Integrated Optical Wireless
Communication and Positioning
by using Distributed
Multiple-Input Multiple-Output
Topology
vorgelegt von
M. Sc.
Sepideh Mohammadi Kouhini
an der Fakultät IV – Elektrotechnik und Informatik
der Technischen Universität Berlin
zur Erlangung des akademischen Grades
Doktorin der Ingenieurwissenschaften
-Dr.-Ing.-
genehmigte Dissertation
Promotionsausschuss:
Vorsitzender: Prof. Dr.-Ing. Falko Dressler
Gutachter: Prof. Dr.-Ing. Ronald Freund, MBA
Gutachter: Prof. Dr. rer.nat. habil. Volker Jungnickel
Gutachter: Prof. Ali Khalighi, Ph.D.
Tag der wissenschaftlichen Aussprache: 24. Mai 2023
Berlin 2024
Zusammenfassung
Zahlreiche Forschungsarbeiten haben das große Potenzial der optische Drahtlostechno-
logie sowohl für die Kommunikation als auch für die Ortung aufgezeigt. Die optische
Drahtlostechnologie benutzt das Licht als neues Medium für die mobile Vernetzung. Der
Bedarf an drahtloser Kommunikation und Ortung in Innenräumen wird aufgrund
der aufkommenden Anwendungen des Internets der Dinge drastisch steigen. Das
Funkspektrum ist jedoch überlastet, es kann nur begrenzt im Raum wiederverwendet
werden. Der herkömmliche Ansatz zur Verdichtung des Netzes reicht nicht aus, da
Interferenzen auftreten, die aufgrund der Ausbreitungseigenschaften von Funkwellen
innerhalb von Gebäuden nicht mehr effizient kontrolliert werden können.
Eine Möglichkeit, dieses Problem zu überwinden, ist der Einsatz der optischen
Drahtlostechnologie. Sie kann hohe Datenraten über Sichtverbindungen bereitstellen,
wobei die Ausbreitung zum Beispiel durch die Wände räumlich besser abgegrenzt ist. So
lassen sich Interferenzen im Vergleich zum Funk leichter bewältigen. Außerdem hat das
Licht einzigartige Eigenschaften. Es ist unempfindlich gegen, und erzeugt selbst keine
elektromagnetische Störungen und die Mehrwegeausbreitung ist vernachlässigbar. Beides
ist für die Kommunikation und Ortung interessant. Allerdings sind optische drahtlose
Verbindungen anfällig dafür, dass die Sichtverbindung unterbrochen ist. Kamerabasierte
Techniken zeigen, dass Licht für die Ortung genutzt werden kann. Jedoch fordern die
Nutzer eine enge Integration der Ortung mit der drahtlosen Kommunikation, anstatt
zwei Technologien einzusetzen. Es war ein offenes Thema, welche Präzision für die
Ortung mit den heutigen Kommunikationsprotokollen erreicht werden kann.
Ziel dieser Arbeit ist es, die Zuverlässigkeit und Robustheit der optischen drahtlosen
Kommunikation durch den Einsatz einer verteilten Multiple-Input-Multiple-Output
(MIMO)-Topologie zu verbessern und zu demonstrieren, dass dasselbe System auch für
eine präzise Positionierung verwendet werden kann.
Mit verteiltem MIMO kann man das Blockierungsproblem lösen und durch
redundante Pfade zuverlässige, robuste optische Drahtlosverbindungen realisieren.
Für die Verteilung der optischen Signale von der Zentraleinheit zu den verteilten
optischen Frontends wird eine optische Plastikfaser verwendet. Das ermöglicht eine
rein optische Übertragung über die Plastikfaser und dann über den drahtlosen
optische Kanal, was besonders robust gegen elektromagnetische Störungen ist. Die
Leistungsfähigkeit verschiedener Diversitäts- und Raummultiplextechniken wurde für
die Kommunikation mit einem und mit mehreren Benutzern verglichen und die
Ergebnisse zeigen die Vorteile gegenüber dem klassischem Zeitmultiplex auf. Die
Arbeit enthält neue Ergebnisse zur selektiven Kombination der Empfangssignale, zur
Winkeldiversität und zum Mehrnutzermultiplex. Das System wurde in verschiedenen
Szenarien getestet. Die Ergebnisse zeigen, dass ein adaptives Umschalten des räumlichen
Übertragungsverfahrens der beste Lösungsansatz ist, um Mobilität zu unterstützen und
gleichzeitig eine hohe Leistung zu erzielen.
