Energy-Neutral Event Monitoring for Internet of Nano Things

Home , Energy neutral design

Energy-neutral Event Monitoring for Internet of Nano Things

Thesis submitted in fulﬁlment of the requirements for the degree of

Doctor of Philosophy in Computer Science and Engineering

Najm Hassan

Supervisors: Prof. Mahbub Hassan and Prof. Chun Tung Chou

November 2018 PLEASE TYPE THE UNIVERSITY OF NEW SOUTH WALES Thesis/Dissertation Sheet

Surname or Family name: Hassan

First name: Najm Other name/s:

Abbreviation for degree as given in the University calendar: PhD

School: Computer Science and Engineering Faculty: Engineering

Title: Energy-neutral Event Monitoring for Internet of Nano Things

Abstract 350 words maximum: (PLEASE TYPE)

Advancements in nanotechnology promise new capabilities for the Internet of Things (IoT) to monitor extremely fine-grained events with sensors as small as up to a hundred nanometers. Researchers predict that such tiny sensors can be connected to the Internet using graphene-based nano-antenna radiating in the terahertz band, giving rise to the so called Internet of Nano-Things (IoNT). Powering such wireless communications with nanoscale energy supply, however, is a major challenge to overcome. Since in many application domains, different types of events discharge different amounts of energy to the environment, we propose an energy-neutral event monitoring framework, called eNEUTRAL IoNT, that allows the sensors to transmit event information using only the amount of energy harvested from the events. We design and analyse two implementation methods for this framework. The first method uses a single pulse containing the entire energy harvested from the event but manipulates its pulse width (time duration) to create unique pulse amplitude for a given combination of event type and its location. In the second option, the harvested event energy is divided into two pulses so that the energy of the first pulse uniquely defines a location and the second pulse uses the remaining energy to identify event types. To minimize classification error at the receiver, we optimize pulse durations in the single-pulse option and pulse energies in the dual-pulse option. Feasibility of eNEUTRAL IoNT is demonstrated using extensive numerical experiments involving terahertz channels. We find that the dual-pulse approach significantly outperforms the single-pulse approach achieving 99% accuracy for detecting both location and event type in 10-node network monitoring two different event types for a radius of 28 mm. As nanoscale energy harvesters and transmitters are still not available to realize operational event monitoring nodes, we, therefore, evaluate and design a key component of the IoNT system, namely pulse generator, using COMSOL Multiphysics. We first surveyed the literature of different approaches for pulse generation that generate Surface Plasmon Polaritons (SPPs) which lead to femtosecond long pulses in graphene. Based on our analysis, we found that most of the existing configurations require complex structures including a prism, periodic slits and special circuits which may be difficult to implement in severely resource constraint nodes. Using COMSOL Multiphysics, we show that pulses in the terahertz band can be generated using the near-field method by which the matching condition for excitation of SPPs can be easily satisfied. The performance of the proposed near-field excitation method in the terahertz band is studied via numerical simulations. This includes the choice of the source, its phase angle, the chemical potential, and frequency. The proposed model can be a good candidate for a low-complexity realization of a THz pulse generator in tiny IoNT nodes. We believe that our findings will open the door for a new direction of research and development toward the energy-neutral event monitoring systems in IoNT.

Declaration rel ating to disposition of project thesis/dissertation

I hereby grant to the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or in part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all property rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation.

I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstracts International (this is applicable to doctoral theses only).

…………………………………………………………… ……………………………………..……………… ……….……………………...…….… Signature Witness Signature Date

The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditions on use. Requests for restriction for a period of up to 2 years must be made in writing. Requests for a longer period of restriction may be considered in exceptional circumstances and require the approval of the Dean of Graduate Research.

FOR OFFICE USE ONLY Date of completion of requirements for Award:

ORIGINALITY STATEMENT

I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgment is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project’s design and conception or in style, presentation, and linguistic expression is acknowledged.

Signed ...... Najm Hassan November 2018 COPYRIGHT STATEMENT

I hereby grant the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or part in the University libraries in all forms of media, now or hereafter known, subject to the provisions of the Copyright Act 1968. I retain all proprietary rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation. I also authorise University Micro- ﬁlms to use the 350 word abstract of my thesis in Dissertation Abstract Interna- tional (this is applicable to doctoral theses only). I have either used no substantial portions of copyright material in my thesis, or I have obtained permission to use copyright material; where permission has not been granted I have applied/will apply for a partial restriction of the digital copy of my thesis or dissertation.

Signed ...... Najm Hassan November 2018

AUTHENTICITY STATEMENT I certify that the Library deposit digital copy is a direct equivalent of the ﬁnal oﬃcially approved version of my thesis. No emendation of content has occurred and if there are any minor variations in formatting, they are the result of the conversion to digital format.

Signed ...... Najm Hassan November 2018 i

ABSTRACT Advancements in nanotechnology promise new capabilities for the Internet of Things (IoT) to monitor extremely fine-grained events with sensors as small as up to a hundred nanometers. Researchers predict that such tiny sensors can be connected to the Internet using graphene-based nano-antenna radiating in the terahertz band, giving rise to the so called Internet of Nano-Things (IoNT). Pow- ering such wireless communications with nanoscale energy supply, however, is a major challenge to overcome. Since in many application domains, different types of events discharge different amounts of energy to the environment, we propose an energy-neutral event monitoring framework, called eNEUTRAL IoNT, that allows the sensors to transmit event information using only the amount of energy harvested from the events. We design and analyse two implementation methods for this framework. The first method uses a single pulse containing the entire energy harvested from the event but manipulates its pulse width (time duration) to create unique pulse amplitude for a given combination of event type and its location. In the second option, the harvested event energy is divided into two pulses so that the energy of the first pulse uniquely defines a location and the second pulse uses the remaining energy to identify event types. To minimize classification error at the receiver, we optimize pulse durations in the single-pulse option and pulse energies in the dual-pulse option. Feasibility of eNEUTRAL IoNT is demonstrated using extensive numerical experiments involving terahertz channels. We find that the dual-pulse approach significantly outperforms the single- pulse approach achieving 99% accuracy for detecting both location and event type in 10-node network monitoring two different event types for a radius of 28 mm. As nanoscale energy harvesters and transmitters are still not available to realize operational event monitoring nodes, we, therefore, evaluate and design a key component of the IoNT system, namely pulse generator, using COMSOL ii

Multiphysics. We first surveyed the literature of different approaches for pulse generation that generate Surface Plasmon Polaritons (SPPs) which lead to femtosecond long pulses in graphene. Based on our analysis, we found that most of the existing configurations require complex structures including a prism, periodic slits and special circuits which may be difficult to implement in severely resource constraint nodes. Using COMSOL Multiphysics, we show that pulses in the terahertz band can be generated using the near-field method by which the matching condition for excitation of SPPs can be easily satisfied. The performance of the proposed near-field excitation method in the terahertz band is studied via numerical simulations. This includes the choice of the source, its phase angle, the chemical potential, and frequency. The proposed model can be a good candidate for a low-complexity realization of a THz pulse generator in tiny IoNT nodes. This thesis is dedicated to my beloved wife for her love, support and encouragement, To my lovely little son Ammar Hassan and To Ammar’s grandparents for their timeless love iv

ACKNOWLEDGEMENTS

Working at the School of Computer Science and Engineering Computer Sci- ence and Engineering School (CSE) at the University of New South Wales University of New South Wales (UNSW) has been a great pleasure and an incredible privi- lege. First, I would like to express my sincere appreciation and profound gratitude to my supervisor, Professor Mahbub Hassan for his exceptional support, encouragement, and guidance during all stages of this research. His truly valuable aca- demic excellence, scientiﬁc intuition, and a beautiful mind have made him a constant oasis of ideas and passions in science. This has inspired and enriched my growth as a student. I also express my genuine thanks to Professor Chun Tung Chou, as my joint-supervisor, for his kind help, incredible support and valuable discussion. It was an honor for me to work closely with such a talented, polite and creative personality. In addition, I express my sincere appreciation to Dr. Ming Ding for great discussion and support. It was an excellent opportunity for me to have him in my supervision panel. I also express my genuine thanks to our external collaborator Dr. Marios Mattheakis, Harvard University USA, for his incredible support contributing to my thesis. Most of all, I am profoundly and forever indebted to my parents for their never-ending love, and encouragement throughout my entire life. Unfortu- nately, I lost my mother in the third year of my PhD. It was a tragedy for me as she was my soul mate. I am sure my parents’ souls is always praying for me and I also continuously pray for the repose of their souls. Last but no means least, it gives me immense pleasure to thank Dr. Eisa Zare- pour and lab members I worked with in CSE UNSW. I enjoyed my PhD study because of them. I had a great life and study experience. They are wonderful v people and always ready to help. Finally, I would like to thank my family for their unwavering understanding and supports. I thank them mainly for their understanding and the acceptance that I have been so busy during my PhD study that we have hardly been able to spend much time together. They are my source of strength, and without their tremendous support, this thesis would never have been started.

Najm Hassan Sydney, Australia November 2018 Contents

1 Introduction1 1.1 Motivation...... 3 1.2 Problem Statement...... 5 1.3 Contributions...... 6 1.4 Dissertation Organization...... 10

2 Background and the State of the Art 12 2.1 Internet of Nano-Things...... 13 2.2 Building blocks of IoNTs...... 13 2.2.1 Nanosensor Node...... 14 2.2.2 Propagation Channel...... 19 2.2.3 Receiver...... 22 2.3 Applications...... 25 2.3.1 Medical Applications...... 25 2.3.2 Industrial Applications...... 28 2.3.3 Environmental Applications...... 28 2.4 Sensor powering challenges...... 29 2.5 Pulse generation for tiny IoNT nodes...... 31

3 Energy-neutral Single Pulse Transmission 38 3.1 Introduction...... 39 3.2 System Model...... 40

vi CONTENTS vii

3.2.1 Event Model...... 42 3.2.2 Structure of the monitoring node...... 43 3.2.3 Pulse Transmitter Model...... 44 3.2.4 Channel Model...... 46 3.2.5 Pulse Receiver Model...... 46 3.3 Single Pulse Transmission (SPT)...... 48 3.3.1 Problem description...... 48 3.3.2 Pulse Width Allocation (PWA)...... 52 3.3.3 Numerical investigation on PWA...... 54 3.3.4 Impact of pulse widths on error probability...... 54 3.3.5 Pulse widths allocation to more than two nodes...... 58 3.4 Evaluation...... 59 3.4.1 Impact of node-RS distance...... 60 3.4.2 Impact of number of nodes...... 60 3.4.3 Impact of number of event types...... 61 3.5 Chapter Summary...... 62

4 Energy-neutral Dual Pulse Transmission 64 4.1 Introduction...... 64 4.2 Dual Pulse Transmission (DPT)...... 65 4.2.1 Problem description...... 66 4.2.2 Pulse Energy Detection...... 67 4.3 Pulse Energy Allocation (PEA)...... 68 4.4 Performance Evaluation...... 69 4.4.1 Impact of the network radius...... 70 4.4.2 Impact of the number of nodes...... 72 4.4.3 Impact of the number of event types...... 72 4.4.4 Impact of the event energy spread...... 73 4.4.5 Impact of the node placement...... 75 CONTENTS viii

4.4.6 Impact of the event energy ﬂuctuations...... 76 4.5 Discussion...... 77 4.6 Chapter Summary...... 78

5 Pulse Generator for tiny IoNT Nodes 79 5.1 Introduction...... 80 5.2 SPP Preliminaries...... 81 5.3 Discussion...... 83 5.4 Pulse Generation for Event Monitoring Node...... 84 5.5 Simulation and Numerical Results...... 86 5.5.1 Impact of the frequency on the amplitude of SPP...... 87 5.5.2 Impact of the chemical potential on the SPP resonance... 87 5.5.3 Impact of the phase angle of the evanescent source on SPP resonance...... 90 5.5.4 Impact of the type of evanescent source on the SPP resonance 90 5.6 Chapter Summary...... 91

6 Conclusion and Future Works 93 6.1 Concluding Remarks...... 93 6.2 Future Directions...... 95

Bibliography 97

A Acronyms 113

Appendix 113 List of Figures

2.1 An integrated nanosensor node [1]...... 14 2.2 Interface of the HITRAN on the web tool. Molecular absorption coefficient can be calculated directly from HITRAN database.... 21 2.3 Total path-loss in dB of THz band at different distances...... 22 2.4 Molecular Absorption Noise PSD in dB/Hz at different distances in THz band...... 23 2.5 Block diagram of energy based detection where the pulse energy is detected by integrating the received energy over an Integration

window (Tint)...... 23

2.6 In Energy detection, the integration window (Tint) is greater than the pulse width (duration) so the additional noise is also integrated. 24 2.7 Block diagram of a CTMA based detection to detect either peak power (pulse amplitude) or pulse energy...... 24 2.8 Network architecture for the IoNT: (a) human lungs monitoring [2] (b) monitoring chemical reactions [3] and (c) plant monitoring [4]. 26 2.9 Generating SPPs using different excitation methods...... 32 2.10 Schematic representation of Attenuated Total Reflection (ATR) methods which are used for SPP excitation: (a) Kretschmann -Raether and (b) Otto Configuration...... 34

3.1 Illustration of eNEUTRAL event monitoring node which includes an energy harvester and a radio...... 43

ix List of Figures x

3.2 CTMA detector to detect the peak power of the pulse...... 47 3.3 Block diagram of the event monitoring system...... 49 3.4 Illustration of pulse widths vs peak power of transmitted pulses in single-pulse approach...... 51 3.5 (a) Error probability versus pulse width of two nodes at distance d = 30 mm (b) A cross-section of heatmap...... 57 3.6 Energy ratio vs pulse width ratio: (a) Local minimum (b) Global minimum...... 58 3.7 The effect of distance on the classification error in single-pulse approach for 5-, 10- and 20-node cases...... 60 3.8 Effect of number of nodes on the classification error for a 30mm network monitoring two distinct event types with energies 1 aJ and 2 aJ...... 61 3.9 Effect of number of event types on classification error for a 5-node network of radius 30 mm...... 62

4.1 CTMA detector to detect the pulse energy...... 68 4.2 The effect of distance on the classification error in SPT and DPT for 5- and 10-node cases...... 71 4.3 Effect of number of nodes on achievable radius for target error be- 2 3 tween 10− and 10− when monitoring two distinct event types with energies 1 aJ and 2 aJ...... 72 4.4 Effect of number of nodes on the classification error for a 30 mm network monitoring two distinct event types with energies 1 aJ and 2 aJ...... 73 4.5 Effect of number of event types on classification error for a 5-node network of radius 30 mm...... 74 4.6 Effect of number of events types on achievable radius for target 2 3 error between 10− and 10− for a 5-node network...... 74 List of Figures xi

4.7 Effect of placement error for a 5-node network with a radius of 30 mm for 2 distinct event types having energies of 1 aJ and 1.5 aJ.. 76 4.8 Effect of event energy fluctuation for a 5-node network with a radius of 30 mm for 2 distinct event types...... 77

5.1 Illustration of a pulse generator where SPP is excited on graphene sheet using the near-ﬁeld method. A nanoantenna takes the SPP wave and radiates it into a free-space EM wave...... 85 5.2 The magnetic intensity I demonstrates the plasmon resonance on a doped graphene monolayer (dotted white line). The spatial distribution of I¯ is represented in arbitrary values (a.u.) by (a) colorbar in x-y plane and by (b) with a solid blue line along the graphene layer (y = 0). An evanescent TM EM dipole source (n=1) of frequency f = 10 THz and with phase angle θ = π/2 is located at

Xs = 100 µm and Ys = 30 nm above the graphene layer, indicated by a tiny white spot. The chemical potential of graphene is con-

sidered µc = 0.5 eV. The SPP wavelength is calculated to be λsp = 5 µm...... 88 5.3 The maximum value of I¯ for several values of frequency is presented revealing that the higher the frequency the better plasmon resonance is achieved...... 89

5.4 The maximum value of I¯ for several values of doping µc is shown revealing that at a certain frequency of f = 10 THz, the plasmon

resonance is a increasing function of the doping for values µc < 0.4 eV...... 89 5.5 The maximum value of I¯ for several as function as the phase angle θ of the source is demonstrated with blue solid line. The best plasmon resonance is observed for the θ = π/2...... 90 List of Figures xii

5.6 The plasmon proﬁle intensity I¯ along the graphene sheet is represented for several values of evanescent parameter n of the source, that is, for n = 1, n = 2 and n = 3...... 91 List of Tables

2.1 Power consumption of diﬀerent types of nanosensors versus power harvested at nanoscale...... 19

3.1 Table of the most frequently used parameters in single-pulse approach of eNEUTRAL IoNT framework...... 53 3.2 The Composition of normal air...... 54 3.3 Pulse widths and the error probability for 2-5 nodes at distance d = 30 mm...... 59

4.1 Comparison between SPT and DPT...... 66 4.2 Table of the most frequently used parameters in dual-pulse solution of eNEUTRAL IoNT framework...... 70 4.3 Impact of energy gap of diﬀerent event types on the error probability at distance d = 30 mm...... 75

xiii List of Publications xiv

List of Publications

Journal Articles

• Najm Hassan, Chun Tung Chou, Mahbub Hassan,“eNEUTRAL IoNT: Energy- neutral Event Monitoring for Internet of Nano Things,” under revision in IEEE Internet of Things Journal (IoT-J), 2018.

• Najm Hassan, Marios Mattheakis, Ming Ding, “Sensorless Node Architecture for Events Detection in Self-Powered Nanosensor Networks,” Nano Communi- cation Networks (Elsevier) Journal, vol. 19, pp 1-9, November 2018.

Conference Proceedings

• N Hassan, C. T. Chou, M Hassan, “Event and Node Identiﬁcation from a Single-Pulse Transmission in Self-powered Nanosensor Networks,” in Proceed- ings of 4th ACM International Conference on Nanoscale Computing and Communication (NANOCOM), September 27-29, 2017, Washington D.C., DC, USA.

• E Zarepour, N Hassan, M Hassan, C. T. Chou, and M E Warkiani, “Design and Analysis of a Wireless Nanosensor Network for Monitoring Human Lung Cells,” in Proceedings of 10th EAI International Conference on Body Area Networks (BodyNets), 2015, September 28-30, Sydney, Australia. Chapter 1

Introduction

The last decade has witnessed significant advances in nanotechnology which makes it possible to fabricate sensor nodes at nanoscale at or below a hundred nanometers. These nodes are made from novel materials which have unique physical, electrical and optical prosperities. Such nodes have the capabilities to sense molecule level events in immediate surroundings [1]. Recent studies have revealed that these nanosensors can communicate using graphene-based nanoantenna radiating in the terahertz band (0.1-10 THz) [5,6] ushering new Internet of Thing (IoT) capabilities for gathering knowledge at an unprecedented depth and scale. Researcher are now pursuing this new direction of IoT under the banner of Internet of Nano Thing (IoNT)[7] with nanoscale monitoring techniques explored for human body [2,8], plants [4], chemical processes [9], and so on. There is a large volume of published studies related to Wireless Sensor Net- work (WSN), however, communication at the nanoscale is relatively a new research topic. In recent years, the area of nanocommunication has significantly attracted the research community mainly due to its unprecedented applicability in different applications. To name a few, biomedical, industrial, environmental and military are envisaged applications where nanoscale communication is expected to be deployed to empower bottom-up control of molecule level events. In biomedical health monitoring systems, nanosensor networks can be deployed

1 Chapter 1 Introduction 2 to monitor cholesterol, sodium, glucose ions in blood or to monitor different infectious agents [1]. There are some reported applications of nanoscale communication in industrial such as real-time monitoring and controlling chemical processes at the molecular level [2, 10]. In environmental applications, plant monitoring systems, plagues defeating systems are complementary applications of nanoscale communication [1]. On the military and defense side, the nanoscale communication applications are nuclear, biological and chemical (NBC) defenses, where chemical and biological nanosensors can be used to detect harmful chem- icals and biological arsenal in the distributed manner. The other military applications are damage detection systems which can be used to detect very small cracks in textiles, vehicles and rockets etc [1]. In such applications, sensed information and measurements can be transmitted to a distinct receiver by estab- lishing nanoscale wireless communication within the IoNT paradigm. How to balance energy expenditure so that it never exceeds the harvested energy in tiny IoNT nodes is a challenging problem to solve. Therefore, novel sensorless energy- neutral solutions are required to monitor events in the environment without employing onboard sensors. Such solutions can obviate the need for several conventional sensor node elements (i.e. microprocessor, memory, and the sensor) to avoid the problem of powering, however, it can also limit the sensing capability locally. Instead, the molecule level events sensing and monitoring can be accom- plished remotely at a distant receiver which is least explored. If different types of events emit different amounts of energy to the environment, then technically it is feasible to identify events, from their emitted energy, at the receiver. In addition to events detection, one of the major issues is to determine the identity of nodes in severely resource constraint nanoscale communication. In the current IoT system, the packet-based conventional addressing scheme is used where a large number of bits within the payload, header, and the preamble, for each packet, are transmitted. Adopting a packet-based approach of IoT system will require much power to transmit many bits which can be energy inefficient in Chapter 1 Introduction 3 severely resource constraint IoNT systems. The aim of this thesis is to propose novel packet-less energy-neutral solutions to overcome the power consumption limitation. An energy-neutral solution aims to balance the energy expenditure in IoNT nodes so that it never exceeds the harvested energy. In this direction, we propose energy-neutral event monitoring schemes tailored with the limited energy and restricted resources of tiny sensors. At the nanoscale, energy harvesters and transmitters are still not available to realize operational IoNT system. Therefore, this thesis will also model the key component of the tiny IoNT node, i.e., a pulse generator, using COMSOL Multi- physics. The proposed pulse generator is simple to construct and uses a near-field method to generate a pulse in the THz band. We expect that simple structures are more likely to succeed in the future design of severely resource constraint nodes.

