RF Energy Harvesting System Design

UNIVERSIDAD POLITÉCNICA DE MADRID
ESCUELA TÉCNICA SUPERIOR

DE INGENIEROS DE TELECOMUNICACIÓN
MASTER UNIVERSITARIO EN INGENIERÍA DE TELECOMUNICACIÓN
TRABAJO FIN DE MASTER
TITULO: DESIGN OF A RADIOFREQUENCY (RF)

ENERGY HARVESTING SYSTEM FOR LOW-POWER
SENSOR APPLICATIONS AT MICROWAVE BANDS
AUTOR: Antonio Alex Amor
AÑO: 2018
(Página en blanco)
TÍTULO: DESIGN OF A RADIOFREQUENCY (RF) ENERGY
HARVESTING SYSTEM FOR LOW-POWER SEN-
SOR APPLICATIONS AT MICROWAVE BANDS
AUTOR: Antonio Alex Amor

TUTOR: José Manuel Fernández González
COTUTOR: Pablo Padilla de la Torre
DEPARTAMENTO: Señales, Sistemas y Radiocomunicaciones
MIEMBROS DEL TRIBUNAL CALIFICADOR
PRESIDENTE:
VOCAL:
SECRETARIO:
SUPLENTE:
FECHA DE LECTURA:
CALIFICACIÓN:
Contents
1 Introduction 14
1.1 Energy Harvesting. Full System Architecture . . . . . . . . . . . . . . 14
1.2 Radio Spectrum. Bands of Interest . . . . . . . . . . . . . . . . . . . 15
1.3 Project Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4 State of the Art of RF Energy Harvesting Systems . . . . . . . . . . . 17
2 Radiofrequency Harvester 20
2.1 Design of an Archimedean Spiral Antenna . . . . . . . . . . . . . . . 20
2.2 Comparison between a Multiband PIFA and an Ultrawideband
Archimedean Spiral Antenna . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.1 New Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Miniaturization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Microstrip to Parallel Strip Balun. Ultrawideband Design . . . . . . . 32
2.4.1 Back-to-Back . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.2 Balun-Antenna Connection . . . . . . . . . . . . . . . . . . . 36
2.5 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.1 Reflection Coefficient Measurement . . . . . . . . . . . . . . . 37
2.5.2 Power Spectrum Measurement . . . . . . . . . . . . . . . . . . 38
3 Conditioning Circuit. Cockcroft-Walton Multiplier 42

3.1 Design of the Half-Wave Cockcroft-Walton Multiplier . . . . . . . . . 42
3.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4 Matching Circuit 48
4.1 Narrow-band modeling of the Archimedean spiral antenna . . . . . . 49
4.2 Equivalent Circuit Model. Test Circuits . . . . . . . . . . . . . . . . . 51
4.2.1 First Model. Test Circuit #1 . . . . . . . . . . . . . . . . . . 52
4.2.2 Second Model. Test Circuit #2 . . . . . . . . . . . . . . . . . 53
4.2.3 Third Model. Test Circuit #3 . . . . . . . . . . . . . . . . . . 56
4.3 Nonlinearity Effects. Circuit #3 . . . . . . . . . . . . . . . . . . . . . 57
4.4 Comparison Among Different Number of Stages . . . . . . . . . . . . 58
5 Conclusion & Future Work 62

5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4
CONTENTS 5
A Harmonic Balance Method 65

A.1 A Simple Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
B Design of L-Matching Networks 68

B.1 Matching Circuit in the Energy Harvesting System . . . . . . . . . . 69
C Reflection on Ethical, Economic, Social and Environmental As-

pects 70
D Cost Estimation 73
E Publications 75
List of Figures
1.1 Full RF energy harvesting system. . . . . . . . . . . . . . . . . . . . . 15

1.2 Section of the electromagnetic spectrum. Source: [4]. . . . . . . . . . 16
1.3 Three-Element Dual-Band Yagi Rectenna (b) and a single element
(equiangular spiral antenna) of the rectenna array implemented in [3]
(b). Sources: [9] and [3], respectively. . . . . . . . . . . . . . . . . . . 17
1.4 Simulated range of optimal source impedances for one of the diodes
used in [3]. Source: [3]. . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 Full-wave rectenna shown in [12](a), and a rectenna based on a
Cockcroft-Walton multiplier [13](b). . . . . . . . . . . . . . . . . . . . 19
2.1 Current distribution over a miniaturized Archimedean spiral antenna. 21

2.2 Archimedean spiral antenna. . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Reflection coefficient of the Archimedean spiral antenna (normalized
to 188.5 Ω). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Axial ratio (AR) of the Archimedean spiral antenna at different fre-
quencies in a cut in the plane φ = 90o . . . . . . . . . . . . . . . . . . 23
2.5 Total efficiency at different frequencies of the Archimedean spiral an-
tenna. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Farfield radiation pattern at different frequencies of the Archimedean
spiral antenna. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.7 Parts of a dual-band PIFA antenna. . . . . . . . . . . . . . . . . . . . 25
2.8 Reflection coefficient of the dual-band PIFA antenna. . . . . . . . . . 26
2.9 Tri-band PIFA antenna. . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.10 Transversal cut of the proposed multiband PIFA (a) and its top (b)
and bottom (c) planes. . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.11 Inclination of PIFA’s patches. . . . . . . . . . . . . . . . . . . . . . . 27
2.12 Monte Carlo simulation on the reflection coefficient of the proposed
multiband antenna when varying ±2o the inclination of the plane of
both PIFAs. The purple thick line represents the ideal scenario (null
inclination) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.13 Farfield radiation pattern at different frequencies of PIFA antenna. . . 28
2.14 Different designs of the Archimedean spiral antenna: (a) no minia-
turization [14], (b) miniaturization without impedance step, and (c)
miniaturization with an impedance step. . . . . . . . . . . . . . . . . 30
6
LIST OF FIGURES 7
2.15 Simulated reflection coefficient of the Archimedean spiral antenna

(normalized to 188.5 Ω) with different designs: (a) no miniaturiza-
tion [14], (b) miniaturization without impedance step, and (c) minia-
turization with an impedance step (20.72 x 19.77 cm vs 15.22 x 14.27
cm). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.16 Simulated efficiencies (radiation efficiency and total efficiency) of the
miniaturized (one impedance step) Archimedean spiral antenna, and
two examples of the spiral’s 3D radiation pattern at two different
frequencies of interest. . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.17 Simulated axial ratio, at different frequencies, of the Archimedean
spiral antenna in two cuts in the planes φ = 0o (continuous line) and
φ = 90o (dashed line). . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.18 Simulated farfield patterns of the miniaturized Archimedean spiral
antenna at the different frequencies of interest on the (a) vertical and
(b) the horizontal plane. . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.19 Ultrawideband microstrip to parallel strip balun. . . . . . . . . . . . 33
2.20 Electric field in the parallel strips’ port, and the QTEM (a) and the
evanescent hybrid (b) modes. . . . . . . . . . . . . . . . . . . . . . . 34
2.21 Simulated S-parameters of the microstrip to parallel strips balun.
Note how the hybrid mode is quite attenuated with respect to the
QTEM mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.22 Back-to-back configuration of the UWB microstrip to parallel strip
balun. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.23 Simulated S-parameters of the microstrip to parallel strips back-to-
back configuration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.24 Balun-antenna connection. . . . . . . . . . . . . . . . . . . . . . . . . 36
2.25 Reflection coefficient of the joint structure (antenna + balun). . . . . 37
2.26 Prototype of the miniaturized Archimedean spiral antenna. . . . . . 37
2.27 Comparison of the simulated and measured reflection coefficients of
the Archimedean spiral antenna (normalized to 188.5 Ω). . . . . . . 38
2.28 Power spectrum measured by the Archimedean spiral antenna inside
and outside the laboratory with the most relevant bands remarked:
1-FM, 2-DTT, 3-LTE-800, 4-GSM-900, 5-GSM-1800, 6-LTE-2100, 7-
WiFi, 8-LTE-2600, 9-WiFi. . . . . . . . . . . . . . . . . . . . . . . . 39
2.29 Measured reflection coefficient of the manufactured 2.4-GHz patches
(Dimensions: small patch: 4.89 x 3.91 cm, circularly-polarized patch:
5.05 x 3.91 cm, big patch: 5.02 x 4.07 cm. . . . . . . . . . . . . . . . 40
2.30 Comparison between the power spectrum measured by the
Archimedean spiral antenna inside the laboratory and by the three
2.4-GHz patches, with the most relevant bands remarked: 1-FM, 2-
DTT, 3-LTE-800, 4-GSM-900, 5-GSM-1800, 6-LTE-2100, 7-WiFi, 8-
LTE-2600, 9-WiFi. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
LIST OF FIGURES 8
3.1 Circuital scheme of the n-stage half-wave Cockcroft-Walton voltage

multiplier: the Villard circuit and the half-wave rectifier are circled in
orange and in red, respectively, and the Greinacher voltage doubler
is remarked in dashed blue line. . . . . . . . . . . . . . . . . . . . . . 42
3.2 HWCW test set ((a) upper side, (b) bottom side) with six different
combinations of components and number of stages: 33 nF – HSMS
2822 – 3 stages (blue), 33 nF// 33 pF – HSMS 2822 – 2 stages (pur-
ple), 33 pF – HSMS 2850 – 2 stages (green), 33 pF – HSMS 2822 – 2
stages (orange), 33 pF – HSMS 2850 (red) – 2 stages, 33 pF – HSMS
2822 – 5 stages (cyan). . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3 Measurement set-up: Agilent E4438C vector signal generator, Yoko-
gawa DL9240L digital oscilloscope, HWCW test set, and a multimeter. 45
3.4 Output DC voltages in a 2-stage HWCW scheme when using different
components (no-load situation). . . . . . . . . . . . . . . . . . . . . . 46
3.5 Output voltage versus frequency in the 2-stage HWCW scheme (with
C=33 pF) at different input powers. . . . . . . . . . . . . . . . . . . . 47
3.6 Output voltage at each state of the 5-stage HWCW scheme (with
C=33 pF) for an input power of -10 dBm. . . . . . . . . . . . . . . . 47
4.1 General scheme of the RF harvesting system. Note as the source

impedance Zs depends on five terms: the input power Pin , the input
range of frequencies f , the number of stages N in the Cockcroft-
Walton multiplier , the intrinsic impendance of the antenna Zant ,
and the load ZL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 Scheme of the RF harvesting system particularized to the use of
the Archimedean spiral antenna. Notes: *Pin1 = −5.80 dBm and
Pin1 = −1.84 dBm, inside and outside the laboratory, respectively;
and Pin2 = −9.20 dBm and Pin1 = −2.34 dBm, inside and outside
the laboratory, respectively. **fin1 = 807 MHz, and fin2 = 942 MHz. 49
4.3 Zoom over the power spectrum measured by the Archimedean spiral
antenna at 800 and 900 MHz. The delta functions marked in blue
and orange represent the modeling of the Archimedean spiral antenna
as a generator, inside and outside the laboratory, respectively. . . . . 50
4.4 Narrow-band circuital model of the antenna. . . . . . . . . . . . . . 50
4.5 Test circuit #1 (a) and its equivalent model (b). The source
impedance that maximizes the power delivered to the load is Zs =
60 + j200 Ω. Values of the components: Lmat = 33 nH, C = 33 pF,
RL = 6.1 kΩ. Values of the parasitics: Cp = 0.77 pF, Cpd = 0.70 pF,
Lp = 2.64 nH. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.6 Measurements relative to test circuit #1 (blue line) and the simula-
tions of the circuit without parasitic elements (red line) and the first
circuit model (magenta line). Cp = 0.77 pF, Cpd = 0.70 pF, Lp = 2.64
nH. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
LIST OF FIGURES 9
4.7 Test circuit #2 (a), first model applied on test circuit #2 (b), and the
final equivalent model (c) that takes into account the parasitic ele-
ments of the board. The source impedance that maximizes the power
delivered to the load is Zs = 50 + j78 Ω. Values of the components:
Lmat1 = 4.7 nH, Lmat2 = 8.2 nH, C = 33 pF, RL = 2.34 kΩ. Values
of the parasitics: Cp1 = 0.70 pF, Cp2 = 0.40 pF, Cpd1 = 0.70 pF,
Cpd2 = 0.10 pF, Lpd = 1.5 nH, Lp = 1.5 nH, Lvia = 0.5 nH. . . . . . . 54
4.8 Measurements relative to test circuit #2 (blue line) and the simu-
lations of the circuit using the first circuit model (red line) and a
redesigned second circuit model (gold line). . . . . . . . . . . . . . . . 54
4.9 Measurement set-up of test circuit #2. . . . . . . . . . . . . . . . . . 55
4.10 Circuit #3 (a) and the third model applied on it (b). The source
impedance that maximizes the power delivered to the load is Zs =
18 + j56 Ω. Values of the components: Lmat1 = Lmat22 = 6.8 nH,
Lmat21 = 3.3 nH, Cmat1 = 1.2 pF, Cmat21 = Cmat22 = 1.5 pF, C = 33
pF, RL = 2.34 kΩ. Values of the parasitics: Cp1 = Cp22 = 2.2 pF,
Cp21 = 1.3 pF, Cpd1 = 0.75 pF, Cpd2 = 0.10 pF, Lpd = 1.5 nH,
Lpd1 = Lpd2 = 0.5 nH, Lp = 1.5 nH, Lvia = 0.55 nH. . . . . . . . . . . 56
4.11 Measurements relative to test circuit #3 (blue line) and the simula-
tions of the circuit using the first circuit model (red line), the second
circuit model (gold line), and a redesigned third model (purple line). . 57
4.12 Nonlinearity in circuit #3 caused by different input power levels. . . . 58
4.13 Measured efficiency in circuit #3 as a function of the input power. . . 58
4.14 Efficiency of the simulated circuit that includes the optimum source
impedances as a function of the load RL . . . . . . . . . . . . . . . . . 60
4.15 Efficiency of the simulated circuit that implements the final L-
matching networks as a function of the load RL . . . . . . . . . . . . . 61
A.1 Partitioning of a microwave circuit into linear and non-linear subcir-

cuits. The source Zs (ω) and load ZL (ω) impedances have been placed
into the linear circuit. The circumflex accent indicates a non-linear
term. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
A.2 A diode D connected to a lumped element with an sinusoidal excita-
tion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
B.1 The two possible configurations of an L-matching network. . . . . . . 69

