Вы находитесь на странице: 1из 85

Simultaneous Wireless Transmission of

Power and Data Using a Rectenna

Jeroen Theeuwes
Student

Eindhoven, University of Technology


The Netherlands

Report of a graduation project


performed in association with

TNO Science and Industry


De Rondom 1
5612 AP Eindhoven
The Netherlands

Supervisors:
ir. H.J. Visser of TNO, Micro Device Technology group
dr. ir. M.C. van Beurden of TU/e, Electromagnetics chair

period: January 2005 - January 2006

6th March 2006


ii
Abstract

An analytical method is presented to model a system for simultaneous transfer of power and
data. This system consists of two patch antennas, to receive the power and data, a splitter
to extract as much power as possible from the received signals, whilst still being able to
reliably receive the data, and a rectifier to convert the received RF-power in DC-power.

These subsystems are modeled, analyzed, designed, manufactured and measured individu-
ally. The antenna model is used to design a dual-polarized patch antenna with a bandwidth
of 5.9% around 2.45GHz. Another, single-polarized, antenna with a complex impedance
matched to the rectifier has been designed. The presented model is able to predict the
input impedance and radiation pattern within a few percent. This is accurate enough for
our initial design.

A nonlinear diode model is presented which is used to design a voltage-doubler rectifier.


This model is able to determine the diode impedance for the different harmonics within
five percent difference between measurements and modelling results.

A Wilkinson splitter has been designed and compared to an analytical model based on
microstrip transmission line theory. This has resulted in reflection and isolation coefficients
that are accurate within a few percent.

The antenna and rectifier are combined to form a rectenna with a rectenna efficiency of
52% for 0dBm received power. The unloaded DC-voltage is larger than 1.0 V at 0 dBm
received power. The matching circuit between antenna and rectifier has been eliminated
resulting in a rectenna with smaller losses and which is smaller in dimensions than the
conventional rectenna with external matching circuit.

All the subsystems have been combined, which makes it possible to wirelessly receive power
and (AM) data. This system has a rectenna efficiency of 25%. The combination of the
individual models predicts the DC output voltage within 8% percent.
ii
Contents

1 Introduction 5

2 Wilkinson Splitter 9
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Microstrip line theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Microstrip Patch Antenna 19


3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Cavity model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Radiation Pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.1 Radiation of a Single Side . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.2 Total Radiation Pattern . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Input impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4.1 Lossless . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.2 Including Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.3 Input impedance model . . . . . . . . . . . . . . . . . . . . . . . . 31
3.5 Effective dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4 Rectifier 43
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Schottky diode model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3 Nonlinear analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3.2 Runge-Kutta method . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

1
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5 Total System 59
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 Single Rectenna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2.1 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 A rectenna-powered wall clock . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4 Prototype design of the total system . . . . . . . . . . . . . . . . . . . . . 66
5.4.1 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.4.2 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

6 Conclusions and Recommendations 77

2
Acknowledgements

This is the final report of my Master project, performed at TNO Science and Industry, in
Eindhoven, which took me twelve months to complete.

This graduation project concludes my studies of Electrical Engineering at the Eindhoven


University of Technology (TU/e). I have chosen to specialize in the area of electromagnet-
ics, and to perform my graduation project from the Electromagnetics chair. I would like
to thank prof. dr. A.G. Tijhuis of TU/e and ir. J.C. Sirks of TNO Science and Industry
for creating the opportunity to do this project.

During my research I received support from numerous people. Especially I would like to
thank my supervisor ir. H.J. Visser of TNO for his daily support throughout this project,
in particular with the analytical challenges and for his confidence in my capabilities. My
gratitude goes out to dr. ir. M.C. van Beurden of TU/e who, as supervisor, has put a lot
of time and effort in this project, making sure every detail was covered.

Further, I would like to thank ing. G.J. Doodeman of TNO, for sharing his enormous
practical knowledge regarding design issues, and A. Reniers of TNO for his help with the
different measurements.

Finally I would like to thank M. van Lierop, my girlfriend, for all the times she convinced
me that I was able to cope with all the obstacles one stumbles upon during a Master project.

3
4
Chapter 1

Introduction

Nowadays more and more electronic applications become wireless. This adds an enormous
degree of freedom to the experience of the application for the user. Transferring required
data wirelessly is very common nowadays, but electronics need power as well. The power
source is usually included in the device in the form of a battery. This brings along the
disadvantage that the power source becomes exhausted after using it for a while and will
have to be replaced or recharged. It is desirable to transfer power to the device wirelessly
as well, thus bypassing the need for a battery, or to recharge the battery.

In this project it is attempted to supply power to the application using a rectenna. The
word rectenna is a combination of the words antenna and rectifier. Electromagnetic waves
are used to transfer power to the application. Besides the power we would like to transmit
data as well. It is tried to use the same antenna to transmit power as well as data, to
minimize the occupied space.
Another issue with batteries is that they sometimes demand precious space in an appli-
cation. Some antennas can be designed in such a way that they can be integrated in the
application more easily. It is tried to make the rectenna as small as possible, to make it
easier to include a rectenna in a bigger system design.
Sometimes replacing or recharging the battery of an application is hard to do, for example
for active sensors at places that are hard to reach. When these sensors do not need much
power to function, the power can be supplied wireless and with this system the measured
data can be read wirelessly as well. In this way it is no longer necessary to reach for the
sensor once it is placed and no wires have to be connected to it. An example is a sensor
measuring the air pressure in a tire or other sensors in enclosed volumes.

Research in the area of wireless power transfer has been going on for quite a while, see [1].
The idea of combining the data and power is well-known for the normal in-house electricity
network. For wireless applications this is a fairly new research topic. Transferring data

5
and power at the same time can be done by two separate systems, but what we would
like to do is to use one system for both purposes. In this way an optimum combination of
power and data transfer can be achieved.

In this report two frequencies will be considered. In a previous research [2] the license
free 2.4 GHz band was used. Here, we elaborate on this research. Further, 868 MHz will
be considered, because some ultra-low-power applications are available in this band. The
disadvantage of using 868 MHz is that the antenna might become quite large because the
wavelength at this frequency is roughly 35 cm.

A rectenna system consists of three major parts. First of all an antenna is used to collect the
power and the data. After that there is a part that extracts power from the received signals
and recovers the sent data. Finally a so-called rectifier is used to convert the RF power
into usable DC power. More and more circuits need only very little power to operate. It
would be very useful to be able to activate integrated circuits with this system. Integrated
circuits need a certain minimum voltage to operate, some already work at as little as 1.0
Volt. In this project it is attempted to supply an output voltage of at least 1.0 Volt. So we
would like to maximize the output voltage for a given input power. Because only limited
power is allowed to be transmitted in the frequency bands of interest it is assumed that the
received power by the rectenna is 1mW. This power level might seem arbitrary, however it
lies within the range of expected power levels of our applications.

To extract as much power as possible from the received signals while still being able to
collect the data, a so-called Wilkinson Splitter (or combiner) is used. This splitter is used
in a somewhat unusual manner. The typical characteristic of a Wilkinson splitter is that it
possesses a resistive part to dissipate power that would otherwise have been reflected at the
ports. This results in a reflection-free lossy splitter. What we try to do is to use a rectifier
as resistive part and turn the power into direct current (DC ) before it is dissipated. In
this way the dissipated power in the splitter that is normally lost has now become useful.
The splitter is actually used here as a combiner; meaning that two receive antennas will
be connected to the splitters and the combined output of the splitter is the data output.
The power is extracted from within the splitter. This part is analyzed in Chapter 2.
The antenna that will be used is a microstrip patch antenna, which is described in Chap-
ter 3. The main reasons for this choice are its small dimensions, the ease of manufacturing
and the low costs. It is chosen to use a voltage doubler as rectifier, as described in Chap-
ter 4. In this way the output DC-voltage is maximized for a given input power.
Each of these subsystems will be modeled using analytical techniques. For each subsystem
a prototype design will be designed and measured. Finally, in Chapter 5, these subsystems
will be combined to design the total system. The models will be combined and compared
with measurements on the total system.

In this project we have developed mathematical models for the individual systems based on
physical properties. This gives us insight in the factors that determine the behavior of the

6
system. We can rely on computer simulation results, which are in general more reliable.
However, these simulations consume much more time and do not give the desired insight.
The results of our models are verified by these computer simulators or by measurements.

We accept certain accuracy margins for our models. We will use the mathematical models
to design a system optimized at a certain frequency and we accept a frequency offset of
roughly 3 % in a realized design. Further we use the models to predict reflection coeffi-
cients for the individual systems, which define the amount of power that is accepted. For
the antenna design we aim to reach a reflection coefficient of less than -10 dB. For the
Wilkinson splitter we aim to reach a reflection coefficient of less than -20 dB. Both these
values are common in industrial RF-design. We aim to be able to determine the output
DC-voltage and DC-power of the total system within 10 % accuracy.

7
8
Chapter 2

Wilkinson Splitter

As mentioned before we would like to design a rectenna system for simultaneous power and
data transmission. To realize such a system a part is needed which extracts power from the
received signals, while still being able to collect the data that is included in these signals.
A so-called Wilkinson Splitter as described in [3] is chosen for this purpose. A Wilkinson
splitter is used for this purpose because it makes a system more broadband and power can
be extracted from it, to be turned into DC-power, while still being able to receive data. This
power extraction acts as the required resistive part in the splitter.

2.1 Introduction

A Wilkinson Splitter may be realized as a Printed-Circuit-Board (PCB)-based splitter,


formed by microstrip lines. In Fig. 2.1 the schematic layout of a Wilkinson Splitter is
shown. The resistive part, R, in this kind of splitter is used to prevent any mismatches
in the system by dissipating power. We would like to extract this power and turn it into
DC power. We are actually using this splitter as combiner, using two ports of the splitter
as input and the combined port as output. However, due to reciprocity the splitter is
analyzed as splitter instead of combiner, because this is the most common way.

2.2 Analysis

Microstrip lines can be treated as simple transmission lines as shown in [4]. Power division
by transmission lines is achieved by using transmission lines with different characteristic

9
Figure 2.1: Basic layout of a Wilkinson splitter

impedances. A Wilkinson splitter is used to divide power in a predefined way. The power
is provided at port 1 and the output ports are port 2 and port 3. The splitter is designed
in such a way that the power out of port 3 is K 2 times that out of port 2. If Zin2 and Zin3
are the input impedances right after the tee (the point where actual splitting takes place)
and P2 and P3 are the output powers then it can be derived from transmission-line theory
that
P3 Zin2
= K2 = . (2.1)
P2 Zin3

Furthermore, the splitter should be designed in such a way that an impedance match is
seen at port 1. This means that the input impedance at the tee should be equal to Z0 , the
characteristic impedance of port 1. This impedance is the parallel impedance of Zin2 and
Zin3 . So

Zin2 · Zin3
= Z0 ,
Zin2 + Zin3

combining this with Equation (2.1) shows us that

K 2 · Zin3
= Z0 . (2.2)
1 + K2
A resistive element is placed between the ends of arm 2 and arm 3. When the total system
is matched and the lengths of arms 2 and 3 are a quarter of a wavelength the voltages at
the ends of arm 2 and 3 are equal. At the desired frequency no current flows through this
resistive element and no power is dissipated here, because there is no voltage difference
over this resistor. When a mismatch occurs due to a frequency shift or a mismatch in

10
the impedance levels, power is dissipated in this resistive element instead of reflection of
power at the different ports. Further, two extra arms are included after arms two and three
to transform the output impedance levels back to Z0 (or any other desired characteristic
impedance). Quarter-wavelength transmission line impedance transformers are used for
this purpose, the theory of which described in [5].
To satisfy all the conditions stated above the lengths of arms 2, 3, 4, and 5 should be a
quarter of a wavelength. Arms 2 and 3 should be this length to achieve the equal voltage
of the resistor at the desired frequency. The length of arms 4 and 5 should be a quarter of
a wave length to perform the impedance transformation. The wavelength, λ, is defined as

the wavelength in the substrate. This wavelength is determined using λ = c/(f εe ) and
the procedure described in [6]. The characteristic impedances of the different arms should
be
r
p 1 + K2
Z02 = Z0 K(1 + K 2 ), Z03 = Z0 ,
K3
and
√ Z0
Z04 = Z0 K, Z05 = √ . (2.3)
K
The analysis of the impedances of arms 4 and 5 and the impedance of the resistive part is
described thoroughly in [3]. There it is shown that this choice of characteristic impedances
results in a identical voltage level at both sides of the resistor. This results in no power
dissipation at the desired frequency. The resistive part of the splitter should be
1 + K2
R = Z0 . (2.4)
K
This value yields infinite isolation and a perfect match at the center frequency.

