See discussions, stats, and author profiles for this publication at: [Link]

net/publication/305676541

Patch antenna optimization in COMSOL multiphysics

Conference Paper · May 2016

DOI: 10.1109/ELEKTRO.2016.7512045

CITATIONS READS

9 1,952

4 authors, including:

Zdeněk Urban Lubomír Ivánek

VŠB-Technical University of Ostrava VŠB-Technical University of Ostrava
6 PUBLICATIONS 14 CITATIONS 32 PUBLICATIONS 73 CITATIONS

SEE PROFILE SEE PROFILE

Yahia A. Zakaria
Engineering Research Division National Research Centre (NRC), Ministry of Scienti…
18 PUBLICATIONS 87 CITATIONS

SEE PROFILE

All content following this page was uploaded by Lubomír Ivánek on 03 December 2019.

The user has requested enhancement of the downloaded file.

 Patch antenna optimization in
COMSOL Multiphysics
Lukas WEZRANOVSKI, Zdenek URBAN, Lubomir IVANEK, Yahia ZAKARIA
Department of Electrotechnics, Faculty of Electrical Engineering and computer science, VSB–Technical University of Ostrava,
[Link] 15/2172, 708 33 Ostrava-Poruba, Czech Republic

[Link]@[Link], [Link]@[Link], [Link]@[Link], [Link]@[Link]

Abstract— This paper deals with the different kinds of Objective function is an n-dimensional real functions of real
optimization. Especially, it uses COMSOL Multiphysics software variables marked f (x1, .., xn). Plurality means the area, for
and its subsequent utilization at the patch antenna. The article which can be found solutions to be considered correct,
deals with optimization of basic dimensions the patch antenna possibly physically realizable. In our case this is the shape
depending on the parameter S11 and the field strength at a
optimization, it means a change of variable parameters -
defined point. This feature is very important, because it helps
determine whether the antenna is properly built or modelled. mostly geometric dimensions in order to achieve an optimal
From this reason, it helps save time and resources of state of monitored parameters. These parameters are antenna
development teams. gain, signal leakage between elements of antenna arrays, i.e.
between each microstrip antennas positioned on one substrate.
Keywords— optimization; COMSOL Multiphysics; patch
antenna; radiation pattern; far-field; frequency; Although in theory we can define arbitrarily difficult
optimization problem, it is not always possible to find method
I. INTRODUCTION that would solve the problem completely. If we decide use to
The essence of this work was the application of optimization optimize procedures which are implemented in commercial
software, the choice is limited even more. In this thesis we
tasks implemented in software COMSOL when designing
used optimization method, which includes a COMSOL
patch antenna. Optimization consists in finding the extremes
optimization module.
of real functions that describing a real state of the system in
this case to achieve the best possible adaptation of antenna
gain for a given frequency.

Fig.2 Patch antenna with an adaptation element. [3]

TABLE I.
Name of the Value
Description
variable (mm)
D 1.54 Substrate thickness
Fig.1 Optimization module at COMSOL. [2] W_line 3.5 Width of the feed line
W_patch 12.77 Width of the patch
One of the definitions of optimization saying that it is a L_patch 15.75 Length of the patch
mathematical discipline in which we want find minimum or W_stub 2 width of the gap in the patch
maximum of the function f (x) (also called the objective L_stub 5 Longitude plunge into the patch
function or purpose function) on a given plurality X. W_sub 2 * W_patch Width of the substrate
L_sub 2 * L_patch Length of the substrate
This research is supported by the project SP2016/143 “Research of L 25 Distance between spots
antenna systems; effectiveness and diagnostics of electric drives with a.
harmonic power; reliability of the supply of electric traction; issue data Basic values of the antenna design (more in [1])
anomalies.”. The authors would like to express their thanks to the members of
Department of Electrical Engineering at VSB – Technical University of
Ostrava for their valuable instructions and support.

