Professional Documents
Culture Documents
Article views: 16
OPTICA ACTA,
1977,
VOL .
24,
NO .
5, 577-590
M. NOVOTNI
Institute of Instrument Technology, Czechoslovak Academy of Sciences,
Kralovopolska 147, 612 64 Brno 12, Czechoslovakia
(Received 19 March 1976)
Abstract. The form of the recently developed series representation of the
axially symmetrical diffraction fields enables us to establish the pupil function
from the series coefficients yk(u) . A new interpretation of the Strehl intensity
ratio is given. It represents the ratio of the energy coming to the image focus
and that coming to the other foci of the filter . Further, by the use of the threeterm approximation, the transmissivity of the filter giving the desired two-point
resolution and attaining the maximum value of the Strehl criterion is found and
compared with the exact solution proposed by Luneberg. Finally, it is shown
that the Luneberg filter transmissivity represents a common approximative solution of the apodization problems that utilize the integral criteria of apodization .
1.
Introduction
Since the diffraction field G(u, v)-cf. [3, 7],
1
(1)
G(u, v) =
(-1)kyk(u)/k(v),
( 2)
k=0
y k (u),
which have
1
y k(u) =
(3)
We may then suppose that the pupil function g(t) can be expressed from the set
of the diffraction coefficients {y k (u)}, k = 0, 1, . . . . The inverse transformation
of the diffraction coefficients yk (u) to the original diffractorg(t) will be derived in
2 by replacing the functions Qk(t), which represent the renormed Bernoulli
polynomials ([3], equation (24)), with the orthogonal Legendre polynomials .
578
M . Novotny
df
dv(v) =
d k(v)-
(4)
From [3], equation (17), it follows that their asymptotic behaviour is characterized
by the expression (k = 1, 2, . . .)
~,
d k(V) 2t
v 2k-2
for v-_> oo .
Thus, the/ k (v) are unlimited and cannot then have any Hankel transforms which
are suitable when calculating integral intensity characteristics-cf. [13], p . 65 .
This disadvantage can be overcome by using the Parseval theorem-cf . 4
and 5 .
We illustrate the present approach of solving inverse problems in terms of
diffraction coefficients y k (u) by using as an example one of the Luneberg problems .
In 3 the formulation of this problem of finding the filter transmissivity with the
desired value of two-point resolution and the maximum Strehl criterion is discussed and modified. The exact solution 1(t) is found to be a common approximative solution of three apodization problems ( 4) . Section 5 presents the
solution of the problem discussed in 3 in terms of the diffraction coefficients .
2. The pupil function expansion in terms of the diffraction coefficients
For simplicity let us set at first u =0 and write briefly y k (0)- yk . In agreement with (3) the coefficients represent certain scalar products of the pupil
function g(t) and the Bernoulli polynomials Q k (t), i .e . yk = (g, Q k ) . If the set
of polynomials {Q k (t)} formed an orthogonal basis, the pupil function g(t) could
be simply expressed by means of yk . Unfortunately, the Bernoulli polynomials
579
are not orthogonal and we must then use another polynomial basis which is
orthogonal with respect to the considered scalar product .
The orthogonal basis of the space of the functions, which are defined in the
interval <0, 1> and satisfy Dirichlet's conditions, is formed by the set of the
Legendre polynomials p 8 (x) transformed and renormed in an appropriate way
s!
p8(2 t
P8(t) _ (2s)i
-1) ;
s = 0, 1, . . . .
(5)
([4],
p . 267)
1
d8(x2 -1)s
p8(x)= 23 '
s!
dx8
(Pr, Ps) =
f0 Pr(t)Ps(t)dt=wa
s!
WS=
bra,
( 6)
1
(7)
[(2s)!] 2s+1'
(8)
P3 (t) = t3 /6 - t2 /4 + t/ 10 - 1/120 .
Comparing with the polynomials Q8 (t) ([3], equations (25)) we see that the first
three polynomials P8(t) and Q 8 (t) coincide .
