unco

rrecte

d proo

fs

PFG 2013 / X, 001–0015 ArticleStuttgart, XXXX 2013

© 2013 E. Schweizerbart'sche Verlagsbuchhandlung, Stuttgart, Germany www.schweizerbart.deDOI: 10.1127/1432-8364/2013/0199 1432-8364/13/0199 $ 0.00

Advanced High Resolution SAR Interferometry ofUrban Areas with Airborne Millimetrewave Radar

MICHAEL SCHMITT, München, CHRISTOPHE MAGNARD, Zürich, Schweiz, STEPHAN

STANKO, Wachtberg, CHRISTIAN ACKERMANN & UWE STILLA, München

Keywords: synthetic aperture radar (SAR), SAR Interferometry, multi-aspect, airborne, urbanareas, millimetrewaves

Skymed satellites have led to a growing num-ber of applications also for urban areas thathave not been in the scope of lower resolutiondata before (STILLA 2007, SÖRGEL 2010). Whilespaceborne SAR systems are typically oper-ated in L-, C- or X-band, it is reasonable toemploy shorter wavelengths in the millimetre-wave domain (Ka-band or W-band) in order tomaximize the achievable height estimation ac-

1 Introduction

The derivation of the topography of extendedareas by across-track synthetic aperture radar(SAR) interferometry has been operational forquite some years now (BAMLER& HARTL 1998,ROSEN et al. 2000). New spaceborne missionswith resolutions in the metre-range like theGerman TanDEM-X or the Italian COSMO-

Summary: For rural and natural scenes, syntheticaperture radar interferometry (InSAR) has longbeen an operational technique for the generation ofdigital surface models. With the advent of sensorsproviding data in the decimetre resolution domain,also the analysis of densely built-up urban areas hasbecome an increasingly important research topic.Due to the complexity of this kind of scenes, how-ever, advanced interferometric techniques have tobe employed. While usually satellite-borne stacksof multi-temporal data are collected in order tomake use of differential SAR interferometry or theincreasingly popular persistent scatterer technique,this article aims at the utilization of an airbornesingle-pass multi-baseline system working in themillimetrewave domain. Starting from the descrip-tion of the exemplary German MEMPHIS sensor,the complete processing chain from the collectionof necessary navigation data over the focusing ofthe raw SAR data to finally the application of so-phisticated InSAR techniques is shown.

Zusammenfassung: Fortgeschrittene hochauflö-sende SAR Interferometrie urbaner Szenen mitflugzeuggetragenem Millimeterwellen Radar. Fürländliche und natürliche Szenen ist die SAR-Inter-ferometrie seit langem eine operationelle Technikzur Generierung digitaler Oberflächenmodelle. Mitdem Einzug von Sensoren, die Daten im Dezime-ter-Bereich bereitstellen, ist auch die Analyse dichtbebauter städtischer Gebiete ein zunehmend wich-tiges Forschungsthema geworden. Wegen der Kom-plexität dieser Art von Szenen müssen jedoch fort-geschrittene interferometrische Techniken verwen-det werden. Während dazu üblicherweise von Sa-telliten aus aufgenommene Stapel multi-temporalerDaten gesammelt werden, um sich der differentiel-len SAR-Interferometrie oder der zunehmend po-pulären Persistent Scatterer-Technik zu bedienen,zielt dieser Artikel auf die Verwendung eines flug-zeuggetragenen Einpass-Mehrfachbasislinien-Sys-tems ab, das im Millimeterwellenbereich arbeitet.Ausgehend von der beispielhaften Beschreibungdes deutschen MEMPHIS-Sensors wird die kom-plette Prozessierungskette von der Aufnahme derbenötigten Navigationsdaten über die Fokussie-rung der rohen SAR-Daten hin zur Anwendunghochentwickelter InSAR-Techniken gezeigt.

unco

rrecte

d proo

fs

2 Photogrammetrie • Fernerkundung • Geoinformation X/2013

2 Interferometric SAR Analysis ofUrban Areas – State of the Art

Since SAR interferometry was introduced inthe the 1970s (GRAHAM 1974), it has continu-ously attracted the attention of an interdiscipli-nary research community. From plain single-or repeat-pass interferometry of natural andrural scenes via differential interferometricanalysis of geophysical phenomena like earth-quakes or volcanoes (GENS & VAN GENDEREN1996) to the sophisticated persistent scatterertechnique (FERRETTI et al. 2001) it has evolvedto an operational remote sensing technologythat has become indispensable to the earth ob-servation and geoinformation communities.Among the most important developments isthe ongoing improvement of imaging resolu-tions. While state-of-the-art spaceborne mis-sions are now able to deliver data in the metreor even sub-metre domain, modern airbornesensors provide imagery with resolutions ofdown to several centimetres (BRENNER 2010).With the advent of this kind of very high andultra high resolution sensors a detailed inter-ferometric analysis of urban areas has becomepossible. However, due to the SAR inherentside-looking geometry, the exploitation of (in-terferometric) SAR data for the reconstruc-tion of urban digital surface models needs so-phisticated processing strategies and is still acommonly investigated topic in the researchcommunity (STILLA et al. 2003, BAMLER et al.2009, SCHMITT et al. 2011, ROSSI et al. 2012).While much of the early work was based onmodel-driven image analysis in order to copewith or even exploit imaging effects like layo-ver or shadowing (BOLTER 2001, THIELE et al.2007), the derivation of point clouds by persis-tent scatterer interferometry (PSI) has drawngrowing interest during the last decade (GERN-HARDT et al. 2010). The main advantage of thisframework is that only quasi-deterministicscatterers, whose reflectivities remain stableduring a set of multi-temporal acquisitions,are considered for the analysis, thus leadingto an enhanced point quality. Furthermore, inthis way also information about the movementof these persistent scatterers can be derived– and remote sensing of four dimensions be-comes possible. Since many repeat-pass im-ages are needed for PSI, usually spaceborne

curacy even for limited baselines enabling thedesign of single-pass multi-baseline systemsmounted on airborne platforms.Together with the general ability of airborne

