Advanced Methods and Protocols for Wake-up Receivers

Advanced Methods and Protocols for Wake-up Receivers Dissertation zur Erlangung des Doktorgrades der Ingenieurwissenschaften Vorgelegt von M.Sc. Timo Kumberg

Technische Fakultät Institut f¨ ur Mikrosystemtechnik (IMTEK) Albert-Ludwigs-Universität Freiburg Freiburg im Breisgau Deutschland, 2017

ii Institut f¨ ur Mikrosystemtechnik (IMTEK) Professur f¨ ur Elektrische Mess- u. Pr¨ ufverfahren Albert-Ludwigs-Universität Freiburg Freiburg im Breisgau Author

Timo Kumberg [email protected]

Gutachter

Prof. Dr. Leonhard M. Reindl Prof. Dr. Christian Schindelhauer

Dekan

Prof. Dr. Oliver Paul

Tag der Promo- 17.01.2018 tion

To Bojana, Milana and Desanka

Abstract The most demanding task in a wireless sensor network is probably to supply the sensor nodes with energy. This supply is usually limited as it is often realized by batteries. A sensor node draws a certain amount from this limited energy storage for every task it performs. Consequently, the total amount of tasks a wireless sensor node is able to perform, becomes limited by its power reserves. Wake-up receivers can reduce the amount of communication activities as they make idle-listening obsolete. Therefore, wake-up receivers can increase the lifetime of a wireless sensor node. A downside of ultra-low power wake-up receivers is their low sensitivity compared to that of communication radios. As a result of the lower sensitivity of wake-up receivers, there is a reduction in their effective operating range. Furthermore, the low sensitivity causes unreliable wake-up detection in certain cases. The methods and algorithms presented in this work mitigate these effects. Chapter Three presents a novel sensor node design with antenna diversity in the wake-up path resulting in an increased wake-up detection robustness and reliability which is experimentally demonstrated. The design requires only two additional antenna switches, and thus, the power requirements of the node increase only marginally. The first part of Chapter Four presents a novel routing protocol for wireless sensor networks with wake-up receivers. The protocol explicitly utilizes the different ranges of wake-up receiver and communication radio by skipping several nodes during data communication. Furthermore, as the nodes are not bounded to a duty-cycling timeframe, several data messages can be communicated utilizing the open link. Based on a theoretical analysis, the performance of the novel routing protocol is compared to existing state-of-the-art algorithms, and it is demonstrated that this new protocol outperforms the current ones in many scenarios. The second part of Chapter Four presents a novel technique to transmit concurrent wake-up messages by purposefully interfering signals of two or more wireless sensor nodes. This creates a wake-up signal by using the beat frequency of the superposed and slightly out of tune carrier signals. The method is analyzed theoretically and verified experimentally. It is demonstrated that this technique increases the received signal strength thereby increasing the wake-up robustness and reliability, and the wakeiv

v up range. Two algorithms are outlined to include this technique into an existing wireless sensor network. As will be discussed, applying the first algorithm increases the transmitted signal strength and nodes that were formerly out-of-range or responding only unreliable, can be accessed with more confidence. The second algorithm is a wake-up flooding protocol that can be used to quickly wake-up several sensor nodes in a wireless sensor network. Chapter Five presents the deployment of a wireless sensor network with wake-up receivers for the structural vibration analysis of a highway bridge using positioning information from a global satellite positioning system. Applying a short time Fourier transform on the measurement data reveals a resonant mode of the bridge. As there are no models or frequency analysis calculations available for this bridge, the results are compared to measurements achieved with acceleration sensors. The wireless sensor network implements the routing protocol presented in this work to transmit the data. In summary, the algorithms and methods presented in this thesis increase the reliability and robustness of wake-up sensors and their communication messages to better enable large sensor network deployments with longer operational uptime using wake-up receivers that incorporate novel techniques for wake-up communication.

Zusammenfassung Die Energieversorgung der drahtlosen Sensorknoten ist die wahrscheinlich größte Herausforderung im Bereich drahtloser Sensornetzwerke. Ein Sensorknoten benötigt f¨ ur jede durchzuf¨ uhrende Aktivität eine gewisse Menge an Energie welche oftmals von endlichen Energiespeichern, wie zum Beispiel von Batterien, geliefert wird. Somit ist die Anzahl der Aktivitäten die ein Sensorknoten durchf¨ uhren kann endlich. Aktivitäten sind zum Beispiel die Erfassung von Umweltparametern und das Kommunizieren dieser Messungen an eine Senke. Aufweckempfänger können die Anzahl der benötigten Aktivitäten deutlich reduzieren, da das sogenannte Idle-Listening nicht mehr stattfindet. Dementsprechend können sie die Lebenszeit eines Sensorknotens verlängern. Allerdings ist die Sensitivität von Aufweckempfängern niedriger als die der Kommunikationsempfängern. Dies f¨ uhrt dazu, dass die mögliche Aufweckreichweite geringer ist als die mögliche Kommunikationsreichweite und dazu, dass Aufwecknachrichten unter gewissen Umständen nur unzuverlässig detektiert werden. Im dritten Kapitel dieser Arbeit, wird erstmalig ein Aufweckempfänger vorgestellt, der die Detektionswahrscheinlichkeit von Aufwecknachrichten durch Antennendiversität erhöht. Das betrachtete Konzept benötigt nur zwei zusätzliche Antennenschalter, und erhöht somit den Energiebedarf des Sensorknotens nur um wenige nano- bis µA. Im ersten Teil des vierten Kapitels wird ein neuartiges Routingprotokoll f¨ ur Sensorknoten mit Aufweckempfängern vorgestellt. Das Protokoll verwendet explizit die unterschiedlichen Reichweiten von Aufweck- und Kommunikationsempfänger, um beim Senden von Daten mehrere Knoten zu u ¨berspringen und dadurch den Energieverbrauch zu reduzieren. Da die Sensorknoten keine Duty-Cycling Periode einhalten m¨ ussen, können u ¨ber einen bestehenden Link mehrere Pakete gesendet werden, und der Energieverbrauch wird weiter reduziert. Der zweite Teil des vierten Kapitels befasst sich mit dem Versenden von zeitlich synchronisierten Aufwecknachrichten. Das in dieser Arbeit vorgestellte Konzept generiert erstmalig Aufwecknachrichten durch eine Schwebung, erzeugt durch im Frequenzband leicht versetzt sendende Sender. Dadurch erhöht sich die empfangene Signalstärke, was wiederum die Robustheit und Zuverlässigkeit der Aufwecknachrichten erhöht und deren Reichweite vergrößert. Das neuartige Konzept wird in dieser Arbeit erstmalig theorevii

viii tisch analysiert und mittels praktischer Experimente verifiziert. Am Beispiel zweier prototypisch umrissener Kommunikationsprotokolle, eines Flooding-Algorithmus und eines Unicast-Algorithmus, wird aufgezeigt wie sich dieses Verfahren in bestehende drahtlose Sensornetzwerke integrieren lässt. Im f¨ unften Kapitel dieser Arbeit wird ein drahtloses Sensornetzwerks mit Aufweckempfängern präsentiert, welches der strukturellen Schwingungsanalyse einer Autobahnbr¨ ucke mit Hilfe von Satelliten gest¨ utzter Positionserfassung, dient. Dabei wird der in ¨ dieser Arbeit entwickelte Routingalgorithmus zur Ubertragung der Daten verwendet. Mit Hilfe einer Fourier-Analyse der Positionsdaten wird eine Eigenfrequenz der Br¨ ucke bei circa 0.3 Hz bestimmt. Zusammenfassend erhöhen die in dieser Arbeit vorgestellten Methoden und Algorithmen die Roustheit und Zuverlässigkeit von Aufwecknachrichten, ermöglichen den Einsatzt von Aufweckempfängern in großen drahtlosen Sensornetzwerken und zeigen eine neuartige Möglichkeit Aufwecknachrichten zu generieren.

©

Copyright 2017 by Timo Kumberg. The copyright of this thesis rests with the author. No quotations from it should be published without the author’s prior written consent and information derived from it should be acknowledged.

Contents Abstract

iv

Zusammenfassung

vii

List of Figures

xiii

List of Tables

xviii

1 Introduction 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Objectives and Contributions of this Work . . . . . . . . . . . . . . . . 1.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Radio Wave Propagation and Global Navigation 2.1 Radio Wave Propagation . . . . . . . . . . . . . . 2.1.1 Large-scale Fading . . . . . . . . . . . . . 2.1.2 Small-scale Fading . . . . . . . . . . . . . 2.2 Satellite-based Positioning . . . . . . . . . . . . . 2.2.1 GPS and GLONASS Signals . . . . . . . . 2.2.2 Pseudorange Measurements . . . . . . . . 2.2.3 Pseudorange Measurement Errors . . . . . 2.2.4 Differential GNSS . . . . . . . . . . . . . .

Satellite System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 Wireless Sensor Nodes with Wake-up Receiver 3.1 Wake-up Receiver . . . . . . . . . . . . . . . . . . . 3.1.1 False Positive and False Negative Wake-ups 3.1.2 Experimental Analysis . . . . . . . . . . . . 3.2 Wireless GNSS Node . . . . . . . . . . . . . . . . . 3.3 Gateway and Remote Server . . . . . . . . . . . . . 3.4 Wireless Sensor Node with Antenna Diversity . . . 3.4.1 Experimental Results . . . . . . . . . . . . . 3.4.2 Conclusion . . . . . . . . . . . . . . . . . . . xi

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

1 1 2 4

. . . . . . . .

5 5 5 8 13 14 17 20 21

. . . . . . . .

27 27 29 30 32 34 35 37 42

xii

Contents

4 Communication with Wake-up Receivers 4.1 Existing Protocols for Wake-up Receivers . . . . . . . . 4.1.1 Opportunistic routing with wake-up receivers . 4.2 T-ROME . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 T-ROME Protocol . . . . . . . . . . . . . . . . 4.2.2 State Machine . . . . . . . . . . . . . . . . . . . 4.2.3 Experimental Analysis . . . . . . . . . . . . . . 4.2.4 Numerical Analysis . . . . . . . . . . . . . . . . 4.2.5 Results . . . . . . . . . . . . . . . . . . . . . . . 4.2.6 Discussion . . . . . . . . . . . . . . . . . . . . . 4.2.7 Conclusions . . . . . . . . . . . . . . . . . . . . 4.3 T-ROME in Noisy Environments . . . . . . . . . . . . 4.3.1 Routing . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Experimental Results and Performance Analysis 4.3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . 4.4 Exploiting Concurrent Wake-up Transmissions . . . . . 4.4.1 Introduction . . . . . . . . . . . . . . . . . . . . 4.4.2 Wireless interference . . . . . . . . . . . . . . . 4.4.3 Concurrent Wake-up Message . . . . . . . . . . 4.4.4 Two concurrent senders . . . . . . . . . . . . . 4.4.5 More than two concurrent senders . . . . . . . . 4.4.6 Concurrent Wake-up Protocol Design . . . . . . 4.4.7 Experimental Results . . . . . . . . . . . . . . . 4.4.8 Expected concurrency of two senders . . . . . . 4.4.9 Conclusions . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

5 Assessment of the Modal Properties of 5.1 Introduction . . . . . . . . . . . . . . . 5.1.1 Cross-layer Routing Protocol . . 5.2 Experimental Results . . . . . . . . . . 5.3 Conclusions . . . . . . . . . . . . . . .

Bridge . . . . . . . . . . . . . . . . . . . .

the Neckartal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . . . . .

46 47 50 52 53 58 61 63 69 74 76 76 77 78 80 80 80 81 84 84 88 91 94 94 99

. . . .

101 102 103 105 113

6 Summary and Outlook 115 6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 6.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 A Publications

118

B Acknowledgments

120

List of Figures 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8

2.9 2.10 2.11 2.12

2.13

3.1 3.2 3.3 3.4 3.5

Model for multipath propagation including ground reflection. . . . . . . Schematic of a selection diversity system where the receiver is always connected to the antenna with the highest RSSI value. . . . . . . . . . Schematic of a maximal-ratio combining diversity system. . . . . . . . . Schematic of a receiver with maximal ratio combining diversity for two input channels. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simplified block diagram of a generic digital GPS receiver. . . . . . . . Block diagram of a generic digital receiver channel. . . . . . . . . . . . Block diagram of a generic carrier tracking loop. . . . . . . . . . . . . . Illustration of the code correlation assuming idealized input signal shapes and correlator output for reasons of simplicity. (a) replica code 1/4 chip early, (b) replica code aligned, (c) replica code 1/4 chip late. . . . . . . Representation of the position vectors. . . . . . . . . . . . . . . . . . . Determination of GNSS code transmission time T − T S . . . . . . . . . Dilution of precision depending on the geometric appearance of the satellites (a) high dilution of precision (b) improved dilution of precision. . . Geometry of single differencing with known position of receiver A and unknown position of receiver B. Both receiver measure their distance ρ to the same satellite j. . . . . . . . . . . . . . . . . . . . . . . . . . . . Geometry of double differencing with known position of receiver A and unknown position of receiver B. Both receiver measure their distance ρ to satellites j and k. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Schematic of wireless sensor node including a wake-up receiver. . . . . . The low-frequency wake-up message (red) is modulated on the highfrequency carrier signal by On Off Keying. . . . . . . . . . . . . . . . . Photo of sensor node with wake-up receiver. . . . . . . . . . . . . . . . Wake-up pattern of the AS3932 LF wake-up receiver. Pattern and data are optional. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Manchester coded wake-up signal consisting of carrier burst, preamble and address pattern. . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

6 11 11 12 15 16 17

18 19 19 22

24

25 28 28 29 29 31

xiv 3.6

List of Figures Experimentally measured false negative wake-up rate over input signal strength in dBm, nodes connected by cables. . . . . . . . . . . . . . . .

31

3.7

Schematic of the GNSS wireless sensor node. . . . . . . . . . . . . . . .

32

3.8

Photo of the GNSS sensor node. . . . . . . . . . . . . . . . . . . . . . .

33

3.9

Structure of a range data log. . . . . . . . . . . . . . . . . . . . . . . .

33

3.10 States of a sensor node. . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

3.11 Lifetime of the GNSS sensor node.

. . . . . . . . . . . . . . . . . . . .

34

3.12 Schematic of the base station. . . . . . . . . . . . . . . . . . . . . . . .

35

3.13 Photo of the GSM base station node. . . . . . . . . . . . . . . . . . . .

35

3.14 Block diagram of a low-power wake-up receiver with antenna diversity. Each diversity branch consists of antenna, matching network and rectifier. 36 3.15 Photo of the wireless sensor node with equal gain diversity. . . . . . . .

37

3.16 Schematic of the diversity system. . . . . . . . . . . . . . . . . . . . . .

37

3.17 Measured voltage at the rectifier output with one active antenna. The typical sensitivity of the AS3932 is depicted as dotted line and the red circle shows the measured sensitivity. . . . . . . . . . . . . . . . . . . .

38

3.18 Reflections measured at the straight (yellow) and angular (green) antenna input ports over frequency. . . . . . . . . . . . . . . . . . . . . .

39

3.19 Measured voltage at the rectifier output with two active antennas. Typical sensitivity of the AS3932 is depicted as blue dotted line. The red circle marks the measured sensitivity. . . . . . . . . . . . . . . . . . . .

40

3.20 Experimentally measured gain in dB of the two antenna diversity system. 41 3.21 Multipath laboratory setup with randomly placed objects. . . . . . . .

42

3.22 Required power to wake-up the receiver at different positions. . . . . .

44

3.23 Gain achieved by the equal-gain diversity wake-up receiver system over the selection diversity wake-up system for the 20 location in ascending order. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

4.1

4.2

4.3

Communication layer stack consisting of physical layer, link layer, routing layer and application. The cross-layer protocol T-ROME supports functions in the link and the routing layer as depicted in the figure. The Wake-up is embedded in the link layer and supports the RTS/CTS scheme based on MACA to reduce packet collisions. . . . . . . . . . . .

53

Schematic of a simple tree routing protocol with nodes a to g. Communication is only possible from child to parent for example from node b to node a. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

Schematic of the wake-up multi-hop routing protocol developed in this work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

List of Figures 4.4

4.5 4.6 4.7 4.8 4.9 4.10

4.11 4.12

4.13

4.14

4.15

4.16 4.17 4.18 4.19 4.20

4.21

125 kHz wake-up call packet (WUC) including 32 to 100 byte carrier burst, 52 byte preamble and 64 byte receiver ID sent at 8192 byte per second. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wake-up acknowledge (WUC ACK) packet consisting of 3 byte (Protocol ID, Receiver ID low byte and Receiver ID high byte). . . . . . . . . . . Packet flow of wake-up and main radios in Wake-up Layer. . . . . . . . Data packet consisting of 4 byte (Packet Type DATA, Source ID, Destination ID and payload length). . . . . . . . . . . . . . . . . . . . . . . Acknowledge packet consisting of 3 byte (Packet Type ACK, Source ID and Destination ID). . . . . . . . . . . . . . . . . . . . . . . . . . . . . Routing packet embedded into MAC packet. . . . . . . . . . . . . . . . R REQ (Routing Request) packet consisting of 4 byte (Number of slots to send (6 bit) Packet Type REQ (2 bit), Source ID, Destination ID and time to live (TTL)). . . . . . . . . . . . . . . . . . . . . . . . . . . . . DATA (Routing Data) packet consisting of 4 byte (Packet Type DATA, Routing Source ID, Routing Destination ID and payload length). . . . R REQ ACK (Routing Acknowledge) packet consisting of 4 byte (Packet Type ACK, current time to live (TTL), Link Quality Identifier (LQI) and number of available memory slots). . . . . . . . . . . . . . . . . . . Sequence diagram of the routing protocol in case the data is sent to the next neighbor. Decision to where the data are sent is done at node A based on information included in the request acknowledge data. . . . . Sequence diagram of the routing protocol for communication to the twohop distant neighbor. Decision to where the data are sent is done at node A based on information included in the request acknowledge data. . . . Sequence diagram of the routing protocol for communication to a threehop distant neighbor. Decision to where the data are sent is done at node A based on information included in the request acknowledge data. State machine of a sensor node for data transmission. . . . . . . . . . . Complete wake-up packet including calibration and mandatory radio bytes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Radio packet including calibration. . . . . . . . . . . . . . . . . . . . . Current drawn by the sensor node in the different states of the protocol. Sending and receiving of 5 data packets in case of 4 participating nodes. Each node (node 13, node 12, node 11 and node 10) has a sending (upper line) and receiving (lower line) state. Node 13 is source, node 10 is sink. Nodes 12 and 11 forward the wake-up calls. . . . . . . . . . . . . . . . T-ROME meta model for node i attempting to wake-up node i + 1. The message is at node i. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xv

55 55 55 55 55 56

56 56

57

57

57

58 59 60 61 62

62 64

xvi

List of Figures

4.22 T-ROME meta model for node j attempting to wake-up node j +1. The message is at node i. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

64

4.23 T-ROME meta model for node m − 1 attempting to wake-up node m. The message is at node i. . . . . . . . . . . . . . . . . . . . . . . . . . .

64

4.24 T-ROME meta model for node m − 1 attempting to wake-up node m. The message is at node m − 1. . . . . . . . . . . . . . . . . . . . . . . .

4.25 T-ROME meta model for node i attempting to transmit data to node j.

65 65

4.26 T-ROME meta model for node i attempting to transmit data to node m. 65 4.27 Markov chain for node i attempting to send data to node j . . . . . . .

66

4.28 Markov chain for node i (that also has the message to be delivered) attempting to wake-up node i + 1 . . . . . . . . . . . . . . . . . . . . .

67

4.29 Meta model for node i attempting to wake-up to node i + 1 using the naive algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

4.30 Markov chain for node i (that also has the message to be delivered) attempting to wake-up node i + 1 using the naive algorithm. . . . . . .

68

4.31 Markov chain for node i attempting to send data to node i + 1 using the naive algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

69

4.32 Placement of the sensor nodes during model verification. . . . . . . . .

69

4.33 Simulated time required to send 1 data packet (dashed) and 5 data packets (solid) along several nodes assuming p = q = 1. The points are data taken from the test setup . . . . . . . . . . . . . . . . . . . . . . .

70

4.34 Simulated time performance of T-ROME and CTP-WUR compared to the naive algorithm, assuming p = q = 1. Black: T-ROME for 1 (dotted), 2 (dashed) and 5 data packets (solid), red: naive algorithm and blue: CTP-WUR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71

4.35 Simulated time performance of T-ROME and CTP-WUR compared to the naive algorithm, assuming p = 0.75 and q = 0.97. Black: T-ROME for 1 (dotted), 2 (dashed) and 5 data packets (solid), red: naive algorithm and blue: CTP-WUR. . . . . . . . . . . . . . . . . . . . . . . . .

72

4.36 Time, energy and power requirements of the sender node.

. . . . . . .

73

4.37 Simulated energy performance of T-ROME and CTP-WUR compared to the naive algorithm, assuming p = q = 1. Black: T-ROME for 1 (dotted), 2 (dashed) and 5 data packets (solid), red: naive algorithm and blue: CTP-WUR. . . . . . . . . . . . . . . . . . . . . . . . . . . .

74

4.38 T-ROME overhead (blue) compared to the overhead of the naive algorithm (black) for sending one data packet . . . . . . . . . . . . . . . . .

74

4.39 T-ROME overhead (blue) compared to the overhead of the naive algorithm (black) for sending 64 data packets . . . . . . . . . . . . . . . . .

75

List of Figures

xvii

4.40 Meta model for node i attempting to wake-up to node i + 1 including the modification. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4.41 Schematic of the wake-up multi-hop routing protocol developed in this work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.42 Sending and receiving of 5 data packets in case of 4 participating nodes. Each node (node 13, node 12, node 11 and node 10) has a sending (upper line) and receiving (lower line) state. Nodes 2 and 3 are relay nodes but node 2 is simulated to be dead. Node 3 forwards the wake-up call. . . . 79 4.43 Performance of T-ROME with and without modification . . . . . . . . 79 4.44 Node D is out of wake-up range due to an outage of node C. . . . . . . 81 4.45 Overview of Zippy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 4.46 Schematic of an on-off-keying modulated message that implements highfrequency carrier signal frequency randomization. . . . . . . . . . . . . 84 4.47 Node-1 and node-2 send a concurrent wake-up to node-3. . . . . . . . . 85 4.48 Spectrum of the sum of two sinusoidal functions f1 and f2 and the resulting carrier and beat frequency. . . . . . . . . . . . . . . . . . . . . 86 4.49 Graphically added sines. . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.50 Beats of different amplitudes. . . . . . . . . . . . . . . . . . . . . . . . 87 4.51 Envelopes of different amplitudes. . . . . . . . . . . . . . . . . . . . . . 87 4.52 Envelopes at different frequencies. . . . . . . . . . . . . . . . . . . . . . 88 4.53 Beats visualized on the real and imaginary axis. . . . . . . . . . . . . . 90 4.54 Simulated and expected amplitude achieved by concurrently sending nodes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.55 Simulated received signal strength over distance for transmit power of 0 and 3 dBm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.56 Concurrent wake-up protocol. . . . . . . . . . . . . . . . . . . . . . . . 92 4.57 Concurrent wake-up flooding protocol. . . . . . . . . . . . . . . . . . . 93 4.58 Distribution of time offsets between two concurrent wake-up packets. . 94 4.59 Normalized signal amplitudes received in case of (a) two, (b) four and (c) six concurrent senders. . . . . . . . . . . . . . . . . . . . . . . . . . 95 4.60 Schematics of the experimental setups. . . . . . . . . . . . . . . . . . . 96 4.61 Number of successfully received wake ups over distance between sender and receiver for single and concurrent wake-ups. . . . . . . . . . . . . . 97 4.62 Visualization of the states (send and receive) of nodes 1, 2 and 3 over time during the described concurrent wake-up protocol. The data was taken with the help of a logic-analyzer. . . . . . . . . . . . . . . . . . . 98 4.63 Timing of the wake-up flooding algorithm. . . . . . . . . . . . . . . . . 98 4.64 Deployed wireless sensor network to analyze the wake-up flooding algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

xviii 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12

List of Figures

GNSS data packet. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . GNSS Payload structure. . . . . . . . . . . . . . . . . . . . . . . . . . . Photo of the Neckar Valley bridge. . . . . . . . . . . . . . . . . . . . . Deployment of the wireless sensor network. . . . . . . . . . . . . . . . . Photo of a GNSS antenna and a relay node. . . . . . . . . . . . . . . . Position E2 (a) x-, (b) y- and (c) z-displacements over time. . . . . . . Position E3 (a) x-, (b) y- and (c) z-displacements over time. . . . . . . Position E4 (a) x-, (b) y- and (c) z-displacements over time. . . . . . . Position E2 (a) x, (b) y and (c) z residuals over time. . . . . . . . . . . Position E3 (a) x, (b) y and (c) z residuals over time. . . . . . . . . . . Position E4 (a) x, (b) y and (c) z residuals over time. . . . . . . . . . . Two-dimensional frequency spectra of the GNSS sensor at position E2 in directions (a) x, (b) y and (c) z. . . . . . . . . . . . . . . . . . . . . 5.13 Two-dimensional frequency spectra of the GNSS sensor at position E3 in directions (a) x, (b) y and (c) z. . . . . . . . . . . . . . . . . . . . . 5.14 Two-dimensional frequency spectra of the GNSS sensor at position E4 in directions (a) x, (b) y and (c) z. . . . . . . . . . . . . . . . . . . . . 5.15 Spectra of the acceleration sensor at position E4 in directions (a) x, (b) y and (c) z. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

104 104 105 106 106 107 108 108 109 110 110 111 112 113 114

List of Tables 1.1 1.2

Receiver (RX) Sensitivity at 868 MHz and transmit (TX) currents at +10 dBm for some typical RF transmitters. . . . . . . . . . . . . . . . List of some wake-up receivers, their sensitivity and power consumption.

2.1

Error sources and amounts of a single receiver pseudorange measurement. 21

3.1

Output voltage of rectifier at decreasing input signal levels using different antenna configurations. . . . . . . . . . . . . . . . . . . . . . . . . . . . The table shows the wake-up sensitivities of the wireless sensor node achieved by measuring the rectified voltage and by feeding the AS3932 with a 868 MHz input signal . . . . . . . . . . . . . . . . . . . . . . . . Required sender power to wake-up the receiver measured at 20 different multipath locations and different antenna configurations. . . . . . . .

3.2

3.3

4.1

4.2

Energy consumption of a network of four nodes (TTL = 3) in mJ of each node for states sending WUC (wake-up call), delay, receive data and transmit data at 3.3 V, when transmitting 500 byte in 5 data packets (100 byte per packet) directly from source to sink. . . . . . . . . . . . . Frequencies used by the nodes during the flooding test set-up. . . . . .

xix

2 3

40

41 43

73 99

Chapter 1 Introduction 1.1

Motivation

Wireless sensor networks are used in many applications like environmental monitoring, home automation, smart manufacturing, infrastructure monitoring and many others. In this context, a wireless sensor network usually consists of many small self-powered sensor nodes that measure their environment, process data and communicate it to other nodes or to a base station [1]. Message transmission can be done via single-hop transmissions or via multi-hop communication resulting in complex network topologies. The most critical parameter of a wireless sensor node is its energy requirement [2] which is vastly dominated by the power required for communication. A lot of research was already done on efficient communication protocols to reduce power consumption and collisions and to increase the throughput of a wireless network [3]. The authors of [3] categorize communication protocols into four groups: asynchronous, synchronous, frame-slotted, and multi-channel protocols. Asynchronous and synchronous protocols are based on duty-cycling, where nodes switch between sleep and active states in order to save energy. To establish a communication link in synchronous protocols like S-MAC or T-MAC, each participating node has to be awake at the same time. This necessitates clock synchronization messages. Asynchronous protocols like B-MAC or WiseMAC nodes use preamble sampling in combination with duty-cycling to detect the beginning of a communication. To minimize collisions frame-slotted protocols allocate different time slots to nearby nodes. Multi-channel protocols use cross-channel communication to realize higher throughput. All these MAC protocols have in common that their energy requirement is linked to the duration of their sleep periods. Longer sleep periods result in lower energy consumption but also in communication latencies. In addition, these MAC protocols require a certain amount of overhead to organize themselves [2]. Recently, wireless sensor networks [4–8] have been upgraded with low-power wakeup receivers. These wake-up receivers have marginal power consumption and wake up 1

2

1. Introduction

the sensor node if a dedicated signal has been received. So, low-power wake-up receivers can greatly reduce the power consumption of wireless sensor nodes, by eliminating the idle listening time and at the same time reduce communication delays to achieve an almost latency free communication [9]. According to He Ba et al. [2], wake-up radios can be categorized into two groups, active and passive wake-up receivers. Passive wake-up receivers harvest their wakeup energy directly from the wake-up message itself, whereas active wake-up receivers require a permanent, yet very low, power supply. In this approach, a wireless sensor node usually incorporates two radio receivers, the main radio for data communication and a second one for receiving wake-up messages [2]. A sensor node wakes up only when it receives a wake-up message and then it turns on its communication radio. Although Bannoura et al. [10] speak of a paradigm shift for wireless sensor protocols with integrated wake-up transceivers, there exist two major challenges [2,10]: First, active wake-up receivers show a higher sensitivity compared to passive ones [2], but their sensitivity is still lower compared to that of state-of-the-art main communication radio transceivers. Secondly, sending wake-up messages may cost more energy than sending of communication messages. Table 1.1 shows the typical sensitivity of some commonly used radio transmitters and their current consumption during transmit state. In Table 1.2 sensitivity and power consumption of some state-of-the-art wake-up receivers are shown. The discrepancy between main radio and wake-up receiver sensitivity is clearly obvious as is the power consumption. Table 1.1. Receiver (RX) Sensitivity at 868 MHz and transmit (TX) currents at +10 dBm for some typical RF transmitters. RF Transceiver

RX Sensitivity [dBm]

TX current [mA]

Si4468 CC1200 CC1101 SPIRIT1

-104 -107 -95 -105

19.7 36 30 21

1.2

Objectives and Contributions of this Work

The goal of this thesis is to develop and to analyze novel techniques, algorithms and protocols for wireless sensor networks with ultra-low power wake-up receivers. The work addresses reliability, robustness, energy, performance and time challenges arising in wireless sensor networks with ultra-low power wake-up receivers. In summary the

1.2. Objectives and Contributions of this Work

3

Table 1.2. List of some wake-up receivers, their sensitivity and power consumption. Wake-up receiver

Sensitivity [dBm]

Power [µW]

Magno and Benini [11] Nilsson and Svensson [12] Gamm et al. [13] Hambeck et al. [14]

-55 -47 -52 -71

1.3 2.3 5.6 2.4

contributions of this work include: The design of a flexible and ultra-low power wireless sensor node with wakeup receiver. The sensor node can be equipped with several types of sensors. Additionally, a specialized sensor node is introduced for highly precise position measurements using global navigation satellite system (GNSS) receivers. The theoretical study and experimental validation of a novel wake-up receiver design that uses equal-gain diversity of two independent propagation paths received by two antennas at different polarizations. The novel design mitigates small-scale fading effects and significantly increases robustness and reliability of the wake-up communication in multipath environments. The development of a Markov chain based framework to model and analyze wakeup receiver based routing protocols. The model will be introduced theoretically and verified experimentally. The development, theoretical analysis and experimental verification of a simple and energy efficient cross-layer routing protocol for wireless sensor networks with wake-up receivers. The protocol makes use of the different transmission ranges of wake-up and main radios in order to save energy by skipping nodes during data transfer. With respect to energy consumption and latency, it outperforms existing protocols in many scenarios. The theoretical analysis and experimental verification of a novel communication technique that uses concurrently transmitted wake-up signals using the beat frequencies of two or more wireless sensor nodes. It will be shown that the newly introduced method increases robustness and reliability of ultra-low power based wireless sensor networks. Furthermore a communication algorithm and a flooding protocol based on the newly developed approach are introduced. The feasibility of the protocols will be demonstrated by means of an outdoor experiment and an indoor setup consisting of several nodes. Due to the concurrency, the flooding

4

1. Introduction algorithm achieves up to 100 % wake-up rates of multiple sensor nodes in less than 20 ms. A wireless sensor network was deployed at the Neckartal Bridge for short-term assessment of the modal properties of the bridge. By using the previously introduced hardware, a position accuracy in the mm-range could be achieve for logging frequencies of up to 20 Hz.

