the rockfall zone of just over 0.5 by 5 m is obtained. 

All the processing tools developed were utilized for the processing. Some observations 

are presented below. 

∙ Amplitude (Figure 5.19(a)). Observing the photo it is hard to realize that the mountain 

peak b is significantly further away. The mountain peak b situated about 300 

m further away than the rock fall zone. This is clearly seen in the amplitude image 

in Figure 5.19(a) and highlights the importance of geocoding the GB-SAR data to 

interpret them. The amplitude image is obtained from input data constructed from 

multiple images to reduce noise. This is accomplished by calculating the vector sum 

of the quadrature components of consecutive acquisitions. 

∙ Coherence (Figure 5.19(b)). Note the loss of coherence in the central part of of 

the image and the lower part of the scene below the railway d compared with the 

amplitude image. This is due to the non-coherent vegetation surface characteristics. 

∙ Interferometric Phase (Figure 5.19(c)). Note the phase cycle slip due to the atmosphere. 

38


a 

b 

d 

c 

Figure 5.17.: Sensor view, Montserrat. The zone of interest is at point a. Point b is a 

mountain peak situated about 300 m further away from point b. Point c is 

a road and point d is a railway, both damaged by rockfall. 

∙ Unwrapped Phase (Figure 5.19(d)). Note how the unwrapped phase resides in the 

range [-10,30] even though the extreme values are well out of the zone of interest 

and may be considered noise. This is necessary step to be able to estimate the 

atmospheric effects using a plane model. 

∙ Linear Term Corrected (Figure 5.20(a)). Compared with previous step the phase is 

now homogeneous except for the mountain peak. This is due to the separation of 

the peak in the image. The phase unwrapping only works for connected pixels. 

∙ Normalized Phase (Figure 5.20(b)). In this image the aliasing errors from the phase 

unwrapping is clearly seen at the mountain peak and at the bottom of the image. 

However this may be accepted since they are not part of the zone of interest. 

∙ Time Series (Figure 5.21). The data obtained from the processing steps were categorized 

in continuous and discontinuous acquired data. The hypothesis for the 

continuous data is that no deformation is expected. These data served to evaluate 

the noise levels expected for the discontinuously acquired measurements with a 

temporal baseline of about two weeks for each band. This figure shows the spectral 

profile of a pixel in the zone of interest with a temporal baseline between the 

illustrated bands of seven minutes. Note that the vertical axis is in radians. The 

conversion factor from radians to millimeters is λ ≈ 1.4mm obtained with wavelength 

18 mm and Equation 2.7. The maximum deviation from zero deformation is 

4π 

at band four of about 0.15 radians which must be considered noise or error induced 

by the assumptions made during the processing. 

39


Further analyses should be devoted to key rockfall characteristics, such as volume of 

expected rockfalls, study of precursory movements and the duration of such movements. 

The geocoding was done projecting the radar geometry over a Digital Elevation Model 

(DEM). See Figure 5.18. As in the case of the Castelldefels acquisitions the DEM was 

interpolated from 30x30 meters to 5x5 meters using bicubic convolution. The colors 

represent orthometric height but could represent deformation in case of such data were 

available. 

a 

b 

c 

d 

Figure 5.18.: Montserrat geocoding. The letters correspond to the locations shown in 

Figure 5.17. Note point b in comparison with Figure 5.17, which highlights 

the importance of geocoding the GB-SAR data to interpret them. 

40


b 

a 

c 

d 

(a) Amplitude 

(b) Coherence 

(c) Phase 

(d) Unwrapped Phase 

Figure 5.19.: Montserrat processing steps. The letters correspond to the locations shown 

in Figure 5.17. 

41


b 

a 

c 

d 

(a) Linear term corrected 

(b) Phase normalized to reference 

point 

Figure 5.20.: Montserrat processing steps (continued). The letters correspond to the locations 

shown in Figure 5.17. 

Figure 5.21.: A Time Series of a pixel at point a highlighted in Figure 5.20(b). 

42

6. Conclusions 

The system characteristics of a GB-SAR, the processing steps and the preliminary results 

from GB-SAR data acquired in different environments have been presented. The main 

part of this work has been the development of the processing tools. In particular the 

image co-registration and geocoding which was started from zero. 

To accomplish this the field campaigns have been important to collect data needed 

for the development of the processing tools. Through these campaigns critical aspects 

of the data collection and processing have been learned. The key observations of the 

acquisitioning and processing of data collected in the field campaigns are presented below. 

∙ It is the necessary to access a good observation position that allows a good illuminating 

geometry. This is important in many aspects. The first aspect concerns the 

antenna beamwidth since the antenna receives the strongest signal in the pointing 

direction of the antennas. This implies that the system must be placed so the area 

of interest is found in center of the beamwidth. 