sensors to realize almost arbitrary flight trackconfigurations (in contrast to the fixed ascend-ing/descending orbital geometry of most sat-ellite sensors), this allows the utilization ofadvanced SAR interferometry methods likemulti-baseline phase unwrapping, SAR to-mography or multi-aspect data fusion, whichgreatly supports the reconstruction of digitalsurface models (DSMs) of complicated urbanareas without the need to collect multi-tempo-ral data stacks over relatively long periods oftime. This is particularly interesting for time-critical disaster mapping scenarios.Apart from these methodical capabilities,

millimetrewave SAR provides additional con-venient characteristics in comparison to X-band SAR, e.g. inherently low speckle and alarger amount of non-specular backscatteringdue to the higher sensitivity for surface rough-ness. Besides these, with millimetrewave SARit is possible to generate high-resolution im-ages with short synthetic aperture, while thesystems can be miniaturized and therefore beadapted to unmanned aerial vehicles (UAVs),see for example (EDRICH & WEISS 2008, NOU-VEL & PLESSIS 2008, STANKO et al. 2012).This article presents the experimental Ger-

man MEMPHIS system as an example for air-borne millimetrewave SARs, although it hasto be mentioned that first plans for spacebornemissions exist (SCHAEFER et al. 2012). Exploit-ing the high flexibility of airborne systems,the main focus of this paper is put on a highresolution interferometric analysis of urbanscenes with the final goal of DSM reconstruc-tion in mind. Starting from a description ofthe system’s hardware, the sensor characteris-tics and the setup in the aircraft, the completeprocessing chain from the focusing of the rawdata with the aid of a high-precision inertialnavigation system (INS) to the exploitation ofmulti-baseline and multi-aspect data is shown.In this context preliminary results are present-ed based on real test data acquired during aflight campaign over the city of Munich, Ger-many, in 2011.

unco

rrecte

d proo

fs

Michael Schmitt et al., High Resolution SAR Interferometry 3

helps to achieve comprehensive information.Therefore, this paper wants to emphasize thestrengths of airborne single-pass multi-base-line InSAR systems: Using such systems,highly coherent data from almost arbitrary as-pect angles can be acquired in very short time,thus providing the full advantages and flexi-bilities of radar remote sensing.

3 Millimetrewave SAR SystemCharacteristics

The potential of advanced InSAR processingstrategies exploiting millimetrewave systempeculiarities is investigated based on experi-mental data acquired on a flight campaign inMay 2011. During this campaign, an interfero-metric four-antenna configuration with a max-imum baseline span of 27.5 cm was employed.The utilized airborne sensor, MEMPHIS, isdescribed in section 3.1, and details about theinterferometric configuration can be foundin section 5.1. As illustrated in Fig. 1, the testscene, located in the Maxvorstadt neighbour-hood of the city of Munich, Germany, was il-luminated from a full multi-aspect configura-tion consisting of five orthogonal and anti-par-

data is utilized for this technique. By fusingPSI point clouds from image stacks acquiredby both ascending and descending satellitetracks as proposed byGERNHARDT et al. (2012),densely sampled 3D and even 4D point data ofwhole cities can be derived.Another promising development in the field

of SAR interferometry over urban areas is theutilization of array signal processing tech-niques for multi-baseline data, which can ei-ther be employed in order to realize real 3Dimaging by the formation of so-called SAR to-mograms or to separate multiple discrete scat-tering contributions whose signals were mixedin one resolution cell due to the layover effect.Recently, many propitious approaches havebeen proposed in this field, e.g. based on air-borne PolInSAR data for detailed urban scat-terer characterisation (SAUER et al. 2011). Thelatest improvement was the introduction ofcompressive sensing based algorithms, whichprovide an efficient means for the separationof discrete scattering contributions in layoverresolution cells and were even adapted to thefour-dimensional imaging of extended city ar-eas (ZHU & BAMLER 2010, REALE et al. 2011).Although obviously much progress has

been made and a detailed imaging of urban ar-eas by means of synthetic aperture radar hasbecome a valuable remote sensing tool, somedrawbacks remain: First of all, the core ad-vantage of SAR remote sensing is the inde-pendence on daylight and weather conditions,making it especially interesting for time-crit-ical tasks such as disaster mapping. How-ever, this unique capability is lost if repeat-pass data have to be collected over relativelylong periods of time. Furthermore, repeat-pass data suffer from temporal decorrelation,which eventually leads to a loss in coherenceof the InSAR data and thus makes the relat-ed phase measurements less reliable. The sec-ond drawback is the SAR inherent side-look-ing imaging geometry that not only leads tothe already-mentioned layover effect but alsoto radar shadowing causing areas without ex-ploitable information in the images. An intui-tive solution for this problem is the utilizationof data from different aspect angles. However,state-of-the-art satellite missions only allowfor the fusion of ascending and descendingdata stacks, i.e. two aspects, which only partly

Fig. 1: Trajectories of the flight campaign Mu-nich 2011. Note that the two tracks on the rightto the scene were carried out at two differentflying heights. Thus, a full multi-aspect configu-ration with an additional fifth acquisition wasrealized (Optical image ©2012 Google).

unco

rrecte

d proo

fs

4 Photogrammetrie • Fernerkundung • Geoinformation X/2013

is achieved by aspheric lenses in front of thefeed horns. Being an experimental, modularand removable system, MEMPHIS is typi-cally mounted on a C-160 Transall airplaneof the German Armed Forces (see Fig. 2). Dueto various possible antenna shapes and con-figurations, data can be acquired in many dif-ferent SAR modes: single-pass multi-baselinecross-track interferometry with four receivingantennas, dual-pol circular or linear polarim-etry and even monopulse for moving target in-dication (MTI).