2.3 Microstrip line theory

As shown in the preceding section, the splitter that we would like to use consists of diffe-
rent microstrip lines with each its own characteristic impedance. These microstrip lines are
formed by a PCB. A PCB consists of different layers. First there is a conducting ground
plane, usually copper. In our model this layer is assumed to be perfectly conducting. On
top of the ground plane a dielectric medium is present with a certain thickness, h. This
substrate has a dielectric constant, εr , which determines the electromagnetic behavior of
the substrate. On top of that another conducting plane is positioned. By giving the top
layer certain shapes transmission lines can be formed. The characteristic impedance of this
microstrip line is determined by its width, the thickness of the substrate and the permit-
tivity of the substrate. Given the height, permittivity of the substrate, and K 2 , a model

11
for the characteristic impedance of a microstrip line is used to determine the width of each
arm of the splitter.
To determine the characteristic impedance of a microstrip line the effective dielectric con-
stant εe has to be determined. A microstrip line has a dielectric between the strip and the
ground plane but it has air above the strip. The field is traveling in the dielectric as well
as in the air. The effective dielectric constant is an approximation of the average dielectric
constant of the two materials. So the effective dielectric constant will always be bounded
by εr and 1. Using this frequency-dependent effective dielectric constant the wavelength in
the medium can be determined. The effective dielectric constant is determined according
to [7] and the characteristic impedance is determined by the model described in [8]. These
models are numerical models based on empirical relations.
From the effective permittivity the wavelength in the substrate can be determined. If c is
the speed of light and f represents the working frequency then
c
λ= √ . (2.5)
f εe

The models that are used to determine the dimensions of the different pieces of microstrip
line first calculate static values for the characteristic impedance and the wavelength and
from these values the frequency dependent values are determined. The validity of these
models is checked using the full-wave simulator Ansoft Ensemble. This simulator deter-
mines the behavior of a RF structure using a method-of-moments analysis. The model
used agrees very well with the simulation results, as can be seen in Fig. 2.2(b). For our
frequency range of interest the impedance is predicted within 3 %.

2.4 Design

To be able to determine the accuracy of the model presented in the previous sections, several
Wilkinson splitters have been produced. To determine the dimensions of the different
pieces of the splitter, one has to know some parameters of the material to be used. We
have chosen to use a FR4 PCB for the splitter. This material has a height of 1.6 mm, a
top and ground layer made of copper which has a conductivity σ = 5.8 · 107 Ω−1 m−1 . The
substrate has a dielectric constant between 3.5 and 4.5. The dielectric constant is not the
same for each slab because the distribution of the different materials in the substrate can
be different between different batches. This dielectric constant is assumed constant over a
single slab. So first of all the dielectric constant has been determined by measuring a piece
of transmission line etched from this material.
The structure used to determine the dielectric constant consists of a piece of transmission
line of roughly 50 Ohm (determined using the average value of εr for this material), followed
by a piece of transmission line with a certain width, different from the 50 Ohm piece. The
line is terminated with a 50 Ohm load. The initial piece of the 50 Ohm transmission line

12
(a) Layout (b) Results

Figure 2.2: Estimation of εr by comparison op input impedances

is used to minimize unwanted reflections at the connector due to a change in impedance.


In this way the beginning of the line is matched to the connector. For this microstrip-
line measurement the reflection coefficient is measured at the beginning of the line by a
network analyzer. From this reflection coefficient the input impedance is determined. The
transmission line is measured over a broad frequency range, from 10 MHz to 4 GHz, so a
wide range of reflection coefficients is available. The model for determining Z0 and εe is
evaluated for a range of εr and for each value it is evaluated how well the measured and
calculated input impedances fit. In this way the value of εr for the material is chosen to
be the one that coincides best with the measured values. In Figure 2.2(a) the piece of test
microstrip line is shown.

When designing such a Wilkinson splitter one encounters some physical challenges. The
resistor that is placed between the two arms must be connected directly to the microstrip
lines. Otherwise another piece of transmission line has to be included. So after the tee the
two transmission lines should bend back towards to each other. We have chosen not to use
elliptical (or any other techniques that does not use right angle bends) arms, although the
bend backwards can be constructed more easily with them. However, the length and width
of these lines can not be determined unambiguously. Rectangular microstrip transmission
lines are used in our case, as shown in Figure 2.3. In case the splitter is designed to resonate
at 2.4 GHz the wavelength is roughly c/2.4 · 109 ' 12.5cm or less (because of the effective
dielectric constant). So when an arm of the splitter should be λ/4 long, the physical length
of an arm is about 3cm. This length is not enough to make the complete bend backwards.

13
(a) 868 MHz initial design (b) 868 MHz improved design

Figure 2.3: Design of a Wilkinson splitter, K 2 = 2

So the lengths of arms 2 and 3 are chosen to be 5λ/4.

The design of a 2.4 GHz splitter is shown in Figure 2.3(a). A drawback of this design
is that the splitter becomes more narrowband, compared to the design with λ/4 arms,
because we are actually using the second operating frequency. If fdes denotes the desired
frequency, then the first operating frequency is fdes /5, this is the frequency where arms 2
and 3 are exactly λ/4.

Three splitters have been designed and manufactured using this approach. A 1:1, a 1:2
and a 1:3 splitter have been manufactured.

By using the εr found with the previous measurements the dimensions for the splitter
can be determined. The design of this splitter requires the transmission lines to bend a
few times. These right-angle bends introduce some capacitive and inductive effects. To
decrease the influence of these effects, especially the capacitive effect, on the performance
of the splitter a part of the bend is cut off. This cut-off bend is called a miter and is
described in [9]. Furthermore, one should make sure that the supply lines at the ports are
long enough for the higher order modes, arising from steps in width, to vanish. One of the
designed splitters is shown in Fig. 2.4.

14
Figure 2.4: A 1:3 Wilkinson Splitter

2.5 Results

First of all a piece of transmission line is measured to be able to determine εr . The results
of the measurement and the estimation of εr are shown in Fig. 2.2(b). We can see that the
model agrees very well with the measurements, although for frequencies above 2.5 GHz
the model differs more and more from the measurements. For our frequencies of interest
the error is smaller than 2 %. This is mostly due to the fact that this transmission line
is created cutting copper tape in the desired shape. This is a really quick but not very
accurate way of creating these kind of structures. Because we are mainly interested in
frequencies below 2.5 GHz this method for determining εr is satisfying. The estimation of
the dielectric constant resulted in a value for εr of 4.3.

With this dielectric constant the dimensions of the different splitters have been determined
and several splitters with different power ratios have been created. Three splitters have
been created to be used at 2.4 GHz and two splitters have been created for 868 MHz
(K 2 = 1 and K 2 = 2). The splitters designed for 2.4 GHz have been measured over a
frequency range from 2.0 GHz to 2.8 GHz. Some results of these measurements for K 2 = 2
are shown in Fig. 2.5. The reflection coefficient (S11) and the isolation (S23, not shown
here) agree reasonably well with the model, but it can be seen that the S33 parameter
does not show the expected behavior. We see that the levels do agree but the shape does
not. However, the reflection coefficients are small enough to be accepted to be used further
for our purposes as explained in Chapter 1. It turns out that this error can be explained

15
(a) S11 reflection (b) S33

Figure 2.5: Results of the 2.4GHz K 2 = 2 splitter

by the higher order modes introduced at the bend near the resistor (verified by simulation
results). At this bend a change in line width and direction is formed.

To prevent the propagation of higher-order modes in the splitter the initial design is altered
at the bend near the resistor, additional lines are added to sufficiently suppress the higher
order modes caused by the bends. An example of this new design in given in Fig. 2.3(b).
This is a design of a splitter to be used at 868 MHz with K 2 = 2. Furthermore, because
this splitter is designed to be used at 868 MHz, the wavelength increases and it is not
necessary to make arms 2 and 3 five quarters of a wavelength, so this splitter will be more
broadband, as explained before. This splitter design has been used to produce a K 2 = 1
and a K 2 = 2 splitter. These new splitters have also been measured. In Fig. 2.6 some of
the results for the K 2 = 2 splitter are shown. Again, the results of the 2.4 GHz splitter are
shown, however the results for the 868 MHz splitter are similar. Now it can be seen that
all measured scattering parameters show the expected behavior and are within the design
margins of Chapter 1.

16
(a) Reflection coefficient S11 (b) Isolation S23

(c) Reflection coefficient S22 (d) Reflection coefficient S33

(e) Power division

Figure 2.6: Results of the new 2.4 GHz K 2 = 2 splitter


17
2.6 Conclusion

In this chapter the Wilkinson Splitter theory has been discussed and it has been shown how
these splitters have been designed. These splitters are to be used as a part of a system to
simultaneously receive power and data. After an initial design an improved design has been
presented to prevent higher-order modes from propagating. It has been shown that the
model satisfies our design demands. Further it has been shown that an impedance match
is seen at the desired frequency at ports 2 and 3, where the antennas will be connected, so
there will be no power reflected towards the antennas.

18
Chapter 3

Microstrip Patch Antenna

As mentioned before we would like to build a system which is powered and receives data
wireless. To receive this power and data an antenna is needed. We have chosen to use a
microstrip patch antenna. The main reasons for this choice are its small dimensions, the
ease of manufacturing and the low material costs. The antenna described in this chapter
is designed to have an input impedance of 50 Ohm and to work in the license free 2.4 GHz
ISM (Industrial, Science and Medical) band. This is done because of practical reasons:
there are many standard components for the 2.4 GHz band, such as power amplifiers and
measurement equipment. Further, almost all measurement equipment use 50 Ohm charac-
teristic impedances. However, the theory used in this chapter can be used to achieve any
desirable antenna input impedance and resonance frequency (within practical limits).

(a) Microstrip inset feed (b) Microstrip edge feed (c) Probe feed

Figure 3.1: Different ways of feeding a microstrip patch antenna

19
3.1 Introduction

A patch antenna, in its basic form, consists a of conducting ground plane, an intermediate
substrate and a conducting top layer. This top layer is called the patch and is usually
photoetched from a copper clad PCB. Some examples of patch antennas are given in
Figure 3.1. In the model that is presented in this section it is assumed that both the top
layer and the ground plane are perfectly conducting. The size of the patch is denoted by
a and b as shown in Figure 3.2.

A patch antenna can be fed in several ways. In Figure 3.1 some examples are shown; a
probe feed; an edge feed, and an inset feed. All these kind of feeds can be used with the
model presented in this chapter, however the model can only handle the inset feed under
certain conditions. These conditions are given in Section 3.2. The method used assumes a
probe or a piece of transmission line to excite the patch at a certain location, this location
is denoted by (x0 , y0 ). When an edge-fed microstrip patch antenna is used, either x0 or y0
is zero. The feed has certain dimensions denoted by dx and dy . In the case of a probe-fed
antenna these dimensions are chosen in such a way that dx = dy and dx × dy is the cross
section of the probe. When the patch antenna is fed with a microstrip transmission line
either dx or dy (depending on a horizontally or vertically aligned microstrip line) is set to
the width of the microstrip and the other one is set to zero. There are other methods for
feeding a microstrip patch antenna (such as slot coupled feeding) which are beyond the
scope of this report, and these methods require major changes in the model.

Important parameters of an antenna are its input impedance and its radiation pattern.
Both will be treated in this chapter. First a model will be presented to calculate these
parameters and next calculated parameters will be compared with measured values.

The modeling presented in this chapter follows the same approach as the one presented
in [2], that dealt with other aspects of the same subject. However, the model presented
here is more extensive, since the model shown here can handle microstrip feed lines as well,
something that was not included in [2], where only probe fed microstrip patch antennas
were treated.
Further, the model shown here has been improved to determine the radiation and gain
pattern much more reliable, comparing Section 3.6 with Section 2.5 of [2]. The reason for
this improvement is not clear.

20
Figure 3.2: Layout of a microstrip fed patch antenna

3.2 Cavity model

A patch antenna can be seen as a radiating cavity. This way of analyzing a patch antenna
was introduced by Richards et al in [10] and was improved by Carver and Mink in [11]. To
analyze a patch antenna the Maxwell’s equations are used, i.e.,

∇ × E = −jωµH ,
∇ × H = J + jωεE , (3.1)

assuming an ejωt -time dependence. In these equations E denotes the electric field strength,
H the magnetic field strength, ω the radial frequency, J the electric current source and µ
and ε represent the permeability and the permittivity of the medium. The height of the
antenna is small compared to the wavelength. This results in electromagnetic fields in the
antenna that are (almost) independent of z. The resulting fields will be transverse magnetic
(TM) with respect to the z direction, since an electric field is applied in z direction. Since
the electric field is alligned purely vertically the Helmholtz equation for the z direction is
given by
∂Ez ∂Ez
+ + k 2 Ez = jωµJz , (3.2)
∂x2 ∂y 2

where k = ω µε = 2π λ
. In this expression λ represents the wavelength in the dielectric,
between the patch and the ground plane.

The cavity is assumed to have a perfectly electrically conducting patch and ground layer.
The feed is placed in such a way that when the patch is excited, the electric field lines will

21
be vertically directed. When the electrical field lines are vertically directed the side walls
may be assumed to be perfectly magnetically conducting. These assumptions result in a
complete set of boundary conditions for the cavity. First of all the electric field, E, can
only have a z component at the top and bottom layer. Secondly the magnetic field, H,
can only have a component notmal to the side walls of the cavity. Thus,

n × H = 0 at n = ux ,
n × H = 0 at n = uy ,
n × E = 0 at n = uz . (3.3)

These boundary conditions and the Helmholtz equation (3.2) can be used together to
determine the fields inside the cavity.

First, the homogeneous wave equation


∂Ez ∂Ez
2
+ 2
+ k 2 Ez = 0 (3.4)
∂x ∂y
is evaluated. This equation describes the electric field inside the cavity when sources are
absent. The solution of this equation under the boundary conditions of equation (3.3) is
of the form

Ezmn = Amn ψmn (x, y). (3.5)

In this equation Ezmn represents the z component of the electric field of mode mn. The ψmn
functions are the eigenfunctions of the left-hand side of Equation (3.4). The electric field
is split up in different modes. Each mode represents a given number of half patch lengths
in the x and y direction of the cavity. This way only discrete solutions (m,n) exist. An
example of a m = 2 and n = 1 mode, called (2,1)-mode is given in Figure 3.3(a). Applying
the first two boundary conditions of Equation (3.3), using separation of variables, results
in

ψmn (x, y) = cos(km x) cos(kn y), (3.6)

where

km = ,
a

kn = . (3.7)
b
These last two expressions for km and kn represent the (transversal) wavenumber compo-
nents of the m, n mode in the cavity in the x or the y direction. When Equations (3.5)
and (3.4) are combined with the expression for the wavenumber k the resonance condition
can be derived:

k 2 = km
2
+ kn2 = kmn
2
. (3.8)

22
(a) TM21 mode (b) Magnetic surface currents

Figure 3.3: An example of a cavity with E fields and equivalent magnetic surface currents

Inside the cavity multiple modes can coexist. The total electric field is then simply the
sum of the electric fields of the individual modes.

