Sie sind auf Seite 1von 27

Final Thesis Report 2010, UNSW@ADFA

1
Ultra Wideband (UWB) Antennas
Andrew J. Spear
1

University of New South Wales at the Australian Defence Force Academy
The aim of the project was to develop a diagnostic tool for evaluation of Ultra
Wideband (UWB) antenna performance. The evaluation involves determining if the
antenna has minimum phase properties, which allows the use of a passive equalisation
network. This passive equalisation network is designed to linearise the phase response
and hence give a constant group delay which is a desirable property for a UWB antenna.
The diagnostic process evaluates how close an antenna is to minimum phase and how
complex the equalisation network would have to be to give a constant group delay. This
process was illustrated using simulation and measurement of 2 different UWB antenna
designs and gave results for varying degrees of equaliser complexity.

Contents

Ultra Wideband (UWB) Antennas 1
Contents 1
I. Introduction 2
II. The 2-Port Network 2
III. Minimum Phase Systems 3
IV. Antenna Transfer Function 4
V. Equalization Network 4
VI. Antenna Parameters 4
VII. Literature Review 5
VIII. Simulation 6
A. Circular Monopole 8
B. Vivaldi Antenna 10
IX. Procedure 13
X. Results 14
A. Comparison of Simulation to Measurements 14
B. Characterization of Substrate Permittivity 15
C. Measured Data 16
XI. Discussion 21
XII. Conclusion 25
XIII. Recommendations 25
XIV. Acknowledgements 25
XV. References 26



1
OFFCDT, School of Engineering & Information Technology. Electrical Engineering Project ZEIT4299.


Final Thesis Report 2010, UNSW@ADFA

2
I. Nomenclature

UWB = Ultra Wideband
FCC = Federal Communications Commission
LHP = Left Half Plane
= Phase [rad]
= Frequency [rad/s]

g
= Group Delay [ns]
R = Antenna Separation [m]
= Reflection Coefficient
= Propagation Coefficient
= Dielectric Permittivity

o
= Free Space Permittivity [8.854x10
-12

F/m]

o
= Free Space Permeability [4x10
-7
H/m]

II. Introduction
The Federal Communications Commission (FCC), the U.S. telecommunications regulator, has allocated
7.5GHz of spectrum for unlicensed use of ultra wideband devices (UWB) in the 3.1GHz to 10.6GHz frequency
band with a maximum transmit power spectral density of -41dBm/MHz [1]. The FCC defines UWB as any
signal that occupies a bandwidth of more than 500MHz or 20% of the centre frequency [26] in the 3.1GHz to
10.6GHz band. Due to the broad bandwidth, UWB technology allows both high-data-rate personal area network
(PAN) wireless connectivity and longer-range, low-data-rate applications [2]. Using signals of a large
bandwidth adds an increased level of complexity when it comes to characterizing UWB antennas. Traditional
antennas are designed to operate at either a particular frequency or over a small range of frequencies depending
on their applications because antenna parameters change significantly over frequency. Operating over a large
frequency band has the effect of incorporating these significant changes in antenna parameters hence traditional
antenna theory is only applicable to a certain degree. The most important parameter of UWB antennas which
changes with frequency is the phase response. It is desired to have a linear phase response which gives a
constant group delay (given in (1)). A constant group delay results in all spectral components of a UWB signal
being received at the same time. Having a non constant group delay results in time dispersion of signals
meaning that different spectral components of a UWB signal arrive at different times which gives distortion in
the signal. This makes the signal processing very complicated to recreate the original signal. For this reason, a
diagnostic tool is developed in order to determine whether a particular UWB antenna can be equalised over the
UWB frequency band to achieve a constant group delay given in (1).

( )
( )

= -
g
(1)

Where is the phase response of the antenna and is the frequency in radians per second. This project took
antenna designs from [6] to [16] and analysed and characterized these antennas according to the data that was
published along with these designs. From this point, several designs were selected and simulated using CST
Microwave Studio and the antennas with the best performance were then constructed and tested. The antenna
designs chosen were the circular monopole antenna proposed by [15] and the Vivaldi antenna proposed by [13].
These were then tested in an anechoic chamber using a network analyser to determine the antenna S-parameters
over the UWB frequency range. The diagnostic tool was then developed using MATLAB to take the S-
parameters of the antenna link and determine how close the antennas are to minimum phase. If an antenna is
minimum phase then it can, in principle, be equalised over a wide frequency band using a passive network. The
diagnostic tool then determines what effect equalisation has on the antenna phase response and how the
complexity of the equalisation network effects the equalisation over the UWB frequency band.

III. The 2-Port Network
The 2-port network in the case of this project consists of two antennas separated by a free space gap as
shown in Figure 1.



Final Thesis Report 2010, UNSW@ADFA

3

Figure 1. 2-Port Antenna Network
The 2-port network has four S-parameters; these being S
11
, S
12
, S
21
and S
22
. In the case of an antenna link,
S
11
describes the reflections at the input of port 1, S
12
describes the transmission from port 1 to port 2, S
21

describes the transmission from port 2 to port 1 and S
22
describes the reflections at the input of port 2. In the
case of an antenna 2-port network, the distance R is not a fixed distance. This is because over the UWB
frequency range, the phase centre of the antennas can move with respect to the antenna structure. The phase
centre of an antenna is the point at which the far field radiation appears to be radiated from. In this project, the
phase centre is assumed to be at a single point over the entire UWB frequency range; this point is on the antenna
structure itself. The free space gap has a linear phase response associated with it. In air, the phase response of
the signal is equal to

( )
jkR
e

= (2)

Where k is the frequency in radians per second divided by the speed of light. To remove this linear phase
term, the distance R was set to the separation of the antennas. It was later found that R had to be changed to
account for the movement of the phase centre and another linear phase term was removed to account for signal
propagation through the antenna structure itself.
IV. Minimum Phase Systems
A minimum phase system is defined as a system which has all poles and zeroes of a transfer function in the
j domain in the left half plane (LHP) [4]. [4] also states that any rational system can be separated into a
minimum phase system and an all-pass system

( ) ( ) ( ) j G j G j G
ap min
= (3)

A minimum phase system can, in principle, be equalised to have a linear phase over unlimited bandwidth
using an equaliser physically realisable from passive components [5]. This is done by creating a passive
network consisting of capacitors and inductors to give LHP poles and zeroes to compensate for those (placing a
pole for every zero and a zero for every pole) of the minimum phase system. The minimum phase component
of a system is defined as the negative Hilbert transform of the attenuation of the system [21], given by

( ) ( ) [ ] j G H j G ln
min
= (4)

Where ln|G(j)| is the attenuation of the system. The minimum phase function can also be represented by
the symbol
m
. This can also be defined mathematically from [20] as

( )
( ) [ ] ( ) [ ]

d
j G j G
j G

=
0
2 2
min
'
ln ' ln
2
(5)

Where S
21
is the transmission phase response of the system. Once the minimum phase term has been
calculated, the all-pass phase is the transmission phase response minus the minimum phase response. The
magnitude response of the all-pass system is unity by definition and if the system is minimum phase, the all-
Port 1
Port 2
R

Final Thesis Report 2010, UNSW@ADFA

pass phase response will be zero. If the all
equalised over a wide frequency range to ideally give this all

The property of minimum phase can then be applied to the antenna 2
transfer function of a single antenna.
transmission characteristic (S
21
parameter) in terms of the components in the system; according to [20], the S
parameter is equal to

S

where H
TX
() and H
RX
() are the transfer functions of the transmit and the receive antennas respectively, R
is the separation of the 2 antennas and c
identical antennas with a single polarisation then the antenna transfer function can be given as



If the antenna were minimum
magnitude of H() and the phase response would be equal to
are minimum phase, it is unlikely that this will be
when it comes to the design of the equali

Any non minimum phase components
reason, the antenna transfer function, for the purposes of equali
of H() and phase equal to
m
/2.
using a passive network, the best possible outcome of the equali
the system. Using this transfer function, the equali
minimum phase transfer function and impl
the complexity has a significant effect on the
equalisation network, the synthesis becomes very difficult with higher order transfer functions due to the fact
that the order of the transfer function reflects upon th
higher the order, the more capacitors and inductors are required.
complexities will be investigated within this project.
transmit antenna and after the receive antenna as shown in