‹,((( 
 II. DIFFERENT METHODS OF OPTIMIZATION This method involves modifying the form of a regular multi-
dimensional polygon (simplex), whose peaks are formed by
A. Monte Carlo method combining of optimized parameters.
In Monte Carlo method performs numerical solution of These adjustments are made based on the values of
problems using several repeated random experiments. This optimization criteria identified in the individual peaks.
method represents a broad class of computational algorithms Simplex is a simplified generalization of a triangle in any n-
based on repeated selection of random file to obtain a dimensional space, but may be a line segment, triangle on a
numerical result. They are often used for physical and plane, tetrahedron in three dimensional space and so on.
mathematical problems, particularly when it is difficult or Generally, this is a special type of polynomial with n + 1
impossible to use other mathematical methods. Monte Carlo vertices in N dimensions. The essence of the method, are the
methods are mainly used in three different problem groups: rules under which alters the dimensions and orientation of the
optimization, numerical integration, and generate random data whole simplex, and thus the value of optimized parameters.
derived from the probability distribution at nonlinear The entire process has progressive (sequential) character.
problems. Relevant change of peak is always carried out based on the
evaluation of the success of previous adjustments.
B. BOBYQUA
D. Coordinate search
Bound Optimization by Quadratic Approximation is a new
algorithm for general unrestricted optimization calculation This algorithm starts with an initial guess parameter at space
which takes into account the curvature of the objective in which the function calculates the score. On the basis of this
function - creates the quadratic interpolation models. It is a space begins search for the reach of better score than was the
method that have base on the approximation of smooth initial estimate. In the framework of this method, the search
unimodal objective function f (x) - polynomial of the second algorithm selects one point in each axis control parameter for
stage - the parabola. The approximation occurs around the better value scores. Corresponding distance searching from the
minimum of the function. If we denote the point values of the current base can be defined at some point during the search of
objective function f (x) xa < xb < xc and their corresponding the local environment. This point is selected as the basic
functions fa, fb and fc, we can use Lagrange interpolation position for further searches in environment. If no better
polynomial. results are not available. In environment around the current
base location, is search distance for algorithm reduced by a
factor defined with the user.
(1) III. OPTIMIZATION ACCORDING TO MATCHED ANTENNA IN
COMSOL MULTIPHYSICS
Then it will be approximately quadratic parabola f (x) and its
sharp global minimum will be in the stationary point xk. This To optimizing the dimensions of patch antenna we can
point is the only real root of the equation proceed as follows.
To ensure the best possible transfer of wave from cable to
antenna, we adjusted the size of the power port using the
(2) parameters S11.
Due to the high demands on the computing power, the own
C. Nelder-Mead
optimization is first performed at one frequency 5.6GHz -
Nelder-Mead technique is heuristic search method that can identified as f_x. After we get a general overview of the
converge to non-stationary points. It is easy to operate, and it optimization module, we performed the calculation for given
is used for a large class of problems. The method was range frequency from f_min up to f_max. The basic variance
designed by John Nelder & Mead Roger in 1965. [4] of dimensions optimization based on [5]. The emitters of
length L in the direction of its longitudinal axis in the range
from 0.7 to 1.0 λ, and the width W in the direction
perpendicular to the longitudinal axis in the range of 0.5 to 0.8
λ, are created symmetrically relative to the middle radiator.
A. Input port optimizing
In the first part, we manually calculated the estimated
dimensions of the patch antenna exactly according to available
literature and mathematical models.
Now we use COMSOL with the exact simulation of
parameters we can verify the manual calculation.
Subsequently, with optimizing adaptation of the input port
tried to achieve the best results.

Fig.3 Example of Nelder-Mead algorithm. [4]


Calculations for simplification we perform at one frequency, possible to operate in a range 5.5-5.7GHz, however, during the
due to the required application is chosen 5,6GHz. Optimized construction of the antenna we must also examine the
variable is W_line - the width of the power line. immediate work frequency around to rule out possible
interference and parasitic effects. Therefore, we calculate a S11
f = abs([Link].S11dB) value for the anticipated extended operating range of the
antenna ie. 5.1-6GHz. The result is a chart below, it is a
The setup of program COMSOL allows to seek maximum or directional characteristics for a range of frequencies and the
minimum, so we take advantage of the fact that our best result value of the S11 depending on the frequency.
of optimization provides the absolute value from the S11 to one In section Study 1, Step 1, Frequency domain in Study settings
frequency f_x.
and subsection click to Frequencies for entering frequency
Next steps:
range.
In the top menu, select the Optimization Study.
Enter the start frequency f_min.
In the menu, choose Optimization Study and set the following
Enter the number of steps to calculate.
parameters of optimization: The number of steps should be at least equal to the raster of
The method used for optimization - Nelder-Mead frequency what is used for the 5GHz frequency range, ie. enter
Tolerance Optimization - 0.01.
5MHz to the end frequency f_max.
Objective Function - select the right corner Electromagnetics
Then click on Replace to overwrite values for the calculation.
waves, ports, S parameter dB and double-click, we insert
After pressing the Compute, will run calculation.
variable into the table.
The resulting values can be found in bookmark Tables > New.
Then we change expression function to abs
([Link].S11dB), because we need to optimize the
absolute value S11 in dB.

