Anda di halaman 1dari 58

ECE 6272

Fundamentals of Radar Signal Processing

Lecture #22:
Synthetic Aperture Radar
School of Electrical and Computer Engineering
Georgia Institute of Technology
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture# 22, Slide 1

Distance Learning Final Exam


Due Date
Due to GT holiday schedule, due
date for DL final exams will be
Wednesday, Dec. 22, at noon EST
This is less than the normal oneweek delay, which would have it
due on Dec. 23.
You will hear more reminders from
GT DL.
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 2

ECE6272

Fundamentals of
Radar Signal Processing

Module #67
Introduction to
Synthetic Aperture Radar Imaging,
Continued
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 3

SAR is About Cross-Range Resolution


A good 2D image requires good resolution in both dimensions
Range (a.k.a cross-track) resolution is obtained with appropriate
high-bandwidth waveforms and matched filtering
SAR is a technique for obtaining good resolution in the other
dimension, namely cross-range (a.k.a. along-track)

range (y)

cross-range (x)

Image courtesy of Sandia National Laboratories


Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 4

SAR is About Imaging Stationary Clutter


Basic SAR is about imaging terrain
often considered the clutter in many other radar modes

The target is therefore stationary


moving objects in scene (vehicles) require special treatment
space-based SAR must deal with earth rotation

Image courtesy of Sandia National Laboratories


Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 5

The Synthetic Aperture Concept:


Synthesizing a Virtual Large Array
future transmit positions

The physical antenna is one element of the synthetic


array
Data is collected at each position sequentially, and then
processed together
Effective aperture size is determined by the distance
traveled while collecting a data set

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 6

Sidelooking Stripmap SAR


Phased array aperture is formed
parallel to the velocity vector

Effective
aperture size
DSAR or aperture
time Ta
determined by
how many
pulses
contribute to
each pixel

Therefore array factor boresight


is perpendicular to the velocity
vector
phase steering can change this
steering of the physical antenna
squints the element pattern

Physical antenna does not


scan
physical beam dragged
along with platform motion
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

resolution range bins

typical sinc-like antenna


pattern

DSAR
= avTa
DSAR
= vT

high cross-range
resolution
range bins
high cross-range

Element pattern not broad

high cross-range
resolution range bins

DSAR = vTa

This basic
SAR
operational
mode is
called
stripmap SAR
Lecture #22, Slide 7

Aperture Time, Size, and


Resolution
Suppose we integrate data over an aperture
time Ta; what resolution is obtained?
SAR =

well explain this in


the next module

DSAR= vTa

SAR=

2vTa

2 DSAR

CR= R SAR

CR=

R
2vTa

To get constant resolution, aperture time


must be made proportional to range
Ta =
Fall 2010

R
2vCR

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 8

Example Airborne Aperture


Time vs. Frequency
aperture time (sec)

10

v = 150 m/s,
R = 10 km
Note aperture
time ranges
from fractions of
a second, to
tens of seconds
Mainstream
range is
usually on the
order of 0.1 to
1 seconds

1 m = cross-range resolution

3m
1

0.1

10 m

10

100

frequency (GHz)
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 9

Space-Based Radar Aperture


Times
2
1

Low earth orbit:


v = 7.6 km/s
770 km
altitude

aperture time (sec)

5 m = cross-range resolution

30 m

0.1

100 m

0.2
1

10

100

frequency (GHz)
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 10

Maximum Synthetic Aperture Size


target

Maximum
synthetic
aperture size is
the maximum
distance traveled
while target is
illuminated

az
R

direction
of flight

DSARmax = R az
Fall 2010

Raz

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 11

Best Case SAR Strip Map


Resolution
Cross-range resolution corresponding
to maximum synthetic aperture size:

Daz

=
= =
SAR
2 DSAR 2 R az 2 R

CRmin = R SAR
Fall 2010

Daz
=
2

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 12

Where Did the R Dependence Go?


CRmin

target

Daz
=
2

w
R

Targets at longer ranges


( ) are illuminated longer
than shorter-range targets ( )
larger potential aperture finer
resolution
direction
of flight

Rw

Increase in effective aperture size exactly


cancels increase in beamwidth with range
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 13

What Have We Gained?