Für die Integration von Positionierung und Kommunikation wird die Time-of-Flight-
Methode gewählt. Als Beispiel wird die Positionierung unter Verwendung des ITU-T
G.9991-Standards für optische drahtlose Kommunikation implementiert. Die Laufzeit
wird mit Hilfe der groben und feinen Zeitsynchronisation extrahiert, die bereits in
der physikalischen Schicht verfügbar sind. Am mobilen Gerät wird die Entfernung
zu mehreren Sendern gemessen und die 3D-Position mittels Trilateration berechnet.
Simulationen zeigen eine Genauigkeit kleiner 1 cm. Das System wurde im Labor, in
einem Konferenzraum und in einem Fertigungsszenario getestet und zeigt eine 3D-
Positionsgenauigkeit im Zentimeterbereich. Schließlich wurde auch die Online-Verfolgung
eines mobilen Objekts in einem Industrieszenario demonstriert.
iv
Abstract
A vast number of research works demonstrated the high potential of optical wireless
technology for both, communication and positioning. Optical wireless technology can
use the light spectrum as a new medium for mobile networking. The demand for wireless
communication and positioning for the indoor environment will increase dramatically
due to the emerging Internet of Things applications. However, the wireless spectrum at
radio frequencies is congested and it needs to be reused more frequently in the spatial
domain. However, the traditional approach to densify the network will not be enough
because of interference which becomes hardly manageable.
One way to overcome this problem is to use optical wireless technology. It can
provide high data rates through line-of-sight links, and confine the propagation in space,
e.g. by the walls in an indoor environment. Thus, interference is managed more easily
compared to radio. Moreover, the light has unique features. It is robust against and
does not create electromagnetic interference and has limited multi-path as well. Both
are unique features that make light interesting for communication and positioning.
However, optical wireless links are vulnerable to blockage. Although camera-based
techniques show that light can be used for positioning, end users requested tight
integration with wireless communications rather than deploying two technologies, one
for each purpose. It was an open issue of what precision could be achieved in this way.
The objective of this thesis is to enhance the reliability and robustness of optical
wireless communications by using a distributed multiple-input multiple-output topology
and demonstrate that the same system can also be used to realize precise positioning.
It is known that multiple links combat the blocking problem. Plastic optical fiber is
used to distribute the optical wireless signals from a central unit to the distributed
optical wireless frontends. This convergent, all-optical approach is robust against
electromagnetic interference. The performance of different diversity and spatial
multiplexing modes were compared for efficient communication with single and multiple
users the gains over classical time-division multiple access were quantified. The
thesis includes new results on selection combining, angular diversity, and multi-user
multiplexing. The system was tested in several environments. Results show that
adaptive switching of the spatial mode is the best candidate to support mobility and
realize high performance.
For integrating positioning with communication, the time-of-flight method is selected.
As an example, positioning is implemented using the ITU-T G.9991 standard for optical
wireless communications. The propagation time is extracted using coarse and fine time
synchronization which are already available at the physical layer in modern wireless
communication protocols. At the mobile device, the distance to multiple transmitters
is measured and the 3D position is calculated by means of trilateration. Simulations
indicate an accuracy of less than 1 cm. The system was tested in the lab, in a conference
room, and in a manufacturing scenario demonstrating 3D position accuracy in the
centimeter range. Finally, online tracking of a mobile object was demonstrated in an
industrial environment.
vi
Dedicated to Sasan, Majid, and Pouya and my dearest parents
Acknowledgements
This work present a summary of my research at OSRAM and Fraunhofer Heinrich Hertz
Institute (HHI) between July 2018 and February 2022. Writing a finishing touch to
my Ph.D. thesis and I cannot stop thinking: what an amazing journey! And there are
wonderful individuals to thank for supporting me to grow personally and professionally
during this journey. First and foremost, my supervisors, Prof. Dr. rer. nat. habil.
Volker Jungnickel and Prof. Dr.-Ing Ronald Freund for giving me the opportunity and
guiding me throughout this journey with their immense knowledge, critical reviews,
and constructive feedback. And for always being responsive despite their busy schedule.