1.1 Motivation

Recent studies confirm that nanosensors may be able to communicate in the Terahertz (THz) band as the transmission band, giving rise to the so-called IoNTs [5–7]. As THz band is the resonance frequency of molecules, communication in this band is severely affected by attenuation and molecular absorption noise [10, 11]. Although, there has been some recent work in the literature to address these issues. Recently, novel communication elements, channel modeling, and modulation schemes, and network architectures have been investigated for IoNTs [2,6, 10, 12–15]. In this direction, graphene-based THz nanoantenna and nanotransceiver which can generate THz pulses suitable for nanoscale communication are investigated in [6,13]. Novel communication protocols and THz channel models are developed [2,10,12]. To enable communication between nanosensors, novel network architectures are investigated in [2, 14, 15]. Recently, novel information coding techniques [16, 17] and energy harvesting schemes [18] are also proposed. Amongst all, pulse-based communication which is based on the Chapter 1 Introduction 4 transmission of hundred femtoseconds long pulses is proposed in [19]. Power supply at the nanoscale is a challenging problem. As batteries are difficult to be built into and replaced in nanosystems, researchers are developing nanoscale energy harvesters (a.k.a nanogenerators) [20–25], that can scavenge tiny amounts of energy from the ambient environment to power event detection and very basic wireless communication where all data is transmitted as a series of extremely short (a few hundred femtoseconds) pulses [19]. The harvested energy can be used to transmit short pulses which can reduce the total energy consumption drastically compared to conventional continuous wave wireless communications. However, as each pulse consumes a finite amount of energy, the harvested energy may not be adequate to transmit a large number of pulses uploading event information and sensor node identification each time events are detected in the environment. Interestingly enough, the information in the pulse based communication can be encoded either in the pulse with (duration), pulse amplitude or pulse position of the transmitted pulse. Therefore it is possible to use a pulse to convey address and event information using any of these parameters of the transmitted pulse. However, pulse position modulation requires synchronization between a transmitting node and a receiver which is difficult to achieve in resource-constrained nanoscale communication. We, therefore, in this thesis, only use pulse width (duration) and transmitted energy of a pulse to convey two pieces (address and event) of information. Another example of using the property of the transmitted signal to identify the transmitter can be found in [26] where the amplitude of the signal is used to identify the transmitter. However the focus of [26] is on medium access control in bacterial communication networks but our focus is on delivering node identity and event information in THz electromagnetic communication. Chapter 1 Introduction 5

1.2 Problem Statement

Powering wireless communication at the nanoscale is a major challenge. Be- cause, nanoscale energy harvesters (a.k.a nanogenerators) [20–25] can not generate enough power to transmit a large number of pulses uploading event information and sensor node identification each time events are detected in the environment. Likewise, the insufficient harvesting rate in embedded environments may not allow continuous operation of sensors [27]. This means that there is a gap between the energy harvesting and the energy consumption which makes it hard to achieve self-powered IoNT systems. How to balance the use of energy so that it never exceeds the harvested energy in tiny IoNT nodes is a challenging problem to solve. Dealing with the finite amount of energy harvested at the nanoscale, the aim of this thesis is to design, develop and evaluate energy-neutral solutions for resource-constrained nanonetworks such as:

• To propose energy-neutral event monitoring solutions that allow the sensors to transmit event information (event type and its location) using only the amount of energy harvested from the events.

• To design pulses that will accurately convey both event type and its location to a distant receiver.

• To develop optimization models for pulse duration and pulse energy with the aim to improve the detection accuracy by minimizing the decoding error at the receiver

• The last contribution in this thesis is to design and evaluate the key component of an IoNT system, i.e., pulse generator, using COMSOL Multiphysics which is simulation software close to the physical reality. Chapter 1 Introduction 6

1.3 Contributions

Using the properties of the transmitted signal to identify node and event types remotely, this thesis conducts a numerical study of the events monitoring associated with resource-constrained nodes in self-powered nanonetworks within the IoNTs paradigm. Before highlighting the key contributions, we ﬁrst start with system model as follows:

• We started with event model to identify event types which can be detected using the energy released by the events. An example application of the energy-neutral is to monitor the events within a chemical production process [11, 27] and demonstrate the feasibility of using emitted energy as the signature for the event. The key idea is to place a monitoring node at a site to monitor the amount of energy released by the chemical reactions taking place at the site. If different chemical reactions release the different amount of energy, then, technically, it is possible to use the amount of energy released by a chemical reaction as a signature to distinguish different reactions. However, in order to extend the concept to multiple nodes, we need to ensure that each node is only monitoring the energy released locally at the site and is minimally impacted by the energy released at its neighbouring sites. This can be realized if the sites are sufficiently far apart from each other.

• We then detailed the structure of monitoring node which consists of two components: an energy harvester and a radio (pulse generator). The purpose of the energy harvester is to convert the energy emitted by the event into electrical energy. We found that there are nanoscale energy harvesters [20–25, 28, 29] that can scavenge tiny amounts of energy from the events in the ambient environment. As such, an energy harvester that harvests the emitted energy from the event can also serve as a sensor for monitoring such events Chapter 1 Introduction 7

• Next, we evaluated pulse transmitter model for monitoring node to generate a pulse using the harvested energy from an event. The purpose of the transmitter is to convert the electrical energy into a radio message. When an event occurs, the energy emitted by the event is harvested by suitable energy harvester and input to the pulse generator to generate a pulse.

• To enable communication among nanosensors, we have a communication channel between a node and a remote station where the transmitted pulse is attenuated. Recent studies conﬁrm that these sensors may be able to communicate in the THz band [6, 30]. We, therefore, reviewed widely used THz channel model for nanonetworks within the IoNTs paradigm.

• Events classiﬁcation is performed at the receiver. We, therefore, provided a detailed study of pulse detection models. We found that Continuous Time Moving Average (CTMA) detector can be used to detect peak power or pulse energy of received pulse [31].

• After building system model, we propose two energy-neutral event monitoring solutions that allow the sensors to transmit event information using only the amount of energy harvested from the events. We then design, evaluate these two implementation solutions for Energy-neutral (eNEUTRAL) IoNT. We also studied the decoding performances of both implementation options of eNEUTRAL IoNT.

The main contributions along with detail descriptions are summarized as follows:

• Single Pulse Transmission (SPT): SPT is the ﬁrst implementation solution of the eNEUTRAL IoNT that uses a single pulse containing the entire energy harvested from the event. However, it manipulates its pulse width (duration) to create a unique pulse amplitude for a given combination of event type and Chapter 1 Introduction 8

its location. This way SPT or simply single-pulse approach enables multiple nodes to communicate event type and its location to a distinct receiver by transmitting a single wireless pulse. However, single-pulse approach imposes the requirement that each node uses a particular pulse width (duration). Therefore, a major challenge is how to allocate pulse widths to multiple nodes to minimize the classification error. This way, we develop optimization model for pulse width allocation with the aim to minimize the classification error. We also study the relation between energy emitted by the events and the choice of pulse widths. From numerical simulation results, We find that single-pulse approach achieving 99% event type and location detection accuracy in 10-node network monitoring two different event types for a distance of 22 mm.

• Dual Pulse Transmission (DPT): In SPT, we manipulate the pulse width to create a unique pulse amplitude for a given combination of event type and its location. This results in the increasing number of classes to classify which is a major variable in the error probability. if we break down the number of classes into smaller class with two pulses then we can improve classification accuracy. We, therefore, propose Dual Pulse Transmis- sion (DPT) in which the nodes will have the same architecture to maintain the sensorless architecture but use a different mechanism for encoding and decoding. In the DPT or dual-pulse approach, the harvested event energy is divided into two pulses so that the energy of the first pulse uniquely defines a location and the second pulse uses the remaining energy to identify event types. We, therefore, encode the address and event type in the energy level of the transmitted pulse. We develop an optimization model for pulse energy allocation to optimize pulse energies. This optimal policy aims to improve the detection accuracy by minimizing classification error at the receiver. Chapter 1 Introduction 9

• Using extensive numerical simulation we evaluate the performances of both single and dual pulse transmissions. The parameters that are considered to evaluate the performances of both systems are the distance, number of nodes, node placement, number of event types, event energy spread and stochastic energy harvesting (energy fluctuations) of different events. We find that the dual-pulse approach significantly outperforms the single- pulse approach achieving 99% accuracy for detecting both location and event type in 10-node network monitoring two different event types for a radius of 30 mm.

• Pulse generation for tiny IoNT nodes: Due to the severe volume restrictions and low complexity of event monitoring nodes, the pulse generator should be kept simple and to generate THz signals without requiring complex structure. We, therefore, propose a graphene-based pulse generator to generate THz pulses. We studied different configurations which are used to generate SPPs on the metallic and graphene surface. The generated SPPs lead to femtoseconds long pulses in graphene. We note that the choice of the most suitable method depends on the frequency, the type of source, the material permittivity, and the system configuration. However, we find that our proposed generator can generate THz pulses directly on the graphene surface without adding any new circuit to event monitoring nodes. We model and evaluate the pulse generator in COMSOL Multiphysics which is close to the physical reality. We use near-field method to generate pulses in the THz band. We expect that simple structures are more likely to succeed in the future design of nano nodes. Chapter 1 Introduction 10

1.4 Dissertation Organization

The remainder of this thesis is organized as follows. In Chapter 2, we detail the state-of-the-art of the research related to this dissertation. We first overview the components of the network architecture of the Internet of Nano-Things. We then present the channel modeling and applications of nanosensor networks within the Internet of Nano-Things paradigm. Finally, we give separate related work sections for technical contributions so as to give the reader an appropriate background respective of that particular chapter. In Chapter 3, we first introduced the system model, in details, for the eNEU- TRAL IoNT framework. We then propose the first implementation solution of the eNEUTRAL IoNT framework that uses a single pulse containing the entire energy harvested from the event. We motivate the need of numerical investigation of pulse width allocation in two node scenario. We also study the impact of pulse widths on error probability. Finally, we investigate pulse width allocation and develop an optimization model to optimally allocate Pulse widths to multiple nodes with the aim to minimize the classification error. In Chapter 4, we introduce the second solution called dual-pulse transmission for the eNEUTRAL IoNT framework with the aim to improve the detection accuracy for longer distance. This approach uses a different encoding mechanism by encoding the address and event type in the energy level of the transmitted pulse. That is, the harvested event energy is divided into two pulses so that the energy of the first pulse uniquely defines a location and the second pulse uses the remaining energy to identify event types. To minimize classification error at the receiver, we develop an optimization model to optimize pulse energies and study the impact of address and event pulse energy on the classification error. In the end, we perform extensive numerical simulation and compare the performance and scalability of both solutions. We find that the dual-pulse approach significantly outperforms the single-pulse approach. Chapter 1 Introduction 11

In Chapter 5, we present a THz pulse generator based on near-ﬁeld excitation for severely resource constraint nodes. We model the pulse generator, using COMSOL Multiphysics. The proposed model is simple to construct without adding any new circuit and can generate pulses in the THz band. The thesis concludes in Chapter 6, followed by main future directions. Chapter 2

Background and the State of the Art

A nanosensor with limited sensing, restricted computational power, and highly constrained storage capabilities can sense, gather and share knowledge at the molecular level, empowering bottom-up control of many applications. It performs very simple computation, sensing and actuation tasks limited to close proximity. To increase their capabilities, these devices can be connected to perform collaborative tasks in a distributed manner and to send the sensed data to the external world for analysis [2] under the banner of the Internet of Nano- Things (IoNTs). Nevertheless, as IoNT is in an early stage of development and has not been successfully demonstrated yet, most work is focused on mathematical modeling, and some conceptual models have been proposed recently. IoNT is a new challenging paradigm for researchers with known nano-speciﬁc challenges as discussed in Chapter 1. For this purpose, a literature survey is car- ried out to identify recent trends and development in the area of IoNT. In this Chapter, we present the discussions in such a way as to lead the readers toward the main contributions of the thesis. We will start with the introduction of IoNTs, followed by the background of the basic building blocks of IoNT system. We will then present a few proposed applications with related work for nanosensor networks within the IoNT paradigm. Finally, we give separate related work sections for this thesis contributions so as to give the readers an appropriate background

12 Chapter 2 Background and the State of the Art 13 respective of that particular contribution.

2.1 Internet of Nano-Things

Internet of Nano-Thing is a new paradigm which connects nanosensors with existing communication networks, and eventually, the Internet which leads to the development of next-generation standard based on Internet of Thing (IoT) called the Internet of Nano-Thing (IoNT) [7]. IoNT will open new doors of research in the area of nanosystems, and nanocommunication to gather knowledge at an unprecedented depth and scale [32]. This development is due to the recent advances in nanotechnology which has provided nanoscale sensing and monitoring solutions to many real-world applications like biotechnology and biomedical, agriculture, and industry [2,4,8,9]. With the emergence of IoNT, researchers propose new communication stan- dards for nanoscale devices to communicate with each other in diverse applications. IoNT uses two broad areas of nanoscale communication such as nano- electromagnetic and molecular communication. Nano-electromagnetic communication is deﬁned as the as transmission and receiving of Electro-Magnetic (EM) radiation among nanoscale devices whereas the molecular communication is re- garded as the transfer of information using molecules and is therefore referred to as molecular communication [7,32]. The focus of this thesis is on electromagnetic communication.

2.2 Building blocks of IoNTs

The Internet of Nano-Thing architecture comprises three main components: nanosensor node, propagation channel and a receiver. In this section, we present the literature for each of these components. Chapter 2 Background and the State of the Art 14

A sensor node senses and monitors molecule level events. These sensed measurements are processes locally and are transmitted via the communication channel to a distinct receiver. A receiver is a macro scale device which receives the data for analysis. In what follows, we provide details about these components.

2.2.1 Nanosensor Node

In recent years, the nanotechnology advances made it possible to develop a new generation of smaller electronics up to a hundred nanometers. Similar to the macro-level sensor nodes, these nanoscale devices will have nanocomponents, including processing, sensing, transmitting, and power storage unit respectively as shown in Figure 2.1. Nano-processor Nano-transceiver Nano-antenna Nano-sensors Nano-memory Nano-actuator Nano-capacitor

Energy-harvester

Figure 2.1: An integrated nanosensor node [1].

Nano-processor

The nano-processor is the main component which drives all the onboard nano- electronics with the exception of the energy harvester and nanobatteries [33]. Due to the form factor of a sensor node, a nano-processor must contain an appropriate number of transistors to accomplish diminutive tasks. For example, the ﬁrst nano-processor called Nano-Sensor Data Processor (NSDP) with 4-bit data Chapter 2 Background and the State of the Art 15 processing is designed in [34]. This nano-processor is based on basic Processing Element (PE) which is a cell with 20 nm by 20 nm in size. Each cell is a square nanostructure with a quantum dot in each of the four corners where two electrons populated in antipodal sites due to the Coulomb repulsion. The proposed nano-processor is only 4-bit processor and can just handle basic tasks.

Nanosensor and Nanoactuator

Graphene and its derivatives have the potential to develop many types of sensors and actuators which can measure and analysis of different, unforeseen essential parameters and magnitudes right bottom at molecules level. These sensors can be used to measure mass, pressure, vibration, the concentration of a given gas or to detect lung cancer and asthma attacks [1]. Different types of chemical and biological nanoactuators have been reported in the literature based on the interaction between nanomaterials, electromagnetic (EM) fields and heat [1]. In this direction, a nanoactuator which is based on the magnetic nanoparticles can kill cancer cells by heating them [35].

Nano-transceiver

A sensor node requires a communication system to communicate the sensed events to a distinct receiver. Conventional antenna, of few centimeters, usually radiates at the Gegahertz (GHz) frequencies. However, scaling down the conventional metallic antenna to nanoscale requires high operating frequency (more than 100 THz) [36]. This approach is not feasible because, at so high operating frequencies, these nodes are expected to experience extremely high path loss and absorption which are not practical to operate for such resource-restricted nodes. On the other hand, designing nanoantennas using nanomaterials such as graphene, reduces the operating frequencies at few THz [6, 13]. Graphene has been proposed Chapter 2 Background and the State of the Art 16 as a building material for plasmonic nano-transceivers [13]. Graphene can support propagation of tightly conﬁned Surface Plasmon Polaritons (SPPs) in the terahertz band (0.1-10 THz) at room temperature [37], enabling the miniaturization of nanoantenna suited for wireless communication among nanoscale nodes. The nano-transceiver will generate surface plasmonic polariton (SPP) wave which leads to femtoseconds long pulse [12, 13, 18, 19]. The nanoantenna converts the SPP wave into EM waves and radiates it into the free-space [13, 18, 19].

Nano-memory and Nano-capacitor

Nano-memory is the data storage unit and plays a vital role in an electronic device because the device operations rely directly on the stored configuration parameters on the available memory. Several types of nano-memories based on different technologies with a length of a few hundreds of nanometers have been reported [33, 38]. A nano-capacitor, on the other hand, is the energy storage unit intermittently powering the different units of a sensor node. The nano-capacitor is connected to the electrodes of the energy harvester to get charged. The maximum energy stored in the capacitor depends on the capacitance of the capacitor, the area of the plates and voltage source through the well-known expression [33]:

1 E = CV 2 (2.1) max 2 c where C is the capacitance, and Vc is the voltage source. For electrostatic ultra- nano-capacitors with capacitance of 9 nF and voltage value, Vc, up to 0.4 V, the maximum energy can be store is approximately 800 pJ [12].

Energy harvesting unit

Due to the severely constrained area, sensor nodes are extremely power restricted. In recent years, researchers have developed diﬀerent techniques to harvest energy from the environment using specialized nanomaterials [39]. These techniques are Chapter 2 Background and the State of the Art 17 piezoelectric [20, 21, 40], thermoelectric [28], triboelectric [22–24] and pyroelectric [25, 41]. in what follows, we provide detail literature of these techniques:

• Piezoelectric: Piezoelectricity, also called the piezoelectric eﬀect, is the ap- pearance of an electrical potential across the sides of certain materials when they are subjected to mechanical stress. At the nanoscale, it is called nano- piezoelectricity. The discovery of piezoelectric nanomaterials provides a new opportunity to develop nanoscale energy harvesters called piezoelectric nanogenerators [20]. The nature of piezoelectricity comes from the non- centrosymmetricity in the crystal [39]. There are 32 crystal classes, and 20 of them exhibit the piezoelectric eﬀect. These materials include Lead zir-

conate titanate (PZT ), barium titanate (BaT iO3), zinc oxide (ZnO), gallium nitride (GaN), zinc sulﬁde (ZnS) and many more [39]. The authors in [21] show that using piezoelectric zinc oxide nanowire arrays, nanoscale mechanical energy can be converted into electricity. In particular, they ﬁnd that a single zinc oxide of diameter 20 nm with the length of 200 nm can produce power up to 0.5 pW at one cycle of resonance [21]. Similarly, ver- tically align zinc oxide nanowire can generate 1.1 pW/µm3 [42].

• Thermoelectric: In this method, the temperature gradient is converted into electricity using Seebeck’s effect. The Seebeck effect, named after the Baltic German physicist Thomas Johann Seebeck, is the conversion of heat flow into power at the junction of two dissimilar electrical conductors [28]. This type of energy harvesting is feasible for portable and pervasive computing devices, and in environments where thermal gradients exist [29]. In particular, human body heat can be converted into electricity using the temperature gradient between the body temperature and the external medium. However, low gradient and limited heat transfer can affect the output power efficiency [28]. Chapter 2 Background and the State of the Art 18

• Triboelectric: The triboelectric also known as the triboelectric effect is a type of contact electrification in which specific materials become electri- cally charged, due to electrostatic induction, after they come into frictional contact with a different material. Rubbing glass with fur, or a plastic comb through the hair can produce triboelectricity. Any materials which exhibit the triboelectrification effect, from metal, to polymer, and to silk can be candidates for fabricating Triboelectric Nanogenerator (TENG). However, the ability of material for gaining/losing electron depends on its polarity. For instance, the organic and inorganic films that exhibit opposite tribo- polarity are used to generate the triboelectricity [22]. TENGs can be used to harvest vibration energy [23], and to convert magnetic force variation to electricity [24]. It is further observed in [43] that using TENG, a power density of 2.04 mW/cm3 equivalent to 0.002 pW/µm3 can be achieved by using micro/nano dual-scale polydimethylsiloxane.