Cross-range resolution is now independent of
range
far superior for mapping purposes

Cross-range resolution is now much smaller,


and comparable to range resolution
0.5 m lower bound vs. 300 m @ Daz = 1 m, R = 10
km, Fc = 10 GHz
5 m lower bound vs. 4.6 km @ Daz = 10 m, R = 770
km, Fc = 5 GHz
far superior for mapping purposes

Huge integration gains


combine hundreds, even thousands of pulses

The cost is much greater signal processing


complexity
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 14

San Francisco, California

SIR-C X-band SAR image


http://photojournal.jpl.nasa.gov/catalog/PIA01791

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 15

RADARSAT Image of Hartsfield


Airport, Atlanta

Produced with commercial


EarthView software
(from Atlantis Scientific)

Fall 2010

Produced with GTRIdeveloped software

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 16

Recent TerraSAR-X Images


Tokyo River
Island, 3/7/2008
with Google
earth photo
inset

Both images from R. Werninghaus and S. Buckreuss,


The TerraSAR-X Mission and System Design, IEEE
Trans. Geoscience & Remote Sensing, vol. 48, no. 2,
Feb. 2010

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 17

Spotlight SAR
What if Daz/2 stripmap resolution isnt good enough?
we need a bigger synthetic aperture!

Real beam is actively scanned to


lengthen illumination time
Resolution is expressed
in terms of angle of
rotation:
R

CR
=
=
2 DSAR
2

DSAR

Continuous mapping sacrificed for resolution


Inverse SAR (ISAR) is similar with roles reversed: radar is
fixed, target moves (e.g., target on turntable, ship on
waves)
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 18

Spotlight SAR Image of the Pentagon


Sandia Twin Otter system
Ku band
1 m resolution

Also evidently used in DoE


Airborne Multisensor Pod
System (AMPS) for
nonproliferation R&D

http://www.sandia.gov/radar/images
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 19

TESAR Optical vs. SAR

http://www.es.northropgrumman.com/products/ISR_overview.htm

Notice the layover and shadow effects in the SAR image!


Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 20

TerraSAR-X 1.1 m Spotlight


Image of Rio de Janeiro

http://www.infoterra.de/image-gallery/images.html
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 21

ECE6272

Fundamentals of
Radar Signal Processing

Module #68
SAR PRF & Area Coverage

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 22

Stripmap Image Depth


Swath length (or depth) Ls extent
is upper bounded by elevation
pattern footprint on ground
Can be less due to
sampling a shorter
range window
unambiguous range
constraints

el

Cross-range extent
is unlimited
so long as data can
be processed or
downloaded in real time
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Rel
sin

Lecture #22, Slide 23

Swath Width Constraint on PRF


SAR is operated with no range
ambiguities
across the swath length
does not need to be unambiguous out
to the absolute range

2
( Ls cos ) +
c
2
( Ls cos )
c
c
PRF
2 Ls cos
PRI

R
Ls cos

Ls

Rel
R
=
sin Del sin
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 24

Yet Again: Doppler Bandwidth


The radial velocity relative to stationary scatterers
differs across the physical beamwidth
Therefore the Doppler shift differs
PHYSICAL
az BEAMWIDTH
The spread in Doppler
across the 3-dB beam
edges is
2v az
BDmax
sin Hz

2v
=
sin
Daz

Hz

DIRECTION OF FLIGHT

This is the bandwidth of the slow time data sequence


obtained from a single range bin over multiple pulses
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 25

SAR Is a Low-PRF Mode


2v
c
PRF
Daz
2 Ls cos
(sidelooking = 0)

PRF must exceed Doppler bandwidth to satisfy the Nyquist


sampling theorem
PRF must provide acceptable unambiguous range:
note: Ls is in slant range, not ground range

Typical bandwidth is 100s to a few 1000s of Hz, so PRF


must be at least 100s to 1000s of Hz
Typical unambiguous range is 10 to 50 km, so PRF must be
less than, e.g., 15 kHz at 10 km
Therefore, SAR is a low PRF mode
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 26

