05 Chapter 6
description
Transcript of 05 Chapter 6
-
Chapter 6
The ESPAR Antenna
6.1 Introduction
The E lectronically S teerable P assive Array Radiator, or ESPAR antenna is a
loaded N port array currently under development at the Advanced Telecommu
nications Research Institute (ATR) of Kyoto, Japan [11]. In its simplest form,
a single active element is surrounded by multiple parasitic elements loaded with
variable reactances. The controlled reactances regulate current on the parasitic
elements, inuencing spatial radiation sensitivity and thus providing the adaptive
trait.
Specically, the ESPAR antenna is being developed for application in wire
less ad hoc networks [96]. Autonomous to cabling and base-station infrastructure,
wireless ad hoc networks are a cheap, simple and dynamic alternative to their
wired, static counter-parts. However, wireless systems are susceptible to signal er
rors arising from multi-path propagation and interference signals from unsolicited
nodes. As data is multi-hopped, signal errors between just a few intermediate
nodes can have a potentially devastating eect on the entire network performance.
In addition, wireless transmission can be a signicant drain on the nodes stored
energy. Node battery life is shortened considerably if high transmission powers
are required for ecient communication. As the ESPAR antenna can direct radia
tion at intended recipients and steer radiation nulls toward the interfering signals,
91
-
92 CHAPTER 6. THE ESPAR ANTENNA
such problems are signicantly reduced. The radiation nulls and main lobe gains
complement each other to maximize the system signal to interference noise ra
tio. Furthermore, the antenna main lobe gain dramatically decreases the required
transmission power for a set range.
The acronym ESPAR allows an arbitrary radiating structure (patch, wire, aper
ture etc.), and is thus synonymous with the N port systems described by Harrington
and Mautz [26, 20, 27, 28]. Their early analysis on radar scattering of generalized
N port loaded networks lead to the formation of a reactively loaded dipole array
[12]. This generated research into equivalent patch and monopole arrays by Dinger
[29, 30, 23], Sibille [31, 32] and Ohira [33, 34].
This existing literature primarily focuses on general loaded N port network an
tenna theory. To a lesser extent, some experimental results of prototype antennas
have been presented. In contrast, the ESPAR antenna is consciously being devel
oped for mass consumer application. Therefore, with general theory understood,
questions regarding the ESPAR antennas practicality for such application can be
addressed. As such, the Adaptive Communications Research Laboratory (ACR)
of ATR is actively involved in the research of practical beam forming algorithms
[35, 97], and, in the authors case, the design of ecient, robust antenna structures
[73, 98].
An ad hoc network would generally comprise mobile terminals (e.g. notebook
computers). It is envisioned the ESPAR antenna will be external to the mobile ter
minal, with design evolution eventually realizing an integrated solution. Regardless
of its position however, the antenna design must inherit similar criteria to the mo
bile terminal. Of particular signicance are the restrictions of power consumption
and physical volume. The volume criterion limits the ground plane, disallowing
Harringtons analytical procedure of the dipole array in [12]. Chapter 5 described
the analytical diculties in dealing with monopole arrays on nite ground struc
tures, leaving simulation the only viable option (excluding experiment) to predict
antenna characteristics.
Such an antenna has hitherto not been considered. It is unknown whether its
-
6.2. THE PHYSICAL ESPAR ANTENNA 93
performance will allow useful communication. For instance, if the ESPAR antenna
has high frequency sensitivity then only small communication bandwidths will be
possible. Therefore, a base ESPAR structure was analyzed through simulation.
Both the antenna structure and loading reactances eect electrical response in the
frequency domain. Accordingly, the antenna frequency sensitivity was measured
to ensure utility. A robust frequency response will additionally translate to sim
ilarly stout electrical characteristics when reactance manufacturing and assembly
tolerances are included.
Once the ESPAR structure was deemed practicable, it was optimized for max
imum azimuthal gain. Consequently, radiation sensitivity at side angles, required
transmitter power and antenna impedance mismatch were inherently reduced. In
creases in principal lobe gain are even more pronounced in a homogeneous ad hoc
network. As both communicating node antenna responses combine to dene system
performance, any design improvement made in the antenna gain will potentially
have double the eect in a system.
This chapter commences with an introduction to the ESPAR antenna structure.
Practical loading methods and considerations are then detailed before presenting
the antenna frequency analysis and optimization procedures.
6.2 The physical ESPAR antenna
The antenna conguration is similar to that of Harringtons dipole array [12].
A single active monopole element (0) is surrounded by six equidistant, parasitic,
monopole elements (1-6) of constant radius from the centre. Each parasitic element
is base loaded with some variable reactance, while the entire array ideally rests
upon an innite ground expanse. The antenna is illustrated in Figure 6.1.
The reactive loads alter the antenna currents through equation (2.15), inuenc
ing its radiation characteristics (2.10). Thus control of the reactive loading allows
radiation beam and null steering. Capacitive varactor diodes have been typically
suggested for the reactances [34]. By applying a reverse dc bias, a depletion region
-
94 CHAPTER 6. THE ESPAR ANTENNA
X2 X3
X5
X4 X6
X1 V1
2 3
4 56
0 RF
Figure 6.1: Ideal monopole ESPAR antenna upon innite ground.
is formed over the diodes PN junction. The depletion region size is determined
by the bias magnitude, and hence the capacitance is controlled. With an applied
reverse bias Vr , capacitance of a PN junction is [16]
C = kV r 1 2 (6.1)
where k is a constant derived from the junctions doping charge densities. The only
additional circuitry on the parasitic elements are RF chokes to isolate the parasitic
elements microwave signal from the dc control lines. The cost savings of passive
components like varactors and RF chokes (inductors or resistors) are apparent
when compared to the active phased array alternatives. A typical varactor control
circuit is presented in Figure 6.2.
