Hollow Waveguide Tapers

On the Analysis of Hollow Waveguide Tapers as
Spot Size Converters for Optical MEMS

Applications
George Isaac and Diaa Khalil
AbstractIn this work, we present approximate formulas for

loss assessment in tapered hollow dielectric waveguide structures
and verify the results numerically. Both the linear and parabolic
taper structures are studied with our technique and the obtained
results are compared with the BPM calculations and a good
agreement is observed. Our results also show that hollow
waveguide tapers may be used as spot size converters.
Index Terms
coefficient.
hollow
waveguide,
Optical
MEMS,
loss
I. INTRODUCTION
OLLOW waveguides are by nature leaky structures. They

do not support guided modes. Treatment of such
structures can be carried out in a very similar fashion to
normally guiding structures through the use of leaky modes
[1]. Leaky modes lose power as they propagate along hollow
structures. However, by exciting the fundamental mode in a
multimode hollow waveguide, the leakage losses can be made
small.
In a recent work [2], hollow waveguides have found an
interesting application in the optical MEMS switch greatly
enhancing its performance. However, lensed fibers are needed
to match the small fiber spot size to the spot size necessary to
excite the fundamental mode in a hollow square waveguide
having dimensions 80mx80m. In this work, we suggest the
use of hollow waveguide tapers as spot size converters, so that
they replace lensed fibers in MEMS applications.
II. TAPER LOSSES
A. Taper Loss Calculation
The power loss coefficients of leaky TE and TM modes of a
hollow slab waveguide having a width w can be derived from
a ray model and are given respectively by [1]
4 h2
w ( h + )
(1)
4 n s 2 ng 2 h 2
w ( ns 2 h + ng 2 )
(2)
where h and are the transverse propagation constants in

the core and cladding respectively and is the longitudinal
propagation constant. These quantities are related by the
equations
2 = ko 2 ng 2 h 2 = ko 2 ns 2 2
(3)
where ko = 2 , is the wavelength; ng and ns are the core

and cladding refractive indices respectively. From the ray
model it is possible to deduce a value for h [1]
h = N w
(4)
with N being the mode number. This result is deduced in a

manner similar to that for a normally guiding dielectric
waveguide, where a mode can only exist only if points on the
same wavefront differ in phase by an integer multiple of 2 .
We shall consider only TE polarization. The TM
polarization can be treated in an exactly similar fashion. If the
tapers are adiabatic, then we can assume that if power is
injected into the fundamental mode of the structure ( N = 1 )
then it will remain in the fundamental mode during
propagation [3]. Therefore, we shall neglect any coupling to
higher order modes in the subsequent analysis. According to
the adiabatic approximation, we can assume that equations (1)
to (4) hold locally at every point inside the waveguide so that
the loss coefficient becomes z dependent. By definition we
have,
(z) =
1 dP
P dz
(5)
It will be more useful for our purposes to manipulate this

equation as follows
1 dP dw 4h 2
(6)
=
P dw dz w
This follows from the approximation of small losses. The right
hand side is just (1) but with h which is the case of small
losses [1]. Rearranging this equation and putting
dz dw = f ( w) we can integrate (6) from the initial width wo
to an arbitrary width w to find the waveguide power at an
arbitrary width as
w 4h '2 f ( w ')
P = Po exp
dw ' = Po exp ( ( w) )
(7)
w w' ' '
w2 B wo 2 + C
ln 2
2
w + C wo B
wo
(11)
wf
L
This integral can be solved analytically under certain
approximations for linear and parabolic tapers. The detailed
calculations can be found in the next section.
Fig. 1. Linear taper geometry.

2
and C =
2
2
ko ns ng
B. Analytical Results
1
where B =
2 ko ng
Using (3) and (4) with N = 1 we have
For the parabolic taper, we have,
1

k o ng
w
k
2

o ng
2
= k o 2 ng 2
(8)
w = wo + ( w f wo )
z
L
and

= ko 2 (ns 2 ng 2 ) +
w
ko ns 2 ng 2 +
1
2ko ns 2 ng 2
f ( w) =

w
2 L( w wo )
( w f wo )2
(9)
which are reasonable approximations for the figures we are

using in this paper. Using (8) and (9) we can write
wo
( w) =
wf
ng ns 2 ng 2
w ' f ( w ')dw '
1 2
1
wo ( w '2
(
) )( w '2 + (
)2 )
2 k o ng
2 ko ns 2 ng 2
(10)
L
z
Let wo be the initial width, w f the final width and L is the