Furthermore the expected deformation direction to measured has to be considered 

when planning a new site, since the measurements are along to the Line Of Sight 

of the antennas. This may require the sensor to be placed on a high or low position 

with respect to to the area of interest to obtain a reasonable height to distance ratio 

needed to calculate a projected displacement. 

A good observation position should also produce images that include stable areas 

that the measurements can be referenced to. Furthermore it is advantageous for the 

co-registration if strong backscatter targets are found throughout the whole scene. 

∙ Co-registration is a key data processing step fundamental to conduct interferometric 

measurements. The co-registration result depends largely on the geometry and the 

reflectivity of the captured scene. Several types of measurements have been conducted 

to study co-registration results with different temporal baselines. The image 

co-registration have been successfully implemented using a free InSAR processor 

intended for satellite data. 

∙ Coherence is a particularly important quality estimator of the measurement. Only 

high level coherence data is useful for deformation measurement. Analyzed data 

collected in the field campaigns have demonstrated that coherence can be preserved 

with a long temporal baseline showing feasible results for deformation measurement. 

∙ Geocoding is necessary to assess where analyzed data is found. To conduct the 

geocoding an accurate surface model is necessary. Depending on the scale of the 

observed scene an existing Digital Elevation Model may be sufficient, otherwise 

surface data can be acquired from external sensors such as an Terrestrial Laser 

Scanner. 

43

6. Conclusions 

As a final word, it is difficult to fully understand a complex system such as the GB-SAR 

despite it being a commercial system. For each application type a lot of work should be 

done to assess the applicability of the system. 

44

References 

Bernardini, G., Pasquale, G. D., Bicci, A., Marra, M., Coppi, F., and Ricci, P. (2007a). 

Microwave interferometer for ambient vibration measurement on civil engineering structures: 

1. Principles of the radar technique and laboratory tests. In Proc. Experimental 

Vibration Analysis for Civil Engineering Structures (EVACES’07). 

Bernardini, G., Pasquale, G. D., Gallino, N., and Gentile, C. (2007b). Microwave interferometer 

for ambient vibration measurement on civil engineering structures: 2. 

Application to full-scale bridges. In Proc. Experimental Vibration Analysis for Civil 

Engineering Structures (EVACES’07). 

Costantini, M. (1998). A novel phase unwrapping method based on Network Programming. 

IEEE Transactions on Geoscience and Remote Sensing, 36(3):813–821. 

Hanssen, R. F. (2001). Radar Interferometry. Kluwer Academic Publishers, Dordrecht. 

Kampes, B. and Usai, S. (1999). Doris: The Delft object-oriented radar interferometric 

software. In Proc. 2nd Int. Symp. Operationalization of Remote Sensing. 

Marambio, A., Pucci, B., Jungner, A., Núñez, M., and Buill, F. (2009). Terrestrial Laser 

Scanner, Terrestrial Synthetic Aperture Radar and Topographic Data: An Integration 

Proposal. In Proc. 8th International Geomatics Week, Barcelona. 

Noferini, L. (2004). Processing techniques of microwave data acquired by Continuous Wave 

Stepped Frequency Radar. PhD thesis, Università degli Studi di Firenze. 

Skolnik, M. I., editor (1990). Radar Handbook. McGraw-Hill Professional, New York, 2nd 

edition. 

45

A. Appendix 

A.1. Co-registration Parameter File 

The co-registration parameters needs satellite pseudo data in order to work with the 

InSAR processor used. In this case ERS-1 data is used with doppler values (Xtrack_f_DC) 

set to zero. 

TU DELFT - DEOS 

===================================================== 

SLAVE RESULTFILE: slave.out 

InSAR Processor: Doris (Delft oo radar interferometric software) 

Version: 

Version 2.6 (debug) 

VECLIB library: not used 

LAPACK library: not used 

Compiled at: Aug 10 2000 15:10:01 

By GNU gcc: 2.95.2 

Creation of this file: Jan 26, 2009 (Monday) 

===================================================== 

Start_process_control 

readfiles: 1 

precise_orbits: 1 

crop: 1 

resample: 1 

filt_azi: 0 

filt_range: 0 

NOT_USED: 0 

End_process_control 

*====================================================================* 

| | 

Following part is appended at: Thu Aug 10 15:48:43 2000 

| | 

*--------------------------------------------------------------------* 

47

A. Appendix 

******************************************************************* 

*_Start_readfiles: 

******************************************************************* 

Volume file: /cdrom/scene1/vdf_dat.001 

Volume_ID: 1 

Volume_identifier: 0004094000014024 

Volume_set_identifier: 19950726 9492285 

(Check)Number of records in ref. file: 26519 

Product type specifier: PRODUCT:ERS-1.SAR.SLC 

Location and date/time of product creation: GENERATED AT I-PAF 

Scene identification: ORBIT 21066 DATE 26-JUL-1995 9:49:22 

Scene location: FRAME 2781 LAT: 40.94 LON: 14.02 

Leader file: 

