0% found this document useful (0 votes)
102 views11 pages

Statistical Modulation Transfer Functions: Modeling Imager Deterministic and

Uploaded by

Laura Nita
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
102 views11 pages

Statistical Modulation Transfer Functions: Modeling Imager Deterministic and

Uploaded by

Laura Nita
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 11

Modeling imager deterministic and

statistical modulation transfer functions

Jerris F. Johnson

A unified two-step approach for evaluation of deterministic and statistical modulation transfer functions
(MTF's) is applied to a time-delay-and-integration charge-coupled imager. The deterministic MTF's
include the well-known spatial and temporal aperture MTF's, as well as the charge-coupled imager
interpixel and intrapixel synchronism MTF's, which are derived here. These latter MTF's originate
from nonsynchronous motion (velocity mismatch) between the image on the focal plane and the charge
packets. The statistical evaluation results in phase and jitter MTF's.
Key words: Modulation transfer function, charge-coupled imager.

1. Introduction tion, as well as sensor parameters, such as integration


The objective of an optical imager (generation of an time and pixel dimensions. Second, some of these
exact reproduction of the input scene) is frustrated by parameters may be identified as random variables.
the low-pass spatial and temporal filter character of If the signal and MTF are functions of these random
the device. Analytically, the degradation of the sig- variables, they also have random components. The
nal with spatial frequency (loss of resolution) is output to be expected by a measurement may be
described by the complex optical transfer function, determined analytically by evaluation of the statisti-
whose modulus is the modulation transfer function cal mean of the signal. This evaluation results in
(MTF). The resolution of an optical imaging sensor additional degrading factors that modulate the sinu-
may be determined experimentally under controlled soidal component of the input in the same manner as
conditions, and analysis may confirm the results. the deterministic MTF's and that are referred to here
However, the performance obtained in the field may as statistical MTF's. The statistical MTF calcula-
vary considerably from that found under the con- tions require probability density functions for all
trolled conditions in the laboratory because of the random variables involved. A measure of the scatter
statistical nature of scene parameters that previously of the expected measurements about the mean is
had been considered deterministic. For example, given by the standard deviation of the signal with
jitter motion between the image on the focal plane respect to the random variables.
and the focal-plane detectors results in degraded In this paper the various aperture MTF's of a
sensor resolution, which may be described by an charge-coupled imager operating in the time-delay-
MTF. and-integration (TDI) mode are determined from a
The evaluation of imager deterministic and statisti- single expression. The optical transfer function of a
cal MTF's proceeds as follows. First, the imager charge-coupled imager contains contributions from
signal output is evaluated analytically for a sinusoidal the optical system, the focal plane, and the finite
input scene. The MTF may, in general, be identified sampling width of the multiphase pixel spatial and
in the output as the factor that modulates the sinusoi- temporal apertures. For completeness the well-
dal component of the input. The output contains known aperture MTF's are derived together with the
scene parameters, such as scene phase and orienta- new expressions for the TDI image and charge-packet
velocity-mismatch (synchronism) MTF's. A form for
a velocity-mismatch MTF has been presented previ-
uosly,12 but this expression does not reduce to the
The author is with the Electro-optical Systems Department,
Sensor Systems Subdivision, Aerospace Corporation, P.O. Box
correct form in the limit that the number of TDI
92957, Los Angeles, California 90009. stages is equal to unity. The synchronism MTF's
Received 22 April 1992. developed here do have the correct limiting form.
0003-6935/93/326503-11$06.00/0. Only the MTF factors that are due to the spatial and
© 1993 Optical Society of America. temporal apertures (pixel geometrical structure and

10 November 1993 / Vol. 32, No. 32 / APPLIED OPTICS 6503


integration time) are considered. To this end, the location; Ro(X) is the pixel spectral responsivity;
transfer inefficiency MTF and the electronics MTF At(t - t), A(x 1 - x), and Ay(y1 - y) are the pixel
are not included. In addition, the focal plane is temporal and spatial aperture functions. Rectangu-
assumed to be of uniform responsivity, and MTF lar space and time apertures are used here, but the
effects occurring within the focal-plane material, e.g., following evaluation could be made for apertures of
the diffusion MTF, are not considered. Of course, all arbitrary shape. The total charge generated within
of the above-mentioned MTF's must be used in the the pixel at the location (x1, Yl, t 1) by the irradiance
evaluation of the overall system MTF. within the spectral bandwidth AXis given by
The parameters in the output signal that are
treated as random variables are the phase and the
position vector of the sinusoidal scene image with Q(pl, t) =f dX Q(pi, t, ). (2)
respect to the focal-plane pixels. A statistical aver- HA
age over the random position vector produces the
jitter MTF, and an average over the random scene This is just an extension of a formula used by Barbe.3
phase produces the phase MTF. Sensor parameters The charge is a function of the wavelength, and the
could also be treated as random variables, but usually integration over the spectral band, Eq. (2), may be
the variation in sensor parameters, e.g., integration performed at any stage of the following development.
time, is small enough to permit these parameters to An analysis of the incoherent imaging of extended
be regarded as constant. objects yields 4
The objectives of this study are twofold: (1) to
describe a unified two-step process for the evaluation H(k, t, X) = Top(k, X)Hg(k, t, X).
of deterministic and statistical MTF's of electro- (3)
optical devices and (2) to apply this process to derive H(k, t, X) is the spatial Fourier transform of the
the MTF's of a TDI charge-coupled imager. The focal-plane irradiance that appears in Eq. (1); k =
organization is as follows: charge-coupled imager xk, + yky is the two-dimensional spatial wave vector
response to a sinusoidal irradiance is determined, the with magnitude k = k I = 2,rrf5 , where fsis the spatial
TDI mode and synchronism error are described, and
the phase and position averages of the signal output frequency; Top(k, ) is the normalized optical trans-
are evaluated. Calculations that would impede the fer function of the optical system that is used to image
flow of the development are relegated to the appendi- the scene onto the focal plane; and Hg(k, t, X) is the
ces. Fourier transform of the image predicted by geometri-
cal optics. This geometrical image is the exact rep-
2. Analysis
lica of the object, magnified and possibly inverted in
the image plane.5 The normalized optical transfer
A. Charge-Coupled Imager Response to Sinusoidal function may be written as
Irradiance
A charge-coupled imager consists of an optical system Top(k, X) = MTFop(k, X)exp[i(D(k, X)], (4)
coupled to an array of photosensitive charge-coupled
elements (focal-plane pixels).3 The signal that is where MTFop(k, X) is the MTF of the optics and
output by the imager is proportional to the electrical cD(k,X)is the optical system phase transfer function.
charge accumulated by each pixel in response to the The optical transfer function of a diffraction-limited
focal-plane irradiance. After an integration time, optical system is always real and nonnegative; i.e., the
the charge accumulated in the array elements is phase transfer function is zero.5
scanned off the focal plane; thus the imager samples The geometrical focal-plane irradiance may be for-
the scene in time as well as in space. mulated in terms of the target spectral radiance.
Consider the time-dependent focal-plane spectral For a sinusoidal target of infinite extent the geometri-
irradiance h(p, t, X), where p = xx + yy is a two- cal image may be written as
dimensional vector that specifies position on the focal
plane, t is the time, and is the wavelength of the hg(p, tX, k, v, p)
optical radiation. The charge generated by this irra-
= T(X)Ho(X)[1 + m cos(ko *p -ko vt + up)], (5)
diance and collected at time t by a single pixel at the
focal-plane coordinates (xl, Yl)
where T(X)is the product of the atmospheric and

