Discussions of dielectric slab waveguide

We discussed how to use the graphical method to find out the modes in the slab waveguide last week.

Here, there are several points you would need to notice:

(1) The intersections of the two curves determine h and  , and thus  . The resulting discrete values of 
determine the “propagation modes” of the waveguide. The values of  so determined are often called the
“eigenvalues” of the modes.
(2) There is always at least one mode (lowest-order symmetric mode) which can propagate in a symmetric
waveguide.
(3) As h increases, higher-order antisymmetric and symmetric modes become allowed (they alternate). As
expected from our wave vector-model discussion, higher-order modes correspond to a larger number of
nodes in the field profile:

(4) Higher-order modes (for a given value of aV ) have a smaller  than the lower-order modes
 they are “less confined” to the core
(also have a larger value of h => more oscillations within the core)
(5) Because different modes will propagate with different velocities (  is different from mode to mode), the
modal dispersion will happen.

Note that multi-mode signals will distort as they propagate => for optical communications one generally
requires single-mode propagation: It is easy to show that single-mode operation is obtained if the core is
sufficiently small for a given index difference and a given wavelength:

a (where  is the free-space wavelength).
4 n2 2 − n12
(6) Suppose we label the modes as m=0,1,2,3,….
(even m for symmetric modes, odd for antisymmetric ).
− i  mz
The propagating field is Em( x, y )e
(Actually, it’s independent of y for a slab waveguide, but the form is usual for fibers and ridge or 2-D,
waveguides). The “spectrum” of  (i.e. the set of all allowed  ) is discrete for the confined modes.
The spectrum of the radiation (unconfined) modes is continuous –i.e. any value of  is possible.

(7) Dispersion ( −  ) diagrams

It is often useful in many optical systems (not just waveguides) to consider a plot of the allowed propagation
constants  as a function of frequency  , or vice versa.
For simplicity, we’ll consider just the symmetric slab waveguide here.
n2
Recall when   n2 ko = ,no modes are possible
c
n1
when   n1ko = ,have radiation modes
c
n n
when 1    2 ,have discrete confined modes
c c
Determined by eigenvalue eqn. or graphical solution

a
Note aV = n2 2 − n12
c
 increasing the frequency  (for a fixed slab thickness a )
Increases the radius of the circle
 get successively more discrete modes
-- using  =  − n1 ko , we can see that as  increases, the corresponding value of  for a particular mode
2 2 2 2

will also increase

-- (an accurate plot of course requires a numerical solution to get the actual value of  )
Thus the modes may be plotted qualitatively as follows

(8) The modes of the waveguide are orthogonal.

(9) The modes form a complete set.

This means that any propagating field distribution in the waveguide can be written as a sum over modes.
E ( x, y, z ) =  amEm( x, y )e − i mz
m
The am (expansion coefficients) are factors determining what “weight” each mode has in contributing to the total
field.

Note some physical consequences:

(a) The modes are orthogonal, so you can’t expand one mode in terms of the others. Thus if an input
beam only excites one mode (e.g. m=4), then only that ( 4th ) mode will propagate.
(b) Remember the modes propagate with different velocities. Suppose you inject a pulse which has a
transverse field profile which excites 3 modes. What comes out the other end of the waveguide? Three
pulses, of course!

We have been discussing these features of modes with respect to the slab waveguide, but the basic concepts still
hold for waveguides which confine the wave in two dimensions (i.e. both x and y)

e.g. rectangular waveguide

We will not treat this problem in any detail, except to note the qualitative form of the modes.
In the case two mode indices are required, one corresponding essentially to the # of antinodes in the x-
direction, the other for the y-direction.
Numerical aperture of Waveguides

Not all source radiations can be guided along an optical fiber. Only rays fall within a certain cone at the input
of the fiber can be propagated through the fiber.
𝛼𝑚𝑎𝑥 gives the total internal reflection at the core-cladding interface. (When 𝛼 = 𝛼𝑚𝑎𝑥 then 𝜃 = 𝜃𝑐 )
If 𝛼 > 𝛼𝑚𝑎𝑥 , the E.M wave will penetrate the cladding layer and will eventually lost.
Now, let’s find the maximum angle (𝛼𝑚𝑎𝑥 )
At the 𝑛0 /𝑛1 interface, the Snell’s law tells us:

