Cross-Layer Simulation Analysis of a High-Precision Radiolocation System

Simulationsbasierte schichtübergreifende Systemanalyse eines hochpräzisen Mikrowellenortungssystems

Der Technischen Fakultät der Universität Erlangen-Nürnberg zur Erlangung des Grades DOKTOR-INGENIEUR

vorgelegt von Ralf Mosshammer

Erlangen – 2010 Als Dissertation genehmigt von der Technischen Fakultät der Universität Erlangen-Nürnberg

Tag der Einreichung: 14.1.2010 Tag der Promotion: 20.5.2010 Dekan: Prof. Dr.-Ing. Reinhard German 1. Berichterstatter: Prof. Dr. tech. Mario Huemer 2. Berichterstatter: Prof. Dr.-Ing. Jörn Thielecke Bedecke deinen Himmel, Zeus, Mit Wolkendunst! Und übe, Knaben gleich,

An Eichen dich und Bergeshöh’n!

Und meinen Herd, Um dessen Glut

Kehrt’ ich mein verirrtes Auge Zur Sonne,alswenndrüberwär

Hast du’s nicht alles selbst vollendet, I have of late–but wherefore I know not–lost all my mirth, forgone al Heilig glühend Herz? indeed it goes so heavily with my disposition that this goodly frame, the earth, seems to me a sterile promontory, this most excellent canopy, the air, look you, this brave o’erhanging firmament, this majestical roof fretted with golden fire, why, it appears no other thing to me than a foul and pestilent congregation of vapours. What a piece of work is a man! how noble in reason! how infinite in faculty! in form and moving how express and admirable! in action how like an angel! in apprehension how like a god! the beauty of the world! the paragon ofanimals!And yet, to me, what is this quintessence of dust? of dust?

of dust

dust

Abstract

In this work, a comprehensive analysis of a competitive and novel, high-precision local positioning system in the 5.8 GHz ISM band is presented. The RESOLUTION platform is built around a secondary-radar FMCW positioning system, supported by a commercial communications solution. The modular and flexible design of the platform allows for the support of various topologies and protocols, which is of supreme interest with regard to the very diverse application fields local positioning can serve. To gain an impression of performance figures with an eye towards actual product deployment, a cross-layer simulation tool was developed. This software allows for analysis of both physical layer properties and network dynamics which occur when multiple receivers are served within a fixed infrastructure. The signal theoretical foundations of secondary Frequency Modulated Contin- uous Wave (FMCW) radar are well established. With regard to this, research on the physical layer is limited to selected effects, with special attention on multipath propagation, which constitutes by far the largest error source. For comparative evaluation, both a model derived from system-specific measurements as well as a standardized model following IEEE 802.15.4a were integrated into simulation. The performance of Medium Access Control (MAC) layer algorithms for multi- user management have been analyzed along the most relevant parameters, such as time-to-fix, update rate, infrastructure utilization and efficiency. The seamless design of the physical and MAC layer simulators allows for complete integration and cross-layer optimization of the platform. Exemplary simulation results are provided. Access procedures derived from known communication models and adapted for the specific needs of positioning systems are described. Utilization of these methods allows for optimal system deployment according to specification parameters. This thesis constitutes an authoritative reference for the performance of the RESOLUTION local positioning system. Novel algorithms with cross-platform effects are investigated. The innovative simulation engine and the techniques used in its implementation are detailed. Comparative benchmarking results of various parameter sets and extreme values are presented and commented. Zusammenfassung

Diese Arbeit präsentiert eine umfassende Analyse eines neuartigen und hochprä- zisen lokalen Positionsbestimmungssystems im ISM-Band bei 5.8 GHz. Die RESOLUTION Plattform besteht aus einem Positionsbestimmungsmodul nach dem Sekundärradar-FMCW Prinzip, unterstützt von einer kommerziellen Kommunikationslösung. Die modulare und flexible Architektur der Plattform unterstützt verschiedene Topologien und Protokolle, was den Einsatz in einem breiten Applikationsfeld ermöglicht. Mit Hilfe einer schichtübergreifenden Simulationssoftware wurden die Parame- ter und Leistungsgrenzen des Systems bestimmt. Die Software erlaubt sowohl die Analyse physikalischer Leistungsparameter als auch der Netzwerkdynamiken, die in Präsenz mehrerer Empfangsmodule auftreten. Die signaltheoretischen Grundlagen von sekundärem FMCW Radar sind hinrei- chend bekannt. In Hinblick auf diese Tatsache beschränkt sich die Analyse der Bitübertragungsschicht auf ausgewählte Effekte mit besonderer Beachtung von Mehrwegeausbreitung, der mit Abstand größten Fehlerquelle im System. Zum Zweck einer vergleichenden Wertung wurden sowohl ein aus Messungen abgelei- tetes, systemspezifisches Kanalmodell als auch das standardisierte IEEE 802.15.4a Modell in die Simulation eingebunden. Die Leistungsgrenzen der Algorithmen der MAC-Schicht für Mehrnutzerzugriff wurden anhand relevanter Parameter wie Time-to-fix, Wiederholrate, Auslastung und Effizienz untersucht. Das ineinandergreifende Design der physikalischen und MAC-Schicht Simulatoren ermöglichte eine komplette Integration und schicht- übergreifende Optimierung der Plattform. Dazu werden relevante Ergebnisse prä- sentiert. Zugriffsverfahren, die von bekannten Modellen aus der Kommunikationstech- nik abgeleitet und für die spezifischen Bedürfnisse der Lokalisierung angepasst wurden werden beschrieben. Die Verwendung dieser Verfahren garantiert eine auf Spezifikationsparameter optimierte Systeminstallation. Diese Arbeit stellt eine verbindliche Referenz für die Leistungsbewertung des Positionsbestimmungssystems RESOLUTION dar. Neuartige Algorithmen, deren Betrachtung durch den Simulator ermöglicht wurde, werden vorgestellt und be- wertet. Die innovative Simulationsumgebung und die Techniken, die bei der Im- plementierung zum Tragen kamen werden im Detail beschrieben. Vergleichende Bewertungen verschiedener Parametersätze und Grenzfälle werden anhand von Simulationsergebnissen dargestellt und kommentiert. Contents

1. Introduction 1 1.1.Stateoftheart...... 2 1.2.Goalsofthethesis...... 3 1.3.Organization...... 3

2. Fundamentals of Wireless Positioning 5 2.1.Applicationclasses...... 6 2.2.Measurementprinciples...... 6 2.2.1.TimeofArrival(ToA)...... 7 2.2.2.RoundtripTimeofFlight(RToF)...... 8 2.2.3.TimeDiﬀerenceofArrival(TDoA)...... 8 2.2.4.AngleofArrival(AoA)...... 9 2.2.5.Fringesolutions...... 9 2.3.Physicallayer...... 10 2.3.1.Non-microwavesolutions...... 10 2.3.2. Microwave based solutions and FMCW ...... 12

3. The RESOLUTION Platform 15 3.1. RESOLUTION servicerequirements...... 15 3.2.Hybridpositioningandcommunication...... 16 3.3. RESOLUTION hardwarebase...... 18

4. Single Node Architecture and Performance Analysis 21 4.1. Basic receiver performance ...... 21 4.1.1.Figuresofmerit...... 24 4.1.2. AWGN performance...... 25 4.1.3.Basebandsignalevaluation...... 28

i 4.2.Hardwareimpairments...... 30 4.2.1.Phasenoise...... 30 4.2.2.Rampnonlinearity...... 32 4.3.Signalingimpairments...... 33 4.3.1.Multipathpropagation...... 34 4.3.2.Positioncalculation...... 43

5. Network Architecture and Quality of Service Aspects 49 5.1.Serviceandnetworkarchitecture...... 49 5.2. The MAC layer...... 53 5.2.1. Static channel access ...... 53 5.2.2. Dynamic channel access and novel access procedures . . . 54 5.3.Integratedperformanceassessment...... 57 5.3.1.Discreteeventsimulation...... 57 5.3.2. RESOLUTION protocols...... 60 5.3.3.Timingmodels...... 63 5.4.Simulationresults...... 69 5.4.1. Basic FIFO and C-ALOHA latencies...... 69 5.4.2.Secondaryperformanceparameters...... 71 5.4.3.Comparisonofpositioningprotocols...... 74 5.4.4.Updaterate...... 75 5.4.5. MAC layerimprovements...... 76

6. Conclusion and Outlook 83

A. The Active Reﬂector 85 A.1.ActivePulsedReﬂector...... 86 A.2. Medium access ...... 87

B. Object Oriented System Simulation Framework 89 B.1.Implementation...... 91 B.2.Deployment...... 92 B.3.Operation...... 93 B.4.Performance...... 93

C. Discrete Event Simulation Framework 95

D. Complex Envelope Simulation 99 Acronyms and Abbreviations

ACK Acknowledge (ﬂow control)

AR Active Reﬂector

A/D Analog to Digital Conversion

AGV Automated Guided Vehicle

ALOHA ALOHA access protocol

AoA Angle of Arrival

AWGN Additive White Gaussian Noise

BER Bit Error Rate

BS Base Station

C-ALOHA Controlled ALOHA

CDF Cumulative Density Function

CIR Channel Impulse Response

CPICH Common Pilot Channel

CSMA Carrier Sense Multiple Access

CSMA/CA Carrier Sense Multiple Access/Collision Avoidance

CTS Clear to Send (ﬂow control)

CW Continuous Wave iv Contents

DCF Distributed Coordination Function

DFT Discrete Fourier Transform

DIFS Distributed Interframe Space

DTFT Discrete Time Fourier Transform

ECB Equivalent Complex Baseband

EIRP Eﬀective Isotropic Radiated Power

EU European Union

FCC Federal Communications Commission

FDMA Frequency Division Multiple Access

FFT Fast Fourier Transform

FIFO First in/First out

FMCW Frequency Modulated Continuous Wave

FSK Frequency Shift Keying

GALILEO GALILEO satellite system

GEL Global Event List

GPS Global Positioning System

GSM Global System for Mobile Communications

HPLS High-Precision Location System

IEEE Institute of Electrical and Electronics Engineers

IF Intermediate Frequency

IFFT Inverse Fast Fourier Transform

IPDL Idle Periods in Downlink

ISM Industrial, Scientiﬁc and Medical

ISO/OSI International Standards Organizsation/Open Systems Interconnection

LBS Location Based Services Contents v

LOS Line of Sight

LPM Local Position Measurement

LPR Local-Positioning Radar

MAC Medium Access Control

MLE Maximum Likelihood Estimation

MMD Multi-Modulus Divider

MS Mobile Station

NACK Not Acknowledge (ﬂow control)

NF Noise Figure

NLOS Non-Line of Sight

PCF Position Calculation Function

PDA Personal Digital Assistant

PLL Phase Locked Loop

PRS Public Regulated Service

QoS Quality of Service

RESOLUTION Reconﬁgurable Systems for Mobile Communication and Positioning

RF Radio Frequency

RFID Radio Frequency Identiﬁcation

RSS Received Signal Strength

RToF Roundtrip Time of Flight

RTS Request to Send (ﬂow control)

RX Receiver

SAW Surface Acoustic Wave

SIRO Serve in Random Order

SNR Signal to Noise Ratio vi Contents

TDMA Time Division Multiple Access

TDoA Time Diﬀerence of Arrival

ToA Time of Arrival

TX Transmitter

UMTS Universal Mobile Telecommunications System

UWB Ultra-Wideband

VCO Voltage Controlled Oscillator

WAIT Wait command (ﬂow control)