Let us study the transformation matrices {(X,r} and {p s r} relating the sets
{P8 (t)} and {Q,.(t)} :
~r
Qr(t)
L. am'Pm( X ),
M=0
P8(t) _
Nn sQn(t) .
n=0
(9 )
(10)
Q samn =3 sm'
Nn
(11)
8
n=m
s>,m .
580
M. Novotny
Thus, knowing, e .g ., matrix {a,,, r } the relations (10) and (11) enable us to
calculate successively the inverse matrix elements 3 fly Nnn+l, . . ., /3 r or respectively, fs8,
8
. .
8S
Any element a,,,r may be evaluated by means of the relation
amr=Wm -1 (Qr,
PM)
following from (9) and (6) . The normalization of the polynomials (5) leads to
the unit values of all diagonal elements of both the matrices, i .e .
r =1,
Prr =1 ;
a r
r=0, 1, 2, . . . .
(13)
r (t)dt=0,
r=1, 2, . . .,
(14)
following from their recurrence property ([3], (24)), the relation (12) gives for
m=0
a0r
=0 r=1,2, . . . .
(15)
Thanks to certain symmetry properties common to the polynomials Qr (t) and
Pr (t) of the same degree r, further elements amr must be zero . From the recurrence relation ([3], (24)) and Rodrigue's formula it follows that both the polynomials Q r(t) and Pr (t) are odd (even) with regard to the axis t= z of the absicissa
<0, 1) if the degree r is odd (even) . Then, because of (14), the scalar products
(12) are zero whenever Qr (t) and Pm (t) show different symmetry . Therefore,
a
mr=0
if r+m=1, 3, . . . .
(16)
if r+m=1, 3, . . . .
(17)
This property facilitates the calculation of the inverse matrix {N8r} from (10) or
(11). Let us write out the elements o4 Br and /3,r of the lowest indexes
1
a0 0
al
a22
a33
a23
a0
1 0
a1 2
aO2
0613
ao 3
1 0 0
1 0
6O
581
Similarly,
{ Rsr} =
1 0
so
Knowing the inverse matrix {p 8r} it is easy to derive the unknown function
g(t) from the diffraction coefficients y o , yl , . . . . Let us expand the function g(t)
in the orthogonal basis (5)
r
g(t)=
8 P8 (t),
(20)
8=0
1
r8=w,
f g(t)P,,(t)dt=w,-1(g, P 8 ) .
(21)
Inserting the expression (9) for the polynomial P8 (t) into (21) and using (3) with
u = 0, we get P 8 in terms of the diffraction coefficients
8
80
re=8
-1
g,
I Nns Qn
n=0
C8
- Ws -1 G Nn s ynn=0
(22)
Hence, the expansion (20) together with (22) enables us to evaluate the pupil
function g(t) from its diffraction coefficients y n .
By the relation (22) the set of the first N diffraction coefficients yn determines
the first N coefficients 17 0 , F1 , . . . , P N _ 1 of the expansion (20) . Therefore, restricting ourselves to solving a problem to the first N diffraction coefficients {y k },
k = 0, 1, . . . , N - 1, the corresponding N-terms pupil function expansion in the set
of the Legendre polynomials (20) represents the best least square approximation .
If we want to establish the pupil function g(t) from its diffraction coefficients
yn (u) associated with any Fresnel diffraction pattern (u # 0), we can use the same
relations (20), (21) and (22), replacing however the polynomials P,(t) and Qn (t)
by the functions Pe (t) exp (- iut/2) and Qn (t) exp (- iut/2) . The invariant form
of (21) and (22) is a consequence of the fact that the new basis {P8 (t) exp (- iut/2)},
s = 0, 1, . . . satisfies the former orthogonality relation (6) and that the transformation matrices in (9) remain unchanged .
3.
2Q
582
M . Novotny
str [g(t)]
= I f g(t) dtI 2/ .j
j
jg(t) 2 dt
(23)
introduced originally by Strehl [6] . The definition of the Strehl intensity ratio
in the form (23) permits an interpretation in energy terms . Using the Fourier
coefficientsg m ([3], equations (4), (7)) we get from (23)
str Ig(t)] = IgoI 2'E,,
(24)
and further
str [g(t)] =
1
jIgml2)/Eg=1o
1 1ggl 2 .