𝑛0 𝑠𝑖𝑛𝛼𝑚𝑎𝑥 = 𝑛1 sin⁡(90° − 𝜃𝑐 )

𝑛 𝑛 𝑛 𝑛 𝑛 √𝑛1 2 −𝑛2 2
=>⁡𝑠𝑖𝑛𝛼𝑚𝑎𝑥 = 𝑛1 sin(90° − 𝜃𝑐 ) = 𝑛1 cos(𝜃𝑐 ) = 𝑛1 √1 − 𝑠𝑖𝑛2 𝜃𝑐 = 𝑛1 √1 − (𝑛2)2 = 𝑛0
0 0 0 0 1

The numerical aperture N.A. can be defined as:

N.A.=⁡√𝑛1 2 − 𝑛2 2
N.A.
=>⁡𝑠𝑖𝑛𝛼𝑚𝑎𝑥 = 𝑛0

Total acceptance angle is 2𝛼𝑚𝑎𝑥

Step-Index Optical Fibers
Of course, technologically the most important and familiar optical waveguide which confines light in both
x and y is the optical fiber. A common form of the fiber is a simple conceptual extension of our slab guide:

n2  n1 , of course
As usual, the propagation equation we have to solve is the Helmholtz eqn.
 2 E + ni 2 ko 2 E = 0 i=1,2
Now, given the symmetry of the problem, we need to work in cylindrical coordinates

Most textbooks simply start by writing down the wave eqn. for a scalar field E in cylindrical coordinates:
1  E 1 2 E 2 E
(r )+ 2 + + ni 2 ko 2 E = 0
r r r r  2 z 2
As usual, we assume the z-dependence is simply e − i z , so E (r ,  , z ) = E ( r ,  )e − i z
1  E 1 2 E
(r )+ 2 + (ni 2 ko 2 −  2 ) E = 0
r r r r  2

This equation may be solved via separation of variables:

E ( r, ) = R( r )( )
2E d 2
= R
 2 d 2
E R
=
r r
1  E 1  E  2 E   dR d 2R
(r ) =  +r 2 = + 2
r r r r  r r  r dr dr
 R  2 R R  2
+ 2 + 2 + (ni 2 ko 2 −  2 ) R = 0
r r r r  2
r2
Multiply through by :
R
 r R r 2  2 R 1  2 2 2 2
+ + + r (ni ko −  2 ) = 0
R r R r 2
  2

Note that each term depends either on r or on  , but not on both. Thus we can write
r R r 2  2 R 1  2
+ + i ko −  ) = −
2 2 2
r 2
( n = constant
R r R r 2   2
(both sides are independent, so they must equal a constant ).

By convention, the constant is called l 2 .

(1) Azimuthal eqn.

1  2
= −l 2
  2
 2 2
+l  = 0
 2
=> ( ) = A cos l + B sin l
(some texts write it as Aeil + Be−il )
Thus we find azimuthal modes, where the field is modulated angularly (i.e. in the intensity, observe an
even # of lobes)

(2) Radial Eqn:

 2 R 1 R l2
+ + i ko −  − )R = 0
2 2 2
( n
r 2 r r r2
Just as with the slab waveguide, in order to have bound solutions, we must have
  n2 ko in the core
  n1ko in the cladding
 define  = n2 ko −  (  =”kappa”)
2 2 2 2

 2 =  2 − n12 ko 2
=> we have radial eqn., for the core and cladding

 2 R 1 R l2
(i) core : + + ( 2
− )R = 0
r 2 r r r2
 2 R 1 R l2
(ii) cladding: + − ( 2
+ )R = 0
r 2 r r r2
The solutions to these equations are the family of Bessel functions. They share some common
features with the sine, cosine, and experimental solutions we found appropriate for the slab waveguide
problem.
(1) As usual, we ignore mathematical solutions which diverge →  as r → 0 or as r →  , since
these are unphysical!
l2
(2) In Eqn.(i), when  − 2  0 , the solution is the Bessel function of the first kind of order l, usually
2