Figure
Because a UWB antenna is only one p
important than others however most parameters are still desired to be as close to ideal as possible. This is where
a trade-off is required to achieve the best performance in the largest am
important parameters when it comes to ultra wideband antennas are size, radiation pattern and impedance
bandwidth. Size is important because UWB communications has a major application in mobile communications

Final Thesis Report 2010, UNSW@ADFA
4
pass phase response will be zero. If the all-pass phase response is non-zero then the system can only be
ed over a wide frequency range to ideally give this all-pass phase response.
V. Antenna Transfer Function
The property of minimum phase can then be applied to the antenna 2-port network where G(j
transfer function of a single antenna. The antenna transfer function can be determined by defining the
parameter) in terms of the components in the system; according to [20], the S
( ) ( ) ( )
o
jkR
RX TX
Rc
e
G G j S


2
21

=
) are the transfer functions of the transmit and the receive antennas respectively, R
tion of the 2 antennas and c
o
is the speed of light. Assuming the 2-port network consists of two
identical antennas with a single polarisation then the antenna transfer function can be given as
( ) ( )
o
c
R j
o
e S
j
Rc
H

21
2
=
If the antenna were minimum phase then the magnitude response of the antenna would be equal to the
) and the phase response would be equal to
m
/2. However, as it is unlikely that the antennas
are minimum phase, it is unlikely that this will be exactly the transfer function of the antenna. This is important
when it comes to the design of the equalisation network.
VI. Equalisation Network
Any non minimum phase components of the system cannot be equalised using a passive network. For this
ion, for the purposes of equalisation, has a magnitude equal to the magnitude
As any non-minimum phase components of the system cannot be equali
possible outcome of the equalisation is to be left with t
s transfer function, the equalisation network can be defined as the inverse of the antenna
minimum phase transfer function and implemented as such. If the equalisation network were to be synthesized,
ignificant effect on the physical equalisation network. For larger complexities of
ation network, the synthesis becomes very difficult with higher order transfer functions due to the fact
that the order of the transfer function reflects upon the amount of components used in the network, that is, the
higher the order, the more capacitors and inductors are required. For this reason several equali
complexities will be investigated within this project. In practice, the equalisation network is placed before the
transmit antenna and after the receive antenna as shown in Figure 2.
Figure 2. Antenna Network with Equalisation Network
VII. Antenna Parameters
Because a UWB antenna is only one particular type of antenna, some of the antenna parameters are more
important than others however most parameters are still desired to be as close to ideal as possible. This is where
off is required to achieve the best performance in the largest amount of parameters possible. The more
important parameters when it comes to ultra wideband antennas are size, radiation pattern and impedance
bandwidth. Size is important because UWB communications has a major application in mobile communications
n the system can only be
port network where G(j) is the
The antenna transfer function can be determined by defining the
parameter) in terms of the components in the system; according to [20], the S
21

(6)
) are the transfer functions of the transmit and the receive antennas respectively, R
port network consists of two
identical antennas with a single polarisation then the antenna transfer function can be given as
(7)
phase then the magnitude response of the antenna would be equal to the
/2. However, as it is unlikely that the antennas
exactly the transfer function of the antenna. This is important
ed using a passive network. For this
has a magnitude equal to the magnitude
of the system cannot be equalised
ation is to be left with the all-pass response of
ation network can be defined as the inverse of the antenna
ation network were to be synthesized,
r larger complexities of
ation network, the synthesis becomes very difficult with higher order transfer functions due to the fact
e amount of components used in the network, that is, the
For this reason several equalisation network
etwork is placed before the

articular type of antenna, some of the antenna parameters are more
important than others however most parameters are still desired to be as close to ideal as possible. This is where
ount of parameters possible. The more
important parameters when it comes to ultra wideband antennas are size, radiation pattern and impedance
bandwidth. Size is important because UWB communications has a major application in mobile communications


Final Thesis Report 2010, UNSW@ADFA

5
hence it is desired for the size of the antenna to be as small as possible. The radiation pattern of an antenna in
practice can either be omnidirectional or directional with the pattern being selected for the intended use of the
antenna. Impedance bandwidth is important because a UWB antenna operates over a wide range of frequencies.
In practice it is not possible to achieve ideal performance in all of these parameters hence a trade-off is required
in the design process to achieve the best performance in these parameters.
For a UWB antenna, the size is desired to be what is known as electrically small. An electrically small
antenna is defined as an antenna that is about to fit into a sphere that has a radius of one electrical radian given
by /2. Because in UWB applications there is a requirement for the antenna to operate at frequencies between
3.1GHz and 10.6GHz, the relative size of the antenna varies depending on which end of the frequency spectrum
is being used. An electrically small antenna over this entire frequency range cannot simultaneously provide the
required impedance bandwidth or efficiency for typical UWB applications. For this reason, it is common for an
antenna to be electrically small at the lower end of the frequency range and have a moderate electrical size at the
upper end of the frequency range.
A common way of defining the impedance bandwidth of an antenna is the frequency range over which the
system has greater than a 10dB return loss; that is, the magnitude of S
11
is less than -10dB. For a UWB
application, a good impedance match can be defined as an antenna system which has a voltage standing wave
ratio (VSWR) of less than 2 [3]. For a perfectly matching system, the VSWR, given as



+
=
1
1
VSWR
(8)

has a reflection coefficient () of zero. This is because it is desired that the transmit antenna radiate all of
the power input to the antenna and that nothing be reflected by the transmit antenna back to the source, hence
is zero. For a good matching system (VSWR<2), (2) shows that has to be less than one third (0.3333) to give
|S
11
| = -9.54dB -10dB.
For the most case, it is desirable that the radiation pattern is as close to isotropic as possible; however there
are applications which require a directional radiation pattern. An omnidirectional radiation pattern means that
the performance of the system is independent of the orientation of the antennas in one two-dimensional plane of
reference and allows for more freedom and mobility of the antennas. The opposite can be said for a directional
antenna, the efficiency of the system is highly dependent upon the orientation of both receive and transmit
antennas. The efficiency of the system is dependent upon the power into the transmit antenna and the power out
of the receive antenna. This is not to be confused with the efficiency of an antenna which is dependent upon the
radiation resistance and loss resistance of a single antenna.
VIII. Literature Review
The first step taken in this project was to conduct a literature review on various sources of UWB antennas
and to analyse antenna performances from these sources. The literature review comprised of antenna designs
from [6] to [16] and the subsequent designs were analysed according to any results that had been published in
each individual paper including the impedance bandwidth and group delay of the antennas. As well as the
performance characteristics of the antennas, the ease of fabrication of the antennas was also taken into account.
This was done because the focus of the project was to develop the diagnostic tool and it would be
counterproductive to spend a significant amount of time constructing an antenna that has only a slightly better
performance.
The outcome of this literature review gave five antenna structures to simulate however limited knowledge in
the use of the simulation software restricted the ability of simulating the antennas to only one of the three
designs chosen. For this reason two simpler designs were chosen to simulate which resulted in the simulation of
three different antenna structures, these being the circular monopole antenna, the Vivaldi antenna and the
microstrip antenna. A copy of the literature review can be found in appendix A.



Final Thesis Report 2010, UNSW@ADFA

6
IX. Simulation
The simulation section of the project utilised Computer Simulation Technologys (CST) Microwave Studio
to perform 1-port simulations of single antennas and 2-port simulations of a link between two antennas for the
various antenna structures discussed in section VIII. The parameters simulated were the impedance bandwidth
(obtained from the magnitude of the S
11
parameter), the transmission characteristic (obtained from the
magnitude of the S
21
parameter), the antenna phase (obtained from the phase of the S
21
parameter) and the
radiation pattern. All simulations were conducted between 1GHz and 12GHz with the radiation pattern being
measured at 4.5GHz, 6.5GHz and 8.5GHz. The simulation frequency range was selected in order to cover the
entire UWB frequency range with a buffer either side of the frequency band. The bandwidth can easily be
increased however the increased bandwidth leads to extra computation time for information that is not required.
The three radiation pattern frequencies were selected in order to show the trend of the radiation pattern over the
UWB frequency range and it was decided that the three frequencies sufficiently demonstrated this. Simulations
were conducted using the adaptive mesh refinement utility enabled in order to give the most accurate results.
The excitation signal used was a Gaussian pulse and can be seen below in Figure 3 (a).
Figure 3 (b) shows the spectral component of the excitation signal used and it can be seen from this plot that
the Gaussian pulse contains spectral components between the simulation frequency range of 1GHz and 12 GHz.