1.3

Outline

This work is divided into six chapters between hardware, software, and application parts including the introductory chapter. Parts of Sections 4.2.4 and 4.4.4 were achieved in cooperation with the Chair of Computer Networks and Telematics, where Prof. Chirstian Schindelhauer contributed to the theoretical analysis. The designs and developments of the wireless sensor nodes as well as the algorithms and the application were mostly done under the BMBF project: Intelligente Br¨ ucken - Infromationssystem zum Strukturmonitoring und Erhaltungsmanagement von Br¨ ucken (IB-ISEB). The rest of the thesis is organized as follows. Chapter 2 discusses the main theoretical backgrounds used throughout this work. The wireless sensor nodes developed and designed in this work are described in Chapter 3. The first part of Chapter 3 introduces a more general design of a wireless sensor nodes with wake-up receiver along with a detailed description of a wireless sensor node for global navigation satellite system (GNSS) receivers. The second part of Chapter 3 presents a novel sensor node design that introduces for the first time a wake-up receiver with equal gain diversity. Chapter 4 mainly introduces the newly developed algorithms throughout this work, namely T-ROME, a simple and energy efficient tree routing algorithm for wireless sensor networks with wake-up receivers, and a novel technique to exploit concurrent wake-up messages. In Chapter 5, a wireless sensor network deployed at the Neckar valley bridge on the A81 near Stuttgart, is presented. The network is based on the hardware presented in Chapter 3 and uses the routing algorithm T-ROME introduced in Chapter 4. Finally, Chapter 6 concludes the main achievements of this thesis and gives an outlook for future work.

Chapter 2 Radio Wave Propagation and Global Navigation Satellite System 2.1

Radio Wave Propagation

Reflection, scattering, and diffraction effects influence radio wave propagation. As a result, radio waves travel multiple paths of different lengths on their way from a sender to a receiver. In the receiver, electromagnetic waves superpose and generate multipath fading effects. In summary, the signal strengths at the receiver decrease as the distance to the sender increases. Propagation models can estimate the averaged received signal strength at a certain distance from the sender. Large-scale propagation models predict the averaged received signal strength. Small-scale propagation models predict the rapid signal strength fluctuations incident for short time periods at a certain distance from the transmitter.

2.1.1

Large-scale Fading

Friis transmission equation, a large-scale model, can be applied to calculate the freespace (line-of-sight) transmission distance d of radio waves as given in Equation (2.1) [15]. PR (d) =

PT GT GR λ2 . (4π)2 d2 L

(2.1)

Here λ is the wavelength, GT and GR are the antenna gains of transmitter (T ) and receiver (R) and PT is the transmitted power. PR resembles the expected received power at distance d. L incorporates additional losses in the transmitter, such as losses in the transmission line or similar. Equation (2.1) does not consider multipath propagation effects. Establishing a more accurate model to estimate received signal strengths over large distances will be presented in the following. 5

6

2. Radio Wave Propagation and Global Navigation Satellite System

Figure 2.1 illustrates the two-ray ground reflection model [16]. Transmitter T is at height ht and receiver R is at height hr . An electrical field Ed propagates in line-of-sight from the transmitter to the receiver. A second wave represented by Ei propagates from the sender and is reflected by the ground under angle θo and travels to the receiver as Eg . At the receiver, both fields Ed and Eg superpose to the total field Et . The waves propagate as sinusoidal functions and along different distances and interfere in the receiver. If they interfere constructively or destructively depends on the phase difference of the waves.

T Ed

R

ds s1

ht

Eg

Ei

d1

Δh

s2 θo

θi

hr

d2

d Figure 2.1. Model for multipath propagation including ground reflection. √ Considering Figure 2.1 it is clear that ds = d2 + ∆h2 . Furthermore, it is obvious that d = d1 + d2 and θi = tan−1 (ht /d1 ) as well as θo = tan−1 (hr /d2 ). Using the dependency of θi = θo , one can see that ht /d1 = hr /d2 which can be rewritten as d1 hr = (d − d1 )ht and solving for d1 reveals Equation (2.2) [17]: d1 = dht /(hr + ht ).

(2.2)

Substituting d1 = d − d2 in Equation (2.2) reveals Equation (2.3) [17]: d2 = dhr /(hr + ht ).

(2.3)

p p From Figure 2.1 one can also see that s1 = d21 + h2t and s2 = d22 + h2r . Inserting Equations (2.2) and (2.3) in these relations, s1 and s2 can be expressed as [17]: s 1+

d2 (hr + ht )2

(2.4)

1+

d2 . (hr + ht )2

(2.5)

s1 = ht and s s2 = hr

2.1. Radio Wave Propagation

7

Furthermore, Equation (2.6) can be extracted from Figure 2.1 [17]: ds =

p d2 + (ht − hr )2 .

(2.6)

As illustrated, two electrical fields (Ed andEi ) propagate from sender to receiver. Ed propagates on a direct path and Eg reaches the receiver by ground reflection. The vectorial sum of both fields reveals Et , which is detected by the receiver. Magnitude and phase of the reflected signal depend on the traversed path length and on the reflection coefficient ρ of the ground [17]. By assuming ds >> ht and ds >> hr the angle θ gets very small and s1 + s2 ≈ ds . So, the reflected signal experiences approximately the same path loss as the direct signal and ρ ≈ −1 [17]. Using the assumption that the direct signal and the reflected signal experience the same path loss, Et , the electrical field at the receiver, can be written as: Et = Ed + Eg = Ed + Ed ρe−j∆φ ,

(2.7)

with ∆φ being the phase difference between direct and reflected signal. And by applying the assumption that ρ = −1, Equation (2.7) can be simplified to: Et = Ed − Ed e−j∆φ = Ed (1 − e−j∆φ ).

(2.8)

P 2 3 1 xn According to the Taylor series ex = 1 + x1! + x2! + x3! + · · · = ∞ n=0 n! . Consequently, by retaining only the first two terms e−j∆φ can be expressed as 1 − j∆φ and Equation (2.8) can be written as: Et = Ed (j∆φ) (2.9) with ∆φ being the phase difference of the two signals. From Figure 2.1 it can be extracted that ∆φ = 2π(s1 + s2 − ds )/λ, where λ is the wavelength of the traveling signals. Using Equations (2.4) and (2.5) as well as Equation (2.6), this relation can be expressed as [17]: ! r r 2πd (hr + ht )2 (ht − hr )2 ∆φ = 1+ − 1+ (2.10) λ d2 d2 √ As pointed out in [17], using the first two terms of the Taylor series 1 + x = 1 + x2 − P 3 (−1)n (2n)! x2 x4 n + x16 − 128 + ··· = ∞ n=0 (1−2n)(n!)2 (4n ) x , the inner terms of Equation (2.10) can be 8 expanded to Equations (2.11) and (2.12) like: r (hr + ht )2 (hr + ht )2 1+ =1+ (2.11) d2 2d2 and

r 1+

(ht − hr )2 (ht − hr )2 = 1 + d2 2d2

(2.12)

8


Substituting Equations (2.11) and (2.12) into Equation (2.10) yields Equation (2.13) that approximates the phase difference of the two signals: ∆φ ≈

4πhr ht . λd

(2.13)

According to [16], the signal power P is related to the magnitude of the electrical field |E| by the intrinsic impedance of free space η as expressed by Equation (2.14): P = |E 2 |/η.

(2.14)

Thus, by substituting Equations (2.13) and (2.14) in Equation (2.9), reveals Equation (2.15): −Ed2 ( 4πhλdr ht )2 (2.15) PR (d) = η Applying Equation (2.14) in Equation (2.1) gives the magnitude of the electrical field that reaches the receiver along the direct path: Ed2 = η

PT GT GR λ2 . (4π)2 d2

(2.16)

Finally, combining Equations (2.16) and (2.15) reveals Equation (2.17) that approximates the field strength at the receiver for large d according to the two-ray propagation illustrated in Figure 2.1 [16, 17]: PR (d) = PT GT GR

h2t h2r . d4

(2.17)

Comparing Equations (2.1) and (2.17) it becomes obvious that for large distances, the power incident at the receiver decreases at a rate of 40 dB per decade, which is much more than expected from free space propagation. Additionally, for large d, the received signal strength becomes independent from the frequency [16].

2.1.2

Small-scale Fading

Deep and rapid amplitude fluctuations caused by the near surroundings of a receiver are called small-scale fading. As discussed above, a signal from an antenna reaches another antenna over several paths with their associated path lengths and attenuations. As a result, many copies of one signal reach the antenna after different delays, where they superimpose each other destructively or constructively [18]. If the bandwidth Bs of a transmitted signal is smaller than the bandwidth BC over which a wireless channel has a constant gain and a linear phase response, the transmitted signal will experience flat fading [16]. A flat fading signal is a locally coherent signal. The symbol amplitudes of a transmission will change over time due


9

to flat fading, as much as 30 dB [16]. The signal amplitudes in flat fading channels are often assumed to be Rayleigh distributed, or if there is in addition to the multipath propagating signals, a dominant and not fading signal present, the channel is assumed to be Ricean distributed [16]. If Bs > BC one speaks of a frequency selective fading which is more difficult to model [16]. On receiver side an automatic gain control (AGC) can be used to mitigate flat fading effects [17] but more so, diversity techniques help to mitigate small-scale fading effects [19]. Assuming sufficiently separated signals in time, frequency, space, or polarization, they are independent of each other [20]. Hence, if one antenna does not receive a signal along one branch due to small-scale fading, the other uncorrelated antenna will likely still receive the signal along a diversity branch [19]. According to [20], a diversity system is a system that provides two or more similar copies of the same signal. For example, a receiver detects signal f1 (t) and stores it locally. Then the receiver detects signal f2 (t). By combining f1 (t) and f2 (t), that is f (t) = f1 (t) + f2 (t), the signal quality can be improved. Resulting from additional noise present in the wireless channel, signal f1 (t) is a combination of the desired signal s1 (t) and a noise signal n1 (t), that is f1 (t) = s1 (t) + n1 (t), likewise is true for signal f2 (t). So, f (t) = s1 (t) + s2 (t) + n1 (t) + n2 (t) is a combination of the desired signal and the present noise. In a diversity system s1 (t) and s2 (t) are similar to each other and combining s1 (t) and s2 (t) increases the desired signal amplitude. Assuming n1 (t) and n2 (t) to be independent additive white Gaussian noise signals having zero means, they partially cancel out each other. Consequently, the SNR in f (t) may increase due to diversity. More generally, a diversity system may combine n signals fi (t) that have due to the uncorrelated channel gains, different signal amplitudes a as described by Equation (2.18) [20]: n X f (t) = ai fi (t). (2.18) i=1

Combining the outputs of several antennas can increase the detected signal strength at a receiver [21] to improve communication robustness and reliability. There are existing several diversity techniques that can be utilized. The example above describes time diversity as the two signals f1 (t) and f2 (t) reach the detector at different times. Separating receiver antennas in space, achieves independent signal paths between sender and receiver. This technique is called space diversity. In frequency diversity systems, signals experience uncorrelated fading resulting from separate transmit frequencies. And in polarization diversity, signals are received either along the horizontal or the vertical polarization. Polarization diversity has the advantage that it requires no spatial separation of the antennas [19].

10


Combine multiple versions of the same signal fi (t) according to Equation (2.18), is common to the above-mentioned diversity methods. But if the multiple versions fi (t), are not in phase during combination, the resulting signal amplitude could even be lower than the originally transmitted signal amplitudes due to destructive interference. Diversity systems usually take care about this by using some sort of a phase-control technique [20]. There exist mainly four diversity systems that can be combined with diversity techniques. Brennan [20] categorized them into scanning diversity, selection diversity, maximal-ratio combining, and equal-gain diversity. The latter two methods combine several independent sinusoidal signals to increase the signals strength. Their phases need to be aligned to avoid destructive interference before summation, to achieve this. Scanning diversity A switch, controlled by a subsequent logic, connects several antennas to a receiver in a scanning diversity system. Switching of the antennas follows a predefined pattern. If an antenna receives a signal with a received signal strength indicator (RSSI) value above a certain threshold, the current antenna stays connected to the receiver until the signal falls below this threshold. Then the scanning sequence is re-initiated. Selection diversity A selection diversity system connects for any incoming signal, the receiver always to the antenna that has the highest RSSI value. All other signals do not contribute to f (t) as illustrated in Figure 2.2 [20] that sketches a selection diversity system consisting of three antennas that receive fading signals. The antenna with the highest RSSI value can be determined for example by comparing the short term averages of the incoming signals on all antennas. This antenna is then connected to the receiver. The average RSSI value detected at a receiver clearly increases compared to that of a single branch system. But selection diversity is not an optimal diversity system, as it does not use all present diversity branches simultaneously [16]. Maximal-ratio combining A maximal-ratio combining system weighs the signals of each diversity branch according to their individual SNR values and then sums all signals [16]. Figure 2.3 shows the generalized block diagram of an maximal-ratio combining diversity system introduced in [22]. Here, each of the M antennas is connected to the cophasing and sum circuitry through its associated amplifier Gi . It can be shown (for example [20], [22], and [16]) that in a maximal-ratio combining system, the power incident at a receiver always has


11

Matching Network

f1(t)

Matching Network

f2(t)

Matching Network

f3(t)

f(t)

Means of Determining Maximum SNR

Figure 2.2. Schematic of a selection diversity system where the receiver is always connected to the antenna with the highest RSSI value. the maximal achievable SNR at all times and consequently has a clear advantage over selection diversity systems. 1

G1

G2

M

cophase and sum

2

output

GM variable gain Gi

Figure 2.3. Schematic of a maximal-ratio combining diversity system. Phase alignment can be achieved by a cophase and sum circuit as illustrated for example in Figure 2.3. The cophase and sum circuit can be realized for example, as a RAKE receiver that has the ability to demodulate each incoming signal independently. By correlating the incoming signals to delayed replica versions, can detect and reduce the relative phase offsets originating from the multipath propagation [23]. To successfully process the signals RAKE receiver usually utilize signals that are coded according to code division multiple access techniques that apply mapping of each data bit to larger sequences utilizing the direct sequence spread spectrum technique.

12


Figure 2.4 illustrates a generic RAKE receiver with maximal ratio combining for two input channels [23]. The incoming signal of each antenna is processed in two channels to correlate the I- and Q-codes depicted as ci in Figure 2.4, where i resembles the I- and Q-parts of the incoming signals. After correlation, the signals are sampled at frequency 1/T , depicted by the switch in Figure 2.4. Then, each signal is amplified by the corresponding gain factor hij , where j resembles the antenna path. Finally, the output is achieved by summing all processed signals [23]. Resulting from its rather complex structure, a RAKE receiver usually has energy requirements in the mW area. h11

Σ

c2

Σ

2

c1

Σ

c2

1/T

1/T

1/T

h21

h12

output Σ

c1 Σ

1

h22

1/T

Figure 2.4. Schematic of a receiver with maximal ratio combining diversity for two input channels.

Equal-gain diversity An equal-gain diversity system is similar to the maximal-ratio combining system depicted in Figure 2.3 but it does not include the variable gain blocks. So, the signals from the diversity branches are first co-phased and then summed at equal gain. An equal-gain diversity system provides a performance that is only marginally inferior to maximal-ratio combining systems and superior to selection diversity [16]. Discussion on diversity systems Equal-gain and maximal-ratio-combining systems can successfully increase the signal strength at their output only if the noise signals are uncorrelated and the data signals are correlated, as already pointed out by [24]. If there is a highly correlated noise signal present at the antennas, equal-gain-combining and maximal-ratio-combining systems degrade in their performance and selection diversity can achieve a higher signal to noise ratio of the output signals then the other techniques. Also, for practical reasons,

2.2. Satellite-based Positioning

13

selection diversity systems often measure the received signal strength indicator and not the signal to noise ratio, which also degrades their efficiency in high noise areas or in case of co-channel interferences. To mitigate this effect, some selection diversity systems additionally measure the present bit error rate and if it falls below a certain threshold, they switch to another antenna, also it has a lower received signal strength indicator [25].

2.2

Satellite-based Positioning

This Section introduces the basic principles of satellite-based positioning systems. The basic algorithms required to achieve a differential positioning information as implemented later in this work, are provided. A more detailed description of satellite-based positioning principles and applications can be found for example in [26–28]. The purpose of satellite-based positioning is to determine the position of an observing site with the help of signals transmitted from satellites that have known positions [27]. To achieve this, the GNSS (global navigation satellite system) consists of the space segment that is the constellation of the satellites in orbit, a control segment on earth consisting of several control stations that monitor the satellites to provide highly accurate time and orbit information. And finally, the user segment that consists of the receiving equipment. There are currently four global navigation satellite systems available, GPS (global positioning system), GLONASS (global navigation satellite system), Galileo and Beidou. The satellites are in an orbit around 20 000 km – 25 000 km above earth [29]. Thus, they are not in geostationary orbits and travel with approximately 4 km s−1 . The relative motion between satellite and user causes a Doppler shift fD of the satellite signal frequencies in the range from fT − fD to fT + fD , where fT is the transmitted signal frequency [16]. According to [30] and [26], the frequency fR detected at a receiver can be expressed as: vlos (2.19) fR = fT 1 + c where vlos = vr · a is the line-of-sight velocity, vr is the line-of-sight relative velocity vector, a is the line-of-sight unit vector and c is the speed of light. Reordering Equation (2.19) yields an expression for vlos [30]: fR vlos = c −1 . (2.20) fT Substituting the Doppler shift fD = fR − fT into Equation (2.20) reveals [30]: fD . (2.21) vlos = c fT

14


Finally, Equation (2.21) can be expressed in terms of the wavelength of the transmitted signal which is λT = c/fT [30]: vlos = λT fd .

(2.22)

According to Equation (2.22), counting the Doppler cycles over a certain short period of time (in the ms range), scaling the sum by the wavelength and dividing it by the integration time, reveals the line-of-sight velocity [30]. Continuously accumulating the Doppler counts achieves integration of vlos . The integral of the velocity is equal to the change of the receiver position relative to its position at the beginning of the integration. But usually, the integration time does not equal 1/fD or an multiple of it and a fraction of a Doppler count remains at the end of the integration. Including this fraction of a Doppler count in the measurement results, yields the integrated Doppler or carrier phase measurement [30].

2.2.1

GPS and GLONASS Signals

GPS signals are transmitted on frequencies L1 = 1525.42 MHz, L2 = 1227.60 MHz, and L5 = 1176.45 MHz. As all available GPS satellites transmit on the same frequencies, the signals are based on code division multiple access (CDMA) and each satellite has its own ID. The GLONASS system transmits on L1 = 1598.0625 MHz – 1609.3125 MHz and L2 = 1242.9375 MHz – 1251.6875 MHz using frequency division multiple access (FDMA) [29]. The basic principles of position measurements are the same for GPS and GLONASS systems, so it will be shown here for GPS signals only. The receivers and antennas utilized in this work support GPS and GLONASS L1 signals and frequencies, so the other systems will not be further discussed here. All GPS signals are direct sequence spread spectrum (DSSS) transmissions based on pseudo random noise (PRN) codes. Each satellite broadcasts on two unique PRN sequences, one as C/A-code (coarse / acquisition) on L1, the other P(Y) (precise) code on L1 and L2. The C/A-code is also open for civilian use, the P(Y) code is encrypted and only accessible to authorized users like the military [29]. GPS receiver carrier and code correlation A C/A-code consists of a 1023 bits long sequence and each bit is called chip. The C/A-code is sent at 1.023 MHz, so a chip duration tc equals 0.98 µs. Using s = c · tc , with c = 299 792 458 m s−1 being the speed of light in vacuum [31], the range s that a radio wave travels during tc can be calculated to approximately 294 m. The P-code’s PRN sequence is transmitted with 10.23 MHz, consequently by using the same relation of s = c · tc , the range s that a radio wave travels during a P(Y) code chip duration tc equals approximately 30 m.


15

The C/A-code is further binary-phase-shift-keying modulated to carry the navigation messages that contain satellite health data, ephemeris, clock bias and almanac data of all GPS satellites in orbit. Navigation messages are transmitted at 50 bit per second, each bit consisting of 20 chips [26]. Usually, GPS receivers apply correlation and tracking techniques to determine the transmitted satellite code and carrier frequency and phase [26]. Maintaining code and carrier phase correlation and tracking with high accuracy and under changing signal conditions involves several sophisticated methods and algorithms such as techniques to mitigate multipath errors [32] or various robust carrier tracking solutions that continue to work also in challenging conditions [33]. A detailed survey of these techniques is out of scope of this work that provides the basic principles of GPS receivers. A more comprehensive discussion can be found for example in [26, 28, 30]. Figure 2.5 visualizes the simplified block diagram of a generic digital GPS receiver from [26] that is basically a single-conversion super-heterodyne receiver [34, 35]. There exist also other architectures for example, based on direct conversion of the satellite signal to decrease the power requirements, although these receivers are usually not as sensitive and selective as the super-heterodyne concepts [36]. AGC LNA

analog IF

ADC

N digital IF

2 1 receiver channel

Frequency synthesizer Local Oscillator

Figure 2.5. Simplified block diagram of a generic digital GPS receiver. According to the simplified generic GPS receiver in Figure 2.5, the satellite signals are received at the antenna and amplified by a low noise preamplifier. After preamplification, the signals are down converted to an intermediate frequency that is large enough, for example 5 MHz, to support the chipping frequency of 1.023 MHz. Down conversion is achieved by mixing the satellite signals with a frequency received from a local oscillator. The down-converted signal is digitized by the A/D converter and additionally conditioned by the automatic gain control unit. The satellite signals are still buried in thermal noise, at this point of the signal processing chain [26]. Figure 2.6 from [30] illustrates exemplary a receiver channel. Usually, there ex-

16

2. Radio Wave Propagation and Global Navigation Satellite System accumulate andLdump accumulate andLdump

ILsample processing

accumulate andLdump carrier NCO

from ADC

late

prompt early

shift-register 90°

delta-range accumulator

IE

IP

IL

PRNLcode generator

code NCO

timeLofLtransmission register

QLsample processing

Figure 2.6. Block diagram of a generic digital receiver channel. ists one channel for each incoming satellite signal. Inside the channel, the carrier numerically controlled oscillator (carrier NCO) mixes its output to the in-phase (I) and quadrature phase (Q) of the signal to determine its carrier phase and the Doppler shift. A second numerically controlled oscillator, the code NCO, generates a replica of the incoming signal with the help of the PRN code generator. A shift register shifts the signal replica two times, to generate three replica versions, an early (E), a prompt (P) and a late (L). The digitized satellite signal is then correlated to the three replica versions E, P, and L. The code tracking loop gives feedback to the code NCO [26,28,30]. The carrier tracking loop provides feedback to the carrier NCO with the objective to keep the phase error between preprocessed satellite signal and prompt replica at zero [26]. At this point, the output from the carrier NCO is accumulated to achieve delta range and accumulated delta range data [30] as discussed above. Figure 2.7 depicts the simplified generic block diagram of the carrier tracking loop [26]. The prompt replica is processed by the carrier tracking loop and once it is locked, its output can be provided to the code tracking loop to improve its performance. The carrier tracking loop as illustrated in Figure 2.7 mainly consists of the carrier loop discriminator and the carrier loop filter. The type of the tracking loop is set by the carrier loop discriminator. In traditional GPS receiver designs, the tracking loop is usually a phase lock loop or a frequency lock loop [26]. Lopez-Salcedo et al. [28] list several other loop types that improve tracking stability if applied under certain conditions. Although long integration times would improve the signal-to-noise ratio, the integration time may normally not exceed 20 ms which is the period of the navigation


17 carrier aiding scale factor

IP QP

accumulate and dump accumulate and dump

IPS QPS

carrier loop discriminator

carrier loop filter

to carrier NCO carrier NCO bias

Figure 2.7. Block diagram of a generic carrier tracking loop. messages. Furthermore, in case of high dynamic signal conditions, the integration time should be short [26]. In code tracking, the phases of the early and the late replicas are typically set to be separated by 1 chip, also narrower correlator spacings were investigated to improve multipath mitigation [32, 37]. The phase of the prompt replica is in the middle [26]. Thus, the correlation of the prompt replica with the incoming signal is at its maximum in case their phases are aligned. Then, the phase of the early replica is half a chip period too early and the phase of the late replica is half a chip duration too late with respect to the incoming code phase. Consequently, the correlators of the early and the late code phases produce only about half the correlation output as the prompt replica correlator [26]. Figure 2.8 from [26] illustrates the code correlation signals for idealized signal shapes for the incoming signal and the correlator output for reasons of simplicity. The shape of a real signal is much less rectangular due to noise. The correlator output of the prompt replica is largest, only when the phases of the incoming signal and of the replica signal are aligned. Braasch and Dierendonck present in [30] the tracking accuracy that can be achieved by the above introduced techniques in the range of 0.001 chips for high quality signals that have a carrier-to-noise ratio of at least 50 dB-Hz to 0.1 chips for carrier-to-noise ratios lower than 30 dB-Hz. By using the above introduced relation between chip duration and the distance a radio wave travels during this time, the code tracking error ranges from 0.3 – 30 m depending on the carrier-to-noise ratio that is typically higher than 35 dB-Hz for GPS [30]. For carrier-phase tracking, Braasch and Dierendonck [30] calculate a tracking error in the range from 3° – 0.5° for carrier-to-noise ratios from 40 – 55 dBm-Hz. Accordingly, by applying the same time to range relations as above, the carrier-phase tracking error ranges from 0.3 mm to 1.6 mm.

2.2.2

Pseudorange Measurements

The main goal of a satellite positioning system is to determine the position of a receiver somewhere on or above the earth. Satellite signals include the instant of time T S when

18

2. Radio Wave Propagation and Global Navigation Satellite System incoming signal

replica signals early

prompt

late normalized correlator output 1

1 P

3/4 1/4 0

L

1

P

3/4

3/4 E

E

L

1/4 0

1/4 0 (a)

P

E

L

(b)

(c)

Figure 2.8. Illustration of the code correlation assuming idealized input signal shapes and correlator output for reasons of simplicity. (a) replica code 1/4 chip early, (b) replica code aligned, (c) replica code 1/4 chip late. they were transmitted to achieve this goal. A receiver samples the satellite signals based on its time frame, whereas the satellite sends signals based on his time frame. If for example, T is the instant of time when the receiver took a measurement, then the receiver can calculate its distance P S from the satellite by P S = c(T − T S ), with c = 299 792 458 m s−1 being the speed of light in vacuum [31]. The basic principles of satellite-based range measurements are described by this simplified model. More detailed, the coordinates xu , yu , zu express the unknown position of a user (a receiver) as depicted in Figure 2.9 [26]. Vector u represents the user position relative to the coordinate system and needs to be determined. The position of the satellite at coordinates xs , ys , zs , is considered to be known and is represented by vector s. This vector can be calculated from the ephemeris data broadcast by the satellite [26]. Vector r, the range between user and satellite is unknown and the goal is to measure it as precisely as possible. Combining r, s, and u yields Equation: d(s, u) = ||s−u|| [26], which is the Euclidean distance between s and u. In the discussion above, it was assumed that the satellite transmits its signal exactly at T S and the receiver samples the signal exactly at T . Figure 2.10 depicts the determination of the instant of time of the transmission more accurately [31]. The ex-


19 satellite

s

r user u

earth

Figure 2.9. Representation of the position vectors. ample adopted from [26], illustrates a specific code sequence transmitted by a satellite at T S . The same code sequence is received at the receiver at the instant of time T . Consequently, the transmission time duration is T − T S . Satellite generated code

Receiver generated replica code T - TS TS= t + τ S

T=t+τ

Figure 2.10. Determination of GNSS code transmission time T − T S . But due to not perfectly synchronized clocks of satellite and receiver, T includes additionally a time offset τ originating from the clock offset of the receiver. By assuming t to be the true system time, T can be expressed as T = t + τ . Likewise, T S includes τ S that represents the clock offset of the satellite from system time, that is T S = tS + τ S , where tS is the system instant of time when the satellite started its transmission. Applying this, the distance P S from satellite to receiver can be expressed as Equation (2.23) [31]: P S (t) = c((t + τ ) − (tS + τ S )) = (t − tS )c + c(τ − τ S ) = ρS (t, tS ) + c(τ − τ S ) (2.23) where ρS (t, tS ) is the geometric distance from satellite to receiver, simplified, as it assumes constant speed of light in the atmosphere and does not consider relativistic effects amongst others. The range P S (t) calculated according to Equation (2.23) that includes the clock errors is called pseudorange. The satellite clocks are highly accurate atomic clocks

20


(usually a satellite has more than one clock) and the control segment determines and uploads constantly their offsets and drifts from system time to the satellites that themselves propagated them in the navigation messages. Therefore, τ S may be assumed to be known for coarse position calculations but not for high precision position determination as required in Chapter 5. But as very precise clocks are expensive, common satellite receivers use more inexpensive clocks and time offset τ of the receiver should be considered in any case [26]. Calculating the distances to three satellites would be sufficient to determine the three unknowns latitude, longitude, and height (xu , yu , zu ). By determining the pseudoranges to four satellites simultaneously, it is additionally possible to resolve the unknown clock offset τ . Then, Equation (2.23) can be expanded to [26]: q (2.24) P j (t) = (xj − xu )2 + (yj − yu )2 + (zj − zu )2 + cτ for j >= 4 and xj , yj and zj being the satellite positions of the jth satellite. This set of equations can be solved either by using i) closed-formed solutions, ii) iterative techniques, or iii) Kalman filtering [26].

2.2.3

Pseudorange Measurement Errors

As shown in previous Section 2.2.2, pseudorange P S includes errors that are partly caused by the ionosphere that influences the electromagnetic waves as they travel from the satellites to the earth and introduces ionospheric propagation delays. As the delays are frequency dependent they can be determined and taken care of, if the receiver detects two frequencies (L1 and L2) simultaneously transmitted from the satellites [29]. A further error source that affects the range calculation are tropospheric delays that are a function of temperature, pressure and relative humidity and are not frequency dependent. Using two frequencies does not help to determine the delays but it is possible to model the troposphere and to compensate for most of the delay [29]. Including the signal delays in the pseudorange calculations and satellite clock bias τ j yields a more accurate range Equation (2.25) [31]: P j (t) = ρj (t, tj ) + c(τ − τ j ) + Zj + Ij

(2.25)

with Zj being the message delay due to the troposphere and Ij being the delay due to the ionosphere. Another error source is the receiver hardware itself that induces thermal noise jitter on the signals during tracking of code or phase [26]. Finally, multipath propagation occurring from reflected signals from surfaces in the vicinity of the receiver antenna, superpose with the line-of-sight signal in the antenna. These additional signals influence the correlation function. Although, there exist several techniques to mitigate multipath


21

effects [30,32,37–39] these errors are cause of one of the most important errors for high precision range measurements [37, 39]. Incorporating measurement errors p related to receiver hardware and multipath effects Equation (2.25) expands to [40]: P j = ρj (t, tj ) + c(τ − τ j ) + Zj + Ij + p .