Q(p1 , t1, X) = Ro(X) f dt At(t 1 - t) f dx A,(x 1 - x)


optics spectral transmission factors, Ho(X) specifies
the scene spectrum, m is a constant modulation
factor, ko is the spatial wave vector of the sinusoid, v
is the velocity of the wave that is traveling on the focal
x dy AY(y - y)h(p, t, X). (1) plane in the x direction (v = xvx + yvy, with vy = 0),
and p is the phase of the wave at the space-time
origin. The argument list of the irradiance in Eq. (5)
where P, = xx1 + YYiis the vector describing the pixel is expanded to include the wave velocity v and the

6504 APPLIED OPTICS / Vol. 32, No. 32 / 10 November 1993


y charge is expanded in analogy with the list used for
the irradiance. Equation (7) will not be evaluated.
Instead, the discussion is extended to an imager
operating in the TDI mode. Both the form of and
the solution to Eq. (7) are obtained from the more
general case below.
B. Time-Delay-and-Integration Implementation,
Multiphase Pixels, and Synchronism Error
An imager with a charge-coupled focal plane offers
the possibility of signal-to-noise enhancement with
the use of the time-delay-and-integration (TDI) mode
of operation.3 In this mode the signal charge pack-
ets are clocked through the imaging CCD stages (TDI
column pixels) at the same velocity that the image is
scanned across the focal plane. At each pixel, signal
charge is generated in response to the image, and this
charge is added to the total signal charge. But
ORIGIN because the noise in each pixel is independent, the
Fig. 1. Portion of a sinusoidal wave of infinite extent (within
variance of the noise is additive;6 i.e., the standard
dashed lines; the solid lines represent lines of constant phase) with deviation of the noise at the end of the TDI column is
spatial frequency vector ko traveling in the x direction with velocity the root-sum-square of the noise from each of the TDI
v across a single pixel located at (xl, yj). column pixels. This results in an improved signal-to-
noise ratio at the focal-planeoutput. However, if
the image and the signal charge packets move nonsyn-
phase p. Figure 1 shows a portion of the wave chronously (with mismatched velocity) across the
described by Eq. (5) impinging upon a single pixel of focal plane, the signal suffers an additional resolution
dimensions (Ix, y) whose geometric center is located loss over that encountered with synchronous motion.
at P. The wavevector ko has polar coordinates This additional loss may be described by an MTF, as
(ko, 0), and the magnitude of the wave vector in the x discussed below.
direction is ko cos 0. The focal-plane irradiance is The TDI concept is well known and is illustrated in
determined by use of the optical transfer function, Fig. 2. The image motion across the focal plane of a
Eq. (4), and the Fourier transform of the geometrical scanning sensor is the result of the uniform sensor
image, Eq. (5), in Eq. (3) and by calculation of the motion across the scene. Shown are two pixels in a
inverse Fourier transform of the resulting product. TDI column, denoted by P1 and P2 , at the times t = 0
The result is and t = Ts, where Ts is the sample period. At t = 0 it
is assumed that there is no charge in any pixel. The
h(p, t, X, ko, v, p)
= T(X)Ho(X){1+ m MTFop(ko, X) SCENE

X)]}. (6)
x cos[ko p - ko vt + p + CD(ko,
OPTICSAND
Inserting Eq. (6) into Eq. (1) yields the signal charge FOCAL PLANE
that is due to the sinusoidal irradiance [Eq. (5)]
produced by a single pixel:
P1
Q(p 1 , t, X, ko, v, p)
/1
\~~~~~~~__ 2
PATCH
= IxlyTRo(X)T(X)Ho(X){1 + M MTFop(k0 ,X)
A

t+T IXi
+ 21x r1y+½21y

x dt dx dy
ti xl-&21x ' 1 - 2ly PATCH SENSOR
B VELOCITY
x cos[ko p - ko vt + p + 1D(koAx)]|, (7)
.

where T is the signal-pixel integration time and the


explicit form of the rectangular space and time aper-
tures is used. The argument list of the collected Fig. 2. Ideal TDI implementation.