List of Tables
2.1 Measurements on the harvested power. . . . . . . . . . . . . . . . . . 41
4.1 Simulated optimum source impedances Zs , load resistances RL , out-

put voltages Vo , output currents Io , output powers Po , and efficiencies
for the different number of stages of the multiplier circuit with the
”outside” modeling of the antenna. . . . . . . . . . . . . . . . . . . . 59
4.2 Simulated optimum source impedances Zs , load resistances RL , out-
put voltages Vo , output currents Io , output powers Po , and efficiencies
for the different number of stages of the multiplier circuit with the
”inside” modeling of the antenna. . . . . . . . . . . . . . . . . . . . . 59
D.1 Cost Estimation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
10
List of Abbreviations and
Acronyms.
AC — Alternating Current
ADS — Advanced Design System
AR — Axial Ratio
CMOS — Complementary Metal-Oxide-Semiconductor
DC — Direct Current
DCR — Direct Current Resistance
DTT — Digital Terrestrial Television
ECG — Electrocardiogram
EM — Electromagnetic
ESR — Equivalent Series Resistance
EU — European Union
FM — Frequency Modulation
FR4 — Flame-Retardant 4
FSA — Fibonacci Spiral Antenna
GSM — Global System for Mobile Communications
HB — Harmonic Balance
HWCW — Half-Wave Cockcroft-Walton
IoT — Internet of Things
LTE — Long Term Evolution
PIFA — Planar Inverted-F Antenna
PCB — Printed Circuit Board
RF — Radiofrequency
SMA — SubMiniature version A
SMT — Surface-Mount Technology
SRF — Self-Resonant Frequency
UWB — Ultrawideband
VNA — Vector Network Analyzer
WiMAX — Worldwide Interoperability for Microwave Access
WSN — Wireless Sensor Network
11
Design of a Radiofrequency (RF) Energy Harvesting System
for Low-Power Sensor Applications at Microwave Bands
Antonio Alex Amor
Keywords: energy harvesting, antenna design, Archimedean spiral antenna,

ultrawideband, matching circuit, Cockcroft-Walton multiplier, storage circuit.
Abstract
Energy harvesting is defined as the process of obtaining energy from external

sources, such as wind, sunlight, heat or radiofrequency (RF) transmitters; which
is later conveniently processed and stored in order to feed, in general, low-power
electronic systems. Generally, a RF energy harvesting system is formed by four
stages:
• RF harvester: an antenna capable of acquiring RF ambient power from the

bands of interest of the radio spectrum.
• Conditioning circuit: a circuit that converts the signal acquired by the antenna,
which is a sum of radio spectrum carriers, into a DC supply.
• Matching circuit: it is situated between RF harvester and conditioning circuit

stages. It maximizes the power delivered from the antenna to the conditioning
circuit.
• Storage circuit: it stores the available energy in order to feed an electronic

system (typically a sensor).
The aim of this project is the design of a radiofrequency energy harvesting system
as a “free” power source for low-power electronic systems. The work is presented
with the following phases and main development tasks:
Firstly, a set of different antennas are being studied. We will focus this
chapter on the study of multiband, wideband, ultrawideband, and frequency
independent antennas. Simulations will be carried out with CST Microwave Studio
Suite, in order to optimize the design that will later be manufactured and measured.
Secondly, we will focus on the conditioning circuit. One of the topologies of

our interest is the Cockcroft-Walton voltage multiplier, a circuit that rectifies and
elevates the signal acquired by the antenna. We will simulate and test different
element configurations on it.
Thirdly, we will maximize the power transfer (from the antenna to the circuit)
by placing an impedance transformer circuit. We must study which is the
LIST OF TABLES 13
source impedance (the receiver antenna can be seen as a generator with a source
impedance) that maximizes the output power, and then transform the antenna
impedance into this one. For this task, we will lay on the use of simulators
that implement “harmonic balance”, a technique commonly used to calculate the
steady-state response of nonlinear differential equations, and it is mostly applied to
nonlinear circuits (as the Cockcroft-Walton multiplier).
Finally, we will design an energy storage circuit in order to conveniently store and
manage the previously conditioned power. It actually serves as a battery for the
low-power electronic system to be fed. Supercapacitors (capacitors of units or tens
of Farads) seem as the best candidate for this purpose.
Chapter 1
Introduction
1.1 Energy Harvesting. Full System Architecture

Energy harvesting is defined as the process of obtaining energy from external
sources, such as wind, sunlight, heat or radiofrequency (RF) waves; which is
later conveniently processed and stored in order to feed, in general, low-power
electronic systems. The most common energy harvesting techniques [1] are based
on the reuse of the sunlight; on the reuse of kinetic energy, through mechanic
movements or deformations inside the device; on the reuse of thermal energy,
through thermocouples that take advantage of the Seebeck effect [1]; and on the
capture of electromagnetic radiation. The constant increase in the number of RF
transmitters has led to a non-negligible level of power at some particular frequency
bands. This fact, joined to the reduction in the power consumption necessities of
many devices, has turned RF energy harvesting into a feasible energy source for
such low-consumption devices. For instance, as discussed in [2], several harvesting
devices are starting to be used for medical purposes, in order to measure intraocular
pressure, temperature, electrocardiograms (ECG), etc.; or in fields as wireless sensor
networks (WSN) and Internet of Things (IoT). Most of these devices consume less
than 30 µW.
Typically, a RF energy harvesting system is formed by four stages [3], as shown

in Fig. 1.1.
• RF harvester: an antenna capable of acquiring RF ambient power from the

bands of interest within the radio spectrum.
• Conditioning circuit: a circuit that converts the signal acquired by the antenna,
which is a sum of radio spectrum carriers, into a DC supply.
• Matching circuit: it is situated between RF harvester and conditioning circuit

stages. It maximizes the power delivered to the conditioning circuit.
• Storage circuit: it stores the acquired power in order to feed a certain electronic
system (typically a sensor).
14
CHAPTER 1. INTRODUCTION 15
Fig. 1.1: Full RF energy harvesting system.
1.2 Radio Spectrum. Bands of Interest

We name by electromagnetic spectrum to the range of all possible electromagnetic
radiations. It extends approximately from radio long waves to gamma rays. The
wavelength λ of a disturbance and the frequency of it, f , are linked by the inversely
proportional relationship
c
λ= (1.1)
f
where c is the propagation speed of the wave in a certain medium. Fig. 1.2 presents
a section of the electromagnetic spectrum as a function of the frequency/wavelength
(top/bottom), and the most typical applications withing the radio spectrum. In
the vast majority of countries, almost all relevant RF commercial applications are
located between 80 MHz and 5 GHz. In our particular case, we should pay special
attention to 500, 800, 900, 1800, 2400, 3500 and 5000 MHz bands [5–7], where DTT
(Digital Terrestrial Television), LTE (Long Term Evolution), GSM (Global System
for Mobile Communications), WiFi and WiMAX (Worldwide Interoperability for
Microwave Access) are located. Another important power contribution comes from
FM band (80 - 100 MHz), but note the size of the antenna becomes quite elevated
to be considered here.
1.3 Project Description

This project presents a radiofrequency ultrawideband energy harvesting system,
based on an ultrawideband planar antenna that captures the spectrum energy, con-
nected to a voltage multiplier circuit that rectifies the signal and converts it into
DC power supply. The document is organized as follows:
Fig. 1.2: Section of the electromagnetic spectrum. Source: [4].
• Chapter 1 introduces the global design of the proposed RF energy harvesting

system and its working scheme.
• Chapter 2 presents the design of the radiofrequency harvester. Firstly, a

comparison between multiband and wideband antennas is made, coming to
the conclusion that it is usually wiser using wideband antennas instead of
multiband ones. Secondly, we focus on the study, design and simulation of
the Archimedean spiral antenna and the ultrawideband balun required to the
measurement of the antenna. Thirdly, several miniaturization techniques are
investigated to try reducing the physical dimensions of the spiral.
• Chapter 3 depicts the design of the conditioning circuit, based on a half-wave

Cockcroft-Walton multiplier. Besides that, a comparison between different low
forward voltage Schottky diodes and RF capacitors is made in order to reduce
voltage ripple and losses of the DC output signal.
• Chapter 4 studies the matching circuit. It is analyzed the effect of modifying

both the input power (power acquired by the antenna) and the load (sensor)
connected to the conditioning circuit in order to maximize the power delivered

from the antenna to the Cockcroft-Walton multiplier circuit.
• Chapter 5 summarizes the work and points out the main conclusions extracted
from it.
1.4 State of the Art of RF Energy Harvesting Sys-

tems
In this section, we illustrate the typical steps carried out in the design of each of
the four stages that form the RF harvesting system, and the most typical problems
derived from them. We also review the current RF energy harvesting systems, show-
ing the most typically used antennas, frequency bands, and conditioning circuits.
The two most commonly used types of antennas as harvesting elements are
multiband and wideband antennas. Multiband antennas are usually narrow-band,
and they are often easier to design due to their well-known performance. To achieve
multiple resonances, it is frequent to make cuts or slots in the structure, which
can be a double-edged sword since can be difficult to control the mutual coupling
between the slots themselves. On the other hand, wideband antennas can be harder
to design, although it depends on the situation. In return, its broadband behavior
assures as that the antenna certainly covers the entire region of interest. Fig. 1.3
shows two different rectenna (rectifying antennas) prototypes: one of them is based
on the three-element dual-band Yagi antenna presented in [9] (Fig. 1.3a), which
operates at 945 MHz and 2.45 GHz; and the other one is based on a equiangular
spiral antenna developed in [3] that covers from 2 GHz to 18 GHz (Fig. 1.3b).
(a) Three-element dual-band Yagi rectenna (b) Equiangular spiral
Fig. 1.3: Three-Element Dual-Band Yagi Rectenna (b) and a single element (equian-
gular spiral antenna) of the rectenna array implemented in [3] (b). Sources: [9]
and [3], respectively.
The most common, but smarter, circuit approximation to the design of the system
consists in finding the optimum source impedance that maximizes the efficiency of
the conditioning circuit. For example, 1.4 shows the simulated range of normalized
optimum source impedances over the Smith chart for a variety of studied cases:
high input power and low frequency (A), low input power and low frequency (B),
low input power and high frequency (C), high input power and high frequency (D).
Note that the shaded area is large, which indicates that the circuit has a strongly
non-linear behavior, that is, its performance depends on the input power and other
factors. Nevertheless, one problem is that the impedance of the antenna does not
usually coincide with the optimum source impedance in the operating frequency
range, so a matching network is typically needed between the antenna and the
conditioning circuit, as Fig. 1.1 illustrates.
The current bottleneck of RF harvesting systems is the conditioning circuit, which

is in charge of eliminating the AC level and rectifying the input signal. This is due
to the intrinsic consumption of the diodes placed in the circuit (they are not ideal),
which at low input power levels dissipate a high relative part of the total power. In
consequence, for input powers of less than 1 mW, the efficiency of a typical half/full-
wave rectifier, defined as the proportion between the output DC power and the input
RF power
PoutDC
η(%) = · 100, (1.2)
PinRF
Fig. 1.4: Simulated range of optimal source impedances for one of the diodes used
in [3]. Source: [3].
is typically under 30%. In [3], a half-wave rectifier circuit is implemented over an

array of equiangular spirals. The results show that the efficiency of the rectifier
ranges from 1% to 20%, depending on the input power level. Some recent (and very
advanced) studies such as [11] have achieved efficiencies of up to 74% for 1 mW input
power, by externally feeding the circuit in question (CMOS configuration) and using
the self-body biasing technique, which allows us to change the threshold voltage Vt
and turn on quicker the transistor. Nevertheless, all schemes that reach high effi-
ciencies in the rectifier circuit are not ”passive”, that is, they are externally powered.
The most used configuration for a conditioning circuit is the half-wave rectifier,
but there are other more complex configurations that are worth mentioning. For ex-
ample, [12] presents a full-wave rectenna, formed by two concentric squared patches
attached to a full-wave rectifier (Fig. 1.5a). The results show an efficiency in the
rectifier circuit of less than 34% for an input power of 1 mW. Another passive (not
externally biased) configurations arise from multiplier circuits. They are capable of
rectifying the signal and elevating the output voltage at the same time, which is of
great interest in our application since the input voltages are of the order of hun-
dreds of mV. In these terms, [13] presents a rectenna based on a coplanar broadband
antenna and a single stage of a Cockcroft-Walton multiplier.
(a) Full-wave rectenna. Source: [12]
(b) Rectenna based on a Cockcroft-Walton multi-

plier. Source [13]
Fig. 1.5: Full-wave rectenna shown in [12](a), and a rectenna based on a Cockcroft-
Walton multiplier [13](b).
Chapter 2
Radiofrequency Harvester
When choosing which type of spiral to use, factors such as the purity of the circular
polarization or the bandwidth of the spiral should be taken into account. In [16],
a comparison between equiangular and Archimedean spiral antennas is made. It
is pointed out that the Archimedean spiral shows an improved axial ratio and a
wider bandwidth for a given outer diameter. The design and manufacturing of the
Archimedes spiral is also easier than in the case of the equiangular spiral, due to
the constant angular increase of its arms. Besides that, wideband spiral antennas
may be configured with single, double or multiple arms [17], the multiple-arm spiral
antenna being a good option to provide squinted beams. In the light of previous
works [16,17], the two-arms Archimedean spiral antenna seems as the proper option
to be implemented in practice.
2.1 Design of an Archimedean Spiral Antenna

The Archimedean spiral is typically classified within the group of frequency inde-
pendent antennas, that is, antennas that maintain some of their radiation parameters
constant, as the input impedance or the bandwidth, in relation to the frequency.
Victor H. Rumsey laid the foundations of this subject. According to him [15],
those antennas that are entirely defined by angles, as the Archimedean spiral or
the equiangular spiral, show a frequency independent behavior. Theoretically, an
infinite-sized Archimedean spiral antenna in a self-complementary design (the arm
width w is equal to the gap width s) presents a constant impedance that is fixed
(according to Babinet’s principle) to the value
Zant = Zcomp = 60π Ω ≈ 188.5 Ω, (2.1)
where Zcomp is the impedance of its complement. The finite size of an Archimedes
antenna causes a finite bandwidth, which becomes more reduced as the antenna
becomes smaller, with its impedance varying nearby 188.5 Ω. This fact can be
noticed in Fig. 2.27.
On the other hand, several studies emphasize [3, 14, 18, 19] that the current distri-
bution over the surface of the Archimedean spiral is concentrated close to the center
20
CHAPTER 2. RADIOFREQUENCY HARVESTER 21
(a) 10 GHz (b) 500 MHz (c) 300 MHz
Fig. 2.1: Current distribution over a miniaturized Archimedean spiral antenna.
of the antenna at high frequencies, and opens to its end at low frequencies. This
current distribution (Fig. 2.1) leads to two equations that estimate the dimensions
of the antenna according to the desired bandwidth (determined by the lower and the
upper cutoff frequencies, fL and fH , respectively). These equations are presented
as follows
co
r1 = √ , (2.2)
2πfH εref f
co
r2 = √ , (2.3)
2πfL εref f
where r1 is the inner radius, r2 is the outer radius, co is the speed of light in the
vacuum, and εref f is the effective relative permittivity of the dielectric used (FR4:
εr = 4.7, tan δ = 0.014 @ 1MHz). Due to the complex geometry of the antenna, this
dimension estimation is revealed as a good starting point to model the Archimedes
spiral on a full-wave electromagnetic simulator. As no ground plane is used, it is
really difficult to estimate in an accurate manner εref f , so it is taken εr2+1 = 2.85 as
the best approximation to it.
Self-complementarity is only reachable if the metallic strip width, w, is equal to

the separation between two adjacent strips, s, that is, w = s. Fixing the upper and
lower cutoff frequencies as 18 GHz and 0.4 GHz, respectively, and applying (2.2)
and (2.3), it was found that the inner radius must be r1 = 1.5 mm and the outer
radius r2 = 70.7 mm. Note that these values are mainly indicative, since, as seen in
Fig. 2.1, not all the current distribution over the surface of the spiral is located at
the center or at the end of the antenna. In fact, as will be seen later, equation (2.3)
works in a worse manner than equation (2.2), due to the current distribution at low
frequencies does not really move to the end of the antenna, but to an intermediate
strip. Therefore, the real value of r2 will be higher than the one calculated in (2.3).
In Fig. 2.2, it is presented the Archimedean spiral antenna, designed over

single-sided FR4 copper clad (no ground plane). As can be noticed, the value of
the inner radius is r1 = 1.20 mm, similar to the calculated value in equation (2.2),
Fig. 2.2: Archimedean spiral antenna.
but the outer radius is slightly higher, with a value of r2 = 97.5 mm. On the
other hand, the metallic strip width w and the gap s were estimated by numerical
simulations, giving us a value of w = s = 3.9 mm.
Figure 2.3 shows the reflection coefficient of the Archimedean spiral antenna (nor-
malized to 188.5 Ω). The antenna offers a bandwidth of 17.57 GHz (|S11 | < −10 dB),
which ranges from 0.44 GHz to 18.01 GHz. It covers DTT, LTE, GSM, WiFi and
WiMAX bands, and it is circularly polarized (AR < 3 dB) in all bands but DTT,
where this value is surpassed, as depicted in Fig. 2.4. A ripple in the lower frequen-
cies of the reflection coefficient is also observed, product of the short electrical size
of the antenna. This means that the Archimedean spiral begins to show a resonant
behavior and its impedance value is no longer constant at the lower frequencies.
Fig. 2.3: Reflection coefficient of the Archimedean spiral antenna (normalized to

