INS/GPS/SAR Integrated Navigation With Image

Geocoding


Martn Espaa,
National Commission of Space Activities (CONAE), Argentina
Department of Research and Doctorate, Faculty of Engineering, University of Buenos Aires, Argentina


BIOGRAPHY

Martn Espaa obtained his Ph.D. at the National
Polytechnic Institute of Grenoble, France, his M.Sc. at the
University of Chile and his B.E. degree in Electronics at
the University of La Plata, Argentina. He has been a full
professor at several academic institutions in Mexico and
Argentina, worked for NASA as an NRC Senior Research
Associate. He is currently with CONAE, the Argentinean
National Space Agency, and teaches at the doctorate
Department of the Faculty of Engineering of the
University of Buenos Aires, Argentina.

ABSTRACT

SAR systems inherently generate highly valuable
navigation data at very high rates. However, this
information is usually totally ignored by the platforms
ad-hoc navigation system required by SAR data
processing methods. On this behalf, a scheme that takes
advantage of SAR data richness in a post processed
INS/GPS/SAR integrated navigation method is here
proposed. The new method, not only may considerably
improve the navigation accuracy, but it also allows for
geo-coding of highlighted points of the scene being
imaged (geo-coding that could easily be propagated to the
rest of the SAR image) by using only flight data. Potential
applications are geodesy and geo-location, roads and civil
works, ducts and transmission lines supervision and
cadastre. For ease of reading, the article summarizes the
basis of SAR processing.

INTRODUCTION

The synthetic aperture radar (SAR) high resolution
imaging technique is now a well matured technology with
several decades of development and experimentation. The
most classical version of this technology, viz. strip
mapping referred to in this work, produces images of a
lateral strip parallel to the linear ground track of SARs
platform straight flying trajectory (aerial or satellite) (see
Fig. 1). The SARs antenna radiates short microwave
pulses at a uniform pulse repetition rate (PRF) throughout
its flight path while sweeping out a strip on the ground.
The pulses echoes, coherently processed, emulate a
virtual antenna array several times longer than the
physical antenna. The relatively unexpected result is a
high-resolution image of the scene illuminated by the
radiated pulses with a spatial resolution independent of
the range. However, lack of precise knowledge of the
platform motion (specially in airborne SAR) may strongly
affect the image quality by causing, among others, image
defocusing -a blurring effect particularly disrupting in
high resolution imaging that prevents discriminate
between neighboring areas of a scene. For this reason,
SAR data processing needs to be "helped" by motion
estimates, such as position and velocity, provided by an
ad-hoc navigation system [1]. SAR and navigation data
are post-processed together within a procedure called
image auto-focusing usually performed off line. Low
quality navigation estimates may imply bad quality
images, particularly when the airborne SAR platform
undergoes strong atmospheric disturbances. On the other
hand, SARs signal echoes clearly contain information
about platforms position and velocity (and even attitude)
relative to the scenes reflectors. However, hitherto, little
effort has been conducted to exploit SAR data to improve
the navigation quality. A new scheme is here proposed
whereby SAR, GPS and inertial data are fused with, as a
result, not only a potential improvement of the platforms
motion estimates but, at the same time, the geo-coding of
a set of specifically chosen cells of the SAR scene. The
approach shows the intimate relationship between SAR
imaging and SLAM (Simultaneous Localization and
Mapping) techniques [2], [3]. Indeed, the SAR antenna is
seen as a navigation sensor (whose data may be fused
with other sensors such as a GPS receiver and an IMU)
and the SAR image as nothing more than a two
dimensional map of the point reflectors sensed through
the echoes of the SAR antenna. Given the specificity of
the subject we judged convenient to introduce the basics
Ground
range
SAR
Flight path
range
nth. pulse
emission
strip
Footprint
Slant strip
Ground
track
azimuth
Nadir
Figure 1: SAR geometry.
of the SAR processing technique (see, e. g.: [4] and [5]) in
the next paragraph.