Fig.4a The course of optimizing input port.

Fig. 5 Radiation pattern depending on the frequency while

optimizing the power line.
Fig.4b The course of optimizing input port.

In menu of Control Variables and Parameters, select the

parameter what you want to optimize, which in this case
W_line and enter numerical values for our optimization.
The default value Initial Value - calculated value of 3.53 mm.
Lower bound – lower back on the 1 mm.
Upper bound – the upper limit to enter 6 mm.
In section Output While Resolving, Probes we select All and
tick Keep Objective values in table and select a table, in which
to store the values - New. After pressing the Compute will run
calculation. The resulting values can be found in bookmark
Tables > New.
B. The calculation of antenna parameters for a range of
frequencies
Above computed simulation we must check even for a wider Fig.6 The course of the S11 depending frequency for
range of frequencies. According to VO-R / 10 / 04.2012-7 it is optimizing the power line.


C. Optimization by eFar The "Objective" parameter in table II. shows the calculation
results for the given parameters and the optimization function.
Following optimization of the essential dimensions of the Example: if the optimization function absolute value S11 then
patch and W_patch L_patch combined with the length of the the value in column "objective" the calculated absolute value
power line L_stub, for achieve the best possible antenna gain. S11 for a given dimensions. When the optimization function
f = comp1.intop1([Link]) directly S11 then the value objective negative number.

Type of a search result - maximum. TABLE III.

Seeking solutions - maximum from the resulting object. freq (Hz) S11 (dB) Lumped port impedance Zlp
The method used - Nelder-Mead limited to a maximum of 50 (Ω)
calculations. 5.6000E9 -12.179 30.284-1.2664i
c.
Optimizing for one frequency f_x. Values S11 and Zlp after optimization by Far-Field.
In the menu Component1 > Definitions right-click and select
Integration Component Couplings and choose integration.
With this action created intop1 for integration calculations.
In the settings can be found section of Geometric entity level
and choose Point. In the Selection we select a point on the
outer boundary conditions perpendicular to the substrate - in
our model it was the point number 33rd. In the Optimization
we set method, tolerance and other calculation parameters
outside variables as in the previous optimization. Then we
select in the Objective Function Model feature, Component1,
Far-field, Far-field gain - [Link]. We adjust the
calculation so it included the setting in Intop1 = comp1.intop1
([Link]). In the Control Variables and
Parameters we select optimization for antenna width and
length of the antenna - and L_patch, W_patch in the range 15
to 35 mm and L-stub 1-10 mm.
Start the calculation for a given frequency. Tables and graphs
below describe the result.

TABLE II.
W_patch L_patch L_stub Objective
0.014230 0.019140 0.0052000 2.34
0.035000 0.019140 0.0052000 3.64
0.014230 0.035000 0.0052000 -6.4806
0.014230 0.019140 0.010000 -4.0812
0.017692 0.027070 0.0060000 -25.759
0.024615 0.019140 0.0052000 3.50
0.024615 0.019140 0.0076000 0.43883
0.024615 0.027070 0.0052000 -25.583
0.030320 0.014000 0.0065185 -4.8178 Fig.7 After optimization by far-field.
0.025833 0.024280 0.0054815 -14.883
0.029199 0.016570 0.0062593 -1.6821 From the above results and the overall behavior of
0.026955 0.021710 0.0057407 -8.5369 optimization is necessary to optimization methods include the
0.028638 0.017855 0.0061296 1.0085 effect of changing the width of the power lines W_line.
0.034220 0.018283 0.0034198 0.096763
0.027016 0.018926 0.0065549 2.0311 D. Optimization by S11
0.029117 0.020282 0.0051737 -3.7025 Followed again by optimizing of the power line. We use
0.028757 0.018462 0.0058906 3.28 dimensions W_patch, L_patch in combination with the length
0.031899 0.018902 0.0043055 0.67040 of the power line L_stub and the width of power line W_line.
0.028237 0.018920 0.0059926 -0.95765 Based on the previous refinement of dimensions using
0.026686 0.018801 0.0055453 5.48 optimization by eFar, we can repeat the optimization for reach
0.029808 0.019140 0.0052000 1.90 the best adaptation of the antenna.
0.025816 0.019033 0.0058775 5.71
0.021604 0.018843 0.0058819 4.37
f = abs([Link].S11dB)
0.025816 0.019033 0.0058775 5.71
b.
Values of the optimization according to Far-Field.
Type a search result - maximum.