( 25)
Eg moo
( Eg
Thus, according (24), the Strehl criterion represents the ratio of the amount of
energy coming to the imaging focus (m=0) to the total energy Eg transmitted
through the filter-cf. [7], 6 . It follows from (25) that strg 51 . The equality
is valid for the uniform transmissivity g(t) = 1 when g o = 1 and the gm = 0, i .e . the
total incident energy is concentrated in the imaging focus . In the case of nonuniform pupil functions the sum 11 g. 1 2, m =A 0 represents the part of the transmitted energy coming to the further foci of the filter . This energy intensifies
the fringes in the focal diffraction pattern due to a super-resolving pupil if compared
with the normal Airy pattern . Therefore, keeping the value of strg as large as
possible we attain the minimum of the total energy in the fringe system .
The super-resolving pupil function that maximizes the Strehl criterion has
been found by Luneberg [8] in the form
1(t) = a(Av) - b(Av)J 0 (Ovt112 ),
(26)
where Ov is the desired two-point resolution and a, b are constants following from
a normalization condition of the fixed transmitted energy through the filter . In
fact, the Luneberg solution (26) maximizes the Strehl criterion (24) so that it
gives the maximum of energy into imaging focus I go l 2 while the transmitted energy
Eg is fixed-cf . [9, 10] . However, the optimum pupil function (26) absorbs
the incident energy in dependence on the required value of Av . Therefore, the
condition of fixed transmitted energy Eg = const has led to the solution g(t) of the
Luneberg problem which gives jg(t)I > 1 . This is, however, unacceptable for a
passive optical system with a constant value of incident light energy . The formal
normalization used in [9] to attain jg(t)I 5 1 interferes with the original requirement
E g = const.
We propose an alternative formulation of the problem discussed . Instead of
fixed transmitted energy Eg = const, let us require the fixed value of the focal
diffraction amplitude, i .e .
1
(27)
583
The desired shape of the central maximum is prescribed in the following way
1
G(v=0v)=
(28)
1
Eg
= f
jg(t)j 2 dt=min.
(29)
Such a function g(t) simultaneously maximizes the Strehl criterion (24) under
the condition (27) . The sense of the proposed reformulation of the original
Luneberg problem consists mainly in that some apodization problems can then be
transferred to our reformulated problem-cf . 4. Thus, the solution of our
problem is also applicable for apodization purposes . Naturally, having found
the form of the solutiong(t), its final normalization to get lg(t)l < 1 is again necessary.
Using the standard method of variation calculus the solution of SE, under the
conditions (27), (28) may be derived in the form (26) . The dependence of the
parameters a, b on AV is rather different from that in the Luneberg solution-cf .
[9] . From (28) and (27) we get
451(AV)
(AV)
- 11 ( )1(
- [01(Ov)]G2
~ b(w)
II (AV) 1( ~A
[ 1(
'
(30)
)]2
where
1
H(Ov)=
[Jo(Ovt1"2)]2dt=[J0(w)]2+[Ji(w)]2
(31)
A.(Ov)=n!(2
J.(v),
n=0, 1, . . . .
(32)
G(v) =
l(t)J0 (vt1"2 ) dt
(33)
Ei =
1 /E1
because of (27) .
(34)
584
M. Novotnj
Ig1 (t)I 2 dt
Ig2(t)I 2 dt
unless g 1 (t) and g2 (t) are proportional . Applying the Schwarz inequality to
g1(t)=k(t), g 2 (t)=1(t) we have
1
I f
(35)
if k(t) is not proportional to 1(t) . We evaluate the left-hand side of (35) . Assuming the validity of the conditions (27) and (28) for k(t) we have subsequently
1
f
(36)
Thus, for any function k(t) satisfying the conditions (27), (28) the left-hand
side of inequality (35) is equal to E12 . Consequently, we conclude that
El < Ek
if k(t) =Al(t) .