(2.26)

Table 2.1 from [29] summarizes error sources and ranges in meters. It can be seen that delays introduced by the atmosphere are dominant. There are also other effects that influence the position calculation, like relativistic effects as well as the so called Sagnac effect that takes care about the earth rotation during the time the radio signals travel from the satellite to the receiver. It is possible to correct for both effects on receiver side, so they will not be further looked at. A more comprehensive and detailed explanation of effects that influence a GNSS signal can be found for example in [26]. Table 2.1. Error sources and amounts of a single receiver pseudorange measurement. source

amount [m]

satellite clocks orbit errors ionospheric delays tropospheric delays receiver noise multipath

± ± ± ± ± ±

2 2.5 5 0.5 0.3 1

Furthermore, the geometric formation of the tracked satellites with respect to the receiver influences directly the position and time calculations. An unfavorable arrangement of satellites can dilute the precision, therefore this metric is called dilution of precision DOP [29]. Figure 2.11 depicts the dilution of precision schematically [29]. In Figure 2.11 (a) it is obvious that it is more difficult to obtain accurately the point of intersection of the range measurements compared to the case shown in Figure 2.11 (b). Due to dilution of precision, determination of zu is usually less accurate than determination of xu and yu .

2.2.4

Differential GNSS

Previous Section 2.2.3 demonstrated that a single receiver system can achieve an accuracy in the meter-range. If a higher accuracy is required a differential global navigation satellite system (DGNSS) can be implemented. DGNSS are either code- or on phaserange measurement based. The accuracy of code-range based differential techniques

22


satellite A

satellite B satellite C

satellite A

satellite D

ranges to satellites A, B, C and D

satellite B satellite C

satellite D

ranges to satellites A, B and C range to satellite D

(a)

(b)

Figure 2.11. Dilution of precision depending on the geometric appearance of the satellites (a) high dilution of precision (b) improved dilution of precision. will improve compared to that of single receiver accuracy but carrier-phase based differential measurements usually achieve higher accuracies. Generally, differential GNSS measurements determine the position of a rover station related to a base station thats position is accurately known. From the discussions in Section 2.2.1 it is known that the receiver mixes the satellite signal with frequency fG (t) and phase ϕG (t) to a locally generated signal with frequency fR (t) and phase ϕR (t) [31]. The phase φ(t) that the receivers detects is equal to the difference of the two signal phases ϕR (t) and ϕG (t) [31, 41]: φ(t) = ϕR (t) − ϕG (t).

(2.27)

Calculating φ(t) generates knowledge about the last centimeters of the distance from satellite to receiver, but the complete range is still unknown. The remaining unknown range is equal to N phase-cycles of the satellite signal that fit into the line-of-sight distance between satellite and receiver. Accordingly, φ(t) can be expressed as [31]: φ(t) + N = ϕR (t) − ϕG (t)

(2.28)

Letting φS (t) be the observed phase at the instant of time T when a receiver detected a signal that was transmitted by satellite S, Equation (2.28) can be written as [31]: φS (T ) = ϕ(T ) − ϕS (T ) − N S .

(2.29)

where ϕ(T ) resembles the phase of the replica signal and ϕS (T ) the phase of the incoming satellite signal both sampled at time T . Phase can also be expressed as the angle of rotation and therefore ϕ(t) is a measure of the angle at instant of time t. If ϕ0 is the angle at time t = 0 and ϕ(t) is the angle at instant of time t, the difference between the angles resembles a certain time like, T (t) = k(ϕ(t) − ϕ0 ) where k is a calibration constant to convert phase cycles


23

into seconds [31]. By assuming a constant phase change rate, the calibration constant k equals the angular velocity which is proportional to the frequency and T (t) can be expressed as [31]: ϕ(t) − ϕ0 (2.30) T (t) = f0 with f0 = 1/k being the frequency at time T (0). Reordering Equation (2.30), yields ϕ(T ) = f0 T +ϕ0 and substituting this relation for ϕ(T ) into Equation (2.29) yields [31]: φS (T ) = f0 T + ϕ0 − f0 T S − ϕS0 − N S

= f0 (T − T S ) + ϕ0 − ϕS0 − N S .

(2.31)

An expression for the carrier phase-range LjA (TA ) in meters is achieved by multiplication of Equation (2.31) by λ0 [31]: LjA (TA ) = c(TA − T j ) + λ0 (ϕ0A − ϕj0 − NAj )

(2.32)

with the subscripts A, B, C, etc. indicating specific receiver quantities and the superscripts j, k, l, etc. indicating satellite specific quantities, hereafter. Observing Equation (2.32) reveals that the first term is the pseudorange from satellite j to receiver A, reported in Equation (2.23). According to the discussion in Section 2.2.3, including the clock biases τA and τ j of receiver A and satellite j, the atmospheric error terms ZAj and IAj and the receiver related errors p , Equation (2.32) can be expanded to [31]: LjA (TA ) = ρjA + c(τA − τ j ) + λ0 (ϕ0A − ϕj0 − NAj ) + ZAj − IAj + p

(2.33)

Please note, that the ionospheric delay has a negative sign in Equation (2.33) and a positive sign in Equation (2.26) which is caused by the ionosphere, where information travels with group velocity and waves propagate with phase velocity [26, 31]. There are several errors included in Equation (2.33) that need to be taken care about before a precise positioning can be achieved. Applying differencing techniques can mitigate several of these effects as will be demonstrated in the following Sections 2.2.4 and 2.2.4. Single differencing Figure 2.12 illustrates the geometry of single differencing. Single differencing is achieved by calculating ∆LjAB that is the difference between the distance from receiver A to satellite j, and the distance from receiver B to satellite j. The distance LjA from receiver A to satellite j is given in Equation 2.33. Substituting subscript A by B in Equation (2.33) yields the distance LjB from receiver B to satellite j. Equation (2.34) expresses the single difference ∆LjAB [31]: ∆LjAB = (ρjA + c(τA − τ j ) + λ0 (ϕ0A − ϕj0 − NAj ) + ZAj − IAj + jA )

− (ρjB + c(τB − τ j ) + λ0 (ϕ0B − ϕj0 − NBj ) + ZBj − IBj + jB ).

(2.34)

24


Expanding Equation (2.34) yields: j j ∆LjAB = ∆ρjAB + c∆τAB + ∆ZAB − ∆IAB + ∆jAB

+ λ0 (ϕ0A − ϕj0 − NAj ) − λ0 (ϕ0B − ϕj0 − NBj )

(2.35)

Observing Equation (2.35) reveals, that the satellite clock bias τ j is removed by calculating the range differences. Additionally, the initial carrier phase ϕj0 of the satellite is eliminated. Furthermore, in case of short baselines, for example if the distance from receiver A to receiver B is less than 10 km, the ionospheric and tropospheric effects are nearly the same. Consequently, ZAj = ZBj and IAj = IBj , which further simplifies Equation (2.34) to [40, 42]: ∆LjAB = ∆ρjAB + c∆τAB + ∆jAB + λ0 (ϕ0A − NAj ) − λ0 (ϕ0B − NBj ).

(2.36)

Single differencing can eliminate several errors, but not the receiver clock biases ∆τAB . This can be achieved by applying double differencing as demonstrated in the following.

satellite j

ρA j

ρB j receiver B unknown position

receiver A known position baseline

Figure 2.12. Geometry of single differencing with known position of receiver A and unknown position of receiver B. Both receiver measure their distance ρ to the same satellite j.

Double differencing Figure 2.13 illustrates the geometry of double differencing. Two receivers A and B track simultaneously two satellites j and k. The single differences ∆LjAB and ∆LkAB can be calculated as demonstrated in Section 2.2.4, for example by applying Equation (2.36). The double difference ∇∆Ljk AB is the difference between the single differences j ∆LAB and ∆LkAB . Please note, the upside-down triangular symbol ∇ is a mnemonic device to emphasizes that the calculated differences are between two points in the sky.


25 satellite k

satellite j

ρB j

ρA k ρB k

ρA j

receiver B unknown position

receiver A known position baseline

Figure 2.13. Geometry of double differencing with known position of receiver A and unknown position of receiver B. Both receiver measure their distance ρ to satellites j and k. Then, the double difference can be expressed as [42]: j k ∇∆Ljk AB = ∆LAB − ∆LAB

= ∆ρjAB + c∆τAB + ∆jAB − (∆ρkAB + c∆τAB + ∆kAB )

+ (λ0 (ϕ0A − NAj ) − λ0 (ϕ0B − NBj )) − (λ0 (ϕ0A − NAk ) − λ0 (ϕ0B − NBk ))

(2.37)

Observing Equation (2.37) reveals that double differencing eliminates the receiver clock errors τAB as well as the initial phase offsets ϕ0A and ϕ0B which yields following simplified expression of Equation (2.38): jk jk j j k k ∇∆Ljk AB = ∇∆ρAB + ∇∆AB + λ0 (NB − NA ) + λ0 (NA − NB )

(2.38)

Equation (2.38) reveals that double differencing transforms the phase ambiguity into an integer value, which is a clear advantage of this method. Following the Least-squares AMBiguity Decorrelation Adjustment (LAMBDA) method presented by Teunissen in [43], the problem of resolving the phase ambiguity can be basically expressed by solving a linear system of observation equations [44]: y = Aa + Bb + e

(2.39)

where y represents the observed minus the calculated carrier phases, vector a consists of the unknown carrier phase ambiguities, b resembles the unknown parameters of the baseline coordinates and e is a noise vector. The matrices A and B are the design matrices that relate the observations to the unknowns [44]. For example, the method of least-squares can solve Equation (2.39) in case it is overdetermined as expressed by Equation (2.40) [44]): min ||y − Aa − Bb||2 , a ∈ Z, b ∈ R a,b

(2.40)

26


Three steps are performed to solve Equation (2.40) [44]. First, the constrain of a ∈ Z is replaced by a ∈ R, which can be achieved by applying the code-range measurements in Equation (2.40). This yields the float solution a ˆ. Then, the float solution a ˆ and its variance matrix Qaˆ are used to minimize the expression given in Equation (2.41) [44]: a ¯ = min(ˆ a − a)T Qaˆ (ˆ a − a), a ∈ Z a

(2.41)

Finally, float estimate ˆb is improved by the difference of (ˆ a−a ¯) to achieve a fixed estimate of b. The preceding paragraph provided the very basics to resolve the integer ambiguities to achieve an accurate position measurement. Further details can be found for example in [26, 40, 42–44]. Once the integer ambiguity is fixed, the carrier phase range measurements result in precise positioning results in the mm-range.

Chapter 3 Wireless Sensor Nodes with Wake-up Receiver This chapter introduces the hardware developed and utilized during this work. The first part of the chapter introduces and analyzes wireless sensor nodes with wake-up receivers from a more general point of view. Then, a wireless GNSS sensor node with wake-up receiver is introduced followed by a wireless sensor node that features antenna diversity in the wake-up path.

3.1

Wake-up Receiver

Low-power wake-up receivers like those introduced in [11, 13, 45–48] consist of an envelope detector and a correlator as sketched in Figure 3.1 that shows schematically a wireless sensor node including a wake-up receiver. The envelope detector demodulates the high-frequency (HF) carrier signal to achieve a low-frequency (LF) wake-up signal as sketched in Figure 3.2 [13] that depicts an On-Off-Keying modulated wake-up signal. The correlator analyzes the LF signal, to verify the validity of a wake-up message. In that case, the main microcontroller of the sensor node is woken up by an interrupt and, depending on the embedded software, a sensor reading might be initiated or the antenna is connected to the main radio to establish further communications. A matching network might be necessary to match the impedances of antenna and wake-up receiver. Blanck et al. presents in [9] an overview of current low-power transceivers. They introduce highly integrated concepts that require 0.1 µW [49, 50] as well as several solutions between 10 and 1000 µW [9]. Only a few receivers are in the range of 1 to 10 µW. Common to the wake-up receivers presented in [9] that have power requirements below 10 µW, is the use of on-off-keying modulated wake-up messages. By using onoff-keying modulation, the wake-up receiver hardware design can be simple and energy efficient. For example, the envelope detector is only composed of diodes and capacitors 27

28

3. Wireless Sensor Nodes with Wake-up Receiver

and a comparator can be used as correlator [11, 45–48]. Envelope Detector

Matching Network Antenna Switch

Controls Communication Radio

Correlator

Interrupt

Microcontroller

Figure 3.1. Schematic of wireless sensor node including a wake-up receiver. Low Frequency Period

High Frequency

Figure 3.2. The low-frequency wake-up message (red) is modulated on the highfrequency carrier signal by On Off Keying. The wireless nodes used in this work are based on the sensor node introduced in [13, 51]. Figure 3.3 shows a photo of the implemented node. The microcontroller utilized on the boards is a 32 bit EFM32G222F128 manufactured by SiliconLabs running at 14 MHz. It provides several low power states to reduce energy consumption. In run mode it needs around 2.5 mA and 0.9 µA in Deep Sleep Mode. Including all peripherals, the sensor node requires around 4.0 mA in run mode. The communication radio is a CC1101 from Texas Instruments. It has a current consumption of 34.2 mA when transmitting at +12 dBm output power at 868 MHz and around 16.4 mA when transmitting at 0 dBm. Its sensitivity is approximately between -95 to -104 dBm, depending on the data rate. The 125 kHz LF receiver (AS3932) from austriamicrosystems has a current consumption of around 3 µA in listening mode. It correlates the incoming signal to a pre-configured address and creates an interrupt if send and stored addresses match. In combination with matching network and an envelope detector, the wake-up receiver has a sensitivity around -51 dBm [13,51]. Figure 3.4 shows the wake-up pattern required to activate the AS3932 chip consisting of carrier burst, preamble and optional address and pattern. The CC1101 transceiver generates the wake-up pattern by modulating the pattern on the 868 MHz signal by means of an on-off-keying modulation as presented in Section 3.1. In addition to some common sensors, the node is equipped with a high precision realtime clock (PCF2129T) and a MicroSD card that can be switch off by the microcontroller. A monopole antenna with a gain of approximately 1.5 dBi is used. Friis transmission equation as introduced in Section 2.1.1 can be used to calculate the freespace transmission distances for wake-up and main radio. But knowing from the

3.1. Wake-up Receiver

29

Figure 3.3. Photo of sensor node with wake-up receiver.

Figure 3.4. Wake-up pattern of the AS3932 LF wake-up receiver. Pattern and data are optional. discussion in Section 2.1.2 that the wireless transmission range is additionally affected by multi-path propagation effects, the communication range of the main radio can be estimated to be well above 300 m when sending at + 0 dBm output power. Transmitting at +12 dBm, the wake-up range can be calculated to be around 45 m. Experiments conducted in [51] using a similar sensor node compared to the one introduced here, implies the wake-up range to be around 45 m even for sending at 10 dBm output power.

3.1.1

False Positive and False Negative Wake-ups

Due to their low power consumption and the fact that they listen always on incoming signals, wake-up receivers are prone to false positive and false negative wake-ups. In the case of false positive wake-ups, a receiver detects a valid signal although the wakeup message was not dedicated to it. False negative wakeups occur when a wake-up receiver stays asleep although a wake-up message was sent to it. Both kinds of false wake-ups can result from interferences on the wireless channel and can possibly lead

30


to an increased power consumption and communication delays. Experimentally, the occurrence of false negative wake-ups, aka missing wake-ups, can be measured for example by counting how many valid wake-ups a receiver detected out of the number of sent valid wake-up messages. The false positive wake-up rate can be experimentally measured by counting how often a wake-up receiver detects a valid wake-up message although the message does not contain a valid address. To reduce the occurrence of false positive and false negative wake-ups some wake-up receivers use active or passive input filter [52], which includes a correlator unit that analyses the received wake-up messages and only creates a wake-up signal in case the addresses match [11, 50, 52], or make use of manchester or similarly encoded wake-up signals [13, 46, 51].

3.1.2

Experimental Analysis

This section presents the experimentally measured wake-up message and furthermore, the occurrence of false positive and false negative wake-ups are investigated in accordance to Section 3.1.1. Wake-up Message Figure 3.5 shows the wake-up message used throughout this work captured at the output of the envelope detector. The wake-up message was Manchester encoded to improve stability and to reduce the false wake-up rate as introduced in Section 3.1.1. In this case, a binary one is Manchester encoded as transition from high to low and a binary zero results from a low to high transition. Consequently, one bit Manchester encoded requires two bit sent. Figure 3.5 clearly shows that the length of the measured wake-up message corresponds very well to the theoretical length of the wake-up message calculated by using the numbers provided by the datasheets as given in Section 4.2.2. False Positive and False Negative Wake-ups To evaluate the occurrence of false negative wake-ups as shown in Section 3.1.1, a laboratory experiment was conducted, consisting of a sender and a receiver connected by cables. An attenuator was placed in the connection between sender and receiver to reduce the incoming signal stepwise from 0 dBm to -60 dBm. During the experiment, the sender sent each possible address (0x00 to 0xFF) 100 times. After sending 100 addresses the sender signaled the receiver via a separate connection. After receiving this signal the receiver incremented its address stepwise from 0x00 to 0xFF. Figure 3.6 shows the averaged false wake-up, aka missing wake-up,

3.1. Wake-up Receiver

31

Figure 3.5. Manchester coded wake-up signal consisting of carrier burst, preamble and address pattern. rate over input signal strength. The experiment illustrates that the wake-up receiver had no false negative wake-ups until the signal strength reached its sensitivity limit at around -50 dBm. Then, the false negative wake-up rate increased quickly to 100 % for signals sent below -52 dBm.

Figure 3.6. Experimentally measured false negative wake-up rate over input signal strength in dBm, nodes connected by cables. To analyze the occurrence of false positive wake-ups an experiment was conducted similar to the one described above. The same test setup was used and the sender was configured to send every address from 0x00 to 0xFF 10 times, while the receiver kept its address. Only each time the sender sent 0xFF the receiver incremented its address by one. Since the receiver’s initial address was 0x00, each address could be cross-checked with all other possible addresses during this test. The receiver just woke up 10 times,

32


exactly what would be expected if no false positive wakeups occur. Throughout the test, the received signal strength was set to -25 dBm.

3.2

Wireless GNSS Node

Figure 3.7 schematically shows the GNSS (global navigation satellite system) receiver node including its hardware blocks. The node basically consists of the same hardware components as the sensor nodes introduced above but is additionally equipped with a GNSS receiver. The NovAtel OEM615 GNSS receiver module supports the measurement of GPS and GLONASS L1 signals (see Section 2.2) with sampling rates up to 20 Hz [53]. The satellite antenna is a GPS-701-GG antenna [53] from NovAtel that also supports the L1 GPS and GLONASS frequencies (see Section 2.2.1. In active mode, the GNSS receiver needs around 400 mA. To reduce the average power consumption, the module is activated only periodically and can be switched off completely during times when it is inactive. Figure 3.8 shows a photo of the GNSS (Global Navigation Satellite System) receiver node. The NovAtel OEM615 GNSS receiver module can be connected to the board by using a simple 20 pin connector. A UART interface is used to configure the GNSS module and to read out the GNSS messages periodically. A GNSS range message

Figure 3.7. Schematic of the GNSS wireless sensor node. generally consists of four fields as depicted in Figure 3.9. The first 28 Bytes are the header that includes information like message type, time, length, etc. The next 4 Bytes large field gives the number of valid observations, the third 24 Bytes large field includes all valid observations and the last field is a 32 Bit CRC [53]. Altogether, a range message varies in its size from 50 Bytes in the case of one valid observation to 516 Bytes for 20 valid observation. At a sampling frequency of 4 Hz, this sums up to 200 Bytes per second for one valid observation and 2064 Bytes per second for 20 valid observations.

3.2. Wireless GNSS Node

33

Figure 3.8. Photo of the GNSS sensor node. At a sampling frequency of 20 Hz, the observation of one satellite results in 1000 Bytes per second, and the observation of 20 valid satellites results in 10320 Bytes per second. A baud rate of 230 kbps (equal to 23 kByte per second for an 8N1 transmission) is used to transmit data from the NovAtel board to the sensor node. The GNSS data is first stored in an internal 10 kByte memory buffer and then transferred periodically onto a 4 GB MicroSD Card. To increase the speed the MCU is running at 28 MHz during this process. By assuming a sampling rate of 4 Hz and 15 valid observations per message (396 Byte), a 4 Hz logging frequency (1584 Bytes per second) for a duration of 30 minutes (1800 s), adds up to 2851.2 kByte that have to be transferred wirelessly. At 20 Hz logging, around 14256 kByte data can be expected during a 30 minutes logging period in case of 15 valid satellites. Header

#obs

range records

CRC

28 Byte

4 Byte

#obs x 24 Byte

4 Byte

Figure 3.9. Structure of a range data log. An important parameter to consider is the lifetime of the GNSS sensor node. The lifetime is the time the node can autonomously operate before it stops to operate due to low energy. As presented in [13], the lifetime is connected to the node’s battery capacity and to the times where the node is actively logging (tlog ) or communicating (tcom ) as the sensor node requires most energy during these periods. During idle times

34


(tidle ) the node requires only marginal power in comparison to active times. Figure 3.10 depicts tlog , tcom and tidle during a certain time interval tint . t_log

t_com

t_idle

t_int

Figure 3.10. States of the sensor nodes separated into active periods (tlog ) and (tcom ) and idle period (tidle ) during a time interval (tint ). To calculate a certain lifetime, we considered one logging interval (tlog ) to be 1800 s and the corresponding communication interval (tcom ) to be 900 s for 4 Hz logging frequency and 3600 s in case of 20 Hz logging. The interval (tint ) is 24 hours, so if the sensor node does one logging in 24 hours, it is in idle state for 83700 s ((tidle = tint −tlog −tcom ). Figure 3.11 shows the lifetime of our GNSS sensor node plotted over increasing number of logging and communication intervals in 24 hours assuming a battery capacity of 40 Ah and no energy harvesting.

Figure 3.11. The lifetime of the GNSS sensor node plotted over increasing number of logging and communications intervals in 24 hours assuming a battery capacity of 40 Ah and no energy harvesting. The black curve illustrates the estimated lifetime for 4 Hz logging and the red curve for 20 Hz logging.

3.3

Gateway and Remote Server

The GSM (global system for mobile communications) node acts as a gateway between local wireless sensor network and the remote server. It has two communication modules, as visualized in Figure 3.12. Figure 3.13 shows a photo of the gateway with its two

3.4. Wireless Sensor Node with Antenna Diversity

35

antennas. The node has an internal memory buffer to transmit several data packets at once to the server via GSM.

Battery

DC-DC

MCU

Wake-upAReceiver MainARadio

PowerASupply

Memory Processing

GSMAModem

Local Antenna GSM Antenna

Communication

Figure 3.12. Schematic of the base station.

Figure 3.13. Photo of the GSM base station node. The remote server was a Raspberry Pi located at the IMTEK building 106 (Laboratory of Electrical Instrumentation) running a MySQL database. The main purpose of the database was to store the data received from the wireless sensor network. Additionally, the database was used to reassemble the previously separated GNSS messages using their measurement timestamps as unique IDs as will be introduced in following Section 5.1.1.

3.4

Wireless Sensor Node with Antenna Diversity

This Section presents a newly designed wake-up node that uses polarization diversity in the wake-up path to improve wake-up robustness and reliability, as introduced and discussed throughout Section 2.1.2. Moreover, the node requires almost no additional active parts that would increase power consumption. The diversity node is based on the same design as introduced above in Section 3.1, which means it also uses the low-frequency wake-up receiver AS3932. This device supports the build up of a selection diversity system as introduced in Section 2.1.2 as it features three input channels. If several antennas are connected to the receiver

36


the AS3932 always selects the channel that has the highest RSSI value [54]. The node introduced in this section uses the AS3932 low-frequency wake-up receiver to implement an equal gain diversity system to further improve the wake-up reliability as discussed in Section 2.1.2. Figure 3.14 shows the block diagram of the proposed diversity system. The system basically combines two sinusoidal signals after the demodulation stage. The two low-frequency signals can be combined directly without the need of additional and energy expansive signal processing to control the signal phases to achieve constructive interference. It is obvious that destructive interference of two sinusoidal signals of the form yi (t) = Ai ej(ω+φi )t with i = 1, 2 happens when the phase shift (φ2 − φ1 ) is in the range 2/3π < φ2 − φ1 < 4/3π. The down converted signal at the wake-up receiver input is at 125 kHz. In case of destructive interference and by using the relation ∆t = ∆ϕ/ω a time offset ∆t in the range of 2.66 · e−6 s < ∆t < 5.33 · e−6 s can be calculated. As the signals are transmitted at speed of light c = 300 · e6 m s−1 , this translates to an antenna separation between 800 m and 1600 m, which is impossible with the hardware used in this work.

Matching Network

Rectifier Wake-up Receiver

Matching Network

Microcontroller

Rectifier

Figure 3.14. Block diagram of a low-power wake-up receiver with antenna diversity. Each diversity branch consists of antenna, matching network and rectifier. As can be seen, the block diagram in Figure 3.14 is similar to the maximal-ratio combining system shown in Figure 2.3 of Section 2.1.2 except for the gain units and consequently resembles an equal gain system (see Section 2.1.2). According to Equation 2.18, the output signal y(t) after combining 2 antenna branches is: y(t) = a1 y1 (t) + a2 y2 (t)

(3.1)

where yi (t) is the corresponding input signal of the ith diversity branch and ai its gain. Figure 3.15 shows a photo of the wireless sensor node with antenna diversity. To feature multiple antennas, the board is equipped with two antenna ports, which are connected to an ADG918 antenna switch from Analog Devices. By inserting a third antenna switch, it is possible to use both antennas as input and output. Due to the


37

three extra antenna switches, the node has an additional power consumption of less than 3 µA compared to the power consumption of the node introduced above. Figure 3.16 illustrates the block diagram of the sensor node. As discussed above, the low frequency wake-up receiver AS3932 has a typical sensitivity of 100 µV RMS.

Figure 3.15. Photo of the wireless sensor node with equal gain diversity.

Figure 3.16. Schematic of the diversity system.

3.4.1

Experimental Results

To verify the design and to test the performance of the wake-up diversity node, several experiments were performed. First, a static experiment was used to verify expected input signal gain of about 3 dB. Then, the performance of the wake-up diversity system was investigated in a multipath environment.

38


Antenna port impedances To verify the impedances of both antennas, a frequency sweep from 820 Mhz to 920 Mhz was performed and the reflections were measured with an Agilent E5071B network analyzer. The measurement results are plotted in Figure 3.17 that illustrates the reflection of the straight antenna in yellow and the reflection of the angular antenna in green. As can be seen, the minimum reflection is at approximately 870 MHz for the straight antenna and at approximately 878 MHz in case of the angular antenna. With a reflection below -15 dB, both antennas perform very well in the frequency region of interest that is around 870 MHz.

0

straight antenna port angular antenna port

reflection in dB

−5

−10

−15

−20

−25

−30 820

830

840

850

860 870 880 890 frequency in MHz

900

910

920

Figure 3.17. Measured voltage at the rectifier output with one active antenna. The typical sensitivity of the AS3932 is depicted as dotted line and the red circle shows the measured sensitivity.

Static measurements At first, a signal generator was connected to the antenna ports of the board, feeding them consecutively with a 868 MHz signal at different input levels. Since both measurements achieved almost equal results, Figure 3.18 shows the output voltage over the input level for the straight antenna, only. The measurements with the angular antenna revealed similar results and can be found in the third column of Table 3.1.


39

An exponential function was used to fit the measurement data, superimposed on a constant noise level, that is y1 = a + b exp(cx) with a = 0.043, b = 26 734 and c = 0.23. The theoretical sensitivity of the AS3932 intersects the fitted curve at -50.6 dBm. The circle in Figure 3.18 shows the sensitivity measured by using the AS3932 receiver: when the receiver did not further react to the input signals its sensitivity limit was reached. Using one antenna input feed, it was found at -51.3 dBm, which fits very well to the results reported by [13] and as discussed above.

Vo in mV

10

fit curve measurement data AS3932 typical sensitivity measured sensitivity

1

0.1

0.01 −70 −65 −60 −55 −50 −45 −40 −35 −30 input signal strength in dBm

Figure 3.18. Reflections measured at the straight (yellow) and angular (green) antenna input ports over frequency. In a second step, a second signal generator was connected to the other antenna input port also feeding it with a 868 MHz signal. Figure 3.19 shows the data curve fitted with an exponential function of the form y2 = a + b exp(cx) with a = 0.031, b = 45 410 and c = 0.226. The intersection of the fitted curve and the theoretical sensitivity line is at -53.7 dBm. The voltage where the AS3932 did not longer sense the input signal was found to be at -53.8 dBm, depicted as a circle in Figure 3.19. This is a gain of around 3 dB compared to the system with one antenna. Table 3.1 lists the experimentally received data that is also plotted in Figures 3.18 and 3.19 plus the measured voltage for the case of only the angular antenna input powered. Table 3.2 summarizes the experimentally identified wake-up sensitivities for the four possible combinations of the experiments above. Figure 3.20 compares the input signal strength with one receiving antenna two the input signal strength with two receiving antennas in dB. The red crosses illustrate the measurement data the solid curve depicts the fitted curve achieved by dividing y1 by y2 . It can be observed that the signal gain

40


Vo in mV

10

fit curve measurement data AS3932 typical sensitivity measured sensitivity

1

0.1

0.01 −70 −65 −60 −55 −50 −45 −40 −35 −30 input signal strength in dBm

Figure 3.19. Measured voltage at the rectifier output with two active antennas. Typical sensitivity of the AS3932 is depicted as blue dotted line. The red circle marks the measured sensitivity.

Table 3.1. Output voltage of rectifier at decreasing input signal levels using different antenna configurations. Input level Measurement (mV) (dBm) straight antenna angular antenna both antennas -

10 20 30 40 45 50 55 60 65

1050 234 26.9 2.76 0.90 0.31 0.12 0.07 0.05

1036 228 26.1 2.69 0.90 0.31 0.14 0.08 0.06

1190 316 52.1 5.49 1.78 0.59 0.22 0.10 0.06

reaches approximately 3 dB in the area where the input signal strength varies from - 50 dBm to - 30 dBm which correlates well to Equation 3.1 in Section 2.1.2 with a1 = 1 and a2 = 1. Below an input signal strength of - 50 dBm the gain decreases, probably because of decreased efficiency of the rectifier diodes.