total length of the tapered section. For the linear taper we
have,
Fig 2. Parabolic taper geometry.
In this case (10) integrates to

w f wo
w = wo +
z
L
f ( w) = 1 dw dz = L ( w f wo )
(10) can be integrated using the technique of partial fractions

to yield,
( w) =
2L
2 ( B + C ) ( w f wo )ng ns 2 ng 2
( w) =
[
2L
2
2
( w f wo ) ng ns 2 ng 2
C1 w2 B wo 2 + C C2 1 w
1 wo
ln (
)(
) +
tan
tan
2 w2 + C wo 2 B
C
C
w + B wo B
ln (
)(
) ]
2 B w B wo + B
C3
where C1 = wo
(B + C) ,
(12)
C2 = C / ( B + C ) and
C3 = B ( B + C ) .
C. BPM Assessment
In this section we compare BPM simulation results to the
analytical formulas (11) and (12). The hollow waveguide is
assumed to have a core refractive index of unity (air) and a
cladding refractive index of 3.5 (Silicon). The operating
wavelength is 1.55m. The waveguide tapers from a width of
10m to a width of 80m. At the end of the taper section, we
place a straight hollow waveguide having a width of 80m in
order to calculate the overlap between the propagated field in
the taper section and the fundamental mode of the hollow
waveguide having the final width. Typical structures for linear
and parabolic tapers are shown in figures 3a and 3b, along
with the field propagation inside the guide. A Pade (4,4)
scheme is employed in the BPM calculations to ensure
accurate results.
The waveguide is excited with the local normal mode of the
taper section at its input, assumed to be the same as that of the
axial field a closed waveguide with the same dimensions. The
justification for this can be found in [4], where the lower order
modes of multimode optical waveguides of arbitrary cross
sections are found to be equivalent to those of the longitudinal
fields in a closed waveguide. Since hollow waveguides can be
treated like normally guiding dielectric waveguides except for
a complex propagation constant we may treat their field
distributions similarly. The power inside the taper section is
monitored along with distance and compared to equations (11)
and (12) for different taper lengths ranging from 1000m to
4000m. Results for the linear and parabolic tapers are shown
in figure 4. In tables I and II we compare the total loss in dB
from simulation for the different taper kinds and lengths and
refractive index contrast, taking into consideration the
propagation loss and also the loss due to the inexact overlap
between the field distribution at the end of the hollow taper
and the fundamental mode of the output waveguide.
As a confirmation that the approximations made are
accurate, a numerical integration is carried out to calculate the
exact dependence of power on distance. The loss coefficient is
used as it appears in equation (1). We plot the results for both
tapers in figures 5 and 6 in the worst case which corresponds
to the fastest variation (i.e. for a taper length 1000m). We
can see that our approximation underestimates the losses in
the case of the linear taper structure while it overestimates it in
the parabolic structure. However in the two cases, the
differences are not quite significant.
We also compare the transmission from the end of the taper
for different taper angles. This is shown for the linear taper in
figure 7 and for the parabolic taper in figure 8. The taper angle
is calculated from
w f wo
2L
= tan 1
Fig. 3a. propagation inside linear taper of length 1000m.
Fig. 3b. Propagation inside parabolic taper of length 1000m.
(13)
Fig4a. Taper length 1000m.
Fig4d. Taper length 4000m.
Fig4b. Taper length 2000m.
Fig. 5. Exact, approximate and simulation results compared for linear taper
length 1000m.
Fig4c. Taper length 3000m.

Fig. 6. Exact, approximate and simulation results compared for parabolic
taper length 1000m.
TABLE II
TAPER COMPARISON FOR DIFFERENT LENGTHS, SILICON DIOXIDE SUBSTRATE
Taper
Length
(m)
1000
2000
3000
4000
P
(dB)
0.64
1.29
1.95
2.61
Linear
O
(dB)
1.32
1.49
2.05
2.67
Parabolic
T
(dB)
1.96
2.78
4
5.28
P
(dB)
O
(dB)
T
(dB)
0.46
0.56
0.683
0.828
1.85
1.38
1.03
0.90
2.31
1.94
1.713
1.728
III. DISCUSSION
Fig. 7. Transmission from the linear taper section for air core and both silicon
and silicon dioxide substrates.
Fig. 8. Transmission from the parabolic taper section for air core and both
silicon and silicon dioxide substrates.
Figure 4 indicates that the theoretical formula for linear

