a fpga-based architecture for in-flight synthetic aperture radar (sar) motion compensation in...
TRANSCRIPT
A FPGA-Based Architecture for In-Flight Synthetic Aperture Radar (SAR) Motion
Compensation in Unmanned Aerial Vehicles
Fernando Ortiz
EM Photonics, Inc.Newark, DE
Ortiz 2 MAPLD 2005/1009
Outline
• Introduction & Motivation
• SAR Reconstruction Basics
• Motion Compensation
• The Hardware Platform
• Architecture for Real-time SAR Motion Compensation
• Conclusion and Future Work
Ortiz 3 MAPLD 2005/1009
SAR Concept
• Radar waves used to visualize objects because of their ability to penetrate a range of materials
Goal: gain the advantages of a large aperture radar by using a smaller, traveling aperture
• Resolution of image improves as aperture size increases
• Unfortunately, increasing aperture size (antenna length) may simply be impractical (antenna lengths in kilometers)
Ortiz 4 MAPLD 2005/1009
SAR Applications
Mining/Space Exploration
Ocean FloorTopography
Air TrafficControl
Target Detection
and Tracking
MedicalImaging
Buried ObjectDetection
Ortiz 5 MAPLD 2005/1009
Motion Compensation
• Problem: cannot guarantee perfect motion paths• Result: degraded images• Solution: motion compensation
• Options for aerial platforms:– Massive onboard computers– Slower processing (secs per frame)– Ground processing
Without Compensation
With Compensation
Complexity of motion compensation is limiting factor in deploying
SAR systems!
Ortiz 6 MAPLD 2005/1009
Motivation
Space-Based Airborne UAV• Simple motion
compensation
• Power/area available for calculations
• Disregard motion compensation (for stable orbits)
• Ground processing practical
How does this impact in-flight systems?
X
UAVs require fast, low area/power motion compensation solversSolution: reconfigurable platforms!
• Advanced motion compensation (erratic path, wind interaction)
• Minimal power/area for processing
Ortiz 7 MAPLD 2005/1009
SAR Geometry
x
y (cross-range domain)
z
xn
yn
ImagedRegion
Reflectivetargets
range
GoalDetermine x,y,for each target
How?•Range Imaging•Cross Range Imaging
Ortiz 8 MAPLD 2005/1009
SAR Basics: Range Imaging
Received signal n
nn c
xtpts )
2()(
Desired information n
nn xxxf )()(0
Combines Range and ReflectivityMatched Filter
x1 x2 x3 x4
1
2 3
4
1
p(t)
x1
2
x2
3
x3
4
x4
1p(t-2x1/c)Yc
2p(t-2x2/c) 3p(t-2x3/c) 4p(t-2x4/c)
Ortiz 9 MAPLD 2005/1009
SAR Basics: Cross-Range Imaging
• Use matched filtering (again) to determine cross-range information
• Put these two together and you have a 2D imaging system
Typical SAR problem
y3
y2
y1
Xc
Fourier Transform(t,u) (w,ku)
2D Matched Filter
Inverse Transform(kx,ky) (x,y)
FFTs are the bottleneck in traditional SAR
c
uYXtputs cc
22 )(2),(
ReceivedSignal
OutputImage
Ortiz 10 MAPLD 2005/1009
MC SAR Processing Flow
Motion Compensation
is the NEW Bottleneck
22
222222
22*
sinsin)(sin)(coscos)(
)()()()()(
)(exp),(
zxk
kyu
uzuyux
zuyxuzzuyuyuxxur
urkkjkkH
x
y
czececze
nnnenenenen
enyxyxexyz
Motion Compensation Filter
Reconfigurable platform permits massive parallelization and pipelining
FFT
ReceivedSignal
ReconstructedImage
IFFT
SARFilter
MCFilter
c
uYXtputs cc
22 )(2),(
SAR Filter
Ortiz 11 MAPLD 2005/1009
Hardware Platform
PLX 9656(External PCI Control)
16 GB DDR SDRAM
PCI 64/66 Interface
Xilinx Virtex-II8000 FPGA
Custom, FPGA-based PCI Card
36 MbDDR SRAM
Ortiz 12 MAPLD 2005/1009
Platform Success
Platform used to develop accelerated solvers for electromagnetic simulations.
Single PC
PC cluster, 30 nodes
EM Photonics Celerity Platform
0
5
10
15
20
25
30
35
0 50 100 150 200Nodes (Millions)
Performance vs. Problem Size
0
5
10
15
20
25
30
35
0 50 100 150 200Nodes (Millions)
Mil
lion
s of
nod
es/s
ec (
Mn
ps)
Key Statistics• 9.5 GB/s Main Memory Bandwidth• 150+ Floating-Point Units @ 133 MHz
Ortiz 13 MAPLD 2005/1009
SAR Motion Compensation Architecture
kx
ky
xy
Round
Cos c
Sin c
yn
xn
REG
U
BRAMxe LUTye LUT
Norm
Norm
Norm
Out
xk
kyu
uyuxuyxuyuyuxxur
urkkjkkH
x
y
cecennenenen
enyxyxexyz
sin)(cos)(-)()()()(
)(exp),(
2222
22*
Ortiz 14 MAPLD 2005/1009
Resource Utilization
26.3921.67% of XC2V8000
3820190
Total
8111288111281FPEXP
0171905733FPSQRT
057305731FPDIV
301890318910FPMUL
04880048810FPADD
MultsLUTs
Total
MultsLUTsQuantity
Three parallel SAR MCUs are feasible within a single chip
Ortiz 15 MAPLD 2005/1009
Conclusion and Future Work
• SAR Motion Compensation requires significant computing power
• Demonstrated FPGA platform capable of in-flight SAR MC
• RC platforms ideal fit for UAV applications – Comm. Bandwidth savings– Airborne processing enables
further applications (e.g. ATR)– Low weight/power– Hardware reusable for other tasks
For the future:
• This solves only one piece– FFTs – Interface
• Form factor has to be converted– Less memory– No PCI– Interface with the rest of the
system– Integrate cooling into the airframe