WGN White Gaussian Noise

WLAN Wireless Local Area Network

WSN Wireless Sensor Network Einleitung

Die Entwicklung der integrierten Schaltung (Integrated Circuit, IC) leitete monu- mentale Veränderungen im Bereich der Datenverarbeitung und Kommunikation ein. Rasch fort- schreitende Verbesserungen in den Bereichen Rechengeschwindigkeit, Komponentenintegration und Stromverbrauch führten zu einer Welle an Produkten und Konsumgütern, die längst Teil industrieller Prozesse und des täglichen Lebens sind: das Internet, Mobiltelefonie, Satellitenna- vigation, Fernseh- und Radiosendungen, tragbare Medienwiedergabe, automatisierte Fertigung, Autopiloten, autonome Steuersysteme und Sensornetzwerke. Die Aussicht auf steigende Profite und anhaltender Absatzdruck führte zu einer zunehmen- den Fokussierung von Forschung und Entwicklung auf die Optimierung von Datendurchsatz, mit dem Ziel, sich der Shannon-Grenze möglichst unter Einhaltung vernünftiger Leistungsauf- nahme zu nähern und die Geräte zeitgleich durch Fortschritte in der Produktionstechnologie zu verkleinern. Getrieben von einer Vision autonomer Maschinenräume und kontextsensitiver Information drängte eine Technologie militärischer Provenienz zunehmend in die öffentliche Wahrnehmung: Positionsbestimmung. Für manche Experten stellen Sensornetzwerke den ultimativen Konvergenzpunkt von Kom- munikationstechnologien dar: stark dezentralisierte Gruppen von energiesparenden Sensorkno- ten mit verteilten Kommunikationsmöglichkeiten. Eine derartige Technologie könnte breite Anwendung in Bereichen wie Landwirtschaft, Umweltüberwachung, Gebäudeautomatisierung, Schlachtfeldüberwachung und industrieller Steuerung finden. Für die meisten dieser Applikati- onsfelder ergeben Sensordaten nur im Zusammenhang mit geographischer oder tolopogischen Information Sinn. Eine weitere Anwendungsmöglichkeit von Positionsdaten ist die Versorgung von Mobilfunk- kunden mit kontextsensitiven Diensten. Zuletzt stellt die industrielle Verwertung von Positionsdaten ein für diese Arbeit herausra- gendes Feld dar. Die steigende Komplexität moderner Industrieanlagen schürt das Bedürfnis weiterer Automation von Transport und Verarbeitung. In dieser Arbeit wird die RESOLUTION Plattform – die Abkürzung steht für “Reconfigurable System for Mobile Communication and Positioning” – vorgestellt und analysiert. Hierbei handelt es sich um ein hybrides Lokalisierungs- und Kommunikationssystem, das sowohl in speziali- sierten Konsumgütern als auch industriellen Umgebung eingesetzt werden kann. Die Plattform umfasst mehrere Konfigurationen, basiert aber in jedem Fall auf dem Prinzip des sekundären linearen FMCW (Frequency Modulated Continuous Wave) Radars für Distanzmessungen. In diesem Fachbereich existiert einiges an Vorarbeit, wie im nächsten Abschnitt dargestellt.

1 Stand der Technik

Chirp-Signale als Kommunikations- oder Radarträger sind seit den Mittfünfzigern bekannt. Wegen der niedrigen Detektions- und Abhörwahrscheinlichkeit ist die Technologie vor allem im militärischen Bereich verbreitet [1]. Lokale Positionsbestimmung kooperativer Ziele, die für diese Arbeit relevante Anwendung, weist einige wichtige Abweichungen zu regulärer Radartechnologie auf. Zum einen versucht das ausgeleuchtete Ziel aktiv, die Detektion zu erleichtern und regeneriert und reflektiert das einfal- lende Signal oder empfängt es und antwortet mit einem neu generierten. Ein breiter Überblick über diese Klasse von Systemen findet sich in [2, 3]. Das in [2, 4] beschriebene Local-Positioning Radar (LPR) ist ein originäres Systemkonzept in diesem Bereich. Es verwendet einen aktiven, gepulsten Reflektor um Ziele zu unterscheiden und die Sichtbarkeit zu erhöhen. Eine ähnliche Technik, allerdings mit passiven Strukturen, wurde zuvor in [5, 6] beschrieben. Ein System mit Surface Acoustic Wave (SAW) Referenz findet sich noch früher in [7]. In jüngerer Zeit erfreuten sich aktive Rückstreumodulatoren und Oszillatoren mit switched injection-locking steigender Beliebtheit. So ein Gerät ist als alternative Konfiguration zu LPR erhältlich und in [8, 9] beschrieben. Der Active Pulsed Reflector,eine alternative Konfiguration für die RESOLUTION Plattform übernimmt dieses Prinzip [10]. Variationen des Grundkonzepts – aktive Rückstreumodulation oder Sekundärradar mit Lauf- zeitmessung durch Chirp-Signale – finden sich in großer Menge in der wissenschaftlichen Li- teratur. Meistens handelt es sich hierbei um algorithmische Verbesserungen des Problems der Mehrwegeausbreitung, wie in [11–13] beschrieben. Eine umfassende Arbeit, die das LPR System im 5.8 GHz Industrial, Scientific and Medical (ISM) Band mit einer Bandbreite von 150 MHz beschreibt ist [14], wobei diese Parameter auch für RESOLUTION gültig sind. Eine anstehende Erweiterung dieses Prinzips ist die Verwendung von Ultra-Wideband Chirps um die Pfadauflösung und damit die Genauigkeit zu verbessern. Ein experimenteller Prototyp mit vielversprechenden Leistungsdaten wird in [15,16] beschrieben. Ein leicht abweichendes Konzept ist Local Position Measurement (LPM), das zwar auf den gleichen physikalischen Prinzipien basiert, jedoch Zeitdifferenzmessungen verwendet. Die Grund- lagen des Systems sind in [17, 18] beschrieben und in [19–21] weiter ausgeführt. Wie das zuvor angesprochene LPR wurde auch dieses System über die Jahre hinweg erweitert und verbessert, vor allem im Bereich der Basisband-Signalverarbeitung [13,22–24]. Ein Mehrwert dieses System besteht in der expliziten Verwendung eines Kommunikationskanals für Telemetriedaten [20]. Die Eigenschaften von sekundären FMCW Radar im ISM Band wurden dank jahrelanger Forschungsaktivitäten auf diesem Gebiet durch Analyse, Simulation und Messung erschöpfend beschrieben. Zentrale Bedeutung kommt hierbei dem Mechanismus zur Rampenerzeugung, d.h. dem Synthesizer, zu. Jeder Phasenfehler, den diese Komponente verursacht hat eine direkte abträgliche Wirkung auf die Leistung des Gesamtsystems. Als Folge daraus widmen sich eine Vielzahl von Studien möglichen Fehlerquellen und Verbesserungen in diesem Bereich [25–30]. Ein dritter Mitbewerber für hochpräzise Positionsbestimmung in Innenräumen ist das Ubi- sense Echtzeitlokalisierungssystem. Obwohl es den selben Applikationsraum wie die zuvor ge- nannten Systeme und RESOLUTION bedient operiert es unter technisch völlig anderen Vorraus- setzungen, nämlich Ultra-Wideband Pulsradar mit Zeitdifferenz- und Winkelmessung. Infor- mationen über dieses System, welches bereits als kommerzielles Produkt verfügbar ist finden sich unter www.ubisense.net (Website zuletzt geladen im Juni 2009). Allgemein lässt sich sagen, dass sowohl in der Positionsbestimmung als auch bei Drahtlos- netzwerken ein starker Trend in Richtung Ultra-Wideband Signalisierung erkennbar ist. Es ist daher nicht verwunderlich, dass die meisten Arbeiten die Mehrnutzerverwaltung betreffend im Kontext von Ultra-Wideband Systemen operieren. Ein guter Überblick über Kanalzugriff in Ultra-Wideband Netzwerken findet sich in [31], und im Detail für den IEEE 802.15.4a Standard in [32].

2 Contents

Generell findet sich Literatur zu Mehrnutzerverwaltung im Bereich Positionsbestimmung nur vereinzelt. Der Grund dafür ist, dass die konkurrierenden Systeme in diesem Gebiet, LPR und LPM in den jeweiligen Varianten statischen Kanalzugriff nutzen, was allerdings ebenfalls eine Reihe von Nachteilen mit sich bringt, die in dieser Arbeit angesprochen werden. Systeme mit wahlfreiem Zugriff werden in [33, 34] und im Besonderen in [35] besprochen.

Zielsetzung

Ziel dieser Arbeit ist eine komplette und referenzierbare Leistungsschätzung der RESOLUTION- Plattform, auch in Hinblick auf Produktionsfähigkeit. In Hinblick auf die ausgiebigen Vorarbeiten, die bereits im Bereich von Sekundärradar mit FMCW-Technik geliefert wurden, besonders und speziﬁsch im 5.8 GHz ISM band, scheint es von verschwindendem wissenschaftlichen Wert, die Plattform auf einer rein signaltheoretischen Ebene zu analysieren. In dieser Arbeit wurde daher ein zweifacher Zugang zur Thematik ge- wählt: die Integration der klassischen Systemsimulation mit einer zeitdiskreten, ereignisbasier- ten Netzwerksimulation, um einen gesamtheitlichen Eindruck der Leistungsgrenzen des Systems in verschiedenen Einsatzszenarios zu erhalten. Physikalische Leistungsgrenzen können durch Literaturstudie abgeleitet werden. Daher wurden die Untersuchungen des Physical Layer wei- testgehend auf Betrachtungen des Problems der Mehrwegeausbreitung eingeschränkt, der bei weitem größten Fehlerquelle im System. Schätzungen der Netzwerkparameter, wie beispielsweise die Akquisitionszeit bei Mehrnutzer- zugriﬀ, stellen einen von der Systemsimulation komplett separaten Forschungsbereich dar. Nichtsdestoweniger ist es möglich, beide Zugänge der Systemanalyse gewinnbringend zu ver- binden, was die Betrachtung optimierter Protokoll- und Algorithmenansätze über Abstrak- tionsgrenzen hinweg ermöglicht. Das kann als erster Schritt in Richtung echter Cross-Layer Optimierung in Hinblick auf eine Massenproduktion des Systems gesehen werden. Zum Erreichen dieser Ziele wurde eine umfangreiche Simulationsumgebung programmiert. In dieser Arbeit werden sowohl die Umgebung an sich und Simulationsresultate auf physikalischer Ebene und Netzwerkebene dargestellt.

Gliederung

Der Rest dieser Arbeit ist um zwei zentrale Kapitel aufgebaut, die sich mit der Analyse der physikalischen und netzwerkbezogenen Parameter des RESOLUTION Systems auseinandersetzen. In Kapitel 4 werden Simulationsergebnisse für einen einzelnen Empfänger gezeigt. Dabei werden ausgewählte Probleme der Hardware und im Besonderen Mehrwegeausbreitung behandelt. Die Systemanalyse wird in Kapitel 5 auf Netzwerkeigenschaften erweitert. Geeignete Maß- zahlen werden deﬁniert und Protokolloptionen für das RESOLUTION System präsentiert. Die integrierte Simulationsumgebung wird vorgestellt, und Ergebnisse für verschiedene Protokoll- optionen dargelegt. Um eine gemeinsame Basis für das Verständnis der besprochenen Technologien im Allgemei- nen zu schaﬀen werden in Kapitel 2 Grundlagen der drahtlosen Positionsbestimmung und in Kapitel 3 die Architektur der RESOLUTION PLattform besprochen. Kapitel 6 schließt die Arbeit mit einer Zusammenfassung ab.

