Modelling Optical Waveguide Bends by the Method of Lines

TELKOMNIKA, Vol.16, No.4, August 2018, pp. 1490~1499
ISSN: 1693-6930, accredited First Grade by Kemenristekdikti, Decree No: 21/E/KPT/2018
DOI: 10.12928/TELKOMNIKA.v16i4.5191  1490
Received November 14, 2016; Revised January 30, 2018; Accepted April 16, 2018
Modelling Optical Waveguide Bends by the Method of
Lines
Ary Syahriar*1
, Nabil Rayhan Syahriar2
, Jusman Syafiie Djamal3
1,3
Electrical Engineering Department, Faculty of Science and Technology University al Azhar Indonesia,
Jakarta Indonesia
2
Mechanical Engineering Department, Bandung Institute of Technology
*Corresponding author, e-mail: ary@uai.ac.id
Abstract
A rigorous analytical and semi analytical method of lines has been used to calculate the
transverse-electric field attenuation coefficient of guided mode as it travels in waveguide bends structure.
Both approaches then were compared to get a better understanding on how the attenuation behaves along
single curve waveguides with constant radius of curvature. The Helmholtz Equation in polar coordinate
was transformed into a curvalinier coordinate to simulate the waveguide bends using the method of line
analysis. The simple absorption boundary conditions are used into the method of lines to demonstrate
evanescent field of the guided mode nature as its travels in waveguide bends structures. The results show
that a reasonable agreement between both theoretical approaches.
Keywords: Optical waveguide bends, Method of lines, The transverse-electric field attenuation coefficient
Copyright © 2018 Universitas Ahmad Dahlan. All rights reserved.
1. Introduction
One of the most important optical integrated devices building blocks is waveguide
bends because it defines the overall size of integrated optics in single subtrates. Waveguide
bends are required in many basic optical structures, including directional couplers, modulators,
ring resonators [1], arrayed waveguide filters [2], optical delay lines [3], S-bend attenuators [4].
and Mach-Zehnder interferometers. However, waveguide bends experience loss as the guided
mode enters the curved section which depends on confinement factors and radius of curvature.
The loss can be minimized by increasing the mode confinement i.e. by increasing the refractive
index differences between core and cladding layers or by decreasing radius of curvatures.
Increasing mode confinement will increase the coupling losses when waveguide is coupled into
the fiber optics and decreasing radius of curvature will increase the overall integrated optics
size. In a silica-based waveguide, there is normally only a very slight variation in refractive index
across the cross-section, to allow low loss coupling to a single mode fibre. The slight variation in
index is most helpful as it permits the vector wave equation to be replaced by a scalar equation
in which the electric field is represented by one vector component. This simplification is known
as the weak-guidance approximation. Therefore a precise knowledge on bend waveguides
characteristics become important to design a compact integrated optical systems.
So far, a number of efficient numerical techniques have been proposed for the analysis
of optical waveguides. These include the finite difference method (FDM), the finite element
method (FEM), the beam propagation method (BPM), and the method of lines (MoL) [5]. The
finite difference method is the oldest numerical method for solving partial differential equations.
It is simple to program and easily applied to non-homogenous refractive index profiles. This
method subdivides the domain into many subregions, in which the partial derivatives are
replaced by finite difference operators. A set of linear equations are then solved to obtain the
eigenvalues. The drawback of the FDM is it offers less flexibility in the modeling of the domain
since the subregion is normally rectangular in shape [6]. The finite element method (FEM) can
model the most intricate domain geometries. In FEM, the waveguide cross section is divided
into surface or volume elements and the field in each element is approximated by a polynomial.
The field continuity conditions are imposed on all interfaces between the different elements.
A variational expression for Maxwell’s equations then is employed to obtain an eigenvalue
matrix equation which is solved by standard methods. This method requires a more complex