I’d like to particularly thank my colleagues Dr.-Ing Christoph Kottke, Dr.-Ing Dominic
Schulz, Dr.-Ing. Anagnostis Paraskevopoulos, Sreelal Mavanchery Mana, Ziyan Ma
and colleagues from electronics group Peter Hellwig, Julian Hohmann and Jonas Hilt.
My examination board member and the Vision project coordinator, Prof. Ali Khalighi:
Thank you as well for your support and guidance, through the subtleties of scientific
writing. It has been a privilege working with, and learning from all of you.
I would like to thank my husband, Sasan, for his endless love and support. For his
constant encouragement throughout my research period. For being with me and for
your appreciated sacrifices. For cheering me up and dusting me off during hard times.
For believing in me at times when I did not. For always showing how proud he is of me.
What a blessing to have you beside me! I would like to present my sincere thankfulness
to my dear father and my beloved mother for their unconditional love and endless
support throughout my life. For without them I would not have had the support, which
I needed to be the person that I am today. Thank you both for giving me strength
to aim for the stars and chase my dreams. Thanks to my dearest brothers Majid and
Pouya, for their positive energy. For always being kind and cheerful and making me
laugh when things would get a bit discouraging. For their wise counsel and sympathetic
ear. Some special words of gratitude go to my colleague and best friend in Munich,
Bianca together with her husband, Helmut and Azadeh. For their endless support,
motivation, and heart-warming words during my Ph.D. journey. Your friendship makes
my life a wonderful experience. You are always on my mind. I would like to thank
my manager at OSRAM, Dr. Bernhard Siessegger and my supervisor Dr. Gerhard
Maierbacher for always supporting me, through transferring my VISION project to
Fraunhofer HHI and also my colleague Herbert Kastle for his valuable technical advises.
Special thanks to my friends who have always been a major source of support: Elnaz,
Nazila, Roya, Mehregan, who always managed to make me feel special. Thanks for
being amazing friends that can magically put everyone at ease. Having you as friends
made me a better and stronger person. Thank you all for always being there for me.
This work was supported in part by European project VisIoN, funded by the
European Unions Horizon 2020 research and innovation program and the EU
Horizon2020 Innovation Action ELIoT under the Marie-Sklodowska Curie grant
agreement number 764461 and 825651, respectively. I would like to thank the project
manager, Melanie Dorier for their support and make this possible. Thanks Vision’s
team, Fary Ghasemlooy, Stanislav Zvanovec, Rafael Perez Jimenez, Luis Nero Alves,
Murat Uysal, and Emrah Kinav for their responsive reviews. I also thank all Vision’s
ESRs for the constructive collaborations.
x
Table of Contents
Title Page i
Zusammenfassung iii
Abstract v
List of Figures xv
List of Tables xxi
1 Introduction 1
1.1 Optical Wireless Communication and Positioning . . . . . . . . . . . . . 1
1.2 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Publications.................................. 7
1.3.1 Publications as main author . . . . . . . . . . . . . . . . . . . . . 7
1.3.2 Publications as contributing author . . . . . . . . . . . . . . . . 11
2 Distributed MIMO for Communications and Positioning 13
2.1 Distributed MIMO Concept . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 CentralUnit ............................. 15
2.1.2 Fronthaul............................... 15
2.1.3 Optical Front-ends . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.4 MobileUnits ............................. 16
2.2 RelatedWork................................. 16
2.2.1 Using light for communication . . . . . . . . . . . . . . . . . . . 16
2.2.2 Using light for positioning . . . . . . . . . . . . . . . . . . . . . . 18
2.3 UseCasesforD-MIMO ........................... 19
2.3.1 Overview ............................... 19
2.3.2 Distributed MIMO for OWC in Industry . . . . . . . . . . . . . . 19
3 POF for Fronthaul 21
3.1 POFandLiFi................................. 21
3.2 Backhaul vs. Fronthaul for LiFi . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.1 POF-based transmission of OWC Signals . . . . . . . . . . . . . 22
3.2.2 Analog D-MIMO over POF . . . . . . . . . . . . . . . . . . . . . 22
3.3 Unidirectional SISO Proof of Concept . . . . . . . . . . . . . . . . . . . 24