Mapping Rate for Stripmap SAR


We map a strip up to Ls meters deep while
moving forward at v m/s
Area coverage rate is therefore A = vLs m2/s
If Ls is large enough to push PRF down to the
Doppler bandwidth limit, Ls cDaz 4v cos

cDaz
A
m 2 /s
4cos
cDaz
2
=
km
/s
6
4 x10
= 0.022cDaz km 2 /day
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 27

Cross-Range Sampling Interval


Recall the Doppler bandwidth

BDmax

2v

Daz

Hz

this must also be the Nyquist sampling rate

Nyquist sampling interval is therefore


Daz
Ts =
2v

sec

At a velocity v, the distance traveled between


cross-range samples is
Daz
=
d s vT
=
meters
s
2
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 28

Maximum Swath Width


Constraint on PRF
If swath length takes on maximum
value:
Del c tan
c
PRF
=
2 Ls cos
2 R
R

Ls
=

Fall 2010

R el
R
=
sin Del sin

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 29

Antenna Area Constraint


cDel tan
2v
PRF
Daz
2 R
If the lower bound equals the upper bound, we get
the antenna area constraint:

4vR
Daz Del >
c tan
need bigger antennas for space-based than airborne
large area antennas are needed for simultaneous maximum
swath length and minimum resolution

Several implicit assumptions to get this result; see


A. Freeman et al, The Myth of the Minimum SAR Antenna Area
Constraint, IEEE Trans. GRS, 38(1), Jan. 2000.
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 30

Image Rate for Spotlight


Suppose we want to image a patch of resolution CR at a
range of R meters
Platform LOS must rotate through /2CR radians
So platform must travel R/2CR meters
This requires R/2vCR seconds (aperture time Ta)

Spotlight image rate is therefore


2vCR
Number of images
images s
R
7200vCR
images hr
=
R
this assumes all images have same resolution and standoff
range

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 31

Resolution vs. Swath Depth and


Area Coverage Rate
Fine resolution requires small
CRmin =
Daz 2
Daz to allow large synthetic
aperture in stripmap:
But small Daz gives large
cDaz
Doppler bandwidth
Ls
meters
4v cos
requires high PRF
limits swath length:
cDaz
A
m2 s
Same is true of area
4cos
mapping rate:
Thus, fine resolution conflicts
with deep swath and/or high
mapping rate!
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 32

RADARSAT-2 Stripmap Modes

http://www.radarsat2.info/

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 33

ECE6272

Fundamentals of
Radar Signal Processing

Module #69
Stripmap Geometry & Range
Migration

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 34

Stripmap SAR Data Collection


Assume pulse compression done on each pulse as it
is received .

Animation courtesy Seung-mok Oh, Georgia Institute of Technology

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 35

Range to a Point Scatterer - 1


z

Platform coordinates:
px = (u=vt,0,h)
Target coordinates:
pt = (xP,ygP,0)
h

RP
x = u = vt

yg

P = (xP.ygP,0)

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 36

Range to a Point Scatterer - 2


Range is the Euclidean
distance:
R = R (u ) =

2
+ h2
( u xP ) + y gP
2

Define RP as range at point of


closest approach (u=xP); then
R ( u ) = ( u xP ) + RP2
2

= RP

Fall 2010

2
u xP )
(
1+

RP2

(a hyperbola)

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 37

Interlude:
Hyperbolas & Parabolas
R=
( u ) RP 1 +

( u xP )

RP2

More advanced SAR algorithms use this


hyperbolic expression for range migration
needed for higher resolutions, deeper swaths,
lower RF frequencies

Many moderate resolution SARs further


simplify this expression to get rid of the
square root
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 38

Range to a Point Scatterer -3


Apply a series expansion of the square
root:
1
1 2
1 + x =1 + x + x +
2
8
Keep only quadratic or lower terms in
(u/RP):
( u x )2
P
RP
R ( u ) 1 +
2

2
R
P

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

(a parabola)

Lecture #22, Slide 39

The Quadratic Variation of


Range (and Phase)
Remember phase of echo is shifted by (4/)R
radians
The quadratic (i.e., chirp) phase spreads the
bandwidth of the slow time data
motion-induced slow time chirp
( u x )2
P
RP
R ( u ) 1 +
2

2
R
P

4
(u ) =

Fall 2010

R (u )

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 40

Pulse Compression in Slow Time


We have just argued that the slow-time data
for a given target is approximately a
geometry-induced LFM
Even without the square root approximation,
we have a hyperbolic phase model fo the
slow-time data
So we can get good resolution by pulse
compressing in slow time with the
appropriate matched filter
So whats the big deal with SAR processing?