TELKOMNIKA ISSN: 1693-6930 
Modelling Optical Waveguide Bends by the Method of Lines (Ary Syahriar)
1491
programming structure and is more demanding in both computer time and memory [7]. The
beam propagation method (BPM) has been used to analyze various two- and three-dimensional
optical devices. The original BPM used an FFT algorithm and solved a paraxial scalar wave
equation. The basic idea of the BPM is to represent the electromagnetic field by a superposition
of plane waves propagating in homogenous media. The advantages of the BPM are that it can
be applied to a structure with an arbitrary cross-section, and that both guided and radiative
waves are included in the analysis. However since the formulation is derived under the
assumption that the refractive index variation in the transverse direction is very small, the FFT-
BPM cannot be applied to structures with large index discontinuities [6].
The method of lines (MoL) has been proved to be a very useful tool for the analysis of
general waveguide systems [8]. It is a semi analytical method, in which the wave equation is
discretized as far as necessary in the transverse direction and solved analytically in the
longitudinal direction, which results in less computational effort. An accurate result can be
obtained since the MoL behaves in a stationary fashion and convergence is monotonic [9].
Discontinuous fields can be described accurately because the interface conditions are included
in the calculation. Furthermore, the MoL is relatively easy to implement using computer
numerical methods. In this paper we compare two different approaches, namely a simple quasi-
analytic theory based on integration of a phenomenological absorption coefficient, and the
method of lines (MoL). Both are applied to a number of different waveguide bend curvatures.
In this paper we applied the method of lines with third order absorbing boundary
condition to analyse weakly guiding optical waveguides bends characteristics. For the first
approximation we have transformed the Helmholtz wave equation in polar co-ordinates to
Cartesian coordinates to simplify the discretisation of waveguide structures. In the process we
compared the results with analytical methods as the correct references that has been developed
previously. We found that the MoL results are in good agreement with analytical methods. The
discrepencies arised from differents radius of curvature used in the calculation and the choices
of absorbing boundaries parameters.
2. Research Method
2.1. Analytical Approach
To analyse the effect of a waveguide bend, consider a bend formed by a circular arc
with radius of curvature r as shown in Figure 1. It is assumed that only the fundamental mode
propagates in the guide. If the radius of curvature is large enough ( r ), then the properties
of the mode are effectively those of a mode traveling in a straight guide. However, as r
decreases, attenuation is expected to occur. Let P(s) be the total power carried by the mode at
any point s along the bend. Assuming that the rate of power loss is proportional to the power
carried by the mode at that point, we can write:
Figure 1. Section of a curved planar waveguide
)(
)(
sP
ds
sdP
 (1)

 ISSN: 1693-6930
TELKOMNIKA Vol. 16, No. 4, August 2018: 1490-1499
1492
where  is the attenuation coefficient. Provided is constant, equation 1 has the solution:
s
ePsP 
 )0()( (2)
Marcatili and Miller have shown that the attenuation coefficient is indeed constant for a fixed
radius, and can be expressed as [10]:
rC
eC 2
1

 (3)
where C1 and C2 are functions of the waveguide parameters but are independent of r.
Equation 3 shows that the attenuation coefficient increases exponentially with decreasing
bending radius; however, if the radius of curvature become large enough the attenuation
becomes negligible. It also shows that the change of with r is dominated by the value of C2,
which (for a weakly-guiding guide) is given by [11]:
2
2
3
2
)2(2
n
n
C
eff


 (4)
Here neff = neff - n2, where n2 is the refractive index of the cladding.
The C2 value also provides a method for characterising mode confinement, which is
useful when investigating techniques for reducing bend losses. Equation 3 shows that  is also
affected (but less strongly) by the value of the coefficient C1, which is defined as [11]:
t
l
cZ
C

'
1
2
1
 (5)
where the parameters Zc, t , and ’
l are given by:
2
2
)
2
cos(2
2 






h
h
n
Zc



(6)
)
2
(cos)sin(
2
1
2
2 h
h
h
t



  (7)
and,



h
l e
h







2
cos
2
2'
(8)
where  is given by:
2
20
21
nk 

(9)
and h, and  have their usual meanings [12].
The C1 coefficient as defined in equation 4 is not a direct function of neff, but is related
to the difference between the propagation constant within the guide and the cladding refractive
index n2. Additionally, the C1 coefficient is strongly model dependent, and so can be used to
calculate the guide shape and other parameters. The above formulation of the C1 and the C2
coefficients was based on the derivation made by Marcatili et.al [10] and later adopted by
Minford et.al [11]. However, two other formulations have been provided by Lee [12] and
Marcuse [13,14]. Lee’s version of the C1 and the C2 coefficients is as follows [12]:

