Anda di halaman 1dari 11

JOURNAL OF THE OPTICAL SOCIETY OF AMERICA VOLUME 55, NUMBER 3 MARCH 1965

Coupling of Optical Fibers and Scattering in Fibers*


ALAN L. JONES
International Business Machtines Corporation, Endicott, New York 13764
(Revision received 22 July 1964)

The phenomena of optical coupling of parallel fibers and scattering of light from a fiber due to rough walls
are considered from a mode point of view. With the use of a Green's function, the problems are cast in the
form of integral equations. Coupled ordinary differential equations are obtained which are used to study the
coupling of modes in parallel fibers, including the case when the diameters are slowly-varying functions of the
axial distance. The analysis of the problem of propagation in an optical fiber having rough walls shows that
the various modes in a fiber will couple and that the roughness will cause radiation through the walls of the
fiber. The character of the radiation may be determined if the spatial spectral density of the surface rough-
ness is known.

INTRODUCTION relating the unknown amplitude factors are obtained.


WHENthe diameter of an optical fiber is large These relations are still in the form of integral equa-
tions; however, they involve only the unknown scalar
compared to the wavelength of light, the geo-
metric theory of optics has been sufficient to study most amplitude factors whereas the original equations are
phenomena observed.' 2 However, when the diameter is vector integral equations in several independent
of the order of a wavelength, light is guided in wave- variables. In some cases, the integral equations involv-
guide modes just as microwaves are guided by a dielec- ing the amplitude factors can be transformed into an
tric waveguide. These modes have been photographed approximate set of coupled first-order differential
and their characteristics have been studied.'- 5 But equations. These in turn yield solutions which can be
little has been done beyond finding the modes that exist shown to be very accurate for a wide range of the
in an infinitely long, circular dielectric cylinder and variables involved.
some of the properties of these modes. Snitzer 5 has To consider the coupling of arrays of parallel optical
reported an approximate solution for the coupling of fibers, the problem of two coupled fibers is first treated.
two parallel optical fibers. Bracey et al.,7 in their studies A set of coupled first-order differential equations relat-
of microwave propagation along dielectric antennas ing the amplitudes of the various modes in both fibers
have also obtained an approximate soultion for this is obtained by using the method outlined above. The
problem. solution of these equations shows that the coupling of
the fibers will result in a complete oscillation of energy
This paper presents a method for the approximate
solution of a class of problems in the theory of wave back and forth if the fibers have eigenvalues which are
propagation in optical fibers and examines, in some identical. If the eigenvalues are slightly different and
detail, the solutions in two particular cases. The the fibers are relatively far apart, there will be little
problems considered are (1) propagation in coupled transfer of energy. The differential equations can also
parallel optical fibers, and (2) propagation in a single be extended to consider arrays of parallel fibers. These
optical fiber whose walls are slightly rough. cases are difficult to solve, in general, but the special
The method of solution consists of casting the case of an infinite line of identical fibers yields a solu-
problem in the form of an integral equation by using a tion for the amplitude factors in terms of Bessel
dyadic Green's function. Solutions of the integral functions.
equation are assumed to be expressible in the form of a The coupled differential equations can also be used
sum over all of the possible modes which can exist in the to study the coupling of parallel fibers whose diameters
fiber. The amplitudes of these modes are taken as vary slowly as a function of the axial distance. For
unknown functions of the axial coordinates. The as- example, a linear variation in diameter yields a solution
sumed form of the solution is substituted into the in terms of parabolic cylinder functions. This solution
integral equation and by equating coefficients, equations again shows that there is little transfer of energy
except where the eigenvalues are about equal. The
* The work in this paper is taken from a thesis by the author solution of the linear case yields information about the
presented to Purdue University in partial fulfillment of the require- coupling in the case of a more general variation.
ments for the degree of Doctor of Philosophy. The other problem, propagation of light in an optical
1 N. S. Kapany, J. Opt. Soc. Am. 47, 413 (1957).
2 R. J. Potter, J. Opt. Soc. Am. 51, 1079 (1961). fiber whose walls are slightly rough, is considered to
N. S. Kapany and J. J. Burke, J. Opt. Soc. Am. 51, 1067 determine if the roughness of the wall can cause the
(1961).
4 E. Snitzer, J. Opt. Soc. Am. 51, 491 (1961).
large attenuation observed in small optical fibers.2 It is
6
E. Snitzer and H. Osterberg, J. Opt. Soc. Am. 51, 499 (1961). found from the analysis that a small amount of rough-
6 E. Snitzer, Advances in Quantum Electronics (Columbia ness can produce rather large attenuation, especially if
University Press, New York, 1961).
7 M. F. Bracey, A. L. Cullen, E. F. T. Gillespie, and J. A. Stani- the mode under consideration is close to cutoff. If the
forth, IRE Trans. Antennas Propagation AP-7, S219 (1959). profile of the surface is taken to be a sinusoidal function
261
262 2ALAN L. JONES Vol. 55