3.2 Millimetrewave Peculiarities

Due to the fact that typical wavelengths ofmillimetrewave frequencies differ from morecommon radar remote sensing bands (L, C, X)in about one order of magnitude, several pecu-liarities have to be considered; some of themcan be exploited advantageously. The mainadvantages of millimetrewave systems cer-tainly are two-fold: First of all, they allow fora significant miniaturization of the hardware,thus enabling the use on unmanned aerial ve-hicles (UAVs) and other small-scale carrierplatforms. Second, very high resolutions maybe achieved with comparably short synthet-ic apertures. One of the advantages resultingfrom a short synthetic aperture is that imagesof vegetation will be better focused, becauseblurring caused by movements of leaves andbranches etc. is reduced.Additional peculiarities of millimetre-

waves in comparison to conventional micro-wave regions occur in the fields of atmospher-ic propagation and surface roughness and areexplained in the following. A more detailedsummary of millimetrewave specifics can befound in ESSEN (2010).

3.2.1 Propagation through the atmos-phere

For millimetrewave radar applications, main-ly the transmission windows around 35 GHzand 94 GHz are employed, whereas high prop-agation losses prohibit long range applications(> 10 km). The millimetrewave region is nev-ertheless an interesting alternative to the morecommon X-band due to considerably different

allel flight tracks. The goal of this campaignwas to create an exemplary dataset of airbornemulti-aspect multi-baseline InSAR data usinga millimetrewave sensor in order to promotethe development of advanced processing strat-egies necessary for urban area analysis.

3.1 The MEMPHIS Sensor

MEMPHIS (Millimeterwave ExperimentalMultifrequency Polarimetric High-resolutionInterferometric System) was developed bythe Fraunhofer Institute for High FrequencyPhysics and Radar Technology FHR in 1998(SCHIMPF et al. 2002). The radar system usestwo front-ends of identical architecture, andoperates at 35 GHz and 94 GHz (Ka-band andW-band), respectively. The primary frequen-cies of 25 GHz and 85 GHz are generated bysuccessive multiplication and filtering of thereference frequency of 100 MHz. For bothsubsystems the waveform is modulated ontoan auxiliary signal at 9.4 GHz, which is up-converted into the respective frequency bandtogether with the primary signal. Depend-ing on the application, the sensor can eitherbe used with polarimetric monopulse feeds oran interferometric set of four receiving anten-nae. The elevation-azimuth asymmetry of thebeam that is necessary for SAR applications

Fig. 2: MEMPHIS radar mounted into a C-160.Top: Synergy infrared optical sensor, center:35 GHz interferometric SAR antenna with 3 x13 beamwidth (box), bottom: 94 GHz polari-metric monopulse antenna with 1 x 13 beam-width.

unco

rrecte

d proo

fs

Michael Schmitt et al., High Resolution SAR Interferometry 5

icantly increases with high temperatures andhumidity; it therefore is often used in weatherradars (LIEBE 1985). For that reason, the choiceof the band eventually depends on the missiongoal.

3.2.2 Surface roughness properties

In the millimetrewave region, the wavelengthis naturally very short in comparison to clas-sical radar bands, i. e. the relating phase reactsvery sensitively on movements of objects orthe radar itself. While this might seem disad-vantageous for a signal processing based im-aging system that relies on the evaluation ofthe phase of the backscattered signal, it can beutilized beneficially instead. The reason is thespecific scattering mechanism, which is domi-nated by a comparably much rougher surface(factor of 10 in comparison to X-band), mak-ing millimetrewave SAR more robust againstuncontrolled movements of the carrier air-craft. In general, the roughness of surfacescauses diffuse scattering, whereas smoothsurfaces result in specular reflections. At mil-limetrewave frequencies, most surfaces ap-pear rough, and diffuse scattering dominatesthe images (see Tab. 1). Diffuse scatteringleads to coherent averaging, an effect similarto multilook processing. Therefore, the inher-ent speckle effect within scenes of homogene-ous surface structure is lower at millimetre-wave frequencies than at X-band for an equalamount of multi-look processing. Besides thisprimary advantage of higher roughness sensi-tivity, another one is the larger extent of roughappearing surfaces in often rather smooth ur-ban environments. This provides a conveni-ent benefit to the analysis of backscatteringcharacteristics, which is often based on theassumption of Gaussian scattering. Since thisassumption only holds for so-called distribut-

propagation properties (SKOLNIK 1980), whichare caused by resonance absorption at thesefrequencies related to energy levels of vibra-tion and rotation states of molecules in the at-mosphere, e. g. water vapour or oxygen.In remote sensing, the propagation through

snow, fog, haze or clouds is one of the mostimportant reasons why SAR sensors are used.While in optical remote sensing the drop sizewithin fog and clouds is in an order of magni-tude where interactions with the electromag-netic radiation of the visible spectrum is mostlikely, these effects are of much minor impor-tance for millimetrewaves. As long as the den-sity of droplets is not too high, and as long asthe liquid water content of snow is not exces-sively high, millimetrewave signals are ableto penetrate most weather phenomena. Onlyhydrometeors with high density of large dropsizes in the order of the electromagnetic wave-length can severely influence the propagationof the signal and thus prevent the desired im-aging of the Earth surface (DANKLMAYER &CHANDRA 2009). This is caused by the fact thatthe drops act as antennae in this case, absorb-ing the energy of the resonant electromagneticwave.For the case of smoke first experiments

show a low attenuation for millimetrewaves,due to the small particle size of smoke in com-parison to e. g. sand or dust. In these lattercases, experimental results can be used for anestimation of the expected propagation loss.These attributes make millimetrwaves almostjust as interesting for any kind of mapping orreconnaissance mission during disaster sce-narios, be it floodings (mostly in concurrencewith clouds and rainfall), dust storms or fires.It has to be mentioned however, that even

within the millimetrewave domain differenc-es between the different frequencies appear.For example, inW-band the attenuation signif-

Tab. 1: Definition of radar roughness categories. The RMS surface height variations (cm) at a localincidence angle of 45° are shown (after LILLESAND et al. 2004)

Ka-band(λ = 0.86 cm)

X-band(λ = 3.2 cm)

L-band(λ = 23.5 cm)