xi
TABLE OF CONTENTS
3.3.1 POF-Link............................... 24
3.3.2 OWC-Link .............................. 24
3.3.3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.3.1 POF SISO - Link . . . . . . . . . . . . . . . . . . . . . 25
3.3.3.2 Analog POF+OWC link . . . . . . . . . . . . . . . . . 26
3.3.4 Link-Level Evaluation . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.5 Summary ............................... 28
3.4 Bidirectional LiFi over POF . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4.1 Gain Variation in a Meeting Room . . . . . . . . . . . . . . . . . 28
3.4.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.3 Performance Results Using LiFi over POF . . . . . . . . . . . . . 32
3.5 Conclusions.................................. 33
4 MIMO Communication 35
4.1 MIMO System Model for LiFi over POF . . . . . . . . . . . . . . . . . . 35
4.1.1 Spatial Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.1.1.1 Downlink.......................... 37
4.1.1.2 Uplink ........................... 38
4.1.2 Spatial Multiplexing . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.2.1 Linear Detection . . . . . . . . . . . . . . . . . . . . . . 40
4.1.3 Throughput Evaluation . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Distributed 2×2MIMO ......................... 42
4.2.1 SDM-over-POF Setup . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2.1.1 LiFiLink.......................... 42
4.2.1.2 Measurement Scenarios . . . . . . . . . . . . . . . . . . 43
4.2.2 Results ................................ 43
4.2.3 Summary ............................... 44
4.3 All-Optical Distributed MIMO: Spatial Diversity vs Spatial Multiplexing 45
4.3.1 Simulation Environment . . . . . . . . . . . . . . . . . . . . . . . 45
4.3.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3.2.1 Spatial Diversity . . . . . . . . . . . . . . . . . . . . . . 46
4.3.2.2 Spatial Multiplexing . . . . . . . . . . . . . . . . . . . . 47
4.3.2.3 MU-MIMO......................... 47
4.3.3 Simulation and Experimental Results . . . . . . . . . . . . . . . 48
4.3.3.1 Spatial Diversity . . . . . . . . . . . . . . . . . . . . . . 49
4.3.3.2 Spatial Multiplexing . . . . . . . . . . . . . . . . . . . . 51
4.4 Conclusions.................................. 53
5 Positioning 55
5.1 Positioning for Industry 4.0 . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1.1 Positioning of Production Resources . . . . . . . . . . . . . . . . 55
5.1.2 Positioning of Transport Systems . . . . . . . . . . . . . . . . . . 55
5.2 Wireless Positioning Techniques . . . . . . . . . . . . . . . . . . . . . . . 56
5.2.1 RF-Based Positioning . . . . . . . . . . . . . . . . . . . . . . . . 57
xii
TABLE OF CONTENTS
5.2.2 Camera-based Positioning . . . . . . . . . . . . . . . . . . . . . . 58
5.2.3 LiFi-based Positioning . . . . . . . . . . . . . . . . . . . . . . . . 61
5.2.3.1 Received Signal Strength Indicator . . . . . . . . . . . . 61
5.2.3.2 Finger Printing/Scene Analysis . . . . . . . . . . . . . . 63
5.2.3.3 Angle of Arrival . . . . . . . . . . . . . . . . . . . . . . 63
5.2.3.4 Phase Difference of Arrival . . . . . . . . . . . . . . . . 63
5.2.3.5 Time of Flight (TOF) . . . . . . . . . . . . . . . . . . . 64
5.2.3.6 Round trip Time of Flight (RTTOF) . . . . . . . . . . 65
5.2.3.7 Time Difference of Arrival (TDOA) . . . . . . . . . . . 65
5.3 Approach and Implementation . . . . . . . . . . . . . . . . . . . . . . . 65
5.3.1 Scenario................................ 66
5.3.2 Digital Signal Processing . . . . . . . . . . . . . . . . . . . . . . 67
5.3.3 MIMOPilots............................. 67
5.3.4 ChannelModel............................ 68
5.3.5 Frame Synchronization . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3.6 MIMO Channel Estimation . . . . . . . . . . . . . . . . . . . . . 70
5.3.7 TOFEstimation ........................... 70
5.3.8 RTTOF................................ 71
5.4 Measurement Setups and Results . . . . . . . . . . . . . . . . . . . . . . 72
5.4.1 Laboratory .............................. 74
5.4.1.1 Simulation Analysis . . . . . . . . . . . . . . . . . . . . 74
5.4.1.2 Measurements Results . . . . . . . . . . . . . . . . . . . 75
5.4.1.3 Summary.......................... 77
5.4.2 ConferenceRoom........................... 79
5.4.2.1 Tx and Rx FOV Characterization . . . . . . . . . . . . 79
5.4.2.2 Ranging........................... 80
5.4.2.3 Summary.......................... 81