• Pyroelectric: A pyroelectric nanogenerator is an energy harvesting device which converts the time-dependent temperature ﬂuctuation into electricity by using nano-structured pyroelectric materials. Unlike thermoelectric, pyroelectric materials do not need a spatial gradient, but it requires temporal temperature changes [28]. Likewise, a pyroelectric nanogenerator made from a single nanowire of zirconate titanate can be used as a temperature

sensor for detecting the change in temperature [25]. Using BaT iO3 ﬁlm of 200 nm thick in pyroelectric nanogenerator, output power of 3 pW/µm3 can be generated [44]. However, in pyroelectric nanogenerator, the amount of harvested power is proportional to the rate of temperature change, which makes it directly useful for nanoscale systems [29]. For example, pyroelectric nanogenerator can be used to power nanosensor nodes deployed in catalyst sites within chemical reactors [11]. Chapter 2 Background and the State of the Art 19

So far we discussed diﬀerent components of a nanosensor node. However, at the nanoscale, the amount of power harvested is many orders of magnitude lower than the power consumption of nanosensors [9, 11]. Table 2.1 shows that the power consumption of the nanosensors is many orders of magnitude higher than what is possibly be harvested by diﬀerent nanoscale energy harvesters. This means that it is not possible to use the harvested energy to drive directly a nanoscale device that includes sensors, processors and wireless communication. Recent work [11, 27] advocates that a node should be composed of as few components as possible. This dissertation, therefore, motivates the need for sensorless event monitoring for IoNT systems which we will discuss in Section 2.4.

Nano sensors Power consumption Energy harvest- Harvested ing options power (pW /um3) Piezoelectric 0.5 [21], 1.1 Hydrogen sensor 1nW [45], 0.1 uW [46] [42] Pyroelectric 0.1 [50], 3 Pressure sensor 1 nW [47], 1 uW [48] [44] Triboelectric 2.1 [43] Temperature sensor 1 nW [49] Megnetoelectric 4.5 [51]

Table 2.1: Power consumption of diﬀerent types of nanosensors versus power harvested at nanoscale.

2.2.2 Propagation Channel

The propagation of sensed information from a sensor node to a receiver, within the Internet of Nano-Things paradigm, is aﬀected by the channel chemical compositions. Recent studies conﬁrm that these nodes may be able to communicate in the terahertz band using graphene as a transmission antenna [6]. The propagation model for THz band communication among sensor nodes is introduced in [52,53] which is based on the radiative transfer theory. In this section, we give details of the THz propagation model. Chapter 2 Background and the State of the Art 20

As terahertz band is the resonance frequency of molecules, communication in this band is severely affected by attenuation and molecular absorption noise. Radio communication is influenced by the chemical compositions of the medium in two different ways in the terahertz band. First, the radio signal is attenuated because molecules in the channel absorb energy in certain frequency bands. Sec- ond, this absorbed energy is re-radiated by the molecules which creates noise in the channel called molecular absorption noise. The attenuation in the terahertz band comes from the two factors which are spreading loss and molecular absorption loss. The spreading loss, which is the function of distance, comes from the signal propagation through channel whereas the molecular absorption loss comes from the molecular absorption of energy as molecules in the channel absorb energy in certain frequency bands. We assume that the radio channel is a medium consisting of X chemical species [x1,x2,.....xN ].

The effect of each chemical species xi on the radio signal is characterized by its molecular absorption coefficient Kx(f ) of species xi at frequency f . The molecular absorption coefficients of many chemical species are available from the HIgh resolution TRANsmission molecular absorption database (HITRAN)[54]. Each ∈ type of molecule xi (xi X) has mole fraction mx in the medium. The medium absorption coefficient K(f ) at frequency f is a weighted sum of the molecular absorption coefficients in the medium:

Xn K(f ) = mxKx(f ) (2.2) x=1 where Kx(f ) is the absorption coeﬃcient of individual molecule species at the frequency f . Figure. 2.2 shows the absorption coeﬃcient of standard air with mean latitude in summer over the Terahertz band (0.1-10 THz) which is equivalent to 1 wave number from 3.3 to 330 cm− . The attenuation at frequency f and a distance d from the radio source is given by [11, 53]: Chapter 2 Background and the State of the Art 21

Figure 2.2: Interface of the HITRAN on the web tool. Molecular absorption coeﬃcient can be calculated directly from HITRAN database.

4πf d 2 A(f ,d) = ∗ eK(f )d (2.3) c where c is the speed of light. Figure. 2.3 shows path loss in dB at diﬀerent distances in THz band At the nanoscale, the thermal noise is very low [1]. This is because the scattering of electrons in nanomaterial creates very low thermal noise. Thus the only noise which aﬀects the communication between sensor nodes is the molecular absorption noise of the channel. The molecular absorption noise, Nabs(f ,d), which is due to the re-radiation of absorbed energy in a random direction by the molecules in the channel, is given [11, 12, 52, 55]:

− K(f )d Nabs(f ,d) = KBT0(1 e− ) (2.4) where T0 is the reference temperature 296K, and KB is the Boltzmann constant. Figure. 2.4 shows molecular Absorption Noise Power Spectral Density (PSD) at diﬀerent distances in the THz band. Chapter 2 Background and the State of the Art 22

Figure 2.3: Total path-loss in dB of THz band at diﬀerent distances.

Let U(f ) be the power spectral density of the transmitted radio signal at frequency f . The signal-to-noise (Signal-to-Noise-Ratio (SNR)) at frequency f and distance d is computed as follows [30].

U(f ) SNR(f ,d) = (2.5) A(f ,d)Nabs(f ,d)

2.2.3 Receiver

For any communication system, a receiver is a key component which receives sensed measurements and decodes the information sent by a sensor node. In this Section, we give details of the state of the art receiver designs which are energy based detector and CTMA (Continuous Time Moving Average) based detector.

Energy-based Detector (ED)

Energy-based Detector (ED) is a simpler circuit, low cost, but it has low sensitivity because it also captures the noise (see Figure 2.5). This is because it integrates Chapter 2 Background and the State of the Art 23

Figure 2.4: Molecular Absorption Noise PSD in dB/Hz at diﬀerent distances in THz band. the signal energy in a single interval of time which is usually greater than the pulse duration . This signiﬁcantly lowers the performance of the receiver since the useful signal power is also averaged with respect to the extra noise power as shown in Figure 2.6.

Band Pass 2 Classifier Filter ( )

Figure 2.5: Block diagram of energy based detection where the pulse energy is detected by integrating the received energy over an Integration window (Tint). Chapter 2 Background and the State of the Art 24

T int

Figure 2.6: In Energy detection, the integration window (Tint) is greater than the pulse width (duration) so the additional noise is also integrated.

CTMA based Detector

The schematic diagram of the CTMA-based detector is shown in Figure 2.7 where the received signal is passed through the bandpass filter to filter the noise outside the THz band. The filtered signal is then squared and then passed through continuous-time integrator which is usually approximated using second order low pass filter [56]. The output of the low pass filter is treated as a continuous time function and input to the peak detector. The peak detector finds the maximum signal energy and treated is an observation for decoding the information sent by the sensor node.

Band Pass 2 CTMA PD Classifier Filter ( )

Figure 2.7: Block diagram of a CTMA based detection to detect either peak power (pulse amplitude) or pulse energy.

CTMA based detector is more robust, and it outperforms the energy-based detector concerning sensitivity [31, 56]. In this dissertation, we, therefore, use the CTMA based detector to detect either peak power (pulse amplitude) or pulse Chapter 2 Background and the State of the Art 25 energy to distinguish the pulses being sent by the monitoring nodes.

2.3 Applications

Sensor nodes are typically used to carry out reasonable functions. These nodes sense and monitor cell level activities and events. Recently, the area of nanoscale communication has attracted the research community to design new algorithm and frameworks for IoNT applications tailored to the peculiarities of the THz band. Such developments have shown drastic eﬀects in monitoring molecules level events [2]. In what follows, we provide some envisaged applications of IoNTs as illustrated in Figure 2.8.

2.3.1 Medical Applications

In the health care system, a sensor node can communicate with a micro-scale device. The sensor node processes the data locally and then can transmit it to the user device for analysis [57]. In this direction, there are significant relevant developments in recent years. For example, a nanorobot that can be introduced into the human body to detect the tumor without causing injury to the patients [58]. They propose an acoustic communication paradigm, NanoBee, in which the nanorobots communicate through acoustic signals [58]. Similarly, researchers at Imperial College of London create robotic pills which have the capabilities to deliver drugs inside complex human body parts, i.e., small intestine where it is difficult for doctors to reach and treat it. The body of such a capsule is designed to have a tiny camera, a wireless chip, and a remote controller [59]. However, there are no specifications given about how to control and position the tiny robotic pills, but remote communication can be a potential solution. Once a nanorobot is deployed inside the human body then they need to prevent itself Chapter 2 Background and the State of the Art 26

(a)

(b)

(c)

Figure 2.8: Network architecture for the IoNT: (a) human lungs monitoring [2] (b) monitoring chemical reactions [3] and (c) plant monitoring [4]. Chapter 2 Background and the State of the Art 27 from the immune system through a chemically inert diamond exterior [58]. Fur- ther, in [60] the authors, inspired by biological systems, propose a nanoparticle system that can target the tumor using molecular pathways. They investigate that such nanoparticle system, with the potential of biological signaling and receiving modules, can communicate information through molecular channels and can target 40 times higher doses of chemotherapeutics to tumors. These nanoparticles can be attached to the cell walls by using some bioengineering techniques such as atomic force microscopy or some form of artificial bacteria [60]. In [61] the authors investigate THz communication in human tissues and measure the absorption path loss of skin tissues. They further propose the propagation model for THz communication in vivo and derive the channel capacity and transmission range for different communication schemes. In the health-care system within IoNT paradigm, the Body Area Network (BAN) can be connected to in-body network [62] where nanomachines patrol in the body to take measurements and send it to the user device. They further investigate application requirements (functional and non-functional) for in-body networks. This propose is somehow different from the BAN that measure all kinds of body parameters from outside using wearable devices. In [63], the authors propose a Bioresorbable Electronic Stent (BES) that is integrated with therapeutic nanoparticles. For data communication, in-vivo and ex-vivo experiments are conducted between stent antenna (5 mm in size) and transmitting antenna (900 MHz and 20 mm in size). Interestingly enough, they show that 10 mW power can be transferred to the stent antenna when the incident power of the transmitting antenna is 1 W. Though the authors experimentally indicate in-vivo communication, however, it can be further extended if molecule level measurements are reported to the external world. This way, it can help to detect diseases at the early stage. These sensors, deployed within a health-care system, can be powered using an onboard energy harvester. Assuming a blood flow scenario, an estimated power of 1.28 pW can be generated when ZnO nanowires bend by a flow movement [33]. Chapter 2 Background and the State of the Art 28

2.3.2 Industrial Applications

Nanoscale communication can transfer a wide range of industrial applications by providing new solutions, easing the manufacturing process and enabling quality control procedures. For instance, nanosensors can be used to develop the touch ultrahigh sensitivity surfaces and haptic interface [1]. They can also be embedded in advanced fabrics to improve the air flow in the structures. The second most promising industrial application of IoNT paradigm is real-time monitoring and controlling chemical processes at the molecular level [2,10]. The authors propose network architecture for chemical reactors and analyse the reliability of terahertz band nanoscale communication [64]. Similarly, the challenges of attenuation and molecular absorption noise, within the terahertz band, are investigated in [30] by introducing the concept of frequency switching based on the predicted composition of the medium over time [30]. In [10], the authors propose a novel concept to monitor chemical reactions and also improve the product selectivity of Fischer Tropsch (FT) catalysis. One of the major issues of nanoscale devices is the limited harvested power. Keeping this constraint in mind, a new self-powered sensing and communication architecture is proposed in [9]. The proposed architecture uses pyroelectric nanogenerator fitted in sensor nodes at catalyst sites where it harvests power from temperature fluctuations on the catalyst surface.

2.3.3 Environmental Applications

Nanotechnology can revolutionize the agricultural and food industry with new advancements in the molecular treatment of diseases and detection [65,66]. Nanonet- works can help in detecting toxic components and bacteria at the molecular level that cannot be detected using traditional sensing technologies. It can also be used to monitor plants processes such as plant emissions, humidity and temperature [1, 67]. As it has been reported in the literature, that plants emit various Herbivore-induced Plant Volatiles (HIPVs) with diﬀerent concentrations based Chapter 2 Background and the State of the Art 29 on the insect type [67, 68]. In this direction, a nanonetwork within the context of IoNT, deploy in agriculture ﬁelds can monitor chemical compounds that are being realized and exchange between neighbor plants. Detecting such processes can help to reveal their life cycle and change patterns. It can also improve the agricultural systems, productivity by identifying the timing of an insect attack and the type of the attacking insect [67, 68].

2.4 Sensor powering challenges

Power supply at the nanoscale is a challenging problem. As it has been discussed previously that nanoscale energy harvesters (a.k.a nanogenerators) [20–25] can not generate enough power to transmit a large number of pulses uploading event information and sensor node identiﬁcation each time events are detected in the environment. How to balance the use of energy so that it never exceeds the harvested energy in tiny IoNT nodes is a challenging problem to solve. Therefore, novel energy-neutral event monitoring solutions are required. A key feature of the energy-neutral system is that it performs event monitoring without using sensors, which helps reducing power consumption of the system. There are other examples of monitoring without using sensors. These examples appear at both macro and nanoscale levels. At the macro level, machine status is monitored using power consumption of machine as signature [69, 70]. Similarly, in [71], variation in the temperature is used as the signature to control the internal temperature of the induction motors. Recently an energy-neutral IoT system is proposed in [72] to monitor pollution using photovoltaic energy harvested from the mini photovoltaic panels. Similarly, piezoelectric vibration energy harvester can be used as a signature to recognize human activity by observing the generated AC voltage [73, 74]. The authors in [75] proposed packet-less pulse switching paradigm for event sensing using Ultra Wide Band Impulse Ra- dio as the physical layer in sensor networks. The proposed concept addresses the Chapter 2 Background and the State of the Art 30 capacity and energy overheads of packet-based communication. Such a paradigm uses MAC-Routing frames to synchronize nodes with a receiver, however, events are localized by a receiver by observing the temporal position of a received pulse to a reference frame structure [75]. Recently, packet-less self-powered event monitoring ultrasonic pulse-based system is proposed in [76]. In the proposed method, a large number of sensor nodes are deployed on a plate-like structure which is being monitored. These sensor nodes, which form a Cellular Pulse Net- working (CPN), sense their local environment and communicate the event information to a base station via ultrasonic pulse links using harvested energy from the ambient vibration. Similar to work in [76], a Scalable Cellular Pulse Network- ing (SCPN) architecture is presented in [77] which retains the energy beneﬁts of CPN while lowering the delivery delay for a larger area being monitored. These studies focus on IoT systems, hence, do not provide solutions to detect events in IoNT systems. Interestingly, energy harvester can also be used as a sensor at the nanoscale.

For instance, a TENG which is based on the contact electrification effect, has been used as a gas sensor [78]. Similarly, the correlation between the output voltage and temperature of nanogenerator can play the role of a temperature sensor [25]. However, these studies do not provide solutions to sense and monitor events remotely because they do not cover the communication aspect. To power a tiny sensor node, there are two possible options. These two solutions are battery and energy harvesting. As batteries are difficult to build into and replace in such tiny sensors, researchers are developing nanoscale energy harvesters (a.k.a nanogenerators) [20–25], that can scavenge tiny amounts of energy from the ambient environment to power event detection and very basic wireless communication where all data is transmitted as a series of extremely short (a few hundred femtoseconds) pulses [19]. However, there is a mismatch between energy harvesting Chapter 2 Background and the State of the Art 31 and energy consumption of a nanonode [11, 18]. This means that it is not possible to use the harvested energy to drive a nanoscale device that includes sensors, processors and wireless communication. To overcome such problem, recent work proposed a sensorless remote event monitoring framework known as SE- MON [9, 11]. The Sensorless Event Monitoring (SEMON) architecture consists of a node and a remote station. The node is located at the place where events occur. The node consists of only two components: an energy harvester and a radio. When an event occurs, the energy emitted by the event is harvested by the energy harvester on the node. This harvested energy drives the radio to produce a pulse whose energy is equal to the amount of harvested energy. At the remote station, the receiver picks up the radio signal and uses an energy detector to measure the amount of received energy. Assuming that different events generate different amount of energy, the remote station can, therefore, use the received energy to identify various events. The limitation of the SEMON framework is that one remote station can only communicate with one node. This is due to the fact that the energy of the received pulse is determined entirely by the event and does not contain information on the identity of the node. This dissertation addresses the powering issue of IoNT system and studies two methods for energy-neutral IoNT, called eNEUTRAL IoNT that allow the sensors to transmit event information (the event type and its location) using only the amount of energy harvested from the events.

2.5 Pulse generation for tiny IoNT nodes

The key component of the event monitoring node is the pulse generator. There are other conﬁgurations which are used to generate SPPs on the metallic and graphene surface. These SPPs lead to very short pulses, just several tens of femtoseconds long. Therefore, in what follows, we provide detail literature of conﬁg- urations that are used to generate SPPs as shown in Figure 2.9. At the end of this Chapter 2 Background and the State of the Art 32 section, we will provide a discussion to elaborate the need for designing a simple model to generate THz pulses suitable for Tiny IoNT nodes.

Kretchmann Configuration [87, 88] Attenuated Total Reflection Otto Configuration [84, 86, 87]

SPP Grating Coupling [80, 84, 85] Generation Methods Electron beam [82, 83]

Electrical Excitation [13, 81]

Optical Excitation [81]

Tip Method [79, 80]

Figure 2.9: Generating SPPs using diﬀerent excitation methods.

Kretschmann-Raether Conﬁguration

The Kretschmann-Raether method is an Attenuated Total Reflection (ATR) method, where Total Internal Reflection (TIR) takes place [88]. In this method, a metallic film is sandwiched between a dielectric, usually a prism, and the air [88]. Light is incident from the dielectric side and SPPs are excited on the metal-air interface as shown in Figure 2.10a). The parallel to interface component of incident light momenta (kx) couples to the SPP wave number (ksp) giving rise to SPP that which propagates along the interface (x direction) as it is shown by Figure 2.10a). A restriction that Kretschmann-Raether method imposes is that it can be applied only for very thin metallic films up to 70nm. Chapter 2 Background and the State of the Art 33

Otto Conﬁguration

Otto method is similar with Kretschmann-Raether conﬁguration but it follows diﬀerent setting of metal and dielectrics, that is, an air gap is located between a metal and a dielectric (prism) [84, 86]. The incidence light is projected from the prism side and SPP wave is excited on the air-metal interface [87] as shown in Figure 2.10b). Likewise, in Kretschmann-Raether method, the parallel to interface component of the momentum of light must couple with the momentum of surface plasmon on the air-metal interface. Due to subwavelength character of

SPP [80], the wave number kx of incident EM wave is smaller than that of SPP wave ksp, that is, a denser medium is used to bring the coupling condition between the light and SPP wave number, since the kx takes higher values in dielectric than in light as Eq. (2.7) states. The coupling condition for surface plasmon waves in the metallic ﬁlm [80] is given by

kx = ksp, (2.6) where the parallel to interface wave number component of incident light is

√ kx = k0 εd sin(θi) (2.7) and the SPP wave number is given by [84, 85]

r εmεd ksp = k0 (2.8) εm + εd where k0 = ω/c is the wave number in free space, ω is the operation angular frequency, c is the light velocity in vacuum, θi is the incidence angle of light wave and εd, εm are the dielectric permittivity for dielectric and for metallic ﬁlm, respectively. The SPP dispersion relation is given by the equation (2.8) at a Chapter 2 Background and the State of the Art 34 metal-dielectric interface [80] where the metal permittivity is frequency dependent (εm(ω)). The Otto method as well as the Kretschmann-Raether, work on ATR conﬁgu- ration where TIR happens on the dielectric interface. That is, the EM wave comes from a denser medium with an angle greater than critical angle [89]. The SPP resonance is sharp and sensitive to the coupling condition and incidence angle. This method is useful when direct contact with the metal surface is undesirable, for instance when sensing molecular absorption of the surface is required.

a) b)

Incident Beam Reflected Beam Incident Beam Reflected TM wave TM wave Beam

(εd) (εd)

Denser dielectric Denser dielectric

Metal (ε2) Dielectric: Air (ε2) SPP SPP

Dielectric: Air (ε3) Metal (ε3)

Figure 2.10: Schematic representation of Attenuated Total Reﬂection (ATR) methods which are used for SPP excitation: (a) Kretschmann -Raether and (b) Otto Conﬁguration.

SPP excitation by grating coupler

A grating coupler is formed by periodic slits or grooves on a metallic surface with lattice constant (grating period) a [80, 84]. In a grating coupler, the momentum of incidence EM wave is coupled with the wave numbers of SPP (ksp) resulting to

SPP excitation [84,85]. The wavenumber kd of the incident light, which is always smaller than ksp, is being increased by an amount proportional to the reciprocal wavenumber of grating kg = 2π/a, until the satisfaction of the coupling condition Chapter 2 Background and the State of the Art 35

± ksp = kd sinθi kgν, (2.9) √ where kd = k0 εd the wavenumber of the incident light in a dielectric environment with permittivity εd, and ν is an integer. The Eq. (2.9) states that when an incidence wave hits a grating surface, then ± the in-plane wave number (kd sinθi) receives an additional momentum of kgν leading to grating coupling. To achieve a successful grating coupling and SPP excitation, we have to tune correctly the parameters of the system, namely θi, a and εd. This method limits the dispersion distribution of the excited SPPs to a very narrow frequency range [90] as well as it requires very sensitive tuning and geometry restriction for meeting the coupling condition.