So far an expression has been derived for the homogeneous Helmholtz equation inside the
cavity with boundary conditions. When a source is present in the cavity the correspond-
ing Helmholtz equation is Equation (3.2). The solutions of the homogeneous Helmholtz
equation can be substituted in this equation. Together with the summation of the different
modes this results in
X X ∂E mn ∂E mn
( z2 + z
2
+ k 2 Ezmn ) = jωµJz (3.9)
m n
∂x ∂y

When Equations (3.4) and (3.8) are combined this results in

∂Ezmn ∂Ezmn 2
2
+ 2
= −kmn Ezmn . (3.10)
∂x ∂y

Substituting this relation in Equation (3.9) gives the expression


XX
(k 2 Ezmn − kmn
2
Ezmn ) = jωµJz (3.11)
m n

This expression can be used to determine the amplitude coefficients, Amn , of Equation (3.5).
To do so we, integrate this expression over the total volume of the cavity and multiply it
by ψm0 n0 , independent of the ψmn of Equation (3.5). This results in
Z XX Z
2 2
(kmn − k )Amn ψmn ψm0 n0 dV = −jωµ Jz ψm0 n0 dV. (3.12)
V m n V

23
The left-hand side of this expression involves an integral of two independent ψmn functions.
This part of the equation is treated first. Because the modes are orthogonal the summation
can be left out and the integral can be evaluated for each single combination of {m, m0 }
and {n, n0 }. The part of Equation (3.12) where ψmn ψm0 n0 is integrated can be written as
Z Z a Z b Z h
ψmn ψm0 n0 dV = ψmn ψm0 n0 dxdydz
V x=0 y=0 z=0
for m 6= m0 n 6= n0

0 or
= (3.13)
abh
χ2mn
for m = m0 and n = n0 .

In this equation

 1√ for m = 0 and n = 0
χmn = 2 for m = 0 or n=0 (3.14)
2 for m 6= 0 and n =
6 0.

Substituting these expressions in Equation (3.12) yields


Z
2 2 abh
(kmn − k )Amn 2 = −jωµ Jz ψmn dV. (3.15)
χmn V

This equation must be satisfied for each single mode, so the amplitude coefficients are given
by

χ2mn
Z
ωµ
Amn = j Jz ψmn dV. (3.16)
abh k 2 − kmn
2
V

Now that the left-hand side of Equation (3.12) has been treated, we are left with the right-
hand side of the equation. This side involves an integral over the current density inside
the cavity. This current density is caused by the feed of the patch antenna. As mentioned
before there are different ways of feeding the patch antenna. The ways treated here are a
probe feed, an inset microstrip feed and an edge microstrip feed.
The model is only capable of accurately analyzing a patch antenna with an inset feed when
the radiating modes of the antenna do not result in a varying Ez field in the direction
perpendicular to the feed line. This way the feed line does not disturb the electric field in
the cavity, compared to a probe feed. Further, some capacitive coupling will arise with an
inset feed, between the feed line crossing the patch and the patch itself, which make the
results of the model less reliable.
Feeding a patch antenna with a microstrip line comes down to impressing a voltage at a
certain place at the patch. This place is defined as (x0 , y0 ) and has dimensions, dx × dy ,
with one of these to set to zero. This impressed voltage is transformed to a current density
J e , using Huygens theorem, J e = n × H. The magnetic field in the cavity, under the
microstrip line is directed perpendicular to the feed line in the horizontal plane, so the
equivalent electric current is directed in the z direction. This process is described in [4].

24
When a probe is used to feed the antenna an electric current can be defined directly as
the physical current flowing through the probe. Again, (x0 , y0 ) is the position of the probe
and the dimensions are dx × dy . When a microstrip line feed is used either dx or dy is set
to one. These dimensions are set as described in Section 3.1. The excitation current is
defined as
∧ |y − y0 | 5 dy
 I0
dx dy
for |x − x0 | 5 dx
2 2
Jz = (3.17)
0 elsewhere.

The integral of the right-hand side of Equation (3.15) can now be determined as
 
nπdy
mπdx sin

sin
Z
2b
−jωµ Jz ψmn dV = −jωµI0 mπd2a x nπdy
V 2a
 2b  
mπdx nπdy
= −jωµhI0 sinc sinc
2a 2b
= −jωµhI0 Gmn . (3.18)

Now the electric field inside the cavity canpbe determined when Equations (3.5), (3.16)
and (3.18) are combined. When using η = µε , the intrinsic impedance, the electric field
inside the cavity of the microstrip antenna can be written as
XX
Ez = Amn ψmn (x, y), (3.19)
m n

with
ηχ2mn k
Amn = jI0 2 2
ψmn (x0 , y0 )Gmn . (3.20)
ab k − kmn

3.3 Radiation Pattern

So far, we have determined the electric field inside the cavity. As mentioned before we
would like to determine the radiation pattern of the microstrip antenna. The electric field
in the cavity is used to find this radiation pattern. To calculate the radiation pattern, the
cavity is modeled by four independent radiating apertures. The radiation pattern can be
determined from the equivalent magnetic current density on the side walls of the cavity.
These equivalent magnetic current densities are defined as M = E × n, as described in
[12]. An example of these magnetic surface currents is shown in Fig. 3.3(b).

25
3.3.1 Radiation of a Single Side

To determine the radiation pattern of a patch antenna, first the radiation pattern of one
single side is calculated. Next, these radiation patterns are combined to form the total
radiation pattern. As mentioned before the equivalent magnetic surface current is used to
determine this radiation pattern. One single side is modeled as an aperture above a per-
fectly conducting ground plane, stretching to infinity. The consequences of the assumption
of an infinite ground plane, as described in [13], are mainly the absence of edge diffractions,
which do not harm the functionality of our antenna . This allows us to define the electric
vector potential at any given location, as described in [14], as
e−jk0 R 0
Z
ε0
F = Ms dS . (3.21)
4π S R
In this equation ε0 represents the permittivity of free space and k0 is the free-space wave
number. M s is the aforementioned surface current density. The integrand is evaluated
over the total area of a radiating aperture and R represents the distance between the
observation point and a point on the side wall.

We are mainly interested in the far-field radiation pattern of the antenna. This allows us
to perform some simplifications in Equation (3.21). We can use r, the distance from the
origin to the observation point, instead of R, in the denominator. The R in the phase term
of the exponent can be approximated by R ≈ r − r0 cos(ξ), as described in [14]. Where r0
represents the distance from the origin to the source point on the aperture and ξ is the
angle between the lines from the origin to the observation point and from the origin to the
source point on the aperture.

To transform the vertically directed electric field lines into the magnetic surface current
the relation M s = 2E × n is used. The fact that the amplitude of the magnetic surface
current correspond with twice the electric field strength is due to mirroring in the ground
plane. When we take for example the x = 0 side of the cavity, without losing generality,
the electric field in this aperture can be written as
 nπy 
E aperture = Amn cos az . (3.22)
b
Here, Amn is the amplitude coefficient as given in Equation (3.19), which depends on the
mode number and the frequency. Combining Equations (3.21) and (3.22) the electric vector
potential is found to be
ε 0 ay h
Z Z b  nπy  e−jk0 [|r|−|r0 | cos(ξ)]
F = Amn cos dy 0 dz 0 . (3.23)
2π z0 =0 y0 =0 b r
To evaluate this integral in Cartesian coordinates the relation
|r0 | cos(ξ) = r0 · ar ,
= y 0 sin(ϑ) sin(ϕ) + z 0 cos(ϑ) (3.24)

26
can be used, with ar the unit vector pointing from the origin to the observation point.
Now, Equation (3.23) can be written as
Z h Z b
ε 0 ay −jk0 r
 nπy  0 0
F =j Amn e cos e−jk0 [y sin(ϑ) sin(ϕ)+z cos(ϑ)] dy 0 dz 0 ,
2πr z=0 y=0 b
ε0 bhay n −j2Y
Y [(−1) e − 1]
=j Amn e−jk0 r 2 2 2
sinc(Z)e−jZ , (3.25)
πr 4Y − n π
with
b
Y = k0 sin(ϑ) sin(ϕ),
2
h
Z = k0 cos(ϑ). (3.26)
2
Now that we have determined the electric vector potential the relation
1
E rad = ∇×F (3.27)
ε0
can be used to determine the radiated electric far field. The radial component of E rad
can be neglected compared to the ϑ and ϕ components, due to the radiation properties of
electromagnetic waves.
Using ay = sin(ϑ) sin(ϕ)ar + cos(ϕ)aϕ + cos(ϑ) sin(ϕ)aϑ the radiated electric field in the
far-field region can be expressed as
k0
E rad = −j Fy (cos(ϑ) sin(ϕ)aϕ + cos(ϕ)aϑ ). (3.28)
ε0

3.3.2 Total Radiation Pattern

We have calculated the radiated field of a single side of the microstrip patch antenna and
now we can combine the radiated fields of the different side walls to determine the total
radiated field. To do so we employ array theory, as the total radiated field consists of
contributions from two arrays. Each array consists of two opposing apertures of the cavity.
Two opposing apertures radiate the same field except for a phase shift. This phase shift
occurs due to the spacing of the radiating slots and another phase shift of π radians might
occur depending on the mode mn. As described in [5] the array factor (AF ) of two identical
radiating sources, with phases α1 and α2 and d as the vector between two similar points
on the different sources is
 
k0 d · r α1 − α2 j α1 −α2
AF = 2 cos + e 2 . (3.29)
2 2

27
The total radiated field can now be determined as
E rad = E arad AFa + E brad AFb . (3.30)
In this expression Equation (3.28) is used to determine the individual radiated fields of the
apertures, expressed as
k0 bh −jk0 r Y [(−1)n e−j2Y − 1]
E arad = e Amn sinc(Z)e−jZ (cos(ϑ) sin(ϕ)aϕ + cos(ϕ)aϑ ),
πr 4Y 2 − n2 π 2
k0 ah −jk0 r X[(−1)m e−j2X − 1]
E brad = e Amn 2 2 2
sinc(Z)e−jZ (cos(ϑ) cos(ϕ)aϕ − sin(ϕ)aϑ ).
πr 4X − m π
(3.31)
Here
AFa = 2 cos(X − αm )ejαm ,
AFb = 2 cos(Y − αn )ejαn , (3.32)
a
X = k0 sin(ϑ) cos(ϕ),
2
b
Y = k0 sin(ϑ) sin(ϕ),
2
h
Z = k0 cos(ϑ), (3.33)
2
and
π

2
for i even,
αi = (3.34)
0 for i odd.
With these expressions the total radiated field of a microstrip patch antenna can be de-
termined for a single mode m, n. The radiated field of multiple modes can be found by
adding the fields of individual modes.

To determine the total radiated power of a patch antenna on an infinite ground plane the
radiated field must be integrated over a hemisphere, so
Z π/2 Z 2π
1
Prad = |E rad |2 r2 sin(ϑ)dϕdϑ, (3.35)
ϑ=0 ϕ=0 2η0
p
where η0 = µ0 /ε0 represents the intrinsic impedance of vacuum.

3.4 Input impedance

Another important property of an antenna is its input impedance. If it is possible to


determine this impedance accurately the feed impedance can be matched to the antenna’s
input impedance and maximum power transfer can be accomplished.

28
3.4.1 Lossless

As described in [15] the input impedance is defined as


Z
1
Zin = − 2 E · J ∗ dV. (3.36)
|I0 | V

In this expression I0 represents the incident current at the feed of the antenna. If we use
Equation (3.19) in the expression and perform the integral over the cavity, the resulting
input impedance will be purely imaginary because no losses have been taken into account.
For each mode a reactance, Xmn , can be determined and the total input impedance is the
sum of these individual impedances.
X X ηχ2 k
mn 2
Zin = −j ψmn (x0 , y0 )Gmn
m n
ab k2 2
− kmn
XX
= −j Xmn . (3.37)
m n

These reactances, Xmn , can be seen as an inductance and a capacitance in parallel. For
each mode such an equivalent circuit can be modeled. The values of the inductance and
capacitance then will be

√ ηχ2mn 2
Lmn = µε
2
ψmn (x0 , y0 )Gmn
abkmn
√ ab 1
Cmn = µε 2 2
. (3.38)
ηχmn ψmn (x0 , y0 )Gmn

Whenever the wavenumber in the dielectric equals a modal wavenumber, so at a modal


resonance, this input impedance reaches infinity, which will not happen in practice. The
reason for this non-realistic behaviour is found in the neglection of the losses thus far,
including the radiation losses, meaning that the fields have been assumed to be restricted
to the cavity.

3.4.2 Including Losses

In the preceding section no losses were taken into account. Some losses have to be included
in the model through, for example radiation loss is an important non-negligible property
of an antenna. Further, there are other losses, such as dielectric losses of the substrate
and conduction losses of the patch and the ground plane. To include the losses in the
aforementioned equivalent circuit of parallel inductance and capacitance for a particular
mode a resistance is added as shown in Figure 3.4. The input impedance of this circuit is

29
Figure 3.4: Impedance circuit model for a single mode

jωRL
Zin =
R− ω 2 RLC + jωL
jω/C
= . (3.39)
ω2 − ωr2 (1 + j/Q)

In this expression ωr represents the resonance frequency ωr = √1 and Q equals the quality
LC
R
factor Q = ωL .