5.4.3 ObjectTracking ........................... 82
5.4.3.1 Scenario and setup . . . . . . . . . . . . . . . . . . . . . 82
5.4.3.2 Results and Discussion . . . . . . . . . . . . . . . . . . 85
5.4.3.3 Summary.......................... 85
5.4.4 Post Optimization in a factory . . . . . . . . . . . . . . . . . . . 86
5.4.4.1 Measurement Scenario and Setup . . . . . . . . . . . . 86
5.4.4.2 Ranging and 3-D Evaluation . . . . . . . . . . . . . . . 86
5.4.4.3 Optimization . . . . . . . . . . . . . . . . . . . . . . . . 87
5.4.4.4 Results and Discussion . . . . . . . . . . . . . . . . . . 89
5.4.4.5 Summary.......................... 92
5.5 Requirements for Chipset Integration . . . . . . . . . . . . . . . . . . . . 92
5.5.1 Chipset Description and Current Capabilities . . . . . . . . . . . 92
5.5.2 Future Chipset Development . . . . . . . . . . . . . . . . . . . . 93
5.5.3 Increase the Precision of NTR Field . . . . . . . . . . . . . . . . 93
5.5.4
Provide Better Access to Lower Layer Information of Upper Layers
93
5.6 Conclusions.................................. 94
xiii
List of Figures
1.1 Electromagnetic spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2
Example of a distributed MIMO system for reliable optical wireless
communication (OWC) between the factory network and a mobile robot.
The robot is equipped with an OWC unit providing omnidirectional
coverage. Multiple optical front-ends (OFE) are arranged at the ceiling,
like in an illumination infrastructure, and provide cohesive and overlap-
ping coverage. A centralized digital signal processor (DSP) connects the
distributed OFEs via fixed links to enable MIMO communication. The
DSP is connected to the factory network. . . . . . . . . . . . . . . . . . 3
1.3
LiFi positioning integrated with wireless communications for intelligent
transport systems (ITS) in a smart factory. . . . . . . . . . . . . . . . . 4
2.1 Topology of distributed MIMO link setup for LiFi in an indoor scenario. 14
2.2
Example of LiFi systems for reliable wireless communications. LiFi
frontends are connected to the central unit (CU) through POF. (a) LiFi in
an aircraft (©istockphoto.com/ONYXprj). (b) LiFi in manufacturing [65].
(c) LiFi in a conference room (©Designed by vectorpocket/Freepik) [36]. 19
3.1
The analog transmission of OWC signals over POF is investigated. The
diagram shows the downlink only for simplicity, the uplink works similarly.
23
3.2
Block diagram of setup to establish the POF link. It shows the stages
for the LED driver, POF, and receiver units. . . . . . . . . . . . . . . . 24
3.3
Hardware implementation of POF and OWC link including POF AFEs
and two OWC units with a point-to-point configuration is shown. The
POF link has 10 meters in length and both OWC units are configurable
to act as an access point or as a client which, can operate for distances
over10m. ................................... 25
3.4
Setup to measure the frequency response of a POF link using a vector
network analyzer (VNA). . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5
Frequency response for the LED only and the LED including POF
for various fiber lengths, all measured with a wideband reference
photoreceiver in the lab. The response for 10 m POF connected to
a first custom-designed POF receiver is shown in Fig. 3.2. . . . . . . . 26
3.6
Diagram of the combination of the POF and OWC links for further
characterization. This corresponds to a SISO link of Fig. 3.1. . . . . . . 27
xv
LIST OF FIGURES
3.7
Frequency response of the OWC unit alone and integrated with the POF
link with various lengths. . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.8
(a) Overview of the measurement scenario in a conference room and
schematic view of the transmitter (Tx - red) and various receiver (Rx -
black) positions placed 215 cm below the Tx. (b) Measured frequency
response for various Rx positions in the meeting room. . . . . . . . . . . 29
3.9
(a) Block diagram of a LiFi over POF link. (b) Experimental setup of
the LiFi over POF link. Wireless distance and POF length are up to
1.5 m and 10 m, respectively. . . . . . . . . . . . . . . . . . . . . . . . . 30
3.10
Measured SNR spectra of the LiFi over POF link under different
configurations (a): Downlink SNR for 10 m POF and 50 cm OWC
link. (b): Uplink SNR for 10 m POF and 50 cm OWC link. (c):
Downlink SNR for 10 m POF and 150 cm OWC link. (d): Uplink
SNR for 10 m POF and 150 cm OWC link. The SNR performance vs.
