Beruflich Dokumente
Kultur Dokumente
A comparison of vectorfiniteelement
formulations for waveguide analysis
S. Seller!, M. Zoboli
Dipartimentodilngegneriadell'Informazione, Parma University,
Viale delle Scienze, 1-43100 Parma, Italy
Abstract
A comparison between vectorial and semi-vectorial finite element
method (FEM) formulations for modal analysis of dielectric waveg-
uides is reported. The influence of the index step and of the mesh
distribution on the solution accuracy is investigated. In particular,
besides on the effective index, attention is focused on the correctness
of the field distributions whose evaluation is a crucial requirement in
waveguide analysis and related applications. As many authors have
been concentrated on the accuracy of the propagation constant, no
useful criteria have been proposed to test the spatial distribution of
the unknownfield.To overcame this lack two error figures have been
introduced and applied to different FEM solutions. In particular the
following formulations have been compared: that one based on the
transverse magnetic field, those based on the so called edge-elements
and a new formulation presented by the authors. Results show the
availability of the proposed approach and the usefulness of the intro-
duced figures for a deep waveguide analysis.
1 Introduction
time and the memory occupancy or the ability to deal with spurious modes,
to directly solve for the propagation constant, to model structures with lossy
media and to provide the desired solution precision. For what concerns this
last aspect, the main attention has always been focused on the effective
index, or on the normalized propagation constant precision, rather than on
the field distribution so that no useful criteria have been assessed to mea-
sure the correctness of the unknown vector components, of their amplitude
ratio or of the polarization direction. Nonetheless these informations are
very important, in particular when accurate field values must be evaluated
in each point of the mesh as in the case of waveguide nonlinear analysis
[1]. This paper attempt to cover this lack through the usage of two error
figuresfirstintroduced by Vassallo [2]: the shape error and the size error.
These figures will be apply to compare the precision of the field distribution
evaluated by means of different vectorial and semi-vectorial formulations.
In particular, through the shape error, the effects of the refractive index
step variation and of the grid distribution on the evaluated field accuracy
will be investigated.
The paper is organized as follow; the next section will be devoted to briefly
summarize the formulations considered and compared in this work. Section
three will present a detailed and comparative error analysis of the solutions
of the different approaches. Conclusion follow.
Vx(r'Vx%)-^M = 0; (1)
(S]{H}-%[T]{H} = 0 (2)
where {//) is the vector of the unknown magnetic field values and [S] and
[T] are two symmetric and sparse matrices. In particular [S] depends on the
phase constant /?. The algebraic system thus obtained provides k* as the
eigenvalue. This is not an efficient approach and direct formulations must
be preferred for practical device design. Four different direct formulations
are considered in the following. The first one has been proposed by Lu et
al. [4]. By imposing the divergence condition on the curl-curl equation, an
accurate and efficient direct formulation which involves only the transverse
Transactions on Engineering Sciences vol 11, 1996 WIT Press, www.witpress.com, ISSN 1743-3533
Software for Electrical Engineering 25
components of the magnetic field and which eliminates the spurious solu-
tions is obtained. This approach avoids to evaluate the longitudinal com-
ponent thus reducing the memory occupancy and the computational efforts
as well. Additionally this formulation has been here improved by using sec-
ond order triangular elements rather than bilinear rectangular ones as in
[4]. Besides allowing to better describe any geometrical shape of waveguide
cross-sections, and in particular the step fibre circular cross-section, these
elements considerably increases the solution accuracy. The second and the
third formulations, based on the edge elements, have been presented by
Lee et al. [5] and by Hay at a et al [6]. Following the definitions given in
[7] they will be referred as T\ and TI.S formulations. In the first case the
transverse field has been evaluated by means of six tangential unknowns for
each triangles [6], while by means of six tangential and two facial ones [5] in
the second one (high order edge element). Thanks to the edge element im-
plementation, Lee et ai introduced a transformation of variable to directly
compute the propagation constant. By substituting the unknown transverse
magnetic field component Ht with 0Ht, being 0 the phase constant, they
achieved an algebraic eigenvalue problem whose eigenvalue is represented
by an explicit known function of /3 itself.
The fourth formulation, presented in detail by the authors in [8], uses the
same transformation; it is applied to a full vectorial node-based FEM formu-
lation with the penalty function techniques [9]. This simple transformation
is useful when using edge elements as in [5] but it is no more sufficient
when node-based elements are considered. In this case, in fact, in order to
move the spurious solution spectrum away from the region of interest, the
functional corresponding to equation (1) must be modified adding the so
called penalty term [9]. This term changes the matrix [S] so that the cited
transformation is no more sufficient to the aim. Further calculations are
necessary [8] and yield:
[S']{H'}-p(r]{H'} = Q. (3)
being t/> a single field component, the transverse one or the whole field vector
as well. In the previous figures || || indicates the L* norm and the subscript
ref refers to the analytical solution. While the shape error measures the
correctness of the unknown field component distribution, the size one gives
informations on the amplitude ratio of the components 0i and t/^.