188.5 Ω).
Fig. 2.4: Axial ratio (AR) of the Archimedean spiral antenna at different frequencies
in a cut in the plane φ = 90o .
In Fig. 2.5, it is plotted the total efficiency of the Archimedean spiral antenna.
As can be seen, the efficiency of the spiral is superior to 82% (0.82 if we normal-
ize it to the unity) in all the bands of interest. Furthermore, the farfield radiation
pattern displayed in Fig. 2.6 shows the omnidirectionality of the antenna and its
backward radiation (due to no ground plane is used). Omnidirectionality and circu-
lar polarization are key factors in RF energy harvesting, due to we do not know the
polarization and the direction of arrival of the incoming energy. Besides that, note
that the directivity of the antenna increases slightly with frequency, having 3.57 dBi
at 830 MHz and 6.15 dBi at 2600 MHz.
Fig. 2.5: Total efficiency at different frequencies of the Archimedean spiral antenna.
Fig. 2.6: Farfield radiation pattern at different frequencies of the Archimedean spiral
antenna.
2.2 Comparison between a Multiband PIFA and

an Ultrawideband Archimedean Spiral An-
tenna
The PIFA (Planar Inverted F-Antenna) is a type of printed antenna commonly
used in mobile communications. It can be considered as a particularization of
a half-wave patch, where its center (the electric field is null at the center of a
λ/2 patch) is short-circuited to the ground plane. This technique is very smart
and allows reducing the size of the PIFA to more than half (≤ λ/4) without
degrading any of its parameters, which permits placing it in small spaces, as
mobile phones. The antenna presents a low directivity (3 – 8 dBi) and a high
efficiency (around 80 %), so, combined with some appropriate slots over its surface,
it can be transformed into a multiband antenna. Nevertheless, it is linearly
polarized and as we do not know the polarization of the incoming signals, the
depolarization losses will be higher than in the case of a circularly-polarized antenna.
The design equation of the PIFA antenna is presented in [21] and has the form
λd
L + W − Ws = , (2.4)
4
where L is the length, W is the width, Ws the short-circuited stub width, and λd
the operating wavelength of the PIFA antenna. Note that reducing Ws permits us
reducing the size of the antenna (L, W ), but the lower the stub width, the lower the
antenna bandwidth (8 %). If we design the PIFA to resonate at 930 MHz (without
any dielectric, in order to reduce losses), and we assume a stub width of Ws = 7
mm and a relation width-length W/L = 0.95, the length and the width of the PIFA
must be L = 44.9 mm and W = 42.7 mm, respectively.
In Fig. 2.7, it is presented a typical dual-band GSM PIFA. It is formed by the

elements described above (quarter-wave patch, ground plane, short-circuited stub),
which permit the antenna to resonate at 930 MHz, and a rectangular cut that
makes the antenna to resonate at 1830 MHz. No dielectric is used in order to reduce
at maximum the losses. The placement of the feeding probe as well as the aperture
of the ground plane are key factors in order to match the antenna. Note that when
the coaxial probe moves towards the short-circuited stub, the input impedance of
the antenna goes to zero (the impedance of a short-circuited stub is zero).
In Fig. 2.8, it is shown the reflection coefficient of the dual-band GSM PIFA. It
covers the 930 and the 1830 MHz bands with a narrow band. Note as well that the
matching at 930 MHz can be improved, although the -6 dB requirement in mobile
communications is fulfilled.
Fig. 2.7: Parts of a dual-band PIFA antenna.

Fig. 2.8: Reflection coefficient of the dual-band PIFA antenna.
2.2.1 New Design

More resonances can be achieved by adding U slots, as shown in Fig. 2.9. But
the fact is that adding several slots greatly hinders the design of PIFA due to the
mutual coupling between elements. Thus, it is not recommended to try to place
more than three slots in the same patch.
Fig. 2.9: Tri-band PIFA antenna.
According to the commented before, we propose a new design where two different
PIFAs are integrated in the same ground plane (90 x 55 mm), which is fed with a
single coaxial probe (SMA connector) through a distribution network printed in a
FR4 substrate (εr = 4.7, tan δ = 0.014 @ 1MHz) and situated below the ground
plane, as shown in Fig. 2.10. The feeding of the new design is more complicated
than in the previous case, but in return the antenna can reach more bands. Two
metallic vias cross the structure, from the end of the distribution network to the
feed points in both PIFAs, in order to feed the antenna. As mentioned in Fig. 2.10,
the PIFA placed on the right side of the image covers all GSM bands, while the
other covers LTE-2100, WiFi, and LTE-2600 bands.
(a) Transversal Cut (b) Top Plane
(c) Bottom Plane
Fig. 2.10: Transversal cut of the proposed multiband PIFA (a) and its top (b) and
bottom (c) planes.
Fig. 2.11: Inclination of PIFA’s patches.
Fig. 2.12: Monte Carlo simulation on the reflection coefficient of the proposed multi-
band antenna when varying ±2o the inclination of the plane of both PIFAs. The
purple thick line represents the ideal scenario (null inclination)
In Fig. 2.12, it is presented a Monte Carlo simulation of the reflection coefficient

in the proposed multiband antenna, where the inclination of the plane of both
PIFAs has set to move in a range of ±2o (look at Fig. 2.11). The purple thick line
represents the ideal scenario (null inclination). Note that there are four resonances
in Fig. 2.12: the first of them caused by PIFA #1; the second one (1450 MHz)
caused by a parasitic resonance either of the ground plane or the metallic vias or
the distribution network; the third one (1830 MHz) caused by an appropriate cut
in PIFA #1; and the fourth one (2100-2700 MHz) caused by a resonance of PIFA
#2. As Fig. 2.12 proves, a slight manufacturing deviation on the inclination of
both PIFAs (≤ 2o ) may mismatch the element in the bands of interest, leaving the
antenna inoperative.
Fig. 2.13 shows the far field radiation pattern of the proposed antenna at different
frequencies. As expected, the multiband PIFA presents a non-directive behavior.
Its total efficiency varies along the frequency according to its radiation efficiency
and the antenna matching, being on average in a 76 % (the maximum is ubicated
at 930 MHz, with a 95 %, and the minimum at 1830 MHz, with a 55 %).
In summarize, we have designed and simulated a multiband PIFA and we have

corroborated that is preferable working with wideband antennas: they are less sen-
sible to slight manufacturing deviations and they are interoperable among countries,
where frequency assignment plans can vary.
Fig. 2.13: Farfield radiation pattern at different frequencies of PIFA antenna.

2.3 Miniaturization
In order to increase the RF harvested power per unit area (µW/cm2 ), it is essential
to try to reduce the physical dimensions of the antenna. Several miniaturization
techniques have been studied so far, most of them based on structural modifications
and lumped element loadings [20, 24, 25]. Recent studies have tried to reduce the
size of spiral antennas by using fractal geometries, as in [26], where a modified Koch
curve is used to shrink a Fibonacci spiral antenna (FSA). We propose increasing
the bandwidth of the Archimedean spiral by extending the arms to the end of the
antenna (maintaining the same PCB area). Consequently, the electrical size of the
Archimedes antenna is higher and therefore, it reaches lower frequencies without
significant performance degradation. Fig. 2.14 depicts different designs following
that approach.
As the simulation results probe (Fig. 2.15), it is preferable to place an impedance

step (Fig. 2.14c) when extending the arms of the antenna, so that reflections are
softened and more bandwidth is achieved (fL1−step = 350 MHz vs fL0−step = 400
MHz vs fLnomin = 510 MHz). In these terms, Fig. 2 (purple line) shows that the
area of the miniaturized antenna can be reduced 5.5 x 5.5 cm2 to meet the lower
cutoff frequency (510 MHz) of the non-miniaturized antenna, that is, the area could
be reduced a 7.38 % without appreciable performance degradation of the antenna.
In Fig. 2.16, the radiation efficiency and the total efficiency of the miniaturized
Archimedes antenna are plotted. The antenna shows a good efficiency behavior,
with values over the 86% at all the frequencies of interest. It also was included in
Fig. 2.16 a comparison of the farfield radiation pattern at two different frequencies
of interest: 830 MHz (LTE) and 2450 MHz (WiFi). As mentioned before, notice
that the 3D radiation pattern tends to be slightly more directive as the frequency
increases. Besides that, Fig. 2.17 presents the axial ratio of the antenna, at the
different frequencies of interest, in two cuts in the planes φ = 0o and φ = 90o . As
it can be seen, the spiral is circularly polarized in a wide range of elevation angles
(-65o to 65o ) at most frequencies, but 500 MHz, where it could not go under 3 dB
and remains elliptically polarized.
(a) No miniaturization (b) 0-step minituarization
(c) 1-step miniaturization
Fig. 2.14: Different designs of the Archimedean spiral antenna: (a) no miniaturiza-
tion [14], (b) miniaturization without impedance step, and (c) miniaturization with
an impedance step.
Fig. 2.15: Simulated reflection coefficient of the Archimedean spiral antenna (nor-
malized to 188.5 Ω) with different designs: (a) no miniaturization [14], (b) minia-
turization without impedance step, and (c) miniaturization with an impedance step
(20.72 x 19.77 cm vs 15.22 x 14.27 cm).
Fig. 2.16: Simulated efficiencies (radiation efficiency and total efficiency) of the
miniaturized (one impedance step) Archimedean spiral antenna, and two examples
of the spiral’s 3D radiation pattern at two different frequencies of interest.
Fig. 2.17: Simulated axial ratio, at different frequencies, of the Archimedean spiral
antenna in two cuts in the planes φ = 0o (continuous line) and φ = 90o (dashed
line).
The farfield radiation pattern, in two cuts in the vertical and horizontal planes,
is shown in Fig. 2.18. As depicted before, the antenna presents its maximum of
radiation on the broadside direction, and since there is no ground plane, it also
radiates symmetrically backwards. As the frequency increases, the 90o radiation
null becomes more accentuated (Fig. 2.18a) and therefore the directivity of the
antenna increases.
(a) Vertical Plane (b) Horizontal Plane
Fig. 2.18: Simulated farfield patterns of the miniaturized Archimedean spiral an-
tenna at the different frequencies of interest on the (a) vertical and (b) the horizontal
plane.
2.4 Microstrip to Parallel Strip Balun. Ultraw-

ideband Design
The two arms of the Archimedean spiral turn the antenna into a balanced
structure and make feeding slightly difficult. This is not a problem in the case
of placing the plus and the minus poles of the rectifier circuit directly attached
to each arm of the antenna, because the balanced structure is preserved, but
when the performance of the antenna is measured independently via an SMA
(Subminiature version A) connector, the situation is quite different. The unbalance
in the connection causes the currents in both arms to not be the same, that is, to
not be symmetric. Subsequently, the radiation pattern changes and the antenna is
no longer omnidirectional. Besides that, the input impedance of the Archimedean
spiral is not referred to 50 Ω, but 188 Ω, as depicted in equation (2.1).
A balun is an electrical device capable of transforming a signal from an unbal-

anced structure (such as a coaxial cable) to a balanced structure (such as a copper
pair). There are many widely studied configurations [18], but most of them are
narrow band solutions. However, different authors have implemented [27, 28] an
ultrawideband microstrip to parallel strip balun oriented to use in spiral antennas,
so that the impedance transformation (from 50 Ω to 188 Ω) is already contemplated.
It is based on a tapered structure, which greatly increases the bandwidth. A typical
design criterion is to set the length of the structure as λo /4 at the lowest operation
frequency, thus, relative bandwidths of 200% are easily achievable. It is also inter-
esting to use high-permittivity substrates in order to reduce the size of the balun,
which can be significant whether we set the lower cutoff frequency at 300 MHz. So,
we are working with Rogers 3006 substrate (εr = 6.15, tan δ = 0.0020 @10 GHz).
In short, the challenge is threefold as commented above: transforming an unbal-
anced structure into a balanced one, and transforming the antenna impedance (188
Ω) into 50 Ω (SMA connector) within a ultra-wide range of frequencies (0.3-18 GHz).
The design of the ultrawideban microstrip to parallel strip balun is shown in Fig.
2.19. The balun is formed by two metallic tapered strips placed in the top and
bottom planes of a Rogers 3006 substrate. The wider section of the bottom strip
acts as a ground plane for the top strip, so that the typical quasi-TEM mode of a
microstrip line is propagated in the structure. The design equation for the metallic
top strip is
wt (x) = wtini eat x , (2.5)
and for the bottom strip is
wb (x) = wbini eab x , (2.6)

1 wp
where wtini is the initial width of the top strip, at = L
ln is the decay factor
wtini

of the top strip, wbini is the initial width of the bottom strip, and ab = L1 ln wwb p
ini
is the decay factor of the bottom strip.
Different simulation and optimization processes set the values mentioned before
in: L = 200 mm, wtini = 1.8 mm, wbini = 12 mm, and wp = 0.15 mm. The initial
width of the top strip wtini = 1.8 mm was calculated from the well-known microstrip
equations [4, 18] in order to obtain 50 Ω in the input port for the chosen dielectric,
while the initial width of the bottom strip wbini = 12 mm was obtained from a
parametric sweep in CST Microwave Studio. The length of the balun L = 200 mm
is about a quarter of wavelength at the lower operating frequency, 300 MHz.
In Fig. 2.20, we can see the two modes present in the port relative to the parallel
Fig. 2.19: Ultrawideband microstrip to parallel strip balun.