Since we have no explicit expression for the losses in the system we use the quality factor
to determine the losses. The quality factor can be determined from field concepts, meaning
magnitude of reactive power
Q= . (3.40)
magnitude of dissipative power
We can determine the quality factor for each individual loss factor and combine these
individual quality factors to a total quality factor which will be explained later. First of
all we can determine the quality factor of the radiation losses as
2ωWE
Qr = . (3.41)
Prad
In this expression Prad is the total radiated power as determined in Equation (3.35). The
stored electrical energy, WE , is determined as
Z
WE = ε|E|2 dV,
V
abhε
= |Amn |2 . (3.42)
χ2mn

30
To determine the quality factor for the conduction losses in the patch and the ground layer
we use the expression that can be found in [16], i.e.,
r
ωµ0 σ
Qc = h . (3.43)
2
Here, σ represents the conductance of the ground plane and patch layer. The last type of
losses we include in the model are the dielectric losses given by
1
Qd = . (3.44)
tan(δ)
σd
In this expression tan(δ) represents the loss tangent of the dielectric. Further, tan(δ) = ωε
,
with σd the conductivity of the dielectric.

Now the total quality factor can be determined using the relation
1 1 1 1
= + + , (3.45)
Q Qr Qc Qd
as described in [15].

Now we can determine the input impedance for each single mode including losses, using
Equation (3.39). To determine the total input impedance the individual impedances can
simply be added up.

3.4.3 Input impedance model

In the way described in the preceding section an accurate estimation for the input impedance
can be found. However, calculating the input impedance in this way requires quite some
computation time, especially solving Equation (3.42) is computationally expensive. This
computation time can be decreased by making some assumptions about the different modes,
because the microstrip patch-antenna’s input impedance is mainly (but not completely)
determined by the dominant or resonant mode(s).

First of all the TM00 mode will have no inductance and a capacitance of C00 = abε h
,
according to Equation (3.38). So the impedance of the TM00 is modeled as a parallel RC
circuit with impedance
j/ωC
Z00 = . (3.46)
1 − j/Q
Here, Q is calculated from Equation (3.45). The TM00 mode will hardly radiate any energy
and therefore Qr can be set to zero.

31
Figure 3.5: Impedance model for a microstrip patch antenna

Next, the radiating modes of course do have radiation losses and also have inductances. So,
for these modes the complete RLC circuit has to be determined. The so-called higher-order
modes, which are non-radiating modes, can all be added up into one single RL parallel
circuit. As described in [10] the inductance of this impedance can be determined as
p √
XL = µ/ε tan(ωh µε) (3.47)

and the resistance of the higher order modes can be determined using Equation (3.45) with
Qr = 0, since these modes do not radiate. The higher-order modes impedance can then be
calculated as
jωL
Zho = . (3.48)
1 + j/Q
Now we can derive the total input impedance of the microstrip patch antenna with sim-
plifications concerning non-radiating modes as shown in Figure 3.5. The losses are used to
determined an effective k to be used in Equation (3.20) so this value can not reach infinity.

3.5 Effective dimensions

The model for the input impedance and the radiation pattern of a microstrip patch an-
tenna as described in the preceding sections has been implemented and analyzed. The
results from this model have been compared to full wave simulations performed with An-
soft Ensemble. It turns out that the model does not agree with the full wave simulations,
the result forms are similar but we observe a considerable offset. The resonance frequency
as well as the radiation pattern of the presented model do not match with the full wave
simulations. Commercially available full-wave simulators have shown to be able to predict
the electromagnetic behaviour of antennas with an excellent degree of accuracy. It has
been checked that the simulations converge. So the discrepancy between the model and
the full-wave simulation is caused by imperfections in the cavity model.

32
Figure 3.6: Electric fringe fields of the cavity

This discrepancy is caused by the so-called fringe fields of the antenna, [17]. These fringe
fields are responsible for the radiation of the patch antenna. It has been suggested before
that the electric field lines are perfectly vertically directed in the cavity of the patch.
Besides these fields some fringe fields outside the cavity will exist as well. These fields
bend outwards of the cavity as shown in Figure 3.6, where a side view of the cavity is
given. To account for these fringe fields an extension of the dimensions of the cavity is
introduced. It is assumed again that all the electric fields inside the (extended) cavity are
vertically directed using these extensions. These same fringe fields occur at an open-end
microstrip line and the effect of these fringe field is described in [10]. The extension length,
δ, for this case is given by

(εef f + 0.3)(w/h + 0.262)


δ = 0.412h ,
(εef f − 0.258)(b/h + 0.813)
(3.49)

with
εr + 1 εr − 1
εef f = + (1 + 10h/b)−1/2 , (3.50)
2 2
where w is the width of the microstrip line. When the effective length in the length or
width direction of the patch is to be determined w is replaced with the length or the width
of the patch. The extra length δ is added twice in each direction because the patch has
two open ends in each direction.

33
The dimensions of the patch as analyzed in the preceding model are now as shown in
Figure 3.7. These effective dimensions are still an approximation to account for the extra
electric field lines in the cavity and the extensions suggested here are not always correct.
In Section 3.6 it is explained how the approximation can be improved.

Figure 3.7: Effective dimensions of the patch

3.6 Results

The model that has been presented in the preceding sections has been used to design
microstrip patch antennas. The starting point of the first design is to create an antenna
that is as broadband as possible to be able to receive power over a frequency range as wide
as possible. Further, it is intended to design a patch antenna that is fed by a microstrip
line, because of the ease of manufacturing (it is even possible to use copper tape on a

34
grounded RF substrate). Further, it is chosen to use an edge feed, because an inset feed
introduces some capacitance between the feed line and the patch itself which can not be
included in the model unambiguously. When an edge feed is used the number of degrees of
freedom to choose the feed location is reduced, but it is still possible to design an antenna
with practically any required complex input impedance needed.

First, an antenna was designed to operate at a frequency of 2.45 GHz with a feed impedance
of 50 Ω. This feed impedance was chosen, because in this way the antenna can easily be
measured with existing equipment and can be used in conventional systems. To make
the antenna broadband, a double resonance was introduced. Around 2.45 GHz resonance
occurs in both perpendicular directions of the patch. These resonances have a slight fre-
quency offset making the bandwidth of the antenna roughly twice as large compared to a
single-resonance microstrip patch antenna. A disadvantage of this approach is the fact that
it is difficult to obtain a 50 Ω input impedance at the edge, when a half wave fits in both
the horizontal and the vertical direction. Therefore, we have made not a half-wavelength
but a full wave fit in the a direction of the patch and now the 50 Ω input impedance is
found at the side of the patch.
The antenna is etched from 1.6 mm thick, grounded FR4 material. This material has a
loss tangent, tan (δ), of 0.016 and the relative permittivity is determined to be εr = 4.28,
by the method described in Chapter 2. The conductivity of the copper on the patch and
ground layer is σ = 5.8 × 107 Ω−1 m−1 .

The determination of the dimensions and feed location of the patch antenna followed the
following approach. The model presented in the preceding sections is used to design a
patch antenna that satisfies the performance demands.
Next, this antenna is analyzed with a full-wave simulator. In general the predicted behavior
by our model and the and the full-wave simulation differ. This difference is mainly caused
by the inaccuracy of the length extensions of the cavity due to the fringe fields. The lengths
extensions are now altered in such a way that the model and full-wave simulation agree.
Now our model can predict the behavior of the patch antenna much more accurate and an
improved design using the cavity model is made that satisfies our performance demands
once again. Again the behavior of the antenna is verified with a full-wave simulator. When
the differences in the predicted behaviors is still too large the process of adjusting the length
extensions can be repeated. However, in all our cases this was not necessary. In this way
a rapid and reliable design of a patch antenna can be reached, since only two full-wave
simulations are needed. The cavity model turns out to require at least a factor of ten times
less CPU time, compared to full-wave simulations.
At 2.45 GHz this design approach resulted in an antenna with dimensions a = 58.3 mm,
b = 29.2 mm, x0 = 20 mm and dx = 3.1 mm, as in Figure 3.2. This antenna was
manufactured, as shown in Figure 3.8, and measured.

The results of this antenna are shown in the Figures 3.9-3.11. No reliable measurements
concerning radiated fields have been performed, due to problems with the measurement

35
Figure 3.8: The 2.45 GHz broadband patch antenna

equipment, probably caused by radiating cables in the anechoic chamber. However, as


mentioned in Section 3.5, we can rely on full-wave simulations performed in Ansoft En-
semble. We can see in Figure 3.9, that the input impedance is predicted within a few
percent, well enough to use the proposed cavity model as a design tool for the antenna of
our rectenna system taking the margins given in Chapter 1 into account. The reflection
coefficient, assuming a 50 Ohm feed, in Figure 3.10(a) shows us that the model is capable
of determining this coefficient with enough accuracy for our purposes. The 10 dB band-
width of the reflection coefficient of this patch antenna is 5.9%, around 2.44 GHz. We can
see that the center frequency of resonance is very close to the 2.45 GHz, for which it was
designed. Compared to other results from the literature [18] and [19] this bandwidth is
large. Comparable bandwidths were only achieved by stacked patch antennas or changing
the shape of the patch. Stacked patch antennas result in an increase in size and costs.
When the patch shape is changed from the rectangular forms presented in this chapter we
can only rely on full-wave simulations because the different models presented in literature
can not cope with exotic shapes, or we have to invest in the development of new models,
which is not desirable.
It can be seen that the reflection coefficient has two minima around 2.45 GHz, these corre-
spond to two different resonating modes, the (1, 0) and the (0, 2) mode. These two modes
correspond to two different polarization states, resulting in elliptic polarization. The form
of the ellipse changes with frequency within the resonance band. The polarization changes
(approximately) from linear to elliptical to linear in the other direction. This makes the
system particularly suitable for frequency-redundant modulated systems.
We can see in Figure 3.10(b) that the radiation pattern determined by the model and the
full-wave simulator show a good resemblance for angles up to ±75 degrees. For angles
larger than this value we see a discrepancy because we assume the absence of dielectric
outside the cavity in our model while the simulator assumes an infinite dielectric layer, but
these angles are outside our main interest and so this discrepancy is not a major concern.

36
The gain patterns as shown in Figure 3.11 also show us that the model is capable of deter-
mining the magnitude of the radiated field within 1 or 2 dB. In the figures the gain in the
φ and θ direction is denoted by Gφ and Gθ . When considering Figure 3.11(d) it should be
noted that the scale shown in this figure is only very small, from -29 to -27 dB, and the
seemingly different behaviour of the model and the full wave simulation is actually of no
concern.

37
(a) Real part

(b) Imaginary part

Figure 3.9: Input impedance of the rectangular patch antenna

38
(a) Reflection coefficient

(b) Normalized radiation pattern

Figure 3.10: Characteristics of the rectangular patch antenna

39
(a) (b)

(c) (d)

Figure 3.11: Gain patterns for angle ϑ of a rectangular patch antenna


(a) Gϕ in the ϕ = 00 surface (b) Gϑ in the ϕ = 00 surface
(c) Gϕ in the ϕ = 900 surface (d) Gϑ in the ϕ = 900 surface

40
(a) Real part (b) Imaginary part

Figure 3.12: Input impedance of patch antenna matched to rectifier

Next, an antenna is designed to be used mainly for rectenna purposes. Therefore we aimed
at matching the input impedance of the antenna directly to the input impedance of a
voltage doubler. More details about this setup is given in Chapter 4, where it will be
shown that this complex input impedance is typically around 40 − j45 Ohm. To match
this impedance to the antenna, the input impedance of the antenna has to be the complex
conjugate of the diode’s output impedance, as described in [14]. Hence, to match this
combination of diode and antenna, the antenna should have an input impedance of 40 +
j45 Ohm. Again, a microstrip edge-fed antenna design is chosen, but now a single-mode
resonance is chosen, resulting in a nearly square patch. This resulted in a patch antenna
with dimensions a = 27.7mm, b = 30.8mm, x0 = 0.4mm and dx = 0.45mm, using the same
substrate as used for the previous antenna. The microstrip feed line width of 0.45mm was
chosen because this is the width of the pins of the intended diode.
In Figure 3.12 it can be seen that, using the presented methods, an antenna with the
required complex input impedance can be obtained. More generally, an antenna with any
desired input impedance (within practical limits) can be quickly and accurately designed.

3.7 Conclusion

In this chapter a model has been presented to analyze a microstrip patch antenna. This
model is based on a radiating cavity. First, the electric field inside this cavity was deter-
mined. From the electric field at the edges of the cavity we have determined the radiated
field. From the electric field inside the cavity together with different losses (radiation,

41
substrate and copper losses) the input impedance is determined.
The results for two different patch antennas have been presented. The input impedance
and the reflection coefficient have been measured and the model is able to determine these
properties with a high degree of accuracy. The input impedance was predicted within a
few percent. Further, the full-wave simulations of the radiated field and the gain pattern
of these antennas were compared to the model. Here, the model was also able to calculate
these patterns accurately. The radiation pattern was predicted within 1 or 2 dB.
The first antenna that was presented has two radiating modes around 2.45 GHz making
it a broadband patch antenna. With the second antenna that was presented it is shown
that it is possible to design a microstrip patch antenna with any desirable complex input
impedance, within practical limits.
The presented model is able to determine the antenna properties with a degree of accuracy
that enables us to use it to design a microstrip patch antenna for our needs.