Smooth < 0.05 < 0.18 < 1.33

Intermediate 0.05 – 0.28 0.18 – 1.03 1.33 – 7.55

Rough > 0.28 > 1.03 > 7.55

unco

rrecte

d proo

fs

6 Photogrammetrie • Fernerkundung • Geoinformation X/2013

focused in range using a chirp replica withthe conventional matched filtering technique.The full bandwidth is reconstructed in the fre-quency domain through an algorithm based onLORD (2000) and WILKINSON et al. (1998). Theazimuth compression is performed with theExtended Omega-K algorithm (REIGBER et al.2006), resulting in a zero-Doppler slant rangegeometry. The block diagram in Fig. 3 showsthe processing chain of the algorithm.One of the most critical steps for reaching

high focusing quality and geolocation accura-cy is the motion compensation, which can bedivided in a first and a second order step; thegeometry is shown in Fig. 4.The first order motion compensation is

achieved as follows:● The navigation data are upsampled to thepulse repetition frequency (PRF) rate.

● A linearized track is defined using a leastsquares method on the position data in

ed scatterers and not for frequently occurringpoint scatterers, millimetrewaves are favour-able over longer wavelengths, where tenden-tially surfaces appear less rough and thereforemore point scattering behaviours or specularreflections are observed.

3.3 Navigation Systems

Since the precision of the aircraft navigationdata is not sufficient for high-precision SARprocessing, for the 2011 campaign the systemwas complemented with a dGPS system com-posed of a GPS L1/L2 antenna (AeroAntennaAT2775-41) coupled to a receiver running at20 Hz sampling rate (Trimble R7) and a pre-cise INS working at 500 Hz sampling rate(iNAV-RQH from the company IMAR). TheGPS, INS and SAR systems were synchro-nised through event markers and secondarymarkers with the GPS time. The realization oftime synchronisation for the IMU was carriedout by the pulse per second (PPS) signal andNMEA information of the GPS receiver. Thenavigation solution of the GPS and IMU datawas then processed with the commercial soft-ware Inertial Explorer using dGPS data fromreference base stations (WAYPOINT PRODUCTSGROUP 2013). The navigation data were finallysmoothed with a Kalman filter to avoid smallvariations in the millimetre range, whichwould introduce artefacts in the focused SARdata. The lever arms between the dGPS an-tenna, the INS and the SAR antennas fixed inoperating position were measured using ter-restrial surveying methods with a few centi-metres accuracy.

4 SAR Raw Data Processing

MEMPHIS is a stepped-frequency radar sys-tem, where the pulse length can be adjusted inthe range of 80 ns – 2 μs. For the high-resolu-tion mode, it successively transmits 8 chirpsof 200 MHz bandwidth with a 100 MHzcarrier frequency shift between each other,thus building together a 900 MHz full band-width, resulting in a range resolution of about16.5 cm. As described in MAGNARD et al.(2012), the raw data from each chirp are first

Fig. 3: Block diagram of the Extended Omega-K algorithm (FFT = fast Fourier transform, IFFT= inverse fast Fourier transform, SLC = singlelook complex).

unco

rrecte

d proo

fs

Michael Schmitt et al., High Resolution SAR Interferometry 7

5 Advanced SAR InterferometryApplications

In this section, the application of MEMPHISdata as an example for a single-pass multi-baseline millimetrewave InSAR system isshown with respect to advanced InSAR analy-sis and a focus on the reconstruction of digitalsurface models of urban areas. Since modernsensors are able to provide SAR imagery inthe decimetre-resolution class, the interfero-metric analysis of densely built-up inner cityareas has challenged the remote sensing com-munity. This is especially caused by the SARinherent side-looking imaging geometry thatproduces effects like layover and shadowing,making the interpretation of (In)SAR image-ry of urban scenes a non-trivial task. Whileshadowing leads to image patches without ex-ploitable information, layover on the one handleads to difficulties in the phase-unwrappingstage of classic SAR interferometry, and onthe other hand causes signal mixtures of dif-ferent backscatterers, e.g. roof and wall of abuilding together with the ground in front ofthe building. A sketch illustrating these ef-fects can be found in Fig. 5.

5.1 Multi-baseline InSAR

5.1.1 Phase unwrapping

The multi-baseline configuration of MEM-PHIS can conveniently be exploited for thephase unwrapping stage during a standard In-SAR processing chain. In order to do so, thefact that the four receiving antennae can be

global Cartesian coordinates. A constantlinearised velocity is defined.

● For each echo, the projection of the vectorPP′→ linking the real and the linearized trackonto the mid-range line of sight is evaluat-ed. The result corresponds to the first orderrange correction ∆r(x,rm) in REIGBER et al.(2006), with x the azimuth position and rmthe mid-range distance. The phase and po-sition in the range direction of the range-compressed (RC) data are corrected ac-cording to this value.

● The RC data are interpolated in azimuth di-rection according to the constant linearizedvelocity, getting a regular (constant) spatialazimuth sampling.The second order motion compensation

consists of the following procedure:● The vector linking the real and linearizedtrack is projected onto the line of sight ofeach range sample (using a flat referencesurface), yielding the second order rangecorrection ∆r(x,r) used in REIGBER et al.(2006), with r being the range distance. Thedifference between this projection and theone applied in the first order motion com-pensation is used for correcting the phaseand position of the data.The motion compensation works best for

objects at the reference surface. The more theobjects are above or below this surface or thelarger the difference between the real and thelinearized flight path is, the larger the geoloca-tion errors and the focusing degradation willbe.