SPP excitation by electron beam

Exciting SPPs using electron beam is a diﬀerent method than using EM radiation. In this method, a beam of electron propagates parallel to a metallic sheet. The phase velocity of the SPP wave must couple with that of the electron beam. It is further investigated in [82] that this condition holds only when periodic dielectric slits are introduced in a substrate and eventually the evanescent wave of the electron beam couples to SPPs. This method requires a continuous electron beam with tunable energy and periodic slits with the proper period.

Optical Excitation

As it has already been mentioned, graphene provides the opportunity to enable communication at the nanoscale in the terahertz band [1,6, 13]. In optical excitation method, graphene plasmonic antenna is fed with a photoconductive source, which is excited on its turn by an optical laser operated in the pulsed mode as reported in [91]. This way, SPPs are excited in the THz band when an external EM source with femtosecond pulse excites SPP on the interface between graphene sheet and dielectric [81]. The main intuitive in this method Chapter 2 Background and the State of the Art 36 is when a biased semiconductor is excited by a laser pulse with photon energy greater than its bandgap then the SPP waves are generated at the interface between the graphene layer and the dielectric material.

Electrical Excitation

In the electrical excitation mechanism, an induced electric voltage (DC) is applied between the source and the drain of a High Electron Mobility Transistor (HEMT). Using this process, a plasma wave is generated, through the Dyakonov- Shur instability, which, in turn, is used to excite a propagating SPP wave at the interface with the graphene layer [81]. In this conﬁguration, SPP wave is generated indirectly by ﬁrst exciting 2D plasmons in the HEMT channel which in turn couples with plasmons in the graphene surface to generate SPP wave.

Tip Method

The tip is a near-field excitation method which uses a nano-prob to excite SPP on the graphene surface. In this method, a short tip act as a point source with dimension d u 25nm is illuminated using infrared beam [79]. Due to the small aperture size of the tip, the EM radiation ensuing from the tip with very high wavenumbers k. Hence, the matching condition between k and ksp is naturally satisfied, and subsequently, SPP waves are excited [80]. Near-field excitation method is easy, robust and used for local excitation of SPPs in several plasmonic systems [79, 80, 92, 93].

Hence, most of the existing configurations require complex structures including the prism, periodic slits and special circuits (i.e., High Electron Mobility Transistor) which may be difficult to implement in severely resource constraint nodes. We will show in Chapter 5 that pulses in the THz band can be generated using the near-field method. Near-field excitation method is used for gate tuning of graphene plasmons in infrared regime [79] and to improve the propagation Chapter 2 Background and the State of the Art 37 length of SPPs in ultraviolet regime [92]. However, using COMSOL Multiphysics, we will show that this method can generate SPPs in the THz band without employing any extra circuits which are usually required in existing methods. Chapter 3

Energy-neutral Single Pulse Transmission

In this Chapter, we propose Single Pulse Transmission (SPT) method for energy- neutral event monitoring that allows the sensors to transmit the complete information about a detected event, including the event type and its location, using only the amount of energy harvested from the events. Since in many application domains, different types of events discharge different amounts of energy to the environment, which technically makes it feasible to identify events from their emitted energy. As such, a nanogenerator that harvests the emitted energy from the event can also serve as a sensor for monitoring such energy emitting events. The remaining challenge is to design pulses that will not only consume exactly the amount of energy harvested from the event but also accurately convey both event type and its location to a distant receiver. The proposed solution uses a single pulse containing the entire energy harvested from the event but manipulates its time duration to create a unique pulse amplitude for a given combination of event type and its location. To minimize classification error at the receiver, we optimize pulse width or durations in the single-pulse.

38 Chapter 3 Energy-neutral Single Pulse Transmission 39

3.1 Introduction

In our proposed single-pulse solution of the eNEUTRAL IoNT, the harvested energy is directly used to power the transmitter to transmit a pulse of proportional amplitude. The recent studies confirm that an efficient way is to use harvested energy directly for node operations [94, 95]. There are other examples of using harvested energy directly to transmit wireless messages [96,97]. For instance, the push of a key/button can be used to deform a piezoelectric material, thereby generating electrical energy which can be used directly to send a short wireless message [96]. Similarly, the harvested energy, extracted from walking, can be used directly to transmit RFID signals [97]. Such an approach can avoid the energy loss due to leakage and self-discharge of batteries and supercapacitors [94]. Our proposed solution can be considered as Harvest-Use approach reported in [95]. However, at the nanoscale, the amount of power harvested is many orders of magnitude lower than the power consumption of nanosensors [9,11]. This means that it is not possible to use the harvested energy to drive directly a nanoscale device that includes sensors, processors and wireless communication. To overcome such problem, sensorless remote event monitoring framework known as SEMON is proposed in [9,11]. As it has been discussed previously, the limitation of SEMON is that it cannot identify the location of nodes since one remote station can only communicate with one node. This is because the remote station uses energy detector, and the energy of the received pulse is determined entirely by the event and does not contain information on the identity of the node. This chapter studies the single-pulse approach to overcome this limitation of SEMON. The key idea of the proposed method is to encode the node identity in some attributes of the pulse, e.g., pulse width. By width, we refer to the time duration of the pulse. In particular, the nodes will have the same node architecture proposed in [11] to maintain the sensorless architecture. There are many Chapter 3 Energy-neutral Single Pulse Transmission 40 other attributes which can be used to convey event information such as amplitude, width, energy, and position of the pulse. However, we choose pulse width to encode the node information since it can be controlled using appropriate electronics (see Section 3.2.3 for more details). Feasibility of eNEUTRAL IoNT is demonstrated using extensive numerical experiments involving terahertz channels. We find that single-pulse approach achieving 99% event type and location detection accuracy in 10-node network monitoring two different event types for a distance of 22 mm. The rest of the chapter is structured as follows. We detail the system model in Section 3.2. We propose the Single Pulse method of eNEUTRAL IoNT and their optimization in Section 3.3 followed by their numerical evaluations in Section 3.4. We summarize the chapter in Section 3.5.

3.2 System Model

This section describes the architecture and components of the eNEUTRAL system. The eNEUTRAL system consists of two types of nodes: the monitoring nodes and a remote station (sink or base station). The purpose of the monitoring nodes, which will merely be referred to as nodes, is to convert the energy released by an event to a radio message. This radio message, which we will see in Section 3.3, consists of one radio pulse. The aim of the radio message is to allow the remote station to infer the identity of the sending node as well as the type of the event. The following assumptions are considered in our system design and analysis:

• We assume that events of interest would release some energy.

• We also assume that diﬀerent types of events would generate diﬀerent amounts of energy. However, actual energy emitted by a particular event type could Chapter 3 Energy-neutral Single Pulse Transmission 41

ﬂuctuate slightly. For example, it was found that chemical reactions release energies which are not 100% deterministic but vary with slight ﬂuc- tuation [11].

• Only one node can harvest the released energy from an occurred event. That is each node is only monitoring the energy released locally at the site and is minimally impacted by the energy released at its neighbouring sites. This can be realized if the sites are suﬃciently far apart from each other.

• The energy of the radio pulse is equal to the energy harvested by the node. There are other examples of using harvested energy directly to transmit wireless messages [96,97]. Energy loss happens in leakage and self-discharges of storage units including batteries and capacitors [94]. To avoid energy loss, the proposed framework can follow Harvest-Use architecture [95] to transmit a signal using the harvested energy from the event.

• We consider a single hop communication between nodes and a remote station.

• We assume that nodes transmit at diﬀerent time slots be using random time delay at each node.

• There are two main models for molecular absorption noise which is due to the re-radiation of absorbed radiation by the molecules in the channel. The first noise model assumes that the magnitude of the noise is not influenced by the amplitude of the transmitted power [53, 98], whereas the second model assumed that the intensity of the transmitted power influences the noise [12, 14]. Note that neither of these noise models has been experimentally tested due to unavailability of nanoscale transceivers yet. However, in this thesis, we use the first noise model which is widely used in literature. Moreover, no internal noise inside the node is considered. Only the THz channel noise has been taken into account in the numerical analysis. This Chapter 3 Energy-neutral Single Pulse Transmission 42

is because at the nanoscale, the thermal noise is very low [1] and there is no noise characterization available for graphene technology.

• We consider a circular deployment where all nodes are equidistant from the remote station. An example of such a network is the cylindrical deployment of nanosensors for intra-body disease detection [8]. Similarly, a remote station can also be placed at the center of the catalyst tube, surrounded by nodes, in a chemical reactor [11].

3.2.1 Event Model

The purpose of the eNEUTRAL system is to detect events by using the energy released by the events. An example application of the eNEUTRAL is to monitor the events within a chemical reactor. This application was described in recent work [11, 27]; however, the earlier framework has the limitation that one remote station can only receive information from one node because the radio message sent by a node contains only information on event type but does not identify the node. This work resolves this limitation so that multiple nodes can communicate with a remote station by including node identity in the radio message. There are events that emit the diﬀerent amount of energy when occurs. For example, in biomedical application, human body provides numerous potential energy sources; a blood ﬂow (pressure) [33], cardiac motions [99], and lung and diaphragm motions [100] are all power sources and driving events. Such conditions within the human body can be monitored directly from the harvested energy. For example, a piezoelectric nanogenerator attached to eyelid can harvest power from eyeball movement [101]. The motion of the eyeball is associated with one’s sleep pattern. Slow and high motion, respectively, of an eyeball, produces distinct output current and such current can be used as a signature to detect the valuable signs, associated with tiredness as well as brain activity [101]. Chapter 3 Energy-neutral Single Pulse Transmission 43

The most promising industrial application of IoNT paradigm is real-time monitoring and controlling chemical processes at the molecular level [2,10] where different events (reactions) take place at the catalytic site which produce different amount of heat energy. An example of the chemical reactor described in [11, 27] consists of a solid catalyst which is composed of discrete catalytic sites. The key idea is to place a monitoring node at a site to monitor the amount of energy released by the chemical reactions taking place at the site. If different chemical reactions release the different amount of energy, then it is possible to use the amount of energy released by a chemical reaction as a signature to distinguish different reactions. The papers [11, 27] used a real-life chemical production process as an example and demonstrated the feasibility of using emitted energy as the signature for the event.

3.2.2 Structure of the monitoring node

Since the amount of energy that a nano-scale node can harvest is severely limited, recent work [11, 27] advocates that a node should be composed of as few components as possible. The structure of the eNEUTRAL monitoring node is shown in Figure 3.1.

Figure 3.1: Illustration of eNEUTRAL event monitoring node which includes an energy harvester and a radio.

The sensor node or simply a node consists of two components: an energy Chapter 3 Energy-neutral Single Pulse Transmission 44 harvester and a radio (pulse generator) module. The purpose of the energy harvester is to convert the energy emitted by the event into electrical energy. Many types of nanoscale energy harvesting methods have been developed for harvesting diﬀerent types of energy, e.g. piezoelectric [20, 21, 40], thermoelectric [28], triboelectric [22–24] and pyroelectric [25, 41] as discussed in Chapter 2. The purpose of the pulse generator or radio module is to generate a radio message. An example of a radio message is a radio pulse. As it has been discussed previously that a pulse contains many parameters: amplitude (i.e., power level), pulse width (duration), transmitted energy (which equals to the square of the amplitude times pulse width) and pulse position. Therefore it is possible to use a pulse to convey two pieces (address and event) of information. In the proposed single-pulse scheme (see Section 3.3) each node converts the energy emitted by the event into one radio pulse. This means the energy of the pulse contains information on the event that has occurred because each event emits a distinct amount of energy. However, for the remote station to determine the identity of the sending node, the node also has to encode its identity in the radio pulse being sent. The proposal of this scheme is that each node will use a diﬀerent pulse width to encode its identity in the width of the radio pulse. We assume that such an encoding mechanism is hard-wired into the nodes when they are manufactured. In the next few subsections, we will provide background information on the pulse transmitter model, communication channel and pulse receiver model.

3.2.3 Pulse Transmitter Model

Nanoscale radio communication is based on the carrier-less pulse-based communication where a nano-transmitter sends ultrashort pulses of durations on the order of hundreds of femtoseconds in the terahertz band. Chapter 3 Energy-neutral Single Pulse Transmission 45

The time proﬁle x(t) of a Gaussian pulse has the form:

! A t2 x(t) = √ exp − (3.1) 2πσ 2σ 2 where t stands for time, and A and σ determine the amplitude and width of the pulse. If a Gaussian pulse given in equation (3.1) is transmitted, due to the derivative characteristics of the antenna, the output of the transmitter antenna can be modeled by the ﬁrst derivative of the Gaussian pulse which is given by [102]:

! At t2 x(1) (t) = √ exp − (3.2) 2π σ 3 2σ 2

For impulse-radio, many diﬀerent pulse waveforms, such as Gaussian, higher time derivatives of Gaussian, Gaussian raised cosine pulse (RCP), etc., have been explored in the literature [102]. Irrespective of the waveform of the pulse, the exact pulse duration or width can be controlled using appropriate electronics. For given pulse energy, variation in pulse width will cause the transmitter to adjust the transmission power, which will essentially adjust the amplitude of the pulse. This is due to the fact that for given amount of the emitted energy repre- senting a given event type, however, manipulating the pulse width will result in achieving a unique pulse amplitude for a given combination of event type and its location. Thus a single pulse can convey both node information (using its width) and event information (using its amplitude). We will therefore simply ex- ploit this fundamental relationship between pulse widths and their amplitudes to keep the proposed approach as generic as possible. Chapter 3 Energy-neutral Single Pulse Transmission 46

3.2.4 Channel Model

Node identity and event type recognition are performed at the remote station. However, there is a communication channel between a node and the remote station. Recent studies confirm that these nodes can be able to communicate in the terahertz band using graphene as a transmitting antenna [6]. The effect of chemical compositions of the medium on the radio signal is characterized by its molecular absorption coefficient Kx(f ) (see equation (2.2)). The molecular absorption coefficients of many chemical species are available from the HITRAN database [54]. The attenuation and molecular absorption noise of the channel are determined by the attenuation (equation (2.3)) and the noise model (equation (2.4)) discussed in Chapter 2.

3.2.5 Pulse Receiver Model

Having described the pulse transmitter model in Section 3.2.3 and the channel model in Section 3.2.4, we will now discuss the pulse receiver model at the remote station. We will see in Section 3.3 that the remote station will make use of pulse amplitude (i.e. power level) detection to distinguish the pulses being sent by the monitoring nodes. We will use a widely used framework for peak power detection, based on the CTMA (Continuous Time Moving Average) based detector described in [31, 103]. The most important component in the CTMA-based detector is the CTMA filter. Intuitively, the ideal CTMA filter integrates its input over a rectangular time window of duration W . The behavior of the ideal CTMA filter can be described as follows: Let sin(t) and sout(t) be respectively the input and output signals of an R t ideal CTMA filter, then s (t) = s (τ)dτ. In practice, the CTMA filter can be out t W in − approximated by using a second order low pass filter [103]. By choosing different values of W¯ , the CTMA-based detector can be used as either an energy detector or peak power detector. Chapter 3 Energy-neutral Single Pulse Transmission 47

In single-pulse transmission, we assume the value of W¯ to be small compared with the width of the pulse, see Figure 3.2, and then the CTMA-based detector can be used to detect the peak power in the received signal. This is because the output of the CTMA ﬁlter is the averaged energy over a small time interval and this averaged energy is proportional to the averaged power in the time interval. Therefore, the CTMA-based detector will return an estimate of the peak power of the pulse.

Figure 3.2: CTMA detector to detect the peak power of the pulse.

An advantage of the CTMA-based detector is that time synchronization is not needed between the monitoring nodes and the remote station. This is because the detector uses a moving time window. The analytical modeling in [31] shows how the probability distribution of the maximum of the CTMA ﬁlter output can be derived. Let Y be the random variable of the maximum of the CTMA ﬁlter output after normalization, then the probability density function f (y,λ) of the peak power is given by:

N 1 p(y|(i,j)) = NF(y,λ) − f (y,λ) (3.3) where N is the number of discretized independent variables, f (y,λ) is the probability density function of each discretized single random variable, and F(y,λ) is its cumulative distribution function. λ is the non-centrality parameter. The probability density function f (y,λ) is usually modeled as a non-central chi-squared distribution [31, 104]. Chapter 3 Energy-neutral Single Pulse Transmission 48

!(u 2)/2 1 y − (y+λ)/2 p f (y,λ) = e− I(u 2)/2( (yλ)) (3.4) 2 λ − where λ is non-centrality parameter and equals to 2γ [105] where γ is signal-to- noise ratio which can be computed using equation (2.5), In(z) is nth order modi-

ﬁed Bessel Function of the ﬁrst kind, and u = 2wWn where w is the pulse duration or width and Wn its bandwidth [31].

3.3 Single Pulse Transmission (SPT)

In this section, we describe our proposed event monitoring single-pulse transmission approach. We ﬁrst present an overview of the event monitoring problem followed by a detailed description of the design.

3.3.1 Problem description

We assume the proposed event monitoring system consists of one remote station and a number of event monitoring nodes distributed in the environment. We use I to denote the number of nodes and use i (where i = 1,...,I) to index these nodes. We assume these nodes are used to monitor J possible events, and we will use j (where j = 1,...,J) to index these events. We assume that when Event j happens, it ˜ releases an amount of energy Ej. We assume that the amount of energy released depends only on the identity of the event but is independent of where the event ˜ ˜ occurs, and different events release the different amount of energy, i.e., Ej1 , Ej2 if j1 , j2. The architecture and components of the event monitoring nodes have been described in Section 3.2. These nodes are passive, and their purpose is to convert the energy emitted by the events into a radio message. We assume that the energy emitted by an event is only picked up by only one monitoring node, see the discussion in Section 3.2.1. We assume that the nodes have an efficiency ξ (where Chapter 3 Energy-neutral Single Pulse Transmission 49

0 < ξ < 1) of converting the energy emitted by events into the energy in the radio message; and the efficiency ξ is independent of the event type and the node. Based on these assumptions, if Event j occurs at Node i, then this event emits ˜ ˜ energy of Ej and the energy of the radio message transmitted is ξEj. We define ˜ ˜ Ej to be ξEj. We define Ej to be ξEj. The block diagram of the event monitoring system is shown in Fig. 3.3. The components of the monitoring system have been discussed in Section 3.2.

Sensor Node Remote Station

Event Energy Band Pass Peak/Energy Radio Radio 2 CTMA Classifier Energy harvester Filter ( ) Detection

Figure 3.3: Block diagram of the event monitoring system.

The aim of the remote station is to decode the radio message to determine the identity of the sending node and the event. Considering the situation that Event j occurs at Node i and Node i sends a radio message to the remote station. After decoding the radio message, the remote station determines that the node and event identities are, respectively, iˆ and jˆ. We say that the decoding is correct if iˆ equals to i and jˆ equals to j, i.e. both node and event identities are correctly determined; otherwise, the decoding is incorrect. We aim to design the radio messages to maximize the probability of correct decoding.

We will see later on that the probability of correct decoding depends on Ej’s, ˜ which are equal to ξEj’s. Since the focus of this paper is on decoding perfor- ˜ mance, the key parameter is Ej, rather than ξ and ξEj. In order to simplify the ˜ description, we will assume that ξ = 1 and Ej = Ej. Finally, we will use the 2-tuple (i,j) to denote the compound event that Event j occurs at Node i. In the Single Pulse Transmission (SPT), the radio message from the monitoring nodes to the remote station consists of one single radio pulse. Since a node Chapter 3 Energy-neutral Single Pulse Transmission 50 needs to convey both its identity as well as event type to the remote station, the node will need to encode both pieces of information in a single pulse. Since the node sends only one pulse, all the energy emitted by an event will be captured in the radio pulse transmitted by the node; therefore, the pulse energy encodes the event type. For SPT, the proposal is that each node uses a unique pulse width, or in other words, the pulse width encodes the identity of the node. As a result, the pulse amplitude is dependent on both event type and node identity. For the decoding of the radio message, we propose that the remote station uses the peak power of the pulse. As an example, consider the case where the number of nodes I = 2 and the number of events J = 2. Figure 3.4(a) shows an example where Node 1 and Node 2 use two different widths and this results in four different peak pulse amplitudes Pij where Pij is the peak power of the pulse transmitted by Node i when Event j occurs. Since the four Pij’s are different, the remote station can use the received Pij to determine both event and node identities. The mechanics of how the remote station can measure peak pulse power has already been explained in Section 3.2.5. Since the remote station uses the peak received power to decode the radio message, it is important that the Pij’s from different node-event pairs be well sep- arated to reduce the decoding error. Figure 3.4(b) illustrates an example of poor choice of pulse widths which results in a small gap between P12 and P21, which means that the remote station will find it hard to distinguish between node-event pairs (1,2) and (2,1). Since the decoding performance depends on the peak received power, and the peak received power depends on the event energy and the pulse widths that the nodes use, therefore we can minimize the decoding error by appropriately choosing the pulse widths used by the nodes. We will call this the Pulse Width Allocation (PWA) problem. For PWA, we assume that the remote station uses the maximum a posteriori (MAP) framework for decoding. Let y be the peak received Chapter 3 Energy-neutral Single Pulse Transmission 51

(a)

Power 1 2 Node 1 Node 2

(b) 1 2 3

≈ Power

Amplitude/Power 1 3 Node 1 Node 2

Figure 3.4: Illustration of pulse widths vs peak power of transmitted pulses in single-pulse approach. power, then estimated node identity iˆ and event type jˆ are:

p(y|(i,j))p((i,j)) (i,ˆ jˆ) = maxP ((i,j)|y) = (3.5) (i,j) PI PJ | i=1 j=1 p(y (i,j))p((i,j)) where p(y|(i,j)) is the probability that peak received power is y given the node- event pair (i,j) and p((i,j)) is the prior probability of Event j occurring at Node i. Assuming IJ joint events equal to M, we compute the joint event error probability by introducing cumulative distribution function (Φµ,σ (α)) such as: Chapter 3 Energy-neutral Single Pulse Transmission 52

 1  P [error] =  1 − Φ (α ) M  µ1,σ1 1,2 MX1 − − + Φµ ,σ (αk 1,k) + 1 Φµ ,σ (αk,k+1) (3.6) k k − k k k=2   + Φµ ,σ (αM 1,M)  M M − 

Where Φµ1,σ1 (α1,2) is the cumulative distribution function for the node-event pair (1,1) at intersection point α1,2, with mean µ1 and standard deviation σ1, which can be derived from peak power distribution of the node-event pair (1,1).