Under the conditions (27), (28) the functional E,, has then an absolute minimum
at the function (26) .
4.
I I G(v)I
E(vm )
2 dv2
vm
= f I G(v)I
_
f IG(v)I
2dv 2
2 dv 2/E g ,
(37)
585
where we have used the Parseval theorem for the Hankel transformation (1)
between the pupil function g(t) and the associated diffraction pattern G(v), i .e .
1
w
f
IG(v)l 2vdv .
(38)
While the encircled energy represents the relative part of the transmitted energy
Efl which is inside the circle v S v m, the spreading factor characterizes the luminous
energy diffracted outside this circle
w
~(vm)=
IG(0)I2
IG(v)1 2 dv2 .
Vm
Assuming further the validity of the condition (27), i .e . G(0)=1, we can write
w
.) =
.(v
f I
V
I'm
IG(v)I2dv2 .
( 39)
I I G(v)I
D(vm , v
2 dv2
vm
M )=
vm
f I G(v)I
(40)
2 dv2
D(vm oo) =
Um
I G(v)I
2 dv 2
The enumerated integral criteria have a common feature : providing the energy in
the circle v S v Y, is constant, i .e .
Vm
I G(v)I
2 dv2 = const,
(42)
586
M. Novotny
SE(Vm )= -SEg ~2
l G(v)I 2 dv2 ,
S -T(vm ) = Mg ,
(43)
As regards the condition (42), it is practically fulfilled for the pupil functions
displaying the same width of the central intensity maximum Ov, i.e . satisfying
(28) . Then, the central intensity maxima have the same basic sizes . As the
equation SE g =0 has the unique solution (26), under the conditions (27) and
(28) are satisfied, the functionals (37), (39) and (41) attain their extremum values
for the unique function g(t) = 1(t) assuming the validity of (43) . Thus, the
original super-resolving pupil 1(t) may be useful even for apodization purposes .
5.
The problem of variation presented in 3 may be solved in terms of the diffraction coefficients y,, . The knowledge of the exact solution (24) enables us to
evaluate the convergence of the pupil function expansion in terms of yn .
Let us rewrite the requirements (27), (28) and (29) in terms of y n . The first
condition (27) means that y o = 1-cf. (2) . According to (2) we can write, instead
of (28),
00
G - fo(w)= I (-1)kYk/k(Ov)
(44)
k=1
Eg=
'0
I
8=0
ws l F812= 1+
~1Ij
rn8Ynl2 .
(45)
8=1 Ws n=1
Our aim is to find the minimum of (45) under the condition (44) . Restricting
ourselves to a finite number of the diffraction coefficients yn, n = 1, 2, . . . N and
introducing one Lagrange multiplicator A in the form
F(-y1, Y2, . . . YN, A) = Eg + 2A [G(Ov) - G]
we transfer our problem to finding an extreme of the function F(y1, Y2, . . . y V , A) .
Differentiating F with regard to all variables we obtain N+ 1 linear equations for
N+ 1 unknown values of y1, Y2, . . . YN, A of the extremum (m=1, 2, . . . N)
L
n =1
w
n=1
g
N am8 n8 )Yn+A( -1 )"1m(w)= 0 ,
ws
a= 1
(-1)1/r
n( v)Yn=G-/0(Ov) .
(46
587
1 + 12y1 2 + 720Y 2 2 .
( 47)
Similarly, the relation (44) in the same approximation turns into (G=0 2)
0 .2'=''/o(Ov)
- Yl/l(Ov)
+Y2/2(Av) .
(48)
Y2"
1 4 15 TrY2
7
(2Tr)2 Y2 1' 5 = 1/
Yo r
(50 )
(2
2 )2 (Y1 2 + 72 12 ) .