2-port measurement of a microstrip printed a fibreglass board were taken to characterise the permittivity of
the FR-4 fibreglass board used for making the antenna structures. The 2-port measurements were taken using an
Agilent E5071C network analyser which gave the 2-port measurements in the form of S-parameters. The
permittivity of FR-4 is given as 4.3 at 10GHz according to [23] however the permittivity of FR-4 changes
significantly between different manufacturers. It is intended for the board to be used over the UWB frequency
range however the permittivity is not constant over the frequency range. Practically, it is not important to know
how the permittivity changes over the UWB frequency range however as the simulation uses an assumed
permittivity characteristic for FR-4, it is important to characterise the actual FR-4 board used for the antenna
construction to improve the fidelity of the simulation. Once the 2-port measurements for the microstrip had
been obtained, 1-port measurements for the connectors and adapters used were taken in order to remove their
effect upon the 2-port measurements. Once left with the 2-port parameters of just the microstrip, the S-
parameters were converted to ABCD parameters according to the conversion given in [17]. From there, by
assuming that the microstrip is a lossy transmission line, the propagation coefficient of the microstrip can be
calculated using the conversion given in (9) from [17].


( ) ( )
( )
( )
(
(
(

=
(

l
Z
l
l Z l
D
C
B
A


cosh
sinh
sinh cosh
0
0
(9)


(a) (b)
0 1 2 3 4 5 6 7
x 10
-19
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Excitation Signal
Time, ns
M
a
g
n
i
t
u
d
e
,

V
2 4 6 8 10 12
0
0.05
0.1
0.15
0.2
0.25
Spectral Density of Excitation Signal
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

V
Figure 3. (a) Time domain plot of the excitation signal used in simulation. (b) The spectral component of the
excitation signal used in simulation.


Final Thesis Report 2010, UNSW@ADFA

7
Where is the propagation coefficient of the transmission line, l is the length of the transmission line and Z
0

is the characteristic impedance of the transmission line. From (9) the propagation coefficient can be calculated
and can be split into its real and imaginary parts with the real part () being the attenuation coefficient and the
imaginary part () being the propagation coefficient. The phase velocity can then be calculated using
according to (10).

=
p
u
(10)

Using the phase velocity, the effective permittivity of the substrate can then be calculated using (11) from
[22].


2
|
|

\
|
=
p
e
u
c

(11)

Where c is the speed of light. The calculated effective permittivity can then be used to calculate the relative
permittivity of the FR-4 board. The relative permittivity is calculated according to (12) from [22].


1
12 1 1
12 1 2

|
|

\
|
+ +
+
=
W
d
W
d
e
r

(12)

The relative permittivity can then be put into CST in the form of = j where the data to be input is
/
o
and /
o
. /
o
is equal to
r
and /
o
is given in (13), from [22].


p o o
u
c

2
' ' =
(13)

Once this data is put into CST, the software package performs a regressive fit in order to produce a model
which represents the input data.



Final Thesis Report 2010, UNSW@ADFA

8
A. Circular Monopole

Figure 4 shows the model of the circular
monopole antenna from [15] used for
simulation in CST with the dimensions
shown below in Figure 5 where W
s
= 42mm,
L
s
= 50mm, W
p
= 20mm, L
f
= 20.3mm, W
f
=
2.6mm and L
g
= 20mm. The red rectangle
at the bottom of the antenna structure in
Figure 4 is the waveguide port used to excite
the antenna structure in place of a coaxial
feed used in practice.







Figure 6 shows the plot of the S
11
magnitude
which indicates that the circular monopole antenna
has an impedance bandwidth which covers the entire
UWB frequency range of 3.1GHz to 10.6GHz. As
discussed in section VII of this report, the impedance
bandwidth is given by the frequency bandwidth over
which the magnitude of S
11
is less than -10dB.
Figure 6 shows that the circular monopole antenna
has an S
11
magnitude of less than -10dB from
approximately 2.25GHz upwards.
The next parameter which was simulated was the
radiation pattern at 4.5GHz, 6.5GHz and 8.5GHz.
The results of this simulation can be seen below in
Figure 7.










L
g

L
s

W
s

L
f
W
p
W
f
Figure 4. Circular Monopole Antenna from [15] modelled in
CST
Figure 5. Dimensions of Circular Monopole Antenna from [15].
Figure 6. S
11
magnitude of Circular Monopole
antenna

0 2 4 6 8 10 12
-40
-35
-30
-25
-20
-15
-10
-5
0
|s11| of Circular Monopole Antenna
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


Final Thesis Report 2010, UNSW@ADFA

9


It can be seen in Figure 7 that the radiation pattern for the circular monopole antenna at 4.5GHz radiates in
main lobes at the front and back of the structure with a portion being radiated to the sides of the structure. At
6.5GHz, the radiation pattern shifts to having to main lobes off towards the upper left and right of the structure.
At 8.5GHz, the radiation pattern is similar to that at 6.5GHz however the main lobes are oriented more upwards
and have a narrower beamwidth. This means that for one particular orientation, the antennas are not going to be
to be situated within the main lobe over the entire frequency band which causes degradation of the received
signal. This demonstrates that the radiation pattern changes dramatically with frequency and will pose
restrictions on the types of applications for which this antenna can be used.
Following the 1-port simulation of the antenna, a 2-port simulation was conducted using 2 antenna structures
separated by a nominal distance. The nominal distance was set corresponding to the antenna separation used in
the practical testing of the antennas which in the case of the circular monopole antennas was 395mm. The
model used for the 2-port simulation in CST can be seen below in Figure 8.






(a) (b)


(c)
Figure 7. Radiation Patterns of Circular Monopole antenna at (a) 4.5GHz, (b) 6.5GHz, (c) 8.5GHz
Figure 8. 2-port simulation model from CST for a link between 2 circular monopole antennas.


Final Thesis Report 2010, UNSW@ADFA

10
The data extracted from the 2-port simulation included the magnitude of the S
21
parameter and also the phase
response of the antenna. The results for both are shown below in Figure 9.



The magnitude plot shown in Figure 9 (a) can be used to determine the minimum phase component of the
antenna which can then be compared to the antenna phase in order to determine whether the antenna can be
equalised using a passive network. The antenna phase shown in Figure 9 (b) was obtained from the S
21
phase
data by removing the phase due to the free space gap between the 2 antennas and also any linear phase due to
signal propagation through the structure itself. The resultant data is then divided by 2 to give the phase response
for one antenna.

B. Vivaldi Antenna

Figure 10 shows the model of the Vivaldi antenna
from [13] used for simulation in CST with the
dimensions shown below in Figure 11 where W
s
=
74mm, L
s
= 107mm, W
ff
= W
fb
= 23mm, L
ff
= L
fb
=
40mm, W
f1
= 1mm and W
f2
= 10mm. Similarly to the
simulation of the circular monopole antenna, the Vivaldi
antenna was also excited using a waveguide port in place
of a coaxial cable.







(a) (b)
0 2 4 6 8 10 12
-60
-55
-50
-45
-40
-35
-30
|s21| of Circular Monopole Antenna
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B
0 2 4 6 8 10 12
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
Antenna Phase of Circular Monopole Antenna
Frequency, GHz
P
h
a
s
e
,

r
a
d
Figure 9. (a) Magnitude of S
21
parameter obtained from simulation. (b) Antenna phase from
simulation.

W
fb

L
s

W
s

L
ff
W
ff
W
f1
L
fb

W
f2
Figure 10. Vivaldi Antenna from [13] modelled in
CST.
Figure 11. Dimensions of Vivaldi Antenna from [13].


Final Thesis Report 2010, UNSW@ADFA

11
Figure 12 shows the plot of the S
11
magnitude which
indicates that the Vivaldi antenna has an impedance
bandwidth which covers the majority of the UWB frequency
range of 3.1GHz to 10.6GHz. Figure 12 shows that the
Vivaldi antenna was not a good match between
approximately 3.2GHz to 3.7GHz. Performance wise, this is
an undesirable characteristic of the antenna however for the
purpose of the analysis to be performed on the antenna in
terms of phase equalization, this poor impedance match
between 3.2GHz and 3.7GHz does not have a significant
impact on the outcome of the analysis.
The next parameter which was simulated was the
radiation pattern at 4.5GHz, 6.5GHz and 8.5GHz. The
results of this simulation can be seen below in Figure 13.