3.3.2 Pulse Width Allocation (PWA)

We formulate PWA as an optimization problem. For this optimization problem, we assume that we are given the event energies Ej, the maximum event rate, the distance between the nodes and the remote station, as well as the channel parameters discussed in Section 3.2.4. The decision variables are wi where wi is the pulse width for Node i. Let ErrorSPT(w1,...,wI ) denote the decoding error when wi’s are used. We can state the PWA optimization problem as:

min ErrorSPT(w1,...,wI ) (3.7) w1,w2,...,wI subject to: ∀ wi1 , wi2 i1,i2 = 1,....,I and i1 , i2 (3.8) ≤ ≤ ∀ wmin wi wmax i1,i2 = 1,....,I (3.9)

The purpose of constraint (3.8) is to ensure that diﬀerent pulse widths are allocated to diﬀerent nodes. The purpose of (3.9) is to impose a lower bound wmin Chapter 3 Energy-neutral Single Pulse Transmission 53

Table 3.1: Table of the most frequently used parameters in single-pulse approach of eNEUTRAL IoNT framework.

Parameter Deﬁnition

I Number of nodes i i is the index of nodes such as i = 1,...,I JJ Number of events j j is the index of events such as j = 1,...,J ∈ ∈ Eij Energy released by event j in node i where i I and j J ξ ξ is the node efficiency such as 0 < ξ < 1 Pij Pij is the peak power of the pulse transmitted by Node i when Event j occurs where i ∈ I and j ∈ J (i,j)(i,j) represents node-event pair that is event j in node i where i ∈ I and j ∈ J Nabs Molecular absorption noise spectral density Wn Wn is the pulse bandwidth λ λ is non-centrality parameter and equal to 2γ γ γ is signal to noise ratio experienced by a remote station In(z) In(z) is nth order modified Bessel Function of the first kind wmin Lower bound of pulse width allocation wmax Upper bound of pulse width allocation ∈ wi Pulse width allocated to node i where i I W¯ represents the integration window in CTMA detector

and an upper bound wmax on the pulse width. Since the pulse-based transmission system for nanoscale devices in [12,19] works in the terahertz band, a lower bound wmin of 0.1ps will ensure that the bandwidth of the pulse does not exceed

10THz. We choose the upper bound wmax to be far less than the reciprocal of the maximum event rate, in order to ensure that there is a small probability that the pulses from diﬀerent nodes will collide. The most frequently used parameters and symbols in single-pulse approach along with their detailed descriptions are given in Table 3.1. Chapter 3 Energy-neutral Single Pulse Transmission 54

3.3.3 Numerical investigation on PWA

This section has two aims. First, we want to study the impact of the pulse widths on the error probability of decoding joint events. Second, we want to investigate the relationship between the energy emitted by the events and the choice of pulse widths. We assume two nodes in Section 3.3.4 and extend to more than two nodes in Section 3.3.5. We consider a channel with standard air having composition of Table 3.2 and normal pressure/temperature of 1atm/296K. We extract the corresponding molecular absorptions from HITRAN [54] for frequency ranging from 0.1 − 10 THz.

Species CO2 H2O O2 N2 Others (CO, CH4,N2O, O3) Ratio (%) 0.033 1.19 20.9 77.88 0.000218

Table 3.2: The Composition of normal air.

We use equation (3.3) to determine the probability density function of the peak power of the received pulse. After that, we determine the error probability using the maximum a-posterior decoding.

3.3.4 Impact of pulse widths on error probability

We assume that the energies emitted by the two events are E1 = 1 atto Joule (aJ) and E2 = 3 aJ. The pulse widths of two nodes are denoted by w1 and w2. The minimum value of w1 and w2 is 0.1 ps due to the maximum bandwidth of 10

THz. We vary w1 and w2 between 0.1 ps and 10 ps. For each pair of w1, w2 we compute the error probability. We plot the results as a heat map in Figure 3.5(a) (best view in colour) where dark blue indicates low error probability and yellow indicates high error probability. Note that the heat map is symmetric about the main diagonal because the roles of w1 and w2 are interchangeable. Chapter 3 Energy-neutral Single Pulse Transmission 55

We ﬁrst consider the case where w1 = w2 which is the main diagonal in Figure

3.5(a). The heat map shows a yellow line along w1 = w2. This is understandable because if both nodes use the same pulse width, the error probability is high. In Figure 3.5(a), we observe two faint yellow lines at an angle of about 45o and 135o. These two faint yellow lines are also regions where error probability is high. We will use a simple example to illustrate how this region of high error arises. First recall that E1 = 1 aJ and E2 = 3 aJ. Let us assume w1 = 0.1 ps and w2 = 0.3 ps. The powers, P11 and P12, of the pulses transmitted by node 1 are 10 µW and 30 µW.

Similarly, the powers, P21 and P22, of the pulses transmitted by node 2 are 3.33 µW and 10 µW. This means that the power of the transmitted pulse for Event 1 at node 1 is the same as that of Event 2 at node 2. Therefore, for this particular choice of pulse widths, it is hard for the receiver to distinguish these two joint events. By generalizing this example, the two faint yellow lines correspond to the lines E1 = E2 . w1 w2 We can also see a number of regions where the error probability is low as indicated by the dark blue color. We can understand this by considering a cross- section of the heat map. Figure 3.5(b) plots the error probability against w2 for w1 = 0.1 ps. Initially, when w2 is close to w1, the error probability is high. We can see another peak in error probability and this corresponds to E1 = E2 . Interesting, w1 w2 we see two regions where the error probability is low. In fact, we can see a local minimum and a global minimum. We will discuss these two minima.

Let us consider the result for w1 = 0.1 ps. The local minimum occurs at about w2 = 0.2 ps. Given these pulse widths, we know that node 1 uses pulses with power 10 µW and 30 µW, and node 2 uses pulses with power 5 µW and 15 µW. If we arrange these four amplitudes in ascending order, we see that the pulses belong alternately to the two nodes. In general, this minimum corresponds to the region where E1 < E1 < E2 < E2 and if the power of the pulses are chosen w2 w1 w2 w1 appropriately by making sure that there is a good size gap between the powers of the pulse, then it is possible to have low error probability. Chapter 3 Energy-neutral Single Pulse Transmission 56

Continuing on the curve corresponding to w1 = 0.1 ps in Figure 3.5(b). The global minimum occurs at about w2 = 0.6 ps. Given these pulse widths, we know that node 1 uses pulses with powers 10 µW and 30 µW, and node 2 uses pulses with powers 1.6 µW and 5 µW respectively. In this case, the powers of the pulses used by node 2 are lower than those of node 1. This corresponds to the strategy of letting node 2 use very different pulse powers compared to those of node 1 to reduce classification error. We want to make a few remarks regarding these results. First, there is a wide range of pulse widths that can give a low probability of error. For example, if 5 w1 = 0.1 ps, then we want a probability error of less than 10− , we can choose w2 to be in the intervals 0.13 ps to 0.25 ps and 0.38 ps to 10 ps. Second, with such a wide range of w2, it means that it may be possible to scale our method of using pulse widths to identify nodes to more than two nodes. This will be further discussed in Section 3.3.5. Third, a factor that restricts the maximum pulse width is the rate at which events occur. A node should finish the transmission of a pulse before another event arrives. Therefore, the event rate imposes an upper bound on the maximum pulse width. We vary the energy levels of the two events to study how they impact on the pulse width allocation. We observe that, for all the energy levels that we have used there is one global and one local minimum. We further observe there is a linear relation between the ratio w2 and E2 . Figure 3.6 w1 E1 shows the relation for these ratios for the global as well as the local minima. We perform a least-squares fit and found the following relations:

w E 2 = 0.4069 2 + 0.5871 w E 1 1 (3.10) w E 2 = 2.7358 2 − 2.0048 w1 E1 Chapter 3 Energy-neutral Single Pulse Transmission 57

(a)

100 Error at w = 0.1 ps 1

10-5

10-10 Error probability

10-15

10-20

0.1 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 w (ps) 2 (b)

Figure 3.5: (a) Error probability versus pulse width of two nodes at distance d = 30 mm (b) A cross-section of heatmap Chapter 3 Energy-neutral Single Pulse Transmission 58

2.4 Data 2.2 Least square fit

2 1

1.8

/PW 2

1.6 PW 1.4

1.2

1 1 1.5 2 2.5 3 3.5 4 4.5 E /E 2 1 (a)

2.4 Data 2.2 Least square fit

2 1

1.8

/PW 2

1.6 PW 1.4

1.2

1 1 1.5 2 2.5 3 3.5 4 4.5 E /E 2 1 (b) Figure 3.6: Energy ratio vs pulse width ratio: (a) Local minimum (b) Global minimum 3.3.5 Pulse widths allocation to more than two nodes

To study the impact of pulse widths of more than two nodes on the error probability of decoding joint events, we develop optimization problem with bound constraints. We aim to minimize the classiﬁcation error probability in a multidi- mensional space for two to ﬁve nodes with upper and lower bound constraints Chapter 3 Energy-neutral Single Pulse Transmission 59

as wmin, and wmax. In Section 3.4, we will study the cases with higher number of nodes. We see in Section 3.3.4 that the pulse-width allocation problem has multiple minima. Since a long pulse width can limit the event rate that the nanomotes can monitor, we want to keep the maximum pulse width small. For the two mote case, we see in Section 3.3.4 we can achieve that by choosing the local minimum. We see from Section 3.3.4 that the local minimum is characterized by E1 < E1 < E2 < E2 . We can generalize this requirement to the cases with w2 w1 w2 w1 more than two nodes. For example, when there are five nodes, we impose the constraint: E1 < E1 < E1 < E1 < E1 < E2 < E2 < E2 < E2 < E2 to limit the maxi- w5 w4 w3 w2 w1 w5 w4 w3 w2 w1 mum pulse width. We summarize the results for the 2-, 3-, 4- and 5-node cases in Table 3.3. We see that by imposing the above constraint in the optimization, the maximum pulse width increases only moderately when the number of nodes increases. Note that when five nodes are used, the error probability is of the order 4 of 10− which means that the system can support five nodes.

No. of nodes Pulse widths (ps) Error probability 17 Two 0.1, 0.19 3.15 × 10− 7 Three 0.1, 0.14, 0.22 1.9 × 10− 04 Four 0.1, 0.13, 0.17, 1.26 × 10− 0.24 03 Five 0.1, 0.12, 0.15, 2.1 × 10− 0.19, 0.25

Table 3.3: Pulse widths and the error probability for 2-5 nodes at distance d = 30 mm.

3.4 Evaluation

In this section, we evaluate the performance of the single-pulse system as a function of distance, number of nodes and event types.The parameter that we do not vary is the mean absorption coeﬃcient is that of standard air (see Table 3.2). Chapter 3 Energy-neutral Single Pulse Transmission 60

3.4.1 Impact of node-RS distance

As a function of the distance, Figure 3.7 plots classification error for three different event types with harvested energy values of 1, 2, and 3 aJ. We observe that with increasing distance, error in single-pulse approaches 100%. This is due to increase attenuation of pulse amplitudes and energies as well as increasing noise power with increasing distance between the transmitter and the receiver. Equa- tion (2.3) and equation (2.4) clearly shows that and attenuation in a terahertz channel increases exponentially with distance; therefore, classification becomes more challenging with increasing distance.

0.9

0.8 SPT

0.7

0.6

0.5

0.4

Error probability Error 0.3 20 nodes 10 nodes 5 nodes 0.2

0.1

10 15 20 25 30 35 40 Distance (mm)

Figure 3.7: The eﬀect of distance on the classiﬁcation error in single-pulse approach for 5-, 10- and 20-node cases.

3.4.2 Impact of number of nodes

The larger the number of nodes in the network, the greater the classification challenge faced at the receiver. We, therefore, study the performance of the proposed Chapter 3 Energy-neutral Single Pulse Transmission 61 single-pulse solution as a function of the number of nodes while considering two distinct event types with harvested energies of 1 aJ and 2 aJ. Figure 3.8 plots classification error for a distance of 30 mm. As expected, performance deteriorates with increasing number of the node; however, we find that single-pulse approach with 10-node network monitoring two event types can achieve 90% event type and location detection accuracy for a distance of 30 mm.

0.35

SPT 0.3

0.25

0.2

0.15 Error probability Error

0.1

0.05

0 5 6 7 8 9 10 11 12 13 14 15 Number of nodes (I)

Figure 3.8: Eﬀect of number of nodes on the classiﬁcation error for a 30mm network monitoring two distinct event types with energies 1 aJ and 2 aJ.

3.4.3 Impact of number of event types

We now investigate how the number of event types affect the classification error of the single-pulse approach. The more distinct types of events are there, the harder it becomes for the receiver to classify them accurately. We keep the number of nodes equal to 5 and distance d = 30 mm. In Figure 3.9 we vary the event types from 1-10 (resp. with energy values from 1-10 aJ) and find that performance drops with increasing number of event types. This is because increasing Chapter 3 Energy-neutral Single Pulse Transmission 62 number of event types result in increasing number of classes to classify. The number of the classes is a major variable in the error probability. If we have a larger number of classes to classify, then the classification error would be higher.

0.4

SPT 0.35

0.3

0.25

0.2

0.15 Error probability Error

0.1

0.05

0 2 3 4 5 6 7 8 9 10 Number of events (J)

Figure 3.9: Eﬀect of number of event types on classiﬁcation error for a 5-node network of radius 30 mm.

3.5 Chapter Summary

In this chapter, we proposed an energy-neutral event monitoring framework, Called eNEUTRAL IoNT, which allows the sensors to transmit the complete information about a detected event, including the event type and its location. Such a framework allows the sensors to transmit event information using only the amount of energy harvested from the events. We, therefore, focused on designing pulses that will not only consume exactly the amount of energy harvested from the event but also accurately convey both event type and its location to Chapter 3 Energy-neutral Single Pulse Transmission 63 a distant receiver. After thorough investigation and analysis, we design single- pulse as the first implementation for this framework. This implementation uses a single pulse to communicate both location and event type using entire energy harvested from the event but manipulates its time duration or width to create a unique pulse amplitude for a given combination of event type and its location. To minimize classification error at the receiver, we optimize pulse durations in the single-pulse. Extensive numerical experiments have shown that single-pulse approach achieving 99% event type and location detection accuracy in 10-node network monitoring two different event types for a distance of 22 mm. Chapter 4

Energy-neutral Dual Pulse Transmission

The previous chapter attempted to achieve energy-neutral event monitoring using only a single pulse. However, in single-pulse transmission, we manipulated time duration or width to create a unique pulse amplitude for a given combination of event type and its location. This results in an increasing number of classes to classify. The number of classes are a significant variable in the error probability. Having a larger number of classes to classify the classification error would be higher. In this chapter, we propose Dual Pulse Transmission (DPT) in which the nodes will have the same architecture to maintain the sensorless architecture. We find that the dual-pulse approach significantly outperforms the single-pulse approach which we will show in Section 4.4.

4.1 Introduction

This chapter proposes a novel concept namely Dual Pulse Transmission (DPT) of the eNEUTRAL IoNT framework which can reliably monitor event types at longer distances. In this approach, we impose the requirement that each node

64 Chapter 4 Energy-neutral Dual Pulse Transmission 65 must communicate both the sensed event and the node identification to the remote station by transmitting an address and event wireless pulse respectively. The energy of the first pulse uniquely defines a location and the second pulse uses the remaining energy to identify event types. In this approach, we formulate the energy allocation to address and event pulse as an optimization problem with the aim to minimize the classification error. Extensive numerical experiments involving terahertz channels confirm that the dual-pulse approach significantly outperforms the single-pulse approach achieving 99% event type and location detection accuracy for a radius of 40mm. The rest of the chapter is structured as follows. We propose dual-pulse implementation of eNEUTRAL IoNT framework in Section 4.2. We present the Pulse Energy Allocation (PEA) optimization model in Section 4.3. The performance analysis using extensive numerical experiments is investigated in Section 4.4.A discussion on the implementation of the dual-pulse method is presented in Sec- tion 4.5. We summarize the chapter in Section 4.6.

4.2 Dual Pulse Transmission (DPT)

To improve the accuracy of event and node identification, we propose the second implementation option of the eNEUTRAL IoNT framework, dual-pulse transmission, which uses a different mechanism for encoding and decoding. In this implementation, the harvested event energy is divided into two pulses so that the energy of the first pulse uniquely defines a location and the second pulse uses the remaining energy to identify event types. We, therefore, encode the address and event type in the energy level of the transmitted pulse. Table 4.1 compares the two proposed schemes SPT with DPT. Chapter 4 Energy-neutral Dual Pulse Transmission 66

Criteria SPT DPT Number of 1 2 pulses in a radio message Pulse width Each node uses a dis- All nodes use the same tinct pulse width pulse width Optimization Tune the pulse width Tune the address pulse energy Decoding Based on peak re- Based on the received ceived power pulse energy

Table 4.1: Comparison between SPT and DPT.

4.2.1 Problem description

The main intuitive of this approach is if we break down the number of classes into smaller class with two pulses then we can improve classification accuracy. To minimize classification error at the receiver, we optimize pulse energies in the dual-pulse approach. For evaluation and scalability we will study the decoding performances of single- and dual-pulse implementations, respectively, in Section 4.4. Recall from Section 3.3, we assume the proposed event monitoring system consists of one remote station and a number of event monitoring nodes distributed in the environment. Similar to single-pulse approach, we use I to denote the number of nodes and use i (where i = 1,..,I) to index these nodes. We assume these nodes are used to monitor J possible events, and we will use j (where j = 1,..,J) to index these events. We assume that when Event j happens, it releases ˜ an amount of energy Ej. We assume that the amount of energy released depends only on the identity of the event but is independent of where the event occurs, ˜ ˜ and different events release the different amount of energy, i.e., Ej1 , Ej2 if j1 , j2. As it has been discussed previously, if an event occurs at a node, the aim of the remote station is to identify both the type of event and the identity of the node. In order to achieve this without changing the principle architecture of the node, Chapter 4 Energy-neutral Dual Pulse Transmission 67 we impose the requirement that each radio message from a monitoring node consists of two pulses. In this approach, we assume that all pulses from all nodes use the same pulse width. The identity of the node is encoded in the energy level of the first pulse, and the event type is encoded in the energy of the second pulse; we will, therefore, refer to these two pulses as the address pulse and event pulse. As an illustration, we consider the case of how Node i encodes the radio message when Event j occurs. Node i will first transmit the address pulse with an energy level of ηi, and then it will transmit the event pulse with the remaining energy − ≥ Ej ηi. These two pulses are transmitted sufficiently apart (i.e. Γ w) where Γ is the time duration between address and event pulse and w is the pulse width. We will discuss how to implement the dual-pulse approach in Section 4.5. In order for the remote station to identify different nodes from the address pulse, a requirement is that different nodes should use different amounts of energy in their address pulse, i.e. ηi1 , ηi2 if i1 , i2. We can see from this example that the design choice to design DPT is to choose the energy levels η1,...,ηI for the nodes. In addition, we require that maxi ηi < minj Ej so that a positive energy is allocated to the event pulse.

4.2.2 Pulse Energy Detection

The decoding at the remote station will be based on the received energy levels of the two pulses. The remote station will make use of the CTMA-based energy detector in [31,103], to determine the received pulse energy. In dual-pulse transmission, the remote station detects the pulse energy where we choose the value of the integration window, W , to be the same as the width of the pulse, then the CTMA-based detector functions as a pulse energy detector. This conﬁguration is identical to the set up in [31] and is illustrated in Figure 4.1. We can see from Figure 4.1 that if the integration window W¯ covers exactly the extend of the received pulse, then the CTMA ﬁlter output has its maximum when the output Chapter 4 Energy-neutral Dual Pulse Transmission 68 corresponds to the energy of the pulse.

Figure 4.1: CTMA detector to detect the pulse energy.

The decoding of the address and event pulses will be done in two steps. In the first step, the remote station uses the received pulse energy of the address pulse in a MAP classifier — similar to (3.5) but with only one decision — to determine the node identity. The second step is similar to the first except that the aim is to determine the event type based on the received event pulse energy.