Fig. 4: Geometry of the motion compensation. The plane represented in the sketch is a planeformed by the linearized flight path and the line of sight.

unco

rrecte

d proo

fs

8 Photogrammetrie • Fernerkundung • Geoinformation X/2013

with Bref and Bfine being the baseline lengthsof the reference and fine interferograms, re-spectively. This offset is necessary since allinterferograms as shown in Tab. 2 are creat-ed, which are not all based on the same masterantenna. Once offset has been determined, thefine interferogram is unwrapped:

, ,

, ,

,0.5 22

i fine unwrapped

finei ref offset i fine

refi fine

BB

φ

φ φ φπ φ

π

=

⎢ ⎥⋅ + −⎢ ⎥

⎢ ⎥+ ⋅ +⎢ ⎥⎢ ⎥⎢ ⎥⎣ ⎦

(3)

Possible errors are corrected with the ad-dition or subtraction of 2π, while taking carethat the gradients between neighbouring pix-els stay in the interval [−π−ε; π+ε], where εis the phase noise. This process is conductedon the first pair of interferograms with base-lines B1 for the reference interferogram and B2for the fine interferogram. It is then repeatedusing the resulting unwrapped phase map asthe reference interferogram and the next in-terferogram with baseline B3 as the fine inter-ferogram, and so on, until the interferogram

used to form five different baselines is exploit-ed. The height ambiguities

( )sin

cosaRh

Bλ θ

θ α=

−(1)

for the different baselines can be found inTab. 2. R is the slant range, θ the off-nadir an-gle, B the baseline, and α the baseline incli-nation.For multi-baseline phase unwrapping, a

coarse-to-fine approach was proposed by ES-SEN et al. (2007). As a reference, the phasegiven by the interferogram with the largestambiguity height, i. e. the shortest baselineis used, as it often does not require being un-wrapped itself. The interferograms generatedusing longer baselines are then successivelyunwrapped with the help of the phase infor-mation from the already unwrapped interfero-grams.The phase unwrapping of the fine interfero-

gram is a two-step process: First, the phase ofthe reference interferogram ϕi,ref is calibratedto match the phase of the fine interferogramϕi,fine by computing a phase offset ϕoffset thatminimizes the following expression:

Fig. 5: Sketch of the imaging effects caused by the SAR inherent side-looking imaging geometry.Note, for example, how the roof CD of the left building partially mixes with the backscattering ofthe façade BC and the ground AB.

[ ]

2

, ,;min mod2

offset

finei ref offset i fine

i ref

BBφ π π

φ φ π φ∈ −

⎡ ⎤⎛ ⎞⎛ ⎞⎢ ⎥⋅ + −⎜ ⎟⎜ ⎟⎢ ⎥⎝ ⎠⎝ ⎠⎢ ⎥⎣ ⎦∑ (2)

unco

rrecte

d proo

fs

Michael Schmitt et al., High Resolution SAR Interferometry 9

5.1.2 Layover separation andtomography

Layover separation can either be seen as amulti-baseline extension of classic single-baseline interferometry or as a special case ofSAR tomography (TomoSAR). TomoSAR isbased on the establishment of a synthetic ap-erture in elevation direction by exploitationof stacked coregistered images acquired fromslightly different viewing angles. In contrastto the synthetic aperture that is employed forazimuth focusing, the tomographic (elevation)aperture can only make use of a sparse andirregularly sampled aperture – which is es-pecially tough for the single-pass MEMPHIScase with just four samples (coming from thefour receiving antennas) per resolution cell.It is well known, e.g. REIGBER & MOREIRA

with the longest baseline is unwrapped. Anexample of the unwrapping is shown in Fig. 6.

Tab. 2: Typical ambiguity heights for the Ka-band antenna of MEMPHIS at mid-range(1631 m), sensor altitude: 715 m above groundlevel, baseline inclination: 65°.

Receivingantennas

Baseline(cm)

Heightambiguity (m)

R1, R2 B1 = 5.5 227.7

R2,R3 or R3,R4 B2 = 11.0 113.9

R1,R3 B3 = 16.5 75.9

R2,R4 B4 = 22.0 56.9

R1,R4 B5 = 27.5 45.6

Fig. 6: The top-left image is an oblique-view aerial image and shows the “Alte Pinakothek” in Mu-nich, Germany. The top-right image shows a SAR amplitude image of the same building acquiredin 2011. The bottom-left image shows the corresponding interferogram with the largest baseline(27.5 cm), and the bottom-right image shows the same interferogram after the phase unwrappingprocess (Optical image ©2012 Google).

unco

rrecte

d proo

fs

10 Photogrammetrie • Fernerkundung • Geoinformation X/2013

scene can be seen in Fig. 7. The processing isorganized as follows:● First, the complex covariance matrix is es-timated for every pixel of the stack using anadaptive filter proposed in SCHMITT & STIL-LA (2013b).

● Next, the covariance matrix is eigen-de-composed in order to separate the eigen-vectors belonging to noise space EN fromthe eigenvectors belonging to signal spaceES.

● Using the steering vector

( ) ( ),2exp

sinnB

a h j hR

πλ θ

⊥⎛ ⎞= − ⋅⎜ ⎟⎝ ⎠

(5)

where λ denotes the wavelength, B⊥, n is theperpendicular baseline between the masterantenna and antenna n, R is the slant rangedistance, θ is the off-nadir angle, and h is

(2000), that the expected height resolution ofsingle-pass multi-baseline InSAR systems isgiven by

( )sinh

RB

λ θρ

Δ= . (4)

Since ΔB equals to the overall baselinelength, and θ ≈ α for MEMPHIS, the heightresolution equals to the height ambiguity ascalculated using (3) and shown in Tab. 2 forthe longest baseline B5, namely ca. 45 m. Sincethis resolution is far worse than the achievableazimuth and range resolutions, the reflectiv-ity profiles of the resolution cells are usual-ly estimated using spectral analysis methodswith super-resolution capabilities, e. g. MU-SIC (multiple signal classification) (SCHMIDT1986). A first example based on application ofMUSIC-based spectral estimation to an urban

Fig. 7: Single-pass TomoSAR results for a test building in the city of Munich, Germany. The top-left image shows an oblique view aerial image of the building, the top-right image displays the SARintensity image and the azimuth profile under investigation (red line). The bottom images showtomographic slices: The continuous MUSIC pseudo-spectrum is displayed on the left, the discretescattering profile is displayed on the right. Note the double scatterers that have been detected inthe layover part (Optical image ©2012 Google).

unco

rrecte

d proo

fs

Michael Schmitt et al., High Resolution SAR Interferometry 11

shown that methods like distributed compres-sive sensing should be able to successfully re-construct discrete backscattering profiles con-taining two point scatterers from a height sep-aration of about 12 m – 13 m for sufficientlyhigh signal-to-noise ratio (SNR).If discrete scatterers are to be detected us-

ing the MUSIC procedure described above,their height can be determined by

ˆ max MUSIChh P= (7)

The resulting profile of discrete scatter-ing points is also shown in Fig. 7. Obviously,some double scatterers caused by layover havebeen detected and resolved successfully, al-though the resolution capability seems to be-come worse with shrinking height differencebetween the scatterers.

