EARTH VIEW THIRD MILLENNIUM: A STUDENT DESIGNED 1 FT. RESOLUTION, INTERFEROMETRIC IMAGING SPACECRAFT

AIAA 2001-4771 EARTH VIEW THIRD MILLENNIUM: A STUDENT DESIGNED 1 FT. RESOLUTION, INTERFEROMETRIC IMAGING SPACECRAFT John Vesecky, Ricardo Torres, Ajay Bharadwaj, Gabriel Loya, Nicholas Avlas and Scott Molton School of Engineering, University of California at Santa Cruz, 1156 High St., Santa Cruz CA 95064 831-459-4099, fax 831-459-4829, email vesecky@soe.ucsc.edu ABSTRACT During the past academic year the University of Michigan (Aerospace Engineering) and the University of California at Santa Cruz (School of Engineering) collaborated in the conceptual design of a low-cost spacecraft capable of imaging the Earth in visible light at 1 ft. resolution. The spacecraft is called Earth View Third Millennium (EV-3M). The aerospace aspects were handled by the University of Michigan Team, while the communication, spacecraft computer and aspects of the imaging system were investigated by the UC Santa Cruz Team. The imaging system uses three 0.4 m apertures operating in an irregularly spaced, interferometric array on a 2-meter long truss. Measurements of the mutual intensity function J (made using combinations of the three apertures as the spacecraft rotates) are used to subpixelate the 1-meter resolution cells resolved by the 80-cm apertures. The communication system design focussed on the trade-off between dedicated ground stations and the Tracking and Data Relay Satellite System (TDRSS). TDRSS was found to be far more cost-effective. However, it appeared advisable to retain a single dedicated ground station for redundancy. The spacecraft computer system was straightforward in terms of spacecraft control, featuring a Honeywell Radiation Hardened Power PC microprocessor. Special purposed processors were needed for the sensor data processing. INTRODUCTION The Earth-View Third Millenium project supports the Air Force Research Laboratory s (AFRL) interest in low cost, imaging satellite design. Innovative designs are needed to improve the affordability, survivability, and availability of space systems for Earth imaging. In response to the AFRL desire for a mission design, the Space System Design class at the University of Michigan has developed a first-order design approach to address this endeavor. In order to further develop this design the University of Michigan sought help from the University of California at Santa Cruz (UCSC). UCSC s design class was responsible for Sensor and Data Processing, Computer Control System, and Copyright 2001 by John Vesecky 1 Communication and Data Compression design. This paper presents a summary of the UCSC contribution that supports the University of Michigan work reported at this conference 1. High resolution imaging of the Earth s surface has many important applications including agriculture, mapping, environmental study and monitoring as well as emergency action in natural disasters. One meter resolution images of urban areas allow monitoring of traffic and parking patterns as well as better identification of vegetation types. At present there are several civil satellites with high resolution imaging capability. A summary of these is shown in Table 1 along with the baseline requirements for the Earth View Third Millennium (EV-3M) mission discussed here. Interesting facts about the Ikonos 1 m resolution system including image samples can be found on the Ikonos website (http://www.spaceimaging.com). Table 1 Comparison of Current High Resolution Missions IKONOS Quickbird OrbView-4 EV-3M Spacecraft Total Mass of s/c 700-726 kg 931kg (wet) Design Life 5 years >5 years 5 years 10years Launch Sep-99 not operational Apr-01 2010 Orbit: Altitude 681km 601 km 470 km 600Km Revisit Time at 2.9 days at 40 deg lat. less than 3 days Best Resolution 1.5days Telescope Telescope Size 1524mm (legnth) x 787mm (diameter) 3x 0.4m Mass of telescope 171 kg - 226 kg 380 kg 3X 2Kg Pan Cromatic Wavelength Range 400-900 nm 450-900 nm 450-900 nm 400-900nm Pan Resolution 1 m 1 m 1 m 1ft (~0.33m) Multispectral Resolution 4 m 4 m 4 m 1m As a future step in high resolution imaging of the Earth we adopted the following design goals for EV-3M: Resolution: 1 ft. or better in the 0.4 to 0.9 µ band Spectral Bands: Panchromatic and multispectral (0.4 to0.9 µ), SWIR MWIR (0.9 to 3.5 µ) SNR in image: 20 db or better Image geolocation: 10 m or better Area coverage: 100 4 x 4 km frames per day Revisit time: 1.5 days or better