SAR DESCRIPTION

Like any radar, the natural coordinates of an
elementary scatterer in a SAR scene are: the rank r and
the azimuth x. In strip mapping SAR, the first one refers
to the distance from the scatterer to the nominal linear
trajectory of the antenna, while the 2nd is a coordinate
running along the nominal path itself also called azimuth
axis (Fig. 2). The plane (r, x) is called the slant plane. A
SAR image is a sufficiently precise correspondence
between the local reflectivity of the scene, illuminated by
the antennas sequence of pulses, and the coordinates on
the slant plane. The power of each microwave pulse is
spread over a frequency band centered on a carrier
frequency. Typical processing of the pulses echoes
combines two consecutive procedures. The first one,
called range compression, is common to any radar and
resolves the scene in the range dimension with accuracy
proportional to the signal bandwidth. Distinctive to SAR
technology is the procedure called azimuth compression
applied in second term to resolve the image in the azimuth
coordinate.

Models of the SAR signals
The complex representation of the signal transmitted
within the pulse interval T
p
centered at the emission time
nT is typically of the form:
0
( ) ( )
j t
s t e S t
e
= , with ( ) S t =
2
/ 2
( / )
j t
p
e t T
o
H the quadratic modulating phase (chirp) and
t measured from nT (Fig. 3). ( / )
p
t T H denotes the unitary
pulse within / 2
p
t T s . Thus, the instantaneous phase and
frequency signals for / 2
p
t T s are, respectively:
2
0
( ) / 2 t t t = e + o and
0
( ) t t e = e + o . The bandwidth is
defined as: AB = / 2 [ ]
p
T Hz o t . If is the carrier
wavelength and L
a
the effective length of the antenna in
the azimuthal direction, it is widely accepted to consider as
the antenna beam width in the same direction its -3dB
boundary: u
a
=/L
a
[rad]. Therefore, the resulting length of
an illuminated segment parallel to the azimuth at the range
r turns out to be: L
r
=r/L
a
. Assuming a uniform platform
speed V, the distance traveled over the azimuth axis
between 2 consecutive pulses is A=VT , with T=1/PRF.

Let
2 2
( ) ( )
n n
r p r x x = + be the range between the
antenna nominal position at the time nT: p
n
=(0,x
n
=nA) and
a scatterer located at p=(r,x) in slant plane coordinates. Let
( , )
n
w r x represent the real two dimensional 2-ways
antenna gain at p. For moderate beam openings,
( , )
n
w r x may be represented as a separable function of
range and azimuth, such as: ( , )
n
w r x ~
( ( )) ( )
r n a n
w r p w x x with ( ) ( / )
a r
w x x L ~ H ; for typical
values:
3
/ / 2 10 [ ]
n a
x x r rad

s u ~ , therefore:
2
( ) ( ) / 2
n n
r p r x x r ~ + . Moreover, it turns out that
( ( , )) ( )
r n r
w r r x w r ~ , i.e: is practically independent of the
azimuth. Let
( , )
( , ) ( , )
j r x
r x r x e

= be the complex
local bi-dimensional reflectivity density at the point p of
the scene and ( ) 2 ( ) /
n n
t p r p c = the pulses round trip
time delay to p. The differential echo over the antenna
coming from p after the n
th
pulse is:
0
0
( ( ))
( ; , )
( ) ( ) ( ( )) ( , )
( ; , )
n
n
j t t p
r a n n
j t
n
dg t r x
w r w x x e S t t p r x drdx
e dG t r x
e
e
=
=
=
(1)

Moderate antenna apertures, also allows us to
consider as independent of n within the time interval in
which the scatterer at p stays within the antenna beam. As
such, the echo produced on the antenna by the set of
scatterers illuminated by the n
th
pulse is:
0
min max
( ) ( ); [ , ]
j t
n n sc
g t e G t t T t t
e
= e (Fig. 3), where, the
complex base band phasor (acquired by quadrature
demodulation, i.e.: signals real and imaginary components
in complex base band) of the returned signal is:

0
min
/2
( )
/2
( )
( ) ( ) ( , ) ( ( ))
max n r
n
n r
n
r x L
j t p
r a n n
r x L
G t
w r w x x r x e S t t p dxdr
+
e

=

} }
(2)