r
written Jl ( r )
Two things to remember:
(a) The Jl functions are oscillatory functions of r, just like sines and cosines. (see plot Figure 5.4,
next page)
In fact, when 𝜅𝑟 >>1 (large r), they are simply damped cosines:
2 1 𝜋
𝐽𝑙 (𝜅𝑟) ≅ √𝜋𝜅𝑟 cos⁡(𝜅𝑟 − (𝑙 + 2) 2 ), 𝜅𝑟 ≫ 1
(b) The Bessel functions are orthogonal, just as the sine and cosine modes of the slab problem.
l2
(3) In eqn. (ii), when  + 2  0 , the solution is the Bessel function of the second kind of order 𝑙,
2

r
usually written l ( r ) .
The l are damped solutions. In fact, when 𝛾𝑟 ≫ 1,
𝑒 −𝛾𝑟
𝑘𝑙 (𝛾𝑟) ≅
√2𝜋𝛾𝑟

Note that higher-order modes (𝑙 >0) will both radial and angular modulation.

As with the slab waveguide, the allowed values of  (i.e. the eigenvalues) of the guided modes
are obtained by matching the fields at the core of cladding boundary (tangential E and H
continuous).

Qualitatively, one can see that the situation is similar to before, e.g.
 Similarly to the slab waveguide case, the  and  values are coupled
 2 +  2 = ko 2 (n22 − n12 )
And a normalized frequency (V-number) is defined:
V = koa (n2 2 − n12 )

Note for slab we defined V = k0 (n2 2 − n12 ) ,

to be consistent, we should have defined it as koa (n2 2 − n12 )

This is an important parameter (usually quoted by manufactures of optical fibers), since it basically
determines the number of modes which can propagate.

The mode structure of multi-mode fibers is quite complex, and not germane to our goals here, so we will
simply quote some results in two important limits
(1) V large. In this case the number of modes which can propagate can be approximated by
V
# of modes= 4
2
(This includes polarization degeneracies)
Of course, multimode fibers have a large core, and are thus good for carrying high power (e.g. for
illumination), but they are poor for optical communication because of the modal dispersion problem.
(2) V small.(i.e. small core ,long wavelength)
If V< 2.405, then only a “single mode” can propagate down the fiber. Clearly this is the fiber
of choice for optical communications, as there can be no modal dispersion.

There is no cutoff for the lowest-order mode; every step-index fiber will support at least one
guided mode. If the fiber parameters are fixed, then the cutoff condition for higher-order
modes is
2 a
c = (n2 2 − n12 )
2.405
Now, it should be noted that the term “single-mode fiber” is a little bit of a misnomer. The fiber is
cylindrically symmetric, so it cannot distinguish input polarizations. In other words, we could inject 
wave polarized either along x̂ or along ŷ , and they both would have the same mode profile and the same
 . (These two modes are “degenerate”)

The functional form of the field profile of the fundamental mode in the core is the J 0 Bessel function.
Modes of the E.M.Field

What is a “mode”? We have been using the term quite a lot and will continue to do so, so we should sharpen
up the concept a bit. Let’s start with the wave equation for the electric of magnet field:
2 E 2H
 2 E =  OR  2
H = 
t 2 t 2
The simplest case is uniform space, so  and  are constants independent of position.
n2
Then  =
c2
We look for solutions that oscillate harmonically, i.e. at a single frequency 
E (r , t ) = E (r )eit or H (r , t ) = H (r )eit
n 2 2 n 2 2
Thus  E (r ) = −
2
E ( r ) or  H (r ) = −
2
H (r )
c2 c2
n
We usually define the wavevector k = , so we have the Helmholtz equations
c
2 E (r ) = −k 2 E (r ) or 2 H (r ) = −k 2 H (r )
 2 : operator; −k 2 : constant = eigenvalue
These are eigenvalue equations.
That is, when a functional operator (  2 in this case) acts on an eigenfunction (a “mode”), it yields the same
spatial function (the mode) multiplied by a constant (the eigenvalue).

Free space: a useful set of solutions are plane waves

eigenfunctions = plane waves (modes )
eigenvalues =wavevector (squared)

Note the crucial physical feature: the spatial function oscillates everywhere in space with the same
frequency and phase.

