Broad-Band Fragmented Aperture Phased Array Element Design Using Genetic Algorithms
Broad-Band Fragmented Aperture Phased Array Element Design Using Genetic Algorithms
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Index TermsBroad-band antennas, finite-difference timedomain (FDTD), genetic algorithm (GA), phased arrays.
I. INTRODUCTION
designed to operate over much larger bandwidths than an individual element. Dense arrays of wire dipoles have been analyzed [5], [6], similar arrays where the dipole arms consist of circular patches [7], [8] and wide crosses [9] have been proposed,
and recently a wide-band dipole array was presented [10]. In
most cases, however, design guide lines and results are scant or
nonexistent. Therefore, this paper explores a design approach
for a planar patch phased array element with octave bandwidth.
Genetic algorithms (GAs) have been used to optimize the
shape of patch elements, both as single elements [11] and as
phased array elements [12]. In [12], the optimization was based
on a GA in combination with a stochastic hill climbing technique and led to a fragmented aperture design. However, most
of the numerical examples given in [12] are for arrays without
a groundplane, which means that the resulting arrays radiate in
two directions. No results are given for groundplane backed arrays scanned off broadside. In this paper, a similar genetic algorithm is used to synthesize a broad-band groundplane backed
fragmented aperture array scanned within 45 from broadside.
This paper focuses on the broad-band properties of the arrays
investigated and not on the optimization method. A detailed description of the GA can be found in [13].
II. GA SYNTHESIS PROCEDURE
A key step in the GA lies in the evaluation of the fitness function. Since we are interested in designing a broad-band array,
it seems reasonable to define a fitness function that minimizes
the standing wave ratio (SWR) over a large frequency range.
This implies that the SWR must be calculated for many frequencies. Furthermore, many geometries will have to be analyzed
during the design procedure which necessitates a fast electromagnetic solver. The problem is simplified by approximating
the large array antenna with an infinite periodic array for which
a unit cell analysis suffices. For the analysis, a finite-difference
time-domain (FDTD) program with periodic boundary conditions is used [14]. The geometry of the unit cell is shown in
Fig. 1.
The design goal is to find a distribution of conducting regions
on the aperture surface, which together with a suitably chosen
permittivity and thickness of the dielectric substrate will produce a well-matched antenna over a large frequency range for
all scan directions. In order to search for the optimum distribution of conducting regions, the unit cell aperture is gridded
into 20 20 pixels where each pixel can be assigned either conducting or nonconducting properties. The element is fed by a
voltage source in the center of the unit cell and the conducting
regions are constrained to be symmetric around the feed point.
Fig. 1. Unit cell geometry for a random fragmented aperture array. (a) Side
view and (b) top view.
where
denotes the discrete sample frequencies over the dedenotes the different sample scan directions.
sign band and
Different types of weighting functions have been used in the
. The
simulations, which are represented by the coefficients
must not be made too large since the fitness
design band
function should correspond to a realistic design objective.
Fig. 2.
3281
GHz.
Apparently the metal distribution obtained is quite different
from the self-complementary geometry in Fig. 3(a). The results
for the reflection coefficients, however, are rather similar compared to the corresponding results for the self-complementary
array. We have been found that different simulations in the end
1The shape of a self-complementary planar antenna is identical to that of its
complementary antenna.
3282
Fig. 3.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Self-complementary array. (a) Unit cell geometry and (b) reflection coefficients.
Fig. 4. Design results for an array antenna without a groundplane. (a) Reflection coefficients and (b) unit cell geometry.
1) Design of Element Configuration and Substrate Permittivity: In this section, a GA is used to design the patch shape
and the dielectric substrate of a groundplane backed fragmented
aperture array. The relative permittivity of the substrate was
allowed to take 16 discrete values in the range 1.14.0, while
the corresponding conductivity always was set equal to zero.
From a broad-band perspective, the largest difference between
groundplane backed arrays and arrays without groundplane is
that the groundplane constitute a bandwidth limiting factor at
low frequencies. At high frequencies the bandwidth is limited,
for both array types, by grating lobe effects. Thus, in an attempt to favor antennas well matched in the center of the design
band, the reflection coefficients were weighted with a half-period of a sine function. The sine function was designed to go to
zero 0.5 GHz outside the bounds of the design band. It was felt
3283
Fig. 5. Design results for a thin array antenna (d = 6 mm, design B = 2:54:5 GHz, obtained BW= 1:43 : 1). (a) Reflection coefficients and (b) unit cell
geometry.
