HIGH-RESOLUTION INTERFEROMETRIC SAR FOR DISCOVERER II

High-Resolution Interferometric Synthetic ApertureRadar for Discoverer II

Michael W. Roth

New requirements have recently emerged for high-resolution digital terrain-elevation data. Such data are required for critical applications such as intelligentpreparation of the battlefield and precision engagement. The recent development ofhigh-resolution interferometric synthetic aperture radar (IFSAR) has made meetingthese requirements a possibility. Because studies have shown that a space-based IFSARmission is feasible, a high-resolution terrain-mapping mode has been included as animportant part of the Discoverer II Technology Demonstration Program. This articlepresents a survey of the current state of research for space-based high-resolution IFSAR.Research and development have focused on key technical challenges such as phaseunwrapping, baseline-tilt estimation, motion contamination, vegetation and urban-areaeffects, synchronization, and error minimization. The application of high-resolutionterrain-elevation data to tactical targeting is also discussed. (Keywords: Interferometricsynthetic aperture radar, Mapping, Synthetic aperture radar, Terrain mapping, Topo-graphic mapping.)

INTRODUCTIONDiscoverer II is a technology demonstration program

jointly sponsored by the Defense Advanced ResearchProjects Agency (DARPA), the U.S. Air Force, andthe National Reconnaissance Office.1 Its purpose is todevelop a system design for satellite surveillance andreconnaissance by producing two prototype satellitesfor performing ground moving-target indication andsynthetic aperture radar (SAR). A constellation ofsuch satellites would be able to meet the needs of next-generation rapid targeting by enabling tasking, collec-tion, exploitation, and dissemination much faster than

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 20, NUMBER 3 (1

currently available (minutes instead of hours to days).In addition, the system would allow targets to be geo-graphically located with great accuracy. The conse-quence would be new capabilities for indication andwarning, intelligent preparation of the battlefield, andprecision engagement.

The recently developed high-resolution interfero-metric SAR (IFSAR) is one of the critical enablingtechnologies for achieving these new battlefield-preparation and precision-engagement capabilities (seethe boxed insert for a brief explanation of SAR and

999) 297

M. W. ROTH

A BRIEF TUTORIAL ON INTERFEROMETRIC SYNTHETIC APERTURE RADARSynthetic aperture radar (SAR) is a technique for per-

forming high-resolution imaging from great distances. SARworks by transmitting coherent broadband microwave radiosignals from an airplane or satellite to “illuminate” an areaand then receiving the reflected signals. These returns aresubsequently stored and processed to simulate signals ob-tained using a large-aperture antenna. The data are thenfocused to form a high-resolution image of the area. Onemajor advantage of SAR over optical imaging techniques isthat it can operate in virtually any weather.

A fundamental component of SAR imaging is that thesignals transmitted and received are coherent. In additionto a precisely measurable time delay, the received signalshave a precisely measurable phase relative to the transmittedsignal. Coherence provides the means for synthetically cre-ating a very large antenna aperture relative to that whichcould be provided with a real antenna. This large syntheticantenna size permits imaging at high resolution.

If a second SAR antenna isnearby, this phase information canalso be used to perform interferom-etry. The classical physics demon-stration of interferometry followsthe work of Thomas Young in1801, who first showed how lightwaves interfere in his double-slitexperiment. In that experiment, acoherent beam of light was split intwo and allowed to recombine.When displayed on a screen, theresulting pattern showed inter-ference fringes. This pattern isknown as the interferogram.

On a flat screen, the interfero-gram varies in a regular fashion. Itcan be exactly predicted based

on the wavelength of the light and the details of the variousoptical distances. However, if the screen is not flat, theinterferogram varies in a fashion that reflects the shape ofthe screen. If the interferogram is measured and analyzed,a mathematical model of the shape of the screen canbe derived. This is the essence of interferometric SAR(IFSAR). The interferogram from two SARs (see the figure)is measured and analyzed to derive a model of the geometricshape of the area illuminated.

Graham2 first proposed the use of IFSAR for topo-graphic mapping. Since the mid-1980s, the concept hasbeen subject to significant research and development.Early work at the Jet Propulsion Laboratory in Pasa-dena, California, focused on demonstrating the conceptfor SAR satellites with relatively low resolution (e.g.,30 m). The article describes recent work by several re-searchers examining the potential for IFSAR at highresolution.