Specializing to the case of a circular cylinder with


the usual cylindrical coordinates r and 0, we find that
the Helmholtz equations can be satisfied by taking
E: and Hz proportional to Bessel functions multiplied
by a factor of eill where n=0, 11, - - -. In the core, the
Bessel function Jn[En(r/a)] is chosen and in the cladding
the modified Hankel function KE[v(rla)] is chosen.
The radius of the cylinder is a. These are required to
satisfy the condition that the field be finite at the axis
and that it decay properly as r becomes large. Substitut-
ing the solutions into the boundary conditions, we
obtain a set of four linear, homogeneous equations for
the determination of four constants. Setting the de-
FIG. 1. Propagation constant vs radius for various modes, terminant of this system equal to zero yields the
nl= 1.8, 12 = 1.5. following dispersion relation':
gn('y)= [-4(-l)n/a'Tr][(vJnKn+UJnKnt)
of distance along the fiber, energy radiates at a charac- X (vk 12 Jn'Kn+uk 2
JnKn') - n
2
'y2J2 2 K. 2 R4 ] = O,
2
teristic angle dependent on the spatial frequency of the
(1)
profile.
The case of random roughness is also considered, and where the prime denotes differentiation with respect to
the expected angular energy distribution of the scat- the argument. The argument of Jn is u and that of Kn
tered radiation can be determined if the spatial spectral is v. The quantities u, v, kj, and R are given by
density of the random surface profile is known.
u= a(k2 - y2 )t, v= a(y2 -k 2 2 )`, R 2 = u2 +V2 , k 32 =W2Mej.
PROPAGATION OF LIGHT IN AN INFINITE Equation (1), when solved for oy, yields the possible
DIELECTRIC CYLINDER
propagation constants-ynm for the cylinder.
To determine the characteristics of light propagation A program has been written for an IBM 7094 com-
in a dielectric cylinder, Maxwell's equations must be puter to compute the roots of Eq. (1) given the indices
solved in the core and the cladding, subject to the of refraction of the core and cladding and the radius-to-
boundary conditions that the tangential components of wavelength ratio. Time required for the computer to
the electric and magnetic field vectors E and H must be find one root is approximately 0.5 sec. The indices of
continuous across the boundary. We assume that the refraction of the two media are related to the quantities
core and cladding materials are nonconducting, iso- kj by
tropic, homogeneous, and linear and that their magnetic nj= k/ko,
permeabilities ,u are equal. The dielectric constants of where
the core and cladding are denoted by e, and e2, respec- ko2 =w2 eo= (27r/X.)2,
tively. A coordinate system is chosen with a z axis in
the direction of the fiber. Assuming a time dependence and eois the dielectric constant for free space and Xois
in the form e-iwt and a z dependence in the form eiyz,we the wavelength of the light under consideration in free
can show that the axial component of E and H must space. With the notation of Snitzer,4 the variation of
satisfy a Helmholtz equation in the transverse co- 'y with radius is shown for the lower-order modes in
ordinates. If these Helmholtz equations are solved for Fig. 1. In this notation the axially symmetric modes are
E, and Hz, the complete field can then be written in designated by TEom (transverse electric) and TMom
terms of these quantities.8 Applying the boundary (transverse magnetic). Hybrid modes are denoted by
conditions, we find that y can take on only certain HEnm and EHnm. (it designates the orders of the Bessel
values. These values, the eigenvalues of the problem, functions that occur in the solution and m indicates
will be designated by ynm. For each ''nm there will be which root is found for each value of n.) As noted, the
associated E and H fields which are the vector eigen- index i can take on the values 0, z4l, i2, *- -. When
functions for the rth mode. Goubau9 has shown that i is a positive integer, it denotes a mode which spirals
the eigenfunctions for a dielectric cylinder of arbitrary clockwise as it progresses along the guide while a nega-
cross section are orthogonal. The orthogonality prop- tive value of X denotes a counter-clockwise spiralling
erty is useful in finding what modes exist at some mode. These spiralling modes are akin to the skew
distance along a fiber if the transverse E or H field at rays in the geometric theory, which follow roughly
one cross section is known. helical paths. A mode with positive n can be combined
with a mode with negative n to form the modes dis-
8 J. A. Stratton, Electromagnetic Theory (McGraw-Hill Book
Company, Inc., New York, 1941). cussed by Snitzer4 that are characterized by sines and
9 G. Goubau, Proc. IRE 40, 865, (1952). cosines in the 0 variable.
March 1965 COUPLING AND SCATTERING IN OPTICAL FIBERS 263

In Fig. 1, note that regardless of how small the guide,


there is always at least one mode which may propagate.
This is the HE11 mode with which electrical engineers
are familiar in studies of dielectric antennas. Notice
that the HE21 and the TMo1 modes have identical
propagation constants at one point. This is of impor-
tance when considering the fiber as a waveguide, since
these two modes will not propagate independently, but
will be strongly coupled in the presence of small
irregularities. Miller10 has shown this in the case of
ordinary waveguides. FIG.3. Geometry of two-fiber problem.
In Fig. 1, the points where the line for any mode
intersects the a/Xo axis indicate the cutoff value for this G. Each term in the integrand has the quantity gn(,y)
mode. The values of u near and far from cutoff are given in the denominator. Using complex integration and
by Snitzer.4 In a geometric-optics sense, "near cutoff" recognizing that the poles of the integrand are zeros
means that the characteristic "rays" of the mode are of gn(-y), we can obtain a representation for G in
close to the critical angle. "Far from cutoff" means that terms of a sum over the propagating modes plus an
the rays are almost parallel to the axis of the fiber. integral which represents the radiated field. The
integral arises from the integration along a branch cut.
SURFACE DYADIC GREEN'S FUNCTION The path of integration in the -y plane is shown in Fig. 2.
The previous section was concerned with the propa- An example of this type of integration is shown by
gation of electromagnetic waves in an infinitely long, Tamir and Oliner1' in their study of waves in a plasma
isolated dielectric cylinder with no regard for how the layer.
waves were initiated. Consider now the effect of a
COUPLING OF PARALLEL FIBERS
point current on the wall of the cylinder as the initiating
source of the waves. The resulting field may be inter- We consider the propagation of electromagnetic
preted as a surface dyadic Green's function which is energy along two parallel, dielectric, circular cylinders.
useful for casting problems into integral equations. In Figure 3 shows the geometry of the problem in which
the theory of scalar waves the Green's function is a the fibers are assumed to extend to infinity in each
scalar quantity. It represents the field produced by a direction. The dielectric constant of the interior of the
scalar point source. When considering electromagnetic fibers will be designated as El while that of the sur-
waves we are concerned with vector fields and vector rounding space will be e2. It is necessary to solve
sources in the form of currents. Therefore, it is necessary Maxwell's equations in- the three regions subject to
to find a quantity which, when multiplied by the source boundary conditions at the surfaces of the cylinders.
and integrated over the volume which encloses the By choosing a time dependence of the form e-id{,
source, will yield a vector quantity. A dyadic is such Maxwell's equations can be converted to the vector
a quantity. Helmholtz equation
We define a dyadic Green's function G which repre-
V2E+k? 2
E= O. (2)
sents the effect of a point current on the wall of an
infinite fiber at the point 6==o, z=zo. This Green's Besides satisfying Eq. (2), the divergence condition
function can be determined analytically by taking the must be satisfied, along with the boundary conditions
solution as a Fourier transform on the z variable and a that the tangential components of E and H must be
Fourier series in the 0 variable. Applying the boundary continuous across the boundaries of the fibers.
conditions, we obtain an integral representation for If the fibers are very far apart, we would expect the
fields associated with each fiber to be independent.
However, when they are close together the field in one
fiber affects the field in the other. This interdependence
can be expressed mathematically in terms of integral
equations. The integral equations are obtained through
the use of the dyadic Green's function, the boundary
conditions, and the vector Green's theorem.
n, xEl s -M

