Using Parametric Simulation

To Optimize WiMAX Antenna Performance

By Martin Andersson
Flomerics, Inc.

The traditional approach to antenna design involves a trial-and-error process consisting of

building a series of prototypes and testing their performance while iterating to an optimized
design. More recently, antenna designers have begun to simulate antennas as software
prototypes, making it possible to analyze alternative designs in a fraction of the time required by
physical prototyping. But normally this approach still follows the iterative process that was
previously used in physical prototyping: model the design, simulate its performance, make
changes to the model in an effort to improve the design, then start the process over again by
simulating the new design.

More recently, a few companies have begun moving to a new approach in which a wide range of
design parameters are evaluated in a single analysis run with the goal of exploring the entire
design space and selecting the optimized design without need for the normal iterative process.
This article will explain this method and present an example of how it was used to design the
feed network of a WiMAX array to provide coverage for the entire frequency range of interest.

Traditional Approaches To Antenna Design

The past decade has witnessed the introduction of many new wireless technologies such as
Bluetooth, WLAN, 2.5G and 3G cellular telephony, RFID, UWB, etc. Each new technology
requires innovations in antenna design to achieve their full potential, and often multiple wireless
technologies are combined, which creates further complications. For example, it’s not unusual
today for a personal computer to have one or more each of Wi-Fi, Bluetooth, and cellular
antennas in close proximity to each other. This creates, in addition to the normal antenna design
issues, a whole new range of very complex concerns caused by coupling between the antennas.

The traditional approach to designing the antenna involves a trial-and-error process consisting of
building a series of prototypes and testing their performance while iterating to an optimized
design. The biggest problem with this approach is that it typically takes about a month to design,
build, and test each prototype. A considerable number of iterations may be required to meet
design requirements and more are generally needed to optimize the design. Another problem
with this approach is that it is usually impossible to achieve the final installed environment on
the bench. So it usually becomes necessary to perform additional rounds of design iterations late
in the design cycle. This sometimes means that the product launch may have be held up for
antenna development, with the potential for losing a substantial amount of revenues or even, in a
worst-case scenario, missing the market window that the product was intended to address.

In this article, we will demonstrate a newer method in which the initial concept design is
modeled and simulated and then key design parameters are replaced by variables. The user
defines ranges for each variable, and the simulation engine creates models and performance
predictions for every possible combination of variables. The time required to optimize the design
is substantially reduced because rather than individually creating each design, the user need only
define the design space of interest and pick the best design from the alternatives created by the
parametric simulation process.

Initial Design Of Feed Network

The purpose of this project was to design a WiMAX antenna array to cover the 3.4 to 3.65 GHz
band. The wavelength is 2.998E+8/3.4E+9=8.818mm. The design strategy is to have a central
feed with equal length distribution to each patch so that the elements radiate in phase. The
network is fed at its center with a 50 coaxial probe connected to the center of a 100 line. Each
end of the line ends in a quarter-wave transformer that transforms the 100 impedance to a
segment that splits into two lines, each feeding a patch antenna element.

The first basic step in the design process is to calculate the edge impedance of the patch and
match it back through the transformer to the 50 line through the feed network. We will do this
using a formula-based transmission line calculator, but it could just as easily be done using trace
impedance formulas from basic microwave theory. Another constraint is that the four radiating
patches need to be sufficiently separated to avoid interfering with each other.

The thickness of the substrate is 1.6 mm and the of the substrate is chosen to 3.58. The next
step is calculating the edge impedance of the patches using approximate formulas. A thin half-
wavelength patch has the corrected side length of:

λ
L = 0.49 ⋅ = 22.18mm
ετ
All trace impedances must be matched to the coaxial probe feed so there is no need to do an
insert feed at the element. The decision was made to make the patch 25 mm wide, which means
the approximate edge impedance can be calculated to:

2
ετ L
Z edge = 90 ⋅ ⋅ ≈ 100Ω
ετ − 1 W

A simple RF calculator is used to calculate the width of a 100 feed on the desired substrate:

W100 = 0.852mm

With the edge impedance known, the other impedances and micro-strip widths can now be
calculated. Two 100 patches will be connected from above and the other two from below to the
feed point. Each connection trace segment must have the impedance:

100
Z= = 50Ω
2

The micro-strip widths are:

W50 = 3.497 mm

In addition, a quarter-wave transformer will be used to connect the 50 segments at each of the
100 line.

Z t = 100 * 50 = 70.07Ω

W70 = 1.96mm

Lt = 11.9mm

The resulting feed network is shown in Figure 1.

 Figure 1: Initial feed network design based on approximative formulas.

Evaluating Performance Of Initial Design

The next step is to evaluate the performance of this initial design. Rather than taking the time
that would be required to build a prototype, we will simulate the antenna as a software prototype
using Flomerics MicroStripes software. This software package uses the transmission line matrix
(TLM) method for solving Maxwell' s equations in the time domain. MicroStripes solves for all
frequencies of interest in a single calculation and therefore captures the full broadband response
of the system in one simulation cycle. The TLM method creates a matrix of equivalent
transmission lines and solves for voltage and current on these lines directly. This approach uses
less memory and CPU time than solving for E and H fields on a conventional computational
grid.