(0, )
n
x
: x azimuth
a
u
0
r
max
r
min
r
( , ) r x -
r
L
Slant strip
Figure 2: Antenna beam in slant plane.
r
( )
n
r p
T=1/PRF
Rx.
Pulse Echo
Time
T
p
min
t
Figure 3 Pulsed SAR signals.
Echo Pulse Pulse
max
t
nT
( )
n
g t
( ) s t
( 1) n T +
sc
T
Impulse response of the SAR processing
The reflectivity information of each of the scatterers
illuminated by the sequence of pulses is distributed among
the corresponding sequence of echoes. SAR processing is
designed to extract from the records of the digitized echoes
cells with uniform reflectivity with diameter given by the
actual SAR image resolution. We chose to illustrate the
principle of SAR processing by studying its response:
( , )
c c
h r r x x to an isolated point target located at
c
( )
c c
p r , x = with complex reflectivity ( )
c
q p eC and 2-
dimensional impulsive density ( ) r, x =
( ) ( )
c c c
q p r r , x x o . Processing of a generic scene (non
impulsive) will thus be equivalent to the 2-dimentional
convolution of the impulsive response ( , ) h with the non-
impulsive generic bi-dimensional density function ( ) r, x
(See Fig. 4). The result is an image with resolution
determined by the range and azimuth impulse response
selectivities (Fig. 5) characterizing the SARs
performance.

By substituting in (2), the echo to the n
th
pulse is:

0
( )
( ; )
( ) ( ) ( ) ( ( ))
n c
n c
j t p
r c a c n c n c
G t p
w r w x x q p e S t t p
e
=

(3)

The next paragraphs describe the SAR processing
of the echoes produced by an impulsive point target with
the assumption that the antenna travels over the azimuth
axis with fixed known velocity V.

1) Range compression
The range compression process consists of the
numerical evaluation of the normalized correlation of each
echo with a replica of the complex pulse which, for the
impulsive point target gives for
sc
t T e (Fig. 3):
( )
0
*
( )
*
1
( ; ) ( ; ) ( )
( ) ( ) ( ) /
1
( ( )) ( )
n c
n c n c
p
j t p
c r c n c c
n c
p
t p G p S t d
T
q p e w r x x L
S t t p S t d
T

t t t =
= H
t t
}
}
=
(4)
By using classical properties of the Fourier transform
and denoting: ( ) { ( )} S j S t e = , the convolution in (4)
may be rewritten as:

* 1 *
( ) 1 *
2
( ) 1
( ( )) ( ) { { ( ( ) ( )}}
{ ( ) ( )}
{ ( ) }
n c
n c
n c n c
j t p
j t p
S t t p S t S t t p S t
e S j S j
e S j

e
e
- = -
= e e
= e

(5)

A property of interest of the complex modulating
signal
2
/ 2
( ) ( / )
j t
p
S t e t T
o
= H is that, for a bandwidth AB
and pulse width T
p
such that . 20
p
AB T > (verified in this
context) its power spectral density may be approximated
as
2
( ) ( / 2 ) S j M AB e ~ H e t . Where, the relationship
between M, T
p
and AB results from Parseval's identity:
/ 2
2 / 2
2 2
/ 2 2 / 2
1
( ) ( )
2
p
p
T
AB
p
T AB
T S t dt S j d M AB
t
t
= = e e=
t
} }
(6)
Form (5) and (6), using:
( ) 1
{ ( / 2 )}
n c
j t p
e AB
e
H e t = sinc( ( ( )))
n c
AB AB t t p t
and the definition of the reflectivity of p
c
back in the
antenna: ( ) ( ) ( )
a c r c c
q p w r q p = , correlation (4) becomes:

0
( )
( ; )
( ) sinc( ( ( ))) (( ) / )
n c
n c
j t p
a c n c n c c
t p
q p e AB t t p x x L
e
=
= t H
(7)

Thus, the peak of the correlation ( ; )
n c
t p
corresponds to a measurement of the complex reflectivity
(affected by the phase
0 n
t e ). On the other hand, the
occurrence of the peak time: 2 ( ) /
n n c
t r p c =

, measures the
range with a precision given by the selectivity of the
function sinc( ( ( )))
n c
AB t t p t which, to a -3dB level, is:
ot=1/AB. This corresponds to a geometric resolution:
/ r c AB o = ; i.e.: ( ) ( ) / 2
n c n c
r p r p r = o . As an example,
for AB = 100 MHz, T
p
=10 sec, AB.T
p
20 and the
theoretical range resolution is: 3 . r mts o ~ Therefore, after
introducing:
2
0 0
( ) ( ) / 2
n c c c n c
t p t x x = e e + o , together
with: 2 /
c c
t r c = ,
0
2 /
c c
cr o e = , we define the
measurement associated with the n
th
pulse:
2
0
( ) / 2)
q ( ) ( ) (( ) / )
c c n c
j t j x x
n c a c n c c
p q p e e x x L
e o
= H (8)
SAR
Scene
( ) r, x
Image
( ) r, x
( , ) ( , ) ( , )
scene
r x h r x d d = t t t
}}
Figure 4: SAR processing seen as a convolution.
azimuth
range
( , )
c c
h r r x x
c
x
c
r
Figure 5: SAR impulse response from a point scatterer.