Typical IFSAR interference pattern for undulating terrain. (Reprinted from Ref. 3 bypermission of Wiley-Liss, Inc., a division of John Wiley & Sons, Inc. 1998.)

IFSAR). Interferometric processing of two SAR imagescan generate both high-resolution and high-accuracydigital terrain-elevation data (DTED). These kinds ofterrain-mapping data have extensive utility for visual-izing important aspects of the battlefield (e.g., culvertsto hide tank columns). In addition, lower-accuracytactical-reconnaissance sensors can achieve very accu-rate target geolocation if the lower-accuracy data areregistered to a previously generated high-resolution andhigh-accuracy terrain-elevation data set. Consequently,a terrain-mapping mode is an important component ofthe Discoverer II Program.

APL has had a lead role in IFSAR technology andsystems development for a number of years. The Lab-oratory led the 1993 Rapid Mapping Project, whichevaluated the potential for rapidly producing worldwideDTED at significantly better resolution and accuracythan were currently available. In follow-on targeting*DTED is a registered trademark of the National Imaging andMapping Agency.

298 JOH

studies, APL developed a complete IFSAR simulation.In addition, the Laboratory has led IFSAR systemsstudies, tests, and development for DARPA, the Army,and the Air Force and has developed numerous inno-vative concepts and processing methods. As a result,APL was selected to be part of the government teamsupporting development of the Discoverer II terrain-mapping mode.

This article presents a survey of the current state oftechnology for high-resolution IFSAR. It is based onDARPA, Army, Air Force, and NASA studies anddiscusses major trends relevant to space-based conceptsfor producing high-resolution terrain-elevation data.The technology for generating topographic datais in flux, and no comprehensive review of the stateof the art was attempted. Rather, selected currenttrends and promising new ideas are emphasized. Final-ly, we do not intended to convey any governmentagency’s preference for any particular architecture orimplementation.

NS HOPKINS APL TECHNICAL DIGEST, VOLUME 20, NUMBER 3 (1999)

TOPOGRAPHIC MAPPINGFUNDAMENTALS

Terrain-elevation data are used to provide accurateelevation maps and other useful products. Several levelsof DTED are defined by post spacing (the distancebetween samples), vertical accuracy, and horizontalaccuracy. Post spacing represents the resolution of thedata. Vertical accuracy is measured by the statisticalquantity known as the 90% linear error (LE90), whichrepresents the maximum value of vertical error for 90%of the data. Horizontal accuracy is measured by the90% circular error (CE90), which represents the max-imum value of the horizontal-distance post-locationerror for 90% of the data. The requirements for postspacing and accuracy vary according to a range of func-tionality. Post spacing can range from 100 to 1 m,relative vertical accuracy from 10 to 0.25 m, and rel-ative horizontal accuracy from 15 to 1 m.

Sets of DTED are defined as levels 1 through 5 data,from lowest to highest resolution. The emerging re-quirements are as follows:

• A level 1 data set must be nearly global. It is definedas having 100-m post spacing, 15-m relative horizon-tal accuracy, and 10-m relative vertical accuracy.Its projected use is for worldwide general missionplanning.

• Level 2 is a higher-resolution near-global data sethaving 30-m post spacing, 10-m relative horizontalaccuracy, and 7-m relative vertical accuracy. Projecteduse is as tactical terrain data for mission planning.

• Level 3 is defined as a regional data set having 10-mpost spacing, 3-m relative horizontal accuracy, and2-m relative vertical accuracy. Projected use is forflight-training simulators and midcourse weaponsguidance.

• Level 4 is also considered a regional data set, with3-m post spacing, 2-m relative horizontal accuracy,and 0.8-m relative vertical accuracy. Projected use isas tactical terrain data for mission planning andweapon terminal guidance.

• Level 5 contains the highest-resolution data. A level5 data set is defined as having 1-m post spacing,0.5-m relative horizontal accuracy, and 0.33-mrelative vertical accuracy. Projected uses includebattlefield visualization and Special Forces planning.