+J [(n xE 2 )*(V x G,)+icA(n 2 xH 2)*Gl]dS 2 , (3)


FIG. 2. Path of integration when z>z0.
T2

10 S. E. Miller, Bell System Tech. J. 33, 661 (1954). llT. Tamir and A. A. Oliner, Proc. IEEE 51, 317 (1963).
264 ALAN L. JONES Vol. 55

n2 xE2 S2=-M2 integral equations and equating coefficients we obtain


a set of coupled integral equations in the unknown
+y E(n xE,) *(v xG2)+ico/u(n x H1) *G2]dS,. (4) functions unm(±+)(z).These integral equations can be
converted to an equivalent set of coupled first-order
differential equations by differentiation
In Eqs. (3) and (4), the quantities Ej and H3 denote the
values of these fields at the boundaries of fiber j; Mi is a dutlm 1+)(z)
source term which represents the field due to the condi- iynmlunml(+) (Z)+i Z [Iklm2(+)UkZ2( (Z)
tions which prevail on a cross section at z= - -. Eqs. dz k,1

(3) and (4) represent two integral equations for the +,3kl,,.2( )UI12( ) (Z)], (7)
four quantities nj x E3 and nj x Hj(j= 1,2). Two addi-
tional integral equations may be obtained by using a
magnetic Green's function instead of the electric one d=-i()- ynmiUnmi )(Z)-i Z E83kl.nm2( )Ukl2()(z)
as was done here. A magnetic Green's function repre- dz k,l

sents the field produced by a point magnetic current +f0klnm2(+)Ukl2( ) (z)], (8)
on the boundary instead of an electric current as in the
case of the electric Green's function. with a similar set of equations relating the derivatives
A solution is obtained in the form of a sum over all of U.nm 2 (±) to the amplitudes of the modes in fiber 1.
of the modes of each fiber. Since the coupling of the The constants f3kl.,j(+) are coupling constants which
various propagating modes is being considered, the are obtained from the representation for the Green's
contribution to the field caused by the continuous function by integrating the field of a mode in one guide
spectrum is neglected. We show later that this is a over the surface of the other guide.
good assumption except when a mode is close to cutoff. Equations (7) and (8) describe the coupling of each
Each vector eigenfunction in the assumed solution is mode in one fiber with all the modes in the other fiber.
multiplied by an unknown function of z. The same The coupling constants l3k.nmj(±) give a measure of the
unknown amplitude functions must be used in the magnitude of the coupling. When the fibers are far
expansion of the E field as for the H field. Because of apart, the coupling constants are small and in the
this, the integral equations obtained using the mag- limit, as the distance between the fibers approaches
netic Green's function are redundant and need no infinity, the equations uncouple; the solution is:
longer be treated. Thus, the two integral equations
(3) and (4) will be used for the determination of the Unj (±) (z) =A nmj( )exp4iYnm jZ. (9)
two sets of unknown amplitude functions.
The E and H fields at the boundary of the jth fiber It can be shown by examining the coupled differential
are expressed as equations that the coupling will not be pronounced
except when the propagation constant of a mode in one
Ej(aj,Oj,z) fiber is close to the value of a propagation constant in
= u (+)(z)E.j(aj,Oj jy.nm/j)nmj the other fiber. In addition, a forward-traveling mode
nm in one guide will be coupled weakly to a backward wave
in the other guide. Therefore, it is only necessary to
consider coupling of forward modes which have approxi-
Hi (aj,oj,z) mately the same propagation constants. This is a com-
mon simplification in the theory of coupled waves.10
Unmj(+)(z)HEmj(ajOj; ynmj)/Nl.nmj The continuous spectrum can also be neglected in the
n,m analysis, because it is characterized by a propagation
+ , n7mj(- (Z)H,,mj(aiOj;Ynmj)/Nvnm, 6
parameter which is a variable but never has a value near
that of the propagating modes. The only exception is
the case when a propagating mode is near cutoff.
where Enmj and Hnmj are the vector eigenfunctions. Because of this and also because the coupling constants
The quantities Unmj(+) (z) and Unmj(-) (z) are the 1nmklj(±)are not necessarily small for modes near cutoff,
variable mode amplitudes which are to be determined. radius-to-wavelength ratios which produce modes near
The plus sign indicates forward-traveling waves and cutoff will be excluded in the following.
the minus sign indicates backward-traveling waves.
Normalization constants Nnm have been used so that
the energy flow in the jth fiber is given approximately COUPLING OF SIMILAR FIBERS
by the expression: It has been shown that the coupling of the modes in
two parallel fibers will be weak unless the propagation
Sustiz)u I( thej(+) (Z)
exp-resionm
j(-) (Z)a H t
constants are close to the same value. This can occur
Substituting the above expressions for E and H in the when the radii of the two fibers are nearly the same.
March 1965 COUPLING AND SCATTERING IN OPTICAL FIBERS 265