It can be seen in Figure 13 that the radiation pattern for the Vivaldi antenna is far more directional than the
circular monopole antenna with the majority of the energy being radiated in the y direction (towards the top of
the antenna structure). As the frequency was increased, the radiation pattern did not change a significant
amount compared to the circular monopole antenna. This shows that the antenna is more stable in terms of its
directivity and the energy delivered over the UWB frequency band is approximately the same for the same
orientation of the antenna.
0 2 4 6 8 10 12
-45
-40
-35
-30
-25
-20
-15
-10
-5
0
|s11| of Vivaldi Antenna
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B
Figure 12. S
11
magnitude of Vivaldi antenna
Figure 13. Radiation Patterns of Vivaldi antenna at (a) 4.5GHz, (b) 6.5GHz, (c) 8.5GHz

(a) (b)


(c)


Final Thesis Report 2010, UNSW@ADFA

12
Once again, following the 1-port simulation of the antenna, a 2-port simulation was conducted using 2
antenna structures separated by a nominal distance. The nominal distance was set corresponding to the antenna
separation used in the practical testing of the antennas which in the case of the Vivaldi antennas was 290mm.
The model used for the 2-port simulation of the Vivaldi antenna in CST can be seen below in Figure 14.






The data extracted from the 2-port simulation included the magnitude of the S
21
parameter and also the phase
response of the antenna. The results for both are shown below in Figure 15.



In the same way as for the circular monopole antenna, the data shown in Figure 15 can be used to determine
the minimum phase component of the antenna and then compared to the antenna phase in order to determine
whether the antenna can be equalised using a passive network. The antenna phase was extracted using the same
process detailed for the circular monopole antenna.


(a) (b)

0 2 4 6 8 10 12
-45
-40
-35
-30
-25
-20
-15
|s21| of Vivaldi Antenna
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B
0 2 4 6 8 10 12
-5
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
Antenna Phase of Vivaldi Antenna
Frequency, GHz
P
h
a
s
e
,

r
a
d
Figure 14. 2-port simulation model from CST for a link between 2 Vivaldi antennas.
Figure 15. Magnitude of S
21
parameter obtained from simulation. (b) Antenna phase from simulation.


Final Thesis Report 2010, UNSW@ADFA

13
X. Procedure
The practical testing consisted of two parts, the first being a 2-port measurement of a simple microstrip
printed on a fibreglass board with SubMiniature version A (SMA) connectors. The second part consisted of the
1-port and 2-port measurements of the circular monopole antennas and the Vivaldi antennas. All 1-port and 2-
port measurements were taken over the frequency range of 2GHz to 12GHz. This range exceeds the UWB
frequency range which allows greater continuity of data for analysis.
The second part of testing consisted of the 1-port and 2-port testing of the Circular Monopole and Vivaldi
antennas. The 1-port testing was conducted using an Agilent E5071C network analyser to measure the S
11

parameter of each antenna structure. The S
11
parameter magnitude was then able to be compared to the S
11

magnitude plot obtained from CST to confirm the accuracy of the simulations. The 2-port testing was
conducted such that each antenna was receiving in the far field. This meant that the antenna separation had to
be at least twice the distance of the calculated far field for a single antenna. The far field is given according to
(14) from [25].

2
2D
R =
(14)

Where D is the size of the antenna and is the wavelength of the transmitted/received signal. As the
frequency is varied over a large bandwidth, this leads the wavelength to also vary and hence the far field
distance varies with frequency. For this reason, it was ensured that the antennas would be in the far field
regardless of the frequency which meant that the far field was calculated at the upper frequency limit (being
12GHz for testing purposes) hence giving the minimum wavelength of signal that will be used. For the circular
monopole antenna, the size of the radiating section of the antenna is 2cm and hence (14) can be used to give a
far field distance of 3.2cm meaning that the antenna separation has to be at least 6.4cm for the antennas to be
operating in the far field. For the Vivaldi antennas, the largest radiating section is approximately 4cm in length.
Once again, using (14) the far field distance can be calculated to be 12.8cm and hence the antenna separation
has to be at least 25.6cm.
The 2-port testing consisted of several iterations of tests conducted to achieve an accurate data set over the
UWB frequency range. The 2-port testing consisted of a Agilent E5071C network analyser with ports 1 and 2
each connected to an antenna and the antennas oriented towards each other separated by a free space distance R.
Figure 16 shows a block diagram of the final experimental setup and depicts the various components that are
incorporated into the measured data. In the final experiment, the data for the circular monopole antennas was
taken with an antenna separation of 395mm and for the Vivaldi antennas, the antennas were separated by
290mm.



Figure 16. Block diagram of experimental setup.
As the desired data from this setup is the magnitude and phase response of a single antenna and the
experimental data contains characteristics associated with each component in Figure 16, the undesired data has
to be removed; this is done both experimentally and mathematically. The 2-port data of each cable was
measured and the results can be seen in Fig. 1 and Fig. 2 in appendix B. The magnitude of the cables (in dB)
was then subtracted from the magnitude of the entire system (in dB) and the phase of the cables (in radians) was
subtracted from the phase of the entire system (in radians) giving the magnitude and phase response of the two
antennas and the free space gap between the antennas. Assuming that free space is a lossless medium, the free
space component can be removed mathematically by measuring the antenna separation R and calculating the
phase according to the free space transmission term e
-jkR
where k is the frequency in radians per second divided
by the speed of light. The phase due to this free space gap can then be removed by once again subtracting the
phase of the free space term (in radians). Whilst removing the free space component due to the antenna
separation R, any other linear phase components due to propagation through the structure are accounted for by
adjusting R. Keeping R constant over the entire frequency band assumes a constant phase reference although
the phase centre of the antenna can move over the frequency band. Finally, what is left is the phase contribution
jkR
e

R


Final Thesis Report 2010, UNSW@ADFA

14
due to the two antennas and the magnitude response of the two antennas. Because the antennas are of the same
design, it was assumed that both antennas behaved identically and hence to obtain the phase and magnitude of
an individual antenna, the magnitude is given as the square root of the magnitude for both antennas and phase
response could be divided by two. The result of this can be seen in section XI.
XI. Results
A. Comparison of Simulation to Measurements



Figure 17 shows the comparison of the simulated and measured antenna magnitude and phase responses for
the circular monopole antenna. It can be seen that the plots show similar trends however there are far more
small scale variations in the phase response of the measured data. The plots also show a significant difference
in the magnitude response of the antenna between the measured and simulated results.






(a) (b)
0 2 4 6 8 10 12
-45
-40
-35
-30
-25
-20
-15
|S21| for Measured and Simulated Data
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


Measured
Simulated
0 2 4 6 8 10 12
-10
-5
0
5
S21 phase for Measured and Simulated Data
Frequency, GHz
P
h
a
s
e
,

r
a
d


Measured
Simulated

(a) (b)
0 2 4 6 8 10 12
-100
-90
-80
-70
-60
-50
-40
-30
|S21| for Measured and Simulated Data
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


Measured
Simulated
0 2 4 6 8 10 12
-6
-5
-4
-3
-2
-1
0
1
2
S21 phase for Measured and Simulated Data
Frequency, GHz
P
h
a
s
e
,

r
a
d


Measured
Simulated
Figure 17. (a) - Magnitude of S
21
for Circular Monopole antenna, (b) - S
21
phase for Circular Monopole antenna
Figure 18. (a) - Magnitude of S
21
for Vivaldi antenna, (b) - S
21
phase for Vivaldi antenna


Final Thesis Report 2010, UNSW@ADFA

15
Figure 18 shows the comparison of the simulated and measured antenna magnitude and phase responses for
the Vivaldi antenna. Similarly to circular monopole antenna results, the plots show similar trends however the
small scale variations are much more prevalent in the measured data. It can be seen in Figure 18 (b) that there is
a large difference between the measured and simulated phase response. This is due to the lack of phase
reference of the measured data.
B. Characterization of Substrate Permittivity
Figure 19 shows the plot of the relative
permittivity of the FR-4 substrate used for the
construction of the microstrip and also used in the
construction of the antennas. The permittivity of FR-
4 is stated to be 4.3 however this plot shows how the
permittivity of the board changes over the UWB
frequency range. The plot shows several
discontinuities in the data and this is believed to be
the points at which the phase of the S
21
parameter is
equal to zero or wraps from pi radians to pi radians.