tapers predicts a behavior that is close to BPM calculations.
However, for parabolic tapers, there is a relatively larger
discrepancy between BPM and theory. This can be attributed
to the fact that coupling to higher order modes takes place,
which is obvious from the field propagation shown in figure
3b and also from the ripples in the graphs shown in figure 4.
Since higher order modes have higher losses, the theory
underestimates the losses in such a situation but not by a
significant amount. Table I clearly shows that as the taper
length increases propagation losses increase. This is because
light propagates in a smaller width for larger distances for
increasing length tapers. For parabolic tapers, this increase is
smaller than for linear tapers, a result also predicted by theory.
For silicon substrate, the propagation losses increased by
0.136 dB for lengths ranging from 1000m to 4000m in the
parabolic taper but increased by 0.63 dB in the linear taper.
The overlap losses are smaller for long parabolic tapers
compared to linear tapers. Overlap loss cannot be calculated
from the theory presented here as it assumes that there is no
coupling to higher order modes. As for overall performance,
our results indicate that long parabolic tapers can act as spot
size converters with a total loss of 0.7 dB for a length of
4000m, if silicon substrates are used.
IV. CONCLUSION
TABLE I
TAPER COMPARISON FOR DIFFERENT LENGTHS , SILICON SUBSTRATE
Taper
Linear
Length
P
O
(m)
(dB)
(dB)
1000
0.2
1
2000
0.41
0.67
3000
0.62
0.75
4000
0.83
0.90
P=propagation loss
O=overlap loss
T=total loss
Parabolic
T
(dB)
1.2
1.08
1.37
1.73
P
(dB)
O
(dB)
T
(dB)
0.15
0.19
0.24
0.286
2
1.33
0.62
0.416
2.15
1.14
0.86
0.70
We have presented approximate formulas for the analysis of

hollow waveguide taper losses based on the adiabatic
approximation. Theoretical formulas for predicting the losses
along arbitrary hollow waveguide tapers have been presented
and two taper types are studied, linear and parabolic. The
formulas can be used to compare the losses in various types of
tapers to a good accuracy and thus may be used efficiently in
optimization of the taper function. We demonstrated also that
the parabolic taper is superior in spot size conversion to the
linear taper, giving 0.7 dB loss for a length of 4000m. This is
also the case in taper guiding structures. An extension of the
simple theory presented here may take into consideration
mode coupling of leaky modes.
REFERENCES
[1]
[2]
[3]
[4]
Dietrich Marcuse, Theory of Dielectric Optical Waveguides,

Academic Press, second edition, 1991.
Kareem Madkour, Hesham Maaty and Diaa Khalil, "Hollow Waveguides
for NxN Optical Cross Connect Switch", ECOC 2003.
W.K. Burns and A.F. Milton, Waveguide Transitions and
Junctions, Guided Wave Optoelectronics, editor: Theodore Tamir,
Springer , 1988.
Allan W. Snyder and Xue-Heng Zheng, "Optical Fibers of Arbitrary
Cross Sections", J. Opt. Soc. Am. A/vol. 3, no.5, May 1986.

Hollow Waveguide Tapers

Uploaded by

Copyright:

Available Formats

Hollow Waveguide Tapers

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hollow Waveguide Tapers

Uploaded by

Copyright:

Available Formats

On the Analysis of Hollow Waveguide Tapers as

Spot Size Converters for Optical MEMS

AbstractIn this work, we present approximate formulas for

OLLOW waveguides are by nature leaky structures. They

where h and are the transverse propagation constants in

where ko = 2 , is the wavelength; ng and ns are the core

with N being the mode number. This result is deduced in a

It will be more useful for our purposes to manipulate this

Fig. 1. Linear taper geometry.

Using (3) and (4) with N = 1 we have

For the parabolic taper, we have,

which are reasonable approximations for the figures we are

w ' f ( w ')dw '

Let wo be the initial width, w f the final width and L is the

Fig 2. Parabolic taper geometry.

In this case (10) integrates to

(10) can be integrated using the technique of partial fractions

Fig. 3a. propagation inside linear taper of length 1000m.

Fig. 3b. Propagation inside parabolic taper of length 1000m.

Fig4a. Taper length 1000m.

Fig4d. Taper length 4000m.

Fig4b. Taper length 2000m.

Fig4c. Taper length 3000m.

Figure 4 indicates that the theoretical formula for linear

We have presented approximate formulas for the analysis of

Dietrich Marcuse, Theory of Dielectric Optical Waveguides,

You might also like