The monopole array of Figure 6.1 obviously derives from the complementary
free space dipole array. However, the monopole arrays ground allows shielding
(electrically and physically) of the control circuitry which can reside below the
plane. It is for this very reason that the original dipole ESPAR antenna [12] has
been reduced to its monopole equivalent. Practically however, the antenna cannot
utilize such a ground. For integration in mobile terminals, the horizontal ground
needs to be strictly limited. Chapter 5 discusses the implications of reducing
the ground size on a monopole array. Typically, radiation is redirected above
-
95 6.2. THE PHYSICAL ESPAR ANTENNA
l l
itic le
DC ControVo tage
Parasmonopo
RF Choke
RF Choke
Varactor
Figure 6.2: Typical varactor control circuit.
the horizon, reducing communication eciency in the azimuth. A solution in
the form of a conductive sleeve or skirt on the perimeter of the lateral ground
is suggested to compensate for the nite ground eects. Figure 6.3 depicts the
practical array while Table 6.1 denes ground radius rg , parasitic element radius
rp, active element height ha, parasitic element height hp and skirt height hs. The
dimensions of Table 6.1 represent those that might be used as an initial 2.5GHz
design. The quarter wavelength monopole heights ensure resonance while their
close spacing guarantees strong mutual coupling. In addition the array is bounded
by a conductive skirt to accommodate a small lateral ground area. When measured
however, it was found resonance of the antenna was at 2.4GHz and therefore all
results and dimensions herein are given for 2.4GHz.
Length Structural Dimensions
rg rp ha hp hs mm 60 30 30 30 30 0.48 0.224 0.224 0.224 0.24
Table 6.1: Initial dimensions of the ESPAR antenna in Figure 6.3 at 2.4GHz.
To obtain a specic radiation pattern, all reactive loads need to be set to some
-
X4X5
96 CHAPTER 6. THE ESPAR ANTENNA
1
X3
V
1
2
1
2
1
2
4 56
3
ha
hp
rp
hs
rg
RF
=90, =0
Ground Skirt
=0
Figure 6.3: Practical ESPAR antenna structure with conductive skirt.
value. The term loading case or loading configuration shall be used henceforth to
describe the set of reactive loads used simultaneously on the antenna.
6.3 Reactance considerations
Employing varactors as the reactive loads raises questions regarding their practi
cality. Firstly, it is important to consider their transmission capabilities. Harmonic
distortion would potentially limit the transmission power they could accommodate.
Antenna elements typically display resonance at higher order modes resulting in
spectrum pollution. Secondly, the reactance range varactors oer can restrict the
adaptive possibilities. As variable capacitors they would produce a limited negative
reactance range.
-
97 6.3. REACTANCE CONSIDERATIONS
6.3.1 Harmonic distortion
Surface mount varactors are not necessarily designed for high power application.
In transmitting mode, the antenna may drive the varactors into non-linearity, cre
ating harmonic distortion. With no lters between the varactors and transmission
medium to counter the eect, unlicensed spectrum will be polluted. A suggested
solution is to distribute the power over multiple varactors [98]. An anti-series var
actor pair conguration is shown in Figure 6.4 with the single varactor alternative.
C
CC
CC
(a) (b)
Figure 6.4: (a) Single varactor and its (b) anti-series varactor pair equivalent.
The varactor pairs create a parallel capacitance twice that of the individual.
In series with an equivalent pair, the total capacitance reduces back to that of the
single varactor equivalent. However, the RF power is divided over the 4 varactors,
exposing each to less and therefore mitigating the harmonic production relative to
the single varactor case. It is analytically shown in [98] that the anti-series varactor
pairs can extinguish second order distortion and boast a third order suppression
ratio (compared to the single varactor) of
HD3 ASV P HD3 SV
= 1
4 3C
2 1
8C0C2 (6.2)
where HD3 ASV P and HD3 SV are the third order harmonic distortion magnitudes of
of the anti-series and single varactor congurations. C0, C1 and C2 are coecients
of the Taylor series expanded, non-linear total capacitance that is a function of
-
98 CHAPTER 6. THE ESPAR ANTENNA
input ac signal v [99]
C = C0 + C1v + C2v 2 . (6.3)
Under varying bias conditions, it was experimentally found [98] that replacing
a single varactor with four can reduce second and third order harmonic distortion
by a worst case average of 18dB and 12dB respectively. Considering the varac
tor under test exhibited a maximum second order harmonic magnitude of -38dBc,
these reductions are not trivial. The measurements were carried out in a compact
anechoic box designed for convenient near-eld measurement [100, 101]. The au-
thors contribution to this research ([100, 101, 98]) was only 10-20% and therefore
extensive results will not be shown.
6.3.2 Reactance range
The reactance of varactors placed at the base of monopole elements will translate
directly to the base reactance of the monopole. Being purely capacitive loads, only
negative reactances can be generated as reactance is dened from capacitance C
jX = 1
jC = j
C 1
X = C
. (6.4)
Varactors considered for prototype ESPAR antennas exhibit capacitive ranges of
0.7pF to 9pF, which from (6.4) at 2.4GHz is a j6.9 to j91.5 reactance range. Therefore, the beam shape possibilities are potentially limited. To improve the
reactance range, and hence the beam-forming ability of the antenna, transmission
lines would have to be included. This is illustrated with fundamental transmission
line theory.