10 November 1993 / Vol. 32, No. 32 / APPLIED OPTICS 6505


signal is integrated for a pixel integration time and single-pixel aperture replaced by a sum of TDI col-
then transferred to the next pixel in the TDI column. umn apertures. The TDI accumulated charge for N
During the time interval beginning at t = 0 and pixels, QN, is
ending at t = Ts, denoted by (0, Ts), pixel 1 (P1 )
continuously scans the scene from an initial scene QN(P1,t1, X,ko, v + AV, p)
footprint, Patch A, to a final footprint, Patch B. At
the time t = Ts the charge packet accumulated in P2 is
transferred to P3 (not shown in the figure); simulta-
neously, the packet in P1 is transferred to P2 ; i.e., the
= Ro(X) f dtAt,,(t - t) f dxAxn,(xl - x)

signal from the scan of A to B that was accumulated


in P1 is now in P2. Pixel P2 at t = Ts is imaging the x dy Ay(yi - y)h(p, t, X, ko, v + AV, p), (8)
same patch (A)that P1 was imaging at t = 0. Pixel P2
now scans from A to B in the interval (TS, 2TS) in
exactly the same manner that P did previously. where the sum is over the TDI pixel apertures, and it
Hence in this ideal case the charge packet in P2 at t = is assumed that the pixels in the TDI column have
2Ts is twice that in P1 at t = Ts; i.e., the signals from identical spectral response.
P1 and P2 are added. The perspective difference Charge transport in a charge-coupled imager is
between the observation of A by P1 at t = 0 and by P2 implemented with multiphase pixels. A pixel con-
at t = Ts is exaggerated in the figure. In reality the sists of a set of electrodes, or phase gates, usually
object distance R is much larger than any other from two to four, whose electrical potential is con-
distance in the figure, and all angles are reduced so trolled by external clock voltages. Clocking wave
that the charge generated by the two samples is forms (voltage as a function of time) induce potential
approximately the same. wells under the electrodes and control their motion as
A portion of a sinusoidal irradiance is shown scan- well as the motion of the electrons residing in the
ning across the TDI focal plane in Fig. 3. The pixel potential minima under each pixel. The pixel spatial
centers are separated in the x direction by the sam- aperture moves with the potential wells.
pling distance s, which is equal to the pixel length 1x A portion of a sinusoidal focal-plane irradiance
for contiguous pixels. The sampling frequency is pattern is shown scanning across two four-phase
1/sx, and the Nyquist frequency in the scan direction pixels in Fig. 4. The TDI signal charge collected by
isfN,X= 1/(2sx). The charge-packet transfer velocity N pixels, each with M phase gates, is
is v = xvx = xsx/Ts, and the image velocity is v + AV,
where AV xAvx + yvy is the velocity mismatch. QN(p1,t, , kov + AV, p)
In the case of ideal TDI, A = 0, and the image
velocity is matched to the charge-packet motion. N M 1tl+(n-1)T+mT/M-V2T/M
The signal charge generated by the TDI array is = Ro(X) I J dt
n=1m=11t+(n-l)T+(m-l)T/M--2T/M
given by an expression similar to Eq. (1) but with the

x
fxi+(n-1)1x+(m-1 dx
xi+(n-1)lx+(m-1)lx/M-½1x
JY+21y
dy h(p, t, , k, v+AV, p), (9)
Y1- Y21y

where (xl, yl) are the coordinates to the center of the


first pixel in the TDI column and the sample time
TS is equal to the single-pixel integration time T.
The temporal aperture is symmetrized to the form
( - 1/
2 T/M, 1/2 T/M) in analogy with the spatial aperture.
The indices m and n in the integral limits cause the
P
pixel spatial aperture, which is always of length , to
move lx/M, i.e., one phase-gate length, between the
subpixel integration intervals of duration T/M. The
four-phase pixel aperture shown in Fig. 4 moves 1x/4
in T/4. The spatial aperture and the charge packet
ORIGIN move with velocity x(lx/M)/(T/M) = xlx/T = v
Fig. 3. Portion of sinusoidal wave of infinite extent (within
(recall that the sample distance is equal to the pixel
dashed lines; the solid lines represent lines of constant phase) with length for contiguous pixels), and the image irradi-
spatial frequency vector ko, traveling with velocity v + Av across ance moves with velocity v + AV. Figure 4 shows a
four TDI column pixels. The first pixel (n = 1) is located at the case in which AV = 0. The (m = 1, n = 1) term in
focal-plane coordinates (xl, yl). Eq. (9) corresponds to Fig. 4(a), and the (n = 1, m = 2)

6506 APPLIED OPTICS / Vol. 32, No. 32 / 10 November 1993


t = -T/8 t = T/8 irradiance, and the result is

(a)
_~~~~,
_ /a y
-,'/
GATES QN(P1, t, X, kov + Av, p)
= N1l.yTRo(X)T(X)Ho(X){1
+ mMTFop(ko, X)
x MTFA(ko)MTFsyNc(ko)cos[ko p
-efx2 0 ex2 - ko- (v + Av)tl + p + 1D(ko,
X)+ 8]I, (10)

t= T/8 0 -' t=3T/8 where the aperture MTF takes the well-known sinc
function form for rectangular apertures [sinc(x) =
sin(x)/x]:
(b) He~~~~ }
MTFA(ko) = sinc('/ 2 kolx cos 0)sinc('/2 koly sin 0)

-fx/4 0 3eQ4 x sinc[1/2k0 - (v + Av)T/M]. (11)

The aperture MTF is the product of the two-


t 3T/8 t =5T/8
dimensional spatial-aperture MTF and the temporal-
aperture, or smear, MTF. The smear MTF, which
(C) depends on the integration time of a single phase
gate, increases as the number of phase gates increases
because the pixel aperture tracks the scanning image
0 o f9
e~~~~~x more closely.
The synchronism (error) MTF derived in Appendix
A,
t =5T/8 ",-t
\ =7T/8

sinc('/ 2Nk 0 AvT)


. sinc('/ 2k0 AvT)
.