5.2 Multi-Aspect Data Fusion

As has been mentioned before, the shadowingeffect leads to gaps in the reconstructed heightdata if only one single dataset is used. Apromising approach to cope with this problemis the utilization of images recorded from dif-ferent aspect angles (see Fig. 1). In this way, onthe one hand potential gaps can be filled withcomplementary data, while on the other handredundant measurements can be exploited toenhance the final reconstruction accuracy.The advantage of single-pass interferometricdata is that shadowing can easily be detectedby looking for pixels showing low coherence.In order to show the significance of multi-as-pect InSAR configurations, the backward ge-ocoding procedure as proposed in SCHMITT &STILLA (2011) is employed: First, the desiredmapping grid is defined in the world coordi-nate system of choice, e.g. UTM. Then, eachgrid element is extended to the height dimen-sion with a pre-defined height spacing in or-der to receive a set of height hypotheses. Eachheight hypothesis is then projected into one ormore available aspects using the inverted, lin-earized range-Doppler equations:

( )2ot

−=p s v

v(8)

the height above the reference surface, theMUSIC pseudo-spectrum

( ) ( ) ( )1

MUSIC H HN N

P ha h E E a h

= (6)

is estimated for every resolution cell. Thesuperscript H denotes the conjugate trans-pose.For the visualization in Fig. 7, the contin-

uous MUSIC pseudo-reflectivities are dis-played for one azimuth slice. Although theseresults are just preliminary, they show a prom-ising perspective for future investigations inmore advanced SAR tomography methods.Basically, both TomoSAR and layover sepa-

ration rely on the same principles. The maindifference is that TomoSAR intends to recon-struct the full, possibly continuous reflectivityprofile for each resolution cell in order to re-alize real 3D imaging for example for sceneswith volume scattering, whereas layover sepa-ration is the utilization of multi-baseline SARinterferometry for the separation of discretescatterers whose backscattering has mixeddue to the layover effect. It is well-known thatsingle-pass systems are theoretically able toretrieve the heights of K = N – 1 scatterers (N:number of receiving antennae), which meansthat MEMPHIS should be able to separate theinformation of up to three height contribu-tions within one resolution cell. Using simu-lated test data, SCHMITT & STILLA (2013a) have

Fig. 8: Zoom into the area of interest used forthe illustrations in Figs. 9 and 10 (Optical im-age ©2012 Google).

unco

rrecte

d proo

fs

12 Photogrammetrie • Fernerkundung • Geoinformation X/2013

displayed in black. It can clearly be seen howthe combination of multiple aspects enhancesthe overall coverage of the area. While 58.6%of the scene are occluded due to shadowing inthe single aspect case, using a joint analysisof multiple aspects leads to a low amount ofnon-information pixels of just 14.4%. Theseremaining non-information pixels are partlycaused by narrow street canyons and possiblyalso by specular reflections.The raw 2.5D DSM resulting from the fu-

sion of five aspects acquired from crossingand anti-parallel trajectories at two differentheights (see Fig. 1) is shown in Fig. 10.In addition to conventional multi-aspect

configurations, PALM et al. (2012) have recent-ly shown how circular trajectories can be usedas the supreme form of multi-aspect SAR im-aging: The circular trajectory is just cut intooverlapping sub-tracks of 2.5°, i. e. in totalmore than 140 MASAR (multiple aspect SAR)images are generated. It is, however, still anopen question how many views are needed inorder to realize the ideal multi-aspect configu-ration.

6 Conclusion and Outlook

In this article, the utilization of an airbornesingle-pass multi-baseline SAR system for ad-vanced interferometric SAR analysis of com-plex urban scenes has been described; further-more, the benefit of radar technology in the

R = −p s (9)

Where p = [x y z]T denotes the 3D hypoth-esis, s0 = [xs0 ys0 zs0]T describes the masterantenna position during acquisition of the firstazimuth bin, and v = [vx vy vz]T is the three-dimensional velocity vector of the master an-tenna. Then, the interferometric phase ϕmeasuredis measured for the projected pixel and com-pared to the expected (simulated) interfero-metric phase

( )2

2simulated M Sl

π

πφλ

= − − − −p s p s , (10)

where sM and sSl denote the positions of themaster and the slave sensor, respectively, and⟨ ⟩2π denotes the wrapping operator. Finally,the height hypothesis that leads to the small-est difference between measured and simu-lated phase is kept as the actual height valueof the grid element. If more than one dataset,i. e. aspect, is available, the height value can bechosen according to the mean difference be-tween simulated and measured phase valuesweighted by the corresponding coherence val-ues, such that redundant observations are ex-ploited for each grid element. The weightingby coherence enables that only reliable phasemeasurements are considered, while shadowpixels (with typically low coherence) do notinfluence the result significantly.Fig. 9 shows the scene converted into a bina-

ry map with all potential shadow pixels being

Fig. 9: Pixels for which no or only unreliable phase measurements are available are shown inblack, high coherence pixels are shown in white. Left: one aspect alone. Right: five aspectsmerged. Note the improvement of the amount of available information by multi-aspect data fusion.

unco

rrecte

d proo

fs

Michael Schmitt et al., High Resolution SAR Interferometry 13

ological developments should aim at achiev-ing maximum information from minimumavailable measurements. This way the timeand cost efficiency of synthetic aperture radarremote sensing will come to its full prestige,while ever improving sensor technology andincreasingly sophisticated processing strate-gies will enable a precise and comprehensivemapping of urban areas.