An incomplete global data set presently exists ateven the lowest-resolution DTED product (level 1).DTED at approximately 30-m resolution (level 2) existfor a very small portion of the world (although 30-mDTED of variable quality are available from theU.S. Geological Survey for most of the contin-ental United States). Furthermore, the capability ofproducing needed regional high-resolution data sets of1- to 10-m post spacing on demand is not available.

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 20, NUMBER 3 (19

HIGH-RESOLUTION INTERFEROMETRIC SAR FOR DISCOVERER II

To improve the global coverage of digital elevationdata, NASA and the National Imagery and MappingAgency (NIMA) are sponsoring a special 11-day SpaceShuttle mission that will be launched in September1999. The Shuttle Radar Topography Mission4

(SRTM) will collect 30-m post-spacing data between60°N and 57°S. The mission will employ IFSAR usingthe Shuttle Imaging Radar-version C and a receive-only antenna deployed at the end of a 60-m boom.Performance is projected to be a relative LE90 of 8 m,an absolute LE90 of 13–16 m, and an absolute CE90of 20 m. The joint NASA/NIMA SRTM Program plansto produce near-global DTED at level 2.

METHODOLOGYTopographic maps are produced using various tech-

niques, including stereophotogrammetry, human sur-vey, and airborne IFSAR and lidar. Stereophotogram-metry is constrained by weather, has a labor-intensiveproduction process, and cannot meet emerging require-ments for timeliness. Human surveys are time-consum-ing, and access can be denied to surveyors during bothcrisis and peace times. Airborne IFSAR has demon-strated the capability for generating level 5 data(Fig. 1). In addition, airborne lidar systems have be-come available that can provide digital elevationmodels for regional areas of interest with even higherresolution and accuracy. However, access for airbornesystems can also be denied in crisis and in peace times.

Consequently, an all-weather and all-access map-ping capability requires the use of space-based SAR.IFSAR processing can be performed on SAR imagescollected from space if a second image of the sametarget area can be obtained. If the two images are takenso that phase-to-target information is preserved, thenthe images are called coherent. Otherwise, the imagesare incoherent. In general, coherence can rapidly de-grade with increasing time between the collection ofthe two images.5,6 However, if two SAR satellites canbe deployed near one another (e.g., within 5 km) sothat coherence can be preserved, then interferometricprocessing can be performed (Fig. 2). IFSAR canestimate the phase differences to a target location be-tween the two SAR images. These phase differencescan be used, in turn, to generate elevation maps withquality up to that corresponding to level 5. However,such accuracy and precision requires accurate knowl-edge of the two antenna phase centers.

The geometrical line segment connecting the phasecenters of two IFSAR antennas is called the baseline.Accurate knowledge of this baseline is critical for theproduction of a high-resolution digital elevation model(DEM). In particular, a knowledge error in the baselinetilt contributes to an elevation error that correspondsto a tilted DEM. The baseline-tilt error can be reduced

99) 299

M. W. ROTH

Figure 1. Sun-shaded and elevation color-coded representation of a high-resolutiondigital elevation model. The model was derived from APL processing of airplane-basedIFSAR data collected over Camp Roberts, CA. The collection system, known as the GlobalTopographic Mapping Digital Collection System (GTM DCS), was developed by ERIMInternational, Inc., Ann Arbor, MI. SAR image-formation processing was performed bySandia National Laboratories. Post spacing is 1.0 m. Red, green, and blue represent 185.0,142.5, and 100.0 m (World Geodetic System [WGS]-84), respectively, with other colors andelevations correspondingly interpolated. The image area, 450 3 300 m, appears three-dimensional when viewed with ChromaDepth glasses.

Figure 2. Collection and exploitation scenario for Discoverer II IFSAR. Shown is thebistatic mode, in which one satellite transmits and receives and the other (about 5 km away)only receives. A high level of geolocation accuracy can be achieved for the IFSAR dataproducts, either by differential Global Positioning System (GPS) processing or by the useof control points such as GPS tags, as shown. Data can be downlinked to the theater usingchannels of the Common Data Link (CDL). The geolocation accuracy of data fromunmanned air vehicle sensors (as shown) can be greatly improved by registration to thecollected IFSAR data (adapted from Hughes1).