-1
As first results [8], Fig.l shows the relative error e% of the effective index
of the fundamental mode versus the normalized frequency v rk^Jn^ n^;
e is defined as (ric// - ^e//)/^e// being n^/ and n^j the numerical and the
Transactions on Engineering Sciences vol 11, 1996 WIT Press, www.witpress.com, ISSN 1743-3533
Figure 2: The punctual distribution of arctg( Hy/H*) within the fibre core
for a numerical and an analytical solutions respectively.
values of the arctg( Hy/H^) are plotted for a numerical and an analytical
solutions respectively. As it can be observed by these distributions, the two
polarization directions are different and the fields must be rotated, as well
as normalized, before any comparison.
The Ht and HZ shape errors thus obtained are reported on Fig.3 versus
the mesh point number, for the fundamental mode HEu at a normalized
frequency i/ = 2.
node-bemad
T,, formula tic
edgeelements
formulation
T, edge
!<
Figure 3: The Ht and H^ shape errors versus the mesh point number. Fibre
parameters as in Fig.l; the HEu mode at v 2 has been considered.
While little differences arise for the minor component #,, the Ht shape
error curves can be clearly distinguished. The new node-based approach
provides the best figure while the two edge element formulations presents
the highest errors. This effect can be due to the nature of their unknowns,
which are related to the interelement interface tangential field rather than
to the field components along the coordinate axes. Finally the transverse
formulation traces a curve parallel to the node-based one as the same inter-
polating functions for the Ht component are used. Nonetheless the lack in
the evaluation of HZ considerably reduces its accuracy.
Notice that no figures involving H^ can be obtained for the transverse for-
mulation. As a consequence, in the following, in order to carry out a full
comparison between all of the cited formulations, particular attention will
be focused only on the Ht component.
It has been observed that the behaviour reported in Fig.3 is repeated for
any value of the normalized frequency. In particular the Ht shape errors
increase from the node-based formulation ones up to the T\ edge element,
passing through the transverse and the TI.S edge element ones. As an ex-
ample Fig.4 shows the errorfigurefor v = 4.
Some differences arise by increasing the index step as can be observed by
comparing Fig.4 and 5. In the last case the core refractive index has been
changed to %*, = 2.0 and %<* = 3.0. The step values thus obtained are quite
similar to the index steps which occur, for instance, in integrated optics.
Transactions on Engineering Sciences vol 11, 1996 WIT Press, www.witpress.com, ISSN 1743-3533
While the node-based approach maintains almost the same values, the other
formulations slightly modify their behaviours. In particular the performance
of the transverse one are considerably degraded of almost and even more
than one order of magnitude in the two cases respectively. At the same
time the 7%.5 edge element errors increase up to the extent of overcoming
the TI ones. Therefore, for what concerns the Ht distribution, the relative
performance order is changed even if the same change does not occur for
the effective index. In fact an analysis of the propagation constant in these
conditions reveals a behaviour similar to that one already reported in Fig.l.
rtocto-based formulation
T,, odge formulation
T, d0 formulation
formulation
nod-baed formulation
T,, dg formulabon
o T, edge formulation
transvm formulation
4 Conclusion
References
[3] M. Zoboli, P. Bassi, The finite element method for anisotropic opti-
cal waveguides, from "Anisotropic and Nonlinear Optical Waveguides",
edited by G.Stegeman and C.G.Someda, ELSEVIER 1992.
[4] Y. Lu, F.A. Fernandez, "An efficient finite element solution of inho-
mogeneous anisotropic and lossy dielectric waveguides", IEEE Trans.
Microwave Theory and Tech., MTT-41, pp.1215-1223, June/July 1993.
[5] J.F. Lee, D.K. Sun, Z.J. Cendes, "Full wave analysis of dielectric waveg-
uides using tangential vectorfiniteelements", IEEE Trans. Microwave
Theory and Tech., MTT-39, pp.1262-1271, Aug. 1991.
[7] B.M. Dillon, P.T.S. Liu, J.P. Webb, "Spurious modes in quadrilateral
and triangular edge elements", COMPEL, vol.13, supplement A, pp.
311-316, May 1994.
[11] A. Jay, Fern Modeling and Preprocessing from "Finite Element Hand-
book", edited by H. Kardestuncer and D.H. Norrie, McGraw-Hill, New
York, 1987.