References

BAMLER, R. & HARTL, P., 1998: Synthetic apertureradar interferometry. – Inverse Problems 14 (4):R1–R54.

BAMLER, R., EINEDER, M., ADAM, N., ZHU, X. &GERNHARDT, S., 2009: Interferometric potentialof high resolution spaceborne SAR. – PFG –Photogrammetrie, Fernerkundung, Geoinforma-tion 2009 (5): 407–419.

BOLTER, R., 2001: Buildings from SAR: Detectionand reconstruction of buildings from multipleview high resolution interferometric SAR data.– PhD Thesis, Technische Universität Graz.

BRENNER, A.R., 2010: Proof of concept for airborneSAR imaging with 5 cm resolution in X-band. –8th European Conference on Synthetic ApertureRadar: 615–618

millimetrewave domain has been discussed.Motivated by the goal to create digital surfacemodels of densely built-up inner city areas,the full processing chain that is necessary hasbeen shown: Starting from an exemplary sen-sor and navigation hardware description, viathe focusing of the raw radar signals to sin-gle look complex SAR imagery up to sophis-ticated interferometric applications exploitingmulti-baseline as well as multi-aspect config-urations, a lot of interdisciplinary expertise isneeded to generate competitive results. Thepotential of airborne single-pass multi-base-line data for time-critical mapping scenarioshas been explained and state-of-the-art pro-cessing outputs have been shown. To sum up,it can be observed that both airborne and spa-ceborne SAR remote sensing have their dis-tinct advantages, which can be exploited fa-vourably subject to the specific task. Beyondthat, cutting-edge airborne sensors may al-ways be used as testing prototypes for futurespaceborne missions.Future work in the field of airborne SAR

will mainly have to focus on effective utili-zation of only sparsely available data. Sinceit is expensive and cumbersome to either col-lect large data stacks or to equip carrier plat-forms with many receiving antennas, method-

Fig. 10: Digital surface model as a result of multi-aspect InSAR data fusion by an enhanced ver-sion of the backward geocoding procedure.

unco

rrecte

d proo

fs

14 Photogrammetrie • Fernerkundung • Geoinformation X/2013

REALE, D., FORNARO, G., PAUCIULLO, A., ZHU, X. &BAMLER, R., 2011: Tomographic imaging andmonitoring of buildings with very high resolu-tion SAR data. – IEEE Geoscience and RemoteSensing Letters 8 (4): 661–665.

REIGBER, A. &MOREIRA, A., 2000: First demonstra-tion of airborne SAR tomography using multi-baseline L-band data. – IEEE Transactions onGeoscience and Remote Sensing 38 (5): 2142–2152.

REIGBER, A., ALIVIZATOS, E., POTSIS, A. &MOREIRA,A., 2006: Extended wavenumber-domain syn-thetic aperture radar focusing with integratedmotion compensation. – IEEE Proceedings Ra-dar, Sonar and Navigation 153 (3): 301–310.

ROSEN, P.A., HENSLEY, S., JOUGHIN, I.R., LI, F.K.,MADSEN, S.N., RODRIGUEZ, E. & GOLDSTEIN,R.M., 2000: Synthetic aperture radar interfer-ometry. – IEEE 88 (3): 333–382.

ROSSI, C., FRITZ, T., EINEDER, M., ERTEN, E., ZHU, X.& GERNHARDT, S., 2012: Towards an urban DEMgeneration with satellite SAR interferometry. –International Archives of the Photogrammetry,Remote Sensing and Spatial Information Sci-ences 39 (B7): 73–78.

SAUER, S., FERRO-FAMIL, L. & REIGBER, A., 2011:Three-dimensional imaging and scatteringmechanism estimation over urban scenes usingdual-baseline polarimetric InSAR observationsat L-band. – IEEE Transactions on Geoscienceand Remote Sensing 49 (11): 4616–4629.

SCHAEFER, C., VÖLKER, M., LOPEZ-DEKKER, P., YOU-NIS, M., DAGAZO-EUSEBIO, E. & LUDWIG,M., 2012:Space-borne Ka-band across-track SAR inter-ferometer. – 1st International Workshop on Ka-band Earth Observation Radar Missions.

SCHIMPF, H., ESSEN, H., BOEHMSDORFF, H. & BREHM,T., 2002: MEMPHIS – a fully polarimetric ex-perimental radar. – IEEE International Geosci-ence and Remote Sensing Symposium: 1714–1716.

SCHMIDT, R.O., 1986: Multiple emitter location andsignal parameter estimation. – IEEE Transac-tions on Antennas and Propagation 34 (3): 276–280.

SCHMITT, M., MAGNARD, C., BREHM, T. & STILLA, U,2011: Towards airborne single pass decimeterresolution SAR interferometry over urban areas.– STILLA, U., ROTTENSTEINER, F., MAYER, H., JUT-ZI, B. & BUTENUTH, M. (eds.): PhotogrammetricImage Analysis. Lecture Notes in Computer Sci-ence 6952: 197–208.

SCHMITT, M. & STILLA, U., 2011: Fusion of airbornemulti-aspect InSAR data by simultaneous back-ward geocoding. – Joint Urban Remote SensingEvent: 53–56.

DANKLMAYER, A. & CHANDRA, M., 2009: Precipita-tion effects for Ka-band SAR. – Advanced RFSensors for Earth Observation.

EDRICH, M. & WEISS, G., 2008: Second-generationKa-band UAV SAR system. – 38th European Mi-crowave Conference: 1636–1639.

ESSEN, H., BREHM, T., BOEHMSDORFF, S.&STILLA, U.,2007: Multibaseline interferometric SAR at mil-limeterwaves – Test of an algorithm on real dataand a synthetic scene. – International Archivesof the Photogrammetry, Remote Sensing andSpatial Information Sciences 36 (3/W49B): 35–39.