Latency: < 45 min. from collection to delivery Design lifetime: 10 years A concept image from the Michigan design study 1 is shown in Fig. 1. Prof. David Hyland 2 has formulated the interferometric imaging process in terms of inversion of the Huygens- Fresnel Principle. Thus, observations on an observation plan O can be used to construct the image I(Q) of an illuminated or luminous object on an image plane Q that is near or at the object. The required observations are the mutual intensity function J(P 1, P 2 ) = <U (P 1 ) U*(P 2 )> at a number of pairs of points P 1 and P 2, such that a range baselines (between P 1 and P 2 ) in terms of both length and orientation are used. For a given angular resolution requirement δθ the set of baselines needs to cover baselines as long as D λ/δθ and a wide range of orientations (mainly near the plane perpendicular to the line of sight between the observer and the object). This method of obtaining high resolution images has been used extensively in radio astronomy over decades 3 and has recently included baselines as long as the distance from the Earth to a LEO satellite. Fig. 1. EV-3M concept of an interferometric imaging satellite viewing the Earth (After Gano et al. 1 ) Project organization flowed from Air Force overall management (Capt. Nick Hage) to the prime contractor, the FXB Corp. (UM Aero 483/583 design class, Profs. David Hyland and Pierre Kabamba) to the subcontractor, California Dreamworks Inc. (UCSC EE 128 design class, Prof. John Vesecky). INTERFEROMETRIC IMAGING CONCEPT As higher resolution imaging systems are required for Earth and astronomical imaging, ever larger apertures are needed. These larger apertures, e.g mirrors, become increasingly massive, requiring ever larger and more expensive spacecraft and launch systems. Lightweight mirrors, sparse apertures and interferometric systems have been considered to develop low-cost, high resolution imaging systems. In an interferometric system multiple, but relatively small, apertures are used in pairs to estimate the mutual intensity function for a number of baselines, distributed in length and orientation. When these baselines cover a region of size D sufficiently well, they allow an image to be formed that has an angular resolution λ/d, i.e. the resolution of a mirror the size of the longest baseline, namely D. Thus, a number of smaller apertures, used in pairs and moved around can form an image with resolution much finer than any of the apertures could by itself. 2 Imaging Geometry Fig. 2. Object s is imaged on imaging surface Q by pairs of coherent observation at points P on observation surface O (After Hyland 2 ). The interferometric scheme shown in Fig. 2 is implemented here as show in Fig. 1 and in Fig.? below. Consider the three apertures attached to a linear truss with irregular spacing (Fig. 1). For a single position of the beam we can obtain three measurements of J(P 1, P 2 ). Now if we rotate the truss we can obtain many orientations of these three baseline lengths and approximately span the space of pairs of points P 1 and P 2. All these measurements are used to approximate the inversion integral given by Hyland, namely I(Q) = (1/λ 2 ) dp 1 dp 2 {α(p 1,Q) α *(P 2,Q) [(e ik(s2-s1) )/(s 1 s 2)] J(P 1,P 2)} (1) where I is the image on the image plane Q, s 1 & s 2 are the ranges from from Q to P 1 or P 2 and α is a complex function dependent on the observational geometry. Thus, the image I(Q) can be recovered by knowing the

observational geometry and measuring enough values of J(P 1, P 2 ) to accurately approximate the integral of (1). Using Hyland s formulation and approximations we have simulated an interferometric imaging situation similar to that of EV-3M. In Fig. 3 we show the observational geometry for interferometric imaging of an object at nadir from a satellite in LEO orbit. 10 D 100 Observation Region DzΖ z = range to _ z satellite = range to orbit geosynchronous orbit 500 Fig. 4. Simulation of interferometric imaging of 3 discrete points from a LEO satellite platform in the visible light. Image EARTH Fig. 3. Obsevational geometry for imaging of a region on the Earth s surface from low Earth orbit (z = 750 km). All three mirrors image a common area which is 1 x 1 km at nadir, yielding 1000 x 1000 1 m pixels. The optical system allows for transferring the image from each aperture to an interferometer subsystem where J can be measured for each pixel. The values of J collected for each 1 m pixel are used with (1) to subpixelate each 1 m pixel with 9 subpixels at 1 ft. For this simulation we took D = 1.1 m, z = 750 km and the wavelength = 0.5 µ. Rather than take a deterministic set of points the set of points were selected randomly within a box of size D. Using values of