5 km

Transmit/receive

GPS tag

Theater

Channelsof CDL

Receive

GPS tag

Theater

300 JOHNS HOPKINS APL TECH

by two methods: (1) precision nav-igation using techniques such asdifferential Global Positioning Sys-tem (GPS) processing7, 8 or (2) useof a different topographic data setto supply tie points to estimate thebaseline tilt. In the second method,differences are computed betweenselected points in a candidate high-resolution DEM and the tie points.These differences can be used toestimate the baseline tilt for thecandidate DEM, which can then becorrected for baseline-tilt errors.

Estimation of the interferomet-ric phase is directly available onlyfrom the complex SAR imagery inthe “wrapped” form (i.e., boundedby zero and 2p). Consequently, thephase must be “unwrapped”; that is,multiples of 2p must be added tothe wrapped phase to derive theinterferometric phase and subse-quently the elevation estimate. Forrelatively smooth terrain, it is ade-quate to use a simple unwrappingalgorithm in which multiples of 2pare added whenever absolutewrapped-phase differences betweenadjacent samples are more than p.This simple procedure is exact aslong as the absolute unwrappedphased difference is less than p.However, for noisy data and roughterrain, simple unwrapping fails,and more complex algorithms arerequired.

The major problem with phaseunwrapping is aliasing (undersam-pling) within the interferogram.Such aliasing is caused by noise,layover (i.e., superposition of sig-nals within the same range bin),and too-large baselines. Severaltechniques have been proposed todeal with phase-unwrapping prob-lems (e.g., Roth9); Ghiglia andPritt3 have recently comparedthem with one another. Anotheraspect of the phase-unwrappingprocess is the need for tie points. Aregion that has been unwrappedcan have the various relative fac-tors of 2p correct but still have anoverall absolute 2p ambiguity for a

NICAL DIGEST, VOLUME 20, NUMBER 3 (1999)

local region. Consequently, a tie point is required to fixthe absolute value of 2p, and special IFSAR processinghas been proposed (e.g., Imel10). However, for high-resolution IFSAR, the effective local region may bedisjoint (divided by boundaries of aliasing) and small(Fig. 3).

One possible way to deal with this problem of dis-joint and small regions for phase unwrapping is tointroduce a short second baseline. This second baselinecan be designed to produce an interferogram that doesnot exceed the range of 2p over the swath of interest.11

This allows computation of a coarse height map with-out use of any form of phase unwrapping. The cyclenumber ambiguities in the phase data derived from thelarge baseline can then be resolved by reference to theheights computed from the small-baseline data. Forairborne IFSAR systems, adding another antenna at ashort baseline is feasible. This approach is used for theDHC-7 aircraft IFSAR system that is being developed

Figure 3. Phase-unwrapping challenge for terrain jumps.Top: Conceptual elevation z profile in azimuth direction x. Be-cause about 5 km of baseline is needed to get the elevationaccuracy for space-based IFSAR, the phase f goes through acycle for approximately every 6 m of elevation change.Center: Corresponding wrapped phase W(f) that would be de-rived from processing two complex SAR images. Bottom: Phase-unwrapping results. Conventional phase-unwrapping algorithms(labeled as actual) work well if there are no isolated terrain jumps.However, jumps can cause such algorithms to fail and give errorscorresponding to the wrong multiple of 2p. The solution is to mergethese data with another digital elevation model (DEM) that may benoisier but lacks this ambiguity problem. Such a DEM can beacquired by processing data from a third antenna at a shortbaseline or by performing SAR stereoprocessing on two-antennadata.

Pha

se, f

10

5

0

Wra

pped

pha

se, W

(f)

Ele

vatio

n, z

(m

)

2π

0

4π

2π

0Azimuth direction, x

Desired

Actual

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 20, NUMBER 3 (1

HIGH-RESOLUTION INTERFEROMETRIC SAR FOR DISCOVERER II

at Sandia National Laboratories under the sponsorshipof the Army Rapid Terrain Visualization Program.

For space systems, however, adding an antenna canhave considerable cost impact. Fortunately, a coarseheight map can be computed by SAR stereoprocessing.In fact, such stereoprocessing (i.e., microregistration) isrequired for high-resolution IFSAR to have acceptablecoherence because of a height-induced misregistrationeffect. Consequently, for a judicious design, the cyclenumber ambiguities for small regions can be resolvedby reference to the heights from the stereo-DEMs. Thisalternative does not require a third antenna, and theauthor has been shown that it works in config-urations of interest to space-based systems.