A transmission line of length l and characteristic impedance Z0 terminated with
a lumped impedance ZL will have an input impedance [80]
ZL + Z0 tanh l Zi = Z0 (6.5)
Z0 + ZL tanh l
-
99 6.3. REACTANCE CONSIDERATIONS
where the propagation constant of the line contained wave
= + j (6.6)
comprises real and imaginary components and representing the attenuation
constant (Np/m) and phase constant (rad/m) respectively.
More specically, if we assume a lossless transmission line (=0, Z0 = Re(Z0) = R0) and utilize the trigonometric relation
tanh jx = j tan x, (6.7)
Zthen (6.5) becomes
L + jR0 tan l Zi = R0 . (6.8)
R0 + jZL tan l
Lossless varactors will produce a purely imaginary load impedance (ZL = jXL),
so (6.8) can be written
jXL + jR0 tan l Zi = R0
R0 XL tan l ( XL + R0 tan l
) Zi = j R0 . (6.9)
R0 XL tan l
Equation (6.9) states a reactively loaded transmission line will have a purely re
active input impedance whose magnitude depends on the line length l. In other
words, a varactor loaded transmission line will create a load at the monopole base
with dierent reactance range to that of just the varactor.
Multiple loads can then extend the possible reactance range. As an example,
consider three parallel varactor loaded 50 transmission lines meeting at point A
in Figure 6.5, where A can be connected to the monopole base. Each varactor has
a reactance range of j0 to j100. Inserting this range into (6.9) with three transmission line lengths of l1 = 0, l2 = 0.176, and l3 = 0.352 give normalized
-
100 CHAPTER 6. THE ESPAR ANTENNA
345
3476
3
48
9;:
FHG
FJI
FJK
:
=@? ABC
-
101 6.3. REACTANCE CONSIDERATIONS
grounded lumped loads (ZL = RL + jXL). Signal induced in the parasitic element
is guided down the transmission line and reected with some phase shift. The total
phase shift (at the monopole base) directly inuences the currents on the monopole
and is determined by the reactive load ZL, the transmission lines characteristic
impedance (Z0) and the transmission lines length. Elementary transmission line
theory states that the reection coecient () of a loaded transmission line is [102]
Z = r + ji =
ZL Z0 (6.10)
L + Z0
where subscripts r and i designate the real and imaginary components of the re
ection coecient. Normalizing with respect to Z0 and assuming no loss in the
transmission line or load (RL = 0 = 1), (6.10) can be rearranged to nd | | the angular component of the reection
i 2xL = arctan = arctan 2 . (6.11)r xL 1
Here xL is the load reactance normalized to the transmission line characteristic
impedance. This relationship is well known and best visualized on a Smith Chart.
When xL increases, its inuence on reduces (d/dxL is small). Therefore, for | | large values of xL , small perturbations will result in minimal phase dierence and | |hence an insignicant eect on the antenna. Conversely, for small xL, equivalent
perturbations will see signicant changes in and hence antenna response.
Equation (6.11) is important when considering the frequency response of a
lumped element (L or C) load. The capacitor reactance relationship (6.4) and
inductor reactance relationship
jXL = jL (6.12)
are linear. Any change in frequency will produce a corresponding change in reac
tance (though inverted for C). If we consider an example operating bandwidth of
10%, all reactances in the loading conguration will vary by 5% of the resonant
-
102 CHAPTER 6. THE ESPAR ANTENNA
frequency. If the loading case comprises capacitive and inductive loads, capacitive
reactances will decrease as frequency increases over the range while inductive reac
tances increase. The antenna response over frequency would normally be solely a
function of structure. However, with the controlling reactances also being frequency
dependent, radiation pattern and antenna match might vary beyond practicality
over a required frequency bandwidth. This eect will be seen to a greater extent
for smaller reactive loads, where changes in XL produce a greater change in .
The second consequence of (6.11) relates to reactance optimization. To real
ize a radiation pattern with particular characteristics (null, lobe locations), the
reactances must be optimized. Therefore, s relationship with reactance (6.11)
inuences our optimization procedure. Clearly, a ne, linear reactance resolution
for large |XL| will be an unnecessary burden on the optimizer. Larger stepping to alleviate this would adversely eect the resolution for small XL values where | | changes in are signicant. In addition, the theoretical reactance range is un
bounded (j j), so a standard search could not include all possibilities. It is consequently more logical to step linearly with respect to . The reactances
can then be calculated from rearranging (6.11):
sin XL = Z0. (6.13)
1 cos
As is naturally bounded between 0 and 2pi, a small, nite set of reactances can
be chosen which represent all possible reactance values that would signicantly
change the characteristics of the antenna.
6.4 Frequency sensitivity
The electrical sensitivity of the antenna is an important measure gauging its prac
ticality. Typically, an antenna will be required to operate linearly over some
frequency bandwidth. While the antenna structure would normally dictate the
electrical sensitivity around resonance, the ESPAR antennas reactive loads bring
additional frequency dependence. Section 6.3.3 discussed how changes in frequency
-
103 6.4. FREQUENCY SENSITIVITY
can cause the reactance to eect the antenna in a non-linear fashion (6.11). There
fore it is important to conrm that the characteristics of the antenna will not
degrade beyond practical limits.