( 51 )
The figure presents these circles of constant transmitted energy in the plane
of yr', Y2". At the same time, they represent the curves of constant value
of the Strehl criterion for the filters with the three-term expansion (20) as
str g(t) =1 /E g . The straight lines Ov = const are constructed using the relation
(48) and the transformations (49) and (50) . They define the points (v2, Y2")
giving a fixed value of the quantity Ov-cf . [11 ] . If we want to achieve a certain
value of the two-point resolution, keeping the value of the Strehl criterion as
large as possible, we must obviously advance along the curve L(y2 , Y2")=0
formed by the points in which the straight lines are tangent to the circles .
The curve L(y1', Y2")=0 defines the desired diffraction coefficients needed for
the three-term expansion (20) of the sought pupil function 13 (t) . With the help
of (22) we get
13 (t) =1 +
1
y
P1(t)
+
V5Y2
1
P2(t) .
( 52 )
The form of 13 (t) is presented in the inset of the figure for Ov = 2 . 5, where an
appropriate normalization is done to attain 1 3 (1) = 1 . The straight line corresponds
to the two-term approximation 12 (t) determined by v2' = 0 and y,' such that
Ov(y1 ', 0) = 2 5 . From the inset of the figure we see that str 1 2 < str 13 but the
differences are slight except for the largest values of Ov,>4 . The differences
588
M. Novotny
2
f
12
Figure 1 . (a) The characteristics Ov (the straight lines) and strg (the circles) of the filter
transmissivity g(t) in the three-term approximation . The small circle points on
the curve L(y1 ', y2 ") correspond to the Luneberg filter l(t)=a-bJo (Ovt'l2 ) giving
the minimum Strehl criterion for arbitrary value of Av . (b) The filter transmissivities l,,(t) in the n-term approximation which have the minimum value of the
Strehl criterion for the value Ov=2 . 5 . The small circle points correspond to the
exact function 1(t) .
between str 1, and str Z are negligible until AV >, 3 . 8 . Therefore l3 (t) practically
coincides with the exact course of 1(t) calculated from (26)-the circle points in
the inset.
A sensible test offers a comparison of the diffraction coefficients stated
for the exact function l(t) and those belonging to 1 3 (t), i .e . defined by the curve
L(y l ', y2 ")=0 in the figure (a) . Let us derive the needed expressions .
Using the recurrence properties of the Bessel functions Jm+1 (v)=
-vmd[v -mJm (v)]/dv and of the Bernoulli polynomials Q k _1(t)=dQ k (t)/dt,
cf. ([3], equation (32)), we can perform the integration of (3) where we set
g(t) =a-bJo (Ovtl/2 ) and u = 0 :
I
Yk = (l,Qk) =
For
k=
Yk_1(AV)=(-1)kb{m=o (k
2E
) v) -
(53)
Thus,
Y1(Ov)= 2{-A1(Ov)+A2(Ov)},
Y2(AV)
= 1
589
(54)
(55)
By means of the transformations (49) and (50) we construct the points yi '(w)
and y2"(Ov), the circle points in the figure (a) . The values of Ov were chosen
the same as for the straight lines of constant resolution in (a) . We find an excellent coincidence of the tangent points of the curve L(yl ', Y 2 ") and the points
constructed with the help of (54) and (55) for all values of Av < 3 .4 . If iv increases
above the value 3 . 4 the three-term approximation gives the values of the Strehl
criterion of l(t) with increasing error : the relative inaccuracy of str l(t) for /Xv = 4 .0
is about 4 per cent .
It seems that the function 1(t), for which it has been found to have an apodizing
property in the previous section, could surpass the known apodizers . The
triangle points in the figure (a) are constructed for the apodizers published by
different authors and summarized in [13], pp . 97 and 99 . Their pupil functions
are in our notation of the form
3
cn (1 -t)n
T(t)=
(56)
n=0
n 1+n'
- ncn
y'
n 2(n+l)(n+2)'
(57)
y2 - I
c n [12(n+1) 2(n+2) (n+3)]'
n
Using (57) and the transformations (49), (50) we compute the transformed diffraction coefficients yi , Y 2 " due to the pupil functions (56) which apodize over
different intervals according to either the minimum dispersion factor (41)-the
five triangle points L in figure (a), or the minimum spreading factor (39)-the
remaining two points p . All the points lie under the curve L(yl ', y2 ") corresponding to the maximum Strehl criterion attained by l(t) =a-bJ (Avtl , 2 ) for
the variable value of Av . Since, according to 3, the pupil function l(t) represents the unique solution of the problem of variation discussed in 3, we may expect
with regard to 4 a somewhat stronger total apodization by the Luneberg filter
l(t) than in the case of the compared four-term apodizers (56) . A more detailed
numerical investigation is however necessary for doing some quantitative conclusions .