3 CHAPTER 1

Introduction

With the advent of the Integrated Circuit came monumental changes to the world of computing and communications. Accelerating improvements in processing speed, component integration and energy consumption led to the surge of professional and consumer products we all see integrated in industry processes and our daily lives: the internet, mobile phones, satellite navi- gation, TV and radio broadcasts, pocket media players, robot factories, autopilots, autonomous control systems, sensor networks. Driven by market demands and the prospect of increasing profits, scientists and engineers have focussed their efforts on optimizing data throughput, edging ever closer towards the limiting Shannon barrier, while maintaining reasonable energy consumption figures and shrinking devices through production technology advancements and integration. More recently, fueled by the vision of autonomous machine spaces and context-aware information systems, a technology from military provenience – as is often the case – has entered the public perception: positioning. For some experts, the ultimate convergence point in the development of communication technology are sensor networks, strongly decentralized groups of ultra-low power sensing nodes with distributed communication facilities. Such a technology could find widespread use in agriculture, environmental monitoring, building automation, battlefield management and industrial control. For most of these applications, sensor data makes only sense in context with a geographical or topological reference. Another legitimation for positioning technology comes from the desire to provide clients of the mobile phone network with context-sensitive services. Lastly, and of outstanding importance for this work, is the field of industrial positioning. The rising complexity and scale of modern industrial environments has bred the desire for further automation of transport and processing. In this work, the RESOLUTION platform – short for “Reconfigurable System for Mobile Com- munication and Positioning” –, a hybrid positioning and communication system for use in both specialized consumer applications and industrial environments is introduced and analyzed. The platform operates in various configurations, but always utilizing the principle of secondary linear FMCW radar for distance measurement. In this area, much prior art exists, as outlined in the next section.

1 1.1. State of the art

Chirp signals as communication or radar carriers have been known since the mid-fifties. The technology is well established in military due to its low probability of interception and detection [1]. Local positioning of cooperative objects, as relevant for this work, usually shows some deviant properties when compared to regular radar. That is, the illuminated target actively seeks to be detected, and either regenerates and reflects the incoming signal or receives it and responds with an originally generated one. A broad overview of this class can be gained by consulting [2,3]. A seminal system concept in this area is LPR, described in [2, 4]. This system employs an active, pulsed reflector to distinguish targets and increase visibility. A similar technique, albeit with passive structures, has been described earlier in [5,6]. A system with SAW reference appears still earlier in [7]. Recently, switched injection-locked oscillators as active backscatterers have seen renewed interest. Such a device is available as alternative receiver configuration in LPR, and its principles have been described in [8, 9]. The Active Pulsed Reflector, an alternative receiver configuration for RESOLUTION, mirrors this principle [10]. Variations on this basic concept – active backscatter modulation or secondary radar roundtrip measurements with chirp signals – can be found aplenty in literature. Mostly, algorithmic improvements to the problem of multipath propagation are shown, as in [11–13]. A comprehensive work describing the LPR system in the 5.8 GHz ISM band and with a bandwidth of 150 MHz – parameters which are also valid for the RESOLUTION platform – is [14]. A forthcoming extension to this is the use of ultra-wideband chirps to increase path profile resolution and, thus, accuracy. An experimental prototype with promising performance has been described in [15, 16]. A slightly deviating concept is LPM, which is based around the same physical principles, but utilizes time difference measurements. The basics of this system are described in [17, 18] and elaborated upon in [19–21]. Like the previously discussed LPR, the system has seen a number of improvements and extensions over the years, mostly pertaining baseband processing [13,22–24]. As added value feature, LPM also explicitly features a communication channel for telemetry data transmission [20]. Owing to year-long research and refinement of those two competing solutions, the properties of secondary radar FMCW systems in the ISM band have been described very exhaustively through analysis, simulation and also measurement results. Of central importance to the system performance is the ramp generation mechanism, i.e., the synthesizer. Any phase error introduced in this component has direct adverse effects on the achievable performance. Con- sequently, the properties, possible error sources and mitigation methods have been studied extensively [25–30]. A third competitor for high-precision indoor positioning is the Ubisense real-time location system. Though serving the same application space as the previously mentioned systems and RESOLUTION, it technically operates under a very different premise, namely ultra-wideband pulse radar with time difference and bearing measurements. Information on this system, which is available as commercial product package, can be found at www.ubisense.net (website re- trieved in June 2009). In general, both positioning and wireless sensor networks, the two broad research areas most closely related to RESOLUTION show a strong trend towards ultra-wideband signaling. It is thus hardly surprising that most works pertaining multi-user access, the second large topical complex of this thesis, operate in the context of ultra-wideband systems. A good overview of medium access control topics for ultra-wideband networks is found in [31], and in particular for the IEEE 802.15.4a standard in [32]. In general, literature specifically treating multi-user access in positioning is few and far between. The reason for this is that the prominent competitors, LPR and LPM and their variants

2 CHAPTER 1. INTRODUCTION

utilize static channel access, which, however, comes with a number of drawbacks, which are also discussed in this work. Systems with random access are described in [33,34] and in particular in [35].

1.2. Goals of the thesis

The goal of this thesis was to provide a complete and comprehensive performance estimation of the hardware developed in the RESOLUTION project, with an eye towards production maturity. With regard to the extensive work done in secondary radar FMCW, especially and specifically in the 5.8 GHz ISM band, there is little scientific worth in carrying on analyses on a signal-theoretical level only. Therefore, a two-pronged approach was taken, integrating classical physical layer system simulation with discrete event network simulation to gain a holistic impression of performance limits in various deployment scenarios. As physical bounds of the system can be readily derived from prior art, the investigative focus for the physical layer simulation was multipath propagation, which constitutes by far the largest remaining error source in the system. Estimation of network parameters, such as time-to-fix, under the premise of multi-user channel access, is a completely distinct field of research from system simulation. Nonetheless, both approaches can fruitfully be combined, making it possible to investigate optimized protocol and algorithm options across abstraction layers. This can be viewed as a first step towards true cross-layer optimization of the system shortly prior to mass production and deployment. To achieve these goals, an extensive simulation framework was implemented. In this thesis, both the framework itself and, more importantly, simulation results both on the direct link level and the network level are presented.

1.3. Organization

The remainder of this work is centered around the two chapters concerned with the analysis of the physical and network properties of the RESOLUTION system. Chapter 4 presents simulation results for the single receiver, highlighting selected hardware impairments and reserving special attention for multipath propagation. Relevant simulation results are presented and commented. The system analysis is expanded to network properties in chapter 5. After a discussion of suitable ﬁgures of merit, several protocol options for RESOLUTION are presented. An integrated simulation environment is introduced and results for several algorithmic and protocol options are given. To establish common ground and foster understanding of positioning technologies in general, chapter 2 deals with fundamentals of wireless positioning, and chapter 3 introduces the architectural basics of the RESOLUTION platform. Chapter 6 summarizes and concludes this thesis.

CHAPTER 2

Fundamentals of Wireless Positioning

Wireless positioning is a field of engineering with an application scope almost as wide as that of wireless communications. It is generally understood to comprise any method or technology that is suitable for automatically determining the position of a target in space by means of wireless transmission. Everything else, the transport medium, protocol, topology and operation scope, are open to definition. This chapter builds the foundation for understanding wireless positioning technology by spotlighting the most important aspects of this engineering field. Given the sheer volume of solutions available today in industry and academia, it can never be exhaustive. Instead, common ground is established to facilitate understanding of subsequent chapters. Beforehand, a common language needs to be established and terms defined. The following attempt loosely adheres to the definitions presented in [36] and [37]. Location in general refers to the semantic understanding of the position of an object in space, thus answering the question “Where is it?”. Location and position are mostly used interchangeably in this thesis. In a more strict sense, position is a technical term, and the question for position always results in a set of coordinates, relative to any frame of reference, whereas location typically references topological features. Positioning thus usually refers to the process of determining the position of an object in 2- or 3-D space, but may also include distance measurement. Range is often used synonymously with distance in positioning literature, which can lead to confusion, since used correctly, range denotes a distance limit, e.g., for which communication still works. Triangulation is often defined as the geometric process of finding a position from measurements, referring the minimal (triangular) layout of devices in the system. Specifically, angu- lation and lateration are technical terms for finding the position from bearing and distance measurements, respectively. In this work, triangulation is taken to include trilateration. Beacons or, more specifically for terrestrial positioning, base stations, are fixed anchor points with known coordinates that serve as measurement reference. The target, terminal or mobile station is the object of which the position is to be determined. It can either have a passive or active role in the positioning process, but it is always mobile with respect to the base stations. Performance figures for positioning systems also merit some attention, which they gain in

5 section 4.1.1. For the current chapter, accuracy is assumed to be the single measure of the “quality” of a positioning system, i.e., its measurement ﬁdelity.

2.1. Application classes

With the advent of near-ubiquitous wireless communications, cheap microprocessors and solid- state frontends came a renewed interest in positioning technology, both for consumer applications and industry solutions [2, 38]. In a first step, positioning efforts can be separated into two broad fields: systems using existing infrastructure to provide location information, mostly as add-on or added value to communications, and dedicated systems with specialized hardware and software for providing position information. The former group includes wireless sensor networks, which by themselves constitute a vast application space. Information from wireless sensor nodes often only makes sense in context with position information: Which room has the least air humidity? Where is the stress fracture? Which patch of soil needs more water? For comprehensive coverage of wireless sensor networks, including positioning techniques, the reader is referred to literature [39–44]. The principal application class for add-on positioning systems are location based services, which are mostly taken to mean commercial services offered by mobile phone providers [45]. The characteristics of this application class are low cost (mostly only software modifications), poor accuracy in the tens of meters regime or even worse, excellent coverage through mobile phone or Wireless Local Area Network (WLAN) infrastructure, and tight coupling with higher-layer semantic processing, such as map projection or location-sensitive billing. Relevant literature is widely available [46–50]. A hybrid approach which involves both existing infrastructure and dedicated hardware is assisted Global Positioning System (GPS). Here, a regular GPS receiver is built into a mobile phone. Azimuth data and satellite lists are provided via the communications service by the base stations to bootstrap the positioning process. This technology is in widespread use today [51]. The class of dedicated positioning systems is led by regular GPS with dedicated receiver systems, soon to be complemented by the European GALILEO effort [52]. Modern GPS receivers achieve an accuracy in the range of several meters in outdoor scenarios, but are notoriously un- derperforming in indoor situations [53]. Applications are widespread, ranging from the original military use to fleet management, hiking, sea and air travel and entertainment [54]. Indoor industrial applications, such as factory automation, automated vehicle guidance and heavy equipment steering call for much higher accuracy than can be provided by GPS even under ideal conditions. Such scenarios fall under the regime of dedicated positioning solutions, which are characterized by comparatively high cost (for infrastructure installment and maintenance) and excellent positioning performance. This class of systems has been widely researched and also seen commercial implementations [2, 15, 17]. The following section will introduce measurement principles which can be found across all applications classes.

2.2. Measurement principles

Several geometric conﬁgurations are known which allow for mobile positioning. In literature, those methods are diﬀerentiated by the measurement data they use for positioning, that is, the target distance ρ, the target bearing (angle) θ, or both [36,37]. Further distinction comes from the roles the mobile and base stations take on in the measurement process. In self-positioning, the mobile unit performs the measurement and calculates its

6 CHAPTER 2. FUNDAMENTALS OF WIRELESS POSITIONING

own position. Conversely, remote-positioning assigns the role of beacon to the mobile, while the measurement takes place in the base stations, and calculations are processed in a central server unit. This has the advantage that baseband logic in the mobile can be kept to a minimum, and complex, energy-consuming algorithms can be implemented in the infrastructure without regard to battery lifetime. If an additional communications link is present, the measurement data can be transmitted to the beacon, which is called indirect-self-positioning and indirect-remote-positioning, respectively.

2.2.1. Time of Arrival (ToA) If the distance to several beacons is known, the position can be calculated by means of rho-rho ﬁxing. The principle is illustrated in ﬁg. 2.1.

S1 (x1, y1) S2 (x2, y2)

1 2

M1 (xm,1, ym,1)

S3 (x3, y3)

Figure 2.1.: Illustration of the ToA measurement principle: the mobile M1 lies on the intersection of three or more circles defined by time-of-flight measurements to or from fixed beacons Sj with known positions.