41

3.5 3

gain in dB

2.5 2 1.5 1 0.5 0 −60

−55

−50

−45

−40

−35

−30

−25

−20

input signal strength in dBm

Figure 3.20. Experimentally measured gain in dB of the two antenna diversity system. Table 3.2. The table shows the wake-up sensitivities of the wireless sensor node achieved by measuring the rectified voltage and by feeding the AS3932 with a 868 MHz input signal

voltage measurement wake-up signal

one antenna

two antennas

- 50.6 dBm - 51.3 dBm

- 53.7 dBm - 53.8 dBm

Measurements in multipath environment For the second test, the diversity node was equipped with antennas and the 868 MHz wake-up message was transmitted wirelessly, but generated by a signal generator to be able to accurately control the output power. Figure 3.21 depicts the test setup. The sender was placed at a fix position in the laboratory, the receiver was positioned at several different locations. Inside the laboratory were several randomly placed objects like chairs, tables, shelfs and general laboratory equipment. In summary, the objects generated an multipath environment suitable to test the diversity wake-up system. Table 3.3 reports the measurement results at the six positions. The first column gives the position, the second column gives the required transmit power to wake-up the node with one antenna connected in straight position. The third column is the required transmit power to wake-up the node with one antenna connected in angular position and the last column are the required transmit powers when both antennas are connected to the node. Assume that a selection diversity system chooses the signal with the highest RSSI

42


transmitter

randomly placed objects

A1 A2

receiver

walls

Figure 3.21. Multipath laboratory setup with randomly placed objects. value. In this particular experiment, the equal-gain diversity system presented here, outperformed both single antenna solutions about approximately 0.2 dB to 2.5 dB and consequently successfully demonstrated its superior performance compared to selection diversity. Figure 3.22 illustrates the required transmitter powers at the 20 locations graphically. The blue circles show the results for the straight antenna system, the green squares for the angular antenna system and the yellow markers depict the equal-gain diversity results. In this particular experiment, the equal-gain system outperformed both other configurations in all measurements, as can be seen in Figure 3.21. To visualize the benefit of the equal-gain diversity wake-up system compared to a selection diversity wake-up system, Figure 3.23 plots the signal strengths differences required to wake-up the wireless node using the equal-gain diversity method compared to selection diversity in ascending order for the 20 observed locations. It can be seen that the gain of the equal-gain diversity system lies between 0.2 and 2.5 dB. Adding signals received by both antennas in an equal manner, is also a possible drawback of the equal-gain diversity system. For example, if one antenna receives a strong noise signal, the effective signal-to-noise-ratio at the receiver input will be reduced compared to the signal strength that would be received with only one receiving antenna that has a high signal-to-noise ratio. On the other hand, most selection diversity systems choose the channel with the apparent highest received input signal strength, but not the channel with the highest signal-to-noise-ratio.

3.4.2

Conclusion

The ultra low-power wireless sensor node presented in this section successfully demonstrated the use of equal-gain-combining in combination with a low-frequency wakeup receiver. Resulting from the wake-up frequencies in the kilohertz range, energy-


43

Table 3.3. Required sender power to wake-up the receiver measured at 20 different multipath locations and different antenna configurations. location straight antenna angular antenna both antennas (dBm) (dBm) (dBm) 1 - 4.2 - 16.3 - 17.1 2 - 7.7 - 3.9 - 9.1 3 + 0.3 + 4.8 - 0.7 4 + 4.8 - 1.1 - 3.6 5 - 0.4 - 4.0 - 5.8 6 - 8.2 - 10.8 - 13.2 7 - 8.2 + 11.0 - 8.9 8 - 11.4 - 1.8 - 11.9 9 - 11.4 - 1.4 - 13.0 10 + 0.2 - 8.9 - 10.1 11 - 2.1 - 12.8 - 13.1 12 - 9.8 - 4.5 - 10.4 13 - 8.0 - 9.2 - 11.6 14 - 11.4 - 5.4 - 12.4 15 - 11.2 - 7.5 - 12.1 16 - 5.9 - 0.4 - 6.9 17 - 5.2 - 9.6 - 10.3 18 - 5.0 - 0.5 - 6.4 19 - 2.4 + 11.5 - 2.6 20 - 1.0 - 12.3 - 13.4 demanding phase control circuits are not required. The implemented and investigated equal-gain wake-up diversity node requires only two additional antenna switches to combine the multiple input signals and that increase the power consumption only marginally. The polarization diversity technique was implemented, although this is no limitation and other diversity techniques like space diversity could also be utilized. It could be verified that the antenna diversity achieved 3 dB gain under ideal laboratory conditions where the signals were free from noise, but not phase aligned. A laboratory multipath environment experiment confirmed the performance gain between 0.2 and 2.5 dB of the equal-gain wake-up diversity system.

44


20

transmitter power in dBm

15

straight antenna angular antenna both antennas

10 5 0

−5 −10 −15 −20

2

4

6

8

10 12 Location

14

16

18

20

Figure 3.22. Required power to wake-up the receiver at different positions.


45

3

2.5

gain in dB

2

1.5

1

0.5

0 2

4

6

8

10 12 location

14

16

18

20

Figure 3.23. Gain achieved by the equal-gain diversity wake-up receiver system over the selection diversity wake-up system for the 20 location in ascending order.

Chapter 4 Communication with Wake-up Receivers Wake-up receivers are the natural choice for wireless sensor networks because of their ultra-low power consumption and their ability to provide communications on demand. A downside of ultra-low power wake-up receivers is their low sensitivity caused by the passive demodulation of the carrier signal. This chapter provides a review on existing communication approaches with wake-up receivers in the first Section 4.1. The next Section 4.2 presents T-ROME, a simple and energy efficient cross-layer routing protocol for wireless sensor nodes with wake-up receivers. The protocol makes use of the different transmission ranges of wake-up and main radios in order to save energy by skipping nodes during data transfer. With respect to energy consumption and latency, T-ROME outperforms existing protocols in many scenarios. The cross layer multi-hop protocol will be described and analyzed by means of a Markov chain model and verified in a laboratory test setup. The following Section 4.3 presents several modifications to T-ROME that does not change the basic behavior of T-ROME, but significantly reduce its requirements with respect to communication energy, time and overhead, especially in noisy environments. Furthermore, Section 4.3 presents a technique to prevent disconnected network nodes in some special cases as will be highlighted. Section 4.4 presents a novel communication scheme by exploiting purposefully interfering out of tune signals of two or more wireless sensor nodes, which produce the wake-up signal as the beat frequency of superposed carriers. Additionally, a communication algorithm and a flooding protocol based on this approach are introduced. Experiments are performed to demonstrate the performance of the newly developed approach to increases the received signal strength up to 3 dB, improving communication robustness and reliability. Furthermore, the feasibility of the newly developed protocols are demonstrated by means of an outdoor experiment and an indoor setup. 46

4.1. Existing Protocols for Wake-up Receivers

4.1

47

Existing Protocols for Wake-up Receivers

Although wake-up receivers have many advantages and are frequently used in wireless sensor networks [4–7], there do not exist many media access control (MAC) or routing protocols that support their use and the majority of existing protocols are only limited to simulations. Some existing protocols for wake-up receivers support single-hop communication only, like E2RMAC [55], WUR-MAC [56], RTWAC [46] and GWR-MAC [57]. These protocols show superior energy requirements compared to synchronous or asynchronous MAC protocols but their performance is only based on simulation results. The main feature of E2RMAC and WUR-MAC protocols is to use the wake-up signal as an RTS/CTS mechanism to avoid the hidden terminal problem. In RTWAC all nodes have a unique and a common wake-up address to support broadcasting and dedicated messages. But the purpose of wake-up messages is only to trigger an event, for example, a sensor reading, at the receiver node. Data communication is realized by a more common CSMA/CA MAC protocol that is not further specified, using the main radio. The protocols presented in [58] and [7] were tested in real applications but are also limited to single-hop communications. Similar to those protocols but designed for body area networks is the work of [45]. The protocol introduces additionally a random back-off time to avoid collisions. The protocol as presented in [58] combines wake-up messages and a low duty cycle TDMA based MAC protocol [59] to increase flexibility. Performance evaluation is done by comparing the proposed protocol with and without a wake-up radio. Recently [52] presented a novel wake-up receiver design together with two flooding protocols FLOOD-WUP and GREEN-WUP. FLOOD-WUP uses different broadcast addresses to forward messages to receivers that are not in range of the first transmitter and to avoid the reception of multiple messages. GREEN-WUP includes additional information about harvested energy at a node coded in its address and nodes with higher energy levels are preferred relay nodes. Evaluation of both protocols is only performed on the basis of simulation and the authors do not evaluate the power requirements of the proposed protocols. [60] presented ZIPPY, an on-demand multi-hop flooding technique based on wake-up receivers. ZIPPY is extensively tested in a laboratory testbed and shows latencies in the range of tens of milliseconds to broadcast multi-hop messages. CTP-WUR, a cross-layer routing protocol for wake-up receivers presented in [61] introduces relaying of wake-up messages by using flagged wake-up messages to inform the receiver about the intended multi-hop wake-up. The relaying node forwards the wake-up call to its parent node that itself starts to wait for data from the first node. In case the node woke up due to a false wake-up, the node goes back to sleep after a

48

4. Communication with Wake-up Receivers

predefined time has passed and no data is received. The protocol allows for relaying of one wake-up message, only. Wake-up messages are not acknowledged but successful data transfer is indicated by an acknowledgment from receiver to the sender. Data communication is done via the CTP routing protocol [62]. The authors of [63] present MH-REACH-Mote, a node based on the Tmote-Sky platform in combination with a wake-up receiver. In their scenario, communication is done from a mobile sink to fix nodes. Wake-up messages are relayed from the nearest fix node to the ones further away from the mobile sink. The protocol assumes no collisions and an existing communication link from the fix sources to the mobile sink. The protocol presented in [64] uses low-power wake-up receivers to create clusters of sensor nodes that exhibit similar sensor readings and only cluster heads transmit information to the sink. Sensor readings and cluster configuration messages are encoded into wake-up messages. The protocol shows promising results for applications with many similar sensor readings. The authors of [47] introduce ALBA-WUR a crosslayer network protocol that supports the use of wake-up receivers. Wake-up addresses are chosen dynamically from a set of predefined wake-up addresses, depending on packet size and on historical node performance. In simulations, ALBA-WUR showed superior power consumption and latency as compared to ALBA-R a geographic crosslayer routing protocol with contention-based MAC [65]. Reducing network latency and power consumption of wireless sensor nodes by minimizing the amount of required wake-up messages are the goals of the Embedding Tree Algorithm and the Greedy Shortest Wake-up Path Algorithm, two algorithms recently presented by Bannoura [66]. Both concepts are similar to the herein proposed protocol and will be reviewed in the following paragraphs to point out parallels and possible drawbacks or advantages. In the Greedy Shortest Wake-up Path Algorithm [66] each node that has data to transmit sends a wake-up message to a node that is positioned nearer to the sink. The newly woken node relays the wake-up message until the sink is reached. A subset of the woken nodes stay active to deliver the data over multiple hops to the sink, making use of the long range communication radio. The Embedding Tree Algorithm [66] proposes a hybrid routing scheme that combines the traditional duty-cycling approach with the wake-up approach. In case a node has data to send, it probes for actively duty-cycling nodes in communication range that form a backbone to the sink. If the node finds such a backbone, it connects itself to it and sends its data packet to the sink utilizing the existing backbone. If the node does not find an existing backbone, it sends a wake-up message to one of its neighbors. This scheme continues until the sink is reached and as with the Greedy Shortest Wake-up Path algorithm, a subset of the woken nodes stay active to form a backbone and to forward the data to the sink [66]. The algorithm organizes the nodes in in clusters to


49

achieve a hierarchical structuring. Both algorithms are analyzed in [66] and their latencies and energy consumptions are compared using Matlab. But from an application-oriented point of view, the concepts leave several questions open to the reader. First, both algorithms assume that the node positions are known to the network, for example achieved from the hierarchical clusters. But there is no solution provided how the clustering is accomplished, Bannoura suggests that the sink node broadcasts this information initially but this leaves the question open how the information are broadcast, especially as there are multiple hops required to reach every node in the network. In the Greedy Shortest Wake-up Path Algorithm, Bannoura suggests that a subset of the waken nodes stays active to forward the data to the sink, but a possibility to determine this subset is not provided. Furthermore, there is no communication algorithm suggested that the backbone nodes can apply to transmit the data. The Embedding Tree Algorithm has the same application-oriented shortcomings as mentioned above. Furthermore, Bannoura does not provide a solution how a newly woken node can detect an active backbone. This could be achieved either by actively sending a request and then listen for a reply, what would also consume energy. Keeping in mind, that the backbone nodes are duty-cycling and not listening always, the required energy to transmit the requests depends on the duty-cycling period. If it is short, the request packets can also be short and few packets suffice to reach a backbone node, thus the node requires only few energy. But, then the backbone nodes need a considerable amount of energy by establishing a fast duty-cycling that is wasted if no messages are pending to be transmitted. If the duty-cycling period is long, the backbone nodes do not need much energy to keep the backbone alive, but the requesting node has to spend more energy to connect itself to the backbone. Additionally, the request could collide with an ongoing communication on the backbone, in which case several messages could fail: the request and the messages currently flowing on the backbone. Also the case of several nodes sending a request concurrently would have to be considered. The possible occurrence of collisions is well known and can be handled with, for example by applying listen-before-talk in combination with a back-off algorithm. Alternatively, the backbone nodes could themselves send beacons to announce their active state. A newly woken node would listen for a beacon and respond to the backbone node that is nearest to the sink. This method would cost the additional energy required by the backbone nodes to send the beacons and the energy required by the just awoken node to listen for a beacon. If the duty-cycling period is fast, the latter energy can almost be neglected but than the backbone nodes would require a considerable amount of energy to transmit the beacons. If the duty-cycling period would be slow, the amount of energy required for a newly woken node to listen for a beacon

50


would increase and it could eventually not be neglected. Furthermore, the backbone nodes themselves require a certain amount of configuration overhead to organize themselves. Additionally, Bannoura [66] leaves the question open how long a backbone stays active after the final message of an ongoing communication was transmitted successfully. A good answer to this question is to let this time be dependent on the network traffic, but in [66], it is not mentioned which metric could be used to measure the network traffic or how this metric, once established, could be used to set the duty-cycling period. In summary, the ideas proposed by Bannoura [66] are promising, but a thorough evaluation and comparison to the algorithm proposed in this work is not practicable as several design issues are left open. From a behavioral point of view, the algorithms have clear parallels to the protocol design proposed herein, although they support sending of only one data packet [66] on an existing link. Resulting from this, the algorithm proposed in this work clearly outperforms the Greedy Shortest Path Algorithm if data aggregation is applied. The Embedded Tree Algorithm mitigates this drawback by implementing the duty-cycling approach. Considering the Embedded Tree Algorithm in a low traffic network, its behavior is similar to that of the herein proposed algorithm, but it requires the additional configuration overhead of the backbone nodes. The algorithm introduces the potential advantage that newly woken nodes do not need to send a wake-up message, but this also introduces new energy demands either on the backbone side or to the newly woken nodes, as discussed above. Considering a high traffic network, the behavior of the Embedded Tree Algorithm becomes similar to that of a traditional duty-cycling network and in case the backbone nodes are not chosen wisely, they could quickly deplete their energy as all traffic tends to flow through them. A solution could be for example, some sort of clustering similar to the LEACH (LowEnergy Adaptive Clustering Hierarchy) protocol [67] and generating the backbone from randomly chosen cluster-heads.

4.1.1

Opportunistic routing with wake-up receivers

As pointed out for example by [68], the quality of wireless links can change quickly due to changes in the environment. To achieve a robust, reliable and efficient routing, stateof-the-art wireless network protocols like CTP [62] estimate the current link quality between nodes and adjust their routing paths accordingly. The link quality estimation can either be achieved by incorporating information from different network layers like the number of received acknowledgments and the link quality indicator provided by the radio or it can be based on the β-factor [68] that measures the burstiness of a wireless link. While link estimation is a common technique in traditional wireless network protocols, it is not standard in all used wireless routing protocols that are based on


51

wake-up receivers since an accurate and timely link quality estimation requires a certain amount of control messages (beacons) to be sent. This is energy-wise expensive due to the high costs of wake-up messages. ALBA-WUR, for example, calculates the link quality by taking into account how many packets have been lost on a specific link in the past. This achieves a good average link quality information but cannot resemble fast or short link quality changes. To avoid collisions and to improve the reliability WUR-MAC chooses dynamically one out of several available channels of the 2.4 GHz ISM band for wake-up transmissions. To choose a channel, the protocol keeps track of all channels used in neighboring nodes for communication and then takes randomly one of the remaining channels for its own communication. This approach does not avoid collisions and like ALBA-WUR only calculates an average channel usage without the possibility to react on rapid channel fluctuations. T-ROME introduces a parameter to assist the sender in order to dynamically choose the best next hop node based on multiple values like distance to the source and link quality estimation. This also enables route adjusting on rapidly changing link conditions. To reduce the number of transmissions from source to sink and to increase network performance, opportunistic routing protocols rely on broadcasting data packets to several nodes (the set of candidates) to forward a message from a source to sink [69]. Usually, the most appropriate forwarder is chosen out of the set of candidates based on local and end-to-end metrics. Local metrics are based on link conditions and geographic positions of the sensor nodes, while end-to-end metrics are usually based on link properties between source and destination [69]. In traditional opportunistic routing, it is necessary that each node of the candidate set receives the broadcast data packet and answers back to the sender. The authors of [69] categorize this candidate coordination into two groups, either being based on control messages or on time-coded sending of data packets. In the latter, a node’s priority is proportional to a time period it waits until it forwards a data packet. If a node overhears a data transmission from another node, it knows that it does not have the highest priority and does not forward the packet. The drawback of this method is that multiple data packets may be transmitted in case a node does not hear the transmission of another one. In case the candidate coordination is based on control messages, acknowledgments or the RTS-CTS frame can be used [69]. In both approaches, the time to sent an acknowledgment or the CTS message is proportional to a node’s priority and if a node overhears an acknowledgment or a CTS message from another node, it backs off. The difference between the two approaches is that in acknowledgment based candidate coordination, the data packet is received by all possible candidates and in the RTS-CTS approach, the data is sent to the most appropriate candidate only. FLOOD-WUP realizes opportunistic routing according to the acknowledgment based

52


approach but forwarding is done after a random period of time has passed. To avoid multiple transmissions of the same data packet, each node changes its wake-up address upon reception of a data packet. Although changing of the wake-up address follows a predefined sequence, it can happen that a node loses the proper sequence due to a false wake-up, and additional control packets are required to reestablish the correct sequence [52]. The opportunistic routing in GREEN-WUP is similar to that of FLOOD-WUP but wake-up addresses are additionally based on the current energy level of a sensor node and the source node goes to sleep after it sent the initial wake-up sequence. A possible relay node has to wake up the source by using a unicast wake-up packet that was provided by the source during initial communication. Due to this, GREEN-WUP requires additional wake-up packets that are usually expensive with respect to energy.

4.2

T-ROME

The cross-layer routing protocol presented in this section is based on existing hardware, in contrast to most of the protocols introduced above. It realizes the RTS/CTS messages similar to those presented above but additionally T-ROME supports multi-hop communication and forwarding mechanisms similar to those presented in ALBA-WUR and GREEN-WUP but T-ROME does not use flooding. The protocols presented in [61] and [63] use relaying of wake-up messages similar to the herein proposed solution but use only one relay node, whereas the number of relay nodes in T-ROME is not limited. Additionally, T-ROME includes a set of decision parameter that can be used to dynamically optimize the relaying process and to choose the optimal relay node similar to the opportunistic RTS-CTS approach shown above, but the candidate set is established during the routing itself. Furthermore, T-ROME introduces a mechanism to send several data packets in a row along an existing link. As already introduced in Tables 1.1 and 1.2, the sensitivity of wake-up receivers is lower than that of communication radios. Consequently, data can be sent over longer distances than wake-up messages as shown in Section 3.1. Due to this, T-ROME is a cross-layer protocol, as visualized in Figure 4.1. Above the physical layer is the link layer that supports single-hop transmissions and waking up of neighboring nodes. This is realized by using an RTS/CTS message exchange to reduce packet collisions as introduced in MACA [70]. In this context, a wake-up message works also as an RTS and the wake-up acknowledgment as the CTS command. The routing layer routes messages along multiple hops according to a static routing table implemented on each node. Following sections introduce the cross-layer protocol and corresponding data packets in more details. The application runs above the communication layers.

4.2. T-ROME

53 Application Routing Layer T-ROME

Link Layer MACA + Wake-up Physical Layer

Figure 4.1. Communication layer stack consisting of physical layer, link layer, routing layer and application. The cross-layer protocol T-ROME supports functions in the link and the routing layer as depicted in the figure. The Wake-up is embedded in the link layer and supports the RTS/CTS scheme based on MACA to reduce packet collisions.

4.2.1

T-ROME Protocol

The static cross-layer protocol is based on the simple Tree Routing algorithm [71]. In this protocol messages can be passed only from child to parent nodes as depicted in Figure 4.2. Every node of a certain depth i is able to communicate with a node of depth i − 1 and vice versa. For example node b is able to communicate to node a.

a

depth 0

depth 1

depth 2

c

b

d

e

f

g

Figure 4.2. Schematic of a simple tree routing protocol with nodes a to g. Communication is only possible from child to parent for example from node b to node a.

The protocol proposed in this work is sketched in Figure 4.3. Sending wake-up messages is similar to the Tree Routing protocol introduced above. It is possible for nodes of depth i to nodes of depth i − 1 where they are in wake-up range. Communication data can cross several levels from depth i to depth i − n with n ∈ N limited by the root node and communication range. In Figure 4.3 node 13 sends for example a wake-up signal to node 12 which forwards the wake-up to node 11 and so on until a defined maximum number n of forwards or the destination is reached. Afterwards, the data can be sent directly from node13 to one of the woken nodes 10, 11 or 12.

54


node 10 data

depth 0 in wake-up range

wake-up node 11

node 21

depth 1

data wake-up data

node 12 wake-up

node 13

in wake-up range

node 22 in wake-up range

depth 2

node 23

depth 3

Figure 4.3. Schematic of the wake-up multi-hop routing protocol developed in this work.

Wake-up Layer

The Wake-up layer is responsible for waking up of neighboring nodes. Each wake-up packet consists of carrier burst, preamble and receiver ID as depicted in Figure 4.4. The carrier burst tunes the detector to the incoming frequency, the preamble is used by the detector to estimate bit length and possible offset. The receiver ID is an up to 16 bit long address to identify the receiver. When sent at a data rate of 8192 bps the wake-up message can be between 148 and 216 bytes long depending on the length of carrier burst and preamble. In a noisy environment, it is recommended to use longer carrier burst and preamble. Before an attempt is started to wake up a neighboring node each node probes the wireless channel (LBT). If a communication is currently going on, the nodes back off and restart the attempt later. After the wireless channel is found to be free each communication is initiated by sending a wake-up message. The receiver acknowledges this wake-up packet (WUC) with an acknowledge message (WUC ACK) that includes the address information of receiver and a protocol ID as can be seen in Figure 4.5. If the address does not match the receiver ID or if the acknowledge message was not received before a certain timeout is reached, waking up is assumed to be unsuccessful and has to be restarted. The packet flow is schematically sketched in Figure 4.6. To reduce collisions, the wake-up layer protocol realizes an RTS/CTS mechanism as depicted in Figure 4.1. The protocol ID is transmitted at an early stage to be able to include newer protocol versions that could react differently upon reception of certain communication packets.

4.2. T-ROME

55 Carrier Burst

Preamble Receiver ID

32 to 100 Byte

52 Byte

64 Byte

Figure 4.4. 125 kHz wake-up call packet (WUC) including 32 to 100 byte carrier burst, 52 byte preamble and 64 byte receiver ID sent at 8192 byte per second. Protocol-ID

Receiver ID Low-Byte

1 Byte

Receiver ID High-Byte

1 Byte

1 Byte

Figure 4.5. Wake-up acknowledge (WUC ACK) packet consisting of 3 byte (Protocol ID, Receiver ID low byte and Receiver ID high byte).

{

timeout

Main Radio Wake-up Radio

ACK

Receiver

{

WUC

Sender

Wake-up Radio

Main Radio

Figure 4.6. Packet flow of wake-up and main radios in Wake-up Layer. Communication MAC Layer The communication MAC layer consists of two types of packets, a data packet, and an acknowledge packet. Each data packet is answered by an acknowledge packet. If the acknowledge packet is not received during a certain time frame, it is assumed that sending of data has failed. Failed data packets are reinserted into the send queue to be resent later. Figures 4.7 and 4.8 show the data and the acknowledge packet. Packet type is used to separate the packets. IDs of the source (Src ID) and destination (Dest ID) are used to verify sender and receiver. The length byte is required internally for packet handling. Packet Type DATA 1 Byte

Src ID

Dest ID Length

1 Byte

1 Byte

1 Byte

Figure 4.7. Data packet consisting of 4 byte (Packet Type DATA, Source ID, Destination ID and payload length).

Packet Type ACK

Src ID

Dest ID

1 Byte

1 Byte

1 Byte

Figure 4.8. Acknowledge packet consisting of 3 byte (Packet Type ACK, Source ID and Destination ID).

56


Routing Layer While the MAC layer is responsible for the communication between neighboring nodes, the routing layer handles communications between nodes that are possibly further apart than only one hop. Routing packets are embedded into MAC layer data packets as depicted in Figure 4.9.

}

optional

MAC Packet Type DATA

Routing Packet Header

4 Byte

4 Byte

Payload 1 - 246 Byte

Figure 4.9. Routing packet embedded into MAC packet. The routing layer takes care of sending, receiving and forwarding packets from source to destination. Figures 4.10, 4.11 and 4.12 show the three available routing packets, namely routing request (R REQ), data (DATA) and acknowledge (R REQ ACK). Each packet consists of four bytes. All data to be sent is managed in data slots that form the message queue. The first six bits of a request type packet are reserved for the number of slots to be sent in the currently ongoing communication. R Src Id and R Dest ID are the routing source and destination IDs of the communicating nodes which could be equal to the MAC IDs but can also be different. TTL (time to live) indicates how many hops a request can be forwarded. Upon reception of a routing request, the receiving node decreases TTL by one, before forwarding the request to the next node. In case TTL is zero the request will not be further forwarded. Forwarding of routing requests is realized with route request packet type packets keeping source and destination ID untouched. Num of slots Packet Type REQ Src ID Dest ID TTL 6 Bit

2 Bit

1 Byte

1 Byte

1 Byte

Figure 4.10. R REQ (Routing Request) packet consisting of 4 byte (Number of slots to send (6 bit) Packet Type REQ (2 bit), Source ID, Destination ID and time to live (TTL)).

Packet Type DATA R_Src ID R_Dest ID Length Payload 1 Byte

1 Byte

1 Byte

1 Byte

1 - 246 Byte

Figure 4.11. DATA (Routing Data) packet consisting of 4 byte (Packet Type DATA, Routing Source ID, Routing Destination ID and payload length). Figures 4.13, 4.14 and 4.15 show the sequence diagrams of the routing protocol in case of four participating nodes. Node A is the source node, nodes B and C are

4.2. T-ROME

57 Packet Type ACK 1 Byte

TTL

LQI

Number of free Slots

1 Byte 1 Byte

1 Byte

Figure 4.12. R REQ ACK (Routing Acknowledge) packet consisting of 4 byte (Packet Type ACK, current time to live (TTL), Link Quality Identifier (LQI) and number of available memory slots). possible relay nodes and node D is the sink. Node A starts by sending a routing request (R REQ) to node B. Node B forwards the request (FWD REQ) to its next neighbor node C who will again forward the request to node D. Each node (B, C, and D) answers the request by sending of a request acknowledge (R REQ ACK) to node A. Node A collects all request acknowledgments and decides based on the information included in the acknowledges to which node the data will be sent. Currently implemented parameters that support the decision, to which node data is sent to, are: available data slots at the receiving node and hop distance from starting node. Further parameters like the available energy at receiver node or various status data like link quality or number of successful wake-ups can be easily used to increase the network stability. B

A

C

D

REQ

CK

FWD_R

A REQ_

EQ FWD_R

_ACK

EQ

REQ

CK EQ_A

R DATA

Figure 4.13. Sequence diagram of the routing protocol in case the data is sent to the next neighbor. Decision to where the data are sent is done at node A based on information included in the request acknowledge data. B

A

C

D

REQ

CK

A REQ_

FWD_R

EQ FWD_R

ACK REQ_ AC REQ_

EQ

K

DATA

Figure 4.14. Sequence diagram of the routing protocol for communication to the twohop distant neighbor. Decision to where the data are sent is done at node A based on information included in the request acknowledge data. Once a communication link to a node is established, up to 64 data packets consisting

58

4. Communication with Wake-up Receivers B

A

C

D

REQ

ACK REQ_

FWD_R

EQ

CK EQ_A

R

FWD_R

EQ

ACK REQ_ DATA

Figure 4.15. Sequence diagram of the routing protocol for communication to a threehop distant neighbor. Decision to where the data are sent is done at node A based on information included in the request acknowledge data. of up to 246 bytes each can be transmitted in a row. After transmission, the link gets closed and the participating nodes fall back to sleep, again. The same routing scheme is repeated until all data has reached their destination.

4.2.2

State Machine

As introduced in [6], the embedded software is implemented as a state machine as depicted in Figure 4.16. At the beginning, a sensor node is in SLEEP state in which it consumes only minimal energy. A low-energy timer transfers the node from sleep either to start a sensor measurement (state MEAS) or to check if there is data available in the memory that is not yet sent (state STORE). In case there are already prepared data slots available, for example from a previously aborted sending, the sleep state will be left and data transfer is initiated by sending a wake-up signal (state SEND WAKE-UP). After a measurement, sensor data is stored in a ringbuffer on the microSD card and data packets are prepared and moved into one of up to 64 available data slots. If there are no free slots available the data is kept in memory to be processed later. After successful filling the message queue, sending of data is initiated with a wake-up signal (state SEND WAKE-UP). Successfully waking of the neighbor node, is indicated by a wake-up acknowledge and a routing request is sent (state SEND R REQ) containing destination ID, number of data packets and max number of wake-up hops. Then, the node listens for route request acknowledgments sent by the woken nodes (state WAIT R ACK). If at least one node that answers has a free slot available, the node starts to send all possible data packets (state SEND DATA). After successful sending, or if any error occurs, the node exits its current state and goes back to sleep. The state machine of the receiver is similar to that of the transmitter. Looking at the state machine, it becomes clear that in case two sensor nodes try to send data at the same time, the data packets would collide and packet transmission

4.2. T-ROME

59 timed out timer

timer timed out or trigger

meas interval timed out start

SLEEP

failed

MEAS

STORE

no data or failure

success or failure

data slots filled timed out filled slots

SEND DATA

SEND WAKE UP

wakeup failed no node or failure

success

wakeup success

failed SEND R REQ

WAIT R ACK success

timed out

Figure 4.16. State machine of a sensor node for data transmission.

would fail. Additionally, T-ROME can encounter self-interference due to the forwarding mechanism of packets that are sent at the same time. To avoid collisions, each source node (but not the relay nodes, as the channel is assumed to be busy during the complete period of data transmission) probes the wireless channel before transmission and if it finds the channel busy it backs off for a certain time period before testing the channel again. To calculate the back off period a simple algorithm is used that calculates the back-off time based on the unique node ids. This means that nodes further away from the sink node have longer back off periods than nodes nearer to the sink. This avoids self-interference and reduces congestions near the sink during periods of high data traffic.

60


Wake-up Message The wake-up signal was received at a data rate of 8192 kbit per second (bit length: 122 µs), which means a 125 kHz period requires 4 byte ones and 4 byte zeros sent in a row at 250 kbit per second, resulting in a bit length of 128 µs. From sender (receiver) side, the wake-up message consisted of 42 byte (10 bit) Carrier Burst which is required at the receiver to detect the presence of a signal and to fine-tune its internal frequency to the incoming signal frequency. The preamble consisted of 48 byte (12 bit). Its purpose is to adjust the receiver offset to be approximately at the level of the averaged input signal and to verify the bit length. The pattern depicts the 16 bit address of the wakeup receiver. It requires sending of 64 byte (16 bit). In the case of Manchester coding, this results in an 8 bit address that can be used to address up to 256 independent devices. For example, the node ID sent in Figure 3.5 is decimal 85.