(d) L l ll lI MTFSyNc(ko) =
sinc('/ 2k0 Av'T) sinc('/2ko AvT/M)

0 fx/4 5Ce/4 (12)


Fig. 4. Portion of sinusoidal focal-plane irradiance scanning depends on the image-charge velocity mismatch and
across multiphase pixels of length l and M = 4 phase gates per is the product of an interpixel synchronism MTF (i.e.,
pixel: (a) Irradiance at t = -T/8 is shown as a solid curve and
the first factor, containing N, the number of TDI
irradiance at t = T/8 is shown as a dashed curve. (b), (c), (d)
Successive portions of the scanning irradiance and the moving
pixels) and an intrapixel synchronism MTF (the
charge packet within a charge-coupled imager. During each sub- second factor, containing M, the number of phase
pixel integration time (of duration T/4) the photocharge is inte- gates per pixel). The result may be simplified by the
grated over the pixel length lx. cancellation of similar terms in the numerator and
denominator. The synchronism MTF derived here
correctly reduces to unity for the number of TDI
term corresponds to Fig. 4(b) with apertures stages and the number of gates equal to one (N = 1
(ti + T/8, t + 3T/8) and (xi - 1x/4, xi + 31x/4). and M = 1), in contrast to previously derived formu-
If Av = 0, then Eq. (9) describes the ideal TDI las.1'2 The synchronism MTF, Eq. (12), has been
operation of a charge-coupler imager. Equation (7) experimentally verified in Ref. 7.
is regained by substitution of Eq. (6) into Eq. (9) and The MTF is the modulus of the optical transfer
by use of N = 1 (single pixel), M = 1 (single phase function and is therefore nonnegative. However,
gate), and Av = 0. Equation (9) contains an addi- the sinc functions appearing in Eq. (11) may be
tional temporal phase shift over Eq. (7), i.e., the negative and are actually optical transfer functions of
symmetrizing - '/2 T/M in the limits of the temporal the aperture, i.e., the Fourier transform of the aper-
integral, which is easily removed by redefinition of t. ture. Following common practice in the imaging
In the case of synchronism error the charge packet community, to the sinc functions in Eq. (11) are
does not move at the same velocity as the image. referred to as MTF's to avoid defining a complicated
Then the TDI aperture velocity is v = xlx/T and the phase function that would account for sign reversals
image velocity is v + Av. The velocity mismatch as the argument of the sinc function increases.
causes the signal from various scene patches to be The result, Eq. (10), contains a replica of the
mixed in the imager. Equation (9) is the basic geometric image [Eq. (5) with velocity v + Av] evalu-
expression to be used in the evaluation of the TDI ated at the location of the first pixel, i.e., at the
charge-coupled imager aperture MTF's in the case of space-time point (x1 , yl, t), and modulated by the
nonzero synchronism error velocity, Av. The evalu- optics, aperture, and synchronism MTF's. In addi-
ation is performed in Appendix A for a sinusoidal tion, the replica is shifted in phase by the optical

10 November 1993 / Vol. 32, No. 32 / APPLIED OPTICS 6507


phase transfer function and by the quantity cal probability density functionfc,(u), then the statisti-
cal average (mean) of F is 6
1 1 M-1I
=- 2 (N - )ko *AvT - 2 M ko AvT, (13) Fb
pp@
= (F(Ct)) = Jdu f2(u)F(u), (14)
which is referred to here as the synchronism phase
shift. Some special cases of Eq. (10) follow.
where the range of d is (a, b) and the angle brackets
denote an average over the density function. The
1. Single Pixelwith Single Phase Gate variance of F is the square of the standard deviation
In this case N = 1 and M = 1, and the result is the and is defined to be the second moment of F about its
solution to Eq. (7). The synchronism MTF is unity mean:
(i.e., the sinc functions cancel giving the correct
limiting form), and the synchronism phase shift
vanishes. However, there is no mechanism for keep-
oF2 = ([F(d) - p]2 )u = jidu fg(u)[F(u) - p2. (15)
ing the charge packets separate with single phase-
gate pixels. For a single multiphase pixel (N = 1 and The evaluation of the mean of the signal charge is
M > 1) the synchronism MTF is not unity. slightly more complicated than that indicated above
because the signal depends on more than one random
variable. In the case that the position and phase are
2. Zero Aperture Width random variables, the mean of the TDI charge [Eq.
If the space and time apertures are reduced to (10)] is
infinitesimal width (lx -> 0, ly -> 0, and T - 0) and if
there is no velocity mismatch, then the MTF's are P = (QN(I1,t 1 , , k, v + Av, P (16)
unity [sinc(0) = 1] and a perfect replica of the focal-
plane irradiance [Eq. (6)], is obtained. However, no xl and Yi are the random components of the pixel
energy is collected by a zero width aperture, and no position vector, p = x + YYi,which terminates on
signal is generated. the first TDI pixel, and pis the phase of the wave at
the space-time origin [see Eq. (5)]. and y, are
random variables arising from jitter, and the use of (p
3. Zero Synchronism Error as a random variable is discussed in the subsection on
In the case that the image-charge-packet velocity the phase MTF. The statistical evaluation must be
mismatch vanishes, Av - 0, the synchronism MTF is performed over the joint probability density of il, Y1,
unity, and the synchronism phase shift 8 is zero. and . However, the complication that is due to the
joint density may be reduced because the scene phase
and position are independent random variables.
C. Statistical Evaluations Then the joint density is the product of the individual
Parameters that describe the scene relative to the probability density functions,6 and Eq. (16) may be
focal plane must be treated as random variables if the written as
above results are to be applied to operational sensors
observing real scenes (whose content is not known a
priori). This means that the charge output of the = dxl dy1 fx1j 1 (x 1, y 1 )
imager [Eq. (10)] is a random variable. The param-
eters that are considered to be random variables in
the following analysis are the position and the phase x dypf(p)QN(pl,t1, X, ko, v + Av, p), (17)
of the sinusoidal scene with respect to the focal plane.
In contrast, charge-coupled imager experiments in
the laboratory usually view a scene whose phase is where f, 1(x1,y 1 ) is the joint density function of
fixed and whose position with respect to the sensor is 1 with Yl and f(p) is the density function for the
fixed and stationary (i.e., with no jitter). phase.
A meaningful question is, what is the mean perfor- The statistical problem could be expanded in scope.
mance of the imager upon viewing a sinusoidal irradi- If the image velocity v + Av were to be treated as a
ance with random phase and position? That is, if the random variable, then Eq. (17) would contain another
observation experiment were performed a sufficient integral over the statistical density function of the
number of times and if the average of these experi- velocity mismatch.
ments were taken, what would be the expected signal?
A related question is, what is the scatter of the values 1. PositionAverageand Jitter Modulation
of the signal about its mean, i.e., what is the variance? Transfer Function
Suppose F is a function of the random variable The scanning focal plane is undergoing uniform
(random variables are denoted by a tilde); then F is a motion with respect to the scene, and jitter motion is
random variable. If ais distributed with the statisti- superimposed upon that uniform motion. The rela-