42
Chapter 4

Rectifier

In the previous chapter it was described how to receive RF signals. In this chapter it
is described how the RF-power in these signals can be converted into usable direct current
(DC)-power, using a so-called rectifier. This conversion is necessary because an application
can normally not be powered with an alternating current (AC) source, especially not at the
RF frequencies that we use.

4.1 Introduction

Rectifier circuits are well known in electronics. In these circuits a diode is used to rectify a
varying voltage waveform. A diode only conducts in one polarity and will therefore create
a DC voltage. When we suppress the components, at frequencies other than the desired
one, using for example a capacitor, we are left with a pure DC voltage. After this, the DC
power is normally dissipated in a load, represented by a load resistance. This circuit, in
its basic form, looks like the one shown in Figure 4.1.

Figure 4.1: Diode rectifier

43
Figure 4.2: Diode V-I relation

In Figure 4.2 the practical and ideal V-I characteristic of a diode is shown. Here, it can be
seen that a diode is a nonlinear device. This makes the complex calculus for linear passive
electronic circuits unsuitable and a non-linear analysis has to be employed.

In this chapter a method will be presented that can cope with the non-linearity of the
diode in the rectifier circuit. The method will be used to determine the input impedance
of the rectifier circuit and to determine the output voltage for a given input power and load
impedance. The input impedance of the rectifier circuit is important because a maximum
power transfer through the rectifier requires an impedance that is matched to the rest of
the circuit.

The method to analyze the nonlinear behaviour of the diode that will be presented in this
chapter will be used to design a so-called voltage doubler rectifier. This voltage doubler is
shown in Figure 4.3, compared to Figure 4.1 a second diode is added. A voltage doubler
uses this second diode to (approximately) double the output voltage of a rectifier, assuming
a large load impedance. The voltage doubler puts the diodes in parallel with respect to the
input RF signal, which lowers the input impedance and this generally reduces the difficulty
of matching the rectifier to the other circuitry. However, the diodes appear in series with
respect to the output load, which (approximately) doubles the output voltage. If the two
diodes are contained in a single package, the cost associated with the second diode is very
small, making the doubler an interesting circuit for wireless powering. More information
concerning voltage doublers is given in [20].
Some rectifiers use a bias current to use a certain working point for the rectifier. However,
because we assume no other power source is present other than the RF input power, we
cannot bias the rectifier circuit. The diode that will be used is a Schottky diode. This
diode has been chosen because it has a low forward bias voltage, meaning that it starts to
conduct at a low voltage, compared to other types of diodes. This makes it more suitable
for low-power rectifier circuits.

44
4.2 Schottky diode model

As mentioned before, a (Schottky) diode is a nonlinear device. This nonlinearity arises


from the composition of the materials in a diode. A Schottky diode is a semiconductor de-
vice, where a piece of metal is placed on top of a n-type or p-type semiconductor substrate.
Here, we use an n-type semiconductor. One of the contacts of the diode is connected to
the semiconductor, the other to the metal. When a semiconductor and a conductor are
brought into contact a so-called contact potential is formed. For a Schottky diode, this
contact potential is called the barrier potential.
Let us assume that the semiconductor contact is connected to ground. When a negative
voltage is applied to the metal contact the free charge carriers in the semiconductor are
expelled from the metal-semiconductor transition and no current can flow from the semi-
conductor material to the metal. This situation is called reversed bias. When a positive
voltage is applied to the metal contact free charge carriers are attracted from the semicon-
ductor to the metal surface and now a current can flow. This situation is called forward
bias.
The area where no free charge carriers are present is called the depletion layer. A typical
property of a Schottky diode is that it has a thin depletion layer giving it a very short
switching time between reverse and forward bias. This property makes this diode even
more suitable for our application.

By using this structure of a Schottky diode a circuit model of the diode can be formed.
The substrate is not a very good conductor (compared to metal) and therefore a substrate
resistance, Rs , is introduced to account for the conduction losses in the substrate. The
depletion layer forms an isolation between two conducting plates and can be modeled as
a so-called junction capacitance, Cj . As the hight of the depletion layer changes with the
barrier voltage, the value of Cj also changes with this voltage. In [21] the expression for
this junction capacitance can be found as
Cj0
Cj (vd ) = . (4.1)
[1 − vd /φ]1/2

Figure 4.3: Voltage doubler diode rectifier

45
In this relation vd represents the voltage over the nonlinear part of the diode as shown in
Figure 4.4. The zero-bias differential barrier capacitance is given by Cj0 and the barrier
potential of the diode is denoted by φ. Both these values can usually be found in the
datasheet of the corresponding diode. The nonlinear conductance of the diode is modeled
as a nonlinear resistance with V-I relation

id = Is (eαvd − 1). (4.2)

In this relation id is the current through the nonlinear impedance d, as in Figure 4.4.
Further, Is represents the saturation current and can be found in the datasheet of the
diode. This saturation current represents the current that can still flow through the diode
when it is reverse biased. This current is caused by thermal emission in the depletion layer.
The factor α is given by
q
α= , (4.3)
nkT
where q is the electron charge, k Boltzmann’s constant and T the absolute temperature in
Kelvin. The ideality factor of a diode is denoted by n and can be found in the datasheet.
For most diodes the value of n is close to one. Now we can use a circuit model for the
diode as shown in Figure 4.4. This model is widely used in literature, as can be seen for
example in [22, 23].

The circuit model that has been described so far represents a single Schottky diode. A
normal ‘off the shelf’ diode is put in a package, so it can be used in an external circuit.
The effects of this packaging can also be included in the model turning the circuit model
of the diode into a package model. This packaging results in parasitic capacitance, Cp , and
inductance, Lp , which are also given in the datasheet of the corresponding diode. This
diode package model will be used in the analysis of the rectifier and is shown in Figure 4.5.

Figure 4.4: Diode circuit model

46
Figure 4.5: Model of a diode in a package

4.3 Nonlinear analysis

4.3.1 Methods

There are many methods to deal with the nonlinear behaviour of the Schottky diode. A
common method is to linearize the system around a working point. However, in our setup
the diode is used in a so-called large-signal region. This means that the system can not be
linearized unambiguously, because the system shows the nonlinear behaviour within the
working region.

A truly nonlinear method is the so-called harmonic balance method. In this method a signal
with a certain discrete frequency is used as input and it is assumed that the nonlinearity
in the system causes the output signal to be composed of different frequency components.
These frequency components are all harmonics of the input frequency (including a DC-
term and the fundamental frequency). To solve the system it is attempted to match
all the harmonics in the system in such a way that the nonlinear constraints, as given
in Equations (4.1) and (4.2), are satisfied. This method tries to find a balance for all
harmonics in the system, so it uses a frequency-domain approach. This method is described
in [2].

There is another truly nonlinear method that employs the time-domain formulation. In this
method, first the (nonlinear) differential equation governing the system is determined. Next
a starting condition is chosen. This starting condition can be chosen relatively arbitrarily,
as long as the system will evolve to a steady state. Now a time-domain input signal is
impressed on the system. An integral, using a small time step over the differential equation
is then used to determine the system state at the next time instance, i.e.
yn+1 = yn + ∆t · f (tn , yn ), (4.4)
where yn is the system state at t = tn and the function f (tn , yn ) represents the derivative
of y at t = tn . The timestep is called ∆t. When this time interval is chosen small enough

47
(a) Circuit (b) Model

Figure 4.6: Analysis setup

this is an accurate way of determining the behaviour of the system. Often, this method
uses a Runge-Kutta integration method and this is described more thoroughly in [24].

We have chosen to use the latter method to analyze the rectifier circuit, because it is
more flexible and it gives the impression to be computationally more efficient. It is more
difficult to include the nonlinearity of the junction capacitance in the harmonic balance
method. Furthermore harmonic balance has already been used in a similar study [2], so a
comparison can be made concerning accuracy and computational costs. This comparison
was requested from within the project for choosing future analysis methods. In Section 4.5
the two different methods are compared.

4.3.2 Runge-Kutta method

To determine the parameters of the diode, the setup shown in Figure 4.6(a) was used. We
assume a voltage source with a single frequency ω as input, i.e.,

Vg = |Vg | cos(ωt). (4.5)

The generator impedance is denoted by Rg . The package model of the nonlinear diode, as
described in the preceding section, is used to analyze this circuit as shown in Figure 4.6(b).

48
The electrical behaviour of this circuit can be described with the following expressions.

Vg = Ig Rg + Lp dIdtg + VCp
VC p = Vd + VRs
VRs = Rs IRs = Rs (ICj + Id ) (4.6)
IC j = Cj dV
dt
d

Id = Is (eαVd − 1)

Ig = Id + ICj + ICp
Id = Is (eαVd − 1)
(4.7)
IC j = Cj dV
dt
d

dVCp
IC p = Cp dt
.
These expressions can be written as two coupled differential equations where the diode
voltage, Vd , and the generator current, Ig , are the unknowns. These differential equations
include a second-order derivative of the generator current. A third differential equation
can be added to form a coupled set of three first-order differential equations. The set of
coupled first-order differential equations that will be used is

Vg = Lp I˙g + Vd + Rg Ig + Rs Cj V̇d + Rs Is (eαVd − 1)






Ig = Is (eαVd − 1) + Cj V˙d + Cp V˙g − Cp Rg I˙g − Cp Lp Ẋ (4.8)


X = I˙g .

The last differential equation is added (resulting in the elimination of second order deriva-
tives) because most techniques to numerically integrate differential equations can only
handle first-order differential equations. In these equations the time derivative of a vari-
able is indicated by a dot above the variable, to make the equations more readable.

4.3.3 Implementation

The equations that have been derived in the preceding section, to analyze the diode have
been implemented in a numerical routine. As mentioned before a method is used where
the next state of the system is determined from the present (known) state of the system
and the differential equation of the system. At time instance t = t0 the system parameters
Vd , Ig and X are known. The system state at t = t0 + ∆t is determined by a Runge-Kutta
method.

49
To do so the expressions in Equation (4.8) are rewritten in such a form that each equation
expresses an unknown derivative in terms of the known parameters. The derivative of the
impressed generator voltage, V̇g , in Equation (4.8) is known. It can simply be determined
from Equation (4.5). Now we have expressions for V˙d (t, Vd , Ig , X), I˙g (X) and Ẋ(t, Vd , Ig , X, V̇d ),
i.e,
1 
Vg − Rg Ig − Vd − XLp − Rs Is (eαVd − 1)

V̇d = (4.9)
Rs Cj
I˙g = X (4.10)
1 h i
Ẋ = Is (eαVd − 1) + Cj V̇d + Cp V̇g − Cp Rg X − Ig (4.11)
Cp Lp

Vg and V̇g are time dependent, as can be seen in Equation (4.5), and therefore the total
system is time dependent.

The Runge-Kutta method multiplies each expression with a time period ∆t and uses these
values to determine the state variables at time instance t + ∆t, as depicted in Equation
(4.4).
Our routine uses a so-called Runge-Kutta4 routine, see [24]. This means that to determine
the state variables at the next time instance, derivative information at the current time
instance, at the end of the interval and twice at the midpoint is used. The midpoint is
evaluated twice, assuming two different system states. Next, these four derivatives are
combined in a predefined way to minimize the error term at the end of the interval. This
approach cancels the first four orders of error terms, making this method a fourth-order
method. The step-size ∆t is adjusted with an adaptive step-size algorithm as described in
[24]. This step-size algorithm compares the result using a certain step size with the result
obtained by using half the step size twice. If the difference between these two results is
below a certain minimum accuracy limit the step size is accepted.

Suppose we know the system state at time t = tn . We call the system variables at this
time point Vdn , Ign and Xn . We would like to determine the system variables at moment
t = tn+1 = tn + ∆t. The derivatives that are used to find these system variables are
determined as follows. First the derivatives are determined at the current point in time.

k1 = V̇d (tn , Vdn , Ign , Xn )


l1 = I˙g (Xn ) (4.12)
m1 = Ẋ(tn , Vdn , Ign , Xn , V̇d (tn , Vdn , Ign , Xn )).

The new variables k, l, m are introduced as temporary variables to store the different deriva-
tives. Next, these factors are used to determine the derivatives at the midpoint of the

50
interval.

k2 = V̇d (tn + ∆t/2, Vdn + k1 /2, Ign + l1 /2, Xn + m1 /2)


l2 = I˙g (Xn + m1 /2) (4.13)
m2 = Ẋ(tn + ∆t/2, Vdn + k1 /2, Ign + l1 /2, Xn + m1 /2, k2 ).

These intermediate derivatives are used to determine a second intermediate set of deriva-
tives.

k3 = V̇d (tn + ∆t/2, Vdn + k2 /2, Ign + l2 /2, Xn + m2 /2)


l3 = I˙g (Xn + m2 /2) (4.14)
m3 = Ẋ(tn + ∆t/2, Vdn + k2 /2, Ign + l2 /2, Xn + m2 /2, k3 ).

The last derivatives that are determined are the ones at the end of the time interval
(t = tn + ∆t). The second intermediate derivatives (k3 , l3 , m3 ) are used for this.

k4 = V̇d (tn + ∆t, Vdn + k3 , Ign + l3 , Xn + m3 )


l4 = I˙g (Xn + m3 ) (4.15)
m4 = Ẋ(tn + ∆t, Vdn + k3 , Ign + l3 , Xn + m3 , k4 ).