18Byte Length8field

7998µs

28Byte Sync8word

Calibration

28Byte Preamble

Figure 4.17 shows the message schematically. Before sending, the radio requires a calibration cycle. Preamble, a sync word, and length field are mandatory bytes which make a wake-up message 6143 µs long. Out of that, the radio is for 5344 µs in sending state and 799 µs in calibration state.

648µs

648µs

328µs

1628Byte Wake-up8Message 51848µs

Figure 4.17. Complete wake-up packet including calibration and mandatory radio bytes.

Communication Packets Figure 4.18 shows schematically the buildup of a complete radio packet including calibration of the radio, sending of the preamble, sync word, length, MAC, status and CRC bytes. Sending of payload and routing bytes is optional. All times (including calibration) in Figure 4.18 are calculated for a baud rate of 250 kbit per second and GFSK (gaussian frequency shift keying) modulation. Generally, sending of data is separated into hardware specific and protocol layer specific parts. In sum, each packet requires the hardware specific calibration, preamble, sync word and CRC which add up to around 991 µs. The rest of the time is required to send protocol messages, either wake-up, MAC or routing. A MAC packet requires 1247 µs and a routing packet without payload requires 1375 µs. The payload is sent in additionally 32 – 7872 µs, depending on payload size. According to the datasheet, the radio requires around 8.4 mA during calibration and when sending at 868 MHz, 0 dBm gain around 16.4 mA.

4.2. T-ROME

61

In receive state, the radio requires around 16.9 mA and for sending a wake-up call at +12 dBm gain the CC1101 needs 34.2 mA.

64Aµs

32Aµs 128Aµs 128Aµs

1A...A246AByte Payload

1AByte Status

1AByteA Length

128Aµs

4AByte MAC 4AByteA RoutingA

2AByte SyncAword

799Aµs

4AByte Preamble

Calibration

DataAfield

32A...A7872Aµs 32Aµs

2AByte CRC 64Aµs

Figure 4.18. Radio packet including calibration.

4.2.3

Experimental Analysis

In order to verify the assumptions on current consumption and timing intervals (as discussed in Section 4.2) communication messages and the protocol on the whole, are analyzed experimentally. The results are presented in the following Section 4.2.3.

Communication Messages Figure 4.19 shows exemplary the current consumption of a sensor node in the different states of the proposed protocol measured via a shunt resistor in the power line. In this example, the node sent 4 data packets consisting of 100 byte each to the next neighbor node. It can be seen that the currents provided in Sections 3.1 and 4.2.2 for microcontroller and radio fit very well to the measurement results for radio calibration, sending and receiving of communication packets, low-power listening, and microcontroller run mode current. It can be further seen, that sending of wake-up packets require less current than expected from the datasheet numbers, only. This is due to the fact that the Manchester encoded wake-up packets consist of an equal amount of zeros and ones and the radio power is reduced during sending of zeros. Furthermore, it can be seen that the timing fits very well to the suggested timing calculated in Sections 4.2.2 and 4.2.2. A logic analyzer was used to visualize all sending and receiving states. As laboratory test setup was the same as introduced in Figure 4.3 with four participating nodes: node 13 as source, node 12 and node 11 as relay nodes and node 10 as sink. Node 13 sent 5 data packets of 100 bytes payload each. According to the protocol nodes 12 and 11 forwarded the request to node 10 that finally received all data packets after around 90 ms. These times intervals can be seen in Figure 4.20 which shows the sending of 500 bytes payload in 5 packets of 100 bytes each over a row of four nodes as sketched in Section 4.2 Figure 4.3.

62


ACK WUC ACK

R_REQ ACK

0.01

0.02

0.03

0.04

0.05 Sec o n ds

ACK

R_REQ

Receive node 10

ACK

ACK

Send node 10

DATA

DATA 0.06

DATA

DATA 0.07

DATA 0.08

ACK

ACK

R_REQ

R_REQ ACK

ACK

WUC

ACK

ACK

R_REQ ACK

WUC ACK Receive node 11

DATA

ACK

WUC ACK Send node 11

0

ACK

R_REQ ACK

ACK

R_REQ ACK ACK

ACK

R_REQ


ACK

WUC

ACK

ACK

Send node 12

R_REQ ACK

R_REQ

WUC ACK

DATA

R_REQ

LBT

ACK

R_REQ ACK

DATA

DATA

ACK

ACK

R_REQ ACK

DATA

ACK

R_REQ ACK

DATA

ACK


ACK

WUC

ACK

Send node 13

R_REQ

Figure 4.19. Current drawn by the sensor node in the different states of the protocol.

DATA

0.09

0.1

Figure 4.20. Sending and receiving of 5 data packets in case of 4 participating nodes. Each node (node 13, node 12, node 11 and node 10) has a sending (upper line) and receiving (lower line) state. Node 13 is source, node 10 is sink. Nodes 12 and 11 forward the wake-up calls.

4.2. T-ROME

4.2.4

63

Numerical Analysis

In order to analyze the performance of T-ROME, and to compare it to other protocols, a Markov chain based model is introduced and meta-models of the routing algorithms T-ROME, CTP-WUR and of an algorithm, here called naive algorithm. It was decided to compare T-ROME especially to these two protocols, as most implemented networks use some derivate of the naive algorithm or a relaying mechanism similar to CTP-WUR. States of the meta-models can either be w or T . States w are states where a node attempts to wake-up another node and states T depict states in which data should be transferred from one node to another node. The models consist of a row of m nodes, subscribed with j and i, where i < j. To model the algorithms, it is further assumed that a message existent at node i at time t0 . Then, a w state could for example be, wi,i,i+1 , which means that the message is at node i (first subscript) and node i (second subscript) attempts to wake up the next node i + 1 (third subscript). In the case of T states the subscripts have the following meaning: Ti,j means a data transmission from node i to node j. Transitions between states are possible along the arrows which are connected to a certain cost that can be probability or time in a more general way. Subscript q describes the probability of a successful wake-up, and p describes the probability of a successful data communication. To simplify the models, equal success probabilities for all nodes is assumed, that is ∀i ∈ m : pi = p, qi = q. The Markov chain based model reflects errors on the medium access level and does not describe the dynamic routing behavior originating from changes in link quality estimations or due to changes in the energy level of certain nodes that could lead to different routes. This could potentially lead to different behaviors of the routing algorithms, and the comparisons presented in Section 5.2 might be influenced by this. An extended Markov chain based model that also reflects the dynamic behavior is clearly more complex and may be part of future research. T-ROME T-ROME, wake-up messages can be forwarded or send directly. In summary, there exist following four possible meta states for a T-ROME branch consisting of m nodes: The message is at node i and node i tries to wake-up node i + 1, for i < m − 1. When awake, node i + 1 tries to wake-up node i + 2. The message is still at node i. If wake-up of node i + 1 fails, the message stays at node i that will initiate another wake-up attempt at a later time. Figure 4.21 depicts this case. The message is at node i and node j tries to wake-up node j + 1, for i < j and j + 1 < m. When awake, node j + 1 tries to wake-up node j + 2. The message

64

4. Communication with Wake-up Receivers is still at node i. If wake-up of node j + 1 fails, node j is ready to receive the message. Figure 4.22 depicts this case. The message is at node i and node m − 1 tries to wake-up node m, for i < m − 1. After reception of the wake-up message, node m is ready to receive the message from node i. If wake-up of node m fails, node m − 1 is ready to receive the message. Figure 4.23 depicts this case. The message is at node m − 1 and node m − 1 tries to wake-up node m. After successfully waking up node m it is ready to receive the message from node m−1. If wake-up of node m fails, the message stays at node m − 1 that will initiate another wake-up attempt at a later time. Figure 4.24 depicts this case. wi,i,i+1

1 − pq 5

pq 5 wi,i+1,i+2

Figure 4.21. T-ROME meta model for node i attempting to wake-up node i + 1. The message is at node i.

wi,j,j+1

pq 5 wi,j+1,j+2

1 − pq 5 Ti,j

Figure 4.22. T-ROME meta model for node j attempting to wake-up node j + 1. The message is at node i.

wi,m−1,m

pq 5 Ti,m

1 − pq 5 Ti,m−1

Figure 4.23. T-ROME meta model for node m − 1 attempting to wake-up node m. The message is at node i. In case of data transmission, there exist following two possibilities:

4.2. T-ROME

65

wm−1,m−1,m

1 − pq 5

pq 5 Tm−1,m

Figure 4.24. T-ROME meta model for node m − 1 attempting to wake-up node m. The message is at node m − 1. Node i tries to transmit the message to node j, for i < j < m. If it succeeds, node j has the message and tries to wake-up node j + 1. If it fails, the message stays at node i that will initiate another wake-up attempt at a later time. Figure 4.25 depicts this case. Node i tries to transmit the message to node m, for i < m. If it succeeds, node m has the message. If the transmission fails, the message stays at node i that will initiate another wake-up attempt at a later time. Figure 4.26 depicts this case.

Ti,j q2 wj,j,j+1

1 − q2 wi,i,i+1

Figure 4.25. T-ROME meta model for node i attempting to transmit data to node j.

Ti,m q2 success

1 − q2 wi,i,i+1

1

Figure 4.26. T-ROME meta model for node i attempting to transmit data to node m. The meta-models shown in Figures 4.21 to 4.26, are composed of several Markov states as depicted in Figures 4.27 and 4.28. It can be seen in both Figures, that there exists a certain probability of success, but the attempts can also fail. In that case,

66


a node enters a fail state that is exited with probability 1 but has a certain delay connected to it. The delay just equals the timeout of the radio which is little more than the time required for the success case. It suffices to show the Markov chains for the cases Ti,j and wi,j,j+1 , as the chains for the cases Ti,m and wi,i,i+1 , wm−1,m−1,m and wi,m−1,m are similar and can be achieved by plugging the Markov model into the corresponding meta model shown above.

Ti,j

1 txij

f ail1

1−q

q

delay1

ackij

1−q

f ail2

delay2 q

1 1

wj,j,j+1

wi,i,i+1

Figure 4.27. Markov chain for node i attempting to send data to node j

Analysis The Markov Models can be analyzed with respect to the expected required time to send a messages via m nodes. From Figures 4.27 and 4.28 the expected times can be extracted for all Ti,j and wi,j,j+1 states. E[Ti,j ] can be expressed by Equation (4.1): E[Ti,j ] = q 2 (Ttx1 + E[wj,j,j+1 ]) +q(1 − q)(Ttx2 + E[wi,i,i+1 ])

(4.1)

+(1 − q)(Ttx3 + E[wi,i,i+1 ]).

Here, Ttx1 to Ttx3 are the times required to send the required communication packets in case of success (Ttx1 ), or the delay times required in case of failure (Ttx2 and Ttx3 ). E[wj,j,j+1 ] and E[wi,i,i+1 ] depict the expected times required in the corresponding TROME meta states which are given in Equations (4.2) to (4.3), below and can be extracted from Figure 4.28: E[wi,j,j+1 ] = pq 5 (Tw1 + E[wj,j+1,j+2 ]) +p(1 − q 5 )(Tw2 + E[Ti,j ])

(4.2)

+(1 − p)(Tw3 + E[Ti,j ]).

Here, Tw1 to Tw3 are the times required to send the required communication packets in case of success (Tw1 ), or the delay times required in case of failure (Tw2 and Tw3 ). As

4.2. T-ROME

67

wi,i,i+1

1 wuc

f ailw

1−p

p wuc ack

1−q

q

f ail2

delayw

1−q delay2

r req 1−q q

ack

1−q

q r req ack 1−q q

ack 1

q wi,i+1,i+2

1

wi,i,i+1

Figure 4.28. Markov chain for node i (that also has the message to be delivered) attempting to wake-up node i + 1 this is the general case of node j attempting to wake-up node j + 1 and the message is still at node i, the message will be send to node j in case of failure (Figure 4.22). Looking at the case where node i has the message and attempts to wake-up node i + 1, leads to Equation (4.3): E[wi,i,i+1 ] = pq 5 (Tw1 + E[Ti,i+1 ]) +p(1 − q 5 )(Tw2 + E[wi,i,i+1 ])

(4.3)

+(1 − p)(Tw3 + E[wi,i,i+1 ]).

Finally, the cases where node m − 1 attempts to wake-up node m, and the case where node i attempts to send data to node m needs to be considered. These two cases are given by Equations (4.4) and (4.5) for the wake-up and communication cases,

68


respectively: E[wm−1,m−1,m ] = pq 5 (Tw1 + E[Tm−1,m ]) +p(1 − q 5 )(Tw2 + E[wm−1,m−1,m ])

(4.4)

+(1 − p)(Tw3 + E[wm−1,m−1,m ])

and E[Ti,m ] = q 2 (Ttx1 ) + q(1 − q)(Ttx2 + E[wi,i,i+1 ])

+(1 − q)(Ttx3 + E[wi,i,i+1 ]).

(4.5)

Equations (4.1) to (4.5) are a set of linear equations, which can be solved for certain Ttxi and Twi for i = 1, 2, 3. Naive Algorithm The naive algorithm wakes up and transmits data from node to node. Here, it is assumed following communication scheme: node i sends a wake-up call to node i + 1 directly followed by the data packet. Node i + 1 acknowledges the data packet if it was received successfully. Figures 4.29 to 4.31 show the corresponding Markov models. wi,i,i+1

1 − pq 2

pq 2 Ti,i+1

Figure 4.29. Meta model for node i attempting to wake-up to node i + 1 using the naive algorithm.

1 wuc

wi,i,i+1 1−p

f ailw

delayw p

Ti,i+1

1 wi,i,i+1

Figure 4.30. Markov chain for node i (that also has the message to be delivered) attempting to wake-up node i + 1 using the naive algorithm. Analysis of the models can be done similar to the analysis of T-ROME shown above in Section 4.2.4.

4.2. T-ROME

69

1

Ti,i+1

txii+1

delay1

q

ackij

f ail1

1−q

1−q

f ail2

delay2 q

1 1

wi+1,i+1,i+2

wi,i,i+1

Figure 4.31. Markov chain for node i attempting to send data to node i + 1 using the naive algorithm CTP-WUR CTP-WUR protocol for unicast type packets is performed as introduced in [61]. As it is possible to relay a message along one node, the CTP-WUR algorithm mainly consists of the meta models shown in Figures 4.21 (for single-hop transmissions) and 4.22 (for relaying, and: j = i + 1). The Markov chain for wake-up is similar to the one shown in Figure 4.28 but it is followed either by transmission (Ti,i+1 ) in case of single-hop, or by the next wake-up (wi,i+1,i+2 ) in case of relaying. As the states are very similar to the already introduced models, they are not presented here in detail. Also, the system of linear equations is similar to that represented by Equations (4.1) to (4.5) above, and can be obtained easily by using the same techniques.

4.2.5

Results

Model Verification To verify the models introduced in Section 4.2.4, several experiments were conducted consisting of two, three, four, five and six nodes as shown in Figure 4.32 that shows the deployment of the sensor nodes during the test setups. As all tests were performed indoors, the nodes were placed at a distance of 1 m to each other and the output power of the main radio was adjusted to - 6 dBm and for the wake-up radio to 0 dBm.

node 1

node 2

node 3

node 4

node 5

node 6

Figure 4.32. Placement of the sensor nodes during model verification.

70


Node 1 was always the source node and nodes 2 to 6 were either relay or sink nodes, depending on the experiment. In the experiment, each node could wake-up only its direct predecessor, that is, node 1 could wake-up node 2, node 2 was able to wake-up node 3 and so on. With respect to data communication, each node could communicate to each other (if awake). In the first experiment, node 1 (source) had one message consisting of 100 byte to deliver to the sink (node 2 to node 6). In the second experiment, the sink had 5 messages each consisting of 100 bytes to deliver. In each experiment T-ROME was used and the required time until the message was delivered at the sink was measured. Figure 4.33 shows the experimental data for 1 message as crosses and for 5 data packets as pluses. The curves in Figure 4.33 show the expected times by using the T-ROME Markov models of Section 4.2.4 above, assuming p = q = 1. Figure 4.33 show that both, the expected and the measured times correspond very well.

Figure 4.33. Simulated time required to send 1 data packet (dashed) and 5 data packets (solid) along several nodes assuming p = q = 1. The points are data taken from the test setup

Performance Analysis After model verification, T-ROME was compared to CTP-WUR and the naive algorithm using the models introduced in Section 4.2.4. Figure 4.43 shows the simulation results with respect to the required time to send 1, 2 and 5 data packets along several nodes, assuming p = q = 1. All results are compared to the performance of the naive algorithm. It can be seen that CTP-WUR performs equally good as the naive algorithm in case of two participating nodes and performs about a constant ratio better than the naive algorithm for more than two participating nodes. This behavior is expected since CTP-WUR uses the same communication messages as the naive algorithm and saves

4.2. T-ROME

71

a constant amount of time by using one relay node. For both algorithms (naive and CTP-WUR), the ratio stays constant, regardless of how many data packets are sent. It can be seen from the results presented in Figure 4.43 for sending one data packet that T-ROME requires more time than the naive algorithm in case of two participating nodes due to the additional messages required in the protocol. In case of four participating nodes, T-ROME performs equally good as the naive algorithm and for more than four nodes, it outperforms it but it does not reach the performance of CTP-WUR. However, T-ROME outperforms CTP-WUR and the naive algorithm, when delivering two or more data packets with two or more participating nodes due to the savings of relaying.

Figure 4.34. Simulated time performance of T-ROME and CTP-WUR compared to the naive algorithm, assuming p = q = 1. Black: T-ROME for 1 (dotted), 2 (dashed) and 5 data packets (solid), red: naive algorithm and blue: CTP-WUR. Figure 4.35 shows the time performance comparison for p = 0.75 and q = 0.97. It can be seen that the differences between the naive and the other (T-ROME and CTP-WUR) are getting smaller and that T-ROME outperforms CTP-WUR and the naive algorithm only when more than one data packet is sent. The sensitivity of TROME originates from the larger amount of communication packets as required by the protocol. Additional issues to consider when analyzing the performance of T-ROME are opportunistic routing approaches and route adjustments based on link quality estimation that are used in some of the wake-up protocols introduced in Section 4.1. As these parameters are not yet implemented in T-ROME, the performance of T-ROME could decrease due to an increased number of packet retransmissions. It can be expected that T-ROME gains a similar or even bigger advantage from these features, compared to other protocols as T-ROME provides additional flexibility during candidate coordination.

72


Figure 4.35. Simulated time performance of T-ROME and CTP-WUR compared to the naive algorithm, assuming p = 0.75 and q = 0.97. Black: T-ROME for 1 (dotted), 2 (dashed) and 5 data packets (solid), red: naive algorithm and blue: CTP-WUR.

Energy Budget Figure 4.36 shows the time, energy and power shares of the source node (node 13) of the proposed protocol calculated with the numbers provided above (TTL = 3, 5 data packets each 100 byte large). It can be seen that the wake-up call requires only 7 % of the time and delay, receive and send share around a third of the time. Looking at the energy shares of each state, it is evident that sending and receiving require most of the energy but the wake-up call still needs almost a fifth of the total energy. Power consumption during wait periods is less than 10 % of the total amount. But due to the high current required during sending wake-up messages, more than 50 % of the allocated power is consumed. Table 4.1 shows the energy consumption of each node using the proposed protocol for a network consisting of four nodes and TTL=3. It can be seen that the source node requires most energy and the sink node fewest, as it does not send a wake-up message and has no delay states. Here, the delay is the time required by the sender to give the forwarding nodes time to wake-up their neighbors. So the delay consists of the times required to send the packets WUC, WUC ACK, R REQ, and ACK. Figure 4.37 shows the simulated energy performance analysis of CTP-WUR and T-ROME compared to the performance of the naive algorithm, for p = q = 1. It can be seen that the energy performance is similar to the time performance. CTP-WUR outperforms the naive algorithm for a communication with more than two nodes. Due to relaying, T-ROME outperforms the naive protocol and CTP-WUR as soon as more than one data packet is sent.

4.2. T-ROME

73

100 33

60

26

41

Wake-up Delay Receive Send

17 5

27 34

40

0

7

52

8 17

En

T

er

Po we r

33

gy

20

im e

Percent

80

Figure 4.36. Time, energy and power requirements of the sender node. Table 4.1. Energy consumption of a network of four nodes (TTL = 3) in mJ of each node for states sending WUC (wake-up call), delay, receive data and transmit data at 3.3 V, when transmitting 500 byte in 5 data packets (100 byte per packet) directly from source to sink. node ID WUC [mJ] Delay [mJ] Receive Send [mJ] total [mJ] [mJ] 13 (source) 12 (relay 1) 11 (relay 2) 10 (sink)

0.6 0.6 0.6 -

0.3 0.1 -

0.8 0.7 0.7 1.5

1.6 0.3 0.3 0.6

3.3 1.7 1.6 2.1

Overhead By using the control overhead ratio OCD as the ratio of control bit sent (CBS ) over data bit delivered (DBD ) (OCD = CBS /DBD ), T-ROME and the naive algorithm can be analyzed with respect to protocol overhead. Figure 4.38 shows OCD over data for transmission of one data packet. It can be seen that transmission of few bytes requires a large overhead due to a large amount of byte required for the wake-up call (162 bytes). It can also be seen that the naive algorithm requires less overhead than T-ROME due to a fewer number of control packets. The naive algorithm reaches the brake-even point at 165 bytes and T-ROME at 192 bytes. Figure 4.39 shows OCD plotted against number of sent packets. As T-ROME requires almost no further control byte after a link is established, ODC decreases quickly for sending of more than one data packet. In the case of the naive algorithm, OCD

74


Figure 4.37. Simulated energy performance of T-ROME and CTP-WUR compared to the naive algorithm, assuming p = q = 1. Black: T-ROME for 1 (dotted), 2 (dashed) and 5 data packets (solid), red: naive algorithm and blue: CTP-WUR.

Figure 4.38. T-ROME overhead (blue) compared to the overhead of the naive algorithm (black) for sending one data packet stays constant as every packet requires the same amount of control byte.

4.2.6

Discussion

Mostly, the goal of a wireless sensor network is to deliver sensor data from a source to a sink. A routing protocol determines the path a messages takes on its way from the source to the sink. Caused by the effects discussed in Section 2.1, and by node energy constraints, communication in wireless sensor networks is usually considered to be unreliable [72] and nodes can get unavailable to other nodes. A routing network should be able to cope with these challenges and provide alternative routes in case the intended route gets inaccessible.

4.2. T-ROME

75

Figure 4.39. T-ROME overhead (blue) compared to the overhead of the naive algorithm (black) for sending 64 data packets

In view of this discussion the static routing table utilized by the herein presented routing protocol is a clear drawback. If a node along the path from source to sink gets unavailable T-ROME cannot find another route. Although, the following Section 4.3 presents a method to mitigate disconnections in certain cases, what clearly improves the routing behavior, a disconnected network is still possible. Furthermore, in case of node mobility it is in many cases impractical to implement the updated routing table on each node. Future research should therefore focus on replacing the static routing of T-ROME by a dynamic routing, for instance as proposed in RPL (Routing Protocol for Low-Power and Lossy Networks) [72] or as suggested by [66]. The aim is to build a logical routing topology over the physical network layout. In RPL [72], the process is initiated by the sink utilizing a DIO (directed acyclic graph information object) message. Depending on certain rules, a recipient of a DIO message can either connect itself to the root to become a leaf-node, or it starts to transmit a DIO message by its own. The resulting topology is a tree structure in which every node has a rank that represents the costs from the node to the sink. In case of changes in the logical network topology, a node can trigger a local repair mechanism to discover new routes in direction to the sink [72]. Applying a dynamic routing topology as just discussed will certainly increase the reliability and robustness of the network, but still disconnected networks can happen, in the simple case when there is no other node available in wake-up range. Building upon the ideas presented by Bannoura [66] a solution to mitigate this scenario becomes apparent as depicted by following example. Assuming the fictive scenario where node A that was earlier linked to node C via node B lost its connection to the network by an outage of node B and there is no other node in wake-up range. If nodes A and B have the means to realize that they are disconnected, they could begin to find each other

76


by applying a duty-cycling scheme similar to that sketched in [66]. Of course, node A could also be disconnected due to an outage of itself or due to node movement. The algorithm should try to consider this.

4.2.7

Conclusions

An energy efficient and simple cross-layer network protocol (T-ROME) for wireless wake-up sensor networks was presented, comprehensively modeled and analyzed in this section. The protocol combines the advantages of wake-up receivers such as lowpower consumption and on-demand communication together with the advantage of long-range communication radios, that is their superior sensitivity. T-ROME makes use of the different communication ranges of communication and wake-up radio. The protocol saves energy by skipping nodes during data communication. Furthermore, T-ROME introduces a set of parameters to optimize the relaying process by dynamically choosing the most appropriate stopover nodes in case the sink is not reachable within one communication hop. The total number of wake-up packets can be reduced with T-ROME by accumulating sensor data and sending up to 64 data packets (16 kbyte) in a row once a communication link is established. Based on Markov chain models, the performance of T-ROME was compared to CTP-WUR and to a naive communication algorithm, that are state-of-the-art communication protocols for wake-up receivers. It could be demonstrated that T-ROME outperforms both protocols in many cases, particularly by utilizing an existing link to transmit several data packets and by skipping nodes during communication. As discussed, future research should focus on replacing the static routing by a dynamic routing algorithm.

4.3

T-ROME in Noisy Environments

It is clear that each communication attempt is connected to a certain probability of failure or success. This probability is obviously connected to the number of messages send during a certain communication attempt. The more messages are sent, the higher the probability of failure. Observing the design of T-ROME reveals that the required number of messages to establish a route in T-ROME can be reduced by omitting several acknowledge packets. As there are fewer packets to be sent, the probability of a successful rout establishing increases, especially in a noise environment. The modifications do not alter the design principles of T-ROME. The core of the improvements is the reduction of the MAC acknowledge messages during the handshaking procedure. Resulting from the reduced amount of messages required to establish a route this simple modification reduces the probability of failure. Especially in an noisy

4.3. T-ROME in Noisy Environments

77

environment where messages are transmitted with low reliability. Consequently, this increases the overall performance of T-ROME. As a result, the Markov chain for node i (that also has the message to be delivered) attempting to wake-up node i + 1 changes compared to that introduced in Section 4.2.4. Figure 4.40 illustrates the modified model. Due to the modification, the probability factor q = 5 in the meta-models depicted in Section 4.2.4 alters to q = 3. As a consequence, Equations in Section 4.2.4 have to be adjusted. As the modification can be achieved by simply replacing q 5 by q 3 , they will not be shown here.

wi,i,i+1

1 wuc

f ailw

1−p

p wuc ack q

1−q

f ail2

1−q

delayw

delay2

r req q

1−q

r req ack q wi,i+1,i+2

1

1

wi,i,i+1

Figure 4.40. Meta model for node i attempting to wake-up to node i + 1 including the modification.

4.3.1

Routing

The routing in T-ROME is based on a static routing table stored on each node. This can lead to disconnected network branches in case a connecting relay node fails due to a temporary or permanent outage. A method is presented here to maintain a connection in the special case where a parent node of the disconnected node is in wake-up range and alive as sketched in Figure 4.41. Nodes of depth i can also wake-up nodes of depth i − n with n = 1, 2, ...T T L. This means if for example node 12 is unreachable due to changes in the link or due to low energy, node 13 is now able to wake-up one of

78


the parent nodes of node 12. As the wake-up range is limited, the number of possible parent wake-ups is also limited to TTL.

node 10 ack

data node 11

wake-up

depth 0 in wake-up range

node 21

depth 1

data wake-up node 12

in wake-up range

node 22

depth 2

wake-up ack node 13

in wake-up range

node 23

depth 3

Figure 4.41. Schematic of the wake-up multi-hop routing protocol developed in this work.

4.3.2

Experimental Results and Performance Analysis

Figure 4.42 visualizes the sending and receiving states of four wireless sensor nodes running the modified T-ROME algorithm. The test setup consisted of four nodes, a source, node 2 and node 3 as relay nodes and a sink. The source had 5 data packets of 100 bytes payload each to sent but node 2 was dead (simply switched OFF). Using a purely static approach this would lead to a disconnected source node but according to the above-introduced behavior, the source now detects the outage of its direct parent node and wakes up its predecessor node 3. Node 3 then forwards the wake-up and the T-ROME request to the sink that finally receives all 5 data packets. Comparing Figures 4.42 and 4.2, one can see the fewer required messages and time due to the omitted MAC acknowledgments. This of course also reduces the overhead of T-ROME. Figures 4.43 (a) and (b) compare the modified and the original versions of T-ROME using the updated models introduced in Sections 4.3 and 4.2.4. Figure 4.43 (a) shows the simulation results with respect to energy-ratio to send 1, 2 and 5 data packets along several nodes, assuming p = q = 1. All results are compared to the performance of T-ROME as presented in Section 4.2. It can be observed that the modified version outperforms the above introduced TROME, especially for sending only a few data packets. This is due to the fewer number of messages to be transmitted during each route establishment. The more packets are sent along an existing link, the less prominent gets the improvement because no new

4.3. T-ROME in Noisy Environments WUC

WUC

79

R_REQ

DATA

DATA

DATA

DATA

DATA

source send WUC ACK

LBT

LBT

WUC ACK

R_REQ ACK

R_REQ ACK

ACK

ACK

ACK

ACK

ACK

ACK

ACK

ACK

ACK

source receive

node 2 send

node 2 receive WUC R_REQ ACK ACK

WUC

R_REQ

node 3 send R_REQ

WUC R_REQ ACK DATA ACK

node 3 receive WUC R_REQ ACK ACK

ACK

sink send R_REQ

DATA

DATA

DATA DATA

DATA

sink receive

Figure 4.42. Sending and receiving of 5 data packets in case of 4 participating nodes. Each node (node 13, node 12, node 11 and node 10) has a sending (upper line) and receiving (lower line) state. Nodes 2 and 3 are relay nodes but node 2 is simulated to be dead. Node 3 forwards the wake-up call. handshaking is required and then, modified and original version of T-ROME are equal.

(a)

(b)

Figure 4.43. Performance of T-ROME with and without modification for 1 (dashed), 2 (dotted) and 5 data packets (solid). (a) p = q = 1 (b) p = 1 and q = 0.8. Figure 4.43 (b) shows the energy comparison for p = 1 and q = 0.8. It can be seen that the performance of the modified version of T-ROME further increases compared to that of the earlier version of T-ROME. This is again due to the reduced number of communication messages required to establish a link between two nodes, and as the probability of success is decreased the chance of a successful route establishment is higher the fewer messages are exchanged. Due to this, the modified version of T-ROME

80


outperforms its original version. Again, if more than one packet is to be transmitted along an existing link, the improvement of sending fewer packets during handshaking becomes less prominent.

4.3.3

Conclusion

This section introduced a small modification to the routing algorithm T-ROME presented in the previous Section 4.2. A small modification was suggested to save additional resources in terms of time, energy, and overhead especially in an noise environment. Additionally, a routing method was presented to mitigate disconnected nodes in cases when parent nodes are still alive and in wake-up range.

4.4

Exploiting Concurrent Wake-up Transmissions

As already discussed, their ultra-low power consumption and their ability to provide communications on demand make wake-up receivers an interesting option for wireless sensor networks. A downside of wake-up receivers is their low sensitivity caused by the passive demodulation of the carrier signal. This section presents a novel communication scheme by exploiting concurrent wake-up signals using the beat frequencies of two or more wireless sensor nodes. Additionally, a communication algorithm and a flooding protocol based on this approach are introduced and the approaches are demonstrate experimentally. It is shown that the received signal strength increases up to the expected 3 dB and as such improves communication robustness and reliability. Furthermore, the feasibility of the newly developed protocols by means of an outdoor experiment and an indoor setup consisting of several nodes, is demonstrated. The flooding algorithm achieves almost 100 % wake-up rate in less than 20 ms.