lines. The first one has a quasi-TEM structure, with the field lines emanating
from the bottom strip and ending up on the top strip, ensuring a 180 degree phase
shift within the two strips. On the other hand, the second mode is a non-desired
evanescent hybrid mode that must suppressed. On the other hand, Fig. 2.21 shows
the S-parameters of the balun. It works fine from 0.3 GHz to more than 18 GHz,
with transmission losses of less than 1.9 dB over the entire band (in the frequencies
of interest, up to 5.2 GHz, less than 0.8 dB). Note also that the hybrid mode is
sufficiently attenuated, more than 20 dB, so that it does not degrade the behavior
of the balun.
(a) QTEM Mode
(b) Hybrid Mode
Fig. 2.20: Electric field in the parallel strips’ port, and the QTEM (a) and the
evanescent hybrid (b) modes.
Fig. 2.21: Simulated S-parameters of the microstrip to parallel strips balun. Note
how the hybrid mode is quite attenuated with respect to the QTEM mode.
2.4.1 Back-to-Back
The model presented in Fig. 2.19 is now ready to be manufactured and to be
used in the antenna, but first, it would be desirable to measure its response in
order to prevent possible deviations in the final prototype. However, note that the
balun is not measurable in our laboratory since it is a parallel strip configuration
referred to an impedance of 188 Ω. In situations of this kind where measurement
is not possible, as for example in the case of microstrip-to-slotline transitions (we
cannot easily measure the slotline), the typical approach is to return to the original
(measurable) line, in a scheme known as back-to-back. Figure 2.22 depicts the idea
applied to our balun. Once the parallel strips are reached, we are going back to the
original microstrip line by mirroring the tapered structure.
The losses in the back-to-back configuration are approximately double those of

the balun. This is reflected in the transmission parameter (green line) of Fig. 2.23,
which indicates that losses are inferior to 3.8 dB in the whole band.
Fig. 2.22: Back-to-back configuration of the UWB microstrip to parallel strip balun.
Fig. 2.23: Simulated S-parameters of the microstrip to parallel strips back-to-back

configuration.
2.4.2 Balun-Antenna Connection

Once we have designed the antenna, the balun, and its back-to-back configuration,
the last simulation step is to unite the entire RF chain. It is not worth trying to
simulate the joint structure, since it implies long simulations times due to a rather
strong meshing (278 million cells), the long extension of the balun (the signals take
time to propagate), and a large bandwidth (0.3 - 18 GHz). Besides that, welding is
very poorly modeled in CST. Based on the fact that the balun minimally modifies
the radiation pattern of the Archimedean spiral antenna, which was corroborated in
simulation, the smarter way to simulate the joint structure is to model the antenna
and the balun as two ”black boxes” (via its S-parameters) and to use CST Design
Studio to study the composed response. This is reported in Fig. 2.24.
As can be seen in Fig. 2.25, the antenna is well-matched from 0.38 GHz to
18 GHz, showing the good performance of the balun. Note now that the joint
structure (antenna + balun of Fig. 2.24a) is fed via a 50-Ω SMA connector, the
unbalance caused by attaching the connector directly to the arms of the antenna
being eliminated by maintaining the ultrawideband response of the antenna.
(a) ”Physical” connec-(b) Use of CST Design Studio to make the

tion. connection.
Fig. 2.24: Balun-antenna connection.

Fig. 2.25: Reflection coefficient of the joint structure (antenna + balun).
2.5 Measurements
Figure 2.26 presents a prototype of the miniaturized Archimedean spiral antenna.
The antenna of the picture was constructed making use of a LPKF ProtoMat S100
milling machine [16] on a low-cost FR4 substrate (εr = 4.7, tan δ = 0.014 @1 MHz).
As previously mentioned, the dimensions of the antenna are 19.77 cm x 20.72 cm.
Fig. 2.26: Prototype of the miniaturized Archimedean spiral antenna.
2.5.1 Reflection Coefficient Measurement

We characterize in the laboratory the performance of the Archimedean spiral
antenna in two measurement steps. In the first of them, we extract the reflection
coefficient of the antenna from an Agilent 8722ES network analyzer, in order to
check its bandwidth. Subsequently, the power spectrum is acquired with a N9020A
MXA signal analyzer in two different scenarios: inside and outside the laboratory.
In Fig. 2.27, it is shown a comparison between the simulated and measured

reflection coefficients of the Archimedean spiral. As the experimental results probe
(red line in Fig. 2.27), the antenna is well-matched (< −10 dB) from 350 MHz
to 16 GHz, presenting an ultrawideband behavior. Besides that, its return loss is
superior to 8.5 dB in the entire frequency range shown (0.3 – 20 GHz). On the
other hand, although the measured reflection coefficient is slightly higher than the
Fig. 2.27: Comparison of the simulated and measured reflection coefficients of the
Archimedean spiral antenna (normalized to 188.5 Ω).
simulated one, note how the lower and upper cutoff frequencies coincide almost
perfectly, being their values 350 MHz and 16 GHz, respectively.
2.5.2 Power Spectrum Measurement

In order to obtain a realistic measurement of the harvested power, the antenna
is connected to a signal analyzer capable of integrating the power spectrum over
a prefixed bandwidth. Fig. 2.28 shows the power spectrum acquired by the
miniaturized Archimedean spiral antenna in two different scenarios: inside and
outside the laboratory (our laboratory is located inside the Higher Technical School
of Telecommunication Engineers, Technical University of Madrid, and the nearest
base station (LTE-800, GSM-900, GSM-1800) is situated at a distance of 60 meters
and at a height of 8 meters from the receiving Archimedean antenna). As it can
be noticed in Fig. 2.28, there is no significant contribution of power in any of
them from 5200 MHz onwards, where the -65 dBm noise level prevails. From
this frequency value, only militar and satcom applications can be found, but the
directivity of the antenna is so reduced that it does not allow us to acquire anything
remarkable. Despite capturing noise in a large frequency range (5.2 GHz – 16 GHz),
its contribution to the total power harvested is negligible, hence it is not worthy to
try to raise the upper cutoff frequency of the antenna (although, in this particu-
lar case, it is something intrinsic to its inner radius and it does not hinder the design).
As seen in Fig. 2.28, the measured power level of the spectrum peaks is generally
higher outside the laboratory than inside the laboratory, due to the radio signal
attenuation caused by the building itself (position of the room, walls, windows, etc.
[18]). Most of the energy captured comes from cellular bands, reaching almost -10
dBm for the 830 and 930 MHz bands. However, the measured FM spectral peak is
higher than expected (-25 dBm outside the laboratory), even though the antenna is
not specially designed to cover FM band. On the other hand, power contribution
Fig. 2.28: Power spectrum measured by the Archimedean spiral antenna inside and
outside the laboratory with the most relevant bands remarked: 1-FM, 2-DTT, 3-
LTE-800, 4-GSM-900, 5-GSM-1800, 6-LTE-2100, 7-WiFi, 8-LTE-2600, 9-WiFi.
of WiFi bands is surprisingly smaller than expected, leaving them in the background.
Furthermore, it could be of interest comparing the power harvested by the

Archimedean spiral antenna with other type of baseline antennas, such as dipoles
or microstrip patches. In these terms, three different patches that operate within
the 2.4 GHz band were designed and manufactured, as shown in Fig. 2.29. One of
them (the one in the center of the image that is matched with a λ/4 transformer)
is circularly polarized due to the slot inserted in the center of the patch, while the
other two are linearly polarized.
Fig. 2.30 shows a comparison between the power spectrum measured by the
Archimedean spiral antenna (inside the laboratory) and by the three 2.4-GHz
patches. The limited bandwidth of the microstrip patches, even within the WiFi
band, naturally conditions the measured spectrum peaks to be smaller than in the
case of using the Archimedean spiral antenna.
The total harvested power, which is the available power between the arms of
the Archimedean spiral antenna, is calculated as the sum of all carriers present
in the radio spectrum. The result of integrating the power spectrum shown in
Figs. 2.28 and 2.30 leads to the values presented in Table 2.1. We should remark
that all measurements have been carried out in a realistic scenario, that is, where
no directional sources have been intentionally put into scene. As previously
commented, the power acquired outside the laboratory (1.86 dBm) is higher than
the one acquired inside the laboratory (-3.19 dBm), both of which are sufficient to
provide the power required (1–30 µW) by the elements discussed in Chapter 1.
Fig. 2.29: Measured reflection coefficient of the manufactured 2.4-GHz patches (Di-
mensions: small patch: 4.89 x 3.91 cm, circularly-polarized patch: 5.05 x 3.91 cm,
big patch: 5.02 x 4.07 cm.
Fig. 2.30: Comparison between the power spectrum measured by the Archimedean
spiral antenna inside the laboratory and by the three 2.4-GHz patches, with the most
relevant bands remarked: 1-FM, 2-DTT, 3-LTE-800, 4-GSM-900, 5-GSM-1800, 6-
LTE-2100, 7-WiFi, 8-LTE-2600, 9-WiFi.
The best way to compare the amount of harvested power among antennas with
different sizes is to express the acquired power per area unit, that is, the physical
dimensions of the Archimedes spiral are 19.77 cm x 20.72 cm, which leads to 3.75
µW/cm2 and 1.17µW/cm2 outside and inside the laboratory, respectively. In Table
2.1, it is also presented the total power harvested by the three microstrip patches
and an ultrawideband horn used in [30]. Note that despite the fact that the patch
area is much smaller than the spiral area, the harvested power per area unit inside
the laboratory is about ten times lower than in the case of the Archimedes spiral.
On the other hand, it is also mentioned in the table that some measurements in [30]
show that a power level of -10 dBm was acquired by a broadband horn, the nearest
cell station being situated in this case at 150 meters.
Table 2.1: Measurements on the harvested power.
INSIDE OUSIDE
THE THE
LAB. LAB.
Power Power/Area Power Power/Area
(dBm) (µW/cm2 ) (dBm) (µW/cm2 )
Archimedean Spiral
-3.19 1.17 1.86 3.75
(0.3 – 16 GHz)
Big Patch
-26.01 0.12 - -
(2.4 GHz)
Small Patch
-28.68 0.071 - -
(2.4 GHz)
Circularly- Polarized
-26.69 0.11 - -
Patch (2.4 GHz)
Horn (0.8-18 GHz) -10
- - -
[30] (max)
Chapter 3
Conditioning Circuit.
Cockcroft-Walton Multiplier
3.1 Design of the Half-Wave Cockcroft-Walton

Multiplier
A single stage of the half-wave Cockcroft-Walton (HWCW) multiplier, which is
also called a Greinacher voltage doubler, is basically formed by two subcircuits [32]:
a Villard circuit (circled in orange in Fig. 3.1), which is a diode clamping circuit,
that is, a circuit capable of shifting the DC value of the input waveform; and a
half-wave rectifier (circled in red in Fig. 3.1) that retains the peak value of the
shifted signal. Essentially, the first capacitor charges to the peak AC voltage (Vpk )
on the negative half cycles, so that the output of the Villard circuit is the sum of
the input waveform and the steady DC voltage of the capacitor in question, being
ideally elevated to twice the voltage peak value (2Vpk ) on the positive half cycles,
and ideally clamped to zero on its negative half cycles. Subsequently, the half-wave
rectifier mitigates the enormous voltage ripple of the Villard circuit (2Vpk ) and
smooths the clamped AC signal.
Fig. 3.1: Circuital scheme of the n-stage half-wave Cockcroft-Walton voltage multi-
plier: the Villard circuit and the half-wave rectifier are circled in orange and in red,
respectively, and the Greinacher voltage doubler is remarked in dashed blue line.
42
CHAPTER 3. CONDITIONING CIRCUIT. COCKCROFT-WALTON MULTIPLIER43
Cascading several stages allows us to multiply the output DC voltage value Vo in

a proportional way to the number of stages n placed, that is
Vo = 2nVpk − 2δV − ∆V, (3.1)
where 2δV is the output voltage ripple and 2∆V is a term that quantifies the voltage
drop due to the incomplete charge of all capacitors when a load is connected (ideal
components are assumed). In a no-load situation, charge and discharge times of the
capacitors are nominally zero, so expression (3.1) can be simplified to Vo = 2nVpk .
As depicted in [33], in a n-stage HWCW multiplier where all capacitors are identical
(Fig. 3.1), the peak-to-peak voltage ripple is calculated as

q n(n + 1) Il n(n + 1)
2δV = = , (3.2)
C 2 fC 2
where q is the storage charge of all capacitors, f is the frequency of the input
waveform, and Il is the load current. On the other hand, the voltage drop ∆V is
calculated taking into account the amount of charge q that loses every stage at every
cycle, that is, ∆V1 = Cq n, ∆V2 = Cq (2n + (n − 1)),..., ∆Vn = Cq (2n + 2(n − 1) + ... +
2 · 2 + 1). The sum of all contributions leads to

q 2 3 1 2 1 Il 2 3 1 2 1
∆V = n + n − n = n + n − n . (3.3)
C 3 2 6 fC 3 2 6
Note that the first capacitors are most responsible of the voltage drop ∆V , as in
the case of the voltage ripple 2δV , according to the progression ∆V1 , ∆V2 ,...,∆Vn
mentioned above. Therefore, whether we increment their capacitances, both terms
reduce and the output voltage Vo rises according to equation (3.1). However, as well
mentioned in [32], it is only convenient doubling the first capacitor, since this one
just has to face half of the voltage than the rest of capacitors. In that situation, ∆V1
is decreased by 0.5 Cq n, which reduces the voltage drop at every stage proportionally,
leading to
Il 2 3 1
∆VC1 =2C = n − n . (3.4)
fC 3 6
In a similar manner, it can be demonstrated that the voltage ripple is now of the
form
Il n2
2δVC1 =2C = . (3.5)
fC 2
We have not considered yet the losses that the components present. In a first
approximation, we can neglect the capacitor losses and focus only on the forward
voltage drop of the diode, VF , which is very harmful in this particular case due to
the low input voltages we are dealing with. As there are two diodes in every stage,
there is a voltage drop of 2nVF due to the non-ideality of the diodes, being the
output voltage