It would be impossible to test every reactance combination, most of which
would not produce practical antenna characteristics anyway. Therefore a sample
of loading cases that produced potentially useful radiation patterns were tested over
a frequency bandwidth. However, in order to nd the loads required to generate
these patterns an optimization of the reactance values needed to be performed.
6.4.1 Reactive optimization
Optimization was required to determine a set of load values that would produce
directional radiation patterns and acceptable reection responses. Computer sim
ulation determined antenna electrical characteristics for a given load conguration.
The three packages available were XFDTD v5.1, HFSS v7.0 and NEC. Of the three,
XFDTD and HFSS were the most accurate with their capability to simulate the
skirted ground plane. However, these packages had simulation times in excess of
1.5 hours. In comparison, NEC considered the antenna on innite ground, but
its simulation term was only several seconds. This made it the ideal choice for
optimization purposes.
The genetic algorithm discussed in Chapter 3 employed NEC as the solver from
which a tness value could be determined. The cost function tness included min
imizing the antenna S11 and maximizing front to null (radiation minimum) ratio
of a = 0 30 primary lobe and = 90 270 null. Antenna symmetry allowed such patterns to be repeated with 60 steps through the horizontal.
Mathematically, the tness was calculated using
f = FNRmn S11. [dB] (6.14)
where FNRmn is the front to null ratio between a main lobe directed along m
and null at n. The negative sign for S11 enabled minimization of the negative
-
104 CHAPTER 6. THE ESPAR ANTENNA
Case 1
Load Reactance (j) 2 3 4 5 6
Main lobe angle ()
Null angle ()
S11 (dB)
-10dB BW (MHz)
1 -150 250 78 26 78 250 0 180 -9.1 -2 184 -97 89 46 89 -97 0 180 -12.4 217 3 -183 226 103 81 59 125 15 245 -12.0 206 4 212 -156 87 56 85 144 30 170 -12.3 217 5 -246 -225 179 43 87 125 30 210 -11.9 201
Table 6.2: Five loading congurations with dierent H-plane radiation characteristics (main lobe and null locations). Reection summary is calculated with XFDTD.
logarithmic value in a maximizing function.
Five dierent loading cases for a 2.4GHz ESPAR antenna were optimized. Each
case represented dierent radiation characteristics that might be desirable in the
nal design. The main goal of the optimization was simply to nd loading con
gurations that produced dierent, directional radiation patterns. Whether they
were completely optimal with respect to (6.14) was not critical. Table 6.2 describes
the dierent loading cases, their corresponding principal lobe and null directions,
and reection responses. While cases 1 and 2 appear equivalent, the main lobe
beamwidth of case 2 was considerably larger than case 1 and was thus included.
The S11 results in Table 6.2 were computed with XFDTD so the nite ground
structure could be accounted for and are shown in Figure 6.6. The resonant depths
range from -9.1dB to -12.4dB. Those under -10dB manage an impedance bandwidth
of better than 200MHz (8.3%). The resonant frequencies ranged between 2.355GHz
and 2.402GHz. This 47MHz change represents a 2.2% shift for the dierent reactive
loading cases, although each S11 did not degrade by more than 0.52dB over this
range. With such wide impedance bandwidths, it was concluded that the antenna
reection response was practical.
XFDTD, HFSS and NEC were used to calculate and compare radiation patterns
of the dierent cases at 2.4GHz. Both E-plane and H-plane cases were considered
and are presented in Figures 6.7 and 6.8. As NEC simulated the structure on an
innite ground plane, it was expected some error between it and the remaining
-
105 6.4. FREQUENCY SENSITIVITY
2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3
12
10
8
6
4
2
0
Frequency [GHz]
S11
[dB]
Case 1 Case 2 Case 3 Case 4 Case 5
Figure 6.6: Reection response of the loading congurations from Table 6.2 calculated with XFDTD.
packages would be present.
Excellent agreement between XFDTD and HFSS was observed for all cases, the
independence of the packages suggesting the results shown were a likely prediction
of the experimental case. As expected, NEC results showed the most dierence,
however lobe and null locations remained constant and thus its use for optimizing
the beam shapes was validated. The ground limitation of NEC is apparent in
Figure 6.8 where maximum radiation is not elevated. Conversely, both XFDTD and
HFSS predict elevation. Monopoles on nite ground structures typically exhibit
elevated radiation (see Chapter 5). While ground sleeves in Chapter 5 were shown
to control this elevation, the hs dimension in Table 6.1 was merely an initial guess
and not optimized.
6.4.2 Experimental verification
To ensure the simulation procedures were correct, the antenna and loading cases
were physically built and measured. An existing solid metal ground structure was
-
106 CHAPTER 6. THE ESPAR ANTENNA
90 90
0dB
3
6
9
12 0
30
60120
150
180
210
240
270 300
330
HFSS Gain: 5.84dBi NEC Gain: 10.71dBi XFDTD Gain: 6.14dBi
0dB
3
6
9
12 0
30
60120
150
180
210
240
270 300
330
HFSS Gain: 4.7dBi NEC Gain: 9.66dBi XFDTD Gain: 4.69dBi
90 90
0dB
3
6
9
12 0
30
60120
150
180
210
240
270 300
330
HFSS Gain: 4.46dBi NEC Gain: 9.54dBi XFDTD Gain: 4.62dBi
0dB
3
6
9
12 0
30
60120
150
180
210
240
270 300
330
HFSS Gain: 4.56dBi NEC Gain: 9.61dBi XFDTD Gain: 4.7dBi
90
0dB
3
6
9
12 0
30
60120
150
180
210
240
270 300
330
HFSS Gain: 5.02dBi NEC Gain: 10.27dBi XFDTD Gain: 5.31dBi
Figure 6.7: Normalized H-plane radiation pattern comparison between HFSS, XFDTD and NEC. Cases 1 to 5 from top left to right to bottom (refer to Table 6.2).