4.4.1

Introduction

Resulting from the limitations introduced by the low to ultra-low power consumption, the sensitivity of wake-up receivers is usually lower than that of state-of-the-art radio transceivers. In cases of fading channels, it can happen that a wireless sensor node gets out-of-reach of any other sensor node, for example, due to fading simply caused by changes in the environment or in the network topology that increase path losses. In these cases, a sensor node might get temporarily or permanently disconnected from the wireless sensor network, although it is still alive. Figure 4.44 shows this scenario where node-3 is out of wake-up range of either node-1 and 2 due to an outage of node-4. To improve reliability and to reduce packet losses, some existing network protocols [73] use concurrent packet transmissions that interfere in a constructive way. The

4.4. Exploiting Concurrent Wake-up Transmissions

81

Wake-up Range

Sender node-2 node-1 node-4 node-3

Figure 4.44. Node D is out of wake-up range due to an outage of node C. feasibility of concurrent transmissions is demonstrated mainly in the area of IEEE 802.15.4 compliant radios. The IEEE 802.15.4 protocol maps 4-bit symbols to 32-bit pseudo-noise sequences that have a length of 0.5 µs at 868 MHz. In order to achieve constructive interferences, this imposes a time constraint of 0.5 µs which is not easily accomplished. This section presents a novel communication protocol to send wake-up messages by purposefully interfering signals of two or more wireless sensor nodes, which produce the wake-up signal as the beat frequency of superposed slightly out of tune carriers. The method will be investigated theoretically and its feasibility will be demonstrated with the help of several experimental setups. The rest of this section is structured as follows. Section 4.4.2 briefly introduces the principles of concurrent transmissions and furthermore introduces the main features of Zippy [60], a flooding protocol that is based on concurrently transmitted wake-up packets. Then, concurrent wake-up messages are discussed theoretically in Section 4.4.3 followed by two newly introduced communication protocols that are based on concurrent wake-up messages in Section 4.4.6. Section 5.2 presents the experimental results and Section 5.3 provides some conclusions.

4.4.2

Wireless interference

Unintentional interference is often called collision and occurs usually when two or more nodes start to send at the same time and a receiver node detects the signals from both transmitters. But although the packets collide, it is possible in some cases to correctly receive the strongest signal which is usually called capture effect [74, 75]. Intentionally generated interference can lead to constructive interference that can be used to improve link quality or to decrease the transmission energy required by a certain node. Usually, protocols that are based on constructive interference can be found in the area of flooding protocols as they achieve very low latencies due to the concurrently transmitted packets. Dutta et al. present in [76] Backcast, a link-layer protocol that relies on identi-

82


cal and concurrently transmitted acknowledgments as answers to a so-called probe. A probe is sent by an initiator and each node that receives this probe sends an identical acknowledgment exactly 192 µs after receiving the probe. This means of course, that acknowledgments from several responders might reach the initiator concurrently. According to Dutta et al., the observed robust detection probability of the acknowledgments cannot be explained only by the capture effect. One of the first flooding protocols in IEEE 802.15.4 compliant wireless sensor networks that use constructive interference is Glossy [73]. IEEE 802.15.4 uses direct sequence spread spectrum modulation by grouping binary data into 4-bit symbols that are mapped to one of 16 possible and nearly orthogonal 32-bit pseudo-noise sequences [77]. The length of one pseudo-noise sequence is 0.5 µs. This also reflects the maximum temporal displacement of two simultaneously transmitted messages allowed in Glossy to achieve constructive interference [73]. This demonstrates and Ferrari et al. [73] also conclude that constructive interference is only possible because of the 802.15.4 modulation scheme and the redundancy generated by the pseudo-noise sequences. There exist several works that are based on or build upon and analyze the performance of Glossy, such as [78–84]. These works have all in common that they are based on radios that use the IEEE 802.15.4 standard and constructive interference of radio packets can be achieved within the synchronization limit of 0.5 µs. But a synchronization of 0.5 µs is still very demanding. In Glossy these temporal constraints are met by synchronizing message transmissions based on signals from the radio to the micro-controller that indicate the reception of a complete packet, and by carefully balancing the number of required clock cycles before a transmission begins. There do not exist many constructive interference based protocols that are based on other modulation schemes than direct sequence spread spectrum. This is obviously due to the limitations posed by the timing constraints. Zippy, recently presented by Sutton et al. [60] is a flooding protocol that uses wake-up transceivers to synchronize neighboring nodes and to rapidly disseminate data. To ensure the timely waking of all wireless sensor nodes, Zippy transmits concurrent wake-up packets. As these packets are on-off-keying modulated and the wake-up receivers are of low complexity to ensure ultra-low power consumption, destructive interference has to be avoided and the capture effect cannot be used. Zippy solves destructive interference by using carrier frequency randomization as will be further discussed in the following paragraph.

Zippy Zippy, introduced by Sutton et al. [60], is a flooding protocol that uses wireless sensor nodes with wake-up receivers without address correlation. The wake-up receiver tests the frequency of the demodulated signal and in case it matches 125 kHz it interrupts the


83

main controller from deep sleep. As illustrated in Figure 4.45, the node then transmit a wake-up call, here called preamble, by itself. Figure 4.45 schematically shows a 2-hop Zippy network consisting of initiator node A and participating nodes B and C [60]. As node A continues to send its wake-up packet after nodes B and C are awake all three nodes transmit part of their wake-up packets concurrently to generate a 1-hop neighbor time synchronization in the order of milliseconds [60]. After sending the preamble, all nodes go to sleep state for a certain period of time that depends on the number of hops in a multi-hop network. Then, the initiator node A wakes up again to transmit a synchronization bit (SYNC) as visualized in Figure 4.45. Node A transmits the SYNC bit in the same matter as the preamble, by using the on-off-keying modulated carrier signal. This SYNC bit then is received by the wake-up receiver of node B. As the internal automatic gain control unit of node’s B wake-up receiver has already settled during preamble reception, the detection of the SYNC bit is achieved in the order of microseconds [60]. After receiving the SYNC bit of node A, node B itself starts to transmit a SYNC bit. This sequence then continues until no further nodes participate. Using this method, Zippy achieves a mean per-hop synchronization of 34 µs in a 2-hop network [60]. After achieving neighborhood synchronization, node A starts to transmit data using the same on-off-keying modulation scheme as before. To ensure fast relaying of the data, the data is transmitted bit-wise and not byte-wise. After node B receives a bit, it immediately forwards it to node C. Asynchronous Wake-up

Neighborhoodg Synchronization

Datag Propagation

CarriergFrequencygRandomization

1-bit S y n c

Preamble Initiator

A

Preamble Participant

B

Preamble Participant

C

TX

S y n c

0-bit

RX

S y n c

OFF OFF

TX

RX

TX

RX

RX

RX

RX

RX

Figure 4.45. Overview of Zippy. To mitigate destructive interference during concurrent transmissions, Zippy uses carrier frequency randomization [60]. Figure 4.46 illustrates exemplary an on-offkeying modulated message that implements carrier frequency randomization. Each low-frequency one-bit consists of several high-frequency periods which are transmitted at q different frequencies. In the example illustrated in Figure 4.46, each low-frequency

84


bit consists of q = 4 frequencies f that can be calculated by f = fc + β∆, with fc being the center frequency of 446.8 MHz, ∆ = 135.41 kHz resembling a minimum offset between two frequencies and β >= 2 being a uniformly distributed random variable. Using carrier frequency randomization, Sutton et al. experimentally verified a packet reception rate of almost 100 % by using six concurrent senders. Using an indoor testbed, Zippy achieves an end-to-end latency between 17.8 ms and 24.4 ms using 8 bit packets and between 29.8 ms and 41.6 ms for 16 bit packets. Here, end-to-end latency is the time elapsed between the start of flooding by the initiator and end of flooding at each participant [60]. An unsolved challenge of Zippy remains the existence of false positive wake-ups as Zippy wake-up receivers cannot use address correlation. 1

1

0

Tb /q

1

0

1

Tb

Figure 4.46. Schematic of an on-off-keying modulated message that implements highfrequency carrier signal frequency randomization.

4.4.3

Concurrent Wake-up Message

As introduced above, one of the major challenges of ultra-low power wake-up receivers is their low sensitivity. Due to this, they are susceptible to signal outages caused by fading channels and their input signal quickly gets below their sensitivity threshold. As shown in previous Section 4.4.2, constructive interference can improve the reliability of transmissions. When two senders transmit simultaneously the same message to a common receiver, constructive or destructive interference can happen. Constructive interference helps the common receiver to detect a message and improves the reliability of the transmission [79]. Considering the example given in Figure 4.47, node-1 and node-2 send simultaneously a wake-up message to node-3 to achieve constructive interference.

4.4.4

Two concurrent senders

In case of two concurrently sent wake-up packets, a possible and probable existing phase-shift φ of one of the signals has to be considered. There will be a superposed


node-2

85

node-3

node-1

Figure 4.47. Node-1 and node-2 send a concurrent wake-up to node-3. signal y(t) of the form of Equation (4.6): y(t) = A1 ejω1 t + A2 ejω1 t = (A1 + A2 )ejω1 t

(4.6)

with A1 and A2 ∈ C such that A1 = r1 ejφ1 and A2 = r2 ejφ2 with r1 , r2 ∈ R are the signal amplitudes and φ2 − φ1 the phase shift between the two signals. Furthermore, ω1 = 2πf1 is the angular frequency of the two superposed sinusoidal waves. But due to the periodicity of the sinusoidal signals, a destructive interference will take place in the range of 2/3π < φ2 −φ1 < 4/3π, which will finally lead to an unreliable communication. The superposed signal y(t) as introduced in Equation (4.6) above, is the sum of two sinusoidal signals having the same frequency f1 . The following Equations (4.7) to (4.9) all show the same sum of two sinusoidal functions having different frequencies f1 and f2 , but highlight different aspects. Equation (4.7) expresses y(t) as the sum of two sinusoidal functions with angular frequencies ω1 = 2πf1 and ω2 = 2πf2 : y(t) = A1 ejω1 t + A2 ejω2 t .

(4.7)

By replacing ω2 = ω1 + (ω2 − ω1 ) Equation (4.7) can be rewritten to achieve Equation (4.8): y(t) = ejω1 t (A1 + A2 ej(ω2 −ω1 )t ). (4.8) And with ω1 = ω − ∆ω and ω2 = ω + ∆ω Equation (4.7) can also be expressed as: y(t) = ejωt (A1 e−j∆ωt + A2 ej∆ωt )

(4.9)

From Equation (4.9), it can be seen that the amplitudes A1 and A2 of the original sinusoidal signals are preserved by the addition. By replacing ∆ω with (ω2 − ω1 )/2 and substituting ω with (ω1 + ω2 )/2 in Equation (4.9) it can be seen that the summation generates two new frequencies that are here called fc and fb . The first frequency, fc , equals (f1 +f2 )/2 and the second frequency, fb , equals (f2 −f1 )/2. The lower frequency fb is called beat, as it appears in Figure 4.48 that shows the frequency domain of y(t). Equations (4.8) and (4.9) also show that a phase-shift φ of one of the high-frequency signals, is a phase-shift of the beat-signal, but no destructive interference between A1 and A2 takes place. With the help of Equations (4.8) and (4.9) it was just demonstrated that a lowfrequency signal can be generated by adding two sinusoidal signals with f1 and f2 .

86

4. Communication with Wake-up Receivers A

f2

(f2 - f1)/2

(f1 + f2)/2 f1

F

Figure 4.48. Spectrum of the sum of two sinusoidal functions f1 and f2 and the resulting carrier and beat frequency. Now, assume that a wake-up receiver listens on a certain frequency fw , for example 125 kHz, for incoming messages. From the discussion above, it follows that fw can be generated by adding two sinusoidal signals with frequencies f1 and f2 that are just separated about the right amount from each other. In the example above with fw = 125 kHz this would be 75 kHz since the envelope detector, as long as it consists of two diodes, doubles the frequency during rectification. This wake-up frequency can now be further on-off-keying or frequency-shift-keying modulated to achieve a certain required wake-up pattern. Figure 4.49 depicts Equation (4.8) graphically, with the horizontal axis representing the real part and the vertical axis representing the imaginary part of y(t). It can be seen that there exist generally two cases: (a) |A1 | = |A2 | and (b) |A1 | > |A2 |. The case |A1 | < |A2 | is the same as case (b) which may be achieved by reordering A1 and A2 in Equation (4.8). Furthermore, it can be deduced from the two graphs (a) and (b) in Figure 4.49 that the beat signal has an amplitude of A2 and A1 resembles an offset that shifts the beat signal from the origin. Im

Im w2 A2 A1

A2

w2

A1 Re

Re

w1

w1

(a)

(b)

Figure 4.49. (a) A1 = A2 (b) A1 > A2 . Figure 4.50 visualizes Equation (4.8) in the time domain and fc = (f1 + f2 )/2 is shown as blue curve and fb is visualized in red. In Figure 4.50 (a) the two sinusoidal signals have equal amplitudes A1 = A2 = 1 and in Figure 4.50 (b) the amplitudes are A1 = 0.7 > A2 = 0.4. In both figures, the resulting beat (fb ) has amplitude 2A2 . The amplitude discrepancy between fb and fc visible in Figure 4.50 (b) shows the offset generated by A1 . Please note that the signals plotted in Figures 4.50 (a) and (b) were


87

obtained from frequencies f1 = 868 MHz and f2 = 868.125 MHz, although fc appears to have a lower frequency in the plots, which is due to massive undersampling that was done for reasons of illustration. 2

carrier frequency: (f1 + f2 )/2 beat frequency: 2A2 cos(2π(f1 − f2 )/2)

carrier frequency: (f1 + f2 )/2 beat frequency: 2A2 cos(2π(f1 − f2 )/2)

0.8 amplitude

amplitude

1 0

0.4 0

−0.4

−1 −0.8 −2 0

4e − 06

8e − 06 1.2e − 05 1.6e − 05 time in s

0

4e − 06

(a)

8e − 06 1.2e − 05 1.6e − 05 time in s

(b)

Figure 4.50. (a) Beat for equal amplitudes A1 = 1 and A2 = 1 (b) beat for A1 = 0.7 > A2 = 0.4. Figure 4.51 depicts the same wake-up signal as it appears at the input of the wakeup receiver after envelope detection. As discussed above, the signal has amplitude 2A2 and frequency 2fb . Figure 4.51 (b) also illustrates the dc-offset of A1 − A2 on the wake-up signal in accordance to the expectations from Equation (4.8). beat envelope A1

1

beat envelope A1

1.1 amplitude

amplitude

2

0.7

0.3 0 0

4e − 06

8e − 06 1.2e − 05 1.6e − 05

time in s

(a)

0

4e − 06

8e − 06 1.2e − 05 1.6e − 05

time in s

(b)

Figure 4.51. (a) Envelope for equal amplitudes A1 = 1 and A2 = 1 (b) envelope for A1 = 0.7 > A2 = 0.4. As already introduced in Section 3.1, wake-up receivers usually listen to the wireless channel in the kilohertz range, with a datarate of several kilobits per second. Consequently, timing constraints are in the µs-range. As the wake-up signal appears at the receiver like any other wake-up signal, there are no additional components required and the same hardware can be used to modulate and demodulate the beat signal.

88


4.4.5

More than two concurrent senders

In a wireless sensor network, it is probable that more than two senders are in wake-up range and superposition of more than two sinusoidal waves can occur. In Section 4.4.4 above it was discussed that it is possible to transmit a wake-up message from two concurrent senders having frequencies f1 and f2 = f1 + fw /2 and amplitudes A1 and A2 . The results obtained in Section 4.4.4 demonstrate that the amplitude of the larger signal, for example A1 , generates a constant offset on the wake-up signal. Consequently, if a third node has the same frequency as the node with the lower amplitude the wakeup signal can be increased, as shown by Equation (4.10) that enhances Equation (4.8) by a third sinusoidal signal with amplitude A3 and with frequency f2 : y(t) = ejω1 t (A1 + (A2 + A3 )ej(ω2 −ω1 )t ).

(4.10)

The signal y(t) expressed by Equation (4.10) is graphically reported in Figure 4.52. Note that the amplitudes in plots (a) and (b) are A1 = 0.8, A2 = 0.1 and A3 = 0.4. The frequencies in (a) are f1 = 868 MHz, f2 = 868.125 MHz and f3 = 868 MHz. In this case, where f3 = f1 and A1 + A3 > A2 , the amplitude A3 of the third node adds to the dc-offset, but not to the the beat. In Figure 4.52 (b) the frequencies are f1 = 868 MHz, f2 = 868.125 MHz and f3 = 868.125 MHz. In this case A3 adds to the beat and increases its amplitude. beat envelope A1 + A2

1

0.5

0

1.5

amplitude

amplitude

1.5

beat envelope A1

1

0.5

0

4e − 06

8e − 06

time in s

(a)

1.2e − 05

1.6e − 05

0

0

4e − 06

8e − 06

time in s

1.2e − 05

1.6e − 05

(b)

Figure 4.52. Envelopes for A1 = 0.8, A2 = 0.1 and A3 = 0.4. In (a), f1 = 868M Hz, f2 = 868.125M Hz and f3 = 868M Hz. In (b) f1 = 868M Hz, f2 = 868.125M Hz and f3 = 868.125M Hz. But due to inaccuracies of the crystal oscillators, frequency f3 of the third node will always be slightly different from f2 or f1 . Utilized radio crystals often have an accuracy of 10 to 40 ppm in the kilohertz range and Equation (4.10) has to be rewritten. Using ω1 = ω − ∆ω, ω2 = ω + ∆ω and ω3 = ω + α∆ω yields: y(t) = ejωt (A1 e−j∆ωt + A2 ej∆ωt + A3 ejα∆ωt )

(4.11)


89

where α indicates the frequency offset of the third node. Equation (4.11) is similar to Equation (4.9) and indicates that the frequency offset of the third node generates an additional beat (in the lower kilohertz range) that might disturb the wake-up signal. Additionally, the phase of the third node is probably shifted, compared to that of the other nodes and destructive interference can occur. Due to these limitations, each additional node will likely generate additional frequencies and offsets that make an undisturbed wake-up signal more unlikely, although several wake-up receivers [13, 51] have a build-in automatic gain control unit that can help to mitigate this effect. By expanding Equation (4.8) by an additional pair of nodes sending at 2πf3 = ω3 and 2πf4 = ω4 reveals Equation (4.12): y(t) = ejω1 t (A1 + ej(ω2 −ω1 )t (A2 + ej(ω3 −ω2 )t (A3 + A4 ej(ω4 −ω3 )t )))

(4.12)

Considering Equation (4.12) reveals that four concurrently sending nodes generate the desired superposed beat frequencies in case f2 −f1 = f4 −f3 . There is also an additional frequency generated by the offset of frequencies f3 − f2 . This offset is present due to oscillator inaccuracies. By setting f3 − f2 purposefully in the MHz-range these effects can be reduced, as the additional beats will also be in the MHz range and will be low-passed by the envelope detector. What is left are the inaccuracies of f4 − f3 not being exactly f2 − f1 . Equation (4.12) indicates, that n nodes can generate n/2 beat wake-up frequencies. But due to frequency inaccuracies, there will be superposed additional frequencies. Furthermore, mitigating the destructive interferences of the beats needs to be considered, what could be achieved by randomizing the frequencies of node pairs similar to the concept introduced in Zippy [60], although randomizing on node level, not on bit level. Figures 4.53 (a) and (b) visualize exemplary beats on the real and imaginary axis. Figure 4.53 (a) visualizes two sinusoidal signals with frequencies f1 and f2 = f1 + ∆f and with amplitudes A1 = 1 and A2 = 0.6. It can be seen that the amplitude of the first signal is larger than that of the second, in accordance to Equation (4.8) and Figure 4.49. Figure 4.53 (b) graphically depicts four sinusoidal signals with frequencies f1 , f2 = f1 + ∆f , f3 = αf1 and f4 = f3 + ∆f as given in Equation (4.12). Here, the amplitudes were A1 = 1, A2 = 0.6, A3 = 0.3, A4 = 0.2. The figure demonstrates the additional frequencies as expected from the discussion above. Figure 4.54 visualizes amplitudes A over number of nodes n for concurrently sending nodes as calculated by Matlab. The black solid curve are the averaged amplitudes Y (n) for n sinusoidal signals (transmitted by n nodes) all having the same frequency f , an random phase-shift φn , amplitude A = 1 averaged over i = 100 runs, as calculated by: i

n

1 XX max(A sin(f + φn )) Y (n) = i 1 1

(4.13)

imag


imag

90

real

(a)

real

(b)

Figure 4.53. Beats visualized on the real and imaginary axis. (a) two sinusoidal signals at f1 and f2 = f1 + ∆f (b) four sinusoidal signals with f1 , f2 = f1 + ∆f , f3 = αf1 and f4 = f3 + ∆f . where the function max() extracts the magnitude of the amplitude of each sinusoidal signal. The black dashed curve shows Y (15), exemplary for one sample run. The black dotted curve depicts the expectation value E(Y (n)). Assume the phases φ and amplitudes a of the sums of n sinusoidal signals are normally distributed random p variables with µ = 0 and σ 2 = n/2. Then R = φ2 + a2 is Rayleigh distributed with p p p E(R) = σ π/2. From σ = n/2 follows E(Y (n)) = πn/4. The red solid curve shows Yb (n) calculated with Equation (4.13) but each second signal had frequency f + ∆f to generate beat frequencies. All signals had a random phase-shift φn and amplitudes A = 1 and were averaged over i = 100 runs. The red dashed curve depicts the exemplary result of one exemplary run (Yb (15)) and the red p dotted curve illustrates the expectation value E(Yb (n)) = πn/2. The expectation value E(Yb (n)) increased by a factor of two as the variance σ 2 of the beat amplitudes also increased by a factor of two. Figure 4.54 does not consider frequency inaccuracies originating from the crystal oscillators, but it demonstrates the potential of concurrent wake-up signals. Performance simulation To analyze the performance of the concurrent wake-ups, Friis transmission equation as introduced in Section 2.1 was used to calculate the freespace transmission distances for the wake-up radio. The expected signal strength was simulated at the receiver over the distance from the receiver to sender as visualized in Figure 4.55. In the simulation, sender and receiver were placed 1.2 m above ground. The black curve gives the expected signal strength for a sending power of 0 dBm and the blue curve for 3 dBm, antenna gain was assumed to be 1.5 dBi. The dashed line shows an exemplary sensitivity threshold

4.4. Exploiting Concurrent Wake-up Transmissions 9 8 7

amplitude

6

91

Y (n) p E(Y (n)) = πn 4 Y (15) Yb (n) p E(Yb (n)) = πn 2 Yb (15)

5 4 3 2 1 0

5

10

15 20 number of nodes

25

30

Figure 4.54. Simulated and expected amplitude achieved by concurrently sending nodes. of a wake-up receiver at -51 dBm. The curves clearly show the advantages of additional sending power. The wake-up range increases and local minima are much less severe.

−20

sender: 3 dBm sender: 0 dBm sensitivity threshold

−25 power in dBm

−30 −35 −40 −45 −50 −55 −60

5

10

15

20 25 30 35 distance in m

40

45

50

Figure 4.55. Simulated received signal strength over distance for transmit power of 0 and 3 dBm.

4.4.6

Concurrent Wake-up Protocol Design

This section presents two prototype network protocols that were developed by applying concurrent wake-up messages based on beat frequencies. The first paragraph introduces a prototype algorithm that can be used to integrate concurrent wake-up messages into existing network protocols, in order to increase their reliability or to decrease the required sending power. The second paragraph introduces a prototype wake-up

92


flooding mechanism based on concurrent wake-up messages to achieve a quick and reliable waking up of several sensor nodes. Integration into existing routing protocols Depicted in Figure 4.56 is a protocol design to demonstrate the feasibility of concurrent wake-up calls and to provide a means to integrate concurrent wake-up calls into existing solutions. To begin a concurrent wake-up, node-1 sends a single wake-up call (WUC) to node-2 that acknowledges (ACK) the successful reception. Node-1 then transmits a request (REQ) for a concurrent wake-up message to node-2 embedding the address of node-3. Upon reception of this dedicated request, indicated by a special bit in the radio packet, node-2 changes its carrier frequency for the transmission period of the concurrent wake-up packet. Based on the timing of this handshaking, nodes-1 and 2 send the concurrent wake-up call (CWUC) to node-3. After node-3 detects the valid wake-up call it is ready to receive data from node-1. Finally, node-3 acknowledges the data reception. This example, assumed a scenario similar to that visualized in Figure 4.44, where node-3 is out of wake-up range. Alternatively, this scheme could also be used when wake-up messages reach node-3 with low reliability, only. Both cases could be detected by node-1 for example, by comparing the number of send wake-up calls to the number of successfully received wake-up calls. In case of low wake-up reliability, node-1 could request node-2 to transmit a concurrent wake-up message to increase its reliability. This basic concurrent wake-up scheme can be integrated into existing routing protocols, like T-ROME as presented in Section 4.2 or CTP-WUR [61], ALBA-WUR [47] or others such as [7, 63].

{

DATA

CWUC

REQ

main radio

CWUC

wake-up radio

wake-up radio

main radio

ACK

node-3

{

main radio

ACK

node-2

{

WUC

node-1

wake-up radio

Figure 4.56. Concurrent wake-up protocol.

Concurrent wake-up flooding Wake-up messages are usually energy-expensive and require a certain amount of time to be sent. In cases of large wireless sensor node deployments, waking-up of all sensor


93

nodes can as such, pose a considerable amount of energy and time to the wireless sensor network. But waking up of all nodes is required for example during the initialization phase of a wireless sensor node deployment or more generally required for network configuration tasks and in cases when data needs to be pulled out of a complete network. In such cases, it is possible to use flooding algorithms, as they provide a tool to quickly broadcast messages throughout a complete wireless sensor network. This paragraph introduces a concurrent wake-up flooding protocol similar to Zippy as introduced by Sutton et al. in [60] and as reviewed in Section 4.4.2. In comparison to Zippy, our newly developed protocol uses both radios, the main and the wake-up radio. The main radio is used to transmit the on-off-keying modulated wake-up calls (WUC), the concurrently transmitted wake-up calls (CWUC) and the synchronization (SYNC) messages. Figure 4.57 schematically illustrates the concurrent wake-up flooding protocol. An initiator node a starts the flooding by sending a normal wake-up message. Each receiver node of this seed turns its radio on and listens to the wireless channel for a synchronization packet (SYNC). Based on the timing provided by the SYNC packet, the nodes transmit a concurrent wake-up packet (CWUC) using the beat frequencies. The initiator retransmits the SYNC packet after each wake-up packet, in order to provide timing information to the newly woken nodes. The algorithm finishes at a predefined maximum number of concurrent wake-up packets. An advantage of generating concurrent wake-up packets from beats is the possibility to encode addresses in wake-up packets, which is not possible in Zippy. Due to the address correlation, the number of false positive wake-ups can be greatly reduced as investigated in [85]. Furthermore, the strength of the beat signal increases due to the concurrency as shown in previous Section 4.4.3, which leads to a more robust and reliable wake-up reception. Synchronizing on the main radio signals provides improved timing.

{

CWUC

SYNC

CWUC

SYNC

main radio

CWUC

wake-up radio

wake-up radio

main radio

Figure 4.57. Concurrent wake-up flooding protocol.

CWUC

node c

{

main radio

CWUC

node b

{

WUC

node a

wake-up radio

94

4.4.7



This section presents the experimental results that were performed to verify the assumptions taken above. Throughout the experiments, wireless sensor nodes were utilized as introduced in Section 3.1.

4.4.8

Expected concurrency of two senders

To further analyze the concurrency of the wake-up packets the time delays between two concurrent packet transmissions were measured. As receiver and transmitter have different internal tasks to perform after packet transmission and to prepare to send a wake-up packet, several nop() cycles were inserted to adjust the wake-up packets in time. Figure 4.58 plots the occurrences over delta time td in µs. A Gaussian fit of the form y = a exp(−(x − x0 )2 /(2s2 )) was applied to evaluate mean td = −1.4 µs and standard deviation s = −3.1 µs. The results of Figure 4.58 show that this method achieves a timing that is sufficient to send concurrent wake-up packets using the beatfrequency at a datarate of 8192 kbps as required by the wake-up receiver that was utilized in this work. 18 16

time offset gaussian fit

14

count

12 10 8 6 4 2 0 −10

−5 0 5 time offset in micro-seconds

Figure 4.58. Distribution of time offsets between two concurrent wake-up packets.

Concurrent wake-up signals Figure 4.59 shows exemplary concurrent wake-up signals taken at a receiver node. The figures show normalized signal amplitudes over time for the cases of (a) two, (b) four and (c) six concurrent senders. As the signal amplitude is strongly connected to the distance between sender and receiver, the signal amplitudes are normalized to the range from 0 to 1, in all cases. The single wake-up packet is transmitted in the time period from 0 to 6 ms. After approximately 10 ms, the synchronization packet is transmitted and the concurrent wake-up packets are transmitted in the time period from 14 to

1

normalized amplitude



0.8 0.6 0.4 0.2 0 0

0.005

0.01

1 0.8 0.6 0.4 0.2 0 0

0.015

time in s

95

0.005

0.015

(b)


(a)

0.01

time in s

1 0.8 0.6 0.4 0.2 0 0

0.005

0.01

time in s

0.015

(c)

Figure 4.59. Normalized signal amplitudes received in case of (a) two, (b) four and (c) six concurrent senders.

20 ms. Nodes 1 to 6 sent at an output power of 0 dBm and were placed approximately at a distance of 1 m around the receiver. Nodes 1 and 2 sent with 868 and 868.125 Mhz, nodes 3 and 4 sent with 866 and 866.125 MHz and nodes 5 and 6 sent with 865 and 865.125 MHz. It can be seen, that in the case of two senders, the signal shape of the concurrent wake-up packet follows the shape as expected in Section 4.4.4. The beat frequency with amplitude A2 is on top of a dc-offset generated by the amplitude A1 − A2 . In cases of four and six concurrent senders, the signal follows the same shape but also includes additional low frequencies that are probably caused by offsets of the 125 kHz as discussed in Section 4.4.5. But although the signal amplitudes are not constant over the duration of the wake-up packet, the receiver is able to detect them as valid wakeup messages. To further analyze the impact of additional nodes and to investigate the concept of frequency randomization further test are required what should be subject to future research.

96


Freespace transmission To verify the concurrent wake-up, two test setups in freespace were performed. The first setup consisted of a single sender and receiver, both in a height of about 1.2 m. The second test consisted of three sensor nodes. Two nodes, placed in distance x = 1 m to each other, sending a concurrent wake-up message. The third node received the concurrent wake-up calls. Both tests consisted of 500 wake-up messages that were sent with a transmit power of 0 dBm. The receiver was placed at different distances d in the range from 1 to 50 m and counted each successfully received wake-up message. Figure 4.60 shows schematically the experimental setups. sender

receiver d

sender 1 x

receiver d

sender 2

(a)

(b)

Figure 4.60. Schematics of the experimental setups for (a) single wake-up and (b) concurrent wake-up.