Il 2 3 1 2 1
VoC1 =2C = 2n (Vpk − VF ) − n + n − n . (3.6)
fC 3 2 6
3.2 Measurements
The half-wave Cockcroft-Walton multiplier is a circuit commonly used in many
areas of electronics, in particular in those areas related to the generation of high DC
voltages (cathode ray tube television, particle accelerators, etc.), but when applied
to energy harvesting in ultra-low power applications, there are two considerations
that should be made. On the one hand, the input voltage peak value may be
situated under 0.2 V in most cases, so it is crucial that the forward voltage drop in
the diode is very low. On the other hand, the circuit operating frequency is about
units of GHz, which requires a diode fast enough to follow the input signal acquired
by the antenna. Two diodes that fulfill these two requirements are the HSMS-2822
(series mode), specially designed for input power levels above -20 dBm at frequencies
below 4 GHz; and the HSMS-2850 (single mode), optimized for use in small signal
(Pin < −20 dBm) applications at frequencies below 1.5 GHz. According to their
datasheets [34,35], the two series diodes encapsulated in HSMS-2822 can provide 0.1
mA with a maximum voltage drop of 0.22 V (@ 25o C), while the HSMS-2850 diode
can provide the same current with a maximum voltage drop of 0.15 V (@ 25o C).
(a) Top plane
(b) Bottom plane
Fig. 3.2: HWCW test set ((a) upper side, (b) bottom side) with six different combi-
nations of components and number of stages: 33 nF – HSMS 2822 – 3 stages (blue),
33 nF// 33 pF – HSMS 2822 – 2 stages (purple), 33 pF – HSMS 2850 – 2 stages
(green), 33 pF – HSMS 2822 – 2 stages (orange), 33 pF – HSMS 2850 (red) – 2
stages, 33 pF – HSMS 2822 – 5 stages (cyan).
Fig. 3.3: Measurement set-up: Agilent E4438C vector signal generator, Yokogawa
DL9240L digital oscilloscope, HWCW test set, and a multimeter.
In order to estimate the capacitor value to be used, a comparison between six

different combinations of components and number of stages of the HWCW is carried
out on the test board shown in Fig. 3.2. As well, Fig. 3.3 presents the measurement
set-up used to characterize the response of each circuit. It is formed by Agilent
E4438C vector signal generator [36], which operates from 250 kHz to 6 GHz; by
Yokogawa DL9240L oscilloscope [37], a 1.5 GHz, 4-channel digital oscilloscope; and
by a multimeter. Despite the quality of the oscilloscope, the probes available in the
laboratory can only measure up to 500 MHz.
Figure 3.4 presents the measured output DC voltage for different combinations of
components in a 2-stage HWCW multiplier (no load connected). The input power
of the sinusoidal waveform was set on -10 dBm, which generates an input voltage
peak of 210 mV (measured with the digital oscilloscope). In a 2-stage ideal scheme,
the output voltage should be 2 · 2 · 210 mV = 840 mV, but note that the output
voltage decays to 500 mV in most schemes due to the voltage drop VF in the diodes.
It is also observed a decay of the output voltage along the frequency, product of
the parasitic elements associated to the capacitors and to the parasitics of the board.
As can be noticed, the scheme that offers worst results is the one that uses
the 33 nF capacitor (orange line), since it has its self-resonant frequency much
lower (30 MHz) than the 33 pF capacitor (2.1 GHz). To palliate this effect and
increase its range of operation, we added a parallel 33 pF capacitor (purple line),
so that the equivalent capacitance is the same (33 nF + 33 pF ≈ 33 nF) but the
self-resonant frequency of the equivalent capacitor is increased. In any case, there
is no significant improvement. Using the same package (HSMS 2822), we obtain
much better results placing the 33 pF capacitor (blue line), but in return, when
a load is connected it shows a higher voltage ripple (20 mV @ 100 MHz). The
circuit that offers better results is the one that uses the HSMS 2850 diode and
Fig. 3.4: Output DC voltages in a 2-stage HWCW scheme when using different
components (no-load situation).
the 33 pF capacitor (green line), even though the datasheet [35] indicates that
the upper operating frequency is 1.5 GHz and that the recommended maximum
input power is -20 dBm. However, when a load was connected to the circuit, the
behavior of the diodes was degraded, so after removing the load and measuring
the output voltage in a no-load situation, the output voltage was inferior to the
measured before. Therefore, we do not recommend its use with an input power of
-10 dBm. On the other hand, Fig. 3.5 shows a comparison of the output voltage
versus frequency as a function of the input power (-10, -20 and – 30 dBm) on
the 2-stage HWCW scheme, where HSMS 2822 diode and 33 pF capacitors are
used. As expected, the lower the input power, the higher the percentage of power,
in proportion, consumed by the diodes. This leads to a reduction of the output
voltage as represented in the red line of Fig. 3.5, where for an input power of
-30 dBm (1 µW) the circuit is not capable of raising the voltage properly (< 35 mV).
Finally, Fig. 3.6 presents the output voltage, at each stage, of the 5-stage HWCW
multiplier. It follows from Fig. 3.5 and Fig. 3.6 that it is not worthwhile placing
several stages when the input power is so reduced. The intrinsic consumption of
the diodes causes that the output power decreases non-linearly as a function of the
input power. In summary, once the power transfer (from the antenna to the multi-
plier circuit) is optimized via the matching circuit, we must choose the appropriate
number of stages that fulfills the output voltage requirement of the sensor (load).
Nevertheless, note that the more stages are set, the higher the output voltage but
the lower the output power. So, there is a trade-off between elevating the output
voltage and maximizing the output power, what we are mainly interested in.
Fig. 3.5: Output voltage versus frequency in the 2-stage HWCW scheme (with C=33
pF) at different input powers.
Fig. 3.6: Output voltage at each state of the 5-stage HWCW scheme (with C=33
pF) for an input power of -10 dBm.
Chapter 4
Matching Circuit
The analysis of the problem, from the point of view of the matching circuit, is
depicted in Fig. 4.1. The source Zs and load ZL impedances of the two different
matching circuit that maximize the power transfer from the RF harvester to the
multiplier circuit initially depends on three parameters: the impedance of the
antenna Zant , the number of stages N of the HWCW multiplier circuit, and the
impedance of the load ZL placed. The non-linearity of the Cockcroft-Walton
multiplier adds two other terms: the input power Pin , and the operating frequency
f ; which greatly increase the complexity of the problem as seen in Fig. 4.1. Note
that the optimum source impedance in the matching circuit is generally different
from the antenna impedance.
For the particular case of the Archimedean spiral antenna (Rant = 188 Ω), the
measurements on the harvested power from the RF spectrum allow us to eliminate
the power and frequency variables. Hence, the complexity of the problem is sub-
stantially reduced, as the source impedance of the matching circuit that ensures
maximum power transfer is now only dependent on two terms: the number of stages
N of the Cockcroft-Walton multiplier, and the selected load (the device to be fed).
Fig. 4.1: General scheme of the RF harvesting system. Note as the source impedance
Zs depends on five terms: the input power Pin , the input range of frequencies f , the
number of stages N in the Cockcroft-Walton multiplier , the intrinsic impendance
of the antenna Zant , and the load ZL .
48
CHAPTER 4. MATCHING CIRCUIT 49
Fig. 4.2: Scheme of the RF harvesting system particularized to the use of the
Archimedean spiral antenna. Notes: *Pin1 = −5.80 dBm and Pin1 = −1.84
dBm, inside and outside the laboratory, respectively; and Pin2 = −9.20 dBm and
Pin1 = −2.34 dBm, inside and outside the laboratory, respectively. **fin1 = 807
MHz, and fin2 = 942 MHz.
Fortunately, the problem can be also divided into N sub-problems, which are only
dependent on the load impedance, considering separately the different number of
stages of the multiplier circuit.
4.1 Narrow-band modeling of the Archimedean

spiral antenna
We already know the impedance of the Archimedean spiral antenna is theoreti-
cally set to be 188 Ω, so the radiofrequency harvester can be modeled by a generator
with and source impedance of Zant = 188 Ω. The generator, in turn, must contain
the information relative to the power spectrum acquired by the antenna. It is not
easy to model the spectrum power, due to the different carrier waveforms present
in the spectrum and the multiple frequencies that are being taken into account
(ultrawideband system). However, we do known that most part of the incoming
power comes from 800/900-MHz bands, as can be noticed in Fig. 2.28.
As Fig. 4.3 shows, both 800 and 900 MHz carriers are far for resembling to
delta functions (sinusoidal excitations). However, modeling the carriers as two delta
functions, placed at 807 MHz and 942 MHz, greatly simplifies the problem. The
power peaks of both delta functions are calculated by integrating the power of both
carriers along their range of frequencies, that is, integrating the power spectrum
from 770 to 820 MHz and from 930 to 954 MHz, respectively. Fig. 4.3 depicts the
process more clearly.
Fig. 4.3: Zoom over the power spectrum measured by the Archimedean spiral an-
tenna at 800 and 900 MHz. The delta functions marked in blue and orange represent
the modeling of the Archimedean spiral antenna as a generator, inside and outside
the laboratory, respectively.
In short, the Archimedean spiral antenna can be modeled from a circuital view-
point as described in Fig. 4.4, the four parameters of interest being Pin1 , f1 = 807
MHz, Pin2 , and f2 = 942 MHz. Note that two different scenarios have been taken
into account: the antenna being placed inside, and outside the laboratory.

 −5.80 dBm if IN SIDE
Pin1 = (4.1)
−1.84 dBm if OU T SIDE


 −9.20 dBm if IN SIDE
Pin2 = (4.2)
−2.34 dBm if OU T SIDE

Fig. 4.4: Narrow-band circuital model of the antenna.

Besides that, it is not smart try simulating the matching circuit (typically
L-networks or π-networks), the conditioning circuit and the load using time-domain
analysis, as for example PSPICE does. The great amount of differential equations
involved in the combined circuit (caused by the presence of several capacitors),
joined to the fact of the non-linearity of the multiple diodes used, leads to large
simulation times (transients included). To solve this situation we make use of the
harmonic balance. It is a frequency-domain method typically used to calculate the
steady-state response of non-linear circuits. It basically decomposes the circuit in
two different parts: a linear one, which includes all linear components (capacitors,
inductors, resistances, etc.), and a non-linear one (diodes, transistors, etc.); and
studies its response in a prefixed number of harmonics (DC, fo , 2fo , etc.). Check
the annex for more information. Many simulators have already implemented this
method, as for example ADS (Advanced Design System).
Generally, periodic excitations (tones, squared signals, etc.) at a single frequency

are used in harmonic balance, although they can be replaced by more complex
excitations at the expense of introducing a more complicated analysis. Multitone
excitations are also allowed. They are of special interest in our modeling, where
multiple frequencies are contemplated (wideband response), but note that placing
several sources implies increasing substantially the number of harmonics considered,
and therefore the simulation time. Hence, there is a trade-off between simplifying
the problem (placing few sinusoidal inputs) and losing the wideband information,
although initially the multiplier circuit was not matched in any case out the fre-
quencies of interest. For this reason, we are pretending to only consider the most
relevant spectral peaks, which are placed at 800 and 900 MHz.
4.2 Equivalent Circuit Model. Test Circuits

As depicted before, since we are dealing with non-lineal circuits and the input
impedances vary with the frequency, the input power and the state of the diodes
at each cycle (ON/OFF), there is no straightforward analytic manner to estimate
the values of the components in the matching circuit. The smartest process to
optimize the output power level consists in placing an equivalent impedance Zs
after the generator (without considering its internal resistance) and looking for the
maximum value of the output power Po by varying Zs and the load RL . Then, the
components of an L or π-matching network are determined in order to transform
the internal generator resistance, which can be Rant = 188 Ω, Rs = 50 Ω or another
vale, into the calculated optimum source impedance Zs .
We need a strong model of our circuit that ensures that all parasitic elements do
not falsify the results. So, it is desirable to manufacture some test circuits before
simulating the final prototypes, in order to extract information from them and refine
the models of more elaborated and complex circuits. Thus, the error committed will
be lower.
4.2.1 First Model. Test Circuit #1

ADS simulations show that the source impedance Zs that maximizes the output
DC power in test circuit #1 is Zs = 60 + j200 Ω. Note that the internal generator
resistance is Rs = 50 Ω, so it is not worthy transforming the real part of the
impedance from 50 Ω to 60 Ω in practice, because the values are very similar.
Therefore, the capacitor of the L-matching network can be eliminated. To obtain a
reactance of Xs = 200 Ω at 870 MHz, we place a 33 nH series inductor. Test circuit
#1 is depicted in Fig. 4.5a.
(a) Test circuit #1 (b) First Circuit Model
Fig. 4.5: Test circuit #1 (a) and its equivalent model (b). The source impedance
that maximizes the power delivered to the load is Zs = 60 + j200 Ω. Values of the
components: Lmat = 33 nH, C = 33 pF, RL = 6.1 kΩ. Values of the parasitics:
Cp = 0.77 pF, Cpd = 0.70 pF, Lp = 2.64 nH.
At the lower frequencies, the parasitics of the components and the board do not
affect the measurement, so the most basic simulation is quite similar to the mea-
surement, but that does not happen as the frequency increases (blue and red lines
in Fig. 4.6). Hence, an accurate modeling of all manufactured circuits is needed
in order to avoid frequency displacements in the measurement. In the first circuit
model shown in Fig. 4.5b, we are mainly taking into account the parasitic elements
of the components, leaving in the background the parasitics of the board. Thus,
Cp = 0.77 pF models the parasitic capacitance associated to the matching inductor
Lmat , Lp = 2.64 nH the parasitic inductance associated to the 33 pF capacitors, and
Cpd = 0.70 pF the parasitic capacitance associated to HSMS 2822 diode package.
Note that the parasitics of the matching inductor and the 33 pF capacitor were cal-
culated taking into account the self-resonant frequency (SRF) of both components,
led by r
1
SRF = . (4.3)
LC
For example, in the case of the 33 nH inductor, the manufacturer set its SRF on 1
GHz, which implies a parasitic capacitance of Cp = 0.77 pF according to (4.3).
The most important parasitics in the circuit model are those related to the RF coil
and the diode package. This can be quickly visualized in Fig. 4.6. The orange line
denotes a hypothetical equivalent circuit where only the parasitic capacitance of the
coil is considered (leaving out the parasitics of the diode and the 33 pF capacitors).
Fig. 4.6: Measurements relative to test circuit #1 (blue line) and the simulations of
the circuit without parasitic elements (red line) and the first circuit model (magenta
line). Cp = 0.77 pF, Cpd = 0.70 pF, Lp = 2.64 nH.
Note the importance of modeling well the RF inductor, since half of the frequency
displacement between simulations and measurements is caused by it. The other half
is completed by taking into account the self-resonant frequency of the capacitors used
and the parasitics of the diode package (not the diode itself), being most significant
the last one. The losses in the capacitors and in the coil, modeled by the ESR
(Equivalent Series Resistance) and DCR (Direct Current Resistance) series resistors,
respectively, were neglected due to their insignificant contribution. However, note
that the HSMS 2822 model in ADS has already internally implemented a series
resistance that models the losses in the diode.
4.2.2 Second Model. Test Circuit #2