Why is this useful? Well, plane waves are easy to manipulate and solve problems with mathematically,
but they do not actually exist in real life! Real physics waves are at best approximations to plane wave.
So, what do we do if we have a real wave oscillating at frequency  but with a complicated spatial
dependence E ( r ) ?

Let’s consider our plane waves as modes of a rectangular box of volume V, with periodic boundary
conditions
 Any wave inside the box must be a sum over the discrete modes

E( r ) = ˆr (  ai e −i kx x ) (  ame ) (  ane −i kz z )

−i ky y

i m n

Where to satisfy periodic boundary conditions

e − ikx ( x + L ) = e − ikxx
So kxL = 2 l ,l=0, 1, 2,...
2 l
kx =
L
2 m 2 n
ky = kz =
Similarly, L , L

Now if we have let the box become infinitely large, a continuum of plane waves is allowed, and the
sum over modes becomes an integral over a continuum of plane waves.

E ( r ) = ˆ  d 3ka (k )e − ik r : (inverse) Fourier transform!

a (k ) =  all d 3rˆ  E ( r )eik r : Fourier transform of the real-space field distribution.
space

In words, the modes of free space are plane waves, and an arbitrary spatial field distribution may be obtained
by summing over those modes, which is mathematically equivalent to an inverse Fourier transform.

The full 3-D Fourier transform mode expansion is actually rarely used in options. It’s nice for rectangular
cavity problems, but those are also of little real relevance to optics. When considering the propagation of
light, we are much more concerned with the following type of problem:

We may know the We may want to We choose the

field in some find the field in propagation to be
Reference plane reference plane B primarily along Z.
A ( ⊥ ẑ )

Notice that we may know the field only in plane A; this means that the 3-D Fourier transform mode
representation of the field is of no use, since the 3-D transform requires that we know the field everywhere!
(The integral is over all space)
Since the field is propagating primarily along the z-direction, let’s represent the field as
E ( r ) =  ( x, y )e − i  z
(Scalar field)
In a manner similar to the Fourier transform, we can write the spatially dependent field as an integral
over plane wave modes, but integrating only over kx and k y :

E ( r ) =  A( kx , ky )e − i ( kyy + kxx +  z )dkydkx
−
This is called the angular spectrum representation of the field; we’ll see why in a moment. It looks a
lot like a Fourier representation, but it’s not quite the same. There are 3 spatial variables on the left, but the
integral runs only over kx and ky .
Note that since E ( r ) must satisfy the Helmholtz eqn.,
(2 + k 2 ) E ( r ) = 0
 − kx 2 − ky 2 −  2 + k 2 = 0
k 2 = kx 2 + ky 2 +  2
We are assuming space is uniform between planes A and B, so we can assume a uniform index of
refraction, so
n 2 2
k2 =
c2
Fixes the magnitude of the total wavevector,
Thus, although the integral runs over all −  kx , ky   , the value of
 are constrained:
n 2 2
2 = 2
− kx 2 − ky 2
c
n 2 2
When   0( k + k 
2 2 2
x y
)
c2
 is real and E ( r ) =  ( x, y )e − i z
Corresponds to a wave propagating in the +z direction.

2
n 2 2  2 
When   0 ( k x2 + k y2 
2
=   ,
  
2
c
 = wavelength in medium)
− z
 is imaginary and E ( r ) =  ( x, y )e
Corresponds to an evanescent wave.

We will return to the meaning of this when we treat the problem of diffraction. For now, we want to
take home two results:
（1）The angular spectrum representation is a mode representation of the field propagating in uniform
space between planes A and B. Decomposing the field into its plane wave components requires knowledge
only of the field in A.
2𝜋
Plane wave components with 𝑘𝑥 2 + 𝑘𝑦 2 > ( 𝜆 )2 are evanescent.

（2）Consider k = kxx + kyy +  z

ˆ ˆ ˆ
 2
k 2 = kx 2 + ky 2 +  2 = ( )2


Thus the decomposition of a field into its angular spectrum corresponds to decomposing it into plane wave
modes propagating at different angles with respect to the z axis.