PHENOMENOLOGY: THE EFFECTSOF ENVIRONMENTAL ANDSCENE VARIATIONS

As with all techniques, topographic mapping usingSAR has limitations that must be recognized. Forexample, just as optical stereoprocessing cannot beperformed in regions of obscuration and deep shadow,SAR-based topographic maps will have data voids inregions of radar layover, shadow, and low cross section.Because radar penetration of foliage (at X band) islimited, SAR-based topographic maps generally corre-spond to the tops of trees rather than to the ground.Urban areas are problematic because of multipath ef-fects. All of these phenomena occur because SAR-based techniques measure the elevation of the radar-reflective surface.

Estimating elevation in vegetation with IFSAR is asubstantial technical challenge. As recently shown inan analysis of high-resolution IFSAR aircraft data(M. W. Roth, unpublished data, 1997), the major ef-fect in vegetated areas is the existence of significantamounts of layover and shadow due to the extremeroughness of the reflective surface. In regions of shadow,elevation cannot be estimated, but in layover regionsthere is significant decorrelation. The layover decorre-lation results from the mixing of radar scattering ele-ments having large differences in elevation within thesame range cell. Such decorrelation causes large errorsin elevation estimates. Because decorrelation is smallerfor small (versus large) IFSAR baselines, one possibleapproach is to estimate elevation in vegetation usinga small baseline.

Even for a small baseline, however, large elevationerrors always occur in layover regions because the erroris fundamental to the physics of radar scattering. Mix-ing returns from scatterers at different elevations withinthe same range bin always produces an elevation esti-mate that is a mixture of the elevations (i.e., the radarcentroid). In layover, the elevation differences can be

999) 301

M. W. ROTH

large, and the resulting elevation mixture can be verydifferent from the elevations of any of the scatteringsurfaces. The smaller decorrelation of small-baselinesystems does not help. In fact, the smaller decorrelationof small-baseline systems can be misleading—the largedecorrelation of large-baseline systems clearly identifiesregions of layover, whereas the decorrelation of small-baseline systems is so small that it does not separatelayover from other sources of decorrelation, such asnoise or misregistration.

One method for reducing holes and improving errorsfrom bad data regions is to re-collect data for the samearea from a different aspect angle and merge the data.In this way, shadowed regions from one aspect need notbe shadowed from a different aspect, and elevation datacan be selected from the nonshadowed region. Thismethod can work if shadowed regions can be clearlyidentified, and identification is possible if the varioussystem and processing noises are low enough. Such ascheme can also work for layover regions if the layoverregions can be clearly identified. In this regard, thegreater decorrelation of large-baseline systems makesidentifying layover regions easier than use of the small-er decorrelation of small-baseline systems.

In addition to the usual effects, such as range walkand azimuth shift, IFSAR has motion-induced effectsthat can occur and that influence SAR images. Whentwo SAR phase centers are separated in the along-trackdirection, target motion along the line of sight causesa differential phase shift between the correspondingcomplex-SAR-image pixels. Although this effect canbe used to measure phenomena such as ocean cur-rents,12 it can also induce errors in IFSAR topographicmapping if the target area has wind-generated motion.Random motion from vegetation can also contribute toloss of coherence, particularly for two-satellite IFSARdesigns with significant time periods of along-trackseparation. In many cases, judicious selection of satel-lite separations can mitigate motion-induced IFSARerrors.

SAR images of cultural areas are complicated byurban effects such as layover, shadows, multipath ef-fects, motions, corner reflectors, and varying reflectiv-ity. Such effects could contribute to significant eleva-tion error for IFSAR. Modeling of urban effects hasshown that characteristic signatures could potentiallybe used to minimize errors in IFSAR elevation estima-tion. For example, multiple heights can occur at a pixellocation for tall buildings. This information can poten-tially be used to detect the front edge of large urbanstructures.

There would be significant advantages if challengingregions (layover, shadow, low radar cross section) couldbe identified automatically. Automatic identificationcould enable special processing to be directed to such

302 JO

areas for minimizing errors. In addition, IFSAR cangenerate high-resolution error maps that can provide aprecise estimate of the error at each DEM post. Con-sequently, because of the great utility of such informa-tion, automatic identification would be useful for esti-mating post-by-post error, as well as for identifying thechallenging regions and forwarding the information tousers as part of the overall DTED product.