-
107 6.4. FREQUENCY SENSITIVITY
0 0
0dB
6
12
18
24
30
60
90
120
150
180 210
240
270
300
330
HFSS Gain: 6.69dBi NEC Gain: 10.71dBi XFDTD Gain: 6.76dBi
0dB
6
12
18
24
30
60
90
120
150
180 210
240
270
300
330
HFSS Gain: 5.76dBi NEC Gain: 9.66dBi XFDTD Gain: 5.59dBi
0 0
0dB
6
12
18
24
30
60
90
120
150
180 210
240
270
300
330
HFSS Gain: 5.32dBi NEC Gain: 8.61dBi XFDTD Gain: 5.02dBi
0dB
6
12
18
24
30
60
90
120
150
180 210
240
270
300
330
HFSS Gain: 5.94dBi NEC Gain: 9.41dBi XFDTD Gain: 5.79dBi
0
0dB
6
12
18
24
30
60
90
120
150
180 210
240
270
300
330
HFSS Gain: 5.57dBi NEC Gain: 9.41dBi XFDTD Gain: 5.46dBi
Figure 6.8: Normalized E-plane radiation pattern comparison between HFSS, XFDTD and NEC. Cases 1 to 5 from top left to right to bottom (refer to Table 6.2). Case 1,2: = 0, case 3: = 15, case 4,5: = 30.
-
108 CHAPTER 6. THE ESPAR ANTENNA
employed. Each monopole element was fed with an SMA panel mount connector,
axed to the bottom of the ground plane. While varactor diodes would typically
be used to produce electrically controlled reactances, their use was inappropriate
for conrming the simulated results. To do so would initially require circuitry to
be designed for varactor mounting and control. Moreover, varactors would require
extensive calibration to conrm the exact reactance they were creating at the
monopole base. Even then, any production variation between units, or asymmetry
in board assembly would potentially degrade results.
Instead, combinations of SMA adapters terminated with either open circuits,
short circuits or surface mount capacitors were used to create the dierent reactive
loads. Existing connectors were measured and all permutations were calculated
from which the case reactances could be selected. Figure 6.9 shows an example
load. At point B, the exact reection phase length (BLoad ) could be measured
with a network analyzer. When added to the ground and panel connector thickness
(AB), the precise reactance at A could be calculated from (6.13). Reactances
were created to within 2 of the required . Detailed information on the load
construction is given in Appendix B. With this procedure, a test antenna with
reactive loads was constructed at a cost incorporating only the panel mount SMA
connectors used as physical mounts for the loading capacitors (see Figure 6.9).
This loading conguration however has a signicant section of transmission line.
A transmission line of physical length L has a of
2L 4pif L = 2pi = radians (6.15)
u
where is the electromagnetic wavelength in the transmission line at frequency
f propagating with velocity u. Thus, the transmission lines change with fre
quency is d 4piL
= radians Hz1 . (6.16)df u
Equation (6.16) shows the change in with respect to frequency is not constant
when L is variable (as is the case with dierent reactances). Therefore, as each
-
109 6.4. FREQUENCY SENSITIVITY
i
itic le
Ri le le
l l
( ialB
A Ground Cross-Sect on
ParasMonopo
ght AngSMA MaAdapter
Pane mount SMA fema e connector part cross-sect
Surface Mount Capac
ions)
itor
Total 50 transmission line length (L)
Figure 6.9: Example SMA loading. The monopole is loaded by a transmission line that provides a reactance at A.
reactance changes at a dierent rate with frequency, the experimental results can
only be compared to their simulated counterparts at the reactance design frequency.
This however, is sucient to validate simulation procedure.
The loss of the SMA loading structures was no greater than 0.3dB. Larger
structures incorporating electrically long phase shifters had losses in excess of 2dB
and as such, the results they produced had signicant error. It was additionally
hoped the relatively low prole SMA loads would have negligible eect on the
antenna radiation characteristics. Figure 6.10 displays the underside of the antenna
when fully loaded. The centre SMA mount would be connected to the RF signal
in operation. The reactances in Figure 6.10 are typical of the loading cases where
some loads are over twice as long as others. Clearly then, (6.16) is signicant
and large experimental frequency sweeps of the antenna would produce misleading
results.
All ve loading cases were constructed and antenna S11, H-plane radiation
patterns, and primary lobe gains were measured for comparison with simulation.
-
110 CHAPTER 6. THE ESPAR ANTENNA
Figure 6.10: Photo of an example loading conguration underneath the ESPAR antenna. Loads of dierent length comprising various SMA adapters are seen. Capacitors terminating the panel mount SMA connectors provide additional phase reection diversity.
-
111 6.4. FREQUENCY SENSITIVITY
S
Measurement procedure is given in Appendix A. Due to (6.16), experimental
11 was only measured at the designed antenna resonance (2.4GHz). All cases
were within 1dB of XFDTDs prediction (Figure 6.6) ranging from -9.7dB to -
12.8dB. The measured radiation patterns and primary lobe gains further veried
simulation accuracy with excellent agreement. Figure 6.11 displays a comparison
of H-plane simulation with experiment. Primary lobe gains were within 1dB of
those predicted and pattern shape was consistent. The similarity of simulation
and experiment simultaneously validates the simulation procedure, and the use of
SMA transmission line loads for accurate reactance creation.