ESSEN, H., 2010: Airborne remote sensing at mil-limeter wave frequencies. – SÖRGEL, U. (ed.): Ra-dar remote sensing of urban areas. – SpringerScience + Business Media.

FERRETTI, A., PRATI, C. & ROCCA, F., 2001: Perma-nent scatterers in SAR interferometry. – IEEETransactions on Geoscience and Remote Sens-ing 39 (1): 8–20.

GENS, R. & VAN GENDEREN, J.L., 1996: Review ar-ticle SAR interferometry – issues, techniques,applications. – International Journal of RemoteSensing 17 (10): 1803–1835.

GERNHARDT, S., ADAM, N., EINEDER, M. & BAMLER,R., 2010: Potential of very high resolution SARfor persistent scatterer interferometry in urbanareas. – Annals of GIS 16 (2): 103–111.

GERNHARDT, S., CONG, X., EINEDER, M., HINZ, S. &BAMLER, R., 2012: Geometrical fusion of multi-track PS point clouds. – IEEE Geoscience andRemote Sensing Letters 9 (1): 38–42.

GRAHAM, L.C., 1974: Synthetic interferometer radarfor topographic mapping. – IEEE 62 (6): 763–768.

LIEBE, H.J., 1985: An updated model for millimeterwave propagation in moist air. – Radio Science20 (5): 1069–1089.

LILLESAND, T.M., KIEFER, R.W. & CHIPMAN, J.W.,2004: Remote sensing and image interpretation.– 5th Edition, John Wiley & Sons.

LORD, R., 2000: Aspects of stepped-frequency pro-cessing for low-frequency SAR systems. – PhDThesis, University of Cape Town.

MAGNARD, C., BREHM, T., ESSEN, H. & MEIER, E.,2012: High resolution MEMPHIS SAR data pro-cessing and applications. – PIERS: 328–332.

NOUVEL, J.F. & PLESSIS, O.R., 2008: The ONERAcompact SAR in Ka-band. – 7th European Con-ference on Synthetic Aperture Radar: on CD.

PALM, S., ORIOT, H.M. & CANTALLOUBE, H.M., 2012:Radargrammetric DEM extraction over urbanarea using circular SAR imagery. – IEEE Trans-actions on Geoscience and Remote Sensing 50(11): 4720–4725.

unco

rrecte

d proo

fs

Michael Schmitt et al., High Resolution SAR Interferometry 15

of target reflectivity spectrum. – 1998 South Af-rican Symposium on Communications and Sig-nal Processing: 101–104.

ZHU, X.& BAMLER, R., 2010: Tomographic SAR in-version by L1 norm regularization – The com-pressive sensing approach. – IEEE Transactionson Geoscience and Remote Sensing 48 (10):3839–3846.

Addresses of the authors:

Dipl.-Ing.MICHAEL SCHMITT, Univ.-Prof. Dr.-Ing.UWE STILLA, Technische Universität München,Photogrammetry and Remote Sensing, D-80333München, Tel.: +49-89-289-22672, -22671, Fax:+49-89-289-23202, e-mail: {michael.schmitt}{stilla}@bv.tum.de

CHRISTOPHE MAGNARD, MSc, University of Zurich,Remote Sensing Laboratories, CH-8057 Zürich,Tel.: +41-44-6355197, Fax: +41-44-6356846, e-mail: christophe.magnard@geo.uzh.ch

Dr. STEPHAN STANKO, Fraunhofer Institute for HighFrequency Physics and Radar Techniques FHR,D-53343 Wachtberg, Tel.: +49-228-9435-704, Fax:+49-228-9435-608, e-mail: stephan.stanko@fhr.fraunhofer.de

Dipl.-Ing. (FH) CHRISTIAN ACKERMANN, TechnischeUniversität München, Institute for Astronomicaland Physical Geodesy, D-80333 München, Tel.:+49-89-289-23187, Fax: +49-89-289-23178, e-mail:ackermann@bv.tum.de

Manuskript eingereicht: Februar 2013Angenommen: Juni 2013

SCHMITT, M. & STILLA, U., 2013a: Compressive-sensing based layover separation in airbornesingle-pass multi-baseline InSAR data. – IEEEGeoscience and Remote Sensing Letters 10 (2):313–317.

SCHMITT, M. & STILLA, U., 2013b: Adaptive multi-looking of airborne single-pass multi-baselineInSAR stacks. – IEEE Transactions on Geosci-ence and Remote Sensing 51 (8): tbd.

SKOLNIK, M.I., 1980: Introduction to radar systems.–McGraw-Hill, New York, USA.

SÖRGEL, U., 2010: Radar remote sensing of urbanareas. – Springer Science + Business Media.

STANKO, S., JOHANNES, W., SOMMER, R., WAHLEN, A.,SCHRÖDER, M. & CARIS, M., 2012: SUMATRA –A UAV based miniaturized SAR system. – 9thEuropean Conference on Synthetic ApertureRadar: 30–33.

STILLA, U., SÖRGEL, U. & THÖNNESSEN, U., 2003: Po-tential and limits of InSAR data for building re-construction in built-up areas. – ISPRS Journalof Photogrammetry and Remote Sensing 58 (1–2): 113–123.

STILLA, U., 2007: High resolution radar imaging ofurban areas. – FRITSCH, D. (ed.): Photogrammet-ric Week ’07: 149–158.

THIELE, A., CADARIO, E., SCHULZ, K., THÖNNESSEN,U. & SÖRGEL, U., 2007: InSAR phase profiles atbuilding locations. – International Archives ofthe Photogrammetry, Remote Sensing and Spa-tial Information Sciences 36 (3/W49A): 203–208.

WAYPOINT PRODUCTS GROUP, 2013: Inertial Explor-er® User Guide. Online: http://www.novatel.com/assets/Documents/Waypoint/Downloads/InertialExplorer850_Manual.pdf (17.6.2013).

WILKINSON, A.J., LORD, R.T. & INGGS, M.R., 1998:Stepped-frequency processing by reconstruction

unco

rrecte

d proo

fs