Examination of DEMs generated by a level 3 IFSARaircraft system13 and comparison with optical DEMshave shown that there can be other kinds of artifactsof unknown origin. For this system, the radar datarepresented individual trees at 1/4 of their actualheights, closed canopies at 1/2. Individual trees in theradar DEMs were displaced toward the flight path by10 to 20 m. As many as 10% of the hilltops containedlarge errors (>7 m) in spite of high correlation values.Ravines in the radar data were shifted toward the flightpath by an average of 10 m, giving rise to the perceptionthat the ravines in the radar data were too shallow.Radar elevation and slope errors varied systematicallywith the true grazing angle. The better radar elevationdata occurred at a grazing angle of approximately 53°and on a positive terrain slope between 0 and 10°. Mostlikely, these artifacts are the consequence of the specificsignal processing applied and could be essentially re-moved by improved processing.

SYSTEMS ISSUESDTED timeliness requirements are derived from the

Army’s need to rapidly deploy forces into any theaterupon command.14 A global data set at a 30-m resolutionneeds to be collected only once. However, regionalcollection at a 1-m resolution requires a timely collec-tion capability. Such topographic data sets are requiredto meet crisis-response and Force projection require-ments. These timeliness requirements were specified as20 3 20 km in 18 h, 90 3 90 km in 72 h, and 300 3 300km in 12 days. For example, consider the utility of thesesizes by examining Bosnia. A 20 3 20 km area wouldcover the local region around Tuzla, a 90 3 90 km areawould cover the region around and between Sarajevoand Tuzla, and a 300 3 300 km area would cover all ofBosnia. These regional timeliness requirements arechallenging for any mapping technology.

For high-resolution IFSAR system concepts, themost significant sources of elevation error are

• Vertical errors due to interferometric phase noise• Baseline calibration errors• Phase unwrapping errors• Ephemeris errors

Ephemeris errors can be minimized with precise track-ing techniques (e.g., Bertiger et al.15). Less significant

HNS HOPKINS APL TECHNICAL DIGEST, VOLUME 20, NUMBER 3 (1999)

HIGH-RESOLUTION INTERFEROMETRIC SAR FOR DISCOVERER II

sources of error include propagation errors, horizontalerrors due to SAR image fidelity, mosaic boundaryerrors, and long-term receiver phase balance. Interfer-ometric phase noise can come from a number of sourc-es. These include SAR image noises, either thermal ormultiplicative. Decorrelation significantly increasesphase noise and can occur from baseline separation,volume scatter, or temporal instability. Image-pairmisregistration can cause decorrelation. Finally, inter-ferometric phase noise can come from differential chan-nel phase errors. Several models of interferometricphase noise and its influence on elevation accuracyexist (e.g., Rodriguez and Martin16; Carrara et al.17).

In addition, for a bistatic IFSAR operation, somemeans must be provided to ensure that signals are re-ceived at the proper time and frequency. Timing synchro-nization is required to synchronize the collection intervalfor the transmit/receive radar and the receive-only radarso that both radars collect the same swath. This iscritical for appropriate positioning of the data window.Techniques for pulse-timing synchronization includeGPS time transfer and direct reception of the transmit-ted pulses. Frequency synchronization is required toposition the receive-only signal within the azimuthprefilter bandwidth. This need implies that the relativevelocity between the satellites must be measured.

Finally, independent master oscillators on the twosatellites, as well as a bistatic configuration, meansthat common-mode monostatic cancellation of low-frequency stable-local-oscillator noise is not appropri-ate. The result is an increase in the integrated sideloberatio and the consequently multiplicative noise ratio(MNR). Reduction in the relative phase noise betweenthe two oscillators is required to improve the MNRof the SAR map formed by the receive-only radar.Techniques for ensuring phase coherence include usingatomic clocks, phase-locking the two separate receivelocal oscillators, and employing auto-focus to correctlow-frequency phase errors. Several techniques havebeen used in interferometric radio astronomy to ensurelocal-oscillator phase coherence between spatially sep-arated receivers. Most of these techniques involvetransmitting signals derived from the separate localoscillators to opposite receiving sites or to a central siteand measuring phase differences. For closed-loop sys-tems, phase correction is transmitted by a narrowbandlink to phase-lock the remote oscillators. Alternatively,the measured phase differences could be used to correctthe data during processing.