All these derivatives are used to determine the system variables at the next time instance
as follows
∆t
Vdn+1 = Vdn + [k1 + k4 + 2(k2 + k3 )]
6
∆t
Ign+1 = Ign + [l1 + l4 + 2(l2 + l3 )] (4.16)
6
∆t
Xn+1 = Xn + [m1 + m4 + 2(m2 + m3 )] .
6
Combining the different derivatives in this way results in the desired cancelation of the
first four orders of error terms and only 5th and higher order error terms will exist.

Suppose we use a generator with a certain voltage amplitude, |Vg |, and single frequency,
f0 . Under these circumstances we can define a diode impedance for all frequencies. The
nonlinearity of the diode creates spectral components at harmonics of f0 (including DC).
Other frequencies will not be excited and will have a zero diode voltage and current, so
no meaningful impedance can be formulated at frequencies other than the harmonics of
f0 . We can envision the diode as a load Zdf0 for frequency f0 and as a voltage source with
series impedance Zdfn for frequency fn other than f0 .
A typical result of the presented model is shown in Figure 4.7, where an Agilent HSMS-
2852 [25] Schottky diode has been used. In this figure the wave shape of the diode junction
voltage Vd is shown for a 2.45 GHz input signal with 0dBm power. The properties of this
diode are described in the next section.

51
Figure 4.7: Time-domain diode junction voltage

Using the presented Runge-Kutta4 routine the diode junction voltage and the generator
current have been determined. By using the impressed generator voltage, the generator
current and the generator impedance the voltage over the diode, D, as in Figure 4.6(a),
can be easily determined. Using a Fourier transform (e.g. FFT) of the steady state of the
system we can transform the time-domain diode current and voltage to a frequency domain
spectrum, giving us phase and amplitude information for each frequency of interest.
For each harmonic of f0 a diode impedance can now be determined as [20]

Vdfn
Zdfn = . (4.17)
Idfn

The incident power on the diode can be determined from the generator voltage and the
generator impedance as

|Vg |2
Pinc = , (4.18)
8Rg

as described in [12].
Now, we have presented a method that can determine the input impedance of a diode
for all the harmonics of the input frequency for a certain incident power, Pinc . A similar
method as presented in [2] has been employed to determine the DC voltage when this diode
is used in an extensive circuit. It uses the Thévenin theorem to subtract the nonlinear part
of the circuit.

52
Figure 4.8: Diode measurement circuit

4.4 Measurements

To validate the method just presented, a diode was measured. As Schottky diode we have
chosen to use the HSMS-2852 from Agilent Technologies. This is a typical RF Schottky
diode for low-power applications, which is widely available. Further, it does not need a
bias current to operate, which makes it suitable for our application.
The properties of this diode can be found in [25]. The properties that are needed to
determine the input impedance with the method presented before are as follows. The
substrate resistance, Rs , is 25 Ohm. The saturation current is 3 µA, Cj0 is 0.18 pF and the
ideality factor is 1.06. The parasitic package capacitance is 8 · 10−14 F and the parasitic
inductance is 2 nH, the barrier potential, φ, is 0.35 V.

To measure the input impedance of this diode, it is mounted directly on a SMA-connector


(subminiature version A). This SMA-connector is attached to a network analyzer by a
phase-constant coaxial cable. At the network analyzer we measure the complex reflected
power. To compensate for the coaxial cable the network analyzer was calibrated including
this cable. So the measurement data that the network analyzer produces is the reflected
power directly at the diode pins, because the calibration set uses the same SMA-connectors.
The generator impedance as well as the characteristic impedance of the coaxial cable is
50 Ohm.

The network analyzer that was used to measure the input impedance of the diode is a HP
8753D. This network analyzer (and practically any other network analyzer) can not handle
DC voltages at its ports. To prevent this from happening a DC-blocker is placed inside
the network analyzer. The DC signal itself is led to a DC port of the network analyzer.
An RF choke is used to make sure only the DC-component is led to the DC-port of the
network analyzer. When a 50 Ohm load is placed at this port the model setup of Figure
4.6(a) is preserved. The measurement setup resembles the one shown in Figure 4.8.

53
The network analyzer is used to measure the reflection coefficient S11 at the diode. Imagine
that a transmission line with characteristic impedance Z0 is terminated with a (complex)
load ZL . From [5] we can derive that the reflection coefficient at the load is
ZL − Z0
S11 =
Z L + Z0
and (4.19)
1 + S11
Z L = Z0 .
1 − S11
In this way, we can determine the diode input impedance using Zd = ZL . The network
analyzer can only measure the reflection at the fundamental frequency, i.e. at the frequency
that is generated by the network analyzer itself as input excitation. In this way we are able
to determine the fundamental impedance of the diode over a frequency range. However,
it is not possible to determine the impedance of the higher harmonics this way, because
the network analyzer ignores signals at frequencies other than its own generator frequency.
A spectrum analyzer has been used for this goal, but the spectrum analyzer measures
only amplitudes, resulting in missing phase information. A time domain analysis would
deliver the phase information, however an oscilloscope for the required frequencies was not
available at that moment.
To be able to analyze the harmonic behaviour of the diode we deployed a simulation
program. We have used Serenade, a RF network simulator, which is well-known for its
reliable simulation results. The results of Serenade can be treated as true values. Serenade
uses a harmonic balance approach to determine the behaviour of nonlinear components.

4.5 Results

The results of the measurement, as described in the preceding section, will now be compared
to the results of the presented model. The fundamental input impedance of the diode has
been measured over a frequency range of 100 MHz to 4.1 GHz. In Figure 4.9 the real and
imaginary part of the input impedance of the diode are shown for 0 dBm (which equals
1 mW) input power. The noise that appears on the measured impedances is caused by
the fact that the network analyzer is not able to fully suppress the higher harmonics on
its input. When a single diode in a HSMS-2852 chip (containing two diodes) is measured,
as done here, the second diode ‘floats’. This diode is connected to the shared pin of the
chip at one side and is not connected at the other side. However, this floating pin of
the second diode results in a capacitive coupling. The measurement data shown here are
compensated with the elimination of a capacitor of 0.3 pF, which roughly corresponds with
the capacitive coupling of the floating diode. We see that the model is able to predict the
input impedance within a few percent at the frequencies we are interested in. The model
has been used to determine the input impedances at other power levels as well, ranging

54
from -10 dBm to +10 dBm. These input impedances exhibit a similar behaviour as shown
in Figure 4.9. The model still reliably determines the input impedance for the other input
power levels. The only result shown here is for 0 dBm input power, because we have
assumed that this is the received power level, as mentioned in Chapter 1. Using these

(a) Real part

(b) Imaginary part

Figure 4.9: Diode input impedance according to model and measurements

55
Figure 4.10: Diode voltage spectrum

input impedances of a single diode the input impedance of a voltage doubler is determined
to be roughly 40-j45 Ω at 2.45 GHz. This impedance is obtained by halving the input
impedance of a single diode because two diodes are placed in parallel in a voltage doubler
resulting in half the impedance.
In Figure 4.10 the harmonic components of the diode voltage VD are shown. In this figure
the values based on the results of the Serenade simulations are compared with the values
obtained with our model. In this figure a 0 dBm input signal at 2.45 GHz is assumed. We
see good agreement between the model and the simulation.
As mentioned in Section 4.3 the Runge-Kutta method presented in the previous section
is compared to the harmonic balance method here. The harmonic balance method is
somewhat more complex, where the Runge-Kutta method is more intuitive. The Runge-
Kutta method turns out to be slightly computationally more efficient resulting in a faster
prediction of the diode behavior. However, the improvement on the calculation times is
not impressively. Further, the Runge-Kutta method can include the nonlinearity of the
junction capacitance more easily. The accuracy of the prediction of the input impedance
for each of the harmonics is comparable. Further, the Runge-Kutta method can be used
more easily to determine the input impedances for the higher orders, which are necessary

56
when higher power levels are considered.

4.6 Conclusion

In this chapter a method is presented to analyze the nonlinear rectifying part of a rectenna.
The nonlinearity in the rectifier is caused by the diode, which transfers the RF-signals to
DC (and to other frequencies). First a nonlinear diode model has been presented. Next
a time-domain Runge-Kutta method has been presented to determine the electrical be-
haviour of the diode in a circuit with an RF-source. This method is used to determine the
input impedance of the diode for all harmonics at a certain input power and frequency.
Measurements have been performed to analyze the reliability of the method. The model
turns out to be able to predict the fundamental input impedance within a few percent,
which is accurate enough to further use this model.
This method enables us to design a RF-rectifier quickly, flexibly and reliably.

57
58
Chapter 5

Total System

In the preceding chapters the individual parts of the system, which is capable of wireless
power and data transmission have been described. In this chapter the results are shown
when these individual parts are combined.

5.1 Introduction

In this chapter three different systems are described that are composed of the earlier
described components and subsystems, which will both be called subsystems. First of all a
single rectifier was designed by combining a patch antenna and a voltage doubler rectifier.
In this design a stub has been used that will be described in the next section.
A set of eight of these rectennas was combined to increase the received power. Finally a
complete design, including the Wilkinson Splitter is presented.

5.2 Single Rectenna

As a first step in combining the different subsystems a single rectenna is designed. This
comes down to combining the patch antenna and rectifier. A typical rectenna setup is
shown in Figure 5.1. The received energy is collected by the antenna. This received
energy is then rectified and dissipated in a load. A rectenna usually includes an impedance
matching circuit. This matching circuit insures a maximum power transfer by matching
the input impedance of the antenna to that of the rectifier, which differ in general. In
our design we made the matching circuit unnecessary by creating an antenna and rectifier

59
Figure 5.1: Rectenna setup

circuit which match directly.

The absence of the impedance-matching circuit reduces the physical dimension of the
system. Further, the (minor) losses, mainly reflection losses, in the matching network are
now eliminated. We do not have control over the input impedance of the voltage doubler
rectifier. Hence, to match the antenna and rectifier, the impedance of the antenna has to
be adjusted to the impedance of the rectifier. In Section 3.6 it is shown how this input
impedance is achieved. Since the antenna will not be used in a stand alone way there is
no need to have a 50 Ω impedance level at the input port.

The rectifier is a voltage doubler as shown in Figure 4.3. The diodes that have been used
are Agilent HSMS-2852 Schottky diodes. In [2] an inductor is suggested to act as a short
circuit for DC at the rectenna. This inductance can be left out in our design because the
second diode in the voltage doubler already supplies a short circuit for DC.

In Figure 5.2 two realizations of the entire system without the load are shown. A load
(representing an application) can be connected directly at the connector to absorb power.
In the left design a stub is included. This stub is part of the rectifier and is included to
suppress the presence of the fundamental frequency, and to a lesser extent higher harmonics,
in the rectifier’s output signal caused by its input.

Measurements have shown us that this stub can be omitted without significant decrease
of system performance. This is allowed owing to the fact that the capacitor at the output
of the voltage doubler suppresses the RF components at the output totally for the small
power levels we are taking into account. This further reduces the size of the rectenna.
The size of the rectenna is now hardly larger than the dimensions of the patch antenna, as
shown in Figure 5.2(b).

The diode model as presented in Chapter 4 was used to design the voltage doubler. To
determine the accuracy of this model measurements concerning the rectifier have been
performed. The voltage doubler is connected directly at the patch resulting in an input
impedance different from 50 Ohm. Normally to measure a rectifier it is connected directly

60
to a 50 Ω source. Our rectifier requires a source with a different impedance, which could not
be created with the available measurement equipment. Our patch antenna was especially
designed to have this input impedance. Hence, to be able to measure the rectifier reliably, it
is measured including the patch antenna. For different levels of input power the unloaded
DC output voltage is measured. The input power of the rectifier in the model can be
easily controlled using |Vg | in Equation (4.5). On the other hand, the power incident
on the rectifier in the measurement is more difficult to determine, since we do not know
exactly how much power is accepted by the antenna. To determine the power accepted
by the antenna for different transmitted power levels we have used a scaling technique.
If we know the antenna’s output power for a given transmitted power generated by a
transmitter antenna at a fixed distance from our (receiving) rectenna, we can calculate the
antenna’s output power for other transmitted power levels, since this is a linear process.
For example, if the antenna’s output power is 0dBm for a transmitted power of 20dBm,
the antenna output power will be -10dBm for 10dBm transmitted power. For one power
level (-3.75 dBm antenna power output in the model) the model and the measurements
have been matched. The other rectifier input power levels have been determined by this
scaling technique. Now, we are performing a kind of relative measurement.
In Figure 5.3 the results of these measurements at 2.45 GHz are shown. We can see a
good agreement between the model and the measurements. The predicted and measured
output voltages only differ by few percent. This gives us confidence in using the presented
rectifier model as design tool.

This measurement is performed without a load impedance attached to the rectenna. In a


real-life situation a load will always be connected to the output, representing the dissipation
of power. However, typical applications for this system will only require power for a small

(a) With stub (b) Without stub

Figure 5.2: A single rectenna

61
Figure 5.3: Voltage doublers DC output voltage for different input power levels

amount of time and then ‘go back to sleep’. To power such applications a capacitor can be
used, which is constantly loaded by the rectenna and once in a while the power is extracted
from this capacitor into the application. In this way the rectenna is still loaded with a high
impedance resulting in a high output voltage, comparable to the results in Figure 5.3. So,
using this technique, the rectenna can deliver a reasonable amount of power at a relatively
high voltage with a small duty cycle. When 0dBm of power is incident on the patch
antenna, roughly 520 µ W can be dissipated at a voltage of 1.2 V. How these values were
determined will be explained in the next section.