The initial antenna design was evaluated by using the simulation program’s ACIS-based modeler
to construct the WiMAX antenna geometry from primitive shapes. In addition to the feed
network described above, this involved creating patches at the end of each trace and a substrate
and a ground plane with a size of about 110 by 100 mm, chosen for clearance to the edge of the
patches, in order to reduce side lobes. The complete design is shown in Figure 2. The
computational domain was expanded by 30% of the model’s largest dimension to place the
external absorbing boundaries in the farfield region. The simulation is then equal to a
measurement of e and h fields in an anechoic chamber. Then the software automatically
generated the mesh, snapped it to the geometry, and refined it around edges in curved areas and
dielectric regions.
 Figure 2: Complete model of antenna design with patches, substrate, and ground plane surrounded by 30% extra
space. The software makes use of the bilateral symmetry of the design and only needs to compute half the volume.

A common problem with time-domain simulation is that the fine cells bleed out to the boundary
of the computational domain. This greatly increases the number of cells in the mesh and leads to
large memory consumption and long computational times. The TLM software, however, has an
octree sub-grid meshing algorithm that progressively and automatically lumps together
computational cells in regions remote from the geometric detail. The software’s multi-grid
meshing capabilities enabled fine cells to be localized to the space occupied by the antenna,
while the surrounding free space region was modeled using a coarser mesh. The ultimate size of
the lumped cells is limited only by the local permittivity, permeability, and highest frequency of
interest. This enables critical but electrically small detail to be captured with an exceedingly high
resolution mesh, without having a significant impact on the global cell count. The initial design
had 801,600 cells reduced to 71,313 by the octree algorithm. The lumped cells are visualized as
yellow regions in Figure 3.
 Figure 3: Model with lumped cells highlighted.

The simulation was excited with a broadband Gaussian pulse injected into the coaxial antenna
port, and the time signature was captured by stepping through time. Fourier-transformation of
this response yielded frequency-domain results across the entire band of the antenna. The results
are shown in Figure 4. The plot in the upper left-hand corner shows the gain on a 3-D diagram
while the graph in the upper right corner shows the gain over a single cross-section of the 3-D
plot. The plot in the lower left shows the surface current plotted on the conducting bodies and the
electric field on a plane near one edge of the patches. Finally, the chart on the lower right shows
the return loss of the antenna plotted against frequency. The return loss plot illustrates that that
rather than meeting the goal of operating over the entire WiMAX frequency range, the initial
design is actually effective only over a small segment of the band. The return loss is lower than
-6 dB only over the range from 3.38 GHz to 3.48 GHz.

Figure 4: Simulation results for initial design.

Using Parametric Simulation To Improve Frequency Response

The first attempt to improve the frequency response of the antenna was simply to round off the
sharp corners of the lines leading to the patches to reduce unwanted reflections. This provided a
minor improvement but not nearly enough to meet the design objective. The next step was
adding another section to the transformers, which reduces the mismatch that each transformer
corrects and thus provides broader band coverage. A formula-based calculator for multisection
transformer design was used to provide the initial dimensions for a new transformer. In the initial
broadband design, simulating the complete antenna, the simulation results showed the antenna
provided two separate frequency bands at opposite ends of the WiMAX band as plotted in
Figure 5.

Figure 5: The initial simulation showed two frequency bands.

Clearly, the design of the new feeding network needed to be better matched to the input
impedance and resonance frequency of the current patch element design or opposite. Using
conventional simulation methods, this would involve a trial-and-error process that might involve
changing the lengths, widths, and angles of the transformer sections and patch elements until the
engineer was satisfied with the results. This approach is considerably more efficient than
building and testing prototypes, but it still takes a considerable amount of time to construct each
model and wait for the simulation results. With multiple design parameters to consider, the
number of simulations required to fully interrogate the design space grows geometrically. For
example, investigating four different widths, four different lengths, and four different angles of
all segments in this feed network would require a total of 4,096 different simulation runs.
In this application, the design process was streamlined by taking advantage of a software feature
that allows users to substitute variables for design parameters. Users model their concept design,
identify geometric entities as variables, pick upper and lower limits for each variable, and select
a step size. The software then generates as many simulation iterations as are required to
completely explore the design space defined by the user. The results of each simulation are
plotted on a single graph, so users can quickly determine which design parameter values provide
optimal performance. In this case, the engineer varied the lengths of the two different
transformer sections and tried three different values for each variable. The software generated a
design for each combination of variables and produced the frequency domain results for each
design.

When the simulation runs were concluded, it was easy to evaluate the results to compare the
performance of the different designs. The best-case feed network design shown in Figure 6 later
provided gain and return loss values shown in Figure 7 when combined with patch antennas that
were tuned to the desired center frequency by also using the variable sweep functionality. At this
stage, the return-loss values for the entire WiMAX band were safely below the -6 dB
requirements. The simulation results also indicated the potential that additional parametric design
iterations of the lines connected to the patches might be able to provide further improvements.
The results in Figure 8 below show that considerable side lobes exist in the latest design iteration
but not in the first attempt’s narrow band antenna. A reasonable approach to eliminate these side
lobes would consist of setting up another series of parametric simulations that involve changing
variables such as the size of the ground plane and the distance between the upper and lower pair
of patches.

Figure 6: The final feed and transformer design as optimized with parametric simulation.
 Figure 7: Return-loss and gain predictions for final design.

Figure 8: Side lobes in final design could be reduced further through additional simulation iterations.

The example demonstrates how simulation can help engineers improve WiMAX antenna
performance by evaluating many more potential designs than would be possible using traditional
bench methods. Simulation has the potential to improve the reliability of antennas by making it
possible to evaluate a wide range of potential configurations prior to installation in order to
evaluate and optimize system performance inexpensively without disrupting operations. The
latest advance enables a series of simulations to be run automatically while varying one or more
design parameters over a user-specified range. This feature speeds up the design process by
making it possible to, for example, quickly consider a wide range of locations and determine the
ideal feed network dimensions.