/cdrom/scene1/lea_01.001 

Scene_centre_latitude: 40.9400000 

Scene_centre_longitude: 14.0240000 

Radar_wavelength (m): 0.0566660 

First_pixel_azimuth_time (UTC): 26-JUL-1995 09:49:23.394 

Pulse_Repetition_Frequency (computed, Hz): 1679.94931897 

Total_azimuth_band_width (Hz): 1378.0000000 

Xtrack_f_DC_constant (Hz, early edge): 0.0000 

Xtrack_f_DC_linear (Hz/s, early edge): 0.0000000 

Xtrack_f_DC_quadratic (Hz/s/s, early edge): 0.00000 

Range_time_to_first_pixel (2way) (ms): 5.5458330 

Range_sampling_rate (computed, MHz): 18.9662980496 

Total_range_band_width (MHz): 15.5500000 

******************************************************************* 

******************************************************************* 

Datafile: /cdrom/scene1/dat_01.001 

Number_of_lines_original: 4002 

Number_of_pixels_original: 401 

******************************************************************* 

* End_readfiles:_NORMAL 

******************************************************************* 

******************************************************************* 

*_Start_crop: slave step01 

******************************************************************* 

Data_output_file: iq_2009.01.22_13.24.21_montserrat1_tp4_05m.bin 

Data_output_format: complex_real4 

First_line (w.r.t. original_image): 1 

Last_line (w.r.t. original_image): 4002 

First_pixel (w.r.t. original_image): 1 

Last_pixel (w.r.t. original_image): 401 

48

A.1. Co-registration Parameter File 

******************************************************************* 

* End_crop:_NORMAL 

******************************************************************* 

******************************************************************* 

*_Start_precise_orbits: 

******************************************************************* 

t(s) X(m) Y(m) Z(m) 

NUMBER_OF_DATAPOINTS: 27 

35358.000000 5151580.064 1646538.645 4689765.110 

35359.000000 5156717.560 1646224.699 4684240.566 

35360.000000 5161849.442 1645908.227 4678710.927 

35361.000000 5166975.704 1645589.230 4673176.199 

35362.000000 5172096.340 1645267.708 4667636.387 

35363.000000 5177211.345 1644943.664 4662091.497 

35364.000000 5182320.713 1644617.098 4656541.536 

35365.000000 5187424.437 1644288.011 4650986.509 

35366.000000 5192522.513 1643956.405 4645426.423 

35367.000000 5197614.933 1643622.281 4639861.283 

35368.000000 5202701.692 1643285.640 4634291.095 

35369.000000 5207782.784 1642946.483 4628715.867 

35370.000000 5212858.204 1642604.811 4623135.603 

35371.000000 5217927.945 1642260.626 4617550.309 

35372.000000 5222992.002 1641913.928 4611959.992 

35373.000000 5228050.368 1641564.720 4606364.658 

35374.000000 5233103.038 1641213.002 4600764.313 

35375.000000 5238150.007 1640858.775 4595158.964 

35376.000000 5243191.267 1640502.040 4589548.615 

35377.000000 5248226.814 1640142.800 4583933.273 

35378.000000 5253256.642 1639781.054 4578312.945 

35379.000000 5258280.744 1639416.805 4572687.636 

35380.000000 5263299.115 1639050.053 4567057.352 

35381.000000 5268311.750 1638680.800 4561422.100 

35382.000000 5273318.642 1638309.047 4555781.886 

35383.000000 5278319.785 1637934.794 4550136.716 

35384.000000 5283315.175 1637558.045 4544486.596 

******************************************************************* 

* End_precise_orbits:_NORMAL 

******************************************************************* 

*====================================================================* 

| | 

Following part is appended at: Mon Jan 26 11:19:20 2009 

49

A. Appendix 

| | 

*--------------------------------------------------------------------* 

******************************************************************* 

*_Start_resample: 

******************************************************************* 

Shifted azimuth spectrum: 0 

Data_output_file: iq_2009.01.22_13.24.21_montserrat1_tp4_05m.resampled.bin 

Data_output_format: 

complex_real4 

Interpolation kernel: 