APPLICATION EXAMPLETactical reconnaissance (“recce”) data typically con-

tain a target object within either an infrared/optical ora SAR image. Precision targeting can be achieved ifan accurate geolocation for the target object can be

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 20, NUMBER 3 (

obtained. However, recce data typically do not haveadequate geolocation accuracy to meet targeting re-quirements. To get improved geolocation accuracy,recce data are registered to some data set with bettergeolocation accuracy, and the geolocation of the targetis then inferred from the registered recce data. Howev-er, previous databases are sparse, require labor-intensiveproduction, and do not meet the needs for very precisetargeting. Consequently, new data are needed that canmeet precise targeting needs.

High-resolution IFSAR data can meet the needs foran accurately geolocated database for registration andprecise targeting of recce data. For example, becauseIFSAR DEMs can be accurately geolocated (to lessthan 1 m with proper system engineering), and becausethe SAR images that were used to form the DEM areexactly registered to the DEM, these SAR images areaccurately geolocated to the same degree as the IFSARDEMs. These SAR images, in turn, can be comparedwith recce data to accurately geolocate the recce data.

The methodology for accurately registering reccedata to SAR images can sometimes be problematic,however. For example, accurately registering two differ-ent SAR images is difficult if collection geometriesdiffer significantly. In addition, accurately registeringinfrared/optical data with SAR data can also be prob-lematic. Registration also is frequently a manual processin which an operator selects features for common ob-jects in the two images to be registered. Typically,cultural features are selected. If the images do notcontain enough cultural features or if manual opera-tions are too slow, then the methodology cannot meetprecise targeting requirements.

In 1995, APL demonstrated the potential forIFSAR technology to produce an improved registrationmethodology that overcame the problems just listed.An APL IFSAR simulation showed that the IFSAR-derived DEMs could generate synthetic optical imagesthat could be accurately registered to simulated recceoptical images. The principle was established by com-paring synthetic images directly derived from a sourceDEM with synthetic images derived from the corre-sponding IFSAR-simulated DEM over a wide range oftopographic conditions. The registration technique wascompletely automated. The IFSAR system simulatedwas the conceptual system for the Topographic Satellite(TOPSAT), designed to generate level 2 DTED.18

The Discoverer II Program has asked APL to furtherdemonstrate this methodology, both by using realflight-test data and by extending the methodology toSAR images. In particular, a DEM derived from one setof flight-test data will be used to simulate SAR imagesfor a different collection geometry corresponding toanother set of flight-test data. The SAR image fromthe second set of flight-test data will be registered to

1999) 303

M. W. ROTH

the simulated SAR data, and the accuracy will bedemonstrated.

CONCLUSIONSA space-based system for high-resolution topograph-

ic mapping would have to measure elevation on theorder of tens of centimeters at a resolution of 1 m.In addition, collection would have to be at a range onthe order of 1000 km for a contiguous area of the orderof 100,000 km2 within less than 2 weeks. The creationof such a capability would represent a significant tech-nical achievement in the history of science and engi-neering. The impact of the capability would be revo-lutionary for both intelligent preparation of thebattlefield and precision engagement.

The emergence of technology for high-resolution IF-SAR has created the potential for such a capability toexist. The technical challenges, while not insurmount-able, are significant. The Discoverer II Program hasembraced these challenges with risk-reduction efforts.The success of these efforts will lead to the technologymaturation needed for a space-based demonstration.

REFERENCES1Hughes, M. T., Discoverer II: Space-Based HRR-GMTI/SAR DemonstrationProgram, Discoverer II Joint Program presentation, available at http://www.laafb.af.mil/Special_Interest/disc2/Hughes2.pdf (accessed 8 Mar 1999).

2Graham, L., “Synthetic Interferometer Radar for Topographic Mapping,”Proc. IEEE 62(6), 763–768 (1974).

304 JOH

3Ghiglia, D. C., and Pritt, M. D., Two-Dimensional Phase Unwrapping, Theory,Algorithms, and Software, John Wiley & Sons, New York (1998).