Figure 20 shows the complex permittivity of the FR-4
substrate used in the construction of the antennas. The
complex permittivity can be seen in (13) in section IX of
this report. The complex permittivity of the substrate is
used to demonstrate how lossy it is over the frequency
band. Representing the microstrip as a lossy transmission
line, it would be expected to see the complex permittivity
increase with frequency. Figure 20 shows that the
complex permittivity does increase over frequency which
confirms the assumption that the microstrip is a lossy
transmission line.


2 4 6 8 10 12
3
3.2
3.4
3.6
3.8
4
4.2
4.4
4.6
4.8
5
Relative Permittivity vs Frequency
Frequency, GHz
R
e
l
a
t
i
v
e

P
e
r
m
i
t
t
i
v
i
t
y
2 4 6 8 10 12
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Complex Permittivity vs Frequency
Frequency, GHz
P
e
r
m
i
t
t
i
v
i
t
y
Figure 20. Complex Permittivity of FR-4 Substrate
Figure 19. Relative Permittivity of FR-4 Substrate


Final Thesis Report 2010, UNSW@ADFA

16
C. Measured Data
Figure 21 shows the S
11
magnitude of the antennas measured in experimentation compared to the simulated
S
11
magnitude for both antennas. It can be seen from both plots that the data shows similar trends however there
are differences between them. In terms of impedance bandwidth, the antennas show better performance in the
measured data than in simulation.







Figure 22 shows the comparison of the S
21
magnitude and S
12
magnitude of the antenna link for both
antennas. These figures show the reciprocity of the antennas and given that S
21
is very similar to S
12
for both
antennas, it is safe to assume that the antennas are reciprocal. A slight difference can be seen between the S
21

and S
12
of the Vivaldi antennas however this is insignificant in that the difference is at a maximum of
approximately 0.5dB at approximately -44dB leading to a difference of 0.00051 in linear magnitude. This
difference can be attributed to the physical differences between the antennas in either the manufacturing process
or the connection of the SMA connector. Both antennas were constructed using a computer automated milling
machine using the same set of design files however no system is perfect and imperfections may have arisen
during this process. Also, to connect the antenna structure to a coaxial cable, an SMA connector was soldered

(a) (b)
2 4 6 8 10 12
-110
-100
-90
-80
-70
-60
-50
-40
-30
|S21| compared to |S12|
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


S21
S12
2 4 6 8 10 12
-46
-44
-42
-40
-38
-36
-34
-32
-30
-28
-26
|S21| compared to |S12|
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


S21
S12

(a) (b)
2 4 6 8 10 12
-60
-50
-40
-30
-20
-10
0
|s11| Measured and Simulated
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


Simulated
Measured
2 4 6 8 10 12
-60
-50
-40
-30
-20
-10
0
|s11| Measured and Simulated
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


Simulated
Measured
Figure 21. Comparison of Simulated and Measured S
11
magnitude data for (a) Circular Monopole
antenna, (b) Vivaldi antenna
Figure 22. Comparison of measured S
21
magnitude to S
12
magnitude for (a) Circular Monopole
antenna, (b) Vivaldi antenna


Final Thesis Report 2010, UNSW@ADFA

17
onto the structure. Although due care was taken when soldering the connectors, imperfections may have arisen
during this process.
To extract the antenna phase, it was necessary to assume reciprocity of the antennas, that is, each antenna
behaves identically. Figure 22 shows that the antenna links has almost exactly the same S
21
and S
12
parameters
which confirms this assumption of reciprocity between each pair of antennas.
Figure 23 is an example of the smoothing function
used in MATLAB to smooth the group delay data in
order for it to be readable. The function smoothes the
data contained in a matrix using a mean filter over a
size specified by the user. It can be seen that the
smoothing function produces a smooth curve that
accurately represents the original data. The smoothing
function itself was written by [25].
This smoothing function was used because in the
raw group delay data there was a lot variation in the
plot which can be seen as the blue curve in Figure 23.
The reason there is a lot of variation in the group delay
is due to the small inaccuracies of the network analyser
which cause the phase to appear to have a general trend
due to the large amount of data points however closer
inspection reveals that there are small scale variations
between the data points. When this data is
differentiated, the small scale variations become
positive and negative numbers which leads to the large
variations which can be seen in Figure 23.





Figure 24 shows the magnitude data of the antenna link with the free space gap separation. A Hilbert
transform is performed on this data to obtain the minimum phase function for the antenna.

2 4 6 8 10 12
-30
-20
-10
0
10
20
30
Measured Group Delay vs Smoothed Group Delay
Frequency, GHz
G
r
o
u
p

D
e
l
a
y
,

n
s


Measured
Smoothed
Figure 23. Effect of Smoothing Function on data
Figure 24. (a) - Magnitude plot of two Circular Monopole antennas separated by free space, (b) -
Magnitude plot of two Vivaldi antennas separated by free space

(a) (b)

2 4 6 8 10 12
-100
-90
-80
-70
-60
-50
-40
-30
|s21| for 2 Antennas Separated by Free Space
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B
2 4 6 8 10 12
-40
-38
-36
-34
-32
-30
-28
-26
-24
-22
|s21| for 2 Antennas Separated by Free Space
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


Final Thesis Report 2010, UNSW@ADFA

18






















Figure 25 shows the measured and minimum phase response and group delay of the circular monopole
antenna. The plot of group delay has been smoothed using the previously mentioned smoothing function. It can
be seen in Figure 25 that there is a discontinuity at approximately 9.8GHz. This is due to the S
21
magnitude
(seen in Figure 24) pushing against the threshold of the network analyser. At this frequency, the magnitude
drops to a point at which the data is too small to make sense of and hence the phase reference is lost causing this
discontinuity. It can also be seen that the phase does not begin at zero. This is because a proper phase reference
is not obtained due to the fact that the frequency band begins a 2GHz rather than DC. Figure 25 (b) shows
points at which the group delay is negative. This is not physically possible as it would indicate that the signal
sent at that frequency is arriving before it is actually sent. This is believed to be due to the effect of the
smoothing function on the data combined with the small scale inaccuracy in the network analyser. The points at
which the group delay goes negative are at points where the actual (not smoothed) group delay becomes very
noisy due to slight changes in the phase and hence when put through a mean filter, the average becomes an
arbitrary number in the noisy data range.
It can be seen that the minimum phase response/group delay is similar to the measured phase response/group
delay however the results are still slightly different. This dissimilarity is highlighted in Error! Reference
source not found. which shows the all-pass group delay of the circular monopole antenna where the all pass
group delay is the minimum phase group delay subtracted from the antenna group delay. If the antenna were
minimum phase then the all-pass phase would be constant however it is quite clear that this is not the case; the
all-pass group delay of the circular monopole antenna varies a significant amount with frequency making
equalisation of the antenna particularly difficult due to a passive network only being able to equalise the
minimum phase component of the antenna.


(b) (b)

2 4 6 8 10 12
-0.5
0
0.5
1
1.5
2
2.5
Measured Phase vs Minimum Phase
Frequency, GHz
P
h
a
s
e
,

r
a
d


Measured
Minimum Phase
2 4 6 8 10 12
-3
-2
-1
0
1
2
3
4
Measured Group Delay vs Minimum Phase Group Delay
Frequency (GHz)
G
r
o
u
p

D
e
l
a
y

(
n
s
)


Measured
Minimum Phase
Figure 25. (a) Measured and Minimum Phase Responses of Circular Monopole antenna, (b) Measured and
Minimum Phase Group Delay of Circular Monopole antenna.

(a) (b)
2 4 6 8 10 12
-4
-3
-2
-1
0
1
2
Measured Phase vs Minimum Phase
Frequency, GHz
P
h
a
s
e
,

r
a
d


Measured
Minimum Phase
2 4 6 8 10 12
-3
-2
-1
0
1
2
3
4
5
6
Measured Group Delay vs Minimum Phase Group Delay
Frequency (GHz)
G
r
o
u
p

D
e
l
a
y

(
n
s
)


Measured
Minimum Phase
Figure 26. (a) Measured Phase Response and Minimum Phase Response of Circular Monopole antenna,
(b) Measured Group Delay and Minimum Phase Group Delay of Vivaldi antenna.