The exact way in which the distance is measured is irrelevant for this method, but mostly, Time of Arrival measurements are assumed. If the time of ﬂight to several beacons is known, then the distance from the mobile i to base station Sj is

ρi =(t0,j − t0,i) · c, (2.1)

where c is the signal propagation speed, and t0,j and t0,i the transmission and arrival instants, respectively. Obviously, this mandates exact synchronization between the beacons and the mobile stations. From the distances, circle equations of the form

2 − 2 − 2 ρi =(x xi) +(y yi) (2.2)

are postulated and solved for the unknown mobile position (x, y), given the beacon coordinates (xi,yi). The need for over-the-air clock synchronization is a major drawback of ToA,andlargely impossible to guarantee in real-world deployment scenarios. It can be overcome algorithmically by using Roundtrip Time of Flight (RToF) and Time Diﬀerence of Arrival (TDoA), described in the following.

7 2.2.2. Roundtrip Time of Flight (RToF) Instead of directly evaluating the incoming beacon signal, the mobile can use it as synchronization reference and respond with its own positioning signal. The need for further clock synchronization is thus obviated. Assume the beacon transmits its signal at time t0, and it impinges on the mobile after the time of flight at t0 + τ. After a fixed wait-time T , which is known system-wide, the mobile returns its own signal, which arrives at the beacon at t0 +2τ + T . The time of flight can now easily be calculated.

2.2.3. Time Difference of Arrival (TDoA) A slightly more intricate approach to solving the synchronization problem is TDoA. Here, instead of absolute times, time differences between beacons are calculated, which leads to hyperbolic equations, as illustrated in fig. 2.2.

S1 (x1, y1) S2 (x2, y2) 2 1

3 1

M1 (xm,1, ym,1)

S3 (x3, y3)

Figure 2.2.: Illustration of the TDoA measurement principle: the mobile calculates only runtime diﬀerences between beacons, thus eliminating the need for synchronization between beacons and mobile.

When the initial transmit instant t0 is unknown and the incident times ti and tj are measured, the time diﬀerence Δt = tj − ti =(tj − t0) − (ti − t0) (2.3) can be calculated, which is proportional to the distance between two beacons Δd = c · (tj − ti). The locus of points whose focal diﬀerence is constant describes a hyperbola, expressed as

x2 y2 − =1, (2.4) a2 b2 where in the case at hand, 2 2 a = Δd/2 2 2 2 b = Di,j/2 − a . (2.5)

Here, Di,j is the (ﬁxed) distance between two beacons, and it is assumed that the stations lie along the x-axis, which is valid because any such coordinate system can be rotated and translated into a more general one.

8 CHAPTER 2. FUNDAMENTALS OF WIRELESS POSITIONING

AsetofN base stations gives N! K = (2.6) 2(N − 2)! time diﬀerence sets, of which N − 1 are independent. Compared to ToA, an additional beacon is necessary per dimension to calculate a position ﬁx.

2.2.4. Angle of Arrival (AoA) Directional antennas and beam steering allow for determination of the signal bearing θ.Ifthis value is known for several ﬁxed beacons, Angle of Arrival (AoA)ortheta-theta ﬁxing can be used to determine a position.

M1 (x, y)

Ĭ1 S2 (x2, y2)

Ĭ2 x

S1 (0, 0)

Figure 2.3.: Geometric setup of a 2-D AoA measurement. Angles are always measured with respect to geometric “north”, i.e., the direction of the y-axis.

Assuming one beacon at the coordinate origin and the other at (x2,y2), and two angle measurements θ1 and θ2 between beacons and mobile, as shown in ﬁg. 2.3, the mobile coordinates are given by

y tan (θ ) − x y = 2 2 2 tan (θ2) − tan (θ1)

x = y · tan (θ1). (2.7)

While AoA by itself is used rarely in contemporary positioning systems, it is fruitfully employed as add-on to distance measurements, a technique which is consequently called rho-theta ﬁxing. Given the distance ρ and angle θ, the mobile coordinates are simply found to be

x = ρ · sin (θ) y = ρ · cos (θ) (2.8) if the beacon is assumed to lie at the origin.

2.2.5. Fringe solutions Besides the methods mentioned above, there are a number of specialized solutions which generally utilize existing hardware to determine the position of the mobile.

9 The existing mobile phone infrastructure offers daunting possibilities: coverage in developed countries is almost complete, the signal properties of both Global System for Mobile Communications (GSM) and Universal Mobile Telecommunications System (UMTS) are well- known, and handsets are readily and cheaply available. The most simplistic approach to locating mobile handsets is cell-ID. Each base station (or “Node B” in the case of UMTS) emits a specific, unique identifier, which can be used by the mobile to determine its position within the network. The accuracy of this method is limited by the cell size, which in dense urban areas can be as low as 100 m, while in rural settings it may grow to several km in size [55]. There is a standardized method to support cell-ID by RToF measurements. This would fix the position of the mobile to a circle around the base station. All commonly available handsets lack the possibility to determine the bearing of the base station signal, so rho-theta fixing is generally not possible. In GSM and UMTS, the possibility for real TDoA positioning exists. To overcome the problem of overshouting, the UMTS standard even proposes the introduction of blank times called Idle Periods in Downlink (IPDL). The Common Pilot Channel (CPICH) signal is correlated within the mobile receiver to estimate the time of flight. With this method, accuracies to within the Federal Communications Commission (FCC) limit, i.e., in the range of less than 100 m can be achieved [56]. A common method to make use of existing WLAN infrastructure is Received Signal Strength (RSS). Here, the mobile performs signal strength measurements, a facility which is by default included in most clients. This information, together with an access point identifier, can serve to get a distance estimate. To this end, a path loss equation is solved for the unknown distance using the power measurement. Due to small-scale fading, this method usually leads to very poor results especially in indoor environments. A different approach to handling the power measurement is the use of fingerprinting [57,58]. The power value from several access points is correlated against a database, which has to be pre-calibrated for the area in question before operation can commence. The mobile is then assumed to be at the position which yields the closest match. This method has two drawbacks. First, it is prone to changes in the environment which affect the propagation properties and, thus, the power patterns for a specific spot. Second, the database has to be built beforehand, which entails traversing the entire area, taking spot measurements and entering the corresponding coordinates. Such an approach is usually not considered to be “true” positioning. In the light of the insights gained so far, tab. 2.1 presents a selection of real-world positioning applications and solutions, along with approximate performance figures.

2.3. Physical layer

Though microwave-based solutions spring to mind when positioning is concerned, there are several other options for the physical transport medium, most prominently ultrasound and optical systems.

2.3.1. Non-microwave solutions Ultrasound, operating with sound waves in the range of 20 kHz–100 MHz, oﬀer the principal characteristic of not being able to penetrate walls. This can be used to good eﬀect in applications like asset location in bureaus and hospitals. However, ultrasound is prone to interference due to the state of the transport medium.

10 CHAPTER 2. FUNDAMENTALS OF WIRELESS POSITIONING

Application Operating Medium Accuracy Coverage Value principle Nintendo Wii ToA,Sen-Optical/Infrared sor fusion sonitor — Ultrasound Ekahau RSS Microwave, WLAN ISM Ubisense AoA, TDoA Microwave, 7GHz UWB Symeo TDoA Microwave, 5.8 GHz Symeo UWB TDoA Microwave, 7.5 GHz UWB ABATEC TDoA Microwave, 5.8 GHz ISM GPS TDoA Microwave, 1227.60 MHz, 1575.42 MHz Galileo TDoA Microwave, 1164–1214 MHz, 1563–1591 MHz GSM LBS CELL-ID, Microwave, RSS, TDoA 1800 MHz, 1900 MHz UMTS LBS TDoA Microwave, 2100 MHz A-GPS TDoA,Sen- Microwave sor fusion

Table 2.1.: Selection of common positioning applications, with a comparison of utilized technology and rough performance estimates. The “value” column refers to the installation and maintenance cost of the system, so high value means low cost. Sources: [15, 45, 47, 49, 52], product brochures (partially available online).

Optical systems can either refer to infrared transceivers, such as utilized in the popular Wii gaming console by Nintendo. Here, two infrared beacons mounted to a TV set are evaluated by a hand held controller to calculate a position on a virtual x-y plane. Laser systems are the second large application class in optical systems. With the extremely small wavelengths oﬀered by optical light, very high accuracies are possible.

11 Optical systems are, in addition, prone to interference through external light sources, most notably daylight. Also, it is not possible to track multiple targets with a laser, because only objects down its main ray axis can be located.

2.3.2. Microwave based solutions and FMCW Microwaves, which denote electromagnetic waves in the frequency band from 300 MHz–300 GHz, have a number of advantages compared to ultrasound and optical/laser solutions. They are robust and resilient against dust particles and air pollution, because their wavelength is much larger than typical particles. Microwave systems offer the possibility of using a broad detection cone to illuminate multiple targets. This advantage is bought with the drawback of multipath propagation and interference, which is the principal error source of microwave positioning systems. Given the availability of cheap transceivers, the multi-target ability and unparalleled flexi- bility of microwaves, they are the primary choice for real-time 3-D positioning systems. The principle of FMCW radar has long been known [1]. The advent of solid-state transmitters and, especially, the digital signal processor, has renewed interest in this technique. Compared to pulse radar, FMCW has several beneficial properties. First, the basic frontend is very simplistic, as shown in fig. 2.4. A Voltage Controlled Oscillator (VCO) generates the modulation signal, which is fed to the antenna and a local mixer. The reflected wave is mixed with the local signal to produce a phase/frequency difference which is proportional to the target distance.

VCO Circulator

Baseband

Figure 2.4.: Basic FMCW circuit. The VCO generates a frequency-modulated signal, which is fed to the antenna and to the mixer. The phase difference of transmitted and incident waves is evaluated in a baseband processor.

Second, the target resolution ΔR of FMCW radar is proportional to the inverse of the bandwidth of the modulated ramp only, and given by c ΔR = , (2.9) 2B where B is the bandwidth and c the signal propagation speed. A further advantage, which is of primary interest in military and security applications, is that the signal time-bandwidth product is typically very high, making it hard to intercept and detect the transmission. Modern digital signal processing allows for evaluation of the phase/frequency diﬀerence of the signal by means of Fast Fourier Transform (FFT), which is trivial compared to more complex correlators required for pulse radar.

12 CHAPTER 2. FUNDAMENTALS OF WIRELESS POSITIONING

The aforementioned advantages are also put to use in local positioning, where the FMCW signal form is mostly used in secondary radar configurations, i.e., where the tracked object is not passively reflecting, but actively receiving and returning a signal of its own. Regardless of the operating principle, the basic waveform generated by the FMCW transmitter is written as 4πB s (t) = cos ((2πf + 1/2 )t + φ), (2.10) TX 0 T where f0 is the center frequency, φ the phase angle, B the sweep bandwidth and T the total sweep duration (up- and downsweep), which is much greater than the expected signal runtime τ. The above and all following statements regarding the FMCW signal form are true within the extent of a half-period (upsweep), so −T/4 ≤ t ≤ T/4. As can be seen in fig. 2.5, which also summarizes the signal parameters, the time and frequency differences between transmitted and incident ramp are proportional to each other with the ramp steepness.

t f B

t T

Figure 2.5.: Graphical illustration of the FMCW principle. The time offset experienced by the reflected ramp is proportional to a frequency difference in both the up- and downsweep.

If a moving object is the detection target, a Doppler shift occurs, which imposes an additional frequency oﬀset on the incident ramp proportional to the movement speed. The frequency shift, given the target velocity v and signal frequency f0,is

v fDoppler = f0 · /c. (2.11) As can be seen in fig. 2.6, this results in different Intermediate Frequency (IF) values for the up- and downsweep. The range and velocity of the target can then be found through [59] Δf +Δf 2B f = 1 2 = R. (2.12) Range 2 cT Δf − Δf 2f f = 1 2 = 0 v. (2.13) Doppler 2 c The secondary-radar FMCW principle is of supreme importance for this work, as the positioning module of the RESOLUTION! (Reconfigurable Systems for Mobile Communication and Positioning) platform is built around this technology. The signaling specifics and platform are described in the next chapter.

13 f

B f1

t T

Figure 2.6.: Velocity measurement with FMCW ramps. The Doppler shift causes deviations in the frequency diﬀerences on the up- and downsweeps.