J from 10, 100 and 500 pairs of points within D we simulated the imaging for three points of mid-level intensity against a dark background. Fig. 4 shows the results of the simulation. The color scale in the middle shows intensity level of a given pixel on a scale from 0 to 20 linear units. With only 10 values of J the image is noisy, but with 100 values it is good and with 500 it is excellent. SENSOR AND DATA PROCESSING In this conceptual design we use an interferometric imaging approach. We note that wavefront reconstruction may well prove to be a better imaging approach 1 due to the tolerence and stability problems inherent in an interferometric approach. The issue will likely be decided by technological advances in the instrumentation used to implement either approach. An overview of the spacecraft concept of Fig. 1 is shown in Fig. 5. This concept features a 2 m truss to which three optical apertures (mirrors) are attached. 3 Irregularly spaced apertures optical aperture truss beam 2m Spin Axis (2 cps) Solar Cells on bottom of truss Light paths (typical) Interference Detector System 40 cm aperture (typical) Fig. 5. Overview of interferometric imaging satellite. Apertures are irregularly spaced to yield a good distribution of baselines and orientations as discussed in the section above. The satellite design and most subsystems are discussed by Gano et al. 1. resolution. A variety of baseline lengths are collected by using the three irregularly spaced apertures to collect 3 different baseline lengths. A variety of baseline orientations are collected by spinning the spacecraft about an axis perpendicular to the truss as shown in Fig. 5. The concept for the interferometric estimation of the mutual intensities J is shown in Fig. 6.

EV-3M Interferometric Optical and Detector System Light Beam Paths, Optical Path Delay Path Delay, ( ) Modulator (t) Sensor Control & Measurement Computer 40 cm aperture telescopes Detector Array Pixel Tracker & <U U*> Measurement System Incoming light from Earth's surface, Optical Path Delay (t) Detector Array Beam Steering Secondary Mirrors Detector Array, Optical Path Delay (t) Beam Steering (typical) Fig. 6. EV-3M interferometric and optical detector system. The light from the scene to be imaged arrives from the left. The three mirrors feed three interferometers with optical path delays in one leg. The 1000 x 1000 pixel beam is focussed such that interference takes place on a pixel by pixel basis and the resultant interference signal is incident on the three 1000 x 1000 pixel detector arrays. The light from the scene to be imaged arrives from the left. The three mirrors feed three interferometers with controllable optical path delays in one leg. The 1000 x 1000 pixel beam is focussed such that interference takes place on a pixel by pixel basis and the resultant interference signal is incident on the three1000 x 1000 pixel detector arrays. The optical path delays are modulated in time so that over a short time period the detected signal allows observation of the resultant interference pattern and from this pattern J can be estimated. The values of J estimated for each pixel through a half rotation allows a variety of baseline orientations. Thus, each 1 m pixel resolved by the 40 cm apertures can be subpixelated at 1 ft resolution and a 1 ft resolution image of a 1 x 1 km scene can be 4 created. By repeated applications of this procedure through a number of revolutions of the spacecraft about its spin axis a sequence of images can be created to image a larger area or by applying filters a multispectral image of a given area can be created. As the spacecraft spins the interference images on the detectors rotate. Since the beams from all the mirrors rotate in concert the interference image is consistent, but rotating in time. This requires that the pixels on the detectors be tracked so that values of J estimated from the interference patterns (in time) are associated with the correct pixel on the ground. The pixel tracker subsystem in Fig. 6 performs that function. We also note that we need to sample the pixels in the array

