For both of these constraints to be met simultaneously, we
must have
W = EAR therefore W zc
D
4Lv 2v 23 Ce)
Thus, the achievable azimuth resolution is proportional to
the swath width. Narrow swaths are necessary to achieve fine
resolution.
As a consequence of the side-looking configuration, SAR images
have an unusual geometry. Look at figure 6, which shows a cross
section at constant azimuth (along track position). Since cross track
position is determined by a range measurement, the SAR effectively
projects points on the terrain onto a line parallel to the line of
sight from the radar. (We must be more careful if a large cross track
swath is imaged due to changes in incident angle across the swath.) We
first notice that, due to shadowing, at certain ranges not returned
signal will be present. We next notice that small hills will be
foreshortened, that is, their peaks will be displaced towards smaller
ranges than will their bases. In an image, they will seem to be bent
towards the radar. For large mountains, the situation can be so
extreme that their peaks can be imaged completely to one side of their
bases. The mountains appear to be lying on their sides; this effect is
called layover.
In addition to being distorted in this manner, SAR images are
subject to a characteristic type of noise. This results from the
coherent nature of the imaging process. The signal at any point on the
image is the vector sum of the phasors representing the radar returns
from all sources of scattering within the resolution of the radar.
This is shown in figure 7. By the central limit theorem, the resultant
phasors will be random normal distributions for the real and imaginary
components. As a result, we can show that the power of the return (the
squared magnitude) will have an exponential distribution
pos nye tb (7)
An exponentially distributed variate has a standard deviation equal to
its mean. This is indeed a noisy type of noise. This type of multi-
plicative noise is called speckle, because of its visual appearance on
SAR images. It can be quite troublesome. One way of dealing with it
is shown in figure 8, Instead of using the entire available SAR data
span to produce one image, the SAR data are divided into sub-spans each
of which is used to create a different statistically independent
image. These images are averaged to quiet the noise. Because
296
ee le E r7? EY
rer