14 CHAPTER 3

The RESOLUTION Platform

The previous chapter has provided a glimpse of the multitude and diversity of the field of positioning, ranging from aviation radar to mobile phone tracking. An area of positioning which has attracted enhanced interest from both industry and academia is high-precision local positioning with specialized, dedicated hardware. The remainder of this work is concerned with the simulative and analytical description and evaluation of such a platform, designed and implemented during the course of the EU-project Reconfigurable Systems for Mobile Communication and Positioning (RESOLUTION) [60–65]. The project idiom has become synonymous with the platform itself and is used accordingly in this work. The remaining sections of this chapter describe the application field and service requirements targeted by the RESOLUTION platform, the hardware base, signaling specifics and requirements. Subsequent chapters will then proceed with simulative performance analysis of both hardware and software aspects of this system.

3.1. RESOLUTION service requirements

The RESOLUTION platform is conceptually intended to serve a market for high-precision radiolocation with dedicated hardware. There are three broad application fields which are intended for service by the platform: Person guidance includes all applications in which the receiver of the position information, usually in some sort of processed form, e.g., projected to a map, is a human. De- ployment scenarios for this class include tourist guidance, assisted living for impaired persons, smart spaces such as large shopping malls, targeted advertising in such confines and interactive games. Special care must be taken to provide the user with semantics corresponding to his position, i.e., location-sensitive information. This usually mandates a comparatively high-bandwidth communication link. Asset tracking specifically pertains the location of indoor items. High-precision location, due to elevated costs of mobile tags, is clearly not suitable for bulk tracking of goods. This remains a classic area of Radio Frequency Identification (RFID) tags. Possible

15 deployment options usually involve costly, singular pieces of equipment such as medical and emergency devices in hospitals. Such items are tracked only on-demand, with high reliability requirements. Robot control is a broad term which is generally taken to mean applications where the recipient of the position information is an automated, usually mobile device such as an Automated Guided Vehicle (AGV). The classical application is the steering of transport vehicles for containers in a port. In such a scenario, the robots do not receive direct position information, but rather control commands from the infrastructure to avoid collisions and navigate them to their destination. Each of these applications obviously has different requirements pertaining the accuracy, number of position updates per second, energy efficiency, reliability and scale, i.e., number of supported mobiles per service area. Tab. 3.1 identifies robot control as the most demanding application class. Fig. 3.1 outlines the basic use cases for those applications. The typical use case for the

Requirements Application class Accuracy Updates Eﬃciency Reliability Scale Person guidance Asset tracking Robot control

Table 3.1.: Requirement map of the application classes supported by the RESOLUTION platform. The size of the rectangle indicates the importance of the respective parameter for the application class.

robot control is shown in fig. 3.1a. The infrastructure, which is the controlling instance of the entire system, requests on-demand position from the robots. Position data is then evaluated and a corresponding command is issued. This process is periodically repeated to ensure constant tracking of the robots. Conversely, in person guidance, the position request is posted by the mobile/user, as seen in fig. 3.1b. Typical for this use case is the evaluation of the position information at the user site, e.g., in a Personal Digital Assistant (PDA) or similar device. Also, the request interval is usually unforeseeable, i.e., random: the user pressing a button, moving on to some other exhibit and so on. A special case is illustrated in fig. 3.1c. This use case is known from GPS: the infrastructure periodically provides measurement signals which the user can optionally process or discard. The position semantic is processed at the user site. It is clear from the above considerations that successful integration of positioning in a wireless network invariably requires a communications link. At the very least, this link must enable the exchange of control messages. Often, additional semantics such as streaming audio and video are transferred. Consequently, the RESOLUTION platform is designed as hybrid communication and positioning solution, with exchangeable communication modules, as detailed in the next section.

3.2. Hybrid positioning and communication

There are several well-established communication standards available which are suitable for use in a sensor network with positioning. For the speciﬁc requirements of RESOLUTION, the sought

16 CHAPTER 3. THE RESOLUTION PLATFORM

Position request Evaluation/Projection Command Position request

...

Position data (a)

Position data

...

Position request Evaluation/Projection Position request

(b)

...

Position data Evaluation/Projection Position data (c)

Figure 3.1.: Use cases and message exchange between infrastructure and mobile. The dashed arrow indicates measurement data exchange. Random and fixed waiting times are illustrated as clocks with or without arrow, respectively. (a) “AGV” use case (b) Classical user request (c) Periodic downlink-only measurement . after key characteristics were • compatibility with the positioning subsystem, i.e., minimal interference on both sides, • reasonable efficiency, so the overall power consumption stays within the bounds dictated by the application, • a proper channel contention scheme, independent of positioning operations, • unlicensed access and • appropriate data rates. The question of what is an appropriate data rate can be answered in context with the application. For simple control or transfer of positioning data, very low data rates are sufficient. Applications such as person guidance might require significantly more bandwidth, however, to provide context-sensitive data like streaming audio and video.

17 The two prime candidate standards for those requirements are IEEE 802.11 (WLAN)and IEEE 802.15.4 (ZigBee). Both operate in free ISM bands, around 2.4 GHz for ZigBee and from 5.25 GHz upwards for WLAN, which is shown in the spectrum allocation plot in ﬁg. 3.2.

ZigBee/WLAN WLAN WLAN Positioning f / GHz 2.400 2.485 5.250 5.350 5.470 5.725 5.875

Figure 3.2.: Spectrum allocation of communication standards suitable for RESOLUTION.

There is also an option for WLAN in the 2.4 GHz band. The WLAN sub-standards in question are characterized in tab. 3.2.

Standard Band Max bit rate 802.11a 5 GHz 54 Mbit/s 802.11b 2.4 GHz 11 Mbit/s 802.11g 2.4 GHz 54 Mbit/s (802.11n) 5 GHz/2.4 GHz 600 Mbit/s

Table 3.2.: Sub-standards of IEEE 802.11 (WLAN) and their characteristics. Note that 802.11n is a draft standard only at the time of this writing.

To ensure suﬃcient band isolation between communication and positioning, it is reasonable to select a standard in the 2.4 GHz band. This makes it impossible to use a single, wide-band antenna for both operations, however, which has an impact on the form factor of the device. In comparison to the high data rates provided by WLAN, ZigBee supports a data rate of only 250 kbit/s. This makes it suitable for transmission of control commands and sparse content packets only. However, ZigBee is optimized for low duty cycle operation and low power consumption, a signiﬁcant advantage over WLAN [66]. Both systems use a Carrier Sense Multiple Access (CSMA) contention scheme to deal with multiple access. Due to the much higher data rates, contention is generally assumed to be a more critical issue in WLAN. For the remainder of this work, and especially in chapter 5, WLAN is assumed to be the communication standard of choice, because it represents a worst-case lower bound on network performance while providing a powerful, high-bandwidth data link. ZigBee remains a viable option for low-power, machine-to-machine operations. For a complete rundown of WLAN functionality, the reader is referred to the relevant standards documents [67, 68].

3.3. RESOLUTION hardware base

Fig. 3.3 shows the conceptual block diagram of the RESOLUTION hardware platform. The FMCW-based positioning subsystem HPLS (High-Precision Location System) consists mainly of the Radio Frequency (RF) front-end, plus baseband logic to evaluate the position. The signal-theoretical foundations of the positioning process are detailed further on.

18 CHAPTER 3. THE RESOLUTION PLATFORM

A D

Baseband- FPGA Synthesizer

Commercial communication chip Interface (WLAN, ZigBee ...)

Figure 3.3.: Conceptual block diagram of the RESOLUTION hardware base, including the HPLS front-end, the communications chip, baseband processing and interface.

Parameter Shorthand Value

Center frequency f0 5.8 GHz Bandwidth B 150 MHz Ramp period T 0.5 ms EIRP –max.14dBm

Table 3.3.: Central physical layer speciﬁcations for the RESOLUTION platform.

Tab. 3.3 lists the central physical layer specifications of HPLS. Operation in the ISM band at 5.8 GHz allows for a license-free output power of 14 dBm, which guarantees a strong range advantage over current Ultra-Wideband (UWB)systems[16]. The communication and positioning signals are multiplexed via higher-layer flow control to ensure minimal interference. The use of separate antennas obviates the need for an antenna switch or circulator. In the current configuration of the hardware, the communications link is regulated via the interface block in the baseband section. The central hardware component of the HPLS frontend is the synthesizer, which is responsible for generating highly linear frequency ramps. The synthesizer is based around a fractional-n Phase Locked Loop (PLL) design with ΣΔ- modulated Multi-Modulus Divider (MMD). This design currently achieves phase noise better than -117 dBc/Hz at only 100 mW output power. Detailed information can be found in [65,69]. The measurement process follows the secondary-radar principle with FMCW signals. In the transmit path, the synthesizer generates a ramp of the form sTX(t)=cos (ω0 + 1/2μt)t + φ , (3.1) where μ is a shorthand for the ramp steepness 4πB/T and φ a constant phase term.

19 Arriving at the receiver, this signal is affected by noise and possibly multipath propagation, an effect which is treated in section 4.3.1. The received signal is thus a sum of multiple copies of the transmit signal, affected by specific attenuation and time delays. It can be written as

Nc−1 sRX(t)=α0sTX(t − τ0)+ αisTX t − τi + n(t). (3.2) i=1

Here, Nc is the total number of path components, with specific amplitudes αi and time delays τi,andn(t) a Gaussian noise term. The multipath components also experience phase shifts, which have a destructive effect on the measurement process. This is elaborated upon in section 4.3.1. Phase terms have been omitted in (3.2) for sake of simplicity. After band selection and amplification, this signal is mixed in the receiver with a local copy of the transmit signal. After low-pass filtering to get rid of high-frequency components at 2ω0,

Nc sBB(t)= α˜i cos (μτi)t + φi + nBB(t) (3.3) i=0

results. Here,α ˜i are the modiﬁed amplitudes, now including also the wanted Line of Sight (LOS) signal with index 0, and nBB(t) the ﬁltered noise. This signal is now fed to Analog to Digital Conversion (A/D) and handed to the baseband processor. As can be seen in (3.3), the frequencies of the baseband cosine terms are directly proportional (with μ) to the respective signal runtimes τi. Frequency analysis in the baseband can now produce the wanted runtime of the LOS term, τ0, which is easily translated to a distance value. The FFT has long been the preferred method for frequency analysis. It is also the default analysis method in this work, so subsequent discussions of baseband analysis always assume the FFT as underlying algorithm. Note that the above observations disregard absolute timing in the system. Generally, the measurement takes place between two identical stations A and B. Station A produces the ramp at the time instant t0,A, which is unknown to station B. In absolute time, the transmit signal is then sTX(t)=cos (ω0 + 1/2μ(t − t0,A)) · (t − t0,A)+Φ , (3.4)

which will make station B produce a measurement which includes the unknown time offset of station A. One approach to resolve this is to use ramp synchronization in a RToF protocol configuration. Here, station B mixes the received ramp with its own, started at the specific time offset t0,B. After a system-wide constant wait time T , it returns a ramp which is frequency shifted to include the measurement result t0,B − t0,A as well as the one-way time of flight. This signal returns to station A, where it is again mixed to produce the runtime, which is possible because both t0,A as well as the difference t0,B − t0,A are now known at station A. There are other methods to resolve the timing offset situation, such as the application of a TDoA protocol. The operation principle remains the same, except that a third station is introduced which provides the synchronization (reference) ramp. The measuring station then receives ramps from one or several other stations which are synchronized to this reference signal. This approach has several advantages, such as obviating the need for two-way signaling. Details on the protocol implementations and their implications on the system performance are found in chapter 5.

20 CHAPTER 4

Single Node Architecture and Performance Analysis

This chapter is concerned with the performance analysis of the HPLS positioning hardware. As opposed to the next chapter, the setup under consideration is that of an isolated node in exchange with a single base station. Suitable ﬁgures of merit are deﬁned and the basic performance under AWGN conditions is given. Selected hardware impairments as well as adverse signaling conditions – multipath propagation being the most prominent – are considered.