6.4.3 Frequency variation
With condence in the simulation procedure, XFDTD and HFSS were used to
vary the operating frequency of the antenna from 2.3GHz to 2.5GHz (8.3%). Both
packages use lumped circuit elements for the reactive load so similarity with a
varactor response can be expected. From Figure 6.6, the worst reection response
over this range was case 1 at -7.5dB. The remaining cases did not rise above -
9.6dB. The variation in H-plane and E-plane radiation for case 1 can be seen in
Figure 6.12. While Figure 6.12 displays some variation in ancillary lobes, the
main lobe beamwidth, gain and position remain relatively constant. Table 6.3
summarizes these results and those for the remaining cases.
Case Gain ( = 0) (dB) FNR (dB) 3dB Beamwidth () 1 0.02 -3.80 0 2 0.23 -0.49 0 3 0.44 -1.01 +10 4 0.42 -1.20 +5 5 0.32 -1.29 +20
Table 6.3: Variation in ESPAR radiation characteristics from 2.3GHz to 2.5GHz. Results shown are averaged between XFDTD and HFSS.
Critical characteristics of the radiation such as gain, front to null ratio and 3dB
beamwidth varied at worst by 0.44dB, -3.8dB and 20 respectively. As a result, it
-
112 CHAPTER 6. THE ESPAR ANTENNA
Figure 6.11: Comparison between measured (green), HFSS (red) and XFDTD (blue) H-plane radiation patterns. Cases 1 to 5 from top left to right to bottom (refer to Table 6.2).
-
113 6.4. FREQUENCY SENSITIVITY
90 90
7dBi
2
3
8
13 0
30
60120
150
180
210
240 300
330
7dBi
2
3
8
13 0
30
60120
150
180
210
240 300
330
270 270
(a) (b) 0 0
7dBi
2
3
8
13
30
60
90
120
150210
240
270
300
330
7dBi
2
3
8
13
30
60
90
120
150210
240
270
300
330
180 180
(c) (d)
Figure 6.12: Case 1 H-plane patterns at (a) 2.3GHz and (b) 2.5GHz and E-plane patterns at (c)2.3GHz and (d) 2.5GHz. Results for both HFSS (red) and XFDTD (blue) shown.
-
114 CHAPTER 6. THE ESPAR ANTENNA
was concluded that the radiation characteristics remained relatively constant over
the frequency span.
While only a small sample of loading congurations were considered, the fact
that none had signicant variation over the 8.3% frequency bandwidth is a good in
dication that other congurations will be similarly robust. Signicantly, the sample
loading congurations are relatively important cases due to their practical radia
tion features. The frequency insensitivity also suggests the antenna characteristics
would be equally robust to small assembly related reactance errors.
6.5 ESPAR optimization
The analysis performed in the previous section was critical to the condence in
a practical implementation of the ESPAR antenna. However, consideration of
the ve loading conguration results concluded the antenna dimensions were not
optimal. E-plane radiation in Figure 6.8 displayed elevations of up to 25 resulting
in a gain reduction at the horizontal of 1.55dB. In addition, it is not known what
the true potential for main lobe gain is. The gain of an antenna is a dening feature
in the performance of an ad hoc wireless network. Thus the optimization goal was
simply to maximize directional gain in a single azimuth direction. Antenna gain is
dened as the product of its directivity D and eciency e [9]
G = eD. (6.17)
Conductor ohmic losses are considered negligible, thus only the antenna impedance
mismatch is considered in the eciency. Equation (6.17) can then be rewritten [9]:
( ) G = 1 S11|2 D (6.18)|
which shows optimizing the gain simultaneously optimizes the return loss and
antenna directivity.
Considering the ESPAR conguration, there are two main areas requiring opti
-
115 6.5. ESPAR OPTIMIZATION
mization; the structural dimensions and reactive loads. Both structure and loading
reactances couple to determine the radiation characteristics of the antenna, and
as such, an optimum set of reactance values is unique to a single structure only.
Therefore, it was desirable to optimize both reactance values and structure. As
this presented a very large solution space, the optimization was partitioned into
several consecutive stages that alternated optimization of reactive loads and struc
ture. At each successive stage, optimization variables underwent an increase in
resolution with a decrease in range. The genetic algorithm technique of Chapter
3 was used as the optimization method. Its cost function employed HFSS to solve
each structure to ensure the returned tness would be an accurate representation
of reality.
6.5.1 Reactance optimization
With an expensive cost function (HFSS), the optimization process needs to be as
ecient as possible. Referring to Figure 6.3, there are six dierent reactance values
that require attention. However, because the optimization goal is to maximize gain
in a single azimuth direction, the problem can be simplied by selecting either
= 0 or = 30 as the direction of gain optimization. Reactance symmetry
around this angular axis can then be used. The two axes reduce the required
optimization variables to four and three respectively. It was not known which
case would provide a higher gain, but it was suspected the = 0 case might as
radiation was along the elemental axis.
The second method to reduce the optimization time of the reactances was that
outlined in Section 6.3.3. was discretized between 0 and 2pi from which reactance
was calculated using (6.13). During the initial optimization stage, each reactance
was discretized into 64 levels (5 bits) separated by 5.6 in . In the later stages
of optimization, stepping was reduced to less that 1.