1493
h
o
e
h
hnk
C 


)
2
(cos
)2(
2 2
2
2
1

 (10)
and,
2
2
2
)(2
n
nn
C
eff 


(11)
Here, ko,  and h have their usual meaning. In a similar way, Marcuse’s version of the
coefficients is as follows [14]:
)exp(
)()2(
2
22
2
2
1
22
1 h
knnh
C
o





 (12)
and,
2
3
2
3
2


C (13)
here, the propagation constant is defined as effonk . It might be expected that these
different formulations would give similar results; however, this was not found to be the case.
2.2. Method of Lines
In this analysis we begin by considering the behaviour of a guided mode as it travels
around a bend of constant curvature. Figure 2 shows a schematic of the geometry. The
waveguide has a constant radius of curvature r, which is measured from the centre of the guide.
The guide is of width h, which is assumed to be much less than r and is centred on a
computational window of width w. The core and cladding refractive indices are given by n1 and
n2 respectively.
Figure 2. Discretisation of a planar waveguides bends by the MoL
Assuming a y-polarised electric field, the Helmholtz wave equation can be written in
polar co-ordinates as [15]:

 ISSN: 1693-6930
1494
0
11 22
2
2
22
2
 yo
yyy
Enk
EEE







(14)
To implement a numerical solution of equation 14 a modification must be made.
It involves changing the co-ordinates to a local co-ordinate system that follows the centre of the
waveguide along the propagation direction [15]. This change also allows the possibility of
analysing the field profile at the local cross-section. By making the substitution:


rs
rx


(15)
Equation 14 may be transformed to:
0)()1()1()1( 22
0
2
2
2
2
2
2
 y
yyy
Exnkcx
x
E
cxc
x
E
cx
s
E






(16)
where the constant c = 1/r represents the waveguide curvature. Note that it is easy to see how
equation 16 reduces to the scalar wave equation for a straight waveguide when c=0. There are
some advantages using equation 16. Firstly, the computational window can be restricted
because the centre is along the path of the waveguide. Secondly, the index profiles need not be
altered as the radius of curvature changes, in contrast to the other methods which use the
modified index profile [15]. To solve equation 16 by the MoL, the equation is now discretised
using the finite difference operator, by putting:
2
11
2
2
)(
2
x
EEE
x
E iiiy


 


(17)
and:
x
EE
x
E iiy


 
2
11


(18)
If this is done, equation 16 can be written in matrix form as:
02
2
 EQ
ds
Ed 

(19)
where = [E1, E2, E3, ......, EN]t
is a column vector containing discretised values of the field
Ey(x), at the points x1, x2, .... xN, and Q

is a tri-diagonal matrix defined by :






















































22
22
2
22
1
0
22
1
22
2
2
2
2
2
2
2
1
2
1
2
2
)1(000
........
00)1(0
000)1(
0)1(00
........
)1(0)1(
00)1(0
2
1
)1(2)1(00
........
0)1()1(2)1(
00)1()1(2
)(
1
ncx
ncx
ncx
k
cxc
cxccxc
cxc
x
cxcx
cxcxcx
cxcx
x
Q
NN
NN

(20)

E

1495
Assuming that there is no back-reflection, the general solution for constant matrix elements has
the form:
inp
sj
ETeTE
 1
 
(21)
where is a matrix containing the eigenvectors of arranged in columns, 

is a diagonal
matrix containing the eigenvalues of , and inpE

is the input field vector. One of the most
important parameters associated with the waveguide is the fractional power that remains in the
core at point z. This power is approximately given by the overlap integral:
2
),()0,()( 


 dxzxExEzP (22)
where E(x,0) is the input field and E(x,z) is the field at point z.
2.3. The Absorbing Boundary Condition
To calculate the modal field of the curved waveguide, it is necessary to restrict the
extent of the computational window. Once again, this is done using absorbing boundary
conditions. Derivation of appropriate boundary conditions in polar co-ordinate is generally rather
complicated [16]. In the calculation described here, we have adopted a simpler approach,
applying a straight guide boundary condition to the curved waveguide equation, by using the
assumption that the radius of curvature is large enough that the mode inside the bend is similar
to that of straight guide.
The absorbing boundary condition is inserted into the edge of the matrix components of
equation 20 [16]. In this case, we have used the more effective third-order absorbing boundary
condition, where the radical is approximated by:
2
20
2
202
1
Sqq
Spp
S