6508 APPLIED OPTICS / Vol. 32, No. 32 / 10 November 1993


tive jitter motion between the scene and sensoir may that the random position of the first TDI pixel, P1,is
be modeled by consideration of an image that rmoves replacedby its mean value, tjpl = () = Xii- + yu-
with no jitter and a pixel position vector P, (see F ig. 3), In addition, the factors on the right-hand side of Eq.
which is a random variable whose statistical dEensity (21) are exactly the same as those in Eq. (10), except
function is that of the jitter. That is, the pixel for the inclusion of an additional modulation factor,
spatial apertures are jittering as the image r noves the jitter MTF, which is given by (Appendix B)
over them. The x1 and Yi statistical averages of the
signal, i.e., Eq. (10) substitutedin Eq. (17), ar(e now MTFJIT(ko)
evaluated. In the remaining portion of this st bsec- = exp(- / 2ko2 2 cos2 O)exp(- /2k 2rU2 sin 2 0). (22)
tion the subscript on the pixel position coordineates is
suppressed in order to simplify the notation; i
and Yi are denoted as i and j. If x and 9 are ji int ly The signal [Eq. (21)] is now modulated by optics,
normal random variables, then the joint doensity aperture, synchronism, and jitter MTF's. The jitter
function of x andy is given by6 MTF derived here is the result of a Gaussian probabil-
ity density function for the relative position of the
exp- (iX2 - 2rxy + tY2)/[2(1-r2 scene on the detector. Other density functions could
fg,(x, Y) = produce different dependencies of the jitter MTF on
21Tu-oy(1 _ r2 )1/2 the jitter parameters (the mean and the standard
(18) deviation). The jitter MTF is not an MTF in the
usual sense because it is not the modulus of the
where Fourier transform of a signal chain transfer function.
However, the jitter MTF does appear in the same
(X =(X ~)
FX / UtX (19a) location as the deterministic MTF's, and the replica
=x (x), form of the wave is maintained. The jitter MTF
(19b) determined above is an example of what in this paper
is called a statistical MTF.
2 ((i- 2
(19c) The subscript denoting the pixel location, e.g.,
and r is the correlation coefficient of x and 9. The subscript 1 on x1, is dropped from the standard
parameters ax and ux are the mean and the stairadard deviation of the jitter because it is assumed that all
deviation of x. Similar expressions hold for 9. The pixels are rigidly connected in the focal plane, and
joint density function, which relates the ra ndom hence they jitter together with the same cr. The
variables x and 9, gives a Gaussian density furiction variance of x is given by the frequency integral of the
for each variable; i.e., the integral of Eq. (18) ovefr ally one-sided power spectral density for jitter motion in
gives a normal (Gaussian) density for x, ancI vice
, - the x direction
versa. The correlation coefficient r indicate s the
amount of correlation between these random vari- a2 = fdfSe(f). (23)
ables, where 0 < r < 1.
If the jitter motions in the two direction s are
independent, as is assumed here, then i and j are A similar expression holds for y. If 2the variances in
uncorrelated (r = 0), and the joint density furiction the two directions are equal, oy,2 = a = .Y2, then Eq.
factors into individual density functions (22) reduces to a standard form 2 :
2 2
fl(x, A) = f1(x)f9(y), (20) MTFJIT(ko) = exp(-k 0 0 /2), (24)
each with Gaussian form, as can be seen from Eq. and the MTF is now independent of orientation angle.
(18). The x and y integrations in Eq. (17) may now Equation (24) is the jitter MTF for uncorrelated
be performed independently; this is done in Appendix equal-variancejitter.
B. Then the position averaged signal is given by (the Jitter degrades high-frequency spatial image contri-
subscripts on the pixel position are now explicit) butions more than low-frequency contributions, as
shown by Eq. (24), because of the statistical averaging
(QN(01, t1, X,ko,v + AV,(P))j
1,y1 (smearing) of the wave over jitter positions. For a
+ m MTFop(ko, X)
= NlxlyTRo(X)T(X)Ho(X){1
given amount of jitter, characterized by r, scene
components with small spatial frequency ko << 1/u
x MTFA(ko)MTFsyNc(ko)MTFJIT(ko) are not greatly degraded. For focal planes with little
jitter (small (), only very-high-frequency spatial scene
X cos[ko* - ko (V + AV)tl + (p + P(ko,X) + a]1, components are degraded.
(21)
2. PhaseAverageand Phase ModulationTransfer
where the replica of the geometric image is main- Function
tained in Eq. (21); i.e., the form of the cosine in Eq. In an operational system that views real, uncon-
(10) is not changed by the above calculation except trolled scenes, as opposed to a system setup in the