4.3 Pulse Energy Allocation (PEA)

Similar to SPT, we can deﬁne an optimization problem to minimize the decoding error probability of the dual-pulse approach. For DPT, we assume that we are given the event energies Ej, the distance between the nodes and the remote station, the channel parameters and a design parameter α which we will discuss later. The goal of the optimization is to optimize the energy allocations for the ﬁrst pulse. The decision variables for the optimization problem are the energy levels of address pulses such as η1,...,ηI . Let ErrorDPT(η1,...,ηI ) be the decoding error probability as a function of the decision variables. We can state the optimization problem as: Chapter 4 Energy-neutral Dual Pulse Transmission 69

min ErrorDPT(η1,...,ηI ) (4.1) η1,η2,...,ηI subject to: ∀ ηi1 , ηi2 i1,i2 = 1,....,I and i1 , i2 (4.2) ≤ ∀ α ηi < minEj i1,i2 = 1,....,I (4.3) j

The design parameter α is to set a lower limit on the ηi. We compute the joint error probability as:

− ErrorDPT(η1,...,ηI ) = 1 CorrectDPT(η1,...,ηI ) (4.4)

Let address pulse is represented by γi which using energy ηi and event pulse is − described by βi,j using energy Ei,j ηi then the correct probability of joint address and event is computed as

CorrectDPT(η1,...,ηI ) = XX h | i P decoded βi,j decoded γi &(γi,βi,j sent) (4.5) i j h | i h i P decoded γi (γi,βi,j sent) P γi,βi,j sent

where P [γi,βi,j sent] is joint prior probability. The most frequently used parameters and symbols in dual-pulse approach along with their detailed descriptions are given in Table 4.2.

4.4 Performance Evaluation

In this section, we evaluate and compare the performance of single and dual pulse approaches of the eNEUTRAL IoNT framework as a function of network Chapter 4 Energy-neutral Dual Pulse Transmission 70

Table 4.2: Table of the most frequently used parameters in dual-pulse solution of eNEUTRAL IoNT framework.

Parameter Deﬁnition

I Number of nodes i i is the index of nodes such as i = 1,...,I JJ Number of events j j is the index of events such as j = 1,...,J ∈ ∈ Eij Energy released by event j in node i where i I and j J ξ ξ is the node eﬃciency such as 0 < ξ < 1 Pij Pij is the peak power of the pulse transmitted by Node i when Event j occurs where i ∈ I and j ∈ J (i,j)(i,j) represents node-event pair that is event j in node i where i ∈ I and j ∈ J ΓΓ is the time duration between address and event pulse γi γi is the address pulse of node i ⊂ ηi Energy level of address pulse of node i where ηi Ej βi,j βi,j is the event pulse of event j in node i α A design parameter used to set a lower limit on the ηi

radius, the number of nodes and distinct event types, and the spread of emitted energies among different event types. We also study the effect of node placement errors on the classification accuracies of the proposed systems. We simulate a channel with standard air (see Table 3.2) and normal pressure/temperature of 1 atm/296 K by extracting the corresponding molecular absorptions from HITRAN [54] for frequency ranging from 0.1 − 10 THz.

4.4.1 Impact of the network radius

As a function of network radius, Figure 4.2 plots classiﬁcation error for three diﬀerent event types with harvested energy values of 1, 2, and 3 aJ. We observe the following:

1. With increasing network radius, error for both single and dual pulse approaches 100%. This is due to increasing attenuation of pulse amplitudes Chapter 4 Energy-neutral Dual Pulse Transmission 71

100 SPT DPT 10-1

10-2

10-3

10 nodes 5 nodes 10 nodes 5 nodes Error probability Error

10-4

10-5

10 15 20 25 30 35 40 Distance (mm)

Figure 4.2: The eﬀect of distance on the classiﬁcation error in SPT and DPT for 5- and 10-node cases.

and energies as well as increasing noise power with increasing distance between the transmitter and the receiver. This is because of attenuation and molecular absorption noise in the THz channel (see Chapter 2 for more details) increase exponentially with distance, therefore, classiﬁcation becomes more challenging with increasing distance.

2. Dual pulse transmission significantly outperforms single pulse either in terms of classification error for a given network radius, or in terms of achievable network radius for a target reliability. As it has been discussed in Section 4.2, the breakthrough of this improvement of the dual-pulse is the number of classes to classify. If we break down the number of classes into smaller class with two pulses then classification accuracy can be improved. Chapter 4 Energy-neutral Dual Pulse Transmission 72

4.4.2 Impact of the number of nodes

The larger the number of nodes in the network, the greater the classification challenge faced at the receiver. We, therefore, study the performance of the proposed methods as a function of the number of nodes while considering two distinct event types with harvested energies of 1 aJ and 2 aJ. Figure 4.3 shows achievable 2 3 network radius for a target error between 10− to 10− , while Figure 4.4 plots classification error for a network radius of 30 mm. As expected, performance deteriorates with increasing number of nodes, but dual pulse outperforms single pulse significantly.

38 DPT SPT 36

28 Distance ( mm ) mm ( Distance

20 2 3 4 5 6 7 8 9 10 Number of nodes ( I )

Figure 4.3: Eﬀect of number of nodes on achievable radius for target error be- 2 3 tween 10− and 10− when monitoring two distinct event types with energies 1 aJ and 2 aJ.

4.4.3 Impact of the number of event types

The more distinct types of events are there, the harder it becomes for the receiver to classify them accurately. Using a 5-node network, we study the eﬀect of an Chapter 4 Energy-neutral Dual Pulse Transmission 73

100

SPT DPT 10-1

10-2

10-3

10-4 Error probability Error

10-5

10-6

10-7 5 6 7 8 9 10 Number of nodes (I)

Figure 4.4: Effect of number of nodes on the classification error for a 30 mm network monitoring two distinct event types with energies 1 aJ and 2 aJ. increasing number of event types on the performance of the proposed system. In Figure 4.5 we vary the event types from 1-20 (j = 1,..,20) with corresponding energy values as 1,..,20 aJ respectively. Figure 4.5 and Figure 4.6 plot classification error for a 30 mm network and achievable network radius for target error 2 3 between 10− and 10− , respectively. Again, we find that performance drops with increasing number of event types, but the dual-pulse significantly outperforms the single-pulse approach.

4.4.4 Impact of the event energy spread

A key requirement for the proposed framework is that distinct event types emit distinct amounts of energy. It is clear that wider the spread of energies, easier it is for the receiver to classify the events, and vice versa. In this section, we study the rate at which we can reduce classiﬁcation error by increasing event energy spread. For two distinct event types, Table 4.3 shows the performance for Chapter 4 Energy-neutral Dual Pulse Transmission 74

100

-1 10 SPT DPT 10-2

10-3

10-4 Error probability Error

10-5

10-6

10-7 2 4 6 8 10 12 14 16 18 20 Number of events (J)

Figure 4.5: Eﬀect of number of event types on classiﬁcation error for a 5-node network of radius 30 mm. 35

30 SPT DPT 25

15 Distance (mm) Distance

1 2 3 4 5 6 7 8 9 10 Number of events (J)

Figure 4.6: Eﬀect of number of events types on achievable radius for target error 2 3 between 10− and 10− for a 5-node network. Chapter 4 Energy-neutral Dual Pulse Transmission 75 a network with a radius of 30 mm as we vary the diﬀerences in event energies from 10% to 50% for 5-node and 10-node networks. We observe the following:

1. As expected, with increasing energy spread, classiﬁcation error drops for both single and dual approaches.

2. The rate at which classiﬁcation error reduces with increasing energy spread among the two events is signiﬁcantly higher for dual pulse compared to the single pulse.

Nodes Events En- Error probability ergy (aJ) SPT DPT 1 2 Five 1, 1.1 1.88 × 10− 1.48 × 10− 2 4 1, 1.2 5.87 × 10− 6.46 × 10− 2 8 1, 1.3 1.75 × 10− 4.43 × 10− 3 9 1, 1.4 7.60 × 10− 2.72 × 10− 3 9 1, 1.5 5.76 × 10− 2.71 × 10− 2 Ten 1, 1.1 0.3178 2.95 × 10− 3 1, 1.2 0.2043 7.087 × 10− 3 1, 1.3 0.1669 7.022 × 10− 3 1, 1.4 0.1575 7.021 × 10− 3 1, 1.5 0.1720 7.020 × 10−

Table 4.3: Impact of energy gap of diﬀerent event types on the error probability at distance d = 30 mm.

4.4.5 Impact of the node placement

Because we use the received pulse energy for the classification of events and the nodes, any variation in the distances between nodes and the central receiver can make the classification more challenging. If the nodes are placed perfectly on the circle, then the distance between any node to the receiver is the same, the radius of the network. However, the manufacturing process may cause some deviations from the exact placement, which we refer to as placement error. With modern Chapter 4 Energy-neutral Dual Pulse Transmission 76 nanoscale lithography, it is expected that such placements errors will be on the order of a few nanometers at most [106]. To study the effect of such placement error on the classification performance, we assume that placements errors are Gaussian distributed with zero mean and a standard deviation of σplacement mm. For two event types, Figure 4.7 compares the impact of placement error when σ is varied from zero (perfect placement) to 80 µm for a 5-node network with a radius of 30 mm. We can see that for this range of placement errors, dual pulse still outperforms single pulse significantly. The asymmetry of the error bar in Figure 4.7 at each data point is due to the log scale.

100

10-2 SPT DPT

10-4

Error probability Error 10-6

10-8

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 (mm) placement

Figure 4.7: Eﬀect of placement error for a 5-node network with a radius of 30 mm for 2 distinct event types having energies of 1 aJ and 1.5 aJ.

4.4.6 Impact of the event energy ﬂuctuations

So far, we have considered purely deterministic event energies. In real scenarios, however, actual energy emitted by a particular event type could ﬂuctuate slightly. Chapter 4 Energy-neutral Dual Pulse Transmission 77

For example, it was found that chemical reactions release energy which is not 100% deterministic but vary with slight fluctuation [11]. We, therefore, study the effect of event energy fluctuation on the performance of the proposed framework. Note that higher the variation, the more challenging it becomes for the receiver to classify different events for a given energy spread. In Figure 4.8, we compare the impact of energy fluctuation, with zero mean, we vary a standard deviation

σenergy from zero to 0.06 aJ for a 5-node network monitoring two event types of 1 aJ and 1.5 aJ. We can see that for this range of energy ﬂuctuations, dual pulse still outperforms single pulse signiﬁcantly.

100

10-2 SPT DPT

10-4

-6

Error probability Error 10

10-8

0 0.01 0.02 0.03 0.04 0.05 0.06 (aJ) energy

Figure 4.8: Eﬀect of event energy ﬂuctuation for a 5-node network with a radius of 30 mm for 2 distinct event types.

4.5 Discussion

In our proposed framework we assume that all nodes are equidistant from the remote station. However, in the case of distance heterogeneous networks where Chapter 4 Energy-neutral Dual Pulse Transmission 78 nodes are deployed randomly then the advantage of having distinct energies are not relevant because the received energy distributions matter for the classification. In such a scenario, the received power or energy observations of some nodes may not be distinct which can make the classification difficult. Implementing the dual-pulse approach is more complex as it requires to transmit two pulses from the energy emitted by an event type. The dual-pulse approach can follow Harvest- Store-Use architecture proposed in [95]. However, instead of storing the entire energy harvested from the event, an amount of harvested energy is used to generate the address pulse directly and the remaining energy, diverted and stored in the storage component temporarily, can be used to generate the event pulse.

4.6 Chapter Summary

In order to improve the detection accuracy at longer distance, we design and evaluate the dual-pulse implementation of energy-neutral event monitoring framework which has same sensorless architecture but uses two pulses; the ﬁrst conveying the location and the second conveying the event type. We develop an optimization model for pulse energy allocation to optimize pulse energies. Extensive numerical experiments have shown the dual-pulse signiﬁcantly outperforms the single-pulse method. In the next chapter, we will propose a simple architecture of pulse generator which can generate THz pulses directly on the graphene surface. Using COMSOL Multiphysics, we will show that the proposed architecture can generate SPPs in the THz band without employing any extra circuits which are usually required in existing methods. Chapter 5

Pulse Generator for tiny IoNT Nodes

In previous chapters, the eNEUTRAL framework was studied considering certain assumptions. In particular, we assumed that pulses in the THz band can be generated without using any additional components to maintain the sensorless node architecture intact. In this chapter, we investigate the key component of the proposed IoNT framework, i.e., pulse generator, using COMSOL Multiphysics to generate THz pulses without adding any new circuit to the node architecture. We first surveyed the literature of different approaches for pulse generation that generate Surface Plasmon Polaritons (SPPs) which lead to femtoseconds long pulses in graphene. Based on our theoretical analysis, we find that most of the existing configurations require complex structures including a prism [84, 86], periodic slits [80, 84] and special circuits (i.e., High Electron Mobility Transistor) [13, 81] which may be difficult to implement in simple structure of monitoring node. This is because the tiny IoNT event monitoring node consists of two components: an energy harvester and a radio (pulse generator). Using COMSOL Multiphysics, we show that pulses in the THz band can be generated using the near-field method by which the matching condition for excitation of SPPs can be easily satisfied without adding any new component to the monitoring node which is usually required in existing methods.

79 Chapter 5 Pulse Generator for tiny IoNT Nodes 80

5.1 Introduction

A pulse generator has the potential to enable carrier-less wireless communication for nanometer or subwavelength scale. However, due to the limited energy bud- get, such communication can happen in the terahertz band (0.1 - 10 THz) [1,6]. By allowing wave propagation in the terahertz band, the communication can happen with less power which might be feasible for event monitoring nodes operated by nanoscale energy harvesting systems. The terahertz pulses then propagate towards the antenna and are finally radiated [29]. A pulse generator generates SPPs which lead to femtoseconds long pulses [13, 18, 19]. The nanoantenna converts THz pulses into EM free-space radiation [13, 18, 19]. Due to the simple structure of event monitoring node (see chapter 3 for more details), a pulse generator requires an easy method to excite SPPs. Using a near- field excitation method, we find that SPPs can be excited in the THz band. Such method does not require any special structure including prism [84, 86], periodic slits [80,84], or high-electron-mobility-transistor (HEMT) [13,81], and can easily excite SPPs suitable for THz nanocommunication. Near-field excitation method is already reported in the literature for gate tuning of graphene plasmons in infrared regime [79] and to improve the propagation length of SPPs in ultraviolet regime [92]. However, using COMSOL Multiphysics, we show that this method can generate SPPs in the THz band without employing any extra circuits which are usually required in existing methods. Inspired from tip method in [79], where a laser irradiates a metallic tip with the radius much smaller than operation wavelength, we introduce a point source of nano dimension with radius R = 20 nm as a boundary condition for exciting SPPs on graphene sheet [93]. In this method, a Transverse Magnetic (TM) polarized evanescent nanoscale source, such as a magnetic dipole or quadrupole, is located above graphene. This kind of sources with dimensions much smaller than Chapter 5 Pulse Generator for tiny IoNT Nodes 81 the wavelength of EM radiation, produce EM radiation with very high wavenumbers k. Hence, the matching condition between the wave number of EM radiation

( k ) and the SPP wave number (ksp) is easily satisfied, and consequently, SPPs can be directly excited. However, such method excites SPPs locally, which requires SPPs resonance improvement in practice [13,107] to enhance the antenna radiation. Thus we numerically investigate how operating frequency, the doping amount of graphene and the properties of the evanescent source impact the plasmon resonance of SPPs. The performance of the proposed near-field excitation method in the THz band is optimized to maximize the SPP resonance via numerical simulations. This includes the choice of the source, its phase angle, the chemical potential, and frequency. The proposed model can be a good candidate for a low-complexity realization of a THz pulse generator in events monitoring node. The rest of the chapter is organized as follows: The fundamentals of SPP are reviewed in Section 5.2. A discussion on the state of the art configurations for exciting SPPs is presented in Section 5.3. In Section 5.4, we present the proposed pulse generator for event monitoring node. In Section 5.5 we provide the simulation results and discussion. Finally concluding remarks are given in Section 5.6.

5.2 SPP Preliminaries

Plasmon is the quantum state of plasma oscillation while surface plasmon is the collective oscillation of valence electrons at metal surfaces [108]. The coupling between surface plasmons and photons forms a quasi-particle called Sur- face Plasmon Polariton, which is an EM wave that travels along a metallic surface. SPPs in regular metals are always TM electromagnetic waves since a Transverse Electric (TE) polarized EM field would have its electric component normal to the interface as a consequence it could not interact with the free electrons of the Chapter 5 Pulse Generator for tiny IoNT Nodes 82 plasmonic materials [109]. Plasmonic materials are fascinating research field which has attracted the at- tention from multiple disciplines and has applied in a lot of applications because of their unique properties. Chemical and biological sensors have been developed by utilizing SPPs resonance as a sensitive function of the surrounding material [109]. In addition, SPPs have also been used in photon-scanning microscopy and imaging [109]. One of the essential features of plasmonic materials such as graphene is to enable THz band nanocommunication among nanosystems [6]. Groundbreaking experiments regarding two-dimensional (2D) materials have shown that graphene is an exceptional 2D plasmonic platform and it can be used as an alternative to noble metals [110]. Graphene is a one-atom thick layer of carbon atoms arranged in a honeycomb 2D crystal lattice possessing unique electrical and optical properties, that is; graphene has the largest known electrical conductivity. Graphene has been proposed as a building material for plasmonic nanoantennas [6] and nanotransceivers [13]. Moreover, doped graphene can support propagation of tightly confined SPPs in the terahertz band (0.1-10 THz) at room temperature [37, 79], enabling the miniaturization of nanotransceivers suited for wireless nanocommunication. The excitation of SPPs can happen in several ways, which we discussed in chapter 2. The generated SPPs lead to femtoseconds long pulses. Each technique requires energy to generate SPPs. The energy can be harvested using a suitable nanoscale energy harvester. Many types of nanoscale energy harvesting methods have been developed for harvesting different types of energy, e.g. piezoelectric [20,21,40], thermoelectric [28], triboelectric [22–24] and pyroelectric [25,41]. The principle idea of generating SPPs is that we need to be on the resonance frequency and wavenumber to have the excitation. The resonance frequency at which SPPs are excited depends on the material conductivity. For instance, noble metals such as gold or silver support SPP waves in visible frequencies [111,112]. Unlike, noble metals, graphene supports SPPs at frequencies in the THz band Chapter 5 Pulse Generator for tiny IoNT Nodes 83

(0.1-10 THz). By allowing wave propagation at the THz band (0.1-10 THz), the SPP waves can be generated with less power which might be feasible for severely resource constrained nodes operated by nanoscale energy harvesters. The terahertz wave then propagates towards the antenna and is finally radiated [27]. Recent studies confirm that SPPs provide a dominant contribution to the antenna radiation [107, 113]. However, SPPs experience damping and, in turn, low resonance which can eventually affect the antenna radiation. We suggest how to improve the SPP resonance using our proposed pulse generator in Section 5.5.

5.3 Discussion

We discussed different configurations for SPP excitation in Chapter 2. The choice of the most suitable method depends on the operating frequency, the type of source, the metal permittivity, and the system configuration. However, integrating any of such configurations in event monitoring node further imposes size, complexity, computational and power limitations. As it has been discussed previously in Chapter 2, in visible regime, Kretschmann Raether configuration is used with noble metals, silver and gold films, for photoemission effect [114], SPP imaging for bio-molecular interaction analysis [85], for testifying of blood cell analysis, serology analysis, and im- munology [115]. On the other hand, Otto configuration is a suitable method in infrared range because of the thickness d of the gap between the denser dielectric and metal, increases accordingly to incidence wavelength [115]. Further- more, the finite film thickness in Otto or Kretschmann configurations induces losses which reduces SPP intensity and propagation [116]. The grating coupler method is suitable for restricting the dispersion distribution of excited SPP to a very narrow frequency window [90]. This method further requires the special structure of fabricating very fine slits on the dielectric interface which may be difficult to produce at the nanoscale. These studies do not provide solutions Chapter 5 Pulse Generator for tiny IoNT Nodes 84 to generate SPPs in graphene in the THz band (0.1 - 10THz). To excite SPPs in the THz band, HEMT circuit is introduced with the graphene-based gate in [13]. However, our work is different from these studies. The focus of our work is to investigate a novel concept for pulse generator which should be simple and can directly excite SPP on graphene surface in the THz band without the need to fabricate specialized circuits on the graphene surface. Using COMSOL Multiphysics, we show that SPPs in the THz band can be generated using the near-field method by which the matching condition for excitation of SPPs can be easily satisfied. The performance of the proposed near-field excitation method in the THz band is optimized via numerical simulations (see Section 5.5). This includes the choice of the source, its phase angle, the chemical potential, and frequency in the THz range. The proposed method is a good candidate for a low complexity realization of a THz pulse generator in event monitoring nodes presented in Chapter 3 and 4.