Figure 4.61 depicts the number of successfully received wake-up packets over distance. The black dashed curve illustrates the single sender case and the red curve visualizes the data for the concurrent wake-ups. Both curves correspond well to the simulated freescpace transmissions reported in Figure 4.55, although the local minimum was located at around 15 m distance for the single wake-up calls and according to simulation was located at a distance around 8 m. This discrepancy could be caused by a differing ground reflection coefficient from simulation to experiment. The receiver successfully detected each single wake-up message (black curve) until a distance of around 20 m, then the wake-up rate decreased to zero at 30 m. Between 30 and 35 m the receiver started again to successfully detect wake-up messages that probably originated from additional multipaths generated by surrounding trees or buildings. The red solid curve in Figure 4.61 visualizes the amount of successfully received wake-up messages for the concurrent wake-up calls. The wake-up range increased as expected. Additionally, the minimum at around 15 m distance to the senders, did not exist. The receiver successfully detected each concurrent wake-up call up to a distance of around 25 m. Then, the number of received packets decreased but not to zero and increased again at around 30 m distance. Located at distances of 40 and 45 m, were two additional maxima that probably also originated from additional multipath propagations due to scattering effects of surrounding trees, buildings, etc.


concurrent wake-up single wake-up

600 wake-ups received

97

500 400 300 200 100 0

5

10

15

20 25 30 35 distance in m

40

45

50

Figure 4.61. Number of successfully received wake ups over distance between sender and receiver for single and concurrent wake-ups. Concurrent wake-up protocol To verify the protocol described above, the T-ROME protocol [85] was enhanced by the concurrent wake-up scheme described in Section 4.4.6 and implemented it on three wireless nodes to verify its performance. The nodes were placed similarly to the scenario depicted in Figure 4.44. Node-2 could receive wake-up messages from node-1 but was not able to receive them from node-3. Node-2 had data to be sent to node-3. Figure 4.62 visualizes the states (send or receive) of the three nodes. Node 2 started to send a wake-up call (WUC) to node-3 that did not respond as it was not in wakeup range. Then, node-2 woke up node-1 and send it a concurrent wake-up request, including the address of node-3. Node-1 acknowledged (ACK) the reception of the concurrent wake-up request (REQ), and based on this acknowledgment, both nodes (1 and 2) sent a concurrent wake-up call to node-3. Finally, node-3 verified the wake-up address that is included in the wake-up call and and started the handshaking procedure as introduced in [86], followed by data communication from node-2 to node-3. Concurrent wake-up flooding To test and analyze the wake-up flooding algorithm as introduced in Section 4.4.6 , an indoor test consisting of several wireless sensor nodes as depicted in Figure 4.64 was conducted. The initiator (node 1 ), started to broadcast a seed wake-up packet. All nodes (nodes 2, 3, 4 and 5 ) that received this seed packet woke up and listened for the synchronization message transmitted directly after the seed packet by the initiator. Based on the timing of the synchronization packet, the nodes then transmitted con-

98


REQ ACK

node 1 receive 10 REQ

CWUC

node 1 send 8

activity

LBT

WACK LBT

REQ

ACK ACK

ACK

ACK

ACK

ACK

ACK

node 2 receive 6 WUC

WUC

REQ CWUC

REQ

DATA

DATA

DATA

DATA

DATA

REQ

DATA

DATA

DATA

DATA

DATA

node 2 send 4

node 3 receive 2 ACK ACK

node 3 send 0 −0.02 −0.01

0

0.01

0.02

0.03

0.04

ACK

ACK

0.05

ACK

0.06

ACK

0.07

ACK

0.08

time in seconds

Figure 4.62. Visualization of the states (send and receive) of nodes 1, 2 and 3 over time during the described concurrent wake-up protocol. The data was taken with the help of a logic-analyzer. current wake-up packets on different center frequencies as introduced in Section 4.4.5. Figure 4.63 visualizes the timing of the wake-up flooding algorithm. As can be seen, the nodes wake-up after around 20 ms and are ready to receive data.

Figure 4.63. Timing of the wake-up flooding algorithm.


99

Table 4.2 lists the frequencies that were utilized by the nodes during the flooding experiment. The third row in Table 4.2 indicates the type of sensor node during the flooding test and the last two rows report the amount of sent and received wake-up packets. Types could either be initiator, sender or receiver. The initiator, initiated the flooding and the sender nodes retransmitted the wake-up packets. The receiver nodes were placed in the building at different places where they had no direct link to the initiator, except for node 5. At each node, the amount of successfully received wake-ups were counted. Table 4.2. Frequencies used by the nodes during the flooding test set-up. node

frequency in MHz

type

wake-ups sent

wake-ups received

1 2 3 4 5 6 7 8 9 10 11

868 868.125 865 865.125 – 866.125 866 – – – –

initiator sender sender sender receiver sender sender receiver receiver receiver receiver

500 500 500 499 – 500 483 – – – –

– 500 500 499 1000 500 483 500 499 495 500

As reported in Table 4.2 that all nodes had a high wake-up packet reception ratio of almost 100 %. As node 5 could receive all seed wake-up calls as well as the following concurrent wake-up messages, it woke up for 1000 times.

4.4.9

Conclusions

This section presented a novel communication approach by exploiting purposefully interfering out of tune signals of two or more wireless sensor nodes, which produce the wake-up signal as the beat frequency of superposed carriers. The theoretical approaches of wake-up messages generated from beat frequencies were discussed theoretically for the first time and their suitability to improve wake-up robustness and reliability was demonstrated. The theoretical approaches were experimentally verified and two novel network protocols that apply this technique were designed. First, a simple and highly reliable algorithm that includes concurrent wake-ups into existing protocols, was developed. A quick and reliable dissemination of wake-up messages in a large wireless sensor network could be achieved, by designing a novel flooding protocol based on concurrent

100


Figure 4.64. Deployed wireless sensor network to analyze the wake-up flooding algorithm. wake-up transmissions. The algorithms were implemented their performance was verified experimentally with the help of an outdoor test setup and an indoor deployment. The tests confirmed the expected gain of up to 3 dB for two concurrent senders and nearly 100 % wake-up rates in less than 20 ms achieved by the flooding protocol. Future research should be aimed to further investigate the presented protocols and compare their performance to direct transmissions based approaches.

Chapter 5 Assessment of the Modal Properties of the Neckartal Bridge This chapter presents a novel structural health monitoring system based on a wireless sensor network for GNSS (global navigation satellite system) receivers as introduced in Section 3 and the communication protocol introduced in Section 4.2. The GNSS network presented here consists of three GNSS rover stations and one base station that are deployed at the Neckartal bridge on the Autobahn A81 in southwest Germany. By performing differential post-processing, precise positioning information in the millimeter range could be achieved, as introduced in Section 2.2. Using the GNSS sensors, a resonant frequency could be determined at 0.33 Hz, mainly in the lateral direction of the bridge. To verify the GNSS results, an accelerometer was additionally placed on the bridge. The frequencies detected by the acceleration sensor correspond well to the frequencies found by the GNSS sensors, although the accelerometer measured further higher frequencies as it is probably more sensitive to small amplitudes. The chapter is structured as follows. Section 5.1 provides a short introduction on existing structural health monitoring systems based on wireless sensor networks with special attention on structural health monitoring of bridges. Another point of focus will be wireless sensor networks with GNSS receivers. Section 5.1.1 briefly reviews the routing protocol introduced in Section 4.2 with focus on the transmissions if the large GNSS data packets. Section 5.2 presents and discuss the measurements taken with the GNSS sensors and analyze the results. Furthermore, the GNSS results are compared to measurements taken with an accelerometer and, finally, Section 5.3 concludes this chapter. 101

102

5.1

5. Assessment of the Modal Properties of the Neckartal Bridge

Introduction

Structural health monitoring of bridges can provide important information about structural performance and may help to detect anomalies or threats originating from damages or deteriorations at early stages. Structural health monitoring can also be used to estimate remaining lifetime, to assist in bridge maintenance planning, to verify construction designs and to deliver important data in the case of disasters or extreme events [87, 88]. According to [89], there exist three different techniques of structural health monitoring for bridges that can be classified into in situ, on-site and remote monitoring techniques. In situ monitoring techniques use sensors that are installed directly at a structure. On-site monitoring techniques are based on more complex sensors that are brought to a bridge during the monitoring campaign. Remote monitoring is done from a greater distance, for example by analyzing photographs of the structure taken from satellites or airplanes [90]. Especially in situ monitoring systems are often realized with the help of wireless sensor networks, as they are inexpensive and easy to install on existing structures. Several wireless sensor network systems have been introduced during the last few years and decades for structural health monitoring of bridges. O’Connor gives in [91] a comprehensive overview of installations for short-term monitoring and long-term monitoring. Short-term installations like [92–98] mostly measure accelerations, but some also include sensors for strain, temperature and velocity. Limitations arise mostly from the high power demands of the sensor nodes or base stations. Hu et al. [98], for example, presents a wireless sensor network to monitor the Zhengdian Highway Bridge based on wireless sensor nodes, which use an MSP430 microcontroller and a CC2420 radio. The nodes are running TinyOS, which uses MintRout to send data over multi-hops to a base station. Including an energy storage of 7500 mAh, a node is able to monitor strain or acceleration continuously for around 168 h, or when choosing a sampling period of 1 h/day, the lifetime can be extended to 168 days. The base station is connected via a USB connector to a powerful host computer. Amongst the wireless monitoring systems reviewed in [91], there are also three long-term monitoring installations, one on the Jindo Bridge [99], one on the Stork Bridge [100] and one on the New Carquinez Bridge [101]. The work in [91] summarizes the challenges arising in long-term installations originating from power supplies, communication reliability, sensor reliability and challenging environmental conditions. As Casciati and Chen point out in [102], large and dense sensor network deployments are often affected by large transmission delays as sensors cannot transmit their data concurrently. To increase communication reliability and to balance communication range, power consumption, data rate and link quality, Casciati and Chen [102] introduce a wireless sensor node that uses frequency-division multiplexing to transmit

5.1. Introduction

103

data simultaneously using different frequencies. Furthermore, in [103], Chen introduces a wireless sensor platform for structural health monitoring applications. The wireless sensor platform includes a power management unit to reduce the platform’s energy consumption and can be equipped with several types of sensors ranging from low power to energy-hungry devices. Chen also proposes an adaptive radio transmission power control algorithm along with a single-hop communication strategy. The power control algorithm adjusts the radio transmission energy to the link quality to save energy during the wireless data acquisition [103]. With respect to structural health monitoring, one other important measurement is the three-dimensional displacement of a structure or of parts of it. These measurements can be done by using on-site sensors [89, 104] or by using in situ measurements provided by Global Communication System (GPS) receivers [105–109]. Here, GPSbased monitoring systems provide several advantages, such as weather independence, absolute displacement measurements, autonomous operation and no need for a lineof-sight connection between different measurement points. Additionally, GPS provides both static and dynamic structural response information, and the position information is free from any measurement drifts and error accumulation [110]. By applying differential data processing of two or more concurrently logging GPS receivers, displacements in the range of 5 mm or below can be detected, as well as oscillation frequencies of up to and above 4 Hz. The authors of [106, 107, 109], for example, recorded the dynamical response of the Wilford Bridge in Nottingham, using GPS receivers in combination with accelerometers to validate the data and to improve the monitoring system’s performance. More recently, Kaloop et al. [111] presented GPS-based positioning data taken at the Mansoura railway bridge using Trimble-5700 dual-frequency GPS receivers logging at a rate of 1 Hz. The authors also present data taken from the long-term SHM system installed on the two towers of the Yonghe Bridge. Here, the GPS receivers are logging with a sampling frequency of 20 Hz. Reported standard deviations are in the mm range. However, these systems were all either tethered installations or short-term measurement campaigns. Although there are several existing wireless sensor networks that support the use of GNSS (global navigation satellite system) receivers [112–118], to the author’s knowledge, there is no such system available to monitor the dynamical behavior of bridges. Difficulties are in the high power demands of GNSS receivers and in the large amounts of data that need to be transmitted wirelessly.

5.1.1

Cross-layer Routing Protocol

The communication radio utilized in this work supports messages up to a size of 256 Byte. This means GNSS messages that usually consist of several kilobytes had to

104


1 ... 246 Byte Payload

1 Byte Status

4 Byte MAC 4 Byte Routing

1 Byte Length

be split into smaller packets before sending. Additionally, the expected total amount of data received from the GNSS receiver is quite high. For this reasons, the cross-layer routing protocol introduced in Section 4.2 was used that is able to transmit large data packets with only few control byte as depicted in Figure 5.1.

Figure 5.1. Data packet consisting of 8 control byte required by the routing algorithm and 1 to 246 payload byte. To be able to split large GNSS messages into several smaller radio packets and to be able to reassemble them later, every GNSS measurement had a unique ID based on its measurement time. Additionally, each packet number was related to the total number of packets of which a complete GNSS message consisted. Figure 5.2 depicts a payload packet to visualize this. Payload Sensor Packet Total.Num Timestamp GNSS.Data Length Type Number of.Packets 1.Byte

1.Byte

1.Byte

1.Byte

4.Byte

1...238.Byte

Figure 5.2. The structure of the payload to be transmitted in a routing packet. The payload consists of a length field, type of sensor, the number of the current packet, the total number of packets belonging to this GNSS message, the measurement timestamp and the GNSS message (part or complete) itself. From Figures 5.1 and 5.2 it becomes obvious that the routing protocol is able to transmit up to 238 Bytes GNSS data per packet. As the expected number of bytes to be transmitted, is in the order of several kilobytes, this is still challenging. But the routing algorithm is able to transmit up to 64 data packets along an existing link (see Section 4.2, which adds up to 15232 Byte payload. The data rate could be changed between 38.4, 100 and 250 kBit per second. At a data rate of 38 kBit per second (equal to 4.75 kByte per second) a transmission of 64 data packets takes roughly 4 s. The transmission of a 30 minutes period of 15 observations logged at 4 Hz (2852 kByte) takes around 10 minutes. But due to the need of additionally required control messages before each sending of 64 packets, the required time is even longer. The transmission of a 30 minutes logging period at 20 Hz (14256 kByte) takes around 50 minutes at a baud rate of 38.4 kBit per second. Using higher data rates, the transmission times can be decreased obviously, but the possibility of bit errors increases.

5.2. Experimental Results

5.2

105


Figure 5.3 shows a photo of the Weitinger Neckartal Bridge on Highway 81 from Stuttgart to Singen. The bridge is 900 m long and around 127 m above ground at its highest point. The bridge is made of five spans and its two end spans consist of inverted cable stay towers that support massive beam spans of 263 m on the southern side and 234 m on the northern side. The bridge has four lanes, two for each driving direction. The total width of the bridge deck is 31.5 m.

Figure 5.3. Photo of the Neckar Valley bridge near Weitingen in southwest Germany. Figure 5.4 shows the schematic of the Neckartal Bridge indicating the deployed GNSS receiver nodes together with the two relay nodes and the GSM base station. Three GNSS receivers were placed on the deck of the bridge, one above the first pillar (node E2), one just in the middle of the span between pillars one and two (node E3) and one in the middle of the bridge (node E4). A GNSS reference (node E1) was placed on the ground before the bridge. As the distances from E1 and E4 to the base station were quite large, they only logged the data. Due to the strong attenuation of the radio waves by the steel girder box, the data could not be sent directly from the bridge deck to the GSM node located inside the steel girder box. So the data was sent from the GNSS sensor node to a relay node located below the deck on top of one pillar as shown in Figure 5.5 (b). Figure 5.5 (a) exemplarily shows a GNSS receiver antenna attached to the bridge by magnets. It can be seen that the antennas have a relatively unobstructed view of the horizon on one side. The above-described points E1, E2, E3 and E4 have been installed with GNSS equipment and measured for a period of 40 min. Then positions E2 and E3 were

106


Figure 5.4. Deployment of the wireless sensor network consisting of three rover nodes, a reference node, two relay nodes and the GSM base station.

(a)

(b)

Figure 5.5. (a) Photo of a GNSS antenna attached to the side of the steel bridge by magnets. (b) Photo of relay node on top of the first pillar. logging and sending data for four consecutive days applying a logging period of three times 30 min per day. For reasons of clarity, the following processing examples are showing selected samples of 10 min for each of the mentioned points taken from the first 40 min interval. At first, the GNSS raw data are processed by applying the software RTKLib [40] developed by T. Takasu and A. Yasuda at the Laboratory of Satellite Navigation, Tokyo University of Marine Science and Technology, Japan. The RTKLib offers a sophisticated post-processing module, RTKPOST, that applies ambiguity fixing strategies, to achieve the required accuracy for structural monitoring. Depending on the distance between the GNSS base station and the rover point, accuracies in the region of one to


107

(a)

(b)

(c)

Figure 5.6. Position E2 (a) x-, (b) y- and (c) z-displacements over time. five millimeters are possible. To eliminate the main error sources of GNSS position, this package applies double differencing within an extended Kalman filter. A detailed description of the applied algorithms may be found in [40]. The resulting point positions from GNSS processing are usually represented as Earth Centered Earth Fixed coordinates, referring to the WGS84 ellipsoid. As these coordinates have no direct relation to the observed structure, the resulting coordinates have been rotated, in a manner that the x- and y-coordinates are representing the longitudinal and transversal axes of the Neckartal bridge, respectively and the z-component represents the vertical axis. Figures 5.6 to 5.8 show the resulting time series of the transformed point positions x, y and z for E2, E3 and E4. Observing Figures 5.6 to 5.8 one realizes immediately the higher noise apparent on the z-direction signal as would be expected from the discussion in Section 2.2. Furthermore, it can be observed in Figures 5.6 to 5.8, besides several short wave vibrations, there are also superposed significant low-frequency oscillations resulting from unknown sources, but that are probably caused by mutlipath effects. Furthermore, the wind forces the bridge to swing, and temperature effects cause strains in the load-carrying concrete pillars. Oscillations induced by the heavy traffic on the Autobahn A81 and a varying dilution of precision as indicated by Casciati and Fuggini in [119] could be further sources of these long wave signals. For example, Ni et al. compare in [120] displacements detected by a GPS system to the displacements detected by a vision-

108


(a)

(b)

(c)

Figure 5.7. Position E3 (a) x-, (b) y- and (c) z-displacements over time.

(a)

(b)

(c)

Figure 5.8. Position E4 (a) x-, (b) y- and (c) z-displacements over time.


109

(a)

(b)

(c)

Figure 5.9. Position E2 (a) x, (b) y and (c) z residuals over time. based system. Both systems correlate well with each other, but Ni et al. observed additional low-frequency oscillations in the GPS measurements that were not present in the vision-based measurements. Keeping this in mind, it becomes obvious, that for professional and efficient vibration monitoring, samples are best to be taken during night times, when there is almost no traffic on the bridge, or the bridge should be closed during this measurement campaigns. As the time series data are apparently noisy, a bandpass filter was applied to remove noise below 0.03 Hz and above 9.90 Hz. The remaining residuals are reported in Figures 5.9 to 5.11 that illustrate the filtered time series data for the measurement points E1, E2, and E4. The residuals can then be further analyzed by applying a Discrete Fourier Transform analysis. The Fourier analysis reveals the frequency response but there is additional noise present on the positioning data. Further analyzing the GNSS data by using a short-time Fourier transform on the measurement data and summing the spectrogram along the time-axis, reveals the two-dimensional frequency spectrograms for the x-, yand z-frequencies. Figures 5.12 to 5.14 illustrates the frequency spectra for positions E2 to E4. The analysis reveals an clear frequency response in x- and y-direction at approximately 0.33 Hz, visible in Figures 5.12 (a) and (b) to 5.14 (a) and (b). Furthermore, the significances of the amplitudes at position E2 are lower compared to the observed

110


(a)

(b)

(c)

Figure 5.10. Position E3 (a) x, (b) y and (c) z residuals over time.

(a)

(b)

(c)

Figure 5.11. Position E4 (a) x, (b) y and (c) z residuals over time.


111

(a)

(b)

(c)

Figure 5.12. Two-dimensional frequency spectra of the GNSS sensor at position E2 in directions (a) x, (b) y and (c) z. amplitudes at positions E3 and E4. This reflects the fact that E2 is above a pillar and as such is expected to make the fewer movements, whereas node E3 and E4 were placed between two pillars. The biggest movements were expected at position E4 as it is in the middle of the bridge. And indeed, the largest amplitudes in comparison to the responses at the other position, can be observed at E4.

Comparing GNSS and Acceleration Measurements As there are no structural response data available for this bridge, an accelerometer was placed at Position E4, to measure its movements in the x -, y- and z -directions, to verify the GNSS measurement data. The following Figures 5.15a to 5.15c show the frequency responses over time from 0 to 5 Hz as measured by the accelerometer. In the x -direction (Figure 5.15a), the sensor clearly shows a 2.1-Hz response. Furthermore, the x -direction acceleration data show a significant frequency at 0.33 Hz, as well as several frequencies below 3 Hz. In the y-direction, the accelerometer measurements indicate a resonance frequency

112


(a)

(b)

(c)

Figure 5.13. Two-dimensional frequency spectra of the GNSS sensor at position E3 in directions (a) x, (b) y and (c) z. around 2 Hz and between 4 and 5 Hz. In the z -direction, there appears to be several frequencies in the range from 1 to 5 Hz. Although the acceleration measurements in the z -direction are generally noisier, about four frequencies can be observed, the first at around 1.3 Hz, the second around 2.5 Hz, the third at 3.7 Hz and the fourth around 5 Hz. The 0.33-Hz resonance is also visible in the GNSS dataset as for example visualized in Figures 5.14a to 5.14b. But the frequencies above 1 Hz could not be detected by the GNSS sensors in this experiment. Using this set of measurements, the accelerometer data appears to have a lower resolution than the GNSS measurements but a higher sensitivity. In summary, the data presented here demonstrate the feasibility of the GNSS measurements to detect low frequencies and underpins the assumption that the detected frequencies are real structural resonance frequencies of the bridge. The comparison of the accelerometer and GNSS sensor measurements shows that the GNSS sensor has a better sensitivity for lower frequencies than the acceleration sensor, but is limited in its response to higher frequency signals probably due to their lower amplitudes. The

5.3. Conclusions

113

(a)

(b)

(c)

Figure 5.14. Two-dimensional frequency spectra of the GNSS sensor at position E4 in directions (a) x, (b) y and (c) z. two-dimension spectra further demonstrate the good frequency resolution that can be achieved by utilizing the GNSS sensor.

5.3

Conclusions

This chapter presented a novel wireless GNSS (global navigation satellite system) sensor network for bridge monitoring based on the wireless sensor nodes that are equipped with NovAtel OEM615 GNSS receivers introduced in Section 3.2. The wireless GNSS nodes support sampling frequencies up to 20 Hz and intermediate data logging on SD cards. Due to the asynchronous communication scheme provided by the wake-up receiver, the nodes reside in deep-sleep during times when no tasks need to be done, but are fully responsive at any time. Their ultra-low-power requirements during inactive periods, in the range of a few µW, provide lifetimes of almost 200 days without energy harvesting. The data was transferred wirelessly to a GSM gateway that transmitted the measurement data to a remote server. By applying corrections for residual satellite clocks, ionospheric and tropospheric delays, and by performing a differential data post

114


acceleration sensor x-direction

acceleration sensor y-direction

150

140 significance

significance

145 140 135

130 125

130

120

125 0

135

1

2 3 frequency in Hz

4

5

0

1

2 3 frequency in Hz

(a)

4

5

(b) 165

acceleration sensor z-direction

significance

160 155 150 145 140 135 130 0

1

2 3 frequency in Hz

4

5

(c)

Figure 5.15. Spectra of the acceleration sensor at position E4 in directions (a) x, (b) y and (c) z. processing, positioning accuracies in the millimeter range could be achieved. Analyzing the GNSS data, resonance frequency at 0.33 Hz could be detected. To verify the GNSS measurements, an acceleration sensor was placed on the bridge and compared the results of both sensors. Although the accelerometer is more sensitive as it is more sensitive to small amplitudes, the results of both sensors correspond well in the range below 1 Hz.

Chapter 6 Summary and Outlook Wireless communication in the area of wireless sensor networks is a mature technology. Wireless sensor networks are deployed in many applications and in many different areas. But still, providing energy to wireless sensor nodes is a major challenge as energy resources are often limited or not always present. Particularly, wireless communication requires many energy resources and a major field of research are energy efficient and high-performance communication algorithms. The asynchronous communication provided by wake-up receivers makes idle-listening obsolete. Consequently, ultra-low power wake-up receivers can save precious energy on a wireless sensor node. But, accompanied by their ultra-low power consumption, wake-up receivers have inferior sensitivity and there exists a gap between the effective operating range of wake-up and communication radios. Additionally, small-scale fading easily leads to unreliable wake-up receptions especially when operated in a dynamic environment. Robust and reliable wake-up detection is the key to long network lifetimes.

6.1

Summary

By introducing antenna diversity in the wake-up path along with novel communication algorithms and approaches, this thesis aims the above-mentioned challenges of unequal wake-up and communication ranges as well as the unreliable wake-up detection caused by small-scale fading. The implemented antenna diversity technique was an equal-gain combining system. The reduction of small-scale fading effects could be demonstrated experimentally. Under laboratory conditions, the received signal strength increased about approximately 3 dB and in a more realistic environment, between 0.8 and 1.2 dB above selection diversity. Furthermore, the antenna diversity system requires only two additional active components which increase the power requirements of the wireless sensor node only marginally. 115

116

6. Summary and Outlook

T-ROME, a simple cross-layer routing protocol developed during this work, makes explicit use of the different wake-up and communication ranges. Energy is saved by leapfrogging sensor nodes during communication and by transmitting several data packets along an established route in a row. Therefore, sensor nodes can for example accumulate measurements to large data packets and transmit them at once which largely reduces the energy required for communication. Accompanied by the routing algorithm, an analytical framework based on Markov chains was developed that can be used to analyze the performance of routing protocols. Using this analytical framework T-ROME was compared to two other state-of-the-art routing algorithms for wireless sensor networks with wake-up receivers. The results demonstrate the superior latency and energy requirements of T-ROME in many scenarios. Sending of a wake-up message requires a certain amount of energy. Utilizing concurrent wake-up transmissions, a sender can reduce its transmit power or make use of the increased robustness and reliability, as concurrent wake-up messages are transmitted at higher signal strengths. This thesis presented a novel method to generate concurrent wake-up packets as beat frequencies produced from two slightly out of tune carrier frequencies. The method was analyzed theoretically and verified experimentally. The concurrently transmitted wake-up signal strength increased about 3 dB what could be verified by simulations and a freespace experiment. A novel algorithm was developed to include concurrent wake-up messages into existing routing protocols. Furthermore, an algorithm was sketched that utilizes concurrently transmitted wake-up messages in a wake-up flooding protocol. The flooding achieved a wake-up rate of almost 100 % in less than 20 ms. The last part of this work presented a deployment of a wireless sensor network with wake-up receivers. The wireless sensor network consisted of wireless sensor nodes that were developed during this work and were equipped with GNSS (global navigation satellite system) receivers that measured the vibrations of a large highway bridge at a sampling rate of almost 20 Hz. To achieve the required position accuracy, a differential GNSS was applied that measured the movements of the receivers on the bridge relative to a receiver at a fixed position next to the bridge. T-ROME, the routing protocol developed during this work was implemented on the wireless sensor nodes to transmit the data to the sink. Since the position information logged by a GNSS receiver can be larger than 1 kByte, but the largest possible data packet that can be transmitted with T-ROME consists of 246 Byte, each GNSS data packet was split into several smaller packets and resembled in a database. Applying a Short Time Fourier Transform of the data revealed a resonance frequency of the bridge at 0.33 Hz mainly in the lateral direction, which was verified by a measurement with an accelerometer.

6.2. Outlook

6.2

117

Outlook

The novel methods and algorithms presented in this work can be implemented to build a robust and reliable wireless sensor network with wake-up receivers. Supporting multihop messages, T-ROME provides all means to realize large-scale wireless sensor network deployments, although the routing approach of T-ROME is static. A slightly more dynamic approach was presented in this work, but the next step has to be the integration of a fully dynamic routing to T-ROME. Dynamic routing prevents disconnected nodes in cases when further nodes in wakeup range exist, but that belong to a different branch and are, as such, out of reach of the currently implemented routing algorithm. Additionally, it could be shown that the routing algorithm has the capability to generate energy awareness, this feature is not yet implemented as well as the utilization of link quality measurements. Incorporating these routing metrics will lead to a more stable network and will optimize network throughput, energy and time as they leverage directly at the base level of communication. Furthermore, the concepts of diversity, concurrent wake-up, and multihop wake-up should be brought together. This work already started this challenge, by successfully including concurrent wake-up messages in T-ROME, although only on an experimental level. The ability of concurrent wake-up flooding to assist in the initial configuration of a wireless sensor network with wake-up receives should be investigated. Concurrent wake-up flooding can be used for example, by T-ROME during the initialization phase of a dynamic routing protocol. This work demonstrated that long sensor node lifetimes can be achieved by utilizing wake-up receivers also with sensors that have high energy demands such as GNSS receivers. The results demonstrate the feasibility of a long-term autonomous structural monitoring, although they are based on short-term measurements, only. Consequently, the next step should be a permanent wireless sensor network deployment at the Neckartal Bridge with several kinds of sensors such as accelerometers, GNSS receivers, and sensors to measure environmental parameters. Different modes of operation of the GNSS receivers, for instance by using high sampling rates (20 Hz or above) and short observation times (minutes to hours) or by using medium sampling rates (1 to 10 Hz) and long observation times (hours to days), more insights to the static and dynamic movements of the structure can be achieved. Additionally, the measurements should be combined with a model of the bridge to evaluate its structural behavior for example in addition to a Kalman filter utilized to improve the position information accuracy.

Appendix A Publications Journal publications Kumberg, T.; Schneid, S.; Reindl, L. A Wireless Sensor Network Using GNSS Receivers for a Short-Term Assessment of the Modal Properties of the Neckartal Bridge. Appl. Sci., 2017, 7, 626. Kumberg, T., Schink, M., Reindl, L. M., Schindelhauer, C. (2017). T-ROME: A simple and energy efficient tree routing protocol for low-power wake-up receivers. Ad Hoc Networks, 59, 97-115.