Seeking solutions - maximum of the resulting object. Table calculated values optimization is shown in disertation
The used method - Monte Carlo with the maximum limited to thesis annex 2. of Mr. Wezranovski [1].
500 random calculations. By this optimization we get the following values S11 and Zlp.

Table calculated values optimization is shown in disertation TABLE VI.

thesis annex 1. of Mr. Wezranovski [1]. freq (Hz) Lumped port S11 (dB)
By this optimization we get the following values S11 and Zlp. impedance (Ω)
5.6000E9 46.383-1.8062i -27.549
TABLE IV. f.
Values after S11 optimization improvement.
freq (Hz) Lumped port S11 (dB)
impedance (Ω)
5.6000E9 44.342+4.7259i -22.153
d.
Values after first S11 optimization.

Fig.9 After optimization improvement by S11 (dB).

E. Verification of antenna parameters for a range of

frequencies
After optimization at one frequency, we once again returned to
the conversion of antenna parameters to the estimated working
Fig.8 After optimization by S11 (dB). frequency range and its surroundings.

Subsequently we perform more accurate calculation of the

above-mentioned conditions of optimization and the following
values.

TABLE V.
The initial Minimum Maximum
value (mm) (mm) (mm)
L_patch 19.7 17 21
W_patch 22.96 21 24
W_line 1.43 1.2 1.7
L_stub 6.66 5.66 7.66
e.
Value of each parameters range. Fig.10 The graph of calculated values S11 and
impedance Zlp power port.


 After proper optimization it showed that the patch antenna
In section Study1, Step1, Frequency domain under Study cannot be solved as independent parameters and values -
settings at subsection frequencies click for entering frequency matching of the input and intensity of field, because these
range. Specify the initial frequency f_min. Then enter the values are strongly linked together. Because of that we are
number of steps to calculate. The number of steps should be at dealing with solving complex optimization sizes L_patch,
least equal to the raster of frequency which is used for the W_patch, W_stub, W_line, L_stub.
5GHz frequency range, ie. 5MHz. Then enter the end
frequency f_max. Click on Replace to overwrite values for the IV. CONCLUSION
calculation. After pressing the Compute button, it will run the Optimization must be done gradually and Multicriterial.
calculation. The resulting values can be found in bookmark Individual optimization methods have a great tendency to
Tables > New. remain in simulations at the local extremes, for this reason is
very important to have at least a basic understanding of the
Before proceeding, it is highly advisable to perform the above behavior of the mathematical and physical model. The
simulations for the wider around operating frequency to be consecutive steps of calculation have confirmed the expected
able to restrict possible undesirable effects. Therefore, we will behavior of simulation and errors in manual calculation given
calculate the values S11 again from 4GHz frequency range to by the simplification of the calculation.
7GHz.

Fig.13 Non-optimized antenna (left side) comparing with

optimized antenna (right side).
Fig.11 Impedance of power port in the wider area.
REFERENCES
TABLE VII.
[1] Wežranowski,L:Building and configuring Wi-Fi networks in terms of
freq (Hz) S11 (dB) LP impedance (Ω) antennas and propagation antenna design technology 3 x 3, Disertation
5.60E+09 -27.912 46.448-1.5600i thesis, VŠB-TU Ostrava , FEI, 2015.
[2] https //[Link]/conference/2014/data/ConfTokyo2014Mini_Opt
6.48E+09 -24.71 45.266-2.8801i [Link]
g.
Local minimum features and value matching. [3] DANĚK, J., Anténa pro univerzální vysílač. Diplomová práce. Brno:
FEKT VUT v Brně,2010/2011.
From the above calculations, we obtained an optimized shape [4] [Link]/~msolufse/[Link]
of patch antenna for a given frequency range, which we used [5] [Link]
to further optimization. From this graph of matching we can [6] Li Sun et all, Axial Ratio Bandwidth of a Circularly Polarized
recognize, the antenna is optimally tuned to the desired Microstrip Antenna, http //[Link]/10.5772/54664, chapter10
frequency and very 5.6GHz and sufficiently to 6.48GHz. [7] Vítečková M., Jedlička D. Static optimization system, VŠB-TUO Faculty
of Mechanical Engineering, 2013, In.
http //[Link]/StatickaOptimalizace/2fjedne/[Link]
[8] RAIDA, Z. Computer Solutions communication systems. Electronic
textbook.. Brno FEKT VUT v Brně, 2004.

Fig.12 Graph of S11 values in the frequency range from 4GHz

up to 7GHz.



View publication stats