Since the propagation constants are discrete in most


cases, we only have to consider coupling of the two
HEm or EHnm modes in fiber 1 to the two HEnm or
EHnm modes in fiber 2. There are two of each mode, one
characterized by a positive value of n and the other by a
negative value. There are certain cases (Fig. 1) when
the propagation constants for two different modes are
equal. These cases will be excluded in the following
analysis.
By converting from modes characterized by e-i-n to
modes characterized by sin nO or cos nO, the coupled
differential equations can be written in the following
form:
dul/dz= iYlUl+i/U2 (10) FIG. 5. Comparison of solution for coupling constant with experi-
ment of Bracey. Open circles indicate experimental values of 613
while solid circles indicate experimental values of 3s. n,=1.6,
du2/dz= iY2"2+i3U 1 , (11) n2 = 1.0, a/Xo = 0.2146.

where yj represents the propagation of the mode under


consideration in the jth fiber and / is the coupling apart, the solutions are nearly equal. In order to find
constant. The above set of equations can be written for how close the slabs can be and still have reasonable
the sinnOmodes and the cosnOmodes with only a change accuracy, the coupling constants were compared nu-
in the magnitude of A.The modes characterized by sinnO merically; good agreement resulted even when the
in one guide do not couple with the cosnO modes in slabs were touching.
the other. Bracey et al.7 have determined theoretically and
We can easily solve the pair of equations (10) and experimentally the coupling between the HE,, modes of
(11). The variation of the energy in each guide is given parallel dielectric waveguides. Their theoretical results
by the absolute-value-squared of the mode amplitudes. were obtained by using the resonator-action theorem of
These quantities are plotted in Fig. 4, where K= MacLean.'2 The results obtained here have been used
23
(-Yl-Y2)/ . Notice that when the guides are identical, to compute the coupling constants (see Appendix) and a
K=O and there is a sinusoidal oscillation of energy, the comparison is made with the experimental results of
frequency of the oscillation being fl/7r cycles per unit Bracey in Fig. 5. Again there is good agreement even
length. When the diameters of the guides are not the when the guides are touching. The theoretical curves
same, K#DOand the energy transfer is not complete. agree with those of Bracey, but because of the com-
To justify, in part, some of the assumptions made in plexity of the analytic expressions, it cannot be deter-
the formulation, the problem of the coupling of two mined if they are equivalent in all cases. Bracey's result
parallel dielectric slabs was studied. Since the slab is just for the HE,, modes in identical guides. The
problem can be solved exactly, it was possible to com- solutions developed here may be used to compute the
pare the solution with an approximate solution ob- coupling between any modes in dissimilar guides. The
tained in the manner used in the present problem. It coupling constants 3c and /3- are plotted as functions
was found analytically that when the slabs are far of d/a in Fig. 6 for the HE,, modes.
The agreement between the exact and approximate
solutions in the case of two parallel slabs and between
the approximate solution and the experimental results
in the case of two cylinders helps support the assump-
tions made. In particular we can conclude that (1)
the continuous spectrum does not contribute much to
the coupling and (2) coupling with only one mode at a
time need be considered.

INFINITE LINE OF IDENTICAL FIBERS


We can extend the above analysis to the case of more
than two fibers by considering the coupling between
E2 every pair of fibers in the array. First, we assume that
the presence of a third fiber does not appreciably alter
.5 S.0
the coupling between two fibers. Then, in general, the

FIG. 4. Energy in two mrallel coupled fibers. 12W. R. MacLean, Quart. Appl. Math. 2, 329 (1945).
266 ALAN L. JONES Vol. 55

individual fibers will vary with distance along the


bundle. Since there is strong coupling only in those
regions where the diameters are the same, the net
coupling between two parallel fibers will be less than in
the case in which the diameters are the same through-
out the length of the bundle. This effect is discussed by
Kapany and Burke.'
When the rate of change in diameter of a fiber is
small compared to the diameter, we can neglect the
coupling of modes within a fiber and consider the
propagation constant to be a function of the diameter
I IX'
and hence of the length along the fiber. This can be
shown to be a valid assumption as long as the diameter
change is restricted so that the mode under considera-
tion does not go through cutoff.
Again, considering one mode in each of two parallel
fibers, the coupled equations become:
dul/dz= it12+i±YI(z)U1,
du2/dz=iiu+ i-Y2(Z)U
2, (14)
where the yj are now functions of the axial coordinate.
Equations (14) cannot be solved for a general varia-
FIG. 6. Coupling constants for HE,, modes in two tion of ryIand 'Y2. The case when Yi and -Y2 are linear
parallel fibers. i,= 1.8, n2 = 1.5. functions of z is solvable, however, and lends insight
into the more general case. It would be expected that
the energy would propagate in the two guides almost
coupled equations become independently except in the region where the propaga-
tion constants are close to the same value. A position
deij
-=z='jluj+i E fjkllk, (12) where the propagation constants are equal will be
dz kidj referred to as a "zero crossing."
Consider the case where 7y and 'Y2 are given by
where we are considering only the coupling of ssome
mode in the jth fiber with the same mode in all olL the
other fibers. If it is assumed that there is significant 1.0II
interaction only with a fiber's nearest neighbors and
consider the case of an infinite line of identical fil)ers, .8
the equations can be solved in terms of Bessel functiLIons.
Considering the case when the energy in fib er 0 . -\
equals 1 at z= 0 and there is no energy in the o,ther
guides at this point, the energy in the jth guide can .4
be written
Ej(z)= itj(z)I2=Jj2(20z). (13) .2