6 point truncated sinc 

First_line (w.r.t. original_master): 1 

Last_line (w.r.t. original_master): 4002 

First_pixel (w.r.t. original_master): 1 

Last_pixel (w.r.t. original_master): 401 

******************************************************************* 

* End_resample:_NORMAL 

******************************************************************* 

Current time: Mon Jan 26 11:19:53 2009 

A.2. Co-registration Statistics 

Degree of model: 1 

Threshold on data (correlation): 0.8 

Oversmaplings factor used in fine: 16 

This means maximum can be found within [samples]: 0.03125 

A priori sigma azimuth (based on experience): 0.15 

A priori sigma range (based on experience): 0.1 

Number of observations: 8008 

Number of rejected observations: 160 

Number of unknowns: 3 

Overall model test in Azimuth direction: 0.00586746 

Overall model test in Range direction: 0.0522292 

Largest w test statistic in Azimuth direction: 1.26765 

for window number: 539 

Largest w test statistic in Range direction: 1.90333 

for window number: 539 

Maximum deviation from unity Normalmatrix*Covar(unknowns): 

Estimated parameters in Azimuth direction 

x_hat std 

(a00 | a10 a01 | a20 a11 a02 | a30 a21 a12 a03 | ...) 

0.0006 0.0152 

2.70143e-16 

50

A.2. Co-registration Statistics 

-0.0004 0.0140 

-0.0023 0.0448 

Estimated parameters in Range direction 

(b00 | b10 b01 | b20 b11 b02 | b30 b21 b12 b03 | ...) 

0.0146 0.0002 

0.0096 0.0002 

-0.0184 0.0020 

Covariance matrix estimated parameters: 

--------------------------------------- 

0.0002 0.0001 -0.0001 

0.0001 0.0002 0.0001 

-0.0001 0.0001 0.0020 

******************************************************************* 

Current time: Tue Apr 14 13:53:10 2009 

******************************************************************* 

*_Start_resample 

Data_output_file: slave32.resampled.bin 

Data_output_format: complex_real4 

Interpolation kernel: 6 point truncated sinc 

Resampled slave size in master system: 1, 4003, 1, 401 

******************************************************************* 

Current time: Tue Apr 14 13:53:27 2009 

51

Reports in Geographic Information Technology 2009 

The TRITA-GIT Series - ISSN 1653-5227 

2009 

09-001 Ahmed Abdallah. Determination of a gravimetric geoid if Sudan using the KTH method. Master of 

Science thesis in geodesy No.3109. Supervisor: Huaan Fan. Janaury 2009. 

09-002 Hussein Mohammed Ahmed Elhadi. GIS, a tool for pavement management. Master of Science thesis 

in geoinformatics. Supervisor: Hans Hauska. February 2009. 

09-003 Robert Odolinski and Johan Sunna. Detaljmätning med nätverks-RTK – en 

noggrannhetsundersökning (Detail surveying with network RTK – an accuracy research). Master of 

Science thesis in geodesy No.3110. Supervisor: Clas-Göran Persson and Milan Horemuz. March 2009. 

09-004 Jenny Illerstam och Susanna Bosrup. Restfelshantering med Natural Neighbour och TRIAD vid byte 

av koordinatsystem i plan och höjd. Master of science thesis in geodesy No. 3111. Supervisor: Milan 

Horemuz and Lars Engberg. March 2009. 

09-005 Erik Olsson. Exporting 3D Geoinformation from Baggis Database to CityGML. Supervisors: Peter 

Axelsson and Yifang Ban. April 2009. 

09-006 Henrik Nilsson. Referenssystemsbyte i Oskarshamns kommun – en fallstudie (Change of reference 

systems in Oskarshamn – a case study). Master’s of Science thesis in geodesy No.3112. Supervisor: 

Huaan Fan. May 2009. 

09-007 Chi-Hao Poon. Interaktiv Multikriteria-Analys (Interactive Multi-Criteria Evaluation). Supervisor: 

Mats Dunkars and Yifang Ban. May 2009. 

09-008 Emma Lundberg. Fastighetsdokumentation – en jämförelse mellan två geodetiska tekniker. Master’s 

of Science thesis in geodesy No.3113. Supervisor: Milan Horemuz, Karin Klasén and Ivar Andersson. 

May 2009. 

09-009 Andenet Ashagrie Gedamu. Testing the Accuracy of Handheld GPS Receivers and Satellite Image for 

Land Registration. Master’s of Science thesis in geodesy No.3114. Supervisor: Milan Horemuz and 

Lars Palm. May 2009. 

09-010 Abubeker Worake Ahmed and Workaferahu Abebe Mergia. Determination of transformation 

parameters between WGS 84 and ADINDAN. Master’s of Science thesis in geodesy No.3115. 

Supervisor: Huaan Fan. May 2009. 

09-011 Andreas Jungner. Ground-Based Synthetic Aperture Radar Data Processing for Deformation 

Measurement. Master’s of Science thesis in geodesy No.3116. Supervisors: Milan Horemuz and 

Michele Crosetto. May 2009.


ISSN 1653-5227 

ISRN KTH/GIT/EX--09/011-SE

Ground-Based Synthetic Aperture Radar Data Processing for ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?