-
116 CHAPTER 6. THE ESPAR ANTENNA
6.5.2 Structural optimization
To reduce optimization time, the limited set of antenna dimensions in Table 6.1
were optimized. Monopole radius (1mm) and skirt sheet thickness (0mm; a 2-D
sheet) were kept constant. It was thought the time cost involved to include these
dimensions in the optimization, would outweigh the marginal performance increase
they would net.
In contrast to the reactive optimization, the optimization of structural dimen
sions had to be performed over a restricted range of values. A maximum of 0.2
variation from the initial dimensions was chosen as the optimization range for the
dimensions hs, rp and rg . Monopole heights ha and hp were only varied by 0.06
(7.5mm) during optimization1 .
The antenna electrical characteristics were not uniformly sensitive to all struc
tural dimensions. For instance, variation in ha or hp would produce greater changes
than the same variation in rg . Consequently, the optimization resolutions were set
empirically for each dimension. During the initial optimization stages, the resolu
tion of rg was 2mm, rp and hs were stepped by 1.5mm and the element heights ha
and hp were resolved to 1mm. The resolutions of ha and hp were rened to 0.5mm
when the optimization was in its nal stages.
6.5.3 Optimization result and analysis
The antenna solution was attained through four stages of optimization. Each phase
adopted a smaller optimization variable range with an increase in resolution. The
optimum chromosome in the nal population of each stage seeded the ranges and
resolutions of the following stage. Employing a dual PIII 850 with 1Gb RAM, the
total optimization involving 3799 unique cost function computations took close to
4 weeks. Table 6.4 and Table 6.5 present the reactances and structural dimen
sions of the antenna at termination of the optimization. The predicted antenna
characteristics are given in Section 6.5.4.
1This work was performed before Chapter 5, thus it was not thought to reduce rg to rp
-
117 6.5. ESPAR OPTIMIZATION
X1 X2 X3 X4 X5 X6 -14 -65 39 17 39 -65
Table 6.4: Optimized reactances (j) for the ESPAR antenna.
Length Structureal Dimensions
rg rp ha hp hs mm 60 38.5 27 27 34.5 0.48 0.308 0.216 0.216 0.276
Table 6.5: Optimized structure of the ESPAR antenna at 2.4GHz.
Comparing the dimensions of Table 6.5 and Table 6.1, the most signicant
change is the parasitic radius rp , which increased by 9.5mm to almost a third of
a wavelength. Similar to Section 5.4.2, all unique solutions (chromosomes) were
recorded so a sensitivity analysis could be made. Figures 6.13 and 6.14 show the
sensitivity plots of all optimization variables. The reactances are replaced with
their equivalent reection phase for illustrative purposes.
The tness was not sensitive to variables hs and rg in Figure 6.13. As such,
after the second stage of optimization, the skirt height and ground radius were kept
constant to reduce the solution space. It is for this reason, their sensitivity plots
do not exhibit tness of equal magnitude to the remaining variables. Conversely,
the parasitic elemental radius and element heights had signicant eect on tness.
These sensitivity results validate the dierent resolutions used between variables
mentioned in Section 6.5.2.
Figure 6.14 conrms the solution space does not have a single well dened opti
mum but is rather populated with multiple sub-optimum solutions. Each reactance
plot shows two to three signicant solutions. The sensitivity plots suggest the so
lution obtained was not necessarily the best possible. In addition, the stochastic
basis of the genetic algorithm requires multiple optimization runs be performed for
increased accuracy. However, with a single optimization taking 4 weeks, this was
not deemed appropriate for the few tenths of a dB likely to be gained.
-
118 CHAPTER 6. THE ESPAR ANTENNA
1 0 1 2 3 4 5 6 7 850
55
60
65
70
r g [m
m]
1 0 1 2 3 4 5 6 7 835
30
25
20
h s [m
m]
1 0 1 2 3 4 5 6 7 820
25
30
35
h a [m
m]
1 0 1 2 3 4 5 6 7 820
25
30
35
h p [m
m]
1 0 1 2 3 4 5 6 7 820
25
30
35
Fitness
r p [m
m]
Figure 6.13: Sensitivity plots of the ESPAR antenna dimensions.
-
6.5. ESPAR OPTIMIZATION 119
1 0 1 2 3 4 5 6 7 8200
100
0
100
200
1
[]
1 0 1 2 3 4 5 6 7 8200
100
0
100
200
2
[]
1 0 1 2 3 4 5 6 7 8200
100
0
100
200
3
[]
1 0 1 2 3 4 5 6 7 8200
100
0
100
200
Fitness
4
[]
Figure 6.14: Sensitivity plots of the ESPAR antenna reactive loads.
-
120 CHAPTER 6. THE ESPAR ANTENNA
Figure 6.15: Top and bottom view of constructed antenna (Tables 6.4 and 6.5) with xed reactive loading.
6.5.4 Experimental verification
The antenna was built in a similar fashion to that explained in Section 5.5.1. The
only addition was the plated vias through the ground plane. The SMA panel mount
dielectric extended through the PCB that made up the ground. It was then cut
ush with the top ground surface. However, to mitigate loss through the PCB, the
vias were plated with copper sheet. Therefore, a perfect 50 line was produced
from the monopole base through to the load termination. Top and bottom views
of the antenna are presented in Figure 6.15. All solder joins are only signicant
visually. Their discontinuity with the antenna surface was negligible at a maximum
of 0.2mm.