This energy distribution is plotted for the first few


fibers in such an infinite array in Fig. 7. o .5 1.0
The solution of the infinite array is of limited im- I
portance in the field of fiber optics; it is quite difficult .2

to make identical fibers, and we are usually more E,02


interested in two-dimensional arrays of fibers rather
than in a single line. However, the above result would 0 .5 1.0
help in the analysis of large arrays of dielectric antennas.

TWO GUIDES OF VARYING PROPAGATION


CONSTANTS
Each fiber has been considered to have a constant
diameter. In an actual fiber array, the diameters of the FIG. 7. Energy in center fibers of an infinite line of fibers.
March 1965 COUPLING AND SCATTERING IN OPTICAL FIBERS 267

Yyi-o+zai, 72=eyo+za2. The solution of Eq. (14) can


be expressed in terms of parabolic cylinder functions." 3
---

If we assume that at z= - o the energy in fiber 1 equals


.8.28
unity while the energy in guide 2 equals zero, the energy E,
o4-- 2 -- .

partition as a function of z is given by the absolute- I 0 1 - \ --- I


RZ
value-squared of these functions
El(z) = I UI(z) 12= c Di, (OZV-10i/4)l2 12
E 2 (Z) = I U2 (Z) 12= ve-,rv 2
ID 2i-.,(fzv-Je3 ri/4 ) l2 (15)

where v=32/ (al-a2) and D,(z) is a parabolic cylinder


function. Ei(z) and E2 (z) have been plotted as func-
tions of z in Fig. 8. These graphs were plotted by using FIG. 8. Transfer of energy in two fibers which have linearly
the asymptotic expansions when |z is large, and the varying propagation parameters.
power series for the functions for small z. If the slopes
of the two propagation constants differ greatly, there
will be little energy transfer. Alternatively, if a, and ca2 can be cast in the form of an integral equation. Since
are nearly equal, almost all of the energy will be we are considering only axially symmetric waves, we
transferred to the other guide. will use only the axially symmetric part of the Green's
function (i.e., the n=O terms). This modified Green's
While this analysis is restricted to the case of a linear
function represents the field due to a ring current on
variation in the propagation parameters, it lends in-
the surface of the cylinder. Since the Green's function
sight into the more general case. It was shown that,
and the field E are defined in slightly different regions
as suspected, there is little transfer of energy except in
(due to the surface roughness), the region of definition
those regions where the propagation parameters are
of E is extended to aid in the formulation. To do this,
about equal. Given two parallel optical fibers: if the
we assume that the field quantity is sufficiently smooth
number of places where the radii were the same, and
near the rough boundary so that its value can be extrap-
the rate of change of the diameter at each zero crossing
olated to the boundary of a fictitious smooth guide of
could be determined, an estimate of the energy trans-
radius ao. It is at this boundary r= ao that the source
fer could be obtained.
in the Green's function is located.
A procedure similar to that used in the section on the
PROPAGATION IN A FIBER WITH ROUGH WALLS
coupling of fibers is used to obtain an integral equation
Consider an infinitely long dielectric "cylinder" of for the field at the surface of the smooth guide. Using
dielectric constant E, imbedded in a medium of di- the fact that the boundary perturbation ef(z) and its
electric constant E2. The surface of the cylinder is taken derivative are assumed small, the boundary conditions
at a variable distance from the axis. We assume that can be expanded in a Taylor series. Only terms of the
the distance from the axis to the rough surface is only a order of E are retained. A discussion of the procedure is
function of the length along the guide. In addition, only given by Miles.'4 Performing this expansion and insert-
axially symmetric waves will be considered. ing the result into the integral equation, we obtain an
The distance from the axis to the rough wall a(z) is integral equation in terms of the field evaluated at the
expressed as
fictitious smooth boundary. This expansion also in-
troduces the roughness function f(z) and its derivative
a(z) = aO[I+ ef (z)] (16) into the integrand. The resulting integral equation can
be split into two separate problems, one involving TE
where ef (z) is the roughness perturbation of the smooth waves and the other TM waves. The following will be
wall. It will be assumed that devoted to solutions for the case of TE waves.
|E «<<1 and that I daldz I<<I. The solution is taken as a sum over the propagating
modes plus an integral which represents the continuous
To find the electric field both inside and outside of spectrum. Each propagating mode is multiplied by an
the fiber, the vector Helmholtz equation must be unknown amplitude function um(±)(z) while the con-
solved subject to the conditions that the tangential tinuous spectrum involves the unknown amplitude
electric and magnetic fields must be continuous across function v(+) (z ;a).
the rough boundary. Substituting the assumed form of the solution into the
Using the dyadic Green's function for the perfect integral equations, we obtain a set of coupled integral
cylinder and the vector Green's theorem, the problem equations in the unknowns um(+)(z) and v(±)(z;a). We
give here one of the integral equations as an example of
13W. Magnus and F. Oberhettinger, Special Functions of Mathe-
matical Physics (Chelsea Publishing Company, New York, 1949). 14J. W. Miles, J. Acoust. Soc. Am. 26, 191 (1954).
268 ALAN L. JONES Vol. 55