In order to test the equivalent model introduced in the previous section, a new
scheme is proposed. We are again looking forward to finding the value of the source
impedance Zs that maximizes the power delivered to the load, but now on the
basis of the previous parasitic elements’ modeling (look at Fig. 4.7b). As it can
be seen in Fig. 4.8, the frequency displacement in the modeling (red line) is less
than the initial case where no parasitic elements were considered, but it is still being
significant. Note that we are pretending to work in the 800/900 MHz band, so a
non-negligible amount of power (a 6.84 %) is being dispatched due to the frequency
displacement in the measurement. Therefore, a finer readjustment of the model is
necessary.
(a) Test circuit #2 (b) First model applied on test circuit #2
(c) Second model applied on test circuit #2
Fig. 4.7: Test circuit #2 (a), first model applied on test circuit #2 (b), and the final
equivalent model (c) that takes into account the parasitic elements of the board. The
source impedance that maximizes the power delivered to the load is Zs = 50+j78 Ω.
Values of the components: Lmat1 = 4.7 nH, Lmat2 = 8.2 nH, C = 33 pF, RL = 2.34
kΩ. Values of the parasitics: Cp1 = 0.70 pF, Cp2 = 0.40 pF, Cpd1 = 0.70 pF,
Cpd2 = 0.10 pF, Lpd = 1.5 nH, Lp = 1.5 nH, Lvia = 0.5 nH.
Fig. 4.8: Measurements relative to test circuit #2 (blue line) and the simulations
of the circuit using the first circuit model (red line) and a redesigned second circuit
model (gold line).
We must take into account the parasitic elements of the board, as well as other
parasitics relative to the diode package. The most significant parasitic that was
neglected in the first model is the parasitic inductance related to the vias that are
part of the board. At frequencies near 1 GHz, every via can be approximately
modeled as a Lvia = 0.5 nH inductor. When we used a 33 nH coil, these parasitics
associated to the vias were neglected without performance degradation due to
Lvia << 33 nH, but now the values of the RF coils are of the order of the parasitics.
We also did not take into account the parasitic capacitance related to the two
pins furthest apart in the diode package, denoted as Cpd2 , nor the series parasitic
inductance Lpd in the diode package. From the considerations depicted, we form the
more complex circuit model shown in Fig. 4.7c. As it can be seen in the gold line
of Fig. 4.8, the second model approximates in an accurate manner the measured
frequency response of test circuit #2. Due to the consumption rate of all diodes,
the lower the input power, the lower the efficiency of the circuit (matching network
+ half-wave Cockcroft-Walton multiplier). In the case of applying an input power
of -3 dBm, the measured output voltage (at 870 MHz) is 0.495 V, which implies
2
an output power of Po = 0.495 2.34 k
= 0.105 mW (-9.79 dBm). Thus, the measured ef-
ficiency of the circuit in these condition is η (Pin1 = −3 dBm, f1 = 0.87 GHz) = 21 %.
In Fig. 4.9, it is presented the measurement set-up that implements the antenna
model of Fig. 4.4 within test circuit #2. It is basically formed by a combiner that
receives two tones at frequencies of 807 and 942 MHz (from two different signal
generators), and subsequently feeds the test circuit #2 with an output power levels
of -1.84 dBm and -2.34 dBm, respectively, the sum of both tones giving a total
power of 0.90 dBm. In these terms, the output voltage measured in the laboratory
2
was 0.997 V, which implies an output power of Po = 0.997 2.34 k
= 0.425 mW. Hence,
the efficiency of the circuit is now η (Pin1 = −1.84 dBm, f1 = 0.807 GHz, Pin2 =
−2.34 dBm, f2 = 0.924 GHz) = 34.6 %. Even though the optimum power peak is
shifted in frequency, note how the efficiency increases with the input power level.
Fig. 4.9: Measurement set-up of test circuit #2.

4.2.3 Third Model. Test Circuit #3

Circuit #3 comes up in order to finally recreate the real scenario of the energy
harvesting system, which includes the circuit model of the antenna, the matching
circuit, the multiplier circuit and a load that acts as our sensor. However, the
fact is that the internal generator resistance is not 188 Ω, but 50Ω, so a previous
transformation stage is needed to obtain the antenna impedance.
There are two L-matching networks in circuit #3 (Fig. 4.10). The first of them
is formed by Lmat1 = 6.8 nH and Cp1 = 1.2 pF, and is in charge of transforming
the internal generator resistance (50 Ω) into the impedance of the antenna (188
Ω). The second of them is formed by Lmat21 = 3.3 nH, Lmat22 = 6.8 nH, and
Cmat21 = Cmat22 = 1.5 pF: two series inductors that act as the equivalent inductor
Lmat2 = Lmat21 + Lmat22 , and two shunt capacitors that act as the equivalent
capacitor Cmat2 = Cmat21 + Cmat22 ; and it is in charge of transforming the antenna
impedance (188 Ω) into the optimal source impedance (Zs = 18 + j56 Ω). The lack
of components in the laboratory has led us to place in the second matching network
two more components than necessary, which increases the number of parasitics and
modifies the performance of the circuit. As a result of this and the fact that the
values of the elements of the matching network are a bit different from the used
in test circuit #2, two values of our previous model has been slightly readjusted
again: Cpd1 = 0.75 pF, Lvia = 0.55 nH.
(a) Circuit #3
(b) Third model applied on circuit #3
Fig. 4.10: Circuit #3 (a) and the third model applied on it (b). The source
impedance that maximizes the power delivered to the load is Zs = 18 + j56 Ω.
Values of the components: Lmat1 = Lmat22 = 6.8 nH, Lmat21 = 3.3 nH, Cmat1 = 1.2
pF, Cmat21 = Cmat22 = 1.5 pF, C = 33 pF, RL = 2.34 kΩ. Values of the parasitics:
Cp1 = Cp22 = 2.2 pF, Cp21 = 1.3 pF, Cpd1 = 0.75 pF, Cpd2 = 0.10 pF, Lpd = 1.5 nH,
Lpd1 = Lpd2 = 0.5 nH, Lp = 1.5 nH, Lvia = 0.55 nH.
Fig. 4.11: Measurements relative to test circuit #3 (blue line) and the simulations
of the circuit using the first circuit model (red line), the second circuit model (gold
line), and a redesigned third model (purple line).
Figure 4.11 shows the output voltage of circuit #3 measured in the laboratory and
the different approximations made in simulation with the three proposed circuit
models. The refinement in the process is clearly appreciable. The third model
adjusts almost perfectly the resonance of circuit #3, in frequency and in amplitude.
4.3 Nonlinearity Effects. Circuit #3

We have already mentioned that the non-linearity of the diodes had serious
implications for the performance of the circuits. In Fig. 4.12, it is shown the
normalized output power along the frequency when applying two different input
power in circuit #3. Firstly, note that the normalized power is generally higher in
the case of working with a higher input power level. This can be seen more clearly
in the ”flat” zone (up to 0.8 GHz) of the graph, where the blue line (0.94 dBm) is
18 % above the red line (-3 dBm). Secondly, the bandwidth in the resonant zone
that caused the L-matching network (0.8 - 1 GHz) is a bit higher in the case of
applying a higher power level at the input. This is somewhat shocking, but it is to
be expected taking into account the non-linearities of the circuit.
Furthermore, Fig. 4.13 directly shows the efficiency of circuit #3 (at 870 MHz,
the central frequency) as a function of the input power. It can be observed a drop
of 7-8 % of the efficiency when the input power is reduced by half (-3 dB) from
5 dBm downwards. Note also that the efficiency approximately saturates from 10
dBm onwards, and that even a drop in efficiency beyond 15 dBm can be seen since
the diodes are designed to operate in small signal region.
Fig. 4.12: Nonlinearity in circuit #3 caused by different input power levels.
Fig. 4.13: Measured efficiency in circuit #3 as a function of the input power.
4.4 Comparison Among Different Number of

Stages
Once we have created a reliable circuit model of all parasitic elements in the board,
the objective is to search in simulation the optimum source and load impedances
for the different number of stages, and then designing the corresponding matching
networks of every stage. Tables 4.1 and 4.2 show the optimum impedances, as well
as the maximum efficiency, as a function of the number of stages in the HWCW
multiplier in the two analyzed scenarios: the antenna being inside or outside the
laboratory. Overall, the efficiency of the circuit is higher in the case of placing the
antenna outside the laboratory than inside it. Moreover, it is observed that the
output voltage does not rise as we increase the number of stages. This is because
we are only worrying about maximizing the power at the output, without worrying
about the voltage level obtained as a result. On the other hand, we could increase
the output voltage at the expense of not getting maximum power at the output. As
commented before, there exists a kind of trade-off between elevating the voltage and
maximizing the output power. Another type of evaluation must be done in the case
of fixing a desired output voltage and trying to optimize the power for that prefixed
value. Anyway, Tables 4.1 and 4.2 give us an idea of the maximum reachable values
of efficiency, in a hypothetical situation where the matching network has an infinite
bandwidth.
Table 4.1: Simulated optimum source impedances Zs , load resistances RL , output

voltages Vo , output currents Io , output powers Po , and efficiencies for the different
number of stages of the multiplier circuit with the ”outside” modeling of the antenna.
OUTSIDE
(Pin1 = −1.84 dBm, Pin2 = −2.34 dBm)
(f1 = 807 MHz, f2 = 942 MHz)
# Stages Zs (Ω) RL (kΩ) Vo (V) Io (mA) Po (dBm) Efficiency(%)
1 11.5 + j54.6 6.0 1.96 0.327 -1.94 51.67
2 9.0 + j15.5 6.6 1.73 0.262 -3.43 36.66
3 8.0 7.6 1.40 0.184 -5.83 21.10
Table 4.2: Simulated optimum source impedances Zs , load resistances RL , output

voltages Vo , output currents Io , output powers Po , and efficiencies for the different
number of stages of the multiplier circuit with the ”inside” modeling of the antenna.
INSIDE
(Pin1 = −5.80 dBm, Pin2 = −9.20 dBm)
(f1 = 807 MHz, f2 = 942 MHz)
# Stages Zs (Ω) RL (kΩ) Vo (V) Io (mA) Po (dBm) Efficiency(%)
1 11.3 + j59.5 5.8 0.941 0.162 -8.16 39.86
2 5.4 + j20.0 9.0 0.90 0.100 -10.47 23.42
3 4.0 + j5.6 11.0 0.66 0.060 -13.97 10.46
There are two more things that are worth mentioning. The first one is that the
optimum impedances slightly vary in case of being in the outside or in the inside,
caused by the non-linearity of the circuit. The second one is that the reactance of
the optimum source impedance decreases as the number of stages elevates. Each
Fig. 4.14: Efficiency of the simulated circuit that includes the optimum source
impedances as a function of the load RL .
stage of the HWCW multiplier has a predominant capacitive behavior, so the

addition of more stages can be seen as placing several series capacitors, resulting in
an equivalent capacitor of lower capacity. This, together with the fact that each
stage adds an equivalent higher parasitic inductance, makes the reactance to be
canceled out smaller.
To see the effectiveness of the impedance matching network is interesting to check

which would be the output power in the case of not placing it, that is, in the case
of having Zs = Zant = 188 Ω. With a single stage in the HWCW multiplier, the
output power would be -23.47 dBm in the case of attaching the antenna directly to
the circuit on the outside. Hence, the efficiency would be 0.33 %, a very poor value
and much lower than the one seen in Table 4.1.
One of the biggest drawbacks in the circuit is that the sensor’s intrinsic resistance
cannot be transformed (we are in the DC region) into the optimum value presented
in the previous tables, since in general these values will not coincide. Therefore,
we must check how much the efficiency of the circuit is reduced by moving away
from the optimal value RL . Fig. 4.14 shows the depicted before. Note that the
efficiency is almost flat once the load resistance is higher than 4 kΩ, but there is a
considerable reduction of the efficiency for small values of RL .
Taking into account the circuit model of the parasitics and the optimum
impedances shown in Tables 4.1 and 4.2, we estimate the values of the capacitor
and the inductor that are part of the L-network (look at Annex B for more informa-
tion). As noted above, the values of these components depend on the power value
supplied at the circuit input. Therefore, they will differ slightly if the antenna is
Fig. 4.15: Efficiency of the simulated circuit that implements the final L-matching
networks as a function of the load RL .
located inside or outside the laboratory. However, we are interested in designing

a single network that covers both cases, for which it is interesting to look at the
intermediate powers Pin1 = −3.82 dBm and Pin1 = −5.75 dBm in order to design
it. Then, we return to the original cases (inside/outside the laboratory) in order
to check the efficiency of the circuit, which is shown in Fig. 4.15 as a function of
the load resistance RL together with an outline of the circuit. As expected, the
efficiency is lower than in the ideal case of Fig. 4.14 since the bandwidth of the
L section is limited. In this regard, it would be interesting to try more complex
techniques in order to improve the bandwidth of the matching network, and subse-
quently the efficiency of the circuit. It is also worth noting that the impedance step
in the matching process is very pronounced (from 188 Ω to Zs ≈ 8 + j15 Ω), which
drastically reduces the bandwidth of the antenna. To palliate this effect, one can
try to place
p an intermediate impedance step (geometric mean), that is, going from
188 Ω to 188 · (8 + j15) ≈ 48 + j29 Ω, and then to Zs ≈ 8 + j15 Ω.
Chapter 5
Conclusion & Future Work
5.1 Conclusion
Firstly, we saw in Chapter 2 how wideband antennas are a better choice when
implementing the radiofrequency harvester: they are more tolerant of small devi-
ations in the manufacturing process, and they are interoperable among countries,
where frequency assignment plans can be different. That is not the case for
multi-band antennas, due to they are strongly resonant and therefore narrowband.
It was also commented that circular polarization is preferred for this application,
as we do not know the polarization of the incident wave. An antenna that meets
these characteristics is the Archimedean spiral, so it emerges as one of the ideal
candidates for energy harvesting applications.
Furthermore, we experimentally confirmed that mobile assignment bands impose

the most important contribution of energy from the radio spectrum, especially from
the 800 and 900 MHz area, where the measured power values are more than eight
times higher than the next most relevant spectral peak (2100 MHz, LTE). Thus, it
was observed that 98 % of the total power captured by the antenna comes from the
mobile bands, of which an 82 % belongs either to 800 MHz or 900 MHz bands. The
remaining 2 % is a sum of the power coming from FM, DTT, WiFi and WiMAX
bands. Therefore, a smart manner to rise the harvested power per antenna unit
area is to design a new reduced prototype that takes advantage of the proposed
miniaturization technique and that operates from 0.8 GHz onwards.
According to the commented before, we have seen that is a reasonable approxi-

mation to focus the design of the matching circuit at 800 and 900 MHz. Otherwise,
it would be tricky to find the optimum source impedance that maximize the output
power through the harmonic balance method, which is pretty sensible to the wave-
form and number of excitations applied. Besides that, it would also be extremely
complicated to design such a wideband matching network. For these reasons,
we modeled the RF harvester at the circuit level as two series sinusoidal gener-
ators of input frequencies 807 MHz and 942 MHz, with a source impedance of 188 Ω.
62
CHAPTER 5. CONCLUSION & FUTURE WORK 63
Thereupon, we designed, manufactured and measured three test circuits, which

served as a reference for modeling the behavior of the parasitic elements in ADS
(with the harmonic balance method) and create a reliable circuit model of the
system. It was possible to prove experimentally and in simulation that the most
harmful element is the parasitic capacitor referred to the inductor, being able to
displace hundreds of MHz the resonance peak in case of not being well modeled.
In addition, we experimentally tested how the efficiency of the circuit drops as

the input power reduces (due to the non-linearity of the circuit) or the number
of stages increases, that is, the output DC voltage rises with a major number of
stages, but the output current drops due to the power consumption of the diodes.
Thus, it is convenient to use as few stages as possible, enough to ensure that the
load receives the voltage level it needs.
Finally, we applied the refined model of the circuit parasites to obtain the source
and load impedances that maximize the power transfer from the antenna to the
load, and subsequently we calculated the values of the components that form the
L-matching network. We saw as the efficiency of the system is limited to 30%, since
the input power is of the order of 1 mW and the diodes mainly limit its performance.
5.2 Future Work