2) Azimuth compression
When the range compression process is applied to
each of the echoes coming from p
c
, the result is a sequence
of complex measures (8) of the reflectivity ( )
n c
p q for,
n=1,2,N, with N the total number of pulses transmitted
from the azimuths x
n
, (n=1,2,N) that illuminated the
scatterer in p
c
. The complex sequence { } ( )
n c
p q should be
seen as a periodic sampling of the complex continuous
function of the azimuth:

0
( ; ) ( ) ( )
c
j t
c a c c c
x p q p e u x x
e
= q , (9)
where,
2
/ 2
( ) ( / )
c
j x
c c
u x e x L
o
= H is a modulating pulse
signal with quadratic (spatial) phase ; / 2
c c
x x x L s :
2
0
( ; ) ( ) / 2
c c c c
x p t x x = e + o and (spatial) bandwidth
2 /
c c c a
AB L L = o = . As in range compression, azimuth
compression starts from the normalized (spatial)
correlation:
0
1
1
( ) ( ; ) ( )
( ) ( ) ( )
c
c c c c
j t
c a c c c c
x L q x p u x
L q p e u x x u x

e
= -
= -

(10)
Also, in this case, the condition 20
c c
AB L >
(typically 300
c c
AB L > ), allows for the approximation:
2 2
( ) { ( )} (( ) / 2 )
c c c c
U j u x M AB v = ~ H v t , with M
c

defined by
c c c
L AB M = . Hence, according to (5) through
(7), the correlation (10) is:

0
( ) ( ) sinc( ( ))
c
j t
c a c c c
x q p e AB x x
e
= t (11)

Thus, the peak of correlation (11) is a measurement of the
complex reflectivity and, is located at the point scatters
azimuth with a resolution given by the selectivity of
function sinc( ( ))
c c
AB x x t , i.e.: 1/ / 2
c a
x AB L o = = ,
determined, only by the effective antenna length,
independently of the range!

INS/GPS/SAR INTEGRATED NAVIGATION

An inertial navigator, based on an inertial
measurement unit (IMU) and a inertial navigation
algorithm (INS) of the kinematic equations, provides
estimates of position, velocity and attitude at rates only
limited by the sampling frequency of the IMUs analog
signals. However, measurement errors, uncertain initial
conditions and numerical approximations induce errors
increasing with time. To limit errors and improve
accuracy, integrated navigation systems merge inertial
measurements with exoceptive ones provided at arbitrary
sampling times. In this way it is possible to combine high
information rates with good precision.
Given that the Extended Kalman Filter (EKF) ([6],
[7]) remains the most popular causal fusion algorithm used
in navigation, we shall use it as a reference in this paper.
However, the proposed concept applies equally well to
other recently developed methods ([8], [9] and [10]) which
may compete advantageously with the EKF in certain
applications. Besides, since in practice the SAR processing
is performed ex-post, more precise non-causal estimators
(smoothers) should be considered.
It is assumed that the IMU and the GPS antenna are
collocated at the phase center of the SAR antenna. This
simplifying assumption is adopted for the clearness of the
presentation, however the methodology may easily be
extended to nonzero lever arms, which, as pointed out in
[13], besides being more realistic it may considerably
contribute to improve the qualitative and quantitative
gyroscopic information.
Figure 6 shows a general scheme of an integrated
navigation system that fuses the measurements
b
m
(acquired in body frame b) of the inertial magnitudes m
with external (unspecified) measurements
k
y (altimeter,
GPS, SAR, compass, etc.) acquired at arbitrary sampling
times. The system estimates the kinematic state vector
(navigation parameters) x, together with the calibration
parameters p
i
and p
e
, respectively, of the inertial and
exoceptive sensors all of them aggregated in the
augmented state [ , , ]
i e
' ' ' ' _= x p p . With the estimates
k
x ,
,

e k
p and
,

i k
p as initial conditions, the INS integrates the
kinematic equations (navigation equations, see below)
between 2 consecutive exoceptive measurements
acquisition times (k, k+1). Once the new external
measurement is available the algorithm calculates the
innovation by subtracting from the external
measurement its a priori estimate calculated with the
sensor model:
1 1 1 , ,
( , , )
k k k e k i k
h

+ + +
= y x p p .
1

+
x being the a
priori estimate of
1 k+
x .