their form. HYNhymI -2-Q is small. When the coupling between the
Nth and mth modes is strong, there will be little
it,,,) (z) = AN53NmeXP(iyNZ)+iE f (zo) coupling between the Nth mode and any other mode as
long as the propagation constants of the mth and any
other mode are not close to the same value. The graph
X - )(Zu)]+ |
0nmlu-n(+)(zo)+un ,(a) of Ynm Vs a/Xo (Fig. 1) shows that for TE modes, the
propagation constants are always fairly far apart
X [v(+) (zo; a)+v(-) (zo; a)]da} eiTm(2-O)dzo. (17) except when the diameter of the guide is large. There-
fore, we need only consider coupling between two modes.
The quantities fnm and am(a) are known. In this case, the coupled integral equations can be
The problem has been reduced to solving a set of replaced by the equivalent set of differential equations
integral equations. While it is not likely that an exact (letting n= 1, mn=2, and, 312= 3 for convenience):
solution is obtainable from the set, the form of the
duj/dz= iTyiui+ief(Z)13u2,
equations suggests that an iterative technique will be
profitable. However, it will be observed that in certain du2 /dz=iVy2"2+ieff(Z),13. (21)
cases the first-order term may become quite large.
Again with the consideration of sinusoidal roughness,
When this happens, the iterative solution is no longer
an approximate solution has been obtained for the
valid. An approximate solution may be found, in those
set of Eqs. (21) by an expansion technique. The
regions where the first-order term is large, by recourse
solution obtained is valid in those regions where the
to a set of differential equations obtained from the
iterated solution becomes large. We find that there is an
integral equations.
oscillation of energy from one mode to the other in a
For the zero-order solution to the integral equations,
manner similar to the energy transfer in parallel fibers.
a single propagating mode of amplitude AN is chosen.
By referring to Fig. 4, which shows the oscillation of
The first iteration then yields values for umi±)(z)
energy in parallel fibers, we can see how energy oscil-
and v(+)(z ;a) once f(z) is known. For example, Um(+)(z)
lates between modes. In this case, the two curves repre-
is given by the expression
sent the energy in modes 1 and 2. The parameter K is
It.(+) (z) = A mJNmei'ym given by the expression

+ieil N/3NmeeiazJ f(zo)exp[(iYN-yjl~zo]dzo. (18)


and the 1 in the abscissa must be replaced by 0.5 efi.
A form of roughness which is convenient to work with In the limiting case when K= 0, there will be a complete
but is rather idealized when considering physical transfer of energy from one mode to the other. A
situations is a sinusoidal roughness. Let f (z) be taken as numerical example has shown that there can be a com-
f(z) = sin(Qz), Q&=27r/X, plete transfer of energy from one mode to another in
0.011 cm with a value of 0.01 for e in a 1-,-diam fiber.
where X is the wavelength of the roughness. If this seems like a fairly short distance, it should be
We find upon substituting Eq. (19) into the iterated realized that 0.011 cm is 110 times the diameter of the
solution that the first-order terms are small except fiber.
when Q is near the values
Q = |N-Ym , Q =YN±_Ym. (20) RADIATION FROM THE WALL OF A FIBER
That is, if the spatial frequency of the roughness of the To determine if the wall roughness can cause energy
wall is equal to the sum or difference of the frequencies to be scattered from the fiber, the far field (the field
of any two propagating modes, there will be large when r is large) must be considered to see if it repre-
coupling of these modes. In this case, a finite amount of sents radiation. Since the propagating modes are
energy will be transferred from the Nth mode to the characterized by a mode shape which decays exponen-
mth mode, thus negating the assumption that the Nth tially with increasing r, they can be neglected in the
mode travels undiminished through the length of the representation for the far field.
fiber. When Q is near one of the values mentioned above, With use of the saddle-point approximation, the far
a more general solution must be sought. This can be field can be obtained in terms of an integral involving
accomplished by considering the coupled differential- the wall-roughness function
integral equations instead of the iterated solution of the
integral equations.
Eo (Y,z) = i cA f(Zo)exp(iyNzo) C1 (r, z-zo)dzo, (22)
COUPLING OF MODES
As has been noted, strong coupling between the where Ci(r,zo) is a transfer function obtained from the
Nth and mnth modes only occurs when the quantity saddle-point approximation.
March 1965 COUPLING AND SCATTERING IN OPTICAL FIBERS 269