4.1. Basic receiver performance

The primary goal of system simulation is to get an estimate of receiver performance without having to implement actual prototype hardware. To this end, mathematical abstractions of system components are developed and implemented in a suitable simulator or programming language. As models can only be an approximation of real-world behavior, it is important to have an idea of the question the simulation should be designed to answer. Given current simulator technology and conventional server performance, it is unfeasible to run long-time simulations with very high modeling detail. Fig. 4.1 illustrates simulation abstraction layers with examples from the structural and functional domains. For an initial performance assessment as part of a feasibility study or concept design, the highest (“conceptual”) abstraction layer will be the correct choice in most cases. Electronic components are modeled as transfer functions and diﬀerential or algebraic equations on this level. To integrate device-speciﬁc characteristics later in the prototyping process, electrical properties are determined in dedicated physics simulations and implemented as mathematical models [70]. Fig. 4.2 shows the principal setup of a basic system simulation of the HPLS system. Seman- tically, the simulation is organized in three parts:

Transmitter consisting of a synthesizer block for generation of the FMCW ramp, followed by an (ideal) power ampliﬁer block for signal power selection.

Channel which in the most basic conﬁguration adds White Gaussian Noise (WGN)tothe signal and fades the signal power according to an underlying large-scale fading model.

21 Structural Functional

Processor-Memory-Switch Algorithm

Gate Boolean Equation

Transistor Differential Equation

Physical Voltage, current

Logical Discrete levels, bits

Behavioral Bit vectors, data blocks

Conceptual Signals

Simulation domain

Figure 4.1.: Simulation abstraction levels exempliﬁed by structural and functional domains. (Adapted from P. J. Ashenden et al.: The System De- signer’s Guide to VHDL-AMS, Morgan Kaufmann Publishers, 2003)

Receiver as the actual component under scrutiny. In a ﬁrst abstraction, it includes the mixer to downconvert the incoming RF ramp with a local signal replica and the baseband processing to evaluate the IF/baseband signal.

DSP

Synthesizer PA Path loss

WGN

Figure 4.2.: Conceptual block diagram of the principal system simulation setup.

Nyquist’s theorem suggests that aliasing-free representation of a signal mandates to sample at least at twice the highest frequency represented in the signal [71]. Assuming a f0 of 5.725 GHz, this would suggest a simulation sampling rate fs of 11.75 GHz, which would yield exceptionally large vectors and very long simulation times. It is thus customary to limit simulation to the band of interest, which in the case of HPLS amounts to the 150 MHz band containing the actual ramp signal, omitting the carrier [70]. As the required transformation yields a complex signal, this method is generally known as Equivalent Complex Baseband (ECB) or complex envelope representation.

22 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

The characteristics of this transformation and its application to the FMCW signal are described in Appendix D. The signal produced by the complex synthesizer is written as

√ 1 2 x˜(t)= Aej 2 μt , (4.1) where μ is the ramp steepness defined in (3.1), and A the complex signal power. This signal is both used as transmit signalx ˜TX(t) and local oscillator outputx ˜LO(t) in the mixing process. As this time-continuous signal can not be represented in a digital computer, sampling is used to transform it into a digital sequence: | x˜(t) t=nTs =˜x[n] √ 1 2 = Aej 2 μ(nTs) , (4.2) where Ts is the sampling period, i.e. the inverse of the sampling frequency fs. As per Nyquist’s theorem, the minimum sampling frequency is given by twice the maximum signal frequency. Nonideal effects in the receiver can, however, lead to an increase in the signal band of interest. Additionally, due to the aperiodic nature of the ramp signal, spectral leakage is to be expected. It is thus customary to radically oversample analog signals for simulation representation. The sampling frequency for representation of RF signals fs,RF has been chosen as eight times the signal bandwidth B. To simulate a specific distance d between transmitter and receiver, the transmit signal is delayed by the (one-way) time-of-flight τ: √ 1 − 2 x˜ [n − τ]= Aej 2 μ(nTs τ) TX √ 1 − 1 2 = Aej(( 2 μnTs μτ )nTs+ 2 μτ ). (4.3)

Note that τ can only be implemented as a discrete sample delay, which theoretically induces an error in the position fix compared to the continuous signal. However, as the RF sampling rate fs,RF is very large, this error is significantly smaller than the one introduced by the limited frequency domain resolution (see 4.1.3). The minimal effects the channel imposes on the transmit signal are signal power fading (path loss) and noise. Path loss can in a first approximation be described as

P λ2G G P = TX TX RX , (4.4) RX (4π)2dn where PRX and PTX are the received and transmitted signal power; λ the signal wavelength; GTX and GRX the transmit and receive antenna gains; and d the distance between transmitter and receiver. The path loss exponent n is 2 for free space conditions. However, it is known that in real environments, especially indoors, this number can be signiﬁcantly larger [72]. Gaussian noise is added to the signal to account for ﬂat thermal noise in the receiver and other sources [52]. It is generated as a complex sequence of random numbers with normal distribution and added to the transmit signal:

x˜CH[n]=A¯x˜TX[n]+ν[n] σN σN = A¯x˜TX[n]+√ N(0, 1) + j√ N(0, 1). (4.5) 2 2

¯ 2 2 Here, A is the faded signal amplitude, σN the desired noise power and N(μ, σ ) a vector of independent Gaussian random variables with mean μ, standard deviation σ and identical length as the transmit signal. When the signal impinges on the receiver, it is mixed with a local (non-delayed) copy of the transmit signalx ˜LO to produce an IF signalx ˜IF. Note that in a real RF simulation, the (real)

23 mixing process would yield signals at both IF and the double carrier frequency 2ω0, which would require subsequent low-pass ﬁltering. In an ECB simulation, complex mixing is applied which yields only the IF signal and is thus equivalent to RF mixing with subsequent ideal low-pass ﬁltering. The IF signal is then given as

x˜IF[n]=˜x[n] ⊗ x˜CH[n] · ∗ =˜x[n] x˜CH[n] A¯ = ej((μτ )nTs+Φ) +¯ν[n], (4.6) 2 where Φ is a constant phase term andν ¯[n] the modiﬁed noise term.

4.1.1. Figures of merit An important distinction in the evaluation of positioning systems is between accuracy, precision and resolution. The latter term describes the ability of the system to perceive two targets as distincts objects. In HPLS, this is a function of bandwidth and frequency domain resolution, i.e., a purely deterministic parameter, given default detection methods. Accuracy, on the other hand, describes the deviation of a position ﬁx from the expected (true) value, i.e., the absolute position error (in any dimension). In the trivial case, it is simply the distance error = |d − dm|, (4.7)

where d and dm are the actual and measured distances, respectively. According to the central limit theorem, for a large number of trials, the distribution of the error vector

e = { 1 2 ... N} (4.8)

2 will follow a Gaussian distribution N(μ,σ ). Barring any systematic errors (i.e. hardware faults or software glitches), the position measurement will be distributed around the true distance d, and thus, μ =0. Precision describes the mean deviation of a number of positioning attempts from the true target position, quantiﬁed by the standard deviation of the underlying probability density. Due to the limited frequency domain resolution, the error is quantized with the FFT bin size, which makes error histograms impracticable and possibly inaccurate. Furthermore, performance analysis usually mandates sweeping design variables, such as the noise power, which would yield several histograms for each datapoint. In radar signal processing, hypothesis testing for the evaluation of signals is commonplace [73]. Given a vector of measurements t y =[y0y1 ...yN−1] , (4.9)

the N-dimensional joint pdfs py(y|H0)andpy(y|H1) are deﬁned with

H0 The measurement is the result of interference

H1 Themeasurementistheresultofinterference and target echoes and py(y|H0)pdfofy given that the target was not present

py(y|H1)pdfofy given that the target was present Based on these pdfs, two probabilities are deﬁned:

PD Probability of detection, i.e., a target that is actually present is also declared.

24 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

PFA Probability of a false alarm, i.e., a target is declared which is in fact not present. A radar is constantly “detecting” signals, so the design of criteria for these probabilities is of vital importance. These parameters are not directly suitable for local positioning, however. The sensor in a local positioning setup only receives specific signals from the beacon within a very limited time frame, so there is no possibility of false alarms. Rather, the probability of detection PD and its complementary, the probability of a miss PM =1− PD, i.e., a target was present but not detected, are suffice to define the performance of a local positioning system. The trivial definition of H1 would be that the detected signal is within the frequency bin known to contain the (quantized) real distance d. This is, however, impractical because it would scale PD with the frequency resolution. Thus, while the resolution of path components sinks with a shorter FFT, the error tolerance rises. This is clearly not ideal from the end-user perspective on accuracy. A more sensible definition is

H0 : > L

H1 : ≤ L, (4.10)

i.e., if the measured error is less or equal than a tolerance level L, the signal is detected. For the detection probability, given a vector of measurements y, this means that

PD = P (|y − d|≤ L). (4.11)

The actual selected error tolerance L will be given with the simulation results. Note that the detection probability is a measure of both accuracy and position. However, effects which shift the mean error from the expected value of zero can produce surprising effects in the detection probability. Fig. 4.3 illustrates a paradox: despite showing far worse precision than fig. 4.3c, the error pdf in fig. 4.3d has a better detection probability due to the fact that some values randomly fall into the tolerance limit. This would suggest that the system actually performs better at lower Signal to Noise Ratio (SNR) values, which is counterintuitive. Note that the extent of this effect hinges on the choice of the tolerance limit, however. If it was extended to include the bulk of the values in fig. 4.3c, detection probability of this pdf would again be better than in fig. 4.3d. Compared to an ideal reference, it is clear that additive noise effects will shift the detection probability to the right. This amounts to a loss in range, and is thus labeled sensitivity loss. Non-Gaussian error sources such as multipath propagation can have the additional effect of saturating the detection probability at values lower than one, suggesting that the system performance suffers even with no noise present.

4.1.2. AWGN performance The most basic impairments experienced by the positioning system are noise and signal power fading over distance. The combined eﬀect can be conveniently captured as the SNR measured at the receiver antenna, given as 2 σS SNR = 2 , (4.12) σN where σS and σN are the standard deviations of the wanted signal and noise, respectively. If the SNR should be referred to a real hardware, the transmitter and receiver Noise Figure (NF) must be taken into account. The NF for a two-port is generally deﬁned as [74] SNR NF = in , (4.13) SNRout

where SNRin and SNRout are the SNR values at the input and output of the system, respectively. The NF can be measured for each element of a circuit, and will usually be given in

25 PDF(e) PDF(e)

100% 60%

e e (a) (b)

PDF(e) PDF(e)

0% 10%

e e (c) (d)

Figure 4.3.: Illustration of the effects of mean and variance fluctuations of the error absolute value e on the detection probability. (a) Very high accuracy and precision (b) Reduction of precision (e.g. due to noise) leads to lower detection probability (c) Despite high precision, the mean value of the error is far off, leading to zero hits (d) Accuracy is again very bad, but due to the worse precision, some values fall within the tolerance limit.

the speciﬁcation. The composite noise ﬁgure of an entire system can be derived from the Friis equation [74] as

NF2 − 1 NF3 − 1 NFm − 1 NFSystem =1+(NF1 − 1) + + + ...+ . (4.14) A1 A1A2 A1 ...Am−1 th Here, NFm and Am are the noise figure and gain of the m stage, respectively. In a simulation, it is customary to define the SNR as a sweep parameter. This is equivalent to iterating over distance values, as the noise power is assumed to be fixed at receiver noise floor BN 0 = σ2 = kT f , (4.15) 2 N 0 s Where fs is the total signal bandwidth in simulation. Thus, the signal power at the receiver antenna is derived as SNR 2 10 2 σS =10 σN . (4.16) A noise vector according to (4.5) is added to the signal, which is then processed through the baseband chain described in section 4.1.3. The resulting detection probability is shown in

26 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

0.8

) 0.6 L <

P( 0.4

0.2

0 -40 -30 -20 -10 0 SNR Figure 4.4.: Detection probability P (<0.15 m) for the reference AWGN case, with FFT signal analysis and brute-force peak search.