1 ,
1 1 1 1 1 1
1
1
,
, ,
;
i e
i e
k e k
k k k k k k
k
k x p p
v
h
h h h

+
+ + + + + +
+
+
o o_ +
c
( =

c
x p
x p p
y y y H
H
= =
=
(12)
In (12), [ , , ]
i e
' ' ' ' o_= o o o x p p ; (0, )
k k
v N R ~ is the
measurement noise and H
k+1
is the Jacobian of the model
IMU p
i
Model of
exoceptive sensors
( )
k
h x, p

Exoceptive
sensors
h
k
(x,p)
EKF
b
m
m
x
x
HW SW

k
y
k
y
k
y o
( ), ( )
k i k
t t o o x p
( )
e k
t op
Vehicle
Figure 6: Navigation system integrating inertial and exoceptive sensors data.
INS
( ) f = x x;p, m `
h
k+1
. The EKF weights the innovation oy
k+1
with the
Kalman gain
1 k
K
+
and updates the a priori estimate
1

+
_ into the a posteriori estimate:
1 1 1 1

k k k k
K

+ + + +
_ =_ + oy .

Kinematic equations and the IMU model
The kinematic (or navigation) equations used by the
EKF are formulated in Earth Fixed Earth Centered
(ECEF) coordinates, denoted e. The benefits of this
formulation are discussed in [11] and [12], Chap. 5. Let P
be the IMU position; V=dP/dt its ground speed; O: the
Earth angular rate; g(P): the gravitational acceleration (all
in frame e);
e
b
C the rotation matrix from b to e
coordinates:
be
the rotational vector angle from frame
e to frame b; x =[ ' P , ' V ,
be
' ] (Fig. 6) . The
kinematic equations are ([11], [12], [13], [14]):

( )
( ) 2
( ) ( )
b
S
= +
=
=
e b
b
e e e
b b b
P
V C f g P P V
P V
C C S C

`
`
`
(13)

Where: exp ( )
b
=
e
b e
C S , S(.) is the matrix operator
of the vector product,
b
is the inertial angular rate and f
b

the specific impulse. References [11] and [153] show an
efficient way to calculate the normal gravity (P) as a
function of P in frame e. Given the IMU
measurements ,
b b
f

, the actual inertial magnitudes are

modeled as:


3
; (0, ); ,
; ; (0, );
b b
m
m
m m N Q m f
N N
= + + = e
= q e q
b
b b
~
`
R ~
(14)

Biases in (14) are grouped in the vector: [ , ]
i f e
' ' ' = p b b
e
6
, modeled as a Brownian processes with , q
gaussian centered independent processes (this model is
enough to illustrate the concept, for more detailed models
see e.g.: [16]).

Error equations and the innovation models
The EKF uses model linearization around the a priori
available estimates:

P ,

V ,

e
b
C ,
i
p ,
e
p and estimates the
deviations:

i i i
o = p p p ;
e e e
o = p p p ;

o = P P P ;

o = V V V ;

e e e
b b b
o = C C C

( )
e
b
S | C =
The latter defines the attitude error | for small enough
values. Denoting [ , , ] ' ' ' ' o o o | x P V = and
P
the Jacobian
of the normal gravity with respect to P, it is possible to
show by using (13) [11], that the deviations may be
approximated by the solutions of the time variant linear
system (15), with forcing functions [ , ]
b b T
o o f coming
out from (14).

( )
3
[ , ]
0 0 0 0

2 ( ) ( ) 0
0 0 ( )
0 b b T
P i
i
e
I
S S
S
o o
( (
( ( o = o + o +
( (

o =
o =
e b e
b b
e
b f
x C f x C p

C
p
p
`
_
`
`
(15)

As known, the EKF has a mathematical core
(covariance update) which is independent of the external
sensors configuration. We thus concentrate on the
specifics introduced by each external sensor, which,
besides its model, is the Jacobian H
k
. intervening in the
Kalman gain update ([4], [14]).