Final Thesis Report 2010, UNSW@ADFA

19



Figure 26 shows the measured and minimum phase response and group delay of the Vivaldi antenna and the
group delay data has been smoothed using the smoothing function. Figure 26 (a) shows that there is a large
difference between the measured and minimum phase response however Figure 26 (b) shows that the group
delays are rather similar. The difference in phase response is due to the lack of phase reference because the
frequency band did not start at DC. For the purposes of the data analysis, the difference in the measured phase
and minimum phase response is insignificant as the group delays are relatively similar.
Figure 26 shows the comparison of the
measured antenna phase response and group delay
to the calculated minimum phase response and
group delay for the Vivaldi antenna. The plot of
the group delay shows that the minimum phase
group delay is very similar to the measured group
delay up until approximately 6GHz at which point
the two begin to differ. Error! Reference source
not found. also confirms this showing the all-pass
group delay of the Vivaldi antenna remaining
relatively constant up until approximately 6GHz.
The all-pass group delay of the Vivaldi antenna
varies significantly less than the circular
monopole antenna hence making the task of
equalization less difficult for the Vivaldi antenna
compared to the circular monopole antenna.
Figure 27 shows the all-pass group delay of
both antennas calculated by subtracting the
calculated minimum phase from the antenna
phase. It can be seen that the circular monopole
group delay goes negative over certain frequencies.
As mentioned previously, this is not physically
possible and is due to the negative group delay seen in Figure 25 as this data was used for the calculation.



Figure 28 shows the effect of equalisation on the group delay of the both antennas for varying equalisation
network complexities. As per previous group delay plots, the data has been smoothed. The negative group
delay components which can be seen are due to the small scale inaccuracies in the phase response which are
reflected in the group delay and then smoothed which produces the negative value.
The data indicates that for the circular monopole antenna, equalisation had a limited effect. This was
expected as the antenna had a significant all-pass phase component which is not able to be equalised using a

(a) (b)
2 4 6 8 10 12
-6
-4
-2
0
2
4
6
Equalised Group Delay vs Measured Group Delay
Frequency (GHz)
G
r
o
u
p

D
e
l
a
y

(
n
s
)


Measured
Equalised (N=5)
Equalised (N=10)
Equalised (N=14)
2 4 6 8 10 12
-2
-1
0
1
2
3
4
Equalised Group Delay vs Measured Group Delay
Frequency (GHz)
G
r
o
u
p

D
e
l
a
y

(
n
s
)


Measured
Equalised (N=5)
Equalised (N=10)
Equalised (N=14)
Figure 28. Equalised Group Delay vs Original Group Delay for (a) Circular Monopole antenna for
various equalisation network complexities, (b) Vivaldi antenna for various equalisation network
complexities
2 3 4 5 6 7 8 9 10 11 12
-6
-4
-2
0
2
4
6
8
3.1GHz 10.6GHz
All-Pass Group Delay
Frequency (GHz)
G
r
o
u
p

D
e
l
a
y

(
n
s
)


Circular Monopole
Vivaldi
Figure 27. All-Pass Group Delay for both antennas.


Final Thesis Report 2010, UNSW@ADFA

20
passive network. Varying the complexity had an effect on the level of equalisation; the 5
th
order equalisation
network had a very limited effect on the antenna phase, the 10
th
order equalisation network had a significantly
greater effect on the antenna phase compared to the 5
th
order equalisation network and the 14
th
order
equalization network did not have a significantly greater effect than the 10
th
order equalisation network. A
similar statement can be made about the effect on equalisation for the Vivaldi antenna; that it did not have a
significant effect on the antenna phase. As expected from the all-pass phase component of the Vivaldi antenna,
equalisation would only have a limited effect and it can be observed that all equalisation networks had the same
effect on the antenna with very little difference between them. It was noticed from the all-pass phase
component of Vivaldi antenna that equalisation may be effective up to 6GHz, for this reason, equalisation
networks with the same orders of complexity were developed over the frequency band of 3.1GHz to 10.6GHz
and the results can be seen in Figure 29. This plot shows that over the smaller frequency range, the 10
th
and 14
th

order equalisation networks had a significant effect on the antenna phase with all larger scale variations in group
delay being removed leaving a relatively smooth curve for the antenna phase. This increase in the effectiveness
of the equalisation is firstly due to the fact that the all-pass phase is almost constant over this frequency range
and secondly due to the fact that 10
th
and 14
th
order transfer functions can more accurately represent the
magnitude and phase response of the minimum phase component over a smaller frequency band which in turn
means that the equalisation networks can more accurately equalise the minimum phase component over a
smaller frequency range rather than a larger frequency range. This indicates that if the complexity of an
equalisation network can be increased above 14
th
order, then the equalisation network can have a far more
significant effect on flattening the group delay over the entire UWB frequency range. However, having this
larger order equalisation network does increase complexity of the network significantly which has an inherent
difficulty in the synthesis of the equalisation network.



Figure 29. Equalisation over a smaller frequency band

3 3.5 4 4.5 5 5.5 6
-0.5
0
0.5
1
1.5
2
Equalised Group Delay vs Measured Group Delay
Frequency (GHz)
G
r
o
u
p

D
e
l
a
y

(
n
s
)


Measured
Equalised (N=5)
Equalised (N=10)
Equalised (N=14)


Final Thesis Report 2010, UNSW@ADFA

21
XII. Discussion
The simulation showed the radiation patterns for both the circular monopole antenna and the Vivaldi
antenna, showing that both had rather different radiation patterns. It can be seen that for the circular monopole
antenna the radiation pattern varies significantly with frequency, meaning that for one particular orientation, the
antennas are not going to be to be situated within the main lobe over the entire frequency band which would
explain why the magnitude of the S
21
parameter dropped significantly at higher frequencies. For the Vivaldi
antenna, very little variation was seen in the radiation pattern over the frequency range which shows that the
Vivaldi antenna has far greater stability than the circular monopole antenna. Figure 24 shows that the
magnitude does not change significantly with frequency (except at approximately 9.1GHz) which contributes to
its better performance compared to the circular monopole antenna.
Part B of the results details the characterization of the FR-4 dielectric substrate used for the construction of
the antenna structures. The relative permittivity shown in Error! Reference source not found. can be seen to
vary between approximately 3.8 and 4.1 over the UWB frequency range. The discontinuities seen in this plot
can be attributed to the points in the data at which the phase of the S
21
parameter passes through zero or wraps
(and hence passes through zero). The data obtained from these measurements was able to be fed back into CST
in order to improve the fidelity of the simulations.


Figure 30. Comparison of Simulation Accuracy for Circular Monopole Antenna
Figure 30 shows a comparison between CSTs model for the FR-4 substrate, the data input into CST for the
FR-4 substrate obtained using the microstrip and the measured data from the antenna structure. There is not a
significant difference between the results showing that CSTs model for the FR-4 substrate is adequate however
this process was important to confirm the accuracy of the model. When the relative permittivity data is input
into CST, a regressive fit is performed on the data automatically by CST in order to produce this model. It was
found that the model which produced the best fit was a seventh order model (determined by CST). Whichever
model for FR-4 used would produce very similar results so it is not important which model is chosen however
the model based on the physical data was used because as the FR-4 substrate performed similarly in this
situation, this may not be the case for all situations.
As the permittivity of the substrate varies significantly over the UWB frequency range, the performance of
UWB antennas may be increased by using a different substrate for the antennas. It could be seen from the
complex permittivity that the FR-4 substrate was lossy and hence using a different substrate such as Teflon may
reduce how lossy the substrate is which would improve the magnitude response of the antennas.
As mentioned in section X, the experiment used to obtain accurate 2-port measurements underwent several
iterations. The first iteration of 2-port testing consisted of connecting the antennas to the network analyser and
the antennas placed on a wooden structure to support the antennas and provide the free space gap for the
antenna link. In this setup, there were no extra cables used other than those connected directly to the network
analyser itself. The network analyser was calibrated such that the results obtained were measured from the end
of each of the cables (hence including the SMA connectors attached to the antenna structure). When the data
was analysed it was noticed that there was a significant amount of back-scatter from objects around the test
setup. For this reason it was decided to conduct further testing inside the anechoic chamber using a Hewlett-
Packard 8719A network analyser.
0 2 4 6 8 10 12
-40
-35
-30
-25
-20
-15
-10
-5
0
Comparison of Simulation Accuracy
Frequency, GHz
M
a
g
n
i
t
u
d
e
,