10 November 1993 / Vol. 32, No. 32 / APPLIED OPTICS 6509


laboratory, it is not known a priori what the phase of form cos(a + p), where a is described above. The
a sinusoidal scene with respect to a sensor pixel will random phase of the image could be anywhere within
be. Then the phase of the scene sinusoid at the the interval (-ir, r). The sinusoid is sampled in
space-time origin, p [see Eq. (5)], which is a constant time by the imager, and a sequence of numbers
during a single imaging experiment, is a random (voltage samples) is generated. (This sequence recon-
variable from one imaging experiment to another. structs the image.) The sinusoidal image is sampled
In addition, for a given imaging experiment, i.e., the every T seconds, and, as can be seen from Eq. (21), the
imaging of a sinusoidal scene, p is independent of the phase difference between two samples is A = ko
pixel position vector (xl, yl). Once a sinusoid is (v + Av)T. One may sample any point on the sinu-
acquired for imaging, however, the phase at the soid (within -rr to rr)by beginning the sampling at an
space-time origin is fixed during the generation of appropriate point within the interval (0, A'p) and
that image. For a given imaging experiment the incrementing by Ap. This reduction in range oc-
phase may take any value within its statistical distri- curs because of the equivalence of sampled images
bution. That is, the phase is a random variable that shifted by an amount A, i.e., by T. The starting
varies as the scene changes, e.g., as in separate range may be translated to the equivalent interval
imaging experiments in which scene sinusoids of (-A'p/2, A'p/2).
arbitrary phase are imaged. The phase MTF that is It is now assumed that the scene phases encoun-
developed here describes the average response pro- tered are uniformly distributed over their range.
duced from multiple imaging experiments and is the Then the normalized phase density function is
expected (mean) MTF that results from field tests on
real sinusoidal scenes with random and unknown
phase. A statistical average over all possible phases -(g)= 1/(A') -/ 2A'P< 'p< 1/
2A'p (25)
may be performed, i.e., the phase integral in Eq. (17) 0 otherwise
is evaluated and is the average of multiple imaging
experiments on the same sinusoidal scene. The re- The phase average of the signal charge is determined
sult of averaging is a modulation factor, a statistical by evaluation of the phase integral in Eq. (17); the
MTF, which describes the average performance of the signal charge is given by Eq. (21). This evaluation is
operational system. performed in Appendix C by use of the density
The sensor response can depend strongly on the function given in Eq. (25), and the position and phase
value of the phase. An extreme case occurs when the averaged signal charge is
frequency of the sinusoid is equal to the focal-plane
Nyquist frequency, i.e., when the focal-plane sample = (QN(01,t, , k, v + Av, ')X1,Y
1,(
distance is one half the sinusoid wavelength. Then, = NxlTRo(X)r(X)Ho(X)t + mMTFop(k 0 , X)
dependent upon the relative phase of the sinusoid
with respect to the pixels, the signal output by the x MTFA(ko)MTFsyNc(ko)MTFJIT(ko)MTFPH(ko)
sensor may have either maximum or zero modula- X cos[ko* -ko (v + Av)t + (D(ko,X)+ 8]).
tion.
In cases in which the phase is not a random (26)
variable from one imaging experiment to the next,
such as in certain laboratory experiments with fixed The differences between the jittered signal [Eq. (21)]
sinusoidal scenes, the phase remains at a single value and its phase average [Eq. (26)]are the elimination of
'po,and the phase statistical density function is given 'p from the replica (the phase in the cosine is replaced
by the Dirac delta function (' - p0). Then the by its mean, zero) and the introduction of a new factor
phase integral in Eq. (17) may be immediately evalu- in the modulated term. This factor,
ated, and the signal output depends on cos( + o),
where a is the argument of the cosine in Eq. (21) MTFPH(ko) = sinc(/ 2Ayp), (27)
excluding p. That is, the replica is evaluated at po,
and there is no modulation factor (the MTF is unity). is referred to as the phase MTF, where Ap = ko
This is analogous, in the case of jitter, to a vibration- (v + Av)T. The phase MTF given in Eq. (27) is the
free experiment in which the standard deviation of result of a uniform density function, and other den-
the jitter is zero and the jitter MTF is unity. sity functions could result in a different dependence
Another possibility, which may be more appropriate onA'p.
for a laboratory setup or for field calibration experi- In the case that 0 = 0 and Av = 0 the phase MTF
ments, is that the phase of the sinusoid is known with reduces to a form that has been used previously.2
some small error. Then the density function could For the uniform density function the phase MTF
be taken as a Gaussian whose standard deviation is a takes the same form as the smear (temporal aperture)
measure of that error. MTF. These two functions are similar in form be-
The phase average in Eq. (17) is now performed. cause the temporal aperture is also uniform (over the
The phase appears in the imaged replica QN in the interval T/M).