1) The GPS observables and their innovations
Under the, so called, loosely coupled scheme, the
GPS receiver is assumed to provide measurements
k
P

,
k
V

of the actual position and velocity in e-frame perturbed by
and additive, Gaussian and independent noise. As such, the
innovations with respect to the corresponding a priori
estimates

P ,

V for | | 0 0 0
gps
k
I I = H become:

, , ,
, , ,

; (0, )

; (0, )
k k k P k P k P k
k k k V k V k V k
v v N
v v N

= o +
= o +
P P P R
V V V R

~


P gps gps gps k
k k k k
V k
v
v
v
( o (
o = + = o_ +
( (
o

P
y H
V
(16)

2) The SAR observables and their innovations
The SAR observables proposed here are the ranges to
a set of distinguishable fixed points within the scene. The
range measurements are performed during the first part of
the SAR processing by compressing the echoes reflected
by those points. Formally, each traceable fixed point
(whose range is being measured) may be associated to an
independent exoceptive sensor. The more they are, the
larger the amount of independent navigation information
available and the better the expected navigation accuracy.
If ( , )
c c
r x are the slant plane components of a fixed point
c
p , assuming a uniform V, its available echoes
correspond (approximately) to pulses transmitted during
the time interval ( / 2) /
c c a c
I x r V = u , in which
c
p
remained within the antenna beam. When
c c
x n = A , the
nominal emission azimuths nA of those pulses are such
that [ / 2 ] [ / 2 ]
c a c c a c
n r VT n n r VT u s s + u .
Let
c
P and
n
P be, respectively, the fixed point and
the antenna position at nT in frame e and denote:
n n c
D P P = , ( , )
n c n
d = P P D , then:
( )
1/2
( ) ( , )
T
n c n c n n n
r p d = = = P P D D D (17)
When unknown or uncertain, the
c
P components
become part of calibration parameters vector for its
associated external sensor, i.e. for this matter:
e c
= p P .
Let (
, ,

, cov( )
c n c n
P P ) be the available estimation at nT. The
initial estimate
,0

c
P of
c
P is obtained form the slant plane
coordinates ( , )
c c c
p r x = of
c
P determined from the SAR
processing and the antennas nominal trajectory.

The model of the SARs p
c
-observable at time nT
is: ( ) ( ) / 2 ( )
n c n c n c n
r p ct p r p = = + c

for (0, )
n r
N Q c ~
assumed independent (Bias or markovian parameters -e.g.:
reflecting atmospheric delay- may be taken into account
by just augmenting the dimension of
e
p ). Thus the
innovation becomes:


, 1
, 1

( ) ( , )

( , ) ( , )
sar
n n c n c n
n c n c n n
r p d
d d

o =
= + c
y P P
P P P P

, (18)

with (

, cov
n n

P P ) the a priori estimation of the antenna
(and IMU) position. By denoting
, 1 , 1

( ) /
n n c n n c n
~ D P P P P

the linearization (12) is

written as:

, 1 , 1

( , ) ( , )
( )
SAR sar
n n c n n n c n c n
n c
T
sar T sar n
n n n n c n
n
d d
v
v v


c c
o ~ o + o +
c c
~ o + = o o +
y P P P P P P
P P
D
D D R R
D


0 0
SAR T T sar SAR sar
n n n n n n n n
v v ( o = o + = o +

y D D H

(19)

INS/GPS/SAR Integration
The EKF updates the augmented state estimate _
with the GPS fix innovation and with the range innovation
at different rates. Since the a priori estimates are
calculated by the INS with a rate limited by the IMU
sampling rate, the latter rate also bounds the rate update of
the a posteriori estimates with the range innovations. As a
practical consequence, the range updates may need to be
downgraded from the original PRF according to the INS
calculation rate. When extended to a set of individualized
fixed point over the scene, the considerable amount of
information introduced by innovations (19) certainly
represents a strong contribution to the navigation accuracy.
Besides, those points end up being geo-referenced by the
method thus allowing the geo-coding to the whole image
by propagation. As a potential practical restrain, it must be
said that the points over the scene have to be chosen
brilliant and isolated enough so as to ensure that the range
measurement through echo compression is performed
without ambiguities. On the other hand, in certain
application, the INS/GPS/SAR integrated navigation
method could be used to geo-code deliberately marked
spots on the scene for instance by corner reflectors.