To design and manufacture the storage circuit. We could initially try using a
C >> capacitor parallel to the load, so that the system load constant (up to 99 %)
would be tL ≈ 5τ = RL C, where RL is the value of the load resistance. Note that
the capacitor operates in DC mode, so it is an open circuit and does not affect the
RF signal.
To manufacture and measure the microstrip to parallel strip balun. It would be

interesting to first manufacture the back-to-back configuration to detect possible
deviations in the design and then redesign the balun, if necessary.
To study other higher-bandwidth adaptive networks. π or T sections offer higher

bandwidth, but typical design equations only include transformations between real
impedances. The inclusion of another element in the network to achieve transforma-
tion from real to complex impedance significantly reduces its bandwidth. It is also
not a good idea to make use of matching networks with lines and stubs, as the max-
imum operating frequency is 0.942 GHz and the lines would have a considerable size.
To extend the narrow-band model of the antenna (807/942 MHz) to a wideband

model that includes the carriers in the 2100, 2400 and 2600 MHz area, the
others with the greatest contribution. This is an extremely complex task, as the
non-linearity of the circuit implies that the optimum source impedance that the
multiplier must see is different for each frequency band.
CHAPTER 5. CONCLUSION & FUTURE WORK 64
To design a PCB that includes the matching circuit, the multiplier circuit, the
storage circuit and the load. In this way we will reduce the parasitics on the board
and optimize the space.
Finally, to join all stages and measure the performance of the full system.
Appendix A
Harmonic Balance Method
The harmonic balance (HB) analysis [38] is a frequency-domain method used to

calculate the steady-state response of non-linear circuits, such as mixers, power am-
plifiers or frequency multipliers. It consists of dividing the circuit into two different
subcircuits: one that includes the linear components, and another one that includes
the non-linear components; and finding through numerical methods the values of
the voltage and current at the different harmonics. The linear part is treated as
a multiport and it is typically described by its Y parameters, while the non-linear
elements are modeled by their I/V or Q/V characteristics. Therefore, the circuit is
divided in N + 2 ports as shown in Fig. A.1, that is, N ports of non-linear elements,
the input port (N + 1), and output port (N + 2). Note that the circumflex accent
indicates a non-linear term in Fig A.1.
Fig. A.1: Partitioning of a microwave circuit into linear and non-linear subcircuits.
The source Zs (ω) and load ZL (ω) impedances have been placed into the linear
circuit. The circumflex accent indicates a non-linear term.
65
APPENDIX A. HARMONIC BALANCE METHOD 66
Fig. A.2: A diode D connected to a lumped element with an sinusoidal excitation.
A.1 A Simple Example

The circuit shown in Fig. A.2 conceptualizes the idea of partitioning the circuit
in both linear and non-linear parts. The excitation Vs (t) consists of a DC (Vs (0))
and a sinusoidal component (Vs (ωp )). From the circuit, it is clear that the current
that flows to the linear part must be:
V (kωp ) − Vs (kωp )
ILIN (kωp ) = . (A.1)
Z(kωp )
On the other hand, the current in the diode can be modeled using the Shockley
equation, which is expressed as
IN L (t) = Isat eδV (t) − 1 .

(A.2)
Then, IN L (kωp ) is obtained by making use of the Fourier transform. Thereupon, we
must check if Kirchoff’s current law,
ILIN (kωp ) + IN L (kωp ) = 0, (A.3)

is satisfied, then we have a solution. Note however that non-linear problems can
have more than one solution, that is, it is not guaranteed that the solution obtained
is the one we are looking for. In short, the process can be summarized as follows [38]:
1. Create an initial estimation of V (kωp ), k = 0, 1, ...K, where the zero term

(k = 0) represents the DC component, and K is the maximum harmonic we
are taking into account.
2. Obtain ILIN (kωp ).
3. Apply the inverse-Fourier transform to V (kωp ) to obtain V (t).
4. Use the specific current expressions in the non-linear elements, IN L (t) =

δV (t)
Isat e − 1 in the case of the diode, to obtain IN L (t).
5. Apply the Fourier transform to IN L (t) to obtain IN L (kωp ).
6. Use ILIN (kωp ) + IN L (kωp ) = 0. Probably it will not be satisfied, so we define

an error function ∆k at each harmonic k:
APPENDIX A. HARMONIC BALANCE METHOD 67
ILIN (kωp ) + IN L (kωp ) = ∆k (A.4)
7. Modify V (kωp ) (using appropriate numerical methods to reduce ∆k ) and re-

peat the process from step 2.
8. Continue until the error ∆k at all harmonics is near to zero.

Appendix B
Design of L-Matching Networks
Impedance matching networks appear in order to transmit the maximum

available power from a transmission line to a load. Besides that, they should
ideally be lossless to avoid unnecessarily neglecting part of the power. With
these premises in mind, there are several ways to transform impedances [4],
but the vast majority of them are based on the use of transmission lines and
stubs, or on the use of concentrated elements. The use of transmission lines is
prohibitive in our particular case because of the low frequency involved in the
design (0.8 - 0.9 GHz) and therefore, the large dimensions of the lines. Thus,
we focus our study on the use of concentrated elements. The simplest match-
ing network of this kind is L section, which is formed by two reactive elements
and is capable of transforming a real impedance into a complex load ZL = RL +jXL .
As shown in Fig. B.1, there are two possible configurations in an L-matching

network. When RL > Zo (zL = ZZLo is inside the 1 + jx circle on the Smith chart),
the circuit of Fig. B.1a must be used, and the design equations are
q p
XL ± RZLo RL2 + XL2 − Zo RL
B= , (B.1)
RL2 + XL2
and
1 Zo XL Zo
X= 1− + . (B.2)
B RL RL
When RL < Zo , the circuit of Fig. B.1b must be used, and the design equations are
r
1 Zo − RL
B=± , (B.3)
Zo RL
and
X = BZo RL − XL . (B.4)
Note that there exist eight different possibilities of the L section depending on the
load placed, since the reactive elements may be either capacitors or inductors and
there are two different configurations.
68
APPENDIX B. DESIGN OF L-MATCHING NETWORKS 69
(a) RL > Zo
(b) RL < Zo
Fig. B.1: The two possible configurations of an L-matching network.
B.1 Matching Circuit in the Energy Harvesting

System
In this particular case, the characteristic impedance Zo of the transmission line
seen in Fig. B.1 corresponds with the Archimedean spiral impedance (188 Ω), or
with the generator internal resistance (50 Ω), depending on the case. However, it
is a little bit tricky to know the value of our load, which corresponds to the input
of the half-wave Cockcroft-Walton multiplier. What we do really know is the value
of the optimum source impedance Zs that maximizes the power transfer to the
load. Nevertheless, the maximum power transfer theorem [4] tells us that the source
impedance Zs must be the conjugate of our load ZL . Hence, ZL = Zs∗ , as shown in
Fig. B.1.
Appendix C
Reflection on Ethical, Economic,

Social and Environmental Aspects
We already defined in Chapter 1 energy harvesting as the process of collecting

energy from external sources, such as wind, sunlight, heat or radiofrequency (RF)
waves; in order to feed a certain electronic system. It is also described as the set of
techniques for obtaining energy from the environment, which is closely related to
the concept of renewable energy.
The effects associated with climate change phenomena are diverse [39]. Nowadays,
it is clear that there is a progressive rise in temperatures, but this is only the most
popular effect. Other less known but equally harmful effects are: desertification,
loss of biodiversity, and the presence of more frequent extreme weather events.
Renewable energies and energy harvesting play a crucial role on this, being
substitutes for more dangerous and polluting sources such as oil, gas and nuclear
fuels.
When assessing the extent of the project, three are the impacts that have to be
taken into account:
• The first one is the scientific-technical impact derived from the study of the
harvesting system itself. One of the greatest technological challenges when
working with renewable energies is to achieve good efficiencies of the energy
conversion. In the particular case of RF energy harvesting, the power acquired
by most antennas is under 1 mW, and since we intend to work with circuitry
that is not externally powered (the half-wave Cockcroft-Walton multiplier),
the amount of harvested power is small enough to ensure high conversion ef-
ficiencies (in the order of 30-40 %, as seen in the body of the document).
This problem and its study can lead to enhancement or creation of more
energy-efficient materials (graphene, conductive ink, etc.) and components
(new diodes, transistors, etc.), so the scientific and technological impact of the
project is evident. Furthermore, a radiofrequency energy harvesting is typ-
ically formed by different stages (four) that are not directly related to each
other or which fall within different areas of knowledge, that is, the study of
70
APPENDIX C. REFLECTION ON ETHICAL, ECONOMIC, SOCIAL AND ENVIRONMENTAL A
the RF harvester can be framed within electromagnetism and antenna theory,

while the conditioning and storage circuits can be framed within the power
electronics. This leads to scientific advances or contributions of various kinds,
such as the study of new antenna miniaturization techniques, the circuit mod-
eling of a wideband spiral antenna, or the study of the number of optimum
stages in the half-wave Cockcroft-Walton multiplier.
• The second one is the beneficial environmental impact derived from using
renewable sources. Some of most prominent ones, as solar or wind energy, have
already been widely studied, and other ones, as radiofrequency energy, have
become popular recently due to the increasing number of RF transmitters such
as FM radio stations, mobile base stations, or WiFi equipments. However, all
of them are aimed at reducing pollution and making a wise use of the available
ambient energy, so they must be encouraged and financially promoted by
governments and public institutions. In this line, Royal Decree 314/2006 [40]
(approving the Technical Building Code (CTE) in Spain) establishes the
obligation to incorporate solar thermal installations and photovoltaic panels
in certain buildings. At the European level, the Renewable Energy Di-
rective [41] establishes an overall policy for the production and promotion
of energy from renewable sources in the European Union (EU). It requires
the EU to fulfill at least 20% of its total energy needs with renewables by 2020.
To facilitate as far as possible the initial study of the RF harvesting system,

common and well-known materials (fiberglass, copper, etc.) have been used.
Nevertheless, one of the future challenges of the project is to guide the study
carried out here to the use of eco-friendly materials. Recent studies lead us
the way to use biodegradable printed circuits [42] in order to manufacture the
Archimedean spiral and the conditioning circuit.
• The third one is the social impact. Energy harvesting can have an important
applications in developing countries, where the consumption, exploitation
or purchase of fossil fuels is limited. However, the acquired power levels in
RF harvesting are not sufficient to properly supply the people, so multiple
antennas are needed. Perhaps the (general) architecture of the system can be
modified to increase the amount of harvested power. The current system has
been developed in order to be used anywhere, and in any situation, without
knowing the source or location of the incoming power. It would be interesting
to try to place the system very close to a high power RF source, or to increase
the directivity of the antenna and point it directly to a known source.
Moreover, something that is important is the ability to compact the system as

much as possible, to simplify the manufacture and to reduce costs. In these
terms, cheap materials can be used (as fiberglass) at the expense of its harmful
effect on the environment. Besides that, new studies [43] are trying to use ink-
jet printing (conductive ink) to manufacture their prototypes. It seems as a
APPENDIX C. REFLECTION ON ETHICAL, ECONOMIC, SOCIAL AND ENVIRONMENTAL A
promising solution to investigate. It can drastically reduce costs and ease the
manufacturing process, which is of imperative need in developing countries.
In summarize, climate change and energy are two sides of the same coin: much
of the greenhouse gas (GHG) emissions comes from the energy sector, including
transport. On the other hand, the energy sector also guarantees the cover of
most basic human needs: lighting, heating, mobility or communications. In
consequence, there must be a sustainable compromise between these two facts, and
renewable energies and energy harvesting (solar, wind, RF, etc.) arise as the way
forward. They are a clear contribution to human being and to the protection of the
environment, apart from being an interesting technological challenge from which
other areas of knowledge can take advantage of.
Appendix D
Cost Estimation
The cost evaluation has been divided into four sections: rental of equipment,
rental of software, purchase of materials, and personnel costs. Each final cost has
been calculated according to the formula
Dedication
F inal Cost = Cost · (D.1)
Devaluation
In descending order, 85.21% of the total budget has been earmarked to cover
staff contracts, 11.94% to the use of circuit/EM softwares, 2.37% to the rental of
equipment, and 0.48% to the purchase of materials. According to Table B.1, the
final cost of the project is set to 28460.75e.
Summary of Table B.1:
• Equipment: 558.1e+ 21%(IVA) = 675.30e (2.37%)
• Software: 2808.33e+ 21% (IVA) = 3398.08e (11.94%)
• Materials: 113.93e+ 21% (IVA) = 137.86e (0.48 %)
• Personnel: 24249.51e (85.21%)
• Total Sum: 28460.75e
73
EQUIPMENT
DESCRIPTION COST (e) DEDICATION (months) DEVALUATION (months) FINAL COST (e)
VNA (Second-Hand) 17000 0.5 84 101.19
Signal Generator 16000 0.5 84 95.24
Plotter LPKF (Rent) 120/hours*3 hours - - 360.00
Multimeter 200 0.5 60 1.67
SOFTWARE
DESCRIPTION COST (e) DEDICATION (months) DEVALUATION (months) FINAL COST (e)
Matlab License 800 9 12 600.00
CST Studio License (University) 2500 9 12 1875.00
ADS License (University) 2000 2 12 333.33
MATERIALS
DESCRIPTION UNITS
APPENDIX D. COST ESTIMATION
UNIT PRICE (e) FINAL COST (e)

PCB (FR-4) A4-size 14.00 1 14.00
SMA connectors 2.33 20 46.60
Test Board 1.60 2 3.20
HSMS 2822 Diodes 0.27 50 13.60
HSMS 2850 Diodes 0.49 25 12.23
33 nF Capacitors 0.19 10 1.90
33 pF Capacitors 0.05 50 2.50
Table D.1: Cost Estimation.