6510 APPLIED OPTICS / Vol. 32, No. 32 / 10 November 1993


3. Other Random Variables output. The statistical MTF's appear in the same
Other scene parameters could be treated as random location as deterministic MTF's, and, as in the result
variables, such as the image velocity or 0 (the orienta- of the deterministic MTF evaluation, the replica of
tion angle of the sinusoidal scene with respect to the the sinusoidal image is preserved. Different density
focal plane). Statistical averages over parameters functions may produce different statistical MTF's.
may not always result in separate factors that can be In general, scene parameters offer a wider range of
identified as MTF's. For example, averaging the variation than sensor parameters.
mean signal charge over the orientation angle intro- The procedure described above was applied to a
duces a 0 integral and a 0 density function into Eq. charge-coupled imager operating in the TDI mode.
(17). The statistical density function that describes The deterministic MTF's that were evaluated include
the orientation of the sinusoidal scene could be taken the well-known two-dimensional spatial aperture and
as uniform in the interval (-7r, r) with density 1/(27r), temporal aperture (smear) MTF's, as well as the new
if no prior knowledge of the scene were known. TDI synchronism (image-charge-packet velocity mis-
Because the previously determined MTF's, as well as match) MTF, which was developed in this paper.
the replica, contain 0 in a complicated manner, evalu- The synchronism MTF consists of a multipixel contri-
ation of the orientation integral over the MTF's and bution (the interpixel synchronism MTF) and a multi-
replica does not produce a separate prefactor that can phase gate contribution (the intrapixel MTF). The
be identified as an orientation MTF. Nevertheless, statistical MTF's that were evaluated were the phase
the integral evaluation may be used to determine the MTF and the jitter MTF, and these reduce to stan-
degradation in response as a function of spatial dard expressions under simplifying assumptions.
frequency. A number of improvements to the calculation,
which could be handled by the formulation, deserve
4. Variance mention. The possibility of TDI pixels with varying
responsivity may be treated by indexing of the spec-
For completeness the variance of the collected charge tral response RO(X)with a pixel index parameter and
is calculated. The amount of scatter of the TDI by inclusion of the indexed response under the sum in
photogenerated charge about the mean value is given Eq. (8). Other parameters, such as image velocity,
by the standard deviation [see Eq. (15)] could be treated as random variables; the only require-
ment is a probability density function for each ran-
go = ((Q - RO)ixj" (28) dom variable and joint density functions for noninde-
pendent random variables. If the orientation angle
where pQ is the mean charge [Eq. (26)] and the were treated as a random variable, then the orienta-
statistical evaluation is over all of the random vari- tion dependence of the optics could be included in the
ables. This expression is evaluated in Appendix D by orientation averages in a straightforward manner.
use of all of the assumptions that were used to
evaluate the mean; the result is Appendix A: Evaluation of Eq. (9)
0 2 = [N1llyTR 0(X)T(X)Ho(X)] 2 Substituting the focal-plane irradiance [Eq. (6)] with
velocity v + AV into the equation for the TDI
x [mMTFop(ko, X)MTFA(ko)MTFsyNc(ko)]2 accumulated charge [Eq. (9)] gives
2 2
x [1/2 + /2 sinc(A'p)exp(-2k 0 u ) t1, X,ko,v + AV, )
QN(P1,
X cos(2ko- pjl+ 2C) - MTFPH2(ko) N M f't1+(n-i)T+mT/M-½2T/M

2 2
= (X)Ho(X)Ro(X) I
n=lm=i t+(n-i)T+(m-i)T/M-½2T/M
dt
X MTFJIT (ko)cos (k 0 - 5lo + C)], (29)
xd+(n-x)l+(m-1)lx/M+Y21x
where the MTF's are defined in Eqs. (11), (12), (22), X eeeeeeeeeeeeeeeeeeeeeeeeeeeeee
dx
and (27); Ap is the quantity in the phase MTF; and C x+(n-)1.+(m-)x/M-21x
appearing in the argument of the cosines is given by
J1+ Y21y
C = -ko (v + Av)tl + I)(ko,X) + 8.
x dy{1 + mMTFop(ko, X)
Y1- 21y
3. Conclusion
The main purpose of this study is the description of a x cos[ko p - ko (v + Av)t + p + cF(ko,A)]}. (Al)
unified two-step process that has been developed for
evaluation of deterministic and statistical MTF's. The integral over y may be written as
Deterministic MTF's are the moduli of the Fourier
transform of signal transfer functions, and they I = dy[1 + A cos(koyy + B)], (A2)
appear in the output as modulating factors to sinusoi- 1- Y21y
dal input. Statistical MTF's are the result of statis-
tical averaging over the density functions of random where koyis the y component of the spatial frequency
variable parameters that appear in the deterministic vector (koy= ko sin 0), the factor A consists of the

10 November 1993 / Vol. 32, No. 32 / APPLIED OPTICS 6511


cosine prefactors, and B contains everything in the uct of the x and y density functions. The statistical
argument that does not depend on y. The result of average overtl andyi may now be performed indepen-
the integration is dently. The average in Eq. (B1) over the random
variable x may be evaluated with the i1 density
I = y[ + A sinc('/ 2 k0yly)cos(k 0yy1 + B)], (A3) function [see Eq. (18) with r = 0 and Eq. (20)].
Explicitly, the average is, with D = koyyl + C,
where sinc(x) = sin(x)/x and the trigonometric iden-
tity (cos(koxic + D))Xj

sin(c + d) - sin(c - d) = 2 cos(c)sin(d) (A4) exp[-i/2(Xi -


is used. As can be seen from Eq. (A3), the aperture
= dxl cos(ko.xl + D)
(2r) 1/2 ox
MTF in the y direction is now a prefactor to the wave
replica (the cosine evaluated aty = yl). Similar evalu- = exp(-k 0 2(r., 2/2)cos(k 0.,pg1+ D). (B5)
ations may be made for x and t; the result is
The statistical average over y, produces a similar
QN(Pl, t, X, ko, v + AV, p) result. Then the average over the random position
is
= T(X)Ho(X)Ro(X)NlxyT{1+ mMTFop(ko, X)
(QN)eiy = A[1 + B MTFJIT(ko)
MC p x cos(koxpxl + koypy, + C)], (B6)
x MTFANkO) mI cskO - kO (V +AV)ti
where MTFJIT(ko)is the jitter MTF [Eq. (22)]and the
cosine term is the replica of the wave evaluated at the
+ p + (D(ko, X) - (n-1+ - 1)11
AvT(5 mean x and y positions. When A, B, and C are
replaced, the result is Eq. (21).
(AS)