rapidly in order sample the interference signals as the optical delays are varied in time. We now consider the detector arrays. Charge coupled diode (CCD), complementary metal oxide semiconductor (CMOS) and charge injection device (CID) arrays were considered. None of these technologies meets our needs very well, but for different reasons. CCD arrays are a mature technology, but are continually improving. back-thinned CCD s have a quantum efficiency of over 90% in the middle of the visible spectrum compared to some 40% for a standard CCD. However, CCD arrays that are fast enough in terms of read out speed have high readout noise. This lowers SNR and is undesirable. Current CMOS array technologies have a high quantum efficiency and low readout noise, but their read out speeds are too slow. CID arrays offer very high read out rates, but the quantum efficiency is low and the technology is not mature. We look forward to improvements in all these imaging array technologies. At present the best detector array we could find is the Sarnoff 1024 x 1024 CCD array. This detector has a 37 electron/pixel readout noise level at 300 frames per second with a quantum efficiency of 55 to 70%. The interferometric imaging system described here requires substantial digital signal processing (DSP) power. However, the available technology is sufficient to the task. In a trade study the Texas Instruments TI6701 was picked as the best current choice. This DSP unit runs at 167 MHz with 1 GFLOP processing capability. It consumes only 1.6 watts of power. The primary issues for the sensor system are as follows: Metrology and stability of the interferometer systems Short sampling time may lead to inadequate signal to noise Detector array technologies available at present all have shortcomings. However, we conclude that an interferometric imaging concept is feasible, but difficult. The wavefront reconstruction concept 1 may prove to be better. DATA COMPRESSION AND COMMUNICATIONS We considered a variety of communications system architectures to achieve the main communication requirement to deliver 100, 4 x 4 km images per day to customers. This is not a particularly taking requirement if one has continuous access to a communications channel 2.2 Mbps. After considering a number of options involving LEO and GEO platforms as well as a ground station network we found that TDRSS offered a very economical solution. For redundancy we also recommended a single ground station. The architecture is shown infig. 7 below. Fig. 7. Communications system architecture. Note the primary use of TDRSS with a single ground station back up. This system is economical to build due to the competitive market in hardware for TDRSS communications. Using the 1998 rates for communication we calculate that the cost for transferring the aforementioned 100 images/day is only $35,000 per year. This is a very small cost compared to running even a single ground station and the delivery time and latency are low. Data compression would further reduce the communications load. We evaluated five data compression schemes and concluded that JPEG 2000 was the best. The advantages include the following: Lossless and lossy compression options Superior low bit rate compression performance Support for large images, up to 64 x 64 kpixels Open architecture. It is a standard that will serve applications for an extended period. 5

In summary our communications system design has the following features: It is easy to see how this scheme could be applied to the EV-3M satellite and potentially increase the resolution. Communications: single store and forward satellite with single ground station and TDRSS Data compression: JPEG 2000 Instrumentation: mainly off the shelf items. COMPUTER CONTROL SYSTEM The computer control system is conventional with an emphasis on conservative design to contribute to robust spacecraft performance. The spacecraft control system (SCS) is required to manage a variety of spacecraft onboard components, e.g. thermal and power systems, schedule DSP operations and coordinate and manage transfer of image data to the ground. After a trade study we picked the radiation hardened power PC (RHPPC) single board computer, running the VxWorks real time operating system. A second board could be added for redundancy, although the IBM 603e processor has only a 1% failure rate for 15 years in space. SUPER RESOLUTION AND EV-3M Since the EV-3M is required to image a small number of areas each day, seven 4 x 4 km areas per orbit, it is generally practical to view a given image scene for a some time, say 100 seconds or more. This would allow collection of a number of images of the same scene, but from slightly different aspects. Such a situation is a good one for the application of multiple image super resolution techniques. Multiple image super resolution (MISR) works by postulating a fine image grid and then overlaying a number of coarser images upon it each coarser image having a slightly different placement. It is as if one dithered the coarser images over the finer pixel grid. One then solves a set of linear equations