5.4 Pulse Generation for Event Monitoring Node.

This section presents the proposed pulse generator of the event monitoring node, as shown in Figure 5.1. However, note that the methodology is general and can be generalized to any sensor node, i.e., monitoring human lung cells [2, 57], chemical reactions [55, 64] and so on. Assuming the energy harvester harvests the δ amount of energy from an event in the immediate environment. This energy can be used to drive the pulse generator to generate a THz pulse (see Figure 5.1). However, due to the simple structure and low complexity of event monitoring node, the pulse generator should be kept simple to generate the pulses in the terahertz band (THz) without any extra circuit on board. In this numerical study, we now investigate the pulse generator, to generate SPPs on the graphene surface using COMSOL Multiphysics. The schematic Chapter 5 Pulse Generator for tiny IoNT Nodes 85 model of the proposed pulse generator is shown in Figure 5.1. The pulse generator generates the SPPs, which then propagates towards the plasmonic antenna. A plasmonic antenna converts SPPs into propagating electromagnetic wave and finally radiates into the free space. The antenna modeling is beyond the scope of this work. As it has been discussed previously, we use near-field excitation method to generate SPPs in the THz band which is simple and can produce electromagnetic radiation with a wide range of high wavenumber. Hence, the coupling condition can be easily satisfied and consequently, the SPP wave can be excited which we will show in the next section. Excitation source EM wave R=20nm Evanescent wave d=30nm y Nanoantenna

Graphene X SPP

Figure 5.1: Illustration of a pulse generator where SPP is excited on graphene sheet using the near-ﬁeld method. A nanoantenna takes the SPP wave and radiates it into a free-space EM wave.

In our modeling, we consider a surface conductivity model for inﬁnitely large graphene sheets. Since graphene is a real 2D material, it is introduced as a boundary between two interfaces, namely as a surface current given by the Ohm law

~J = σgE~, where σg is the complex optical conductivity of graphene [93]. In low frequencies, i.e. THz, σg is dominated by intraband transitions and can be approximated very well by the Drude model as [117]:

je2µ σ (ω) = c , (5.1) g ~2 1 π (ω + jτ− ) where e is the electron charge, ~ is the reduced Planck constant, ω is the angular frequency, τ is the scattering rate equal to τ = 0.5 ps and µc is the chemical Chapter 5 Pulse Generator for tiny IoNT Nodes 86 potential of graphene [93, 117]. In what follows we consider a free standing graphene sheet, viz. graphene is surrounded by air with permittivity εair = 1. As a part of a future work, we will also evaluate the proposed model for finite size graphene sheet on dielectric material. Nevertheless, methodology in this paper is entirely general. For exciting SPPs, we use the near-field method. In particular, very close to the graphene sheet we put a TM evanescent EM source, E~, with electric field components given by        −   Ex   cos(n(φ θ))         −   Ey  =  sin(n(φ θ))  (5.2)          Ez   0  where n is an integer and defines the evanescent character of source, viz. n = 1 corresponds to a dipole source, n = 2 to quadrupole and so on; θ is a constant − − phase and φ is a spatial depended phase defined as φ = arctan2(y ys,x xs) where (xs,ys) are the Cartesian coordinates of the source center and arctan2() is the inverse function of the tangent function with two arguments. Using the near-field excitation method, we aim to explore the dependence of the plasmon resonance on the frequency f ,on the doping amount µc of graphene, on the source phase θ and evanescent parameter n.

5.5 Simulation and Numerical Results

In this section, we present numerical results of exciting SPPs using proposed pulse generator. The numerical simulations have been performed by virtue of COMSOL for a 2D space. The graphene is considered as a surface boundary condition with conductivity given by Eq. (5.1) and extended along to the x axis [93]. The length of graphene, along with the x-direction, is chosen to be large enough (200 µm) to be considered as an inﬁnite plane rather than a graphene nanoribbon. A circle Chapter 5 Pulse Generator for tiny IoNT Nodes 87 of radius R = 20 nm has been located 30 nm above the graphene sheet. On the boundaries of the circle, we apply a TM electric ﬁeld according to the Eq. (5.2), which is acting as a TM polarized excitation source since R << λ (in the THz regime). Furthermore, perfect match layer (PML) boundary conditions have been imposed on all the edges.

5.5.1 Impact of the frequency on the amplitude of SPP

In Figure 5.2, we first present COMSOL result at f = 10 THz, µc = 0.5 eV and for θ = π/2. The magnetic field intensity I, associated with the only non-zero com- | |2 ponent of magnetic field I¯ = Hz , is represented with color in Figure 5.2(a) and with the solid line in Figure 5.2(b), normalized to the maximum value of intensity. Plasmon resonance and propagating SPP waves are observed. The plasmon wavelength is calculated to be λsp = 5 µm, which is six times smaller than the light wavelength revealing the sub-wavelength character of the graphene plasmons. We now proceed by investigating the plasmon resonance (amplitude) for several values of frequency in the THz regime, by keeping constant the rest parameters as in Figure 5.2. The intensity I¯ is an excellent metric to measure plasmon resonance, that is, a higher EM amplitude denotes a better plasmon resonance. Figure 5.3 indicates the amplitude of I,¯ normalized to its maximum value, as a function of frequency. We observed that a better plasmon resonance is achieved in higher frequencies of the THz band.

5.5.2 Impact of the chemical potential on the SPP resonance

Furthermore, we numerically investigate how the plasmon resonance depends on graphene doping at the certain frequency of f = 10 THz and for the evanescent source with n = 1 and θ = π/2. In Figure 5.4, we present the maximum amplitude of the intensity I¯ for a range of chemical potential. We observe from Figure 5.4 Chapter 5 Pulse Generator for tiny IoNT Nodes 88

Figure 5.2: The magnetic intensity I demonstrates the plasmon resonance on a doped graphene monolayer (dotted white line). The spatial distribution of I¯ is represented in arbitrary values (a.u.) by (a) colorbar in x-y plane and by (b) with a solid blue line along the graphene layer (y = 0). An evanescent TM EM dipole source (n=1) of frequency f = 10 THz and with phase angle θ = π/2 is located at Xs = 100 µm and Ys = 30 nm above the graphene layer, indicated by a tiny white spot. The chemical potential of graphene is considered µc = 0.5 eV. The SPP wavelength is calculated to be λsp = 5 µm.

that the plasmon resonance is an increasing function of µc when it is less than

0.4 eV. In addition, we observe an abnormal behavior for values µc < 0.4 eV. For larger values of µc > 0.43 eV, the plasmon resonance is decreasing monotonically with µc. Chapter 5 Pulse Generator for tiny IoNT Nodes 89

Figure 5.3: The maximum value of I¯ for several values of frequency is presented revealing that the higher the frequency the better plasmon resonance is achieved.

Figure 5.4: The maximum value of I¯ for several values of doping µc is shown revealing that at a certain frequency of f = 10 THz, the plasmon resonance is a increasing function of the doping for values µc < 0.4 eV. Chapter 5 Pulse Generator for tiny IoNT Nodes 90

5.5.3 Impact of the phase angle of the evanescent source on SPP resonance

The dependence of plasmon resonance on the source phase angle θ is displayed in Figure 5.5. We observe that the plasmon intensity I¯ follows a normal distribution, with respect to θ, with the best plasmon resonance achieved at θ = π/2. This result indicates the strong dependence of the phase angle of the excitation source on the amplitude of the SPPs. Subsequently, when we use evanescent sources for excitation of SPPs, we have to take into account the optimization phase angle.

Figure 5.5: The maximum value of I¯ for several as function as the phase angle θ of the source is demonstrated with blue solid line. The best plasmon resonance is observed for the θ = π/2.

5.5.4 Impact of the type of evanescent source on the SPP resonance

Finally, by studying how the evanescent parameter n affects the plasmon resonance, we find that as the excitation source becomes more evanescent (n > 1) the Chapter 5 Pulse Generator for tiny IoNT Nodes 91 plasmon resonance decrease dramatically. In Figure 5.6 we present the profile I¯ along the graphene layer for several values of the evanescent parameter. We observe that the plasmon resonance decreases dramatically as n increases. This is because of the quality of the plasmon excitation is inverse proportional to the evanescent parameter of the source (n), thus as larger is the n as faster the SPP amplitude drops.

Figure 5.6: The plasmon proﬁle intensity I¯ along the graphene sheet is represented for several values of evanescent parameter n of the source, that is, for n = 1, n = 2 and n = 3.

5.6 Chapter Summary

In this work, we proposed a new design of the pulse generator for IoNT event monitoring nodes. Using COMSOL Multiphysics, which is a commercial Maxwell’s equations solver in the frequency domain, we show that pulses in the THz band can be generated using the near-field method by which the matching condition for excitation of SPPs can be easily satisfied. Near-field method does not require any special structure including prism [84, 86], periodic slits [80, 84], Chapter 5 Pulse Generator for tiny IoNT Nodes 92 or high-electron-mobility-transistor (HEMT) [13, 81], therefore can be suitable for Tiny IoNT nodes. The performance of the proposed pulse generator in the THz band is optimized to maximize the SPPs resonance via numerical simulations. This includes the choice of the source, its phase angle, the chemical potential, and the frequency. Chapter 6

Conclusion and Future Works

This chapter concludes the thesis by highlighting the key contributions of the research undertaken and discuss several remaining challenges for future work.

6.1 Concluding Remarks

This thesis has conducted a systematic study to solve the power issue associated with resource-constrained nodes in self-powered Wireless Nanosensor Networks. In particular, we focused on designing pulses that will not only consume exactly the amount of energy harvested from the event but also accurately convey both event type and its location to a distant receiver. The key contributions of this thesis are:

1. We propose a truly energy-neutral event monitoring framework for IoNT, called eNEUTRAL IoNT, that allows the sensors to transmit the complete information about a detected event, including the event type and its location, using only the amount of energy harvested from the events.

2. We then propose a single-pulse transmission scheme as the ﬁrst implementation of eNEUTRAL IoNT. Such scheme uses a single pulse containing the entire energy harvested from the event, but manipulates its pulse width or

93 Chapter 6 Conclusion and Future Works 94

duration to create a unique pulse amplitude for a given combination of event type and its location. To minimize classiﬁcation error at the receiver, we develop an optimization model to optimize pulse durations in the single-pulse approach. Numerical results show that single-pulse scheme achieving 99% event type and location detection accuracy in 10-node network monitoring two diﬀerent event types for a radius of 22 mm.

3. The single-pulse transmission enables multiple sensorless nodes to communicate with one remote station. However, increasing number of nodes and event types results in an increasing number of classes to classify which severely affect the performance of the single-pulse scheme. In chapter 4, we present and analyse a more generalized method of the eNEUTRAL IoNT framework that provides accurate events monitoring at longer distances. Such a scheme transmits two pulses using the harvested energy from an event. The energy of the first pulse uniquely defines a location and the second pulse uses the remaining energy to identify event types. We developed an optimization model for pulse energy allocation to optimize pulse energies improve the detection accuracy by minimizing classification error at the receiver.

4. Extensive numerical experiments have shown that the dual-pulse system significantly outperforms single pulse either in terms of classification error for a given network radius or regarding achievable network radius for target reliability. We find that the dual-pulse approach is achieving 99% accuracy for detecting both location and event type in 10-node network monitoring two different event types for a radius of 30 mm.

5. Due to the simple structure and low complexity of event monitoring node, designing a pulse generator for Tiny IoNT node is a challenging problem. Since most of the excitation methods require complex structures and con- ﬁgurations to generate SPPs, which may not be feasible for such resource Chapter 6 Conclusion and Future Works 95

constrained nodes. In this direction, we design and evaluate a pulse generator where we used the near-ﬁeld excitation method to excite SPPs on the graphene surface in the THz band. Such a method can generate SPPs easily on the graphene surface without any prism, periodic slits and extra circuit on board. We further investigated the chemical doping, characteristics of the evanescent source, namely the type, and the phase angle to achieve high resonance (amplitude) SPPs on the graphene surface.

6.2 Future Directions

Some of the future research directions can be summarised as follows:

1. Future work will focus on developing proof-of-concept prototypes of the proposed energy-neutral data updates for IoT. As nanoscale antennas, transmitters, and receivers are still not available; we hope that the available impulse radio ultra-wideband (IR-UWB) hardware can be used for proof- of-concept demonstrations of pulse-based energy-neutral data updates.

2. To design eNEUTRAL IoNT system, we assume that nodes transmit at different time slots be using random time delay at each node. To relax this assumption, we need to study the case when pulses of different events over- lap at the receiver. Such a case makes the classification difficult, but how to do this is a challenge which is a case for future work. One possible solution could be multivariate analysis at the receiver to study the case of the overlapped pulses.

3. In Chapter 3 and 4, we study the feasibility of eNEUTRAL IoNT for networks where sensors are placed in equidistant from the remote station. As a potential future work, the eNEUTRAL IoNT methods can be investigated against distance heterogeneous networks in IoNT systems. Chapter 6 Conclusion and Future Works 96

4. In single-pulse and dual-pulse methods, we focused on designing pulses that will not only consume exactly the amount of energy harvested from the event but also accurately convey both event type and its location to a distant receiver. However, the sensed information can also be encoded in the pulse position of the transmitted pulse. In such a scheme, the pulse position can serve as the address of the receiver. A critical limitation of pulse position modulation is that requires synchronization between transmitting node and a receiver. How to address the synchronization issue in eNEUTRAL IoNT using PPM would constitute an interesting future work.

5. In Chapter 5, we propose and design a simple architecture for pulse generator in COMSOL Multiphysics which can be a candidate for a tiny IoNT event monitoring nodes. Modelling energy harvesting component, channel and a receiver in COMSOL Multiphysics for the realization of an operational event monitoring system could be a basis for future researches. Bibliography

[1] Ian F Akyildiz and Josep Miquel Jornet. Electromagnetic wireless nanosensor networks. Nano Communication Networks, 1(1):3–19, 2010.

[2] Eisa Zarepour, Najmul Hassan, Mahbub Hassan, Chun Tung Chou, and Majid Ebrahimi Warkiani. Design and analysis of a wireless nanosensor network for monitoring human lung cells. In Proceedings of the 10th EAI International Conference on Body Area Networks, pages 139–145. ICST (In- stitute for Computer Sciences, Social-Informatics and Telecommunications Engineering), 2015.

[3] Eisa Zarepour, Adesoji A Adesina, Mahbub Hassan, and Chun Tung Chou. Innovative approach to improving gas-to-liquid fuel catalysis via nanosensor network modulation. Industrial & Engineering Chemistry Research, 53(14):5728–5736, 2014.

[4] Armita Afsharinejad, Alan Davy, Brendan Jennings, and Conor Brennan. Performance analysis of plant monitoring nanosensor networks at thz frequencies. IEEE Internet of Things Journal, 3(1):59–69, 2016.

[5] Josep Miquel Jornet and Ian F Akyildiz. Graphene-based nano-antennas for electromagnetic nanocommunications in the terahertz band. In Pro- ceedings of the Fourth European Conference on Antennas and Propagation, pages 1–5. IEEE, 2010.

97 References 98

[6] Josep Miquel Jornet and Ian F Akyildiz. Graphene-based plasmonic nanoantenna for terahertz band communication in nanonetworks. IEEE Journal on selected areas in communications, 31(12):685–694, 2013.

[7] Ian F Akyildiz and Josep Miquel Jornet. The internet of nano-things. IEEE Wireless Communications, 17(6):58–63, 2010.

[8] Suk Jin Lee, Changyong Jung, Kyusun Choi, and Sungun Kim. Design of wireless nanosensor networks for intrabody application. International Journal of Distributed Sensor Networks, 11(7):176761, 2015.

[9] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, and Adesoji A Adesina. Remote detection of chemical reactions using nanoscale terahertz communication powered by pyroelectric energy harvesting. In Proceedings of the Second Annual International Conference on Nanoscale Computing and Com- munication, page 8. ACM, 2015.

[10] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, and Adesoji A Adesina. Nano-scale sensor networks for chemical catalysis. In Nanotechnology (IEEE-NANO), 2013 13th IEEE Conference on, pages 61–66. IEEE, 2013.

[11] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, and Adesoji A. Adesina. Semon: Sensorless event monitoring in self-powered wireless nanosensor networks. ACM Transactions on Sensor Networks, 13(15), 2017.

[12] Josep Miquel Jornet, Ian F Akyildiz, et al. Fundamentals of electromagnetic nanonetworks in the terahertz band. Foundations and Trends® in Network- ing, 7(2-3):77–233, 2013.

[13] Josep Miquel Jornet and Ian F Akyildiz. Graphene-based plasmonic nanotransceiver for terahertz band communication. In Antennas and Propaga- tion (EuCAP), 2014 8th European Conference on, pages 492–496. IEEE, 2014. References 99

[14] Massimiliano Pierobon, Josep Miquel Jornet, Nadine Akkari, Suleiman Al- masri, and Ian F Akyildiz. A routing framework for energy harvesting wireless nanosensor networks in the terahertz band. Wireless networks, 20(5):1169–1183, 2014.

[15] Qammer H Abbasi, Ke Yang, Nishtha Chopra, Josep Miquel Jornet, Najah Abed Abuali, Khalid A Qaraqe, and Akram Alomainy. Nano- communication for biomedical applications: A review on the state-of-the- art from physical layers to novel networking concepts. IEEE Access, 4:3920– 3935, 2016.

[16] Kaikai Chi, Yi-hua Zhu, Xiaohong Jiang, and Victor CM Leung. Energy- eﬃcient preﬁx-free codes for wireless nano-sensor networks using ook modulation. IEEE Transactions on Wireless Communications, 13(5):2670– 2682, 2014.

[17] Kaikai Chi, Yi-hua Zhu, Yanjun Li, Daqiang Zhang, and Victor CM Leung. Coding schemes to minimize energy consumption of communication links in wireless nanosensor networks. IEEE Internet of Things Journal, 3(4):480– 493, 2016.

[18] Josep Miquel Jornet and Ian F Akyildiz. Joint energy harvesting and communication analysis for perpetual wireless nanosensor networks in the terahertz band. IEEE Transactions on Nanotechnology, 11(3):570–580, 2012.

[19] Josep Miquel Jornet and Ian F Akyildiz. Femtosecond-long pulse-based modulation for terahertz band communication in nanonetworks. IEEE Transactions on Communications, 62(5):1742–1754, 2014.

[20] Zhong Lin Wang. Zno nanowire and nanobelt platform for nanotechnology. Materials Science and Engineering: R: Reports, 64(3):33–71, 2009. References 100

[21] Zhong Lin Wang and Jinhui Song. Piezoelectric nanogenerators based on zinc oxide nanowire arrays. Science, 312(5771):242–246, 2006.

[22] Sihong Wang, Long Lin, and Zhong Lin Wang. Nanoscale triboelectric- eﬀect-enabled energy conversion for sustainably powering portable electronics. Nano letters, 12(12):6339–6346, 2012.

[23] Weiqing Yang, Jun Chen, Guang Zhu, Jin Yang, Peng Bai, Yuanjie Su, Qing- sheng Jing, Xia Cao, and Zhong Lin Wang. Harvesting energy from the natural vibration of human walking. ACS nano, 7(12):11317–11324, 2013.

[24] Mengdi Han, Xiao-Sheng Zhang, Xuming Sun, Bo Meng, Wen Liu, and Haixia Zhang. Magnetic-assisted triboelectric nanogenerators as self- powered visualized omnidirectional tilt sensing system. Scientiﬁc reports, 4, 2014.

[25] Ya Yang, Yusheng Zhou, Jyh Ming Wu, and Zhong Lin Wang. Single micro/nanowire pyroelectric nanogenerators as self-powered temperature sensors. ACS nano, 6(9):8456–8461, 2012.

[26] Bhuvana Krishnaswamy and Raghupathy Sivakumar. Source addressing and medium access control in bacterial communication networks. In Pro- ceedings of the Second Annual International Conference on Nanoscale Comput- ing and Communication, page 1. ACM, 2015.

[27] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, and Adesoji A Adesina. Energy-harvesting nanosensor networks: Eﬃcient event detection. IEEE Nanotechnology Magazine, 10(4):4–12, 2016.

[28] Gael Sebald, Daniel Guyomar, and Amen Agbossou. On thermoelectric and pyroelectric energy harvesting. Smart Materials and Structures, 18(12):125006, 2009. References 101

[29] Eisa Zarepour. Eﬃcient Communication Protocols for Wireless Nanoscale Sen- sor Networks. PhD thesis, University of New South Wales, Sydney, Aus- tralia, 2015.

[30] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, and Adesoji A Adesina. Frequency hopping strategies for improving terahertz sensor network performance over composition varying channels. In World of Wireless, Mobile and Multimedia Networks (WoWMoM), IEEE 15th International Symposium on a, pages 1–9. IEEE, 2014.

[31] Raul Gomez Cid-Fuentes, Josep Miquel Jornet, Ian F Akyildiz, and Eduard Alarcon.´ A receiver architecture for pulse-based electromagnetic nanonetworks in the terahertz band. In IEEE International Conference on Communi- cations (ICC), pages 4937–4942. IEEE, 2012.

[32] Anand Nayyar, Vikram Puri, and Dac-Nhuong Le. Internet of nano things (iont): Next evolutionary step in nanotechnology. Nanoscience and Nan- otechnology, 7(1):4–8, 2017.

[33] Sebastian Canovas-Carrasco, Antonio-Javier Garcia-Sanchez, Felipe Garcia-Sanchez, and Joan Garcia-Haro. Conceptual design of a nano- networking device. Sensors, 16(12):2104, 2016.

[34] Fenghui Yao, Mohamed Saleh Zein-Sabatto, Guifeng Shao, Mohammad Bo- druzzaman, and Mohan Malkani. Nanosensor data processor in quantum- dot cellular automata. Journal of Nanotechnology, 2014, 2014.

[35] Christian Falconi, Arnaldo DAmico, and Zhong Lin Wang. Wireless joule nanoheaters. Sensors and Actuators B: Chemical, 127(1):54–62, 2007.