Use of the formula8 Appendix C: Phase Modulation Transfer Function


n-i The phase average is given by the last integral in Eq.
I cos(x + ky) (17) (the jitter average has already been evaluated):
k=O

= cos
[1 + (n - 1)y sin(ny/2)cosec(y/2) (A6) (QN(WP))X±,y = ' df f((p)(QN('P))X1
91 , (Cl)

gives Eq. (10). where the arguments of QN(pi,t, X,ko,V + AV,(p)


other than p are suppressed. The signal charge is
given by Eq. (21), which may be written in the form
Appendix B: Jitter Modulation Transfer Function
The position average [in Eq. (17)] of the signal charge (QN((PX1y1 = A[1 + B cos(C + (p)], (C2)
[Eq. (10)] is
where A, B, and C are products of factors and the
(QN(P1)P1 = (A[1 + B cos(ko *I5 + CP dependence on p is explicit. Inserting Eq. (C2) into
= A[1 + B(cos(ko0 i + koy9 +
Eq. (Cl) and using the uniform phase density func-
tion [Eq. (25)] gives
(B1)
jA9/2 1
where A[1 + B cos(C + 9)],
(QN((P)X1S = J , /2 d'P/
A = Nl1lyTR 0(X)T(X)H0 (X),
(B2) (C3)

B = mMTFop(ko, X)MTFA(ko)MTFsyNc(ko), (B3) which yields

C = -ko (v + AV)tl + p + D(ko,X)+ &. (B4) 2 sin(/2A) 1


(QN(WP)gl, 9= A[1 + B 2(Ap cos(C)]
In Eq. (Bi), k& = ko cos and koy = ko sin 0 are the x
and y components of the vector ko. If the jitter = A[1 + B MTFpH(ko)cos(C)]. (C4)
motions in the x and y directions are independent,
then their joint density function factors into a prod- When A, B, and C are replaced, the result is Eq. (26).

6512 APPLIED OPTICS / Vol. 32, No. 32 / 10 November 1993


Appendix D: Signal Variance where cos2(x) = [1 + cos(2x)]/2 is employed and A is
The variance is given by Eq. (28), i.e., the same as in Eq. (27). When A, B, and C are
replaced in Eq. (D7), the result is Eq. (29).
2= j dxl j dylfx,91 (xfy,) j d'pf9('p) The author thanks Robert J. Chambers, Terrence
S. Lomheim, and Lee W. Schumann for carefully
reviewing the manuscript. This work was motivated
X[QN(PI, t, X,ko, v + AV,'p) - ,0]2. (DI)
by and related to company-sponsored research at the
Q is given by Eq. (10), which may be written as Aerospace Corporation.
QN = A[ + B cos(ko P1 + p + C)]. (D2)
References
In Eq. (D2), parameters other than the random
1. D. F. Barbe, "Time delay and integration image sensors," in
variables are lumped into C, and Solid-State Imaging, P. G. Jespers, F. Van der Wiele, and M. H.
A = N1llyTR0(X)T(X)Ho(X), (D3) White, eds. (Noordhoff, Gr6ningen, The Netherlands, 1976), pp.
659-671.
B = mMTFop(ko, X)MTFA(ko)MTFsyNc(ko). (D4) 2. R. J. Arguello, "Image chain analysis of high resolution, high
The signal mean [Eq. (26)]may then be written as speed CCD film reader system," presented at the Tenth Annual
Modeling and Simulation Conference, School of Engineering,
1Q(QN) P University of Pittsburgh, Pittsburgh, Pa., 25-27 April 1979.
3. D. F. Barbe, "Imaging devices using the charge-coupled con-
= A[1 + B MTFJIT(ko)MTFPH(kO)cos(kO + C)], cept," Proc. IEEE 63, 38-67 (1975).
(D5) 4. M. Born and E. Wolf,Principles of Optics, 5th ed. (Pergamon,
London, 1975), p. 480.
and the variance [Eq. (28)]may be written as 5. J. W. Goodman, Introduction to Fourier Optics (McGraw-Hill,
New York, 1968), Chaps. 5 and 6, pp. 77-140.
Q (QN2 )PO - IQ * (D6)
6. A. Papoulis, Probability, Random Variables, and Stochastic
Use of Eqs. (D2) and (D5) in Eq. (D6) gives Processes (McGraw-Hill, New York, 1965), Chap. 4, p. 83.
7. T. S. Lomheim, L. W. Schumann, R. M. Shima, J. S. Thompson,
go2 = A2 B 2 [1/2 + 1/2 sinc(A'p) and W. F. Woodward, "Electro-optical hardware considerations
2 2
in measuring the imaging capability of scanned time-delay-and-
x exp(-2k 0 cr )cos(2ko Rujp+ 2C) integrate charge-coupled imagers," Opt. Eng. 29, 911-927
(1990).
- MTFPH2(ko)MTFJIT2(ko)cos 2 (ko *tp1 + C)],
8. I. S. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series,
(D7) and Products, 4th ed. (Academic, New York, 1965), p. 29.

10 November 1993 / Vol. 32, No. 32 / APPLIED OPTICS 6513

You might also like