5.2.1 Efficiency

The efficiency of this rectenna has been measured as well. A linearly polarized horn antenna
at 2.42 GHz was used as transmit antenna. This frequency was chosen because the highest
output voltage was observed for this frequency.
First, the gain of the transmit horn antenna was characterized by measuring the received
power by a standard gain horn in an anechoic chamber. From this measurement the gain
of the horn antenna was determined to be 5.95 dB at the specified frequency.
Next, the optimal load impedance was determined. We have used a purely resistive load.
We have used 100 mW of transmit power. The distance from the transmit antenna to the
rectenna was chosen such that roughly 1 mW of power was incident on the rectenna. This
measurement was not performed in an anechoic chamber, because the load impedance had
to be changed several times, which is a tedious job to do in an anechoic chamber. The

62
Figure 5.4: Dissipated power for different load resistances

minor changes in power level hardly change the diode behaviour. In this way the received
power can not be determined exactly, due to multipath and scattering effects. The distance
between transmitter and receiver was chosen such that the received power without these
effects would have been 1 mW. The dissipated DC-power in the load was measured for
different load impedances. The results of this measurement are shown in Figure 5.4. In
this figure we can see that the optimal load impedance is around 900 Ω.
Next, the horn antenna was placed, in an anechoic chamber, at such a distance from the
rectenna that the power incident on the rectenna was 1 mW. In this case, the measurement
was performed in an anechoic chamber, so the received power could be set much more
precise, by setting the distance between transmitter and receiver for a given power level.
This distance was determined from Equation (5.1), i.e.,
 2
λ0
Prec = Ptr Gtr Grec , (5.1)
4πR

which is the Friis transmission equation, where Prec and Grec are the received power and
the receiver’s gain, Ptr and Gtr are the transmitter’s gain and power, R is the distance
between the transmitter and receiver and λ0 is the free-space wavelength. The gain of the
receive antenna is determined in Chapter 3, and the gain of the transmitter has just been
determined.
The dissipated power with an optimum load resistance (900 Ω) was measured (in an ane-
choic chamber) by measuring the DC-current and voltage using a voltage and a current

63
Figure 5.5: Schematic layout of the rectennas in series

meter. The dissipated power was 516µW . The efficiency, η, can be determined as
PDC
η= ∗ 100%, (5.2)
Prec
where Prec is the incident power on the rectenna and PDC is the dissipated DC-power. This
results in an efficiency of 52 %. This efficiency is an improvement compared to the results
obtained in [2], where an efficiency of 40 % was achieved. In [26] an efficiency of about
80% was achieved for a rectenna operating at 2.45 GHz, but this result was obtained with
an input power of 100 mW resulting in a much more efficient rectifier.
When we compare our efficiency to the one achieved in [2] the improvement is mainly due
to the elimination of the impedance matching circuit, which causes additional reflection
losses. Furthermore, our efficiency is more realistic than the high input power efficiencies
reported in [26], since regulations wil not allow these high input powers in most practical
applications.

5.3 A rectenna-powered wall clock

The rectenna presented in the previous section showed promising results and has been
used in a more extensive system. To further increase the output power eight rectennas
have been connected in series, as shown in Figure 5.5. Since the ground planes of the indi-
vidual antennas act as ground for the rectifiers as well, the ground planes of the individual
patch antennas are not connected. This leave space for DC-circuitry.
In this way the output voltage and power are increased by a factor of eight. Further, the
rectenna system can deliver the same amount of power at a larger distance assuming the
same source and transmitted power.
As mentioned before this system is particularly suitable for applications with a low duty
cycle. A standard electronic wall clock has been used as a demonstrator. Normally a clock
is powered by a battery, requiring power only during an instance each second. When we
are able to collect the required amount of power at the right voltage (usually 1.5 Volt) we
can supply the power for the clock by the rectenna system.

64
(a) Front (b) Back

Figure 5.6: Wall clock powered by eight rectennas

In Figure 5.6 this clock is shown. On the front side we can see the eight patches and
diodes. At the back we can see the individual ground planes, where it can be seen that the
ground planes are not connected. Further, we can see a small electronics board hanging
above the back of the clock. This is included to make sure that the output voltage of
the rectifier never exceeds 2.0 Volts. When the input voltage of the clock exceeds this
value, the clock might get damaged. During experiments a clock ‘deceased’ because these
security electronics were not included at that time and the rectenna system received too
much power resulting in an output level of roughly 15 V.
A typical waveform of the supply voltage of the clock is shown in Figure 5.7. Here we can
see the slow charging of the capacitor and the instantaneous power absorption by the clock.
This waveform is observed when enough power is collected by the rectennas to power the
clock. When more power would be received the output voltage would approach the asymp-
totic level of 2 Volt earlier and would stay almost constant until the clock ticks. When less
power would have been received, the charging would have developed more slowly. Then,
the output voltage will not come near 2 V, when the power is absorbed each second. When
the output voltage at the moment of power absorption is below 1.2 V the clock will not
operate at all.
This system is not suitable to be measured quantitatively since it will let the clock tick
or not. The performance of the individual rectennas were already measured in the prece-
ding section. What can be measured is the distance for which it works assuming certain
circumstances. We used a horn antenna at 2.45 GHz, which was powered with a 20dBm
(100mW) constant wave. The horn antenna was placed in the focal point of a parabolic

65
Figure 5.7: Typical supply voltage of the wall clock

(dish) antenna, in order to create a small radiating beam. The dish has a diameter of
51 cm, resulting a a transmitter gain of roughly 22 dB. Under these circumstances it was
possible to power the clock over a range of 6 meters.
Since WLAN uses the same frequency band it was attempted to let the clock operate,
powered by the radiated fields of a WLAN antenna. It turned out that WLAN delivered
enough power to let the clock tick over a distance of roughly 10 cm from the WLAN an-
tenna. To do so, it was necessary to create a constant data stream between two WLAN
nodes. The throughput was not harmed by the presence of the clock, so the radiated fields
were not altered too severely.

5.4 Prototype design of the total system

So far, the rectenna capabilities of the individual subsystems have been analyzed. In this
way only power is transferred. The goal of this project was to be able to transfer power
and data simultaneously. To do so the Wilkinson splitter was included.
To send the data a modulated carrier was used. The two most obvious modulation options
were frequency modulation (FM) and amplitude modulation (AM). Frequency modulation
is preferable because it is less sensitive to noise. However, it is difficult in practice to
design a FM demodulator which does not require more power than our system can supply,
especially not within the framework of this project.
On the other hand a relatively simple AM demodulator can be used which does not require
any additional power. So, the data has been sent through the system using AM. More
information about the modulation and demodulation will be given in the next subsections.

66
Figure 5.8: Building blocks of the total system

The building blocks of the prototype are shown in Figure 5.8. The individual blocks will
now be discussed.

RF-source, modulator and data

As RF-source, a network analyzer was used to create a carrier signal with a single fre-
quency. The system will be analyzed for carriers varying between 2.25 and 2.65 GHz. The
network analyzer was used to perform a frequency sweep, required for this measurement.
A network analyzer is rather big and for demonstration purposes a more elegant source
can be used. There are constant wave sources (without sweeping capabilities) that can
produce a RF carrier between 2 and 3 GHz which are not bigger than 20x10x4 cm.
An amplitude modulator using an MSWT-4-20 switch has been made, since no ‘off-the-
shelf’ modulator was available to be used at 2.45 GHz. This modulator is shown in Fig-
ure 5.9(a). The modulator attenuates or passes an (RF) input carrier signal, depending
on the input voltages of the four transistors. The resistors attenuate the carrier and the
combination of the three resistors ensures that the impedance level is not changed. By
making the attenuation small (e.g. 1 dB) it is ensured that power is always delivered to
the system. One could say that the data is superimposed on the carrier. The power in the
transmit carrier mostly determines the transmit power. When FM would have been used
the instantaneous transmit power would be (almost) independent on the data.
To create (dummy) data a function generator is used to generate a block wave. The func-
tion generator was used to create a 1 kHz block wave, with an amplitude of 1.25 V and an
offset of -1.25 V. The amplitude and offset have been set in this way to realize the 1 dB
attenuation of the amplitude modulator.

67
(a) Modulator (b) Demodulator

Figure 5.9: AM system

Amplifier and transmit antenna

In the prototype design the transmit antenna (Tx Antenna) is a horn antenna, designed
for the 2.45 GHz band, and creates a linearly polarized field. The amplifier is setup in
such a way that the signal fed to the transmit antenna contains 20dBm of power. This
equals 100 mW and is the maximum output power allowed in this band. A KU 233 BBA
amplifier was used, which has a gain of 30 dB over a frequency range from 500 MHz to
2500 MHz. After the signal has been sent by the transmit antenna it is received by the
receive antenna (Rx antenna). The main focus of this project has been on the parts after
the transmit antenna

Receive antenna

As receive antennas (Rx antennas) two patch antennas have been used. This type of
antenna has been analyzed in Chapter 3. The antenna that was used in the prototype
is the broadband patch antenna described in Section 3.6. The behaviour of this antenna
is shown in Figures 3.9-3.11. The important characteristics of this antenna are its dual
polarization and the fact that it is relatively broadband. This results in a larger frequency
range in which the system can be used and horizontal as well as vertical polarization can be

68
used. It should be noted that the polarization shifts over the frequency band from vertical
to horizontal.

Wilkinson splitter

The signals that have been received by the two patch antennas are led to a Wilkinson
splitter. This splitter is described in Chapter 2. A splitter with K 2 = 1 was used, because
comparable signal strengths are expected from the two antennas. This splitter requires a
100 Ohm resistive element to prevent reflections at the different ports. The rectifier will
be used as resistive element.

Rectifier

A rectifier is used to convert the received RF-power to DC-power. A voltage doubler, as


described in Section 4.1, is used. The implementation of this voltage doubler is the same
as described in Section 5.2, so the HSMS-2852 is used here as well. In Chapter 4 the input
impedance of an diode has been determined. By using the fundamental input impedance
of a single diode, the input impedance at the fundamental frequency of the voltage doubler
can be determined. This impedance is around 40 − j45Ω and differs from the 100 Ohm
required by the Wilkinson splitter. This results in a possible mismatch in the system,
however the time span of the project did not allow us to improve the match between the
voltage doubler and Wilkinson splitter.

Demodulator

The demodulator is a so-called crystal receiver, as shown in Figure 5.9(b). Initially a


crystal was used in such a circuit, but nowadays a diode is used. This receiver is actually
a rectifier itself, however it is not used to subtract power. This demodulator is used at the
receiver side. At the transmitter side the carrier was modulated with the data. Here, the
data is recovered. A series capacitance at the output of the demodulator can be placed
to suppress the DC term of the demodulated signal. In the results of the measurements
which are shown next, the output of the demodulator is filtered with a 50 Hz filter and a
low-pass filter, with a cutoff frequency of 10kHz.

69
5.4.1 Measurements

The prototype system described in the preceding section has been measured and compared
to the described models given throughout this report.
The same Wilkinson splitter that was measured in Chapter 2 with K 2 = 1 for 2.45 GHz
was used, so it has not been measured separately here. The voltage doubler that was used
in the system was designed with the model given in Chapter 4. The same rectifier was used
in the single rectenna described before so this rectifier has not been measured separately
either. The patches that were used have been described and measured in Chapter 3. The
modulation system had not been not measured before so it was measured separately first.
The data source was set to create a constant 1 kHz block wave and the amplitude and
offset of the block wave was set in such a way that the output wave of the modulator was
a 10 or 11 dB attenuation of the input wave of the modulator. This was done because
it turned out that the switch caused some losses itself, so no signal could be transferred
through the modulator without attenuation. However, the extra 10dB attenuation was
made up for by the power levels of the input RF source and the amplifier. The amplifier’s
gain cannot be set and equals 30 dB, so the RF-source is set to create a 0 dBm signal,
resulting in the desired output power of the amplifier of roughly 20 dBm. The modulator’s
performance was measured by means of a network analyzer. The measurements showed
that the modulator modulates the signal up to 100 kHz without any problems, so it could
be used without any problems in our system. Next, the output of the modulator was led
directly to the demodulator, so the original shape of the block wave should be recovered.
In this way it was perfectly possible to recover the original (dummy) data stream. To make
this subsystem more perceptible the output of the demodulator was led to a crystal ear
plug, so the data ‘could be heard’. This earplug creates audible waves for input waves with
amplitudes as low as 5 mV.

Once all the subsystems were measured and gave results that were satisfactory enough
to use them further, the subsystems were connected as shown in Figure 5.8. First of all,
the output of the demodulator was measured with an oscilloscope. This output voltage is
shown in Figure 5.10. The series capacitance, as mentioned before, was used to suppress
the DC-term in the output of the demodulator. A filter has been applied to the received
signals to suppress 50 Hz noise. This was necessary because the measurement signal picked
up a 50 Hz noise due to the measurement cables and the surrounding instruments. The
signal that was measured directly at the output of the demodulator also suffered from noise
at higher frequencies, because the signal levels are quite small. However, as can be seen
in the figure the signals can be used to recover the original data. The period time of 1 ms
(from the 1 kHz input wave) can be observed clearly. When this signal is led to the ear
plug, that was described before, the modulator’s input frequency can be heard.