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 41

Range Migration:
Range Walk and Curvature
Range migration
describes the down-range
curvature of the data as
cross-range position varies
Decompose into
range walk and
range curvature

sidelooking,
far range

Rc

squinted

Fall 2010

Rw
range

Returns from a given


scatterer may be spread
over multiple range bins
This greatly complicates
cross-range processing

sidelooking, near range

cross-range

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 42

Range Walk
Difference in range to scatterer at beginning
(u = xPDSAR/2) and end (u = xP +DSAR/2) of
synthetic aperture
Always zero for side-looking systems:

Rw= R ( xP + DSAR 2 ) R ( xP DSAR 2 )


2
2

DSAR 2 )
DSAR 2 )
(
(
= RP +
RP +

2 RP
2 RP

=0
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 43

Range Curvature
Difference in range to scatterer at beginning or
end (u = xP DSAR/2) and middle (u = xP) of
synthetic aperture

R ( xP + DSAR 2 ) R ( xP )
R=
c
2

DSAR 2 )
(
=
RP +
{ RP }
2 RP
For constant aperture

time, decreases with

range
=
For constant
resolution (Ta R),
increases with range
Fall 2010

2
DSAR
v 2Ta2
=
8 RP
8 RP

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 44

Range Migration in Range Bins


SARs usually operate with square pixels:
R = CR =
Using Ta = R/2v and normalizing to ,
range curvature becomes

Rc R
=

323
2

So improving resolution by a factor of 2x


increases range curvature in terms of range
bins by 8x!
Can be less than one range bin, or can be
hundreds of range bins
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 45

Other Big Deals in SAR Processing


sidelooking,
far range

Shift-varying matched filter

Rc

well see this in next module


range

Motion compensation

squinted

Rw

sidelooking, near range

We assumed straight & level,


constant velocity flight

cross-range

No turbulence, no cross winds

Antenna pattern compensation


Computational load
Moving targets in scene
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 46

ECE6272

Fundamentals of
Radar Signal Processing

Module #70
Stripmap SAR Dataset

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 47

SAR Waveforms
SAR processing achieves high resolution
in the cross-range dimension
Range resolution is achieved by using
wideband waveforms and pulse
compression
Recall range resolution is R = c/2
Examples: 10 m 15 MHz, 1 m 150 MHz

LFM waveforms are very common in SAR


typically 100 or greater
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 48

Time Signature of a Single Point Target


reflectivity
d[m,n]
Reflectivity

point
spread
s[m,n]
Real Part
of theresponse
Signal Returns

-1500

-1500

-1000

-500

Azimuth [m]

Azimuth [m]

-1000

0
500

-500

500

1000

Before pulse compression!

1000

1500
10

10.2

10.4

10.6

10.8

11

1500

Ground Range [km]

10

10.2

10.4

10.6

10.8

11

Ground Range [km]

Real Part of the Signal for Azimuth = 499.75[m


0.8
0.6

Amplitude

0.4
0.2
0
-0.2
-0.4

s [ m, n0 ]

-0.6
-0.8
10

10.2

10.4

10.6

10.8

11

11.2

Ground Range [km]


Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 49

Range Migration Creates Cross-Range Chirp


( u x )2
P
R ( u ) 1 +
2

2 RP

Real Part of the Signal for Ground-Range = 10401.6532[m]


1
0.8

RP

pointPart
spread
response
s[m,n]
Real
of the
Signal Returns

0.4
0.2

-1500

0
-0.2

-1000

-0.4

s [ m0 , n ]

-0.6
-0.8
-1

-2500-2000-1500-1000 -500

500 1000 1500 2000 2500

Cross-Range [m]
Real Part of the Signal for Ground-Range = 10401.6532[m]