d
B


Measured
Actual FR-4
CST FR-4


Final Thesis Report 2010, UNSW@ADFA

22
Having the antenna link in the anechoic chamber required approximately 9m of cable to connect the
antennas to the network analyser. This addition of 9m of cable had two issues associated with it; these being
attenuation in the magnitude of the S
21
parameter and an added linear phase component. The first (and most
apparent) issue noticed was the addition of a linear phase component due to the length of cable. As the network
analyser can take a maximum of 1601 data points over the frequency range of interest, the frequency was swept
in steps of 6.25MHz. Another characteristic of the network analyser is that the measured phase is wrapped
around 180
o
. At larger antenna separations it was noticed that the phase of the S
21
parameter had a positive
gradient. This was quickly realised to be impossible and it was discovered that the issue was to do with the
number of data points, phase wrapping and the large linear phase component introduced by the large amount of
cable and the large free space gap between the antennas. This was because the number of data points over this
frequency range lead to the phase wrapping between subsequent measuring of the phase and causing the phase
to appear as if it had a positive gradient. It was found that this issue could be negated by reducing the free space
gap between the antennas however this lead to the discovery of the second issue. It was noticed that as the
measurements reached the upper limit of the frequency band, the magnitude of the S
21
parameter would reach a
constant value and appear very noisy around this point. This constant value was noticed to be the threshold in
accuracy of the network analyser and hence would give results that were so small that no sense of their phase
could be made. The reason the data was reaching this threshold was due to the large amount of cable which was
used to connect the antennas to the anechoic chamber. 2-port measurements of the cables themselves were
taken and the results of this can be seen in Fig. 1of appendix B. It can be seen in this plot that at the upper
frequency limit the magnitude of the S
21
parameter drops to approximately -30dB. This was the reason that the
data was reaching the threshold of the network analyser. At this point, there were three possible solutions to the
problem.
The first solution was to simply use less cable however as the network analyser being used could not be
moved and the anechoic chamber could certainly not be moved, this was not able to be done. The second
solution was to use the Agilent E5071C network analyser used in the first experiment and use anechoic foam
tiles outside the anechoic chamber to, in essence, create a smaller anechoic chamber closer to the network
analyser and hence requiring a shorter amount of cable to connect the antennas to the network analyser.
Creating an anechoic chamber outside of the anechoic chamber proved to be a more difficult task than initially
anticipated and the results still showed a significant amount of back-scatter. The third solution was to move the
Agilent E5071C network analyser next to the door of the anechoic chamber which meant that the anechoic
chamber could be utilised using a reduced amount of cable compared to that which was initially used with the
anechoic chamber. The data from this setup was significantly better than anything previously obtained in the
project.
The linear phase component of the system due to the free space gap was removed using a nominal distance
R which was assumed to be the physical antenna separation. The data was analysed with the free space
component removed and the distance R was altered to account for any linear phase delay due to propagation
through the antenna structure and also for any movement in the antenna phase centre. It can be seen in section
XI that the group delay goes negative at points and it was suggested that this could be due to the smoothing
function. This negative group delay could also be due to the removal of too much linear phase from the system.
Increasing the value of R increases the amount of linear phase delay that is removed from the system and
removing more linear phase delay results in a reduction in group delay. If too much linear phase delay is
removed from the system then this could cause the group delay of the antenna to appear negative.
A system that is minimum phase can be equalised over a wide frequency band using a passive network. This
is because a minimum phase system has all left half plane (LHP) poles and zeroes. The results show that neither
of the antennas are exactly minimum phase but do exhibit minimum phase properties over certain frequency
ranges. The all-pass component of the system (shown in Figure 27) is the component of the system which
cannot be equalised using a passive network; therefore, the best possible outcome of equalisation using a
passive network is this all-pass group delay. To develop the theoretical equaliser, the transfer function of the
minimum phase component of the antenna response must be obtained. This was done using MATLABs inbuilt
invfreqs function which allows a transfer function to be approximated with varying degrees of complexity. The
limiting factor of the function is that MATLAB is only able to compute a transfer function up to 14
th
order,
anything higher than this MATLAB is unable to obtain a solution. Once a transfer function is found for the
minimum phase component of the antenna, the equalisation network transfer function is equal to the inverse of
this function as shown in (14).

( )
( )

j G
j G
eq
min
1
= (14)



Final Thesis Report 2010, UNSW@ADFA

23
This is the equivalent of placing a LHP pole for every LHP zero in the minimum phase component and vice
versa for placing LHP zeroes for every LHP pole. Once the transfer function of the equalisation network has
been computed, the phase response is then computed and is added to the antenna phase. This is done because
the equalisation network is placed in cascade with the antenna according to Figure 2. Mathematically, this
appears as (15).

( ) ( ) ( ) ( ) j G j G j G j G
ap eq min
= (15)

Assuming that the equalisation transfer function is the exact inverse of the minimum phase transfer function,
the equalised group delay would be exactly the same as the all-pass group delay. However, as the invfreqs
function is limited to a 14
th
order transfer function, an equalisation network transfer function that is exactly the
inverse of the minimum phase transfer function was unable to be obtained.
Using the invfreqs function within MATLAB had an intrinsic flaw associated with it; this being the
difficulty for it to recreate a complex phase response with a maximum of 14
th
order. The definition of a
minimum phase system is that it has all LHP poles and zeroes however the invfreqs function does not always
allow this. It was noticed that when invfreqs function was used to compute the transfer function of the
minimum phase component of both antennas, the resultant transfer function would have one right half plane
(RHP) pole and zero. This is not possible for a minimum phase function or for a passive network therefore
when implementing the transfer function of the equalization network, all RHP poles and zeroes that were
computed by the invfreqs function were neglected from the equalisation network transfer function. This in turn
reduced the effect that the equalisation network had on the antenna phase.
The minimum phase component of the antenna was calculated using MATLABs inbuilt Hilbert transform
function. As stated in section IV, the phase of a minimum phase system is related to the magnitude through the
Hilbert transform. The mathematical solution for the Hilbert transform is given in (5) in section III and can also
be implemented in MATLAB to give the minimum phase component. MATLABs inbuilt function performs a
Fast Fourier Transform (FFT) on the input magnitude data, this data is then multiplied by the signum function
and the inverse FFT is then performed on it, the resultant data is the Hilbert transform of the original data. This
computation time of this function does not change significantly with an increase in the frequency range of the
data however the mathematical solution, which was implemented in MATLAB using for loops, did suffer
from a significant increase in the computation time for an increase in the frequency range. (5) shows the limits
of the Hilbert transform going between zero and infinity however the data measured was between 2GHz and
12GHz. The Hilbert transform is sensitive to truncation of the data according to [20] hence continuity in the
measured data was assumed. This assumption lead to the data being extended on the lower side of the measured
frequency band to DC using the first value of the data set by assuming it was constant from DC to 2GHz and
similarly the last in the data set was assumed to be the constant value between 12 GHz and 400GHz.
The sensitivity of the Hilbert transform was confirmed using a simple transfer function and performing the
Hilbert transform on the data. The results of this can be seen in Figure 31 below.


Figure 31. Test of Hilbert Transform Calculations
0 2 4 6 8 10
x 10
6
-7
-6
-5
-4
-3
-2
-1
0
1
Minimum Phase Response
Frequency, rad/s
P
h
a
s
e
,

r
a
d


MATLAB H Transform
Mathematical H Tranform
Calculated Phase


Final Thesis Report 2010, UNSW@ADFA

24
Figure 31 shows that MATLAB Hilbert transform is rather sensitive to truncation of data at the upper and
lower limits of the data set whereas the mathematical solution for is rather accurate at these limits. However,
truncation of data has the effect of zooming in on a frequency range of interest so in this test example, if the
frequency range of interest was between 2GHz and 4GHz, both functions have the same performance. For this
reason, it was decided to use MATLABs built in Hilbert transform because the frequency range can be easily
changed without an increase in computation time. It was also found that when using the mathematical solution
for the Hilbert transform on the measured data (from experimentation), the solution was very sensitive to the
small scale variations that were seen in the measured data and hence did not produce accurate results. Although
the mathematical solution appears to be a more accurate representation of the minimum phase component in an
ideal case, it did not perform well in a practical case.
Along with the confirmation of the Hilbert transform method of using a simple transfer function of a
minimum phase system and inspecting the results shown such as that depicted in Figure 31, this minimum phase
method was supported by another fact. [19] states that the minimum phase component represents the small
scale changes in phase and hence when the minimum phase component is subtracted from the antenna phase
component, the result is a smooth curve.