ﬁg. 4.4. The detection limit of the reference case is 0.15 m. Tab. 4.1 relates the SNR values to distances for diﬀerent large scale fading exponents, unit antenna gains and 25 mW transmit power. Recall that n = 2 corresponds to ideal free space propagation, which will hardly ever be the case in an actual application.

SNR / dB Free space Indoor -40 380.31 52.49 -35 213.86 35.76 -30 120.26 24.36 -25 67.63 16.60 -20 38.03 11.31 -15 21.39 7.70 -10 12.03 5.25 03.802.44

Table 4.1.: Relation of SNR values to actual distances for two diﬀerent path loss exponents.

It is hard to make a compelling statement about the range of the system given only ﬁg. 4.4. The detection probability makes no statement on how far a “miss” was away from the actual target; there might be algorithms in the baseband to re-request position ﬁx attempts when certain criteria are met; antenna combining may be in use to utilize multiple signal paths; and so forth.

27 In the following sections, the plot in ﬁg. 4.4 will be shown and referred to as “reference case”.

4.1.3. Baseband signal evaluation

Complex xIF[n] xBB[n] xTF[n] X(f) | DFT Real

Resampling Windowing Detection

Figure 4.5.: Detailed signal processing chain in baseband. The incoming IF signal is converted to real, radically downsampled and windowed. After DFT, peak detection is performed to evaluated the signal runtime.

Fig. 4.5 shows a more detailed block diagram of the signal processing steps taking place after mixing. The IF signal in (4.6) is a complex oscillation with the frequency

ω0 = μτ. (4.17) Assuming reasonable values for the distance, this signal is in the MHz range and can thus be converted back to a real signal, yielding A¯ x [n]={ ej((μτ )nTs+Φ) +¯ν[n]} BB 2 A¯ = cos (μτ)nT +Φ + {ν¯[n]}. (4.18) 2 s

Recalling that fs was chosen as 8B to accommodate potential spectral leakage in the RF domain, xBB[n] is extremely oversampled. This is disadvantageous in frequency domain evaluation, as after FFT, the frequency domain resolution is given as f Δf = s . (4.19) N It is thus meaningful to downsample the baseband signal to the lowest possible frequency if N is held constant. As no further frequency domain leakage can be expected at this point, this would amount to fs,BB ≥ 2μτmax, (4.20) where τmax is the expected maximum time of ﬂight, which can be derived from the system speciﬁcation. After resampling, the signal is multiplied with a window function to compensate for spectral leakage due to the unknown period length of the cosine in (4.18). Tab. 4.2 lists the parameters of some common window functions. In reality, the choice of this function is of minor interest in the AWGN case, as the observation length of the incoming signal is much larger than the signal period T0. It becomes crucial for multipath propagation, though (see section 4.3.1). In the following, a rectangular window is assumed as the default choice unless stated otherwise. The windowed signal is fed to an N-point FFT, becoming | ·

x˜TF =˜xBB[n] fs,BB w[n] ¢ X(f)=A˜W(πT(f − μτ)) + N(f). (4.21)

28 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

Type ΔA /dB Δω /rad 4πf Rect -13 s/Nw 8πf Hanning -32 s/Nw 8πf Hamming -42 s/Nw 12πf Blackman -58 s/Nw

Table 4.2.: Parameters of diﬀerent window types. ΔA is the sidelobe suppression relative to the main lobe height and Δω the width of the main lobe.

Here, w[n]andW (·) is the window function in time and frequency domain, A˜ the modified amplitude and N(f) the Fourier transformed noise term. Note that the constant phase term Φ from (4.18) has been omitted here for simplicity. An artistic rendition of the squared absolute spectrum |X(f)|2 can be seen in fig. 4.6. An inherent problem of this approach to frequency analysis is that the underlying spectrum of the Discrete Time Fourier Transform (DTFT) is only sampled at regular, discrete instants by the DFT. This effect is exaggerated in fig. 4.6, where the actual frequency peak is close to the edge of bin five. The maximum error from this effect is equal to half the bin size (if the actual peak

df |X(f)|2

f 123456789101112

Figure 4.6.: Illustration of the frequency domain signal |X(f)|2. The underlying DTFT spectrum is sampled by the DFT, giving an approximation of the true frequency content. is on the outermost edge of the bin) and given by πcΔf = , (4.22) FFT,max μ where c is the signal velocity. The channel proﬁle in frequency domain is handed to peak detection. In the trivial case, this is simply a maximum detection, but more complex approaches are possible [75–80]. From the bin index of the peakp ˆ, the distance is calculated as 2πcpˆΔf d = . (4.23) μ To obtain a ﬁx in two or three dimensions, distances from multiple beacons are then processed in the position calculation engine (see sec. 4.3.2).

29 4.2. Hardware impairments

The following sections describe impairments to the positioning process which are related to hardware nonidealities or physical limitations. Of these, the performance of the integrated synthesizer is of special interest. Glitches in the generation of the frequency ramps have the unpleasant eﬀect of directly aﬀecting the IF measurement. These impairments included phase noise, ramp nonlinearity, and possibly synchronization.

4.2.1. Phase noise

An ideal oscillator exhibits a Dirac spectrum at the oscillation frequency f0. Due to nonideal physical eﬀects, the spectrum of a practical oscillator tends to exhibit a “skirt” spectrum around f0, illustrated in ﬁg. 4.7. Mathematically, this is written as

S(f) S(f)

f f

f0 f0 (a) (b)

Figure 4.7.: Output spectrum S(f) of an (a) ideal oscillator (b) practical oscillator, which exhibits a phase noise “skirt”.

x(t)=A cos (2πf0t +Φ+φ(t)), (4.24) where Φ is a constant and φ(t) a time-variant excess phase term. Phase noise is usually quantiﬁed as the noise power in a 1 Hz band at an oﬀset Δf from the carrier, relative to the carrier power [81]: noise power in 1 Hz band at f +Δf L| =10lg 0 . (4.25) Δf carrier power

Fig. 4.8 shows a typical phase noise shape in frequency domain. The phase noise regions around the carrier exhibit diﬀerent slopes due to diﬀerent physical mechanisms involved [82]:

1/f 3 Related to the physical resonance mechanism of the active oscillator.

1/f 2 Caused by white or uncorrelated timing ﬂuctuations in the oscillation period.

1/f Flicker noise added either by physical resonance mechanism or noisy electronics.

30 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

1/f3

1/f2 1/f

log(f)

f1 f2 f3

Figure 4.8.: Log-log rendition of the classical phase noise power spectral density.

The modeling of 1/f α noise is the topic of several previous works [81–88]. Only the general process is repeated here for the sake of completion. Colored noise is generally generated by filtering of white noise. In [86], a technique for using first order filters to shape different regions of the spectrum is proposed. This method is computationally economic and gives good approximation to the ideal spectrum. Fig. 4.9 shows a computation of the one-sided power spectral density of phase noise.

-20

-40

-60

-80

-100

-120

-140 Power spectral density / dBc/Hz

-160 102 103 104 105 106 f/Hz Figure 4.9.: One-sided power spectral density of phase noise generated with the time-domain method. Good approximation to the theoretical curve is achieved.

The eﬀect of phase noise on the secondary FMCW radar employed by HPLS has been subject to mathematical analysis [65]. It is shown that the error caused by phase perturbation of the frequency ramp is proportional to the time of ﬂight. However, the variance of the localization is very low for reasonable values of phase noise.

31 1

0.8

) 0.6 L <

P( 0.4 L|1MHz = −90 dBc/Hz

L|1MHz = −120 dBc/Hz 0.2

0 -40 -30 -20 -10 0 SNR Figure 4.10.: Detection Probability P (<0.15 m) with phase noise applied. A sensitivity loss is seen only at very high noise levels.

Fig. 4.10, which shows simulation results using the time-domain phase noise generation described above, confirm the assumption that phase noise has only a minor impact on the HPLS system. A low sensitivity loss is seen at a setting of L|1MHz = −90 dBc/Hz. The integrated synthesizer developed for the actual system, described in [69, 89], achieves a closed loop L of -117 dBc/Hz, however. Thus, it is concluded that phase noise has only a minor influence on the system performance, especially when compared to multipath propagation effects.

4.2.2. Ramp nonlinearity In the ideal system, the frequency ramp generated by the transmit and receive synthesizers are assumed to be perfectly linear over the sweep duration T and bandwidth B. Hardware nonidealities, however, may give rise to two distinct eﬀects. The ﬁrst is the distortion of the ramp away from perfect linearity, which can be seen as the superposition of a time-variant frequency term to the instantaneous frequency, written as

ωFM(t)=ω0 + μt + δωFM(t), (4.26) where ω0 is the offset frequency. The impact of this effect has seen thorough treatment by analysis, simulation and measurement in [90], where the authors found that low modulation frequencies generally cause larger errors. A second, unrelated effect is the mismatch that might occur between the slopes of the transmit and receive ramps. This error is very likely to occur in one form or the other, as the synthesizer generates the ramps based on a crystal oscillator reference, which is prone to variations due to voltage fluctuation, temperature change and fabrication tolerance. These fluctuations usually sum up to end in the low ppm range, so it can be said that

Δp f = f (1 + ), (4.27) XTAL1 XTAL2 1e6

32 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

where fXTAL1 and fXTAL2 are the two oscillator frequencies and Δp is the parts deviation in ppm. Assuming they are not used for other timing purposes in the system, the oscillators must then time the ramp periodicity T , which will directly be aﬀected by the frequency deviation. In simulation it is convenient to assume that one oscillator is perfect and the other produces an oﬀset. Thus, the synthesizer phase becomes

1 1 φ(t)= μt2. (4.28) 2 1+1e6Δp

This deviation remains as constant term in the IF signal and causes an offset in the frequency evaluation. Additionally, a residual chirp term will dilute the spectral peak. Thus, this error effects not only the precision, but also the accuracy of the result, i.e., causes a shift of the mean error value. This yields the effect described in fig. 4.3. Fig. 4.11 shows simulation results for

0.8

1 ppm ) 0.6 L 2 ppm

< 3 ppm

P( 0.4 4 ppm

0.2

0 -40 -30 -20 -10 0 SNR Figure 4.11.: Detection probability in the AWGN case, with various settings for oscillator mismatch. It can be seen that this is a systematic error which shifts the mean away from the detection cone. various ppm settings. The detection probability breaks down at 4 ppm: here, it can be seen that the mean value of the error is shifted outside the detection cone. This causes the system to perform better at lower SNR values.

4.3. Signaling impairments

Impairments treated in this section do not pertain shortcomings of the physical hardware, but are related to signal properties or signal processing algorithms. Multipath propagation is generally understood to be the limiting factor in all radiolocation systems, and is thus given special attention.

33 4.3.1. Multipath propagation Propagation over multiple signal paths is a well known problem in microwave transmission. In communications, it accounts for complex receiver structures, but yields the beneﬁcial eﬀect of allowing communication between terminals which have no LOS connection. In local positioning,

|h(t)|2

§ B/T

Figure 4.12.: Illustration of multipath propagation. The signal emitted from the beacon reaches the sensor on multiple paths with increasing delays. multipath propagation is usually by far the largest source of error in the system. Fig. 4.12 illustrates the principal mechanism: A radio signal emitted by the beacon reaches the sensor on multiple signal paths. In local positioning, a LOS connection is essentially mandatory to obtain a meaningful position fix. Multiple secondary rays reach the sensor, each with a specific delay relative to the LOS path (excess delay) and specific amplitudes. The source of these secondary rays are given in [72] as:

• Reﬂection, occurring when the transmit signal impinges on a smooth surface with very large dimensions as compared to the wavelength λ.

• Diﬀraction, which occurs when secondary waves form behind a large obstructing body.

• Scattering refers to the spread of radio waves when the source wave hits small (with respect to λ) objects.