[36] Ignacio Llatser, Albert Cabellos-Aparicio, Eduard Alarcon,´ Josep Miquel Jornet, Albert Mestres, Heekwan Lee, and Josep Sole-Pareta.´ Scalability of References 102

the channel capacity in graphene-enabled wireless communications to the nanoscale. IEEE Transactions on Communications, 63(1):324–333, 2015.

[37] Long Ju, Baisong Geng, Jason Horng, Caglar Girit, Michael Martin, Zhao Hao, Hans A Bechtel, Xiaogan Liang, Alex Zettl, Y Ron Shen, et al. Graphene plasmonics for tunable terahertz metamaterials. Nature nanotechnology, 6(10):630–634, 2011.

[38] Andy Chung, Jamal Deen, Jeong-Soo Lee, and M Meyyappan. Nanoscale memory devices. Nanotechnology, 21(41):412001, 2010.

[39] Zhong Lin Wang and Ying Liu. Piezoelectric Eﬀect at Nanoscale, pages 1–18. Springer Netherlands, Dordrecht, 2014.

[40] Zhong Lin Wang. Energy harvesting for self-powered nanosystems. Nano Research, 1(1):1–8, 2008.

[41] Ya Yang, Wenxi Guo, Ken C Pradel, Guang Zhu, Yusheng Zhou, Yan Zhang, Youfan Hu, Long Lin, and Zhong Lin Wang. Pyroelectric nanogenerators for harvesting thermoelectric energy. Nano letters, 12(6):2833–2838, 2012.

[42] Guang Zhu, Rusen Yang, Sihong Wang, and Zhong Lin Wang. Flexible high-output nanogenerator based on lateral zno nanowire array. Nano letters, 10(8):3151–3155, 2010.

[43] Mengdi Han, Xiao-Sheng Zhang, Bo Meng, Wen Liu, Wei Tang, Xuming Sun, Wei Wang, and Haixia Zhang. r-shaped hybrid nanogenerator with enhanced piezoelectricity. ACS nano, 7(10):8554–8560, 2013.

[44] Bikram Bhatia, Anoop R Damodaran, Hanna Cho, Lane W Martin, and William P King. High-frequency thermal-electrical cycles for pyroelectric energy conversion. Journal of Applied Physics, 116(19):194509, 2014. References 103

[45] LL Fields, JP Zheng, Y Cheng, and P Xiong. Room-temperature low-power hydrogen sensor based on a single tin dioxide nanobelt. Applied Physics Letters, 88(26):263102, 2006.

[46] William Mickelson, Allen Sussman, and Alex Zettl. Low-power, fast, se- lective nanoparticle-based hydrogen sulﬁde gas sensor. Applied Physics Let- ters, 100(17):173110, 2012.

[47] T Helbling, S Drittenbass, L Durrer, C Roman, and C Hierold. Ultra small single walled carbon nanotube pressure sensors. In Micro Electro Mechan- ical Systems, 2009. MEMS 2009. IEEE 22nd International Conference on, pages 575–578. IEEE, 2009.

[48] Maggie QH Zhang, Carmen KM Fung, Gary CT Chow, Wen J Li, and Philip Leong. Parylene-c embedded cnt-based mems piezoresistive pressure sensors using dep nanoassembly. In Nanotechnology, 2006. IEEE-NANO 2006. Sixth IEEE Conference on, volume 2, pages 802–805. IEEE, 2006.

[49] Chien-Chih Lin and Jenn-Gwo Hwu. Performance enhancement of metal- oxide-semiconductor tunneling temperature sensors with nanoscale oxides by employing ultrathin al 2 o 3 high-k dielectrics. Nanoscale, 5(17):8090– 8097, 2013.

[50] Gilhwan Cha, Yanbing Jia, and Y Sungtaek Ju. High-power density pyroelectric energy harvesters incorporating switchable liquid-based thermal interfaces. In Micro Electro Mechanical Systems (MEMS), 2012 IEEE 25th International Conference on, pages 1241–1244. IEEE, 2012.

[51] Yuan Zhou, Daniel J Apo, and Shashank Priya. Dual-phase self-biased magnetoelectric energy harvester. Applied Physics Letters, 103(19):192909, 2013. References 104

[52] Josep Miquel Jornet and Ian F Akyildiz. Channel capacity of electromagnetic nanonetworks in the terahertz band. In Communications (ICC), 2010 IEEE International Conference on, pages 1–6. IEEE, 2010.

[53] Josep Miquel Jornet and Ian F Akyildiz. Channel modeling and capacity analysis for electromagnetic wireless nanonetworks in the terahertz band. IEEE Transactions on Wireless Communications, 10(10):3211–3221, 2011.

[54] YL Babikov, IE Gordon, SN Mikhailenko, LS Rothman, and SA Tashkun. Hitran on the web, a new tool for hitran spectroscopic data manipulation. In Proceedings of the 12th International HITRAN Conference, pages 29–31, 2012.

[55] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, and Adesoji A Adesina. Open-loop power adaptation in nanosensor networks for chemical reactors. IEEE Transactions on Molecular, Biological and Multi-Scale Communi- cations, 1(3):292–307, 2015.

[56] Raul¨ Gomez´ Cid-F. Wake-up transceiver architectures with novel symbol time synchronization schemes for electromagnetic nanonetworks. 2011.

[57] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, and Majid Ebrahimi Warkiani. Characterizing terahertz channels for monitoring human lungs with wireless nanosensor networks. Nano Communication Networks, 9:43– 57, 2016.

[58] Valeria Loscri and Anna Maria Vegni. An acoustic communication technique of nanorobot swarms for nanomedicine applications. IEEE transactions on nanobioscience, 14(6):598–607, 2015.

[59] Sylvain Martel. Magnetic microbots to ﬁght cancer. IEEE Spectrum, 25, 2012. References 105

[60] Geoﬀrey Von Maltzahn, Ji-Ho Park, Kevin Y Lin, Neetu Singh, Christian Schwoppe,¨ Rolf Mesters, Wolfgang E Berdel, Erkki Ruoslahti, Michael J Sailor, and Sangeeta N Bhatia. Nanoparticles that communicate in vivo to amplify tumour targeting. Nature materials, 10(7):545–552, 2011.

[61] Giuseppe Piro, Ke Yang, Gennaro Boggia, Nishtha Chopra, Luigi Alfredo Grieco, and Akram Alomainy. Terahertz communications in human tissues at the nanoscale for healthcare applications. IEEE Transactions on Nanotech- nology, 14(3):404–406, 2015.

[62] Falko Dressler and Stefan Fischer. Connecting in-body nano communication with body area networks: Challenges and opportunities of the internet of nano things. Nano Communication Networks, 6(2):29–38, 2015.

[63] Donghee Son, Jongha Lee, Dong Jun Lee, Roozbeh Ghaﬀari, Sumin Yun, Seok Joo Kim, Ji Eun Lee, Hye Rim Cho, Soonho Yoon, Shixuan Yang, et al. Bioresorbable electronic stent integrated with therapeutic nanoparticles for endovascular diseases. ACS nano, 9(6):5937–5946, 2015.

[64] Eisa Zarepour, Mahbub Hassan, Chun Tung Chou, Adesoji A Adesina, and Majid Ebrahimi Warkiani. Reliability analysis of time-varying wireless nanoscale sensor networks. In Nanotechnology (IEEE-NANO), 2015 IEEE 15th International Conference on, pages 63–68. IEEE, 2015.

[65] Alejandro Perez-de´ Luque and Diego Rubiales. Nanotechnology for para- sitic plant control. Pest management science, 65(5):540–545, 2009.

[66] Carmen I Moraru, Chithra P Panchapakesan, Qingrong Huang, Panl Takhistov, Sean Liu, and Jozef L Kokini. Nanotechnology: a new frontier in food science. Food Technology, 2003. References 106

[67] Armita Afsharinejad. A Cross-Layer Communications Framework for THz Band Plant Monitoring Nanonetworks. PhD thesis, Waterford Institute of Technology, 2016.

[68] Marcel Dicke and Ian T Baldwin. The evolutionary context for herbivore- induced plant volatiles: beyond the cry for help. Trends in plant science, 15(3):167–175, 2010.

[69] Luis Alfonso Franco-Gasca, Gilberto Herrera-Ruiz, Roc´ıo Peniche-Vera, Rene´ de Jesus´ Romero-Troncoso, and Wbaldo Leal-Tafolla. Sensorless tool failure monitoring system for drilling machines. International Journal of Machine Tools and Manufacture, 46(3):381–386, 2006.

[70] W Li, D Li, and J Ni. Diagnosis of tapping process using spindle motor current. International Journal of Machine Tools and Manufacture, 43(1):73– 79, 2003.

[71] Maximiliano O Sonnaillon, Guillermo Bisheimer, Cristian De Angelo, and Guillermo O Garc´ıa. Online sensorless induction motor temperature monitoring. IEEE transactions on Energy Conversion, 25(2):273–280, 2010.

[72] Rossi Maurizio, Tosato Pietro, et al. Energy neutral design of an iot system for pollution monitoring. In IEEE EESMS 2017. Institute of Electrical and Electronics Engineers Inc., 2017.

[73] Sara Khalifa, Mahbub Hassan, and Aruna Seneviratne. Pervasive self- powered human activity recognition without the accelerometer. In Per- vasive Computing and Communications (PerCom), IEEE International Confer- ence on, pages 79–86. IEEE, 2015.

[74] Sara Khalifa, Mahbub Hassan, Aruna Seneviratne, and Sajal K Das. Energy-harvesting wearables for activity-aware services. IEEE Internet Computing, 19(5):8–16, 2015. References 107

[75] Qiong Huo, Jayanthi Rao, and Subir Biswas. Pulse switching: Toward a packet-less protocol paradigm for event sensing. IEEE Transactions on Mo- bile Computing, 12(1):35–50, 2013.

[76] Saptarshi Das, Hadi Salehi, Yan Shi, Shantanu Chakrabartty, Rigob- erto Burgueno, and Subir Biswas. Towards packet-less ultrasonic sensor networks for energy-harvesting structures. Computer Communications, 101:94–105, 2017.

[77] Saptarshi Das and Subir Biswas. A scalable pulse protocol for structural health monitoring. In Communication Systems & Networks (COMSNETS), 2018 10th International Conference on, pages 206–213. IEEE, 2018.

[78] Hulin Zhang, Ya Yang, Yuanjie Su, Jun Chen, Chenguo Hu, Zhenkun Wu, Yan Liu, Ching Ping Wong, Yoshio Bando, and Zhong Lin Wang. Tri- boelectric nanogenerator as self-powered active sensors for detecting liquid/gaseous water/ethanol. Nano Energy, 2(5):693–701, 2013.

[79] Zhe Fei, AS Rodin, GO Andreev, W Bao, AS McLeod, M Wagner, LM Zhang, Z Zhao, M Thiemens, G Dominguez, et al. Gate-tuning of graphene plasmons revealed by infrared nano-imaging. Nature, 487(7405):82, 2012.

[80] Stefan Alexander Maier. Plasmonics: fundamentals and applications. Springer Science & Business Media, 2007.

[81] Josep Miquel Jornet and Albert Cabellos. On the feeding mechanisms for graphene-based thz plasmonic nano-antennas. In Nanotechnology (IEEE- NANO), 2015 IEEE 15th International Conference on, pages 168–171. IEEE, 2015. References 108

[82] Shenggang Liu, Chao Zhang, Min Hu, Xiaoxing Chen, Ping Zhang, Sen Gong, Tao Zhao, and Renbin Zhong. Coherent and tunable terahertz radiation from graphene surface plasmon polaritons excited by an electron beam. Applied Physics Letters, 104(20):201104, 2014.

[83] Ernst Jan R Vesseur, Javier Aizpurua, Toon Coenen, Alejandro Reyes- Coronado, Philip E Batson, and Albert Polman. Plasmonic excitation and manipulation with an electron beam. MRS bulletin, 37(8):752–760, 2012.

[84] Heinz Raether. Surface plasmons on gratings. In Surface plasmons on smooth and rough surfaces and on gratings, pages 91–116. Springer, 1988.

[85] Stefan Lof¨ as,˚ Magnus Malmqvist, Inger Ronnberg,¨ Esa Stenberg, Bo Lied- berg, and Ingemar Lundstrom.¨ Bioanalysis with surface plasmon resonance. Sensors and Actuators B: Chemical, 5(1-4):79–84, 1991.

[86] Andreas Otto. Excitation of nonradiative surface plasma waves in silver by the method of frustrated total reﬂection. Zeitschrift fur¨ Physik A Hadrons and nuclei, 216(4):398–410, 1968.

[87] SP Yushanov, LT Gritter, JS Crompton, and KC Koppenhoefer. Surface plasmon resonance. In COMSOL Conference, 2012.

[88] Erwin Kretschmann and Heinz Raether. Radiative decay of non radiative surface plasmons excited by light. Zeitschrift fur¨ Naturforschung A, 23(12):2135–2136, 1968.

[89] Yu V Bludov, Aires Ferreira, NMR Peres, and MI Vasilevskiy. A primer on surface plasmon-polaritons in graphene. International Journal of Modern Physics B, 27(10):1341001, 2013. References 109

[90] Pawel Karpinski and Andrzej Miniewicz. Surface plasmon polariton excitation in metallic layer via surface relief gratings in photoactive polymer studied by the ﬁnite-diﬀerence time-domain method. Plasmonics, 6(3):541–546, 2011.

[91] Albert Cabellos-Aparicio, Ignacio Llatser, Eduard Alarcon, Allen Hsu, and Tomas Palacios. Use of terahertz photoconductive sources to characterize tunable graphene rf plasmonic antennas. IEEE Transactions on Nanotech- nology, 14(2):390–396, 2015.

[92] Marios Mattheakis, Thomas Oikonomou, Mario I Molina, and George P Tsironis. Phase transition in pt symmetric active plasmonic systems. IEEE Journal of Selected Topics in Quantum Electronics, 22(5):76–81, 2016.

[93] Marios Mattheakis, Constantinos A Valagiannopoulos, and Efthimios Kaxiras. Epsilon-near-zero behavior from plasmonic dirac point: The- ory and realization using two-dimensional materials. Physical Review B, 94(20):201404, 2016.

[94] Stefano Basagni, M Yousof Naderi, Chiara Petrioli, and Dora Spenza. Wire- less sensor networks with energy harvesting. Mobile Ad Hoc Networking: The Cutting Edge Directions, pages 701–736, 2013.

[95] Sujesha Sudevalayam and Purushottam Kulkarni. Energy harvesting sensor nodes: Survey and implications. IEEE Communications Surveys & Tuto- rials, 13(3):443–461, 2011.

[96] Joseph A Paradiso and Mark Feldmeier. A compact, wireless, self-powered pushbutton controller. In International Conference on Ubiquitous Comput- ing, pages 299–304. Springer, 2001. References 110

[97] John Kymissis, Clyde Kendall, Joseph Paradiso, and Neil Gershenfeld. Par- asitic power harvesting in shoes. In Wearable Computers, 1998. Digest of Papers. Second International Symposium on, pages 132–139. IEEE, 1998.

[98] Josep Miquel Jornet and Ian F Akyildiz. Information capacity of pulse- based wireless nanosensor networks. In Sensor, Mesh and Ad Hoc Commu- nications and Networks (SECON), 2011 8th Annual IEEE Communications Society Conference on, pages 80–88. IEEE, 2011.

[99] Bingwei Lu, Ying Chen, Dapeng Ou, Hang Chen, Liwei Diao, Wei Zhang, Jun Zheng, Weiguo Ma, Lizhong Sun, and Xue Feng. Ultra-ﬂexible piezoelectric devices integrated with heart to harvest the biomechanical energy. Scientiﬁc reports, 5:16065, 2015.

[100] Canan Dagdeviren, Byung Duk Yang, Yewang Su, Phat L Tran, Pauline Joe, Eric Anderson, Jing Xia, Vijay Doraiswamy, Behrooz Dehdashti, Xue Feng, et al. Conformal piezoelectric energy harvesting and storage from motions of the heart, lung, and diaphragm. Proceedings of the National Academy of Sciences, 111(5):1927–1932, 2014.

[101] Sangmin Lee, Ronan Hinchet, Yean Lee, Ya Yang, Zong-Hong Lin, Gustavo Ardila, Laurent Montes,` Mireille Mouis, and Zhong Lin Wang. Ultrathin nanogenerators as self-powered/active skin sensors for tracking eye ball motion. Advanced Functional Materials, 24(8):1163–1168, 2014.

[102] Hongsan Sheng, Philip Orlik, Alexander M Haimovich, Leonard J Cimini, and Jinyun Zhang. On the spectral and power requirements for ultra- wideband transmission. In Communications, 2003. ICC’03. IEEE Interna- tional Conference on, volume 1, pages 738–742. IEEE, 2003.

[103] Raul¨ Gomez´ Cid-Fuentes and Ian F Akyildiz. Wake-up transceiver architectures with novel symbol time synchronization schemes for electromagnetic nanonetworks. Citeseer, 2011. References 111

[104] Robert F Mills and Glenn E Prescott. A comparison of various radiometer detection models. IEEE Transactions on Aerospace and Electronic Systems, 32(1):467–473, 1996.

[105] Ajeya Gupta, Michael Medley, and Josep Miquel Jornet. Joint synchronization and symbol detection design for pulse-based communications in the thz band. In Global Communications Conference (GLOBECOM), 2015 IEEE, pages 1–7. IEEE, 2015.

[106] Erez Graitzer, Gunter Antesberger, Guy Ben-Zvi, Avi Cohen, Vladimir Dmitriev, and Stephanie Winkelmeier. Correcting image placement errors using registration control (regc) technology. In Optical Microlithog- raphy XXIV, volume 7973, page 797312. International Society for Optics and Photonics, 2011.

[107] Hongwei Jia, Haitao Liu, and Ying Zhong. Role of surface plasmon polaritons and other waves in the radiation of resonant optical dipole antennas. Scientiﬁc reports, 5:8456, 2015.

[108] JM Pitarke, VM Silkin, EV Chulkov, and PM Echenique. Theory of surface plasmons and surface-plasmon polaritons. Reports on progress in physics, 70(1):1, 2006.

[109] Dror Sarid and William Challener. Modern introduction to surface plasmons: theory, Mathematica modeling, and applications. Cambridge Univer- sity Press, 2010.

[110] AN Grigorenko, Marco Polini, and KS Novoselov. Graphene plasmonics. Nature photonics, 6(11):749, 2012.

[111] Mona Nafari and Josep Miquel Jornet. Metallic plasmonic nano-antenna for wireless optical communication in intra-body nanonetworks. In Pro- ceedings of the 10th EAI International Conference on Body Area Networks, References 112

pages 287–293. ICST (Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering), 2015.

[112] Mubashir Husain Rehmani and Al-Sakib Khan Pathan. Emerging communication technologies based on wireless sensor networks: current research and future applications. CRC Press, 2016.

[113] Chuan Liu, Haitao Liu, and Ying Zhong. Impact of surface plasmon polaritons and other waves on the radiation of a dipole emitter close to a metallic nanowire antenna. Optics Express, 22(21):25539–25549, 2014.

[114] Ch Macek, A Otto, and W Steinmann. Resonant photoemission from alu- minium ﬁlms at 5 ev photon energy due to nonradiative surface plasma waves. physica status solidi (b), 51(1):K59–K61, 1972.

[115] Dominique Barchiesi and Andreas Otto. Excitations of surface plasmon polaritons by attenuated total reﬂection, revisited. Riv. Nuovo Cimento, 36:173–209, 2013.

[116] A Drezet, A Hohenau, D Koller, A Stepanov, H Ditlbacher, B Stein- berger, FR Aussenegg, A Leitner, and JR Krenn. Leakage radiation microscopy of surface plasmon polaritons. Materials science and engineering: B, 149(3):220–229, 2008.

[117] Marios Mattheakis, Giorgos P Tsironis, and Efthimios Kaxiras. Graphene and active metamaterials: theoretical methods and physical properties. arXiv preprint arXiv:1704.01912, 2017. Appendix A

Acronyms

CSE Computer Science and Engineering School ...... iv

UNSW University of New South Wales ...... iv

IoT Internet of Thing ...... 1

IoNT Internet of Nano Thing ...... 1

WSN Wireless Sensor Network ...... 1

THz Terahertz...... 3

SPT Single Pulse Transmission ...... 7

DPT Dual Pulse Transmission ...... 8 eNEUTRAL Energy-neutral ...... 7

CTMA Continuous Time Moving Average ...... 7

EM Electro-Magnetic ...... 13

NSDP Nano-Sensor Data Processor...... 14

PE Processing Element ...... 15

GHz Gegahertz...... 15

HITRAN HIgh resolution TRANsmission molecular absorption database . 20

PSD Power Spectral Density...... 21

113 Appendix A - Acronyms 114

SNR Signal-to-Noise-Ratio ...... 22

ED Energy-based Detector ...... 22

BES Bioresorbable Electronic Stent...... 27

BAN Body Area Network ...... 27

FT Fischer Tropsch ...... 28

HIPVs Herbivore-induced Plant Volatiles ...... 28

SCPN Scalable Cellular Pulse Networking ...... 30

TENG Triboelectric Nanogenerator ...... 18

SEMON Sensorless Event Monitoring ...... 31

HEMT High Electron Mobility Transistor ...... 36 aJ attoJoule...... 54

TM Transverse Magnetic ...... 80

TE Transverse Electric ...... 81