frequency for the uplink and downlink is not symmetric as explained in
thetext..................................... 31
3.11 Uplink data rate vs. OWC distance. Gain is increased in 3 dB steps. . . 32
3.12 Downlink data rate vs. OWC distance. Gain is increased in 3 dB steps. 33
4.1 SDMapproach ................................ 36
4.2
Diversity combining was implemented as analog electronics in the
combiner unit, (a) maximum ratio combining with different weight
factors, (b) equal gain combining, which accumulates the receiving signal
with equal gain, (c) selection combining, which monitors the received
signal strength and select only the strongest ones. . . . . . . . . . . . . . 38
4.3
Single input single output (SISO) link block diagram of SDM-over-POF
setup. ..................................... 42
4.4 Experimental setup for D-MIMO over POF with SDM approach. . . . . 43
4.5
Down-link normalized singular values for SDM (a),
ξ1S1
and
ξ2S1
for
Scenario 1: spatially separated RX
d1
= 100 cm,
d2
= 70 cm,
d3
= 70
cm, and (b) Scenario 2:
ξ1S2
and
ξ2S2
of co-located RX:
d1
= 50 cm,
d2
= 5 cm, d3=35cm.............................. 44
4.6
Simulation Environment for (a) spatial diversity (SDIV), which consists
of four OFEs connected in series pattern and two mobile user (MU)s,
(b) Spatial multiplexing (SMUX) setup with two OFEs and two MUs.
Each MUs is equipped with two OFEs, which are tilted with angle α. . 46
4.7
spatial diversity (SDIV) measurement setup: In this setup, the CU
distributes the signal equally and sends it via POF to D-OFEs. On the
user side, there are two users with varying locations. . . . . . . . . . . . 46
4.8
SMUX setup: for evaluating spatial multiplexing (SMUX) feature in a
LiFi link. Each user is equipped with two OFEs [81]. . . . . . . . . . . . 47
4.9
Simulated SNR vs. Frequency for up-link, including EGC as well as SC
methods for user 1 (a) and user 2 (b), for the SDIV scenarios I, II, and III.
49
xvi
LIST OF FIGURES
4.10
Measured SNR .vs Frequency for uplink, including EGC and SC methods
for user 1 (a) and (b) user 2, for the SDIV scenarios I, II, and III. . . . 49
4.11
(a) Estimated throughput for different users locations considering both
EGC and SC. (b) Measured throughput for different users locations
considering both EGC and SC. . . . . . . . . . . . . . . . . . . . . . . . 50
4.12
Simulation results for singular values in the SMUX cell layout described
in Section V-C for
D3
=5cm for two users without and with angular
diversity for downlink (a) and uplink (b). . . . . . . . . . . . . . . . . . 51
4.13
Simulation results for singular values in the SMUX cell layout described
in Section V-C for
D3
=140cm for two users without and with angular
diversity for downlink (a) and uplink (b). . . . . . . . . . . . . . . . . . 51
4.14
Measured throughput of spatial multiplexing in downlink and uplink for
three defined scenarios and without and with angular diversity. . . . . . 53
5.1
LiFi positioning integrated with wireless communications for intelligent
transport systems (ITS) in a smart factory [91]. . . . . . . . . . . . . . . 56
5.2
Top view illustration of the factory floor, there is a fixed path for the
transport system, stockyards along the fixed path, and alternative paths
through the machine’s own stockyards [91]. . . . . . . . . . . . . . . . . 57
5.3
Scenario for radio-based positioning with moving intelligent transport
systems(ITS) between separate buildings indicating the problems of
radio-based signaling and handover due to interference from other rooms,
reflections at huge machines, walls and surrounding objects [96]. . . . . 58
5.4
Line scanning (yellow) in a camera picture of a ceiling with light source [91].