 (23)
Differences in the choice of the coefficients, p and q, produces different families of
absorbing boundary conditions. These result in differences in the angle of exact absorption of
the incoming wave by the absorbing boundary layer. Table 1 shows a list of the coefficient
values and absorption angles of the approximations that are most commonly used. Here, we
have used the L type of approximation.
Table 1. Coefficients for different third-order absorbing boundary conditions,
after reference [4.16]
Type of approximation p0 p2 q2 angle of exact absorption (
o
)
Pade’ 1.0000 -0.7500 -0.2500 0.0
Chebyshev L
0.9997 -0.8086 -0.3165 11.7, 31.9, 43.5
Chebyshev points 0.9965 -0.9129 -0.4725 15.0, 45.0, 75.0
Least square ( L2) 0.9925 -0.9223 -0.5108 18.4, 51.3, 76.6
Chebyshev-Pade’ (C-P) 0.9903 -0.9431 -0.5556 18.4, 53.1, 81.2
Newman points 1.0000 -1.0000 0.6697 0.0, 60.5, 90.0
Chebyshev L 0.9565 -0.9435 0.7038 26.9, 66.6, 87.0
q0 = 1.0000 for each approximation
3. Results and Analysis
To get a better understanding on how the guided mode evolution during its propagation
inside curved waveguides; we have used two methods, i.e. an analytical theory that is assumed
to be the right approach and the method of lines. In both calculations parameters of a
waveguide bends with a radius 5000 m through an angle of 45o
with different refractive index

T

Q

Q

 ISSN: 1693-6930
1496
changes have been used. Figure 3 shows the variation of C2 calculated using equation 4 with
wavelength for different values of n, i.e. for different degrees of waveguide confinement. These
results show that C2 increases as the confinement becomes higher and also that C2 values
generally decrease slowly at long wavelengths.
Figure 3. Variation of the parameter C2 with wavelength, for different values of n as predicted
by equation 4
Figure 4 (a) and (b) show a comparison of the C1 and the C2 values as a function of
waveguide width, calculated by Marcatili’s approximation of equation 4 and 5; Lee’s
approximation of equation 10 and 11; and Marcuse’s approximation of equation 12 and 13
respectively. Here, the parameters of n1=1.463, n2=1458, =1.525 m, with h varying from
4 m to 7 m, have been used.
(a) (b)
Figure 4. A comparison of different approximations for (a) the C1 value, and (b) the C2 value
Figure 4 (a) shows that the C1 coefficients of Lee’s and Marcuse’s expression are in a
good agreement. However, Marcatili’s equation gives much lower value. This might well be
because of the different approaches used to derive the C1 coefficient. The Marcatili
approximation is obtained from the complex solution of the eigenvalue equation of the
waveguide bend, while both Lee and Marcuse use a different approximation based on the local
rate of power radiation from the bend. Furthermore, for the C2 coefficient, Lee’s and Marcatili’s
approximations give a good agreement, while Marcuse’s approach gives apparently incorrect
values. It can be concluded that Lee’s expression is the most likely to be correct. This
assumption will be validated later by comparison with the rigorous method of lines.
0
0.0005
0.001
0.0015
0.002
0.0025
0.003
1.3 1.35 1.4 1.45 1.5 1.55 1.6
 (m)
C2(m
-1
)
n =0.007
n =0.006
n =0.005
0
10000
20000
30000
40000
50000
60000
70000
80000
4 4.5 5 5.5 6 6.5 7 7.5
Width(m)
C1(m-1
).
Marcatili
Lee
Marcuse
0
500
1000
1500
2000
2500
4 4.5 5 5.5 6 6.5 7 7.5
Width (m)
C2(m-1
).
Marcatili
Lee
Marcuse