to find the intensity values of the pixels on the finer resolution grid. This technique is well known and works well for situations such as the multiple images collected by a video camera. Fig. 8 shows the application of super resolution in such a situation by Prof. Peyman Milanfar in the EE Dept. at the University of California at Santa Cruz. In this case one can easily see the improvement in the image after a dozen or more images were combined in a super resolution process. Fig. 8. Super resolution applied to a sequence of night vision camera images. Note the improvement in the image at right. (After Milanfar 4 ) Most applications of super resolution, such as Fig. 8 are in situations where the imaging system that is collecting the multiple coarse images is not diffraction limited. In the case for EV-3M we seek to obtain resolution below the diffraction limit of our 40 cm collecting aperture which have a 1 m resolution at nadir. We have not been able to find compelling evidence that super resolution, using the techniques discussed here, can achieve resolution much below the diffraction limit of the collecting system. Nevertheless, we recognize that super resolution techniques are useful for removing atmospheric effects and other distortions so we recommend that provision be made for the collection of multiple images of a given area so that super resolution techniques can be applied. Below we show an image of the Martian surface constructed by a super resolution technique using multiple images from a recent NASA Mars Lander. Fig. 9. Super resolution image of the Martian surface using data from the NASA Mars Lander. (After NASA JPL website.) 6

SUMMARY AND CONCLUSIONS The EV-3M system is designed to bring higher resolution (1 ft.) visible and multispectral imagry to the user for new and existing applications in the commercial, government, military and science communities. To make this higher quality imagry more widely available the cost must be kept low for both the system itself and for operation. Our cost goal is under $100 million for the spacecraft and launch. The design of this system was undertaken by student design classes at the University of Michigan and the University of California at Santa Cruz. The novel concepts of the design begin with the use of interferometric and wavefront reconstruction approaches using multiple small apertures rather than a single large aperture. The UCSC team pursued the interferometric approach. This method has yet to be applied in space and has significant challenges in the metrology and stability of the interferometer systems for measuring the mutual intensity J. Existing detector array technologies have significant shortcomings for this application with high speed readout CCD arrays being the current best option. CMOS and CID technologies offer promise for future design inprovements. Super resolution techniques offer opportunities for image improvement and we make provision for collection of multiple images of the same scene. However, we conclude that the use of multiple image super resolution (MISR) for improving resolution beyond the diffraction limit of the optical system is not yet proven conclusively and should not be relied upon to achieve the resolution requirement of EV-3M at present. Overall we conclude with the Michigan team that design and construction of a low cost Earth imaging satellite is feasible and worth pursuing for the many new applications that such a system would enable. at the Naval Postgraduate School was generous with his time and talents in supporting the UCSC design course. REFERENCES [1] S. Gano, D. Hyland & P. Kabamba, A baseline study of a low-cost, high-resolution imaging system using wavefront reconstruction, AIAA 2001, paper AIAA 2001-4751, 2001. [2] D. Hyland, Interferometric imaging concepts with reduced formation keeping constraints, AIAA 2001, paper AIAA 2001-4610, 2001. [3] K. Rohlfs, Tools of Radio Astronomy, Springer- Verlag, Berlin, 1986. [4] P. Milanfar, private communication, 2000. ACKNOWLEDGEMENTS We received much help from faculty and research scientists at a variety of institutions. We thank Capt. Nick Hage of the US Air Force for comments and advice on the overall objectives of the project and Natalie Clark of the Air Force Research Laboratory of technical advice. At the University of Michigan we thank David Hyland and Pete Hansen. At the University of California at Santa Cruz we thank Peyman Milanfar, Steve Petersen, Ali Shakouri and Richard Stover. We received much valuable help from Ron Abileah, Jeff Voss and Boris Venet at SRI International. Fred Levian 7