59
5.5
Signal fragments being stitched together from multiple frames to recover
thelampidentifier .............................. 59
5.6
Autonomous lamps emit unique codes. These are picked up by the phone
camera. An Internet link (Wi-Fi 5G,...) to a Signify–Philips database
allows the phone app to translate received codes and camera-measured
angles into positions. A customer can use these data in a proprietary
application[91]................................. 60
5.7
Improved scenario using LiFi for integrating positioning with wireless
communications solving problems from radio-based scenario regarding
interference and multi-path propagation and simplifying handovers for
moving,ITS[96]................................ 62
5.8 LiFi - based TDOA localization. . . . . . . . . . . . . . . . . . . . . . . 64
5.9
System architecture. (a) block diagram of OFDM transmitter and
receiver for positioning integrated with wireless communications. (b)
PHY frame structure, (c) Time flow chart for RTTOF measurements. . 66
5.10
(a): Block diagram of point-to-point distance measurement, The electrical
synchronization cable keeps a constant time offset between Tx and Rx,
and uses the same clocks, to simplify the measurement. In reality, the
positioning will be implemented as a bidirectional RTToF protocol so
that the cable is not needed. . . . . . . . . . . . . . . . . . . . . . . . . . 71
xvii
LIST OF FIGURES
5.11
Simulation results for combining coarse and fine timing for final distance
measurement. The SNR condition is good at this point, which leads to
estimating the distance without error. . . . . . . . . . . . . . . . . . . . 72
5.12 Simulation results for MSEs of x-, y- and z- dimensions vs. SNR. . . . . 73
5.13
1-D distance measurement (blue) with single Tx and Rx optical frontend
(OFE) and after using 10x averaging window (red). . . . . . . . . . . . . 74
5.14 Influence of signal bandwidth on normalized distance. . . . . . . . . . . 75
5.15
(a) Frequency response of the LiFi signal and (b) channel phase response
withlinearregression. ............................ 76
5.16
Estimated position through 40-iterations (o) and real position (
∗
), and
Txs position (∇)in3Dview. ........................ 77
5.17 Mean square error (MSE) of receiver for each axis after 40x averaging. . 78
5.18 X, Y and Z axis error detection in receiver position [127]. . . . . . . . . 78
5.19 The transmitter and receiver are tilted in various directions. . . . . . . . 79
5.20
Measurement setup including 4Txs and one Rx. Rx is placed in three
different locations of the transmitter cell. . . . . . . . . . . . . . . . . . 80
5.21
MSE of measured distance and actual distance based on Tx and Rx angle
rotation for D:3,4,5 m (D: distance between Tx and Rx) in log scale. Tx
angle rotation reaches beyond 70
◦
due to the large FOV of the photodiode.
81
5.22
MSE of measured distance and actual distance based on the simultaneous
rotation of Tx and Rx for D: 3, 4, 5 m. . . . . . . . . . . . . . . . . . . 82
5.23
MSE of estimated distance when Rx is placed in three different locations.
82
5.24
System architecture: (a) Transmitter signal, (b) Received signal (c) Packet Structure,
(d) Measurement setup showing LiFi transmitter (Tx)s overlap and thirteen location
indexes of mobile user (MU). ........................... 83
5.25
(a) Ranging root-mean-square-error (RMSE) for 2 m Txs setup (b) Ranging RMSE
of 1.5 m Tx setup, (c) x, y, z RMSE of receiver (Rx) for 2 m Txs setup, (d) x, y, z
RMSE of Rx for Txs setup of 1.5 m. ....................... 83
5.26
(a-d) Ranging RMSE heatmap for each individual Tx in the 1.5 m cell configuration,
each color indicates the corresponding error interval (e) online tracking results for a
mobile object. The object follows an arbitrary path, and its location is detected as
shown by circles. ................................. 84
5.27
(a-c) RMSE heatmap for each x, y, and z direction considering 1.5 m cell arrangement.
84
5.28
(a) Block diagram of OFDM transmitter and receiver for proposed LiFi positioning
system. (b) Measurements setup in the factory hall. Red dots show measurement points.