1497
To illustrate the way in which the guided mode evolves in a curved waveguide, we first
compare the input and output mode shapes obtained after travelling round a bend of radius
5000 m through an angle of 45o
. Figure 5 (a) shows results obtained for a guide of core index
1.464, cladding index 1.458 and width 5 m at a wavelength of 1.525 m, while Figure 5 (b)
shows results for a similar but less strongly confining guide which has a core index of 1.463.
The calculations have been done by using the MOL scheme.
(a)
(b)
Figure 5. Input and output field distribution of the fundamental mode after travelling around a
waveguide bend of radius 5000 m through an angle of 450
. (a) n = 0.006, (b) n = 0.005
Figure 5 demonstrates that the output field profile of the mode generally extends into
the cladding and its peak is reduced, so that it is gradually radiating power. The amount of the
power loss depends on the degree of confinement. For example, in Figure 5(a), the output field
extends into the cladding only to a very limited extent, and the input and the output field shapes
are very similar. However, in Figure 5(b) the output field extends much further into the cladding
due to the reduction in confinement. The degree of asymmetry also increases considerably as
the confinement is reduced.
We now use the results of the method of lines calculation to estimate an effective
attenuation coefficient along a uniformly curved waveguide. This can be done by using
equation 1, where the
ds
sdP )(
values are found by evaluating the difference in the integrated
optical power across the mode in the computational area between two adjacent axial
propagation steps. Figure 6 shows the attenuation coefficient found in this way as a function of
bending angle , for the parameters n=0.005, 0.006, and 0.007,n2=1.458, =1.525 m,
h=5 m, and r=5000 m.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-20 -10 0 10 20
Lateral position x (m)
TransverseFieldamplitude(arb.unit) input field
output field
n =0.006
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-20 -10 0 10 20
Lateral position x (m)
Transversefieldamplitude(arb.unit)
input field
output field
n =0.005

 ISSN: 1693-6930
1498
Figure 6. Attenuation coefficient as a function of the bending angle , for a different
degrees of confinement
In each case, the value of is not constant, but rises gradually from zero at the start of
the bend and settles to a steady-state value only after some rapid fluctuations. In early
analyses, the fluctuations have been associated with transition loss [17-18]. However, recently it
was demonstrated that they are merely a mathematical artefact which is inherent in numerical
modelling of bends using beam propagation methods, and the steady-state value is an accurate
estimate of the attenuation coefficient after the mode has settled to its final lateral position.
A comparison of the attenuation coefficients predicted by simple theory and the MoL (at
large axial distance) is shown in Figure 8. In the analytical approximations, the C1 and the C2
coefficients needed to find the values have been calculated by using each of the three
approximations.
Figure 8. A comparison of the attenuation coefficients obtained from the MoL and from the three
different analytical expressions
Figure 8 demonstrates that good agreement is obtained between the analytical form for
the loss coefficient based on Lee’s expression and the MoL calculation. A slight difference,
however, occurs at a small radius. In contrast, Marcatili’s formulation predicts a very low value
of when compared to Lee’s approximation and the MoL calculation, while Marcuse’s
formulation predicts much higher values.
The residual discrepancies between the predictions of Lee’s theory and the MoL may be
explained as follows. In the MoL, the calculation results are highly dependent on a proper
choice of absorbing boundary condition at the edge of the computational windows, so that un-
suitable conditions give rise to significant reflection back into the computational window and
hence lower apparent loss. A similar effect also appears in another numerical scheme [19].
0
0.00005
0.0001
0.00015
0.0002
0.00025
0.0003
0.00035
0.0004
0.00045
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
=s/r
(m
-1
).
n =0.005
n =0.006
n =0.007
0
0.005
0.01
0.015
0.02
0.025
1000 2000 3000 4000 5000 6000
Radius of curvature (m)
(m
-1
).
Numerical (MoL)
Analytic (Marcatili's)
Analytic (Lee's)
Analytic (Marcuse's)