Azimuth [m]

Amplitude

0.6

-500
0
500

0.8

1000

Amplitude

0.6
0.4

1500

0.2

10

10.4

10.6

10.8

11

Ground Range [km]

-0.2
-0.4
-0.6

s [ m0 , n ]

-0.8
-1

-100 -80 -60 -40 -20

20

40

60

Cross-Range [m]
Fall 2010

10.2

80 100

So point target response is a 2D chirp,


curved in range
Curvature and chirp rates predictable functions
of radar and geometry parameters

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 50

Signature of Three Point Targets


Region of Support

Real Part of the Signal for Azimuth = -0.25[m]

-1500

2.5

-1000

1.5
1

Amplitude

Azimuth [m]

-500

500

0.5
0
-0.5
-1
-1.5

1000

-2
-2.5

1500
10

10

10.2

10.4

10.8

10.6

10.2

11

10.4

10.6

10.8

11

11.2

Ground Range [km]

Ground Range [km]

Real Part of the Signal Returns


-1500

Azimuth [m]

-1000

-500

500

1000

1500
10

10.2

10.4

10.6

10.8

11

Ground Range [km]


Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 51

Data Set Size: Number of Range


Samples per Pulse
# of range samples to be processed per pulse =
(swath length)/(range resolution) + (waveform BT
product):
Ls
L=
+ ( 1)
R
not all contribute to a single point target,
but all have to be processed

L-1

Example (SIR-C L band):


Ls = 15 km, R = 15 m,
0
M-1
0
= 10 MHz, = 17 s L = 1170 pulse
# = slow time
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 52

Data Set Size: Number of Pulses


Number of pulses to contributing to the point target
response = (aperture time)(PRF):
M = Ta PRF
Thus

L-1

R 2v az
=

v
CR

R az
R
M
=
CR CR Daz

0
0

M-1

pulse # = slow time

Example (SIR-C L band): R = 250 km, Daz = 10 m,


Fc = 1 GHz, CR = 15 m M = 500
Generally: aperture time on order of 1 second, PRF 100s
to 1000s of Hz 100s to 1000s of pulses to integrate!
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 53

The Goal of SAR Processing

Echo energy from a single point target is distributed in both range and cross-range
(range bins and pulse number, or fast and slow time)
This function is called the point spread response (PSR) or point target response
(PTR) of the SAR sensor
Goal of SAR processing is to compress the PSR in both range and cross-range to a
single (correctly located) point
2D equivalent of pulse compression
Approach is essentially a matched filter
but many variations in implementation and accuracy
Shape of point target response depends on range of scatterer
therefore processing is not shift-invariant
point spread response s[m,n]

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

image d [ m, n ]

Lecture #22, Slide 54

Pulse Compression First


Many SAR algorithms do pulse compression first:

Range (Pulse)
Compression

After pulse compression, SAR point spread response


(PSR) for a target at (xP,R), in terms of cross-range and
down-range spatial coordinates (u,R) is
2
h ( u , t ; R ) = t
c
Fall 2010

( u xP )2 + R 2

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 55

SAR is Not Shift-Invariant


After pulse compression, SAR point spread response
(PSR) for a target at (x,R) is
2
h ( u , t ; R ) = t
c

( u xP )2 + R 2

This is a 2D function with range


migration
non-separable
invariant in cross-range (u)
direction
for a fixed aperture time

variant in down-range (R)


direction

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Illustration courtesy of
Dr. Greg A. Showman

Lecture #22, Slide 56

Assignments
Reading:
Chapter 8.1 8.3

Homework
HW #8
DL: due Thursday Nov. 18

HW #9
On campus: due Thursday, Nov. 30
DL: due Thursday, Dec. 9

Computer Project #4: Detection


On campus: due now
DL: Due Tuesday, Nov. 30
Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture #22, Slide 57

SIR-C L+C band image of Death Valley..


Red=LHH, G=LHV, B=CHV

End of Lecture 22

Next :
Stripmap SAR Image Formation

Fall 2010

Copyright Mark A. Richards . All Rights Reserved.

Lecture# 22, Slide 58

Anda mungkin juga menyukai