86
5.29
(a) Flowchart for evaluating regular (green boxes) and optimized (orange boxes)
distance as well as 3-D location information. (b) Flow chart for the initial calibration
measurement. (c) Ranging error and its linear correction factor as a function of the
angle. (d) Spatial relationship between Tx and Rx pair. .............. 87
5.30
(a) Sector segmentation of 35 marked reference points based on
T x4
.(b) Relationship
between ranging error and angle for four sectors for
T x4
. Blue curves: actual values
from calibration measurement; red curve: linear approximation; yellow curve: ranging
error with applied correction factor. ........................ 88
xviii
LIST OF FIGURES
5.31
(a) Ranging error of regular measurement for
T x4
for each position. (b) Ranging
error information of corrected optimized measurement of T x4for each position. . . . 89
5.32
(a) Original ranging RMSE heatmap of
T x4
. (b) Optimized ranging RMSE heatmap
of T x4...................................... 90
5.33
(a-c) Regular RMSE in X, Y, and Z-direction. (d-f) Optimized RMSE in each X, Y,
and Z-direction. ................................. 91
xix
List of Tables
3.1 Expected capacity according to equation 3.1. . . . . . . . . . . . . . . . 28
3.2
Measured Throughput in down- and uplink of OWC link and distributed
POF plus Optical wireless Link for several distances. . . . . . . . . . . . 33
4.1 Throughput evaluation of D-MIMO set up in two different scenarios. . 44
4.2 LiFi Cell Layouts, Scenarios, and Cases . . . . . . . . . . . . . . . . . . 48
5.1 SystemParameters.............................. 70
5.2 Regular and Optimized Ranging Measurement . . . . . . . . . . . . . . 90
5.3 Regular and Optimized 3-D Measurement . . . . . . . . . . . . . . . . . 92
xxi
Abbreviations and Symbols
AD angular diversity
ADC analog-to-digital converter
AF amplify-and-forward
AFE analog frontend
AGC automatic gain control
AGV automated guided vehicle
AI artificial intelligence
AOA angle of arrival
AP access points
API application programming interfaces
CFR channel frequency responses
CP cyclic prefix
CSI channel state information
CU central unit
D-MIMO distributed-MIMO
D-OFE distributed optical front ends
DAC digital-to-analog converter
DC-OFDM
direct current orthogonal frequency-division multiplex-
ing
DD direct detection
DF decode-and-forward
DMT discrete multi-tone
DSP digital signal processing
EGC equal gain combining
EMI electromagnetic interference
FOV field of view
FSO free-space-optical
FTM fine timing measurements
FWHM full width at half maximum
GPS global positioning system
IM intensity modulation
IoT Internet of Things
ITS intelligent transport systems
LA large area
LC light communication
xxiii
Abbreviations and Symbols
LOS line-of-sight
MD mobile device
MIMO multiple-input multiple-output
MMSE minimum mean square error
MRC maximum ratio combining
MS multi-stream
MU mobile user
MU-MIMO multi-user multiple-input multiple-output
NTR network time reference
OCC optical camera communications
OFDM orthogonal frequency-division multiplexing
OFE optical frontend
OOK on-off-keying
OWC optical wireless communication
OWP optical wireless positioning
PDOA phase difference of arrival
PHY physical layer
PLC power-line communications
PMMA polymethyl methacrylate
POF plastic optical fiber
PPM pulse position modulation
RF radio frequency
RN reference node
RoF radio-over-fiber
RSS received signal strength
RSSI received signal strength indicator
RTT round-trip-time
Rx receiver
SC selection combining
SDIV spatial diversity
SDM space division multiplexing
SISO single input single output
SMUX spatial multiplexing
SNR signal-to-noise ratio
SPI serial peripheral interface
SS single-stream
SU single user
SVD singular value decomposition
TDOA time difference of arrival
ToF time-of-flight
Tx transmitter
URLLC ultra-reliable and low latency communications
VLC visible light communication
VNA vector network analyzer
xxiv
Abbreviations and Symbols
WDM wavelength division multiplexing
WLAN wireless local area networks
ZF zero forcing
Symbols
H(f)Channel matrix
x(f)Transmit signal
y(f)Received signal
n(f)White Gaussian noise
N0Noise power spectral
BBandwidth
NRNumber of receiver
NUNumber of user
NTNumber of transmitter
σVariance
Z(f)Wireless MIMO channel
G(f)POF channel
NPD Number of photodiode
HLEDj(f)Frequency response of LED
gTX(ϕj)Radiation pattern
dj,l Distance between j-th OFE and the l-th user
HPDl(f)PD freuquency response
ϕjTransmitter radiation angle
ϕlReceiver radiation angle
g(dj,l)Wireless path loss
gRX(ϕl)PD sensitivity factor
w Weigth factor
d(1) Data signal
SNR Signal-to-noise-ratio
wMRC Maximum ratio combining weigth
Ggain for each received signal
wSC Selection combining weigth
ˆ︁