Another measurement was performed to measure the unloaded DC voltage for a varying
frequency. The network analyzer was used to perform a frequency sweep from 2.25 to

70
Figure 5.10: Demodulated signal

2.65 GHz and an oscilloscope was used to to measure the DC voltage over this frequency
range. We have used a broadband horn antenna as transmit antenna. As shown in Chap-
ter 3 we have used a patch antenna with two perpendicular linear polarizations. To use the
presented model to determine the input impedance of the patch antenna both polarizations
must be excited equally. Since we only had a linearly polarized horn antenna as transmit
antenna at our disposal we placed this antenna in such a way that the polarization tilt
angle was 45 degrees compared to the horizontal plane. The received wave can be split up
into two linear perpendicular polarized waves,
√ corresponding to the two polarizations of
the patch. Each polarization contains 1/ 2 of the received electromagnetic field and half
of the received power. The measurement was set up in such a way that for each frequency
a 0 dBm signal reached the patch antennas. Because of the broadband horn antenna that
was used, this resulted in a source with constant power over the entire frequency range.
The results of this measurement are shown in Figure 5.11. In this figure the voltages that
were determined from the presented models are shown as well. In the next section it will
be explained how these voltages have been obtained and how reliable the modeling of the
system is.

5.4.2 Verification

The presented models throughout this report have been used to determine the unloaded
DC output voltage of the total system. We have assumed that no power is dissipated in
the demodulator and that all the collected power is transferred to the rectifier.
Let us first look at the patch antennas. The horn antenna was used to create a field
containing 0 dBm of power at the patch antennas divided equally over both polarizations
of the patch antennas. The input impedance of the patch antennas over the frequency
range has been determined from the theory which was presented in Chapter 3.
Next, the input impedance of the Wilkinson splitter at port 2 and 3 (which are equal

71
Figure 5.11: DC output voltage of the prototype

72
because a K 2 = 1 splitter was used) has been determined over this same frequency range.
Here, the input impedance of the voltage doubler with 0 dBm input power is used as
resistive element in the splitter. This simplification is allowed since the voltage doubler’s
input impedance does not drastically change with input power. Further, our system is
designed assuming 1 mW input power, so we especially want to compare measurements
and model for this power level. Now we know the complex output impedance of the patch
antennas and the complex input impedances at the input ports of the splitter we can
determine the accepted power. In [27] the power reflection coefficient for a complex load
and complex source impedance is determined as

ZL − ZS∗ 2

2
|s| =
, (5.3)
ZL + ZS

where ZS is the complex source impedance (patch antenna output impedance) and ZL is
the complex load impedance (splitter input impedance).
By using this accepted power the DC voltage of the voltage doubler is determined. These
DC voltages are shown in Figure 5.11.

5.4.3 Discussion

In Figure 5.11 we see the measured DC-voltage of the total system over a frequency range
from 2.25 to 2.65 GHz. The DC-voltage determined by combining the different models is
also shown in this figure. The distance between transmit and receive antennas was chosen
such that the received power was roughly 1mW, using Equation (5.1). This amount of
power could not be determined very precisely because this measurement was not performed
in an anechoic chamber. We can see that the voltages that are predicted by the models
agree with the measured values. This implies that we can combine the different models,
which results in accurate predictions of the individual outputs, to determine the total
system behaviour. If the system is used at the conditions it is designed for (frequency and
power) the output voltage is predicted within 8%.
We clearly see two peaks in the figure, corresponding to the resonance frequency of each
polarization at a different frequency. When we increase the angle of the polarization of the
transmit antenna we see an increase in the hight of one peak and we see the other peak
decreasing. This results from the increase of power in one polarization of the patch and a
decrease in the perpendicular polarization. When we compare these voltages to the ones
obtained in [2] we can see that this system is much more broadband. This is due to the
Wilkinson splitter, which still accepts power when the system is operated off resonance,
and due to the double resonance of the patch antennas.

The rectenna efficiency of the total system has been measured as described before in
Section 5.2.1. An efficiency of 25% was observed with a load impedance of 900 Ω at

73
2.4 GHz and an input power of 0 dBm per antenna. At a first glance this efficiency might
seem poor. However, if we would have used a separate antenna for power conversion and a
separate antenna for data purposes we would have achieved an efficiency of 26% using the
rectenna described in Section 5.2. Our system can be optimized further to give a higher
rectenna efficiency.

1. The input impedance of the rectifier can be transformed to match the desired impedance
of the Wilkinson splitter to accept more power at the input ports of the splitter. A
simple technique to achieve this is to use pieces of microstrip line with certain lengths.
However, this is a very narrowband solution and more elegant solutions are possible,
as described in [28].

2. Further, a Schottky diode with better rectifying capabilities could be used to optimize
the efficiency. We have used a diode with a substrate resistance of 25 Ω, while in [26]
it is reported that a Schottky diode was used with a substrate resistance of only 4 Ω.
This diode was not available to us, so the HSMS-2852 Schottky diode was used.

3. Using another substrate can also improve the efficiency since the loss tangent of the
used FR4 results in substrate losses. An example of a substrate that could improve
the efficiency is Rogers 4003. We have chosen to use FR4 throughout this project
because it can be easily processed using simple etching techniques and it is relatively
cheap. It is difficult and time consuming to apply a photosensitive layer on standard
microwave laminate while FR4 is normally produced with the photosensitive layer
already on the material.

5.5 Conclusion

In this chapter we have combined the subsystems described in the preceding chapters, first
to form a rectenna and second to form a system capable of simultaneous power and data
transfer.

First, a single rectenna was designed, measured and compared to our modeling. The
unloaded DC-output voltage for different input powers was predicted well, within a few
percent, by combining the different models. Further, the load impedance resulting in
an optimal DC-power consumption was determined to be around 900 Ω. The rectenna
efficiency was measured to be 52%.

Eight of these same rectennas were combined to make a wireless powered wall clock. The
performance of this clock was measured quantitatively. The rectenna-powered wall clock
operates even on WLAN.

74
The total system to transfer data and power wireless was presented. This system used some
new building blocks, such as the modulator and demodulator, which have been explained.
A carrier was modulated with a basic digital signal. The system was able to transmit, re-
ceive and demodulate the modulated carrier, so the original data was recovered. Further,
we have been able to simultaneously transmit and receive power. The rectenna efficiency of
this system turned out to be 25%. The modeling was able to predict the system behaviour.
Recommendations have been given to increase the efficiency of the single rectenna and the
total system.

75
76
Chapter 6

Conclusions and Recommendations

The modeling of a rectenna system capable of simultaneous wireless power and data transfer
has been investigated. The design of such a system has been presented and the modeling
has been compared to measurements on a realized prototype. It has been shown that these
analytical models predict the behaviour of the system accurately enough to be used as
design tool when used appropriately.

The analytical models give insight in the behaviour of the individual parts of the system.
As shown throughout this report the design parameters, such as the dimensions, of the
individual parts of the system are very critical with respect to the behaviour of the total
system. A small change in the length or width of a design can result in a huge change in
behaviour. This implies that this system is not suitable to be designed with trial-and-error
techniques and these models can save designers lots of work and time since our models are
much faster than full wave simulations.

The approach that was followed in this project was to use models that were suitable to be
scaled. This means that the presented models can be used for a wide range of frequencies,
power levels and substrate materials.

A single rectenna was modeled, designed, produced and measured. This rectenna has
small dimensions and a relatively high efficiency of 52% under the given circumstances.
The small dimensions were especially achieved due to the elimination of the impedance-
matching circuit. This miniaturization was allowed because the antenna was matched
directly to the rectifying circuit. Compared to results from other rectenna designs this
rectenna performs quite well, especially when the low power levels are taken into account.
A voltage doubler was used in the rectenna, resulting in an unloaded DC output voltage of
around 1.2 V, with the assumed input power of 1 mW. This voltage is high enough to power
integrated circuits, which makes the rectenna suitable for a wide range of applications.

77
The individual subsystems have been combined to form the total system capable of wireless
power and data transfer. This system was manufactured and measured. The results
showed that the system is capable of transmitting and recovering the data. Further, it
simultaneously converts RF power to DC power. The DC output voltage was measured
and compared to the predicted values by the combined models. The models are capable
of predicting the behaviour of the DC output voltage, within 8% at the frequencies it is
designed for. The rectenna efficiency of this system has been determined to be 25% for a
load of 900 Ω.

Improvements in the system performance can be achieved by matching the input impedance
of the rectifier better to the required impedance of the splitter. Further, a microwave
substrate with lower losses will also result in an improved system performance.

The Schottky diode that was used throughout this project is not an optimal rectifier diode.
Especially the high substrate resistance results in a poor rectifying efficiency. Another
Schottky diode with a lower substrate resistance will result in a better system performance.
To a lesser extent, a diode with a lower junction capacitance can improve the results.

To improve the reliability of the data transfer the amplitude modulation that was used in
the prototype can be replaced by another modulation technique, e.g. frequency modulation,
that is less sensitive to noise.

Further, it could be investigated whether an array of antennas before the rectifier is an


option for rectenna purposes. The array of antennas could collect more energy than one
single antenna making the rectifying process much more efficient.

In this report the design of a system has been presented that, to our knowledge, has not
been presented in literature before. Here, an initial modeling and design approach was
proposed which can be used for further improvement and development.

78
Bibliography
Note: The number(s) following a bibliography item indicate the page(s) where the
reference is used in this report.

[1] .W.C. Brown. The history of power transmission by radio waves. IEEE Transactions
on Microwave Theory and Techniques, 1984 5

[2] J.A.G. Akkermans. Design of a rectenna for wireless low-power transmission. Tech-
nical report, Eindhoven, University of Technology, 2004. 6, 20, 47, 48, 52, 60, 64,
73

[3] L.I. Parad and R.L. Moynihan. Split-tee power divider. IEEE Trans. MTT, pages
91–95, January 1965. 9, 11

[4] K.C. Gupta, Ramesh Garg and Rakesh Chadha. Computer Aided Design of Microwave
Circuits. Artech, 1981. 9, 24

[5] Sadiku. Elements of Electromagnetics. Saunders College Publishing, 2 edition, 1994.


11, 27, 54

[6] E. Hammerstad and Ø. Jensen. Accurate models for microstrip computer-aided design.
Microwave Symposium Digest, 80:407–409, May 1980. 11

[7] W.J. Getsinger. Microstrip dispersion model. IEEE Transactions on Microwave The-
ory and Techniques, pages 34–39, January 1973. 12

[8] A. van de Capelle. Transmission Line Model for Rectangular Microstrip Antennas.
Peter Peregrinus, 1989. 12

[9] David M. Pozar. Microwave Engineering. New York: John Wiley and Sons, 1997. 14

[10] Keith R. Carver and James W. Mink. Microstrip antenna technology. IEEE Trans-
actions on Antennas and Propagaion, pages 3–25, January 1981. 21, 32, 33

79
[11] William F. Richards, Yuen T. Lo and Daniel D. Harrison. An improved theory for mi-
crostrip antennas and applications. IEEE Transactions on Antennas and Propagaion,
pages 38–46, January 1981. 21

[12] C. A. Balanis. Antenna Theory. New York: John Wiley & Sons, Inc, 2nd edition,
1983. 25, 52

[13] J. Huang. The finite ground plate effect on the microstrip antenna radiation patterns.
IEEE Transactions on Antennas and Propagaion, pages 649–653, July 1983. 26

[14] A. Smolders. Elektromagnetische antennes. Syllabus, Eindhoven University of Tech-


nology, 2000. 26, 41

[15] R.F. Harrington. Time Harmonic Electromagnetic Fields. New York: McGraw-Hill,
1964. 29, 31

[16] J. Watkins. Circular resonant structures in microstrip. Electronic Letters, pages


524–525, October 1969. 31

[17] P. Hammer, D. van Bouchatte, D. Verschraeven and A. van de Capelle. A model for
calculating the radiation field of microstrip antennas. IEEE Transactions on Antennas
and Propagaion, 27(2):267–270, March 1979. 33

[18] A. Tavakoli, A. Monajati and R. Moini. A wide band dual-polarized microstrip patch
antenna. Antennas and Propagation Society International Symposium, 2:956–599,
July 1997. 36

[19] C.S. Lee and V. Nalbandian. Planar circularly polarized microstrip antenna with a
single feed. IEEE Transactions on Antennas and Propagaion, 47:1005–1007, June
1999. 36

[20] Agilent Technologies. Application note 1088, designing the virtual battery, 1999. 44,
52

[21] Dominic A. Fleri and Leonard D. Cohen. Nonlinear analysis of the schottky-barrier
mixed diode. IEEE Transactions on Microwave Theory and Techniques, pages 39–43,
January 1973. 45

[22] S. Pendharkar and K. Shenai. A circuit simulation model for high-power high-speed
GaAs Schottky diodes. Conf. Proc. IEEE APEC ’96, pages 246–249, March 1996. 46

[23] G.H. Stauffer. Finding the lumped element varactor diode model. High Frequency
Electronics, pages 22–28, November 2003. 46

80
[24] W.H. Press, S.A. Teukolsky, W.T. Vetterling and B.P. Flannery. Numerical Recipes
in C. Cambridge: Cambridge University Press, 2nd edition, 1992. 48, 50

[25] Agilent Thechnologies. Surface Mount Zero Bias Schottky Detector Diodes, HSMS-
2850 Series, 1999. 51, 53

[26] J.H. Suh and K. Chang. A high-efficiency dual frequency rectenna for 2.45 and
5.8 GHz wireless power transmission. IEEE Transactions on Microwave Theory and
Techniques, pages 1784–1789, July 2002. 64, 74

[27] Pavel V. Nikitin, K. V. Seshagiri Rao, Sander F. Lam, Vijay Pillai, Rene Martinez
and Harley Heinrich. Power Reflection Coefficient Analysis for Complex Impedances
in RFID Tag Design. IEEE Transactions on Microwave Theory and Techniques, pages
2721–2725, September 2005. 73

[28] Agilent Technologies. Impedance matching techniques for mixers and diodes. Appli-
cation note, 1999. 74

81

Вам также может понравиться