As a special case, we again consider a sinusoidal


roughness function. The integration in Eq. (22) can be
carried out (again using a saddle-point approximation).
We find that the radiation from the fiber is very de-
pendent on the wavelength of the roughness. Integrat-
ing the energy flow over a control volume, we find that
the rough wall causes an exponential decay of the mode
under consideration. The attenuation constant B is
plotted in Fig. 9 as a function of the wavelength of the
roughness. Also, the radiation propagates at a charac-
teristic angle. This angle depends on the frequency and
is given by
6=tan {(YN-$Q)[k2 - ('yN-2)2'-}. FIG. 10. Expected angular energy distribution, band-limited
white noise, n,= 1.8, n2= 1.5, ao/Xo= 1.0, ao/X =4.0, arrow indi-
The angle 6 is measured from the normal to the axis in cates direction of light in fiber.
the plane made by the axis and the normal.
The energy in the Nth mode will be reduced to eel of
possible to obtain certain moments, for example, the
its initial value after traveling the distance L=B-1. A
mean value of the Poynting vector. We assume that the
numerical example provides an estimate of the amount
second-order properties of the random process which
of attenuation which roughness can cause. Taking a
describes the rough surface are known. The roughness
fiber with ao/Xo=1, nl=1.8, and n2=1.5 (Fig. 9),
function f(z) is taken as a second-order random process
and choosing the wavelength of the roughness to be
and will be denoted by F(z). It is assumed to have zero
equal to the diameter of the guide, ao/X=0.5. For this
mean and to be weakly stationary."
value, BalE2 equals 2.5 for the first mode. Taking a
The expected value of the Poynting vector (P) is
1-A-diam fiber and choosing E=0.01, we find that
obtained by taking the expectation of the absolute
L=0.20 cm. That is, the first TE mode in a 1-,4-diam
value of Eq. (22) squared. We then see that if the auto-
fiber has its energy reduced by a factor of e-l after
correlation function of the roughness F(z) is known,
traveling 2 mm. (Note: the above analysis shows that (P) can be found. Writing the autocorrelation function
there will be no radiation from the guide if the wave- as the Fourier transform of the spatial spectral
density
length of the roughness is very short compared to the
SF(0) of F(z), we can write (P) in terms of an integral
dimensions of the fiber or if it is very long.)
over SF(0) times a transfer function. This transfer
RANDOM ROUGHNESS function is found to be the radiated field produced by a
sinusoidal roughness of spatial frequency U. By a
Since an actual fiber would not be expected to have a change of variable, (P) can be written as an integral
roughness in the form of a simple sinusoid, the effects of over an angular cordinate 0 instead of the frequency
a randomly rough wall on the transmission properties of variable Q. We then observe that the integrand gives
an optical fiber have been considered. an expression for the expected value of the Poynting
Because of the random nature of the roughness vector in the interval 0, O+dO,where 0 is measured as
function, we cannot expect to obtain explicit repre- above. This expression is called the expected angular
sentations for the field quantities. Instead, it is often energy distribution.
The expected angular energy distribution is plotted
in Figs. 10, 11, and 12 as a function of 0 for different
spatial spectral densities, with AN 2 = 1. Figure 10
shows that the expected angular energy distribution is
symmetric about 0=0 when the spectrum is band-
limited white noise, and the cutoff frequency is greater
than -yN+k2.The quantity X,is the value of the wave-
length at the cutoff frequency. Plots for the two TE
modes which exist when a/Xo= 1 are shown. Photo-
graphs of optical fibers indicate that the radiation is
greater in the forward direction. A more realistic
spectrum, namely
SF(0)= (,,,2)' -1 exp -Q/29 2),
0 I cA a Gaussian distribution, resulted in the expected
FIG. 9. Attenuation constant vs reciprocal of roughness wavelength, 15E. Parzen, Stochastic Processes (Holden-Day, San Francisco,
ns= 1.8, n2 = 1.5, a0 /1X0=1.0. 1962), p. 70.
270 ALAN L. JONES Vol. 55

angular energy distribution shown in Fig. 11. We see


that the distribution is now predominately in the
forward direction as anticipated. Figure 12 shows the
distribution for the second TE mode when the diameter
of the fiber is double that used in Fig. 10, while the
wavelength of the light is held fixed. The number and
position of the lobes in the distribution depend on the
fiber diameter and not on the roughness function; this
could be used as the basis for a method of measuring
the diameter and the uniformity of very small fibers.
Besides causing radiation from a fiber, a randomly
rough wall causes the various modes to be coupled.
This problem is more difficult since we must deal with FIG. 12. Expected angular energy distribution, band-limited
differential equations whose coefficients are random white noise, nj=1.8, n,2=1.5, ao/Xo=2.0, ao/X0 =8.0, arrow indi-
functions. A perturbation technique has been applied cates direction of light in fiber.
to the problem to obtain an estimate of the magnitude
of the mode coupling. This analysis shows that a sig- APPENDIX: COMPUTATION OF COUPLING
nificant amount of energy may be transferred from one CONSTANTS
mode to another in about the same distance that a The computation of the coupling constants used in
mode's energy may be reduced by a factor of e-l due to Eqs. (7) and (8) involves an integration of the vector
radiation. It is apparent that separating the phe- eigenfunctions of fibers 1 and 2 at the surface of fiber 2.
nomena of mode coupling and radiation, as has been Since this integration cannot be performed simply, it is
done here, may not be valid in certain cases. If the convenient to obtain approximate expressions for the
fiber can support only one mode, the analysis should be constants. We have done this by using the asymptotic
quite accurate. Even in the case of multimode fibers, expansions for the Hankel functions and an addition
the above analysis lends insight into the transmission theorem for Bessel functions. Because asymptotic
characteristics. By considering the radiation effect expansions are used, the expressions obtained are good
alone, we might conclude that to minimize energy loss only for the case when the guides are relatively far
we need only transmit light along a fiber in a mode apart. However, there is good agreement with experi-
which is far from cutoff. (Recall that the radiation mental data even when the guides are almost touching.
patterns show that a mode close to cutoff is attenuated The expression for the coupling constants will be
rapidly.) However, a mode which is far from cutoff will written out here for the case of identical guides, where
be continuously coupled to modes closer to cutoff the same mode is considered in each fiber. The coupling

(V)
V2
which have a larger attenuation. Therefore, the initial constants zinc and En are given by the expressions
mode would lose energy faster than would be predicted
by considering only the radiation. f.c= b.(d)k22, (Al)
2 2
inS-= b,(d)P yn , (A2)
ACKNOWLEDGMENTS where
The author gratefully acknowledges the help and 2a,,(- l)nu2e-vdla
advice given to him by Dr. S. J. Citron and Dr. J. C. (A3)
a,, (-y,) K2
Samuels of Purdue University. a,,
g . (^Yn) 2 2
- nJn Kn R2