Conference publications Kumberg, T.; Moharrami, M.; Schindelhauer, C.; Reindl, L. Improving the Performance of the Cross-Layer Wake-Up Routing Protocol T-ROME. Proc. of IWCMC 2017 Wireless Sensor Symposium (IWCMC-Wireless Sensors 2017), 2017. Saez, J., Kumberg, T., Reindl, L.M., Development and characterization of a Robust Differential Wake-up Receiver for Wireless Sensor Networks, Proc. of IWCMC 2017 Wireless Sensor Symposium (IWCMC-Wireless Sensors 2017), 2017. Kumberg, T., Kokert, J., Younesi, V., Koenig, S., Reindl, L. M. (2016, April). Wake-up transceivers for structural health monitoring of bridges. In SPIE Smart Structures and Materials+ Nondestructive Evaluation and Health Monitoring (pp. 98041S98041S). International Society for Optics and Photonics. Kumberg, T., Kokert, J., Tannhaeuser, R., Schink, M., Reindl, L.M. (2016) Development of a Wireless low-power Sensor Node for low latency Sensor Networks using Wake-Up Receivers. Proceedings 11th Future Security Conference, Berlin, Germany. 118

119

Kumberg, T., Tannhaeuser, R., Koenig, S., Schneid, S. (2016) Einsatz eines sensorgestuetzten Informationssystems fuer das Brueckenmonitoring im Bestand – Teil 1, 2. Proceedings of Brueckenkolloquium, at Technische Akademie Esslingen, June 2016. Kumberg, T., Schink, M., Reindl, L. M., Schindelhauer, C. (2016) Analyzing TROME: A Cross-Layer Routing Protocol for Low-Power Wake-up Receivers. Proceedings of the 15. GiITG KuVS Fachgespraech Sensornetze, University of Applied Sciences, Dept. of Computer Science, Technical Reports, September 2016. Kumberg, T., Tannhaeuser, R., Reindl, L. M. (2015) Using Antenna Diversity to Improve Wake-up Range and Probability. In Proc. of Progress In Electromagnetics Research Symposium Proceedings (PIERS). Session 4P6 RF and Wireless Communication, (pp. 2779-2783). Kumberg, T., Tannhaeuser, R, Zimmermann, L., Schink, M., Schindelhauer, C., Reindl, L.M. (2015) Wake-Up Transceivers for Monitoring Critical Infrastructure – a Prototype overcoming Duty Cycling in Wireless Sensor Networks, proceedings of Future Security 2015, Berlin. Kumberg, T., Tannhaeuser, R., Schink, M., Schneid, S., König, S., Schindelhauer, C., Reindl, L. M. (2015). Wireless wake-up sensor network for structural health monitoring of large-scale highway bridges. In International Conference on Performance-based and Life-cycle Structural Engineering (pp. 1393-1401). School of Civil Engineering, The University of Queensland. Kumberg, T.; Tannhaeuser, R.; Gamm, G.U.; Reindl, L.M., (2014) Energy improved wake-up strategy for wireless sensor networks. In proc. Sensors and Measuring Systems 2014 ; 17. ITGGMA Symposium; pp.1,6, 3-4 June 2014.

Appendix B Acknowledgments Besonderen Dank möchte ich meinem Doktorvater Herr Prof. Leonhard Reindl aussprechen, f¨ ur die Möglichkeit an seinem Lehrstuhl meine Arbeit anfertigen zu können. Sein stetes Vertrauen in meine Arbeit und meine Fähigkeiten war mir dabei immer eine große Motivation. Meinem Zweitgutachter Herr Prof. Schindelhauer danke ich f¨ ur die zahlreichen und fruchtbaren Diskussionen und Verbesserungsvorschläge zu meiner Arbeit. F¨ ur finanzielle Unterst¨ utzung bei Vortragsreisen nach Brisbane, Las Vegas und Valencia danke ich der Wissenschaftlichen Gesellschaft Freiburg sowie dem DAAD. Der DFG danke ich f¨ ur die finanzielle Unterst¨ utzung bei der Publikation meiner Forschungsergebnisse. Der Professur EMP möchte ich f¨ ur die schönen vier Jahre danken die ich hier hatte. Bei den Technikern Christoph Bohnert, Uwe Burzlaff und Hans Baumer bedanke ich mich f¨ ur die technische Hilfe und Ratschläge bei Problemen. Besonderen Dank auch an Heidi Schmidt und Beate Botschek f¨ ur ihre Hilfsbereitschaft bei den verschiedensten Anliegen. Insbesondere danke ich auch Daniel Vössing, Taimur Aftab, Adnan Yousaf und Robert Tannhäuser f¨ ur die zahlreichen Kaffeepausen, die hilfreichen Diskussionen, das Korrekturlesen und die generelle Unterst¨ utzung. Ebenso dank ich meinen Hiwis Colin Seibel und Artur Neff und meinen anderen Bachelor/Master-Studenten, die eine große Hilfe waren. Mein besonderer Dank geht an meine Familie, die mir vertraut und mich unterst¨ utz.

120

Bibliography [1] Daniele Puccinelli and Martin Haenggi. Wireless sensor networks: applications and challenges of ubiquitous sensing. Circuits and Systems Magazine, IEEE, 5(3):19–31, 2005. [2] He Ba, Ilker Demirkol, and Wendi Heinzelman. Passive wake-up radios: From devices to applications. Ad hoc networks, 11(8):2605–2621, 2013. [3] Pei Huang, Li Xiao, Sima Soltani, Matt W Mutka, and Ning Xi. The evolution of MAC protocols in wireless sensor networks: A survey. Communications Surveys & Tutorials, IEEE, 15(1):101–120, 2013. [4] Fabian Hoflinger, Gerd Ulrich Gamm, Joan Albesa, and Leonhard M Reindl. Smartphone remote control for home automation applications based on acoustic wake-up receivers. In Instrumentation and Measurement Technology Conference (I2MTC) Proceedings, 2014 IEEE International, pages 1580–1583. IEEE, 2014. [5] Amir Bannoura, Fabian Höflinger, Omar Gorgies, Gerd Ulrich Gamm, Joan Albesa, and Leonhard M Reindl. Acoustic Wake-Up Receivers for Home Automation Control Applications. Electronics, 5(1):4, 2016. [6] Timo Kumberg, Robert Tannhaeuser, Marc Schink, Sascha Schneid, Stefan Koenig, Christian Schindelhauer, and Leonhard M Reindl. Wireless wake-up sensor network for structural health monitoring of large-scale highway bridges. Proceedings of the Second International Conference on Performance?based and Life-cycle Structural Engineering (PLSE 2015), pages 1393–1401, 2015. [7] Gerd Ulrich Gamm and Leonhard Michael Reindl. Smart metering using distributed wake-up receivers. In Instrumentation and Measurement Technology Conference (I2MTC), 2012 IEEE International, pages 2589–2593. IEEE, 2012. [8] T. Kumberg, J. Kokert, V. Younesi, S. Koenig, and L.M. Reindl. Wake-up transceivers for structural health monitoring of bridges. In Proceedings of SPIE - The International Society for Optical Engineering, volume 9804, 2016. 121

122

Bibliography

[9] Johannes Blanckenstein, Jirka Klaue, and Holger Karl. A Survey of Low-Power Transceivers and Their Applications. Circuits and Systems Magazine, IEEE, 15(3):6–17, 2015. [10] Amir Bannoura, Christian Ortolf, Leonhard Reindl, and Christian Schindelhauer. The wake up dominating set problem. Theoretical Computer Science, 608:120–134, 2015. [11] Michele Magno and Luca Benini. An ultra low power high sensitivity wake-up radio receiver with addressing capability. In Wireless and Mobile Computing, Networking and Communications (WiMob), 2014 IEEE 10th International Conference on, pages 92–99. IEEE, 2014. [12] Emil Nilsson and Christer Svensson. Ultra low power wake-up radio using envelope detector and transmission line voltage transformer. Emerging and Selected Topics in Circuits and Systems, IEEE Journal on, 3(1):5–12, 2013. [13] Gerd Ulrich Gamm, M Kostic, Matthias Sippel, and Leonhard M Reindl. Low– power sensor node with addressable wake–up on–demand capability. International Journal of Sensor Networks, 11(1):48–56, 2012. [14] Christian Hambeck, Stefan Mahlknecht, and Thomas Herndl. A 2.4 µw wake-up receiver for wireless sensor nodes with- 71dbm sensitivity. In Circuits and Systems (ISCAS), 2011 IEEE International Symposium on, pages 534–537. IEEE, 2011. [15] Tor-Inge Kvaksrud. Range measurements in an open field environment. Texas Instrum. Incorporated, Dallas, TX, Design Note DN018, 2008. [16] Theodore S Rappaport et al. Wireless communications: principles and practice, volume 2. 1996. [17] John S Seybold. Introduction to RF propagation. John Wiley & Sons, 2005. [18] Holger Karl and Andreas Willig. Protocols and architectures for wireless sensor networks. John Wiley & Sons, 2007. [19] Ross D Murch and K Ben Letaief. Antenna systems for broadband wireless access. IEEE Communications Magazine, 40(4):76–83, 2002. [20] DG Brennan. Linear diversity combining techniques. Proceedings of the IEEE, 91(2):331–356, 2003. [21] Carl B Dietrich, Kai Dietze, J Randall Nealy, and Warren L Stutzman. Spatial, polarization, and pattern diversity for wireless handheld terminals. IEEE transactions on antennas and propagation, 49(9):1271–1281, 2001.

Bibliography

123

[22] Donald Cox. Antenna diversity performance in mitigating the effects of portable radiotelephone orientation and multipath propagation. IEEE transactions on communications, 31(5):620–628, 1983. [23] Mark Wallace and Jay Walton. Method and apparatus for antenna diversity in a wireless communication system, June 6 2001. US Patent App. 09/875,397. [24] JD Parsons, MIGUEL Henze, PA Ratliff, and MICHAEL J Withers. Diversity techniques for mobile radio reception. Radio and Electronic Engineer, 45(7):357– 367, 1975. [25] Stephen Ross Todd. Diversity antenna selection, December 14 1999. US Patent 6,002,672. [26] Elliott Kaplan and Christopher Hegarty. Understanding GPS: principles and applications. Artech house, 2005. [27] Bernhard Hofmann-Wellenhof, Herbert Lichtenegger, and Elmar Wasle. GNSS– global navigation satellite systems: GPS, GLONASS, Galileo, and more. Springer Science & Business Media, 2007. [28] Andrey Soloviev and Jeff Dickman. Extending GPS carrier phase availability indoors with a deeply integrated receiver architecture. IEEE Wireless Communications, 18(2), 2011. [29] C Jeffrey. An introduction to gnss gps, glonass, galileo and other global navigation satellite systems. edn. NovAtel Inc, 2010. [30] Michael S Braasch and AJ Van Dierendonck. Gps receiver architectures and measurements. Proceedings of the IEEE, 87(1):48–64, 1999. [31] Geoffrey Blewitt. Basics of the GPS technique: observation equations. Geodetic applications of GPS, pages 10–54, 1997. [32] Mohammad Zahidul H Bhuiyan and Elena Simona Lohan. Multipath mitigation techniques for satellite-based positioning applications. Global navigation satellite systems: signal, theory and applications. InTech, Rijeka, pages 405–426, 2012. [33] José A López-Salcedo, Jose A Del Peral-Rosado, and Gonzalo Seco-Granados. Survey on robust carrier tracking techniques. IEEE Communications Surveys & Tutorials, 16(2):670–688, 2014. [34] Derek K Shaeffer, Arvin R Shahani, SS Mohan, Hirad Samavati, Hamid R Rategh, Maria del Mar Hershenson, Min Xu, C Patrick Yue, Daniel J Eddleman, and

124

Bibliography

Thomas H Lee. A 115-mw, 0.5-/spl mu/m cmos gps receiver with wide dynamicrange active filters. IEEE Journal of Solid-State Circuits, 33(12):2219–2231, 1998. [35] Christopher Bowick. RF circuit design. Newnes, 2011. [36] Kyoohyun Lim, S-H Lee, Sunki Min, Sungmin Ock, M-W Hwang, C-H Lee, K-L Kim, and Sangwoo Han. A fully integrated direct-conversion receiver for cdma and gps applications. IEEE journal of solid-state circuits, 41(11):2408–2416, 2006. [37] Michael S Braasch. Performance comparison of multipath mitigating receiver architectures. In Aerospace Conference, 2001, IEEE Proceedings., volume 3, pages 3–1309. IEEE, 2001. [38] Philip RR Strode and Paul D Groves. Gnss multipath detection using threefrequency signal-to-noise measurements. GPS solutions, 20(3):399–412, 2016. [39] Lawrence Lau and Paul Cross. Investigations into phase multipath mitigation techniques for high precision positioning in difficult environments. Journal of Navigation, 60(03):457–482, 2007. [40] T Takasu. RTKLIB: An open source program package for GNSS positioning. http://www.rtklib.com/prog/rtklib_2.4.1.zip, 2011. Accessed: 2018-07-28. [41] Mike Golio. Commercial wireless circuits and components handbook. CRC press, 2002. [42] Sandra Verhagen. The GNSS integer ambiguities: estimation and validation. PhD thesis, TU Delft, Delft University of Technology, 2005. [43] Peter JG Teunissen. Least-squares estimation of the integer gps ambiguities. In Invited lecture, section IV theory and methodology, IAG general meeting, Beijing, China, 1993. [44] Peter JG Teunissen, PJ Jonge, and CCJM Tiberius. Performance of the lambda method for fast gps ambiguity resolution. Navigation, 44(3):373–383, 1997. [45] Moshaddique Al Ameen and Choong Seon Hong. An On-Demand Emergency Packet Transmission Scheme for Wireless Body Area Networks. Sensors, 15(12):30584–30616, 2015. [46] Junaid Ansari, Dmitry Pankin, and Petri Mähönen. Radio-triggered wake-ups with addressing capabilities for extremely low power sensor network applications. International Journal of Wireless Information Networks, 16(3):118–130, 2009.

Bibliography

125

[47] Dora Spenza, Michele Magno, Stefano Basagni, Luca Benini, Mario Paoli, and Chiara Petrioli. Beyond duty cycling: wake-up radio with selective awakenings for long-lived wireless sensing systems. In Computer Communications (INFOCOM), 2015 IEEE Conference on, pages 522–530. IEEE, 2015. [48] Stevan J Marinkovic and Emanuel M Popovici. Nano-power wireless wake-up receiver with serial peripheral interface. Selected Areas in Communications, IEEE Journal on, 29(8):1641–1647, 2011. [49] Nathan E Roberts and David D Wentzloff. A 98nW wake-up radio for wireless body area networks. In Radio Frequency Integrated Circuits Symposium (RFIC), 2012 IEEE, pages 373–376. IEEE, 2012. [50] Seunghyun Oh, Nathan E Roberts, and David D Wentzloff. A 116nW multiband wake-up receiver with 31-bit correlator and interference rejection. In Custom Integrated Circuits Conference (CICC), 2013 IEEE, pages 1–4. IEEE, 2013. [51] Timo Kumberg, Robert Tannhaeuser, Gerd Ulrich Gamm, and Leonhard M Reindl. Energy improved wake-up strategy for wireless sensor networks. In Sensors and Measuring Systems 2014; 17. ITG/GMA Symposium; Proceedings of, pages 1–6. VDE, 2014. [52] Chiara Petrioli, Dora Spenza, Pasquale Tommasino, and Alessandro Trifiletti. A novel wake-up receiver with addressing capability for wireless sensor nodes. In Distributed Computing in Sensor Systems (DCOSS), 2014 IEEE International Conference on, pages 18–25. IEEE, 2014. [53] NovAtel Inc., 1120 - 68 Avenue NE Calgary, AB Canada, T2E 8S5. OEM6 Family Installation and Operation User Manual. OM-20000128 Rev 4. [54] ams AG, Tobelbader Strasse 30 8141 Unterpremstaetten Austria. AS39323D Low Frequency Wakeup Receiver. Rev. 1.4. [55] Vivek Jain, Ratnabali Biswas, and Dharma P Agrawal. Energy-efficient and reliable medium access in sensor networks. In World of Wireless, Mobile and Multimedia Networks, 2007. WoWMoM 2007. IEEE International Symposium on a, pages 1–8. IEEE, 2007. [56] S Mahlknecht and M Spinola Durante. WUR-MAC: energy efficient wakeup receiver based MAC protocol. In Proceedings of the 8th IFAC International Conference on Fieldbuses and Networks in Industrial and Embedded Systems, 2009.

126

Bibliography

[57] Heikki Karvonen, Juha Petajajarvi, Jari Iinatti, Matti Hamalainen, and Carlos Pomalaza-Raez. A generic wake-up radio based MAC protocol for energy efficient short range communication. In Personal, Indoor, and Mobile Radio Communication (PIMRC), 2014 IEEE 25th Annual International Symposium on, pages 2173–2177. IEEE, 2014. [58] Stevan J Marinkovic and Emanuel M Popovici. Power efficient networking using a novel wake-up radio. In Pervasive Computing Technologies for Healthcare (PervasiveHealth), 2011 5th International Conference on, pages 139–143. IEEE, 2011. [59] Stevan Jovica Marinković, Emanuel Mihai Popovici, Christian Spagnol, Stephen Faul, and William Peter Marnane. Energy-efficient low duty cycle MAC protocol for wireless body area networks. Information Technology in Biomedicine, IEEE Transactions on, 13(6):915–925, 2009. [60] Felix Sutton, Bernhard Buchli, Jan Beutel, and Lothar Thiele. Zippy: On-Demand Network Flooding. In Proceedings of the 13th ACM Conference on Embedded Networked Sensor Systems, pages 45–58. ACM, 2015. [61] Stefano Basagni, Chiara Petrioli, and Dora Spenza. CTP-WUR: The collection tree protocol in wake-up radio WSNs for critical applications. In 2016 International Conference on Computing, Networking and Communications (ICNC), pages 1–6. IEEE, 2016. [62] Omprakash Gnawali, Rodrigo Fonseca, Kyle Jamieson, Maria Kazandjieva, David Moss, and Philip Levis. CTP: An efficient, robust, and reliable collection tree protocol for wireless sensor networks. ACM Transactions on Sensor Networks (TOSN), 10(1):16, 2013. [63] Li Chen, Jeremy Warner, Wendi Heinzelman, and Ilker Demirkol. MH-REACHMote: Supporting multi-hop passive radio wake-up for wireless sensor networks. In 2015 IEEE International Conference on Communications (ICC), pages 6512–6518. IEEE, 2015. [64] Johannes Blanckenstein, Jirka Klaue, and Holger Karl. Energy efficient clustering using a wake-up receiver. In European Wireless, 2012. EW. 18th European Wireless Conference, pages 1–8. VDE, 2012. [65] Chiara Petrioli, Michele Nati, Paolo Casari, Michele Zorzi, and Stefano Basagni. ALBA-R: Load-balancing geographic routing around connectivity holes in wireless sensor networks. Parallel and Distributed Systems, IEEE Transactions on, 25(3):529–539, 2014.

Bibliography

127

[66] Amir Bannoura. Algorithms and Applications for Low Power Wireless Sensor Networks using Wake-up Receivers. dissertation, University of Freiburg, 2016. [67] Wendi Rabiner Heinzelman, Anantha Chandrakasan, and Hari Balakrishnan. Energy-efficient communication protocol for wireless microsensor networks. In System sciences, 2000. Proceedings of the 33rd annual Hawaii international conference on, pages 10–pp. IEEE, 2000. [68] Kannan Srinivasan, Maria A Kazandjieva, Saatvik Agarwal, and Philip Levis. The β-factor: measuring wireless link burstiness. In Proceedings of the 6th ACM conference on Embedded network sensor systems, pages 29–42. ACM, 2008. [69] Azzedine Boukerche and Amir Darehshoorzadeh. Opportunistic routing in wireless networks: Models, algorithms, and classifications. ACM Computing Surveys (CSUR), 47(2):22, 2015. [70] Phil Karn. MACA-a new channel access method for packet radio. In ARRL/CRRL Amateur radio 9th computer networking conference, volume 140, pages 134–140, 1990. [71] Wanzhi Qiu, Efstratios Skafidas, and Peng Hao. Enhanced tree routing for wireless sensor networks. Ad hoc networks, 7(3):638–650, 2009. [72] J Vasseur, Navneet Agarwal, Jonathan Hui, Zach Shelby, Paul Bertrand, and Cedric Chauvenet. Rpl: The ip routing protocol designed for low power and lossy networks. Internet Protocol for Smart Objects (IPSO) Alliance, 36, 2011. [73] Federico Ferrari, Marco Zimmerling, Lothar Thiele, and Olga Saukh. Efficient network flooding and time synchronization with glossy. In Information Processing in Sensor Networks (IPSN), 2011 10th International Conference on, pages 73–84. IEEE, 2011. [74] Kamin Whitehouse, Alec Woo, Fred Jiang, Joseph Polastre, and David Culler. Exploiting the capture effect for collision detection and recovery. In Embedded Networked Sensors, 2005. EmNetS-II. The Second IEEE Workshop on, pages 45– 52. IEEE, 2005. [75] Dongjin Son, Bhaskar Krishnamachari, and John Heidemann. Experimental study of concurrent transmission in wireless sensor networks. In Proceedings of the 4th international conference on Embedded networked sensor systems, pages 237–250. ACM, 2006.

128

Bibliography

[76] Prabal Dutta, Razvan Musaloiu-e, Ion Stoica, and Andreas Terzis. Wireless ACK collisions not considered harmful. In Proceedings of the 7th ACM Workshop on Hot Topics in Networks (HotNets-VII), pages 1–6, 2008. [77] Ed Callaway, Paul Gorday, Lance Hester, Jose A Gutierrez, Marco Naeve, Bob Heile, and Venkat Bahl. Home networking with IEEE 802. 15. 4: a developing standard for low-rate wireless personal area networks. IEEE Communications magazine, 40(8):70–77, 2002. [78] Dingwen Yuan and Matthias Hollick. Let’s talk together: Understanding concurrent transmission in wireless sensor networks. In Local Computer Networks (LCN), 2013 IEEE 38th Conference on, pages 219–227. IEEE, 2013. [79] Yin Wang, Yuan He, Xufei Mao, Yunhao Liu, and Xiang-yang Li. Exploiting constructive interference for scalable flooding in wireless networks. IEEE/ACM Transactions on Networking, 21(6):1880–1889, 2013. [80] Yin Wang, Yunhao Liu, Yuan He, Xiang-Yang Li, and Dapeng Cheng. Disco: Improving packet delivery via deliberate synchronized constructive interference. IEEE Transactions on Parallel and Distributed Systems, 26(3):713–723, 2015. [81] Manjunath Doddavenkatappa, Mun Choon Chan, Ben Leong, and Others. Splash: Fast Data Dissemination with Constructive Interference in Wireless Sensor Networks. In NSDI, pages 269–282, 2013. [82] Shuying Yu, Xiaobing Wu, Pan Wu, Dingming Wu, Haipeng Dai, and Guihai Chen. Cirf: Constructive interference-based reliable flooding in asynchronous dutycycle wireless sensor networks. In Wireless Communications and Networking Conference (WCNC), 2014 IEEE, pages 2734–2738. IEEE, 2014. [83] Vijay S Rao, M Koppal, R Venkatesha Prasad, T V Prabhakar, C Sarkar, and Ignas Niemegeers. Murphy loves CI: Unfolding and improving constructive interference in WSNs. In Computer Communications, IEEE INFOCOM 2016-The 35th Annual IEEE International Conference on, pages 1–9. IEEE, 2016. [84] Antonio Escobar, Francisco J Cruz, Javier Garcia-Jimenez, Jirka Klaue, and Angel Corona. RedFixHop with channel hopping: Reliable ultra-low-latency network flooding. In Design of Circuits and Integrated Systems (DCIS), 2016 Conference on, pages 1–4. IEEE, 2016. [85] Timo Kumberg, Marc Schink, Leonhard M. Reindl, and Christian Schindelhauer. T-ROME: A simple and energy efficient tree routing protocol for low-power wake-up receivers. Ad Hoc Networks, 59:97–115, may 2017.

Bibliography

129

[86] Timo Kumberg, Mojtaba Moharrami, Christian Schindelhauer, and Leonhard Reindl. Improving the Performance of the {Cross-Layer} {Wake-Up} Routing Protocol {T-ROME}. In IWCMC 2017 Wireless Sensor Symposium (IWCMC-Wireless Sensors 2017), Valencia, Spain, 2017. [87] Peter C Chang, Alison Flatau, and S C Liu. Review paper: health monitoring of civil infrastructure. Structural health monitoring, 2(3):257–267, 2003. [88] J M Ko and Y Q Ni. Technology developments in structural health monitoring of large-scale bridges. Engineering structures, 27(12):1715–1725, 2005. [89] T M Ahlborn, R Shuchman, L L Sutter, C N Brooks, D K Harris, J W Burns, K A Endsley, D C Evans, K Vaghefi, and R C Oats. An evaluation of commercially available remote sensors for assessing highway bridge condition. 2010. [90] Can Balkaya, Fabio Casciati, Sara Casciati, Lucia Faravelli, and Michele Vece. Real-time identification of disaster areas by an open-access vision-based tool. Advances in Engineering Software, 88:83–90, 2015. [91] Sean Michael O’Connor. Wireless Monitoring Systems for Long-Term Reliability Assessment of Bridge Structures based on Compressed Sensing and Data-Driven Data Interrogation Methods. PhD thesis, University of Michigan, 2015. [92] Jerome P Lynch, Yang Wang, Kenneth J Loh, Jin-Hak Yi, and Chung-Bang Yun. Performance monitoring of the Geumdang Bridge using a dense network of highresolution wireless sensors. Smart Materials and Structures, 15(6):1561, 2006. [93] Sukun Kim, Shamim Pakzad, David Culler, James Demmel, Gregory Fenves, Steven Glaser, and Martin Turon. Health monitoring of civil infrastructures using wireless sensor networks. In Proceedings of the 6th international conference on Information processing in sensor networks, pages 254–263. ACM, 2007. [94] Matthew J Whelan, Michael V Gangone, and Kerop D Janoyan. Highway bridge assessment using an adaptive real-time wireless sensor network. IEEE Sensors Journal, 9(11):1405–1413, 2009. [95] Maurizio Bocca, Lasse M Eriksson, Aamir Mahmood, Riku Jäntti, and Jyrki Kullaa. A synchronized wireless sensor network for experimental modal analysis in structural health monitoring. Computer-Aided Civil and Infrastructure Engineering, 26(7):483–499, 2011. [96] Sung-Han Sim, Jian Li, Hongki Jo, Jong-Woong Park, Soojin Cho, Billie F Spencer Jr, and Hyung-Jo Jung. A wireless smart sensor network for automated monitoring of cable tension. Smart Materials and Structures, 23(2):25006, 2013.

130

Bibliography

[97] M J Chae, H S Yoo, J Y Kim, and M Y Cho. Development of a wireless sensor network system for suspension bridge health monitoring. Automation in Construction, 21:237–252, 2012. [98] Xiaoya Hu, Bingwen Wang, and Han Ji. A wireless sensor network-based structural health monitoring system for highway bridges. Computer-Aided Civil and Infrastructure Engineering, 28(3):193–209, 2013. [99] Shinae Jang, Hongki Jo, Soojin Cho, Kirill Mechitov, Jennifer A Rice, SungHan Sim, Hyung-Jo Jung, Chung-Bang Yun, Billie F Spencer Jr, and Gul Agha. Structural health monitoring of a cable-stayed bridge using smart sensor technology: deployment and evaluation. [100] Jonas Meyer, Reinhard Bischoff, Glauco Feltrin, and Masoud Motavalli. Wireless sensor networks for long-term structural health monitoring. Smart Structures and Systems, 6(3):263–275, 2010. [101] Masahiro Kurata, Junhee Kim, Yilan Zhang, Jerome P Lynch, G W der Linden, Vince Jacob, Ed Thometz, Pat Hipley, and Li-Hong Sheng. Long-term assessment of an autonomous wireless structural health monitoring system at the new Carquinez Suspension Bridge. In SPIE Smart Structures and Materials+ Nondestructive Evaluation and Health Monitoring, page 798312. International Society for Optics and Photonics, 2011. [102] Sara Casciati and ZhiCong Chen. A multi-channel wireless connection system for structural health monitoring applications. Structural Control and Health Monitoring, 18(5):588–600, 2011. [103] ZhiCong Chen. Energy efficiency strategy for a general real-time wireless sensor platform. Smart Structures and Systems, 14(4):617–641, 2014. [104] E J Cross, K Y Koo, J M W Brownjohn, and K Worden. Long-term monitoring and data analysis of the Tamar Bridge. Mechanical Systems and Signal Processing, 35(1):16–34, 2013. [105] Alfredo Knecht and Luca Manetti. Using GPS in structural health monitoring. In SPIE’S 8th annual international symposium on smart structures and materials, pages 122–129. International Society for Optics and Photonics, 2001. [106] Ting-Hua Yi, Hong-Nan Li, and Ming Gu. Experimental assessment of high-rate GPS receivers for deformation monitoring of bridge. Measurement, 46(1):420–432, 2013.

Bibliography

131

[107] Xiaolin Meng, A H Dodson, and G W Roberts. Detecting bridge dynamics with GPS and triaxial accelerometers. Engineering Structures, 29(11):3178–3184, 2007. [108] A Nickitopoulou, K Protopsalti, and S Stiros. Monitoring dynamic and quasistatic deformations of large flexible engineering structures with GPS: accuracy, limitations and promises. Engineering Structures, 28(10):1471–1482, 2006. [109] Panos Psimoulis, Stella Pytharouli, Dimitris Karambalis, and Stathis Stiros. Potential of Global Positioning System (GPS) to measure frequencies of oscillations of engineering structures. Journal of Sound and Vibration, 318(3):606–623, 2008. [110] Ting-Hua Yi, Hong-Nan Li, and Ming Gu. Recent research and applications of gps-based monitoring technology for high-rise structures. Structural Control and Health Monitoring, 20(5):649–670, 2013. [111] Mosbeh R Kaloop, Jong Wan Hu, and Emad Elbeltagi. Adjustment and Assessment of the Measurements of Low and High Sampling Frequencies of GPS Real-Time Monitoring of Structural Movement. ISPRS International Journal of Geo-Information, 5(12):222, 2016. [112] Radu Stoleru, Tian He, and John A Stankovic. Walking GPS: A practical solution for localization in manually deployed wireless sensor networks. In Local Computer Networks, 2004. 29th Annual IEEE International Conference on, pages 480–489. IEEE, 2004. [113] Kirk Martinez, Royan Ong, and Jane Hart. Glacsweb: a sensor network for hostile environments. In Sensor and Ad Hoc Communications and Networks, 2004. IEEE SECON 2004. 2004 First Annual IEEE Communications Society Conference on, pages 81–87. IEEE, 2004. [114] L Aguado, Ciaran O’Driscoll, Peiqing Xia, Konstantin Nurutdinov, Chris Hill, and Patrick O’Beirne. A low-cost, low-power Galileo/GPS positioning system for monitoring landslides. Navitec (October 2006), 194, 2006. [115] Claudio Lucianaz, Oscar Rorato, Marco Allegretti, M Mamino, Marco Roggero, and F Diotri. Low cost DGPS wireless network. In Antennas and Propagation in Wireless Communications (APWC), 2011 IEEE-APS Topical Conference on, pages 792–795. IEEE, 2011. [116] Bernhard Buchli, Felix Sutton, and Jan Beutel. GPS-equipped wireless sensor network node for high-accuracy positioning applications. In European Conference on Wireless Sensor Networks, pages 179–195. Springer, 2012.

132

Bibliography

[117] L Benoit, P Briole, O Martin, C Thom, J-P Malet, and P Ulrich. Monitoring landslide displacements with the Geocube wireless network of low-cost GPS. Engineering Geology, 195:111–121, 2015. [118] Sara Casciati, Zhi Cong Chen, Lucia Faravelli, and Michele Vece. Synergy of monitoring and security. Smart Structures and Systems, 17(5):743–751, 2016. [119] Fabio Casciati and Clemente Fuggini. Engineering vibration monitoring by gps: long duration records. Earthquake Engineering and Engineering Vibration, 8(3):459–467, 2009. [120] YQ Ni, KY Wong, and Yong Xia. Health checks through landmark bridges to sky-high structures. Advances in Structural Engineering, 14(1):103–119, 2011.

Advanced Methods and Protocols for Wake-up Receivers

Advanced Methods and Protocols for Wake-up Receivers

Suggest Documents

Methods and Protocols Methods and Protocols

Methods and Protocols Methods and Protocols

Methods and Protocols Methods and Protocols - Laboratorium

Methods and Protocols Methods and Protocols - MAFIADOC.COM

Methods and Protocols Methods and Protocols

Advanced Radio Receivers

Methods and Protocols Methods and Protocols - Google Sites

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols - AMREPFlow

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols

Methods and Protocols