1.5 pF Capacitors 0.26 20 5.20
1.2 pF Capacitors 0.24 10 2.40
Inductors (3.3, 6.8 nH) 60 0.14 8.40
33 nH Inductors 10 0.19 1.90
Resistances 0.20 10 2.00
PERSONNEL
DESCRIPTION COST (e/month) QUANTITY DEDICATION (months) FINAL COST (e)
Junior Engineer 2694.39 1 9 24249.51
28460.75e(w. 21% IVA)
74
Appendix E
Publications
On the basis of the studies carried out for the completion of the master’s
thesis entitled ”Design of a Radiofrequency (RF) Energy Harvesting System for
Low-Power Sensor Applications at Microwave Bands”, two articles have seen the
light (accepted for publication), and another one is currently under review.
• A. Alex-Amor, P. Padilla, J. M. Fernández-González, and M. Sierra-Castañer.

“A Miniaturized Ultrawideband Archimedean Spiral Antenna for Low-Power
Sensor Applications in Energy Harvesting”. Microwave and Optical Technol-
ogy Letters, Manuscript submitted for publication, 2018.
• A. Alex-Amor, P. Padilla, J. M. Fernández-González, and M. Sierra-Castañer.

“Outline of a Radiofrequency Energy Harvesting System Based on an Ultra-
wideband Archimedean Spiral Antenna”. Simposium Nacional URSI 2018,
Granada, Spain.
• A. Alex-Amor, J. M. Fernández-González, P. Padilla, and M. Sierra-Castañer.

“Comparison Between a Multiband PIFA and an Ultrawideband Archimedean
Spiral Antenna for Energy Harvesting in Microwave Bands”. 2nd URSI AT-
RASC, Gran Canaria, Spain, 2018. Manuscript accepted for publication.
75
Bibliography
[1] L. Mateu and F. Moll. “Review of Energy Harvesting Techniques and Ap-
plications for Microelectronics”. Proceedings of the SPIE - The International
Society for Optical Engineering, 5837, June 2005.
[2] L. Tran, H. Cha and W. Park. “RF Power Harvesting: a Review on Designing
Methodologies and Applications”. Micro and Nano System Letters, Feb. 2017.
[3] J. A. Hagerty, F. B. Helmbrecht, W. H. McCalpin, R. Zane, and Z. B. Popovic.

“Recycling Ambient Microwave Energy With Broad-Band Rectenna Arrays”.
IEEE Transactions on Microwave Theory and Techniques, Vol. 52, No. 3,
2004.
[4] David. M. Pozar. “Microwave Engineering”. New York: John Wiley & Sons,
3rd Edition.
[5] Ministerio de Energı́a, Turismo y Agenda Digital, Gobierno de España.

“Registro Público de Concesiones”. Retrieved March 19, 2018, from https:
//sedeaplicaciones.minetur.gob.es/setsi_regconcesiones/
[6] “Frecuencias telefonı́a móvil”. In Wikipedia, Retrieved August 12, 2017

from https://wiki.bandaancha.st/Frecuencias_telefon%C3%ADa_m%C3%
B3vil
[7] J. M. Hernando, L. Mendo, J. M. Riera. “Comunicaciones Móviles”. Editorial

Centro de Estudios Ramón Areces, S. A., 3rd Edition.
[8] M. Sierra-Pérez, B. Galocha, J. L. Fernández-Jambrina, M. Sierra-Castañer.

“Electrónica de Comunicaciones”. Pearson Prentice Hall, 1st Edition.
[9] R. Scheeler, S. Korhummel, and Z. Popovic. “A Dual-Frequency Ultralow-

Power Efficient 0.5-g Rectenna”. IEEE Microwave Magazine, Vol. 15, 2014.
[10] R. Md. Ferdous, A. W. Reza, M. Faisal Siddiqu. “Renewable energy har-

vesting for wireless sensors using passive RFID tag technology: A review”.
Renewable and Sustainable Energy Review, Vol. 58, pp. 1114-1128, May 2016.
[11] A. K. Moghaddam, J. H. Chuah, H. Ramiah, J. Ahmadian, P. Mak, and Rui

P. Martins. “A 73.9%-Efficiency CMOS Rectifier Using a Lower DC Feeding
(LDCF) Self-Body-Biasing Technique for Far-Field RF Energy-Harvesting
76
BIBLIOGRAPHY 77
Systems”IEEE Transactions on Circuits and Systems-I, Vol. 64, No. 4, April

2017.
[12] M. Mattsson , C. I. Kolitsidas, and B. L. G. Jonsson. “Dual-Band Dual-

Polarized Full-Wave Rectenna Based on Differential Field Sampling”. IEEE
Antennas and Wireless Propagation Letters, Vol. 17, No. 6, 2018.
[13] Q. Awais, Y. Jin, H. T. Chattha, M. Jamil, Q. He, and B. A. Khawaja. “A

compact rectenna system with high conversion efficiency for wireless energy
harvesting”. IEEE Access, 2018.
[14] A. Alex-Amor, J. M. Fernández-González, P. Padilla, and M. Sierra-Castañer.

“Comparison Between a Multiband PIFA and an Ultrawideband Archimedean
Spiral Antenna for Energy Harvesting in Microwave Bands”. 2nd URSI AT-
RASC, 2018. Manuscript accepted for publication.
[15] V. H. Rumsey, “Frequency Independent Antennas”, IRE International Con-

vention Record, Vol. 5, pp. 114-118, March 1957.
[16] M. H. El-Feshawy, H. Hammad. “Comparison between broadband equiangu-

lar and Archimedean spiral rectennas for energy harvesting”. Radio Science
Conference (NRSC), 2016.
[17] M. N. Afsar, Y. Wang, R. Cheung. “Analysis and Measurement of a Broad-

band Spiral Antenna”. IEEE Antennas and Propagation Magazine, Vol. 46,
No. 1, Feb. 2004.
[18] C. A. Balanis. “Antenna Theory: Analysis and Design”. John Wiley & Sons,
3rd edition, 2005.
[19] E.D. Caswell. “Design and Analysis of Star Spiral with Application to Wide-
band Arrays with Variable Element Sizes” (Doctoral Thesis). Bradley Depart-
ment of Electrical and Computer Engineering, Virginia Polytechnic Institute
and State University.
[20] A. Mehrabani, L. Shafai. “Polarisation reconfigurable, centre-fed, and low-

profile Archimedean spiral antennas with unidirectional broadside patterns”.
IET Microwaves, Antennas and Propagation, Nov. 2016.
[21] K. Wong. “Planar Antennas for Wireless Communications”. John Wiley &
Sons, 2003.
[22] A. Dadgarpour, A. Abbosh, and F. Jolani. “Planar Multiband Antenna for

Compact Mobile Transceivers”. IEEE Antennas and Propagation Letters, Vol.
10, pp. 651-654, 2011.
[23] S. Al Ja’afreh, Y. Huang, L. Xing. “Low profile and wideband planar inverted-
F antenna with polarization and pattern diversities”. IET Microwaves, An-
tennas & Propagation, 1751-8725, Aug. 2015.
BIBLIOGRAPHY 78
[24] J. L. Volakis, Chi-Chih Chen, K. Fujimoto. “Small Antennas: Miniaturization

Techniques & Applications”. McGraw Hill Education, 1st edition.
[25] Q. Liu, C.-L. Ruan, L. Peng, and W.-X. Wu. “A Novel Compact Archimedean
Spiral Antenna with gap-Loading”. Progress in Electromagnetics Research
Letters, Vol. 3, pp. 169–177, 2008.
[26] C. Sharma, and D. K. Vishwakarma. “Miniaturization of Spiral Antenna

Based on Fibonacci Sequence Using Modified Koch Curve”. Antennas and
Wireless Propagation Letters, Vol. 16, 2017.
[27] K. Vinayagamoorthy, J. Coetzee, D. Jayalath. “Microstrip to Parallel Strip

Balun as Spiral Antenna Feed”. IEEE 75th Vehicular Technology Conference
(VTC Spring), 2012.
[28] K. Pal Singh, R. Ranjan, and R. Deshpande. “Design of a Wideband Balun

for Archimedean Spiral Antenna in Energy Harvesting”. 2015 International
Conference on Applied and Theoretical Computing and Communication Tech-
nology (iCATccT), 2015.
[29] LPKF Laser & Electronics. “Manual ProtoMat S100“. http://www.lpkfusa.

com/downloads/support/docs/man_s100.pdf
[30] H. Sun, Y. Guo, M. He, Z. Zhang. “A Dual-Band Rectenna Using Broad-

band Yagi Antenna Array for Ambient RF Power Harvesting”. Antennas and
Wireless Propagation Letters, Vol. 12, 2013.
[31] P. I. Wells. “The Attenuation of UHF Radio Signals by Houses”. IEEE Trans-
actions on Vehicular Technology, Vol. VT-26, No. 4, Nov. 1977.
[32] D. Kind and K. Feser. High-voltage Test Techniques, 2nd ed., Newnes, 2001.
[33] M. Ruzbehani. “A Comparative Study of Symmetrical Cockcroft-Walton

Voltage Multipliers”. Journal of Electrical and Computer Engineering, Vol.
2017, 4805268, 2017.
[34] “HSMS-282x Surface Mount RF Schottky Barrier Diodes datasheet”. Avago

Technologies, San José, California, USA.
[35] “Agilent HSMS-285x Series Surface Mount Zero Bias Schottky Detector
Diodes datasheet”. Agilent Technologies, Santa Clara, California, USA.
[36] “Agilent E4438C vector signal generator datasheet”. Agilent Technolo-

gies, Santa Clara, California, USA. http://literature.cdn.keysight.
com/litweb/pdf/5988-4039EN.pdf
[37] “Yokogawa DL9240L digital oscilloscope datasheet”. Yokowaga Meters &

Instruments Corporation, Tokyo, Japan. http://www.yokogawa.com/pdf/
provide/E/GW/IM/0000019382/0/IM701310-17E.pdf
BIBLIOGRAPHY 79
[38] Stephen A. Maas. “Nonlinear Microwave and RF Circuits”. Artech House

USA, 2nd Edition, 2003.
[39] X. Labandeira, P. Linares, K. Würzburg. “Energı́as renovables y cambio

climático”. Cuadernos económicos de ICE, No. 83.
[40] Real Decreto 314/2006, de 17 de marzo, por el que se aprueba el

Código Técnico de la Edificación. http://www.boe.es/boe/dias/2006/03/
28/pdfs/A11816-11831.pdf
[41] Directive 2009/28/EC of the European Parliament and of the Council of

23 April 2009 on the promotion of the use of energy from renewable
sources and amending and subsequently repealing Directives 2001/77/EC
and 2003/30/EC. https://eur-lex.europa.eu/legal-content/EN/TXT/
PDF/?uri=CELEX:32009L0028&from=EN
[42] X. Huang, Y. Liu, S. Hwang, S. Kang, D. Patnaik, J. Fajardo-Cortes, J. A.

Rogers. “Biodegradable Materials for Multilayer Transient Printed Circuit
Boards”. Advanced Materials, Sept., 2014.
[43] Do Hanh Ngan Bui. “Printed flexible antenna for energy harvesting”. Optics
/ Photonic. Université Grenoble Alpes, 2017.

RF Energy Harvesting System Design

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

RF Energy Harvesting System Design

Uploaded by

Copyright:

Available Formats

UNIVERSIDAD POLITÉCNICA DE MADRID

ESCUELA TÉCNICA SUPERIOR

MASTER UNIVERSITARIO EN INGENIERÍA DE TELECOMUNICACIÓN

TRABAJO FIN DE MASTER

TITULO: DESIGN OF A RADIOFREQUENCY (RF)

AUTOR: Antonio Alex Amor

AUTOR: Antonio Alex Amor

MIEMBROS DEL TRIBUNAL CALIFICADOR

3 Conditioning Circuit. Cockcroft-Walton Multiplier 42

5 Conclusion & Future Work 62

A Harmonic Balance Method 65

B Design of L-Matching Networks 68

C Reflection on Ethical, Economic, Social and Environmental As-

1.1 Full RF energy harvesting system. . . . . . . . . . . . . . . . . . . . . 15

2.1 Current distribution over a miniaturized Archimedean spiral antenna. 21

2.15 Simulated reflection coefficient of the Archimedean spiral antenna

3.1 Circuital scheme of the n-stage half-wave Cockcroft-Walton voltage

4.1 General scheme of the RF harvesting system. Note as the source

A.1 Partitioning of a microwave circuit into linear and non-linear subcir-

B.1 The two possible configurations of an L-matching network. . . . . . . 69

2.1 Measurements on the harvested power. . . . . . . . . . . . . . . . . . 41

4.1 Simulated optimum source impedances Zs , load resistances RL , out-

D.1 Cost Estimation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

Keywords: energy harvesting, antenna design, Archimedean spiral antenna,

Energy harvesting is defined as the process of obtaining energy from external

• RF harvester: an antenna capable of acquiring RF ambient power from the

• Matching circuit: it is situated between RF harvester and conditioning circuit

• Storage circuit: it stores the available energy in order to feed an electronic

Secondly, we will focus on the conditioning circuit. One of the topologies of

1.1 Energy Harvesting. Full System Architecture

Typically, a RF energy harvesting system is formed by four stages [3], as shown

• RF harvester: an antenna capable of acquiring RF ambient power from the

• Matching circuit: it is situated between RF harvester and conditioning circuit

Fig. 1.1: Full RF energy harvesting system.

1.2 Radio Spectrum. Bands of Interest

1.3 Project Description

Fig. 1.2: Section of the electromagnetic spectrum. Source: [4].

• Chapter 1 introduces the global design of the proposed RF energy harvesting

• Chapter 2 presents the design of the radiofrequency harvester. Firstly, a

• Chapter 3 depicts the design of the conditioning circuit, based on a half-wave

• Chapter 4 studies the matching circuit. It is analyzed the effect of modifying

connected to the conditioning circuit in order to maximize the power delivered

1.4 State of the Art of RF Energy Harvesting Sys-

(a) Three-element dual-band Yagi rectenna (b) Equiangular spiral

The current bottleneck of RF harvesting systems is the conditioning circuit, which

is typically under 30%. In [3], a half-wave rectifier circuit is implemented over an

(a) Full-wave rectenna. Source: [12]

(b) Rectenna based on a Cockcroft-Walton multi-

2.1 Design of an Archimedean Spiral Antenna

(a) 10 GHz (b) 500 MHz (c) 300 MHz

Fig. 2.1: Current distribution over a miniaturized Archimedean spiral antenna.

Self-complementarity is only reachable if the metallic strip width, w, is equal to

In Fig. 2.2, it is presented the Archimedean spiral antenna, designed over

Fig. 2.2: Archimedean spiral antenna.

Fig. 2.3: Reflection coefficient of the Archimedean spiral antenna (normalized to

2.2 Comparison between a Multiband PIFA and

In Fig. 2.7, it is presented a typical dual-band GSM PIFA. It is formed by the

Fig. 2.7: Parts of a dual-band PIFA antenna.

Fig. 2.8: Reflection coefficient of the dual-band PIFA antenna.

2.2.1 New Design

Fig. 2.9: Tri-band PIFA antenna.

(a) Transversal Cut (b) Top Plane