Figure 32. Comparison of Phase Terms for Vivaldi Antenna
Figure 32 shows that the all-pass phase of the Vivaldi antenna is a smooth curve showing that the small scale
changes in phase a represented in the minimum phase component. This supports the method used to calculate
the minimum phase components from the magnitude response of the antenna. Another experiment conducted to
confirm the accuracy of the diagnostic tool was measure the 2-port parameters of a link of horn antennas. This
was performed due to the results published in [20] stating that the double ridged horn antennas were very close
to minimum phase. This experiment suffered from the previously mentioned issue in that the phase response
appeared to have a positive gradient due to the large free space gap between the antennas as well as the 9m of
cable which was used to connect the network analyser to the antennas. In this situation, the antennas were
unable to be moved and hence the free space gap was not able to be shortened in order to remove this problem.
A solution was found by setting an electrical delay on the network analyser which was a crude way of removing
the effect of the cabling on the system. The data was then run through the diagnostic tool however the horn
antennas were not shown to be more minimum phase than the UWB antennas used for experimentation. This is
due to the fact that the results published in [21] were for a pair of double ridged horns whereas this experiment
used regular horn antennas.
As mentioned in section XI, the network analysers were observed to have a threshold at which the phase
data became unrecognizable. This point was noticed to be when the magnitude is at approximately -85dB.
Because the network analyser produces the polar coordinates for the S-parameter of a network, the linear
magnitude at -85dB is equal to 5.610
-5
. With such a small linear magnitude, it becomes difficult to measure
differentiations in the phase of adjacent data points leading to a loss of the phase reference.


2 4 6 8 10 12
-0.5
0
0.5
1
1.5
2
2.5
Phase Terms
Frequency, GHz
P
h
a
s
e
,

r
a
d


Antenna
Minimum Phase
All-Pass


Final Thesis Report 2010, UNSW@ADFA

25
XIII. Conclusion
It was found that the diagnostic tool was able to accurately reproduce the minimum phase component of an
antenna phase response using the principle of minimum phase systems. Also, using this minimum phase
component, a theoretical passive equalisation network can be calculated to equalise the minimum phase
component of the antenna phase response. The antenna phase response was able to be split up into the
minimum phase component and the all-pass phase component. However, only the minimum phase component
of the antenna can be equalised using a passive network and hence with an ideal passive equalisation network,
the best possible result is that the antenna phase becomes the same as the all-pass phase. The diagnostic tool is
also able to calculate passive equalisation networks of varying complexities in order to show the effect that the
complexity has on the amount of equalisation. It was found that the Vivaldi antenna outperformed the circular
monopole antenna in almost every aspect. The Vivaldi antenna had a more stable radiation pattern however the
circular monopole had a better impedance match over the UWB frequency range. The Vivaldi antenna also had
showed a greater potential for equalisation over the UWB frequency range and showed a particularly good level
of equalisation between 3.1GHz and 6GHz.

XIV. Recommendations
For future work, this diagnostic tool can be used to characterize other antenna designs according to their
minimum phase properties and move into the synthesis and fabrication of equalisation networks. As well as
continuing the work done in this project, other projects could be done in UWB antennas including the
design/optimization of design of UWB antennas or the signal processing and modulation schemes necessary for
UWB communications.
XV. Acknowledgements
I would like thank Dr. Greg Milford from the School of Engineering and Information Technology at the
University of New South Wales at the Australian Defence Force Academy for supervising this project. Dr.
Milford was extremely helpful in all aspects of the project whether it be technical knowledge or knowledge of
the underlying theoretical concepts and he also passed on many of his exceptional engineering practices and
procedures.




Final Thesis Report 2010, UNSW@ADFA

26
XVI. References

[1] Aiello, R 2003, Ultra-Wideband Wireless Systems, IEEE Microwave Magazine, June 2003

[2] Aiello, R 2006, Ultra Wideband Systems: Technologies and Applications, Elsevier, Burlington, USA

[3] Stutzman,W. L., Thiele, G. A.: Antenna Theory and Design second edition, John Wiley & Sons, Inc.

[4] Oppenheim, A., Schafer, R., 1989. Discrete-Time Signal Processing, Prentice Hall.

[5] McLean, J., Sutton, R., Foltz, H., 2009, Minimum Phase / All-Pass Decomposition LPDA Transfer Functions, IEEE
International Conference on Ultra-Wideband, September 2009, pp. 525 529.

[6] Sibbile, A 2005, Modulation Scheme and Channel Dependence of Ultra-Wideband Antenna Performance, IEEE
Antennas and Wireless Propagation Letters, vol. 4

[7] Yang, Y, et al. The Design of Ultra-wideband Antennas with Performance Close to the Fundamental Limit, Virginia
Tech Antenna Group, Blacksburg, VA, USA

[8] Wong, K.L. High-Performance Ultra-Wideband Planar Antenna Design, Dept. of Electrical Engineering National sun
Yat-Sen University Kaohsiung, Taiwan.

[9] Chen, S.Y., et al. 2006, Unipolar Log-Periodic Slot Antenna Fed by a CPW for UWB Applications, IEEE Antennas and
Wireless Propagation, vol. 5

[10] Xiao-Xiang, HE, 2009, New band-notched UWB antenna, College of Information Science and technology, Nanjing
University of Aeronautics and Astronautics, Nanjing, P.R. China

[11] Zhao, CD, 2004, Analysis on the Properties of a Coupled Planar Dipole UWB Antenna, IEEE Antennas and Wireless
Propagation Letters, vol. 3

[12] Choi, SH, 2003, A new Ultra-Wideband Antenna for UWB Applications, Microwave and Optical Technology Letters,
vol. 40, no. 5, Mar 2004

[13] Mehdipour, A. 2007, Complete Dispersion Analysis of Vivaldi Antenna for Ultra Wideband Applications, Progress in
Electromagnetics Research, pp 85-96.

[14] Hecimovic, N. The Improvements of the Antenna Parameters in Ultra-Wideband Communications, Ericsson Nikola
Tesla, d.d., Croatia.

[15] Liang, J. 2005, Study of Printed Circular Disc Monopole Antenna for UWB Systems, IEEE Transactions on Antennas
and Propagation, vol. 53, no. 11.

[16] Lim, K.-S. 2008, Design and Construction of Microstrip UWB Antenna with Time Domain Analysis, Progress in
Electromagnetics Research, vol.3 pp 153 164.

[17] Pozar, D. Microwave Engineering, Third Edition, John Wiley & Sons, Inc. 2005, pp 187.

[18] Pedro, L.D., et al. 1998, Transmission Line Modelling: A Circuit Theory Approach, Society for Industrial and
Applied Mathematics, Vol. 20 No. 2 June, pp. 347 352.

[19] McLean, J., Sutton, R., Foltz, H., 2009, Minimum Phase / All-Pass Decomposition LPDA Transfer Functions, IEEE
International Conference on Ultra-Wideband, September 2009, pp. 525 529.

[20] Foltz, H., et al. 2007, UWB Antenna transfer Functions Using Minimum Phase Functions IEEE.

[21] Bode, H.W. Network Analysis and Feedback Amplifier Design, D. Van Nostrand, Co., Inc., New York, N. Y., 1945.

[22] ZEIT4223 Engineering Electromagnetics Course Notes. Course offered in session 1 of 2010 at the University of New
South Wales at the Australian Defence Force Academy. Course taught by and notes written by Dr. Greg Milford.

[23] Computer Simulation Technology, Germany, http://www.cst.com

[24] Balanis, A, 2005, Antenna Theory: Analysis and Design, John Wiley & Sons, Inc., New Jersey.



Final Thesis Report 2010, UNSW@ADFA

27
[25] MATLAB function smooth2a. Written by Greg Reeves, CalTech. Obtained from
http://www.mathworks.com/matlabcentral/fileexchange/23287-smooth2a.

[26] LaComb, J, et al. 2009, Ultra Wideband Surface Wave Communication Progress in Electromagnetics Research C,
Vol. 8, pp 95 105.

Das könnte Ihnen auch gefallen