Fig. 6. Design results for an array antenna with d = 12 mm (design B = 2:05:0 GHz, obtained BW= 1:98 : 1). (a) Reflection coefficients and (b) unit cell
geometry.
that this approach would provide a better indication of achievable bandwidth than trying to force a uniform but, perhaps, poor
match over the entire band. In the previous section, where the
bandwidth was limited only for high frequencies, a uniform
weighting function was used.
Fig. 5 shows results when the metal distribution and the relative permittivity of the substrate were designed for a thin elmm
at center freement (substrate thickness
GHz.
quency) within the frequency band
As expected, the groundplane constitutes a bandwidth limiting factor at low frequencies, and, with the thin substrate considered above, a bandwidth of about 1.43:1 was obtained. As
mentioned before, the performance of the arrays is evaluated
,
for three different scan directions according to Fig. 2 (
,
). In practice, however,
it is often required that the arrays perform well also for intercardinal plane scanning. In Fig. 5(a), numerical results are also
in the diagonal plane
given with the array scanned to
. The diagonal plane is located between the cardinal
planes, and it is therefore not surprising that the dotted curve
in Fig. 5(a) is located between the cardinal plane curves. It
should be emphasized that diagonal plane scanning cannot be
considered as a linear combination of scanning in the cardinal
planes. Nevertheless, based on experience from numerous simulations, arrays that perform well for cardinal plane scanning are
likely to perform well also for diagonal plane scanning for fre-
3284
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Fig. 7. Design results for an array antenna when the substrate thickness was allowed to take values between 6.028.5 mm. (Design B = 1:55:0 GHz, obtained
BW= 2:23 : 1.) (a) Reflection coefficients and (b) unit cell geometry.
Fig. 8. Design results for an array antenna with element contact restricted. (B = 1:55:0 GHz, d = 16:5 mm, BW= 2:21 : 1). (a) Reflection coefficients and
(b) unit cell geometry.
quencies where no grating lobes can propagate. Thus, the simplification of omitting diagonal plane scanning when evaluating
the array fitness may be justified.
The reflection coefficients presented in this section were
obtained by relating the antenna impedance to a characteristic
impedance of 50 . From a broad-band perspective, however,
this value of
might not be the optimal choice. From now
onwards, the characteristic impedance is allowed to take values
between 20300 . This is accomplished by assuming a set
and computing the
of 281 evenly spaced, fixed values for
corresponding fitness according to (1) for each of these fixed
values. For each element configuration, this leads to a set of
values, from which the
fitness values with corresponding
value is chosen. Thus, the characteristic impedance is
best
Fig. 9.
3285
Design results for an array antenna with element contact restricted when the substrate thickness was allowed to take values between 6.028.5 mm. (B =
1:55:0 GHz, BW= 2:32 : 1). (a) Reflection coefficients and (b) unit cell geometry.
3286
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 53, NO. 10, OCTOBER 2005
Fig. 10. Comparison between PB-FDTD and CST-MS for the array in Fig. 6(b)
scanned to broadside. The coarse mesh corresponds to 36 cells/wavelength at
5.5 GHz while the fine mesh corresponds to 436 cells/wavelength at 5.5 GHz.
[16]
Bjrn Thors received the M.Sc. degree in engineering physics from Uppsala University, Sweden, in
1996 and the Ph.D. degree in electromagnetic theory
from the Royal Institute of Technology (KTH),
Stockholm, Sweden, in 2003.
Since 2003, he has been a Research Associate
with KTH. He spent the 2001 to 2002 academic year
with the ElectroScience Laboratory, The Ohio State
University, Columbus, as a Visiting Scholar, working
with development of high-frequency methods for
conformal antennas. From 2003 to 2005, he was
a Research Associate at KTH. He is currently with Ericsson Research,
Stockholm, Sweden, working with human exposure to radio frequency electromagnetic fields. His research interests include conformal antennas, scattering,
high-frequency techniques, and broad-band array antennas.
3287