The antenna was tested in a anechoic chamber with its S11, H-plane and E-plane
radiation patterns recorded. The antenna mismatch impedance was poor with a
measured S11 of -7dB at 2.4GHz, within 0.4dB of the computer result. Simulation
results of the S11 bandwidth showed minimal (within 0.5dB) degradation over
100MHz and hence it is clear the S11 is not too localized around resonance. This
could obviously not be shown experimentally due to (6.15) but as simulation was so
close to experiment, it can safely be assumed. If lower levels of S11 are required, the
problem would have to be re-optimized with weights associated to the directivity
-
121 6.5. ESPAR OPTIMIZATION
and S11 in (6.18).
The antenna produced H and E-plane radiation characteristics that can be
seen in Figure 6.16. In the azimuth, the antennas maximum gain reaches 8.08dBi.
Its front-to-back and front-to-null ratios are 10dB and 18.8dB respectively. This
pattern can be reproduced at 60 intervals through the azimuth due to antenna
symmetry. The main lobe is elevated at an angle of 4 above the horizontal. Cross
polarization patterns were measured and found to be below 14dB of the principal
lobe gain.
There is a clear agreement between the simulated and experimental results. Had
the cost function simulated the antenna with inconsistent error, the GA would have
converged on a simulated optimum rather than its true counterpart. The similar
ity between the simulated and experimental results conrm the cost function was
accurate and hence the chance of a single optimization run nding the optimum
solution was maximized. The simulated gain was degraded by 0.7dB when com
pared to experimental testing. However, this error was consistent over a number
of dierent tests and therefore would not have aected the optimization.
When the original antenna (Table 6.1) was optimized with respect to reactance
only, an experimental gain no greater than 6.7dBi was achieved (case 1). In addi
tion, the main lobe was elevated at 25. Conversely, regardless of the nite ground
plane, the optimized antenna has minimal lobe elevation and a gain of 8.08dBi.
This reduction in main-lobe elevation is seen as the likely reason for the increase in
azimuthal gain. To reject the possibility that the reactance-structure combination
uked a suppression of main lobe elevation, a further 10 random, non-optimized
reactance loading cases were tested. Principal lobe elevations were recorded be
tween 0 and 15 with a mean of 9.7. Horizontal gains were no less than 0.5dB of
the elevated gains. This is in contrast to the original antennas average elevation
of 25 with horizontal gains between 1dB and 1.6dB less than the elevated gains.
These results suggest the optimized structure of the antenna reduced the main lobe
elevation independent of the reactive loading.
While the structural optimization netted an increase in gain of approximately
-
122 CHAPTER 6. THE ESPAR ANTENNA
90
8dBi
0
8
16 0
30
60120
150
180
210
240 300
330
270
(a) 0
8dBi
2
4
10
16
30
60
90
120
150210
240
270
300
330
180
(b)
Figure 6.16: Comparison of HFSS simulation (red) and experimental (blue) H-plane (a) and E-plane (b) radiation patterns at 2.4GHz of the antenna dened in Tables 6.4 and 6.5. HFSS Gain: 7.4dBi, experimental gain: 8.08dBi
-
6.6. SUMMARY 123
1.4dB over the reactive optimization, this could potentially translate to 2.8dB in a
system environment, which is equivalent to a signicant reduction in transmitter
power of approximately 48%. In addition, showing that an accurate structural
simulation works, has laid foundations for immersing the antenna in dielectric to
reduce it to a more practical size as has been proposed previously [103].
6.6 Summary
Loaded N port array theory is well known and documented. However, practical
implementation of the antenna has always been limited; usually only as a means of
theory validation. The ESPAR antenna is being developed specically for consumer
application and therefore its practicality requires consideration. Accordingly, the
ESPAR antennas utility for implementation in a communication system has been
investigated. Practical issues regarding the reactive loads were initially examined.
Varactor harmonic distortion was shown to be reduced by replacing a single var
actor with an anti-series pair. Varactors or their anti-series pair equivalents could
then be combined with dierent transmission lines to encompass the full reactance
range.
Testing and examination of a physical ESPAR structure followed. Simula
tion was used to perform a frequency sensitivity analysis of the ESPAR structure.
It showed that over an 8.1% bandwidth, the antenna had relatively stable gain,
front to null ratio and 3dB H-plane beamwidth possessing a maximum variation of
0.44dB, -3.8dB and 20 respectively for ve separate loading congurations. The
loading congurations were pre-optimized for practical radiation features which
therefore weighted their relative importance. Simulation procedure was validated
with experimental testing at resonance (2.4GHz).
Finally, the design of a horizontal gain optimized ESPAR antenna was pre
sented. The antenna was optimized with respect to gain to reduce interference and
transmission powers in a system. In a homogeneous network, any optimization
in the gain of an antenna will potentially have double the eect in the system.
-
124 CHAPTER 6. THE ESPAR ANTENNA
Situated in a large solution space, a genetic algorithm was used for optimization
because of its robust nature. The GA employed a FEM based cost function in
order to obtain accurate modeling of the antenna. Both physical structure and
loading reactances were optimized. The nal antenna possessed a maximum gain
of 8.08dBi even though S11 was only -7dB at 2.4GHz. Average principal lobe
elevation angles were recorded at 9.7 for 10 random loading cases.
The author was involved in the ASVP testing and was subsequently an ancillary
author in its publication [104, 100, 101, 98]. The frequency sensitivity procedure
and results were presented at [71, 72] and the ESPAR optimized design has been
published [73].