Pu 2v 2kl 2 [J,,/Kn+D4ztD2 ]d/dYE[J 'K.+Di=FD 2]


(A4)
60° d 1 1 n2
-EjnKl+DjFD22=,ya2-J'K,,+- 1--) JnK,
dL U2 u 2l
70e

+-Jn'Kn' +Di':FD2', (A5)


v)
80 where
ya2 eF
D'= -JnKn'(vl2 +2u2) - w24,'Kn
1v3
FIG. 11. Expected angular energy distribution, Gaussian spectral
density, it = 1.8, 12 = 1.5, ao/Xo= 1.0, ao/X0 = 1.0, arrow indicates +vue2(1+-)J,1 Kn ]
direction of light in fiber.
March 1965 M
COUPLING
1965 AND SCATTERING IN OPTICAL FIBERS 271

R2 e= (kl2+k22)12k12
D2'= -[-yaa~u-1J, D'D
21J.D -1D3]
2
2k, a D1= uv-rJ.K,'e,
D 2 = (2k ca2)-'JnR2D
3i
2
D3t=2ya 2Kf+u2v[ v-rK,,+ (1+-)Kn]
K,'v-4 |V{-2 2 2
D3= U V- K ' 2+4u-2v- 4 2
n y 2ki2 K 2 4
a,

and P is the quantity defined by Snitzer.4


-u 2 KK' + 8112 yk1 2 ir2 v- 4Ka
K 4
{ y a V-'Kn'+ Kn
2 2 The s and c superscripts denote modes characterized
by sinnO and cosnO, respectively. For the EH modes,
the upper sign is taken in (A4) and (A5) while the lower
sign is taken for the HE modes. Quantities TALand Ens
O OF A2M2
ERIC have been plotted in Fig. 5 for the HE,, mode.

JOURNAL OF THE: OPTICAL SOCIETY OF AMERICA VOLUME 55, NUMBER 3 MARCH 1965

Infrared Filters of Evaporated Gallium Arsenide


R. P. HowsON
Allen Clark Research Centre, Plessey-UK Limited, Caswell, Towcester, Northzamptonshzire,England
(Received 15 July 1964)

Films of gallium arsenide have been produced by vacuum evaporation on to amorphous substrates and
their optical properties have been measured. A simple, effective way of evaporating gallium arsenide is de-
scribed. The design of absorption filters using evaporated semiconductor films of high refractive index is con-
sidered and the manufacture and properties of such a filter are described.

INTRODUCTION single-crystal and polycrystalline form; the high vapor


SEMICONDUCTORS,
by virtue of their electronic pressure of the volatile component over the molten
compound makes complex apparatus necessary if
structure, act as long-wavelength-pass filters. The
transition between no significant absorption and high stoichiometry is to be maintained when material is
absorption can take place over a very narrow wave- grown from the melt. This difficulty is accentuated if
length range. Notable among the semiconductors for a mixed-compound crystal is to be grown.
this property are the so-called direct-gap III-V com- For optical-filter applications, the high absorption
pounds. A further feature of these materials is the coefficient of these materials would permit thin films
high value of absorption at short wavelengths. Direct- to be used, which may be more easily prepared than
gap III-V compounds, either pure or mixed are known the bulk material. III-V compounds in thin-film form
over a range of energy gaps from 2.0 to 0.15 eV, from have been prepared by vacuum evaporation and vapor
a gallium arsenide-gallium phosphide mixed compound transport.1 ' 2 Vapor transport, using reaction on the
to indium antimonide. The corresponding wavelength substrate surface to form the compound, has the limita-
range is from 6200 A, within the visible spectral region, tion that high temperatures are often needed at the
to 8 A. There is evidence that a continuous variation substrate, which is not compatible with easily available
within this range is possible, using pure and mixed amorphous substrates, like glass, and is not readily
compounds of gallium arsenide, indium phosphide, adapted for producing large-area films, which are
indium arsenide, gallium antimonide, and indium desirable for filters. Vacuum evaporation onto amor-
antimonide.1 An extension to shorter wavelengths may phous substrates has been studied,'l and some optical
be possible by use of compounds with boron or alumi- measurements reported,'' 6 but with little detail.
num as the group III element and nitrogen from group 2 R. K. Willardson and H. L. Goering, Compound Senmicon-
V. ductors Vol. 1. "The Preparation of III-V Compounds" (Reinhold
Optical measurements on III-V compounds are Publishing Corporation, New York, 1962).
3K. G. Gunther, Compound Semiconductors Vol. 1. "Vaporiza-
generally reported for single-crystal materials and tion and Reaction of the Elements," (Reinhold Publishing
have shown absorption coefficients on the short- Corporation, New York, 1962), p. 313.
wavelength side of the absorption edge between 103 4C. Paparoditis, Compound Semiconductors Vol. 1. "Evapora-
tion of the Compound," (Reinhold Publishing Corporation, New
and 104 cm: 1, with refractive indices between 3 and 4. York, 1962), p. 326.
In many cases the compounds are difficult to prepare in 'R. F. Potter and G. G. Kretschmar, J. Opt. Soc. Am. 51,
693 (1961).
I E. K. Miller and J. L. Richards, J. Appl. Phys. 35, 1233 6 S. Martinuzzi, Compt. Rend. 253, 1157 (1961); 255, 110
(1964). (1962).

Anda mungkin juga menyukai