CONCLUSIONS

Air born SAR data processing requires as input high
quality navigation information of the hosting platform.
Traditionally this is done through an on-board ad-hoc
navigation system in post processed mode totally
independent of the SAR system. However, by its own
nature a SAR system generate valuable navigation data at
very high rates which is normally disregarded. A method
is here proposed to fuse SAR, GPS and inertial data
having a strong potential to improve the overall navigation
accuracy with ensuing benefits on SAR performance and
image quality. As a by product, a set of selected points of
the scene become geo-coded, allowing, by propagation, for
the geo-coding of the whole image. Potential applications
of this last feature include milestone geo-location, road
and civil works, ducts and transmission lines supervision,
cadastre and Geodesy. Corner reflector based SAR
calibration and performance evaluation procedures may
also benefit from the approach.
Finally, having shown the complementarities
between SAR and navigation, we hope that the present
work will encourage new research on integrated
approaches unifying SAR processing and navigation with
as a single objective to optimize the image quality.

ACKNOWLEDGMENTS

This work was supported by the National
Commission of Space Activities (CONAE) and the
Secretary of Research and Doctorate, Faculty of
Engineering, University of Buenos Aires, Argentina.

REFERENCES

[1] Lorga, J F; Van Rossum, M W; Van Halsema, E ;
Chu, Q P; Mulder, J A; The development of a SAR
dedicated navigation system: from scratch to the first
test fligh; Position Location and Navigation
Symposium, PLANS, 2004; pp. 249-258.
[2] Durrant-Whyte H. and T. Bailey. Simultaneous
Localization and Mapping: Part I. IEEE Robotics &
Automation Magazine, 13(12), pp. 99110, Jun.
2006.
[3] Bailey T. and H. Durrant-Whyte. Simultaneous
localization and mapping (SLAM): Part II. IEEE
Robotics & Automation Magazine, 13(3) pp. 108
117, Sep. 2006.
[4] Curlander J.C. & R.N McDonough; Synthetic
aperture radar, systems and signal processing; J.
Wiley , NY, 1991.
[5] Munson D.C. Jr., R. L. Visentin; A signal processing
view of strip-mapping synthetic aperture radar ;
IEEE Tr. On Acoustic, Speech & Signal Processing,
Vol. 37, N 12, Dec. 1989, pp 2131-2147.
[6] Grewal M. S., P.A. Andrews; Kalman filtering,
theory and practice using matlab, Wiley Interscience,
2nd. Ed., 2001.
[7] Jazwinski, A.H, Stochastic processes and filtering
theory. Academic Press, New York., 1970.
[8] Arulampalam M. S., S. Maskell, N. Gordon and T.
Clapp, A tutorial on particle filters for online
nonlinear/non-gaussian bayesian tracking; IEEE
Trans. Sign. Proc., Vol. 50, No. 2, February, 2002)
[9] Hall David L. & J Llinas; Handbook of multisensor
data fusion, USA, CRC Press, 2001.
[10] Masson F.; Navegacin autnoma robusta en
grandes superficies Tesis PhD., Universidad
Nacional del Sur, Baha Blanca, Argentina, Sept.,
2003.
[11] Wei M., K.P. Schwartz, A strapdown inertial
algorithm using an earth-fixed cartesian frame, ION
Navigation: Jr., Vol. 37, N 2, Summer 1990.

[12] Espaa M. Fundamentos de la navegacin integrada;
230 pgs. AADECA Edit., Buenos Aires, 2010.
[13] Espaa M., J. Giribet, C. Miranda; Navegacin
integrada y georreferenciacin directa de sensores
remotos aerotransportados; XX Congreso
AADECA, Buenos Aires, Agosto 2006.
[14] Farrel J. A., M. Barth; The global positioning system
and inertial navigation; Mc. Graw Hill, NY, 1995.
[15] Wagner J. F, T. Wieneke Integrating satellite and
inertial navigation:-conventional and new fusion
approaches; Control Engineering Practice 11, 2003.
[16] Titterton D.H. & J.L. Weston, 1997, Strap down
inertial navigation technology, Peter Peregrinus
Ltd., IEE, London.