4SRTM Project Office, SRTM: Shuttle Radar Topography Mission, SharonOkonek, Web site administrator, NASA Jet Propulsion Laboratory, availableat http://www-radar.jpl.nasa.gov/srtm/ (accessed 8 Mar 1999).

5Zebker, H. A., and Villasenor, J., “Decorrelation in Interferometric RadarEchoes,” IEEE Trans. Geosci. Remote Sensing 30(5), 950–959 (1992).

6Zebker, H. A., “Accuracy of Topographic Map Derived from ERS-1Interferometric Radar,” IEEE Trans. Geosci. Remote Sensing 32(4), 823–836(1994).

7Yunck, T. P., TOPSAT Orbit Determination and Satellite–Satellite VectorMeasurement with GPS, Jet Propulsion Laboratory, California Institute ofTechnology, Pasadena, CA (18 Jun 1993).

8Yunck, T. P., and Hajj, G. A., Ionosphere and Multipath Errors in the TOPSATVector Measurement, Engineering Memorandum 335.0-94-001, Jet PropulsionLaboratory, Pasadena, CA (8 Apr 1994).

9Roth, M. W., Phase Unwrapping for Interferometric SAR by the Least-Error Path,Technical Memorandum F1B0-95U-019, JHU/APL, Laurel, MD (30 Mar1995).

10Imel, D. A., “Accuracy of the Residual Delay Absolute Phase Algorithm,”IEEE Trans. Geosci. Remote Sensing 36(1), 322–324 (1998).

11Jakowatz, C. V., Jr., Wahl, D. E., and Thompson, P. A., “AmbiguityResolution in SAR Interferometry by Use of Three Phase Centers,” inAlgorithms for Synthetic Aperture Radar Imagery III, Vol. 2757, Society of Photo-Optical Instrumentation Engineers, Bellingham, WA, pp. 82–91 (1996).

12Goldstein, R. M., and Zebker, H. A., “Interferometric Radar Measurement ofOcean Surface Currents,” Nature 328, 707–709 (1987).

13Norvell, F. R., Evaluation of ERIM’s “FSARE” Digital Elevation Models ofCholame Hills and Camp Roberts, California, in Support of GeoSAR, U.S. ArmyTopographic Engineering Center, Fort Belvior, VA (Dec 1996).

14Arnold, E. J., Army Digital Topographic Data (DTD) Requirements, ATSE-AC(115), Department of the Army, TRADOC (18 Feb 1998).

15Bertiger, W. I., Bar-Sever, Y. E., Christensen, E. J., Davis, E. S., Guinn, J. R.,et al., “GPS Precise Tracking of TOPEX/POSEIDON: Results and Implica-tions,” J. Geophys. Res. 99(C12), 24,449–24,464 (1994).

16Rodriguez, E., and Martin, J. M., “Theory and Design of InterferometricSynthetic Aperture Radars,” IEE Proc.-F 129(2), 147–159 (1992).

17Carrara, W. G., Goodman, R. J., and Majewski, R. M., Spotlight SyntheticAperture Radar, Signal Processing Algorithms, Artech House, Boston, MA(1995).

18Zebker, H. A., Farr, T. G., Salazar, R. P., and Dixon, T. H., “Mapping theWorld’s Topography Using Radar Interferometry: The TOPSAT Mission,”Proc. IEEE 82(12), 1774–1786 (1994).

THE AUTHOR

MICHAEL W. ROTH received a B.A. degree in physics and mathematics fromMacMurray College in 1971, and M.S. and Ph.D. degrees in physics from theUniversity of Illinois in 1972 and 1975, respectively. After serving as a ResearchAssociate at Fermilab, he joined APL in 1977 and is a member of the PrincipalProfessional Staff. He has served as a Project Leader and Chief Scientist fornumerous efforts relating to sensors, signal and image processing, guidancesystems, neural networks, and mapping systems. Dr. Roth has also served onseveral APL and government committees and panels. He has published papers ona variety of topics, has received a number of academic and professionalawards, and has served as the Editor-in-Chief of IEEE Transactions on NeuralNetworks. He is currently the Project Manager for Mapping Systems inAPL’s Power Projection Systems Department. Dr. Roth’s e-mail address ismichael.roth@jhuapl.edu.

NS HOPKINS APL TECHNICAL DIGEST, VOLUME 20, NUMBER 3 (1999)