590
The author is indebted to Mrs . Z . Kucerova for her assistance in the preparation of the manuscript and to Dr . J . Komrska for critical reading of the paper .
La forme due developpement en serie recemment obtenu pour la representation des
figures de diffraction a symetrie axiale nous permet de deduire la fonction pupillaire a
partir des coefficients y k(u) de la serie . On donne une nouvelle interpretation du rapport
des intensites de Strehl . Il represente le rapport entre 1'energie arrivant au foyer image
et celle arrivant aux autres foyers du filtre . De plus, en utilisant l'approximation a trois
termes, on trouve la transmissivite du filtre donnant la resolution desiree pour deux points
et atteignant la valeur maximum du critere de Strehl . Cette transmissivite est comparee
a la solution exacte proposee par Luneberg . Finalement, on montre que la transmissivite
du filtre de Luneberg represente une solution approchee des problemes d'apodisation
qui utilisent les criteres integraux d'apodisation .
Die Form der kurzlich entwickelten Reihendarstellung achsialsymmetrischer Beugungsfelder erlaubt uns die Darstellung der Pupillenfunktion aus den Reihenkoeffizienten
yk(u) . Es wird eine neue Deutung der Strehlschen Definitionshelligkeit gegeben . Diese
stellt das Verhaltnis der Energie im Brennpunkt der Bildebene zur Energie in den anderen
Brennpunkten des Filters dar . Weiterhin wird unter Benutzung der Naherung mit drei
Termen die Transmission des Filters berechnet, das die gewUnschte Doppelpunkt-Auflosung
1 iefert and ein Maximum des Strehl-Krituerims erreicht and die Ergebnisse nut der exakten
Losung nach Luneberg verglichen . SchlieBlich wird gezeigt, daB das Luneberg-Filter
eine allgemeine Naherungslosung der Apodisationsprobleme darstellt, die Integralkriterien der Apodisation benutzen .
REFERENCES
MARCHAND, E . W., and WOLF, E ., 1966, J. opt . Soc. Am ., 56, 1712 .
SOMMERFELD, A ., 1964, Optics (New York : Academic Press) .
NOVOTNY, M ., 1977, Optica Acta, 24, 551 .
LUKE, Y. L ., 1969, The Special Functions and their Approximations, Vol. 1 (New York :
Academic Press) .
WILKINS, J . E ., 1963, J. opt. Soc . Am ., 53, 420 .
STREHL, K ., 1895, Z. InstrumKde ., 15, 364 .
NovoTNY, M ., 1973, Optica Acta, 20, 217 .
LuNEBERG, R. K., 1964, Mathematical Theory of Optics (University of California
Press) .
BARRAKAT, R ., 1962, J. opt . Soc. Am ., 52, 264 .
WILKINS, J . E., 1963, J. opt. Soc . Am ., 53, 420 .
NOVOTNf, M ., 1977, Optica Acta, 24, 567.
JAHNKE, E ., EMDE, F ., and L6SCH, F., 1960, Tafeln hoherer Funktionen (Stuttgart :
Teubner Verlagsgesselschaft) .
JACQUINOT, P ., and ROIZEN-DOSSIER, B ., 1964, Apodisation in Progress in Optics,
Vol . III, edited by E . Wolf (Amsterdam : North-Holland Publishing Co.) .
LANSRAUX, G ., and BOIVIN, G ., 1961, Can . J. Phys ., 39,158.
BoIVIN, A ., 1964, Theorie et Calcul des Figures de Diffraction de Revolution (Paris :
Gauthier-Villars) .