1499
4. Conclusion
We have investigated different analytic approximations to the local loss coefficient in
waveguide bends based on constant radius of curvature, and have uncovered disagreement
between several previously published analytical expressions. To verify the accuracy of the
attenuation coefficient on several publish analytical expression, we have used the beam
propagation algorithm based on the method of lines in polar co-ordinate. We have found
reasonable agreement with the analytic approximation to the local loss coefficient based on
Lee’s approach. This agreement might be used to extend the calculation of loss in waveguide
bends structure in modelling continuously-varying S-bends waveguides using cascaded section
method. Residual disagreement is ascribed mainly to the moderate performance of the
absorbing boundary conditions used to limit the range of the calculation window.
References
[1] BE Little, ST Chu, HA Haus, J Foresi, JP Laine. Microring resonator channel dropping filters. J.
Lightwave Technol. 1997; 15: 998-1005.
[2] M Zirngibl, CH Joyner, LW Stulz, T Gaiffe, C Dragone. Polarization independent 8 × 8 waveguide
grating multiplexers on In P. Electron. Lett. 1993; 29: 201-202.
[3] RR Hayes, D Yap. GaAs spiral optical waveguides for delay-line applications. J. Lightwave Technol.
1993; 11: 523-528.
[4] X Jiang, W Qi, H Zhang, Y Tang, Y Hao, J Yang, M Wang. Loss crosstalk 1 × 2 thermooptic digital
optical switch with integrated S-bend attenuator. IEEE Photon. Technol. Lett. 2006; 18: 610-612.
[5] F Ladouceur, JD Love. Silica-based buried channel waveguides and devices. Chapman & Hall.
London. 1996.
[6] WP Huang. Methods for modeling and simulation of guided-wave optoelectronic devices: part I:
modes and couplings. EMW Publishing, Cambridge, Massachusetts, 1995.
[7] SM Saad. Review of numerical methods for the analysis of arbitrarily-shaped microwave and optical
dielectric waveguides. IEEE Trans. on Microwave Theory and Tech. 1985; MTT-33: 894-899.
[8] T Itoh. Numerical techniques for microwave and millimeter-wave passive structures. John Wiley &
Sons. New York. 1989.
[9] U Rogge, R Pregla. Method of lines for the analysis of dielectric waveguides. IEEE J.Lightwave
Technol. 1993; LT-11: 2015-2020.
[10] EAJ Marcatili. Bends in optical dielectric guides. Bell Syst. Tech. J. 1969; 48: 2103-2132.
[11] WJ Minford, SK Korotky, RD Alferness. Low-loss Ti:LiNbO3 waveguide bends at =1.3 m. IEEE J.
Quantum Electron. 1982; QE-18: 1802-1806.
[12] DL Lee. Electromagnetic principles of integrated optics. John Wiley & sons. New York. 1986.
[13] D Marcuse. Length optimisation of an S-shaped transition between offset optical waveguides. Appl.
Opt. 1978; 17: 763-768.
[14] D Marcuse. Bending losses of the asymmetric slab waveguide. Bell Syst. Tech. J. 1971; 50: 2551-
2563.
[15] SJ Garth. Mode behaviour on bent planar dielectric waveguides. IEE Proc. Optoelectron. 1995; 142:
115-120.
[16] TG Moore, JG Blaschak, A Taflove, GA Kriegsmann. Theory and application of radiation boundary
operators. IEEE Trans. Antenna and Prop. 1988; 36: 1797-1812.
[17] J Yamauchi, S Kikuchi, T Hirooka, M Nakano. Beam propagation analysis of bent step-index slab
waveguide. Elect. Lett. 1990; 26: 822-824.
[18] M Rivera. Lowest-order mode transmission in multimode dielectric S-bends. Opt. Quantum Electron.
1997; 29: 323-333.
[19] J Saijonmaa, D Yevick. Beam-propagation analysis of loss in bent optical waveguides and fibers. J.
Opt. Soc. Amer. 1983; 73: 1785-1791.

Modelling Optical Waveguide Bends by the Method of Lines

More Related Content

What's hot (19)

Similar to Modelling Optical Waveguide Bends by the Method of Lines (20)

More from TELKOMNIKA JOURNAL (20)

Recently uploaded (20)

Modelling Optical Waveguide Bends by the Method of Lines