Mathematically, the transmit signal xTX(t) can be thought of being convolved with the characteristic Channel Impulse Response (CIR) h(t), producing

xMP(t)=xTX(t) ∗ h(t). (4.29)

34 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

The CIR generally takes on the form

Np h(t)= cnδ(t − τn), (4.30) n=1

th where Np is the total number of path components, cn the complex amplitude of the n compo- th nent with magnitude |cn| and phase θn,andτn the speciﬁc delay of the n component. Note th that while τn gives an absolute time delay value, τe,n is the excess delay of the n component, with the LOS component delay as reference (τe,LOS =0). The exact form of (4.30) is strongly dependent on the actual scenario. Outdoors, a strong LOS component may be accompanied by few and widely spaced secondary rays. Indoor scenarios, which are highly relevant for HPLS, are generally more challenging. In factory halls, there is a multitude of highly reﬂective objects, and ample opportunity for shadowing and scattering.

Measurement-based channel model Meaningful modeling of a propagation channel can be based on actual channel measurements followed by model abstraction. The following steps are described in [91], in which a project- speciﬁc channel model was devloped:

1. Measure the channel transmittance S21 (i.e., the frequency response) with a Vector Network Analyzer and suitable antennas with sufficient bandwidth to resolve closely spaced path components. 2. Perform an Inverse Fast Fourier Transform (IFFT) to obtain the discrete time-domain CIR h[n]. 3. Reduce the model order by eliminating low-power or high-delay components. 4. Fit the measurement to a channel model by Maximum Likelihood Estimation (MLE). ◦ ◦ Distribution of the component phases θn is assumed to be uniform over (0 , 360 ). A fitting and flexible model for the magnitudes |cn| is the Nakagami-m distribution. The pdf is given by μ μ 1 − − μ 2 p (x; μ; ω)=2 x(2μ 1)e ω x , (4.31) Nakagami ω Γ(μ)

where μ and ω are called the shape and scale parameters, respectively, and Γ(·) is the gamma function. Excess delays τe,n were ﬁtted to a lognormal distribution:

(ln x−μ)2 1 − 2 pLognormal(x; μ; σ)= √ e 2σ (4.32) xσ 2π Fig. 4.13 illustrates distributions suitable for modeling of the multipath components. In simulation, a set of pre-calculated channels with a very high sampling rate is used. These are resampled ad-hoc to ﬁt the current simulation bandwidth and convolved with the transmit signal. This will in the following be called the Warsaw channel model as a reference to its place of inception.

Alternative channel models Based on statistical analysis of empirical data, general channel models including parameters can be derived for a variety of scenarios. This has a long tradition in mobile communications, where bandwidths are usually very small compared to what is encountered in positioning.

35 1.5 0.4 μ =0.5, ω =1 μ =0.5, σ =1 μ =1,ω =1 μ =1,σ =1 μ =1,ω =3 μ =1,σ =3 1 μ =2,ω =1 μ =2,σ =1 μ =2,ω =2 μ =2,σ =2 0.2 0.5

0 0 1 1.5 22.53 2 46810 (a) (b)

1 1 σ =0.5 σ =1,ν =0 σ =1 σ =1,ν =0.5 σ =2 σ =1,ν =2 σ =3 σ =0.25, ν =1 σ =4 σ =0.25, ν =2 0.5 0.5

0 0 2 4 6 8100 510 (c) (d)

Figure 4.13.: Various distributions suitable for representing parameters of h(t). (a) Nakagami-m distribution (b) Lognormal distribution (c) Rayleigh distribution (d) Rice distribution.

One of the most widely used model is the Saleh-Valenzuela model for indoor radio channels [92]. It is based on the assumption that the CIR can be separated into clusters of singular rays (path components). Relevant parameters for this model are then the distribution of clusters, the amplitude fading within clusters and the (temporal) distribution of rays within each cluster. In the Saleh-Valenzuela model, the CIR is written as ∞ ∞ jΦj,i h(t)= aj,i · e · δ(t − Ti − τj,i). (4.33) i=0 j=0

th th Here, Ti is the arrival time of the i cluster; τj,i is the delay of the j component within and th relative to the i cluster; and aj,i and Φj,i describe the amplitude and phase of the respective components. Clusters and rays are assumed to arrive independently, so the arrival times follow a Poisson process. This mandates that the interarrival times of clusters are exponentially distributed as

−Λ(Ti−Ti−1) P(Ti|Ti−1)=Λe (4.34)

−λ(τj,i−τj−1,i) P(τj,i|τj−1,i)=λe . (4.35)

36 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

Amplitude coeﬃcients are assumed to follow a Rayleigh distribution:

2 a −a a2 j,i j,i/ j,i P(aj,i)=2 e , (4.36) 2 aj,i and the phase coeﬃcients Φj,i are uniformly distributed in [0, 2π). The Saleh-Valenzuela model, however successful in mobile communications, is not directly suitable for simulation of wideband or even ultra-wideband positioning systems, which mandate a path resolution of up to 1 ns. On grounds of the Saleh-Valenzuela model, the IEEE 802.15.4a working group developed a model for UWB channels which takes frequency-selective path loss into account [93]. The IEEE 802.15.4a channel model for UWB channels deﬁnes the CIR in reference to the Saleh-Valenzuela model as

K J jΦj,i h(t)= aj,i · e · δ(t − Ti − τj,i). (4.37) i=0 j=0

The number of clusters is limited and a random variable:

L L P(L)= e−L, (4.38) L! where L is the mean value. Interarrival times between clusters remain unchanged to the Saleh-Valenzuela model, but interarrival times of rays is now the sum of two exponential functions:

−λ1(τj,i−τj−1,i) −λ2(τj,i−τj−1,i) P(τj,i|τj−1,i)=βλ1e +(1− β)λ2e . (4.39)

For LOS scenarios, the mean energy of the cluster is an exponential function:

−τ {| 2 |} ∝ j,i/γi E aj,i Ωie , (4.40)

th where Ωi is the total energy of the i cluster, and γi a time constant, which is in turn deﬁned as γi = kγ Ti + γ0, (4.41) where kγ and γo are constant. For Non-Line of Sight (NLOS) scenarios, the cluster energy is

−τ −τ {| 2 |} ∝ − j,i/γrise j,i/γ1 E aj,i (1 ξe )e , (4.42) where ξ, γrise and γ1 are model-specific parameters, which describe the position of the cluster maximum. The amplitude coefficients follow a Nakagami-distribution, as described in 4.31. The IEEE 802.15.4a model features different sets of parameters which describe varying scenarios, such as office, outdoor and industrial settings. Tabulated parameters for these models, which have been used in simulations, can be found from [93].

System performance The ability to resolve closely spaced paths depends on three parameters: • The system (ramp) bandwidth B, • the type of window function and

37 •theFFT bin size (frequency domain resolution). In fig. 4.14, the effect of increasing B is illustrated in a noise-free channel scenario (100 dB SNR)withaCIR given by ⎧ ⎪ ⎪1forn =0, ⎪ ⎪0.9forn =2, ⎨⎪ 0.8forn =3, h[n]=⎪ (4.43) ⎪0.5forn =9, ⎪ ⎪0.6forn =11, ⎩⎪ 0otherwise. For the sake of brevity, this will henceforth be shortened to a vector of tuples containing excess delays and relative amplitudes, as in h[n]= (0, 1), (2, 0.9), (3, 0.8), (9, 0.5), (11, 0.6) , (4.44) with all other amplitudes assumed to be 0. Note that the sampling rate at the channel is fs,RF, which implies a minimal path spacing of 25 cm. The figure shows the expected peak positions

1 1 2 2 | | ) ) f f ( ( 0.5 0.5 X X | |

0 0 4.9 5 5.1 5.2 0.99 1 1.01 1.02 1.03 1.04 5 6 f/Hz ×10 f/Hz ×10 (a) (b)

1 1 2 2 | | ) ) f f

( 0.5 ( 0.5 X X | |

0 0 1.66 1.68 1.7 1.72 3.32 3.34 3.36 3.38 3.4 6 6 f/Hz ×10 f/Hz ×10 (c) (d)

Figure 4.14.: The effect of bandwidth on the separability (resolution) of paths, exemplified on a simplified CIR with 5 multipath components. (a) B = 150 MHz (b) 300 MHz (c) 500 MHz (d) 1 GHz.

of (4.44) as vertical lines. It can be seen that in the 150 MHz case, only two peaks are clearly

38 CHAPTER 4. SINGLE NODE ARCHITECTURE AND PERFORMANCE ANALYSIS

resolved, while the components at n =2andn = 3 fuse with the LOS component. This has an immediate impact on the resolution of objects. The characteristic main- and sidelobes of a closely spaced second path, whose shape and number depend on the window function, will perturb the LOS main lobe and cause detection errors. This eﬀect is illustrated for the rectangular and Hanning window types in ﬁgs. 4.15 and 4.16. In both simulations, the applied (real) CIR was given by

h[n]= (0, 1), (i, 0.9) , (4.45) with i =[1, 2, 3, 5, 7, 9]. The plots show how the detected peakp ˆDetected oscillates statistically aroundthetruepeakˆpActual. The impression that the maximum error is larger when using the Hanning window is confirmed by fig. 4.17. Here, the error caused by the second path component at normalized excess delays is plotted for different window types. As can be expected from the parameters in tab. 4.2, wide main lobes cause large errors at narrow path spacings. The rectangular function has the smallest Δω, but the worst sidelobe suppression, which results in considerable error values event at 30Δtfs,FE, which corresponds to a target 7.5 m away from the second sensor. A second prominent effect of multipath propagation is the cancellation of incident waves which have deviating phase terms. The phase change experienced by a wave is dependent on the type of reflection (diffraction, scattering, etc.) and the physical properties of the material the wave impinges on [72]. In most channel models, the phase is assumed to be uniformly distributed in the interval [0, 2π). Fig. 4.18 shows the same sequence of received delay profiles as in fig. 4.14, only with a random phase shift applied. Thus, the CIR is complex-valued, with

h[n]= (0, 1), (2, 0.9, Φr,1), (3, 0.8, Φr,2), (9, 0.5, Φr,3), (11, 0.6, Φr,4) . (4.46)

Each element now corresponds to a triplet (Δτ,|a|, ∠a), where Δτ is the excess delay and |a| and ∠a the absolute value and angle of the complex number a.Φr,i represents a uniformly distributed random phase in [0, 2π). The detrimental effect of the phase shift is evident as subsequent path components randomly cancel and reinforce each other as well as the LOS path. The increased bandwidth even merits a detrimental effect in this special case, as the last component of the delay profile is now resolved and, by statistical fluke, has and even greater amplitude than the LOS component. Brute-force peak detection would in this case yield a very large distance error. Fig. 4.19 shows a detection probability plot for multipath propagation according to the Warsaw model, with approximately 10,000 model instances used. In comparison to the reference AWGN case, two effects are observed. First, a shift towards higher SNR values indicates a loss in precision due to the aforementioned effects of peak distortion and cancellation. This effectively reduces the range of the system and can thus count as example of sensitivity loss. The second effect is the saturation of the detection probability towards a value of about 0.9. This implies that no matter how little noise is present in the system, there is a limit to the achievable precision due to multipath propagation effects and limited bandwidth. Fig. 4.20 compares several settings for the system bandwidth B in the Warsaw model. Following fig. 4.14, an increase in bandwidth should mitigate the channel effects. As is evident, this is the case and has the effect of both improving on the saturation error and the system sensitivity. The panel in fig. 4.21 shows detection probabilities for various scenarios of the IEEE 802.15.4a model. They represent environments which may typically be encountered in local positioning applications. It is interesting to note that with this channel model applied, the system generally performs significantly worse than with the Warsaw model, and different scenarios vary widely in their effect on detection probability. In the face of these facts, the solidity of any argument made

39 1 1 2 2 | | ) ) f f ( ( X X | |

0 0 4000 4050 4100 4150 4200 4000 4050 4100 4150 4200 Sample index Sample index (a) (b)