349 
(a) Phase 0 
on 
MTF 
tellite 
ement 
xol in 
nation 
i-orbit 
, their 
'MTF 
Then, 
î-edge 
(duced 
>lution 
l by: 
ng to 
>ed in 
7 ourier 
:tively. 
f OTF 
ly, the 
5 Point 
pacing 
iree is 
: MTF 
isform. 
i use a 
n array 
i these 
to the 
x with 
ligning 
ampled 
(b) Phase 1/4 
(c) Phase 1/2 
(d) Phase 3/4 (c) Reconstructed PSF 
Fig 1 The reconstruction 
2.2 Target deployment/selection standards for point 
source/array method 
For point source method, the primary consideration is the 
at-sensor irradiance intensity. The following standards should 
be obeyed when deploy point source target: 
1) Properly design the irradiance intensity to avoid the 
saturation of image system. Furthermore, to get enough 
contrast and SNR, the point source should be designed so that 
the peak sensor response is on the order of 75% of its dynamic 
range. 
2) A complicating factor for point source method is that 
many high spatial resolution sensors are pointable. Therefore, 
it’s better to use diffusion bodies or convexes as point sources. 
For point array methods, besides the intensity of point 
sources, there are two major considerations need to be 
regarded: 
1) The number of point sources should be as much as 
possible, which is restricted by the size of target, cost, etc. 
Experience indicates that 15-20 point sources can obtain 
reasonable result. 
2) There must be enough distance between different point 
sources so that system’s response to different point sources 
can’t overlap. For many systems, the PSF does not extend over 
3-GSI, thus, a minimum mirror separation of 5-GSI should be 
adequate. 
Figure 2 gives an example of point array target. The point 
sources are diffusion bodies. There are total 16 point sources in 
a 4X4 arrange style. The distance between two nearest point 
sources is 5.25 pixels, and there are 0.25 pixels sample phase 
differences between two nearest point sources. 
Fig 2 Point array target deploy example 
2.3 Data processing steps for point array method 
Usually, the data processing for point array method uses 
parametric approaches. The basic steps include: 
1) Determines peak location of each point source data set to 
subpixel accuracy using a parametric two-dimensional model 
(Gaussian model). 
2) The individual point source data sets are aligned along 
of oversampled PSF 
their model-estimated center positions to a common reference. 
3) The 2-D model is applied again to estimate system’s 
oversampled PSF. 
4) A Fourier Transform is applied to the PSF and normalized 
to obtain the corresponding MTF. 
3 PULSE METHOD 
3.1 The principle of pulse method 
Image system’s response to ideal line is Line Spread Function 
(LSF), which is the integration of PSF in the direction 
perpendicular to the line. According to optical theory, the 
modulus of the Fourier Transform of LSF is corresponding 1-D 
MTF. However, it is impossible to get infinitely narrow line 
source. Thus, the Fourier Transform of LSF obtained from 
image should be corrected according to the width of input pulse. 
The approach divide the spectra of LSF obtained from image 
by the spectral of ideal square pulse with the same width to 
obtain corresponding MTF. 
3.2 Target deployment/selection standards for pulse method 
The target for pulse method should have the following 
characteristics: 
1) Pulse target should consist of a uniformly bright region 
with two homogeneous dark regions. The boundaries between 
the bright and dark regions are straight edges. 
2) The orientation of the pulse target must be maintained so 
that the oversampled pulse response can be obtained from 
imagery (about its principle see section 4.2). 
3) The length of pulse target should be as long as possible, at 
least obtain several ‘slices’ of the PSF in the orthogonal 
direction. 
4) The pulse target width must be designed carefully to place 
the zero-crossings at locations where the value of the MTF is 
not critical to evaluating system performance (especially not at 
Nyquist frequency). 
When calculating system MTF, a Fourier Transform of the 
pulse response and ideal squre pulse function is required. Since 
the well-known Fourier Transform of a square pulse function is 
a sine function, it is necessary to deal with zero-crossings. It is 
apparent that the width of pulse target should be as narrow as 
possible to avoid zero-crossings, say one GSI or less. However, 
the difficulty is that the strength of the signal received by the 
sensor from the narrow pulse width decreases linearly as the 
width of the pulse. As a result, the SNR is compromised and a 
good estimate can not be obtained. Practice has shown that a 
pulse width of 3-GSI is optimal for this type of target. With this 
width, a good tradeoff is reached between obtaining a strong 
signal and maintaining ample distance from placing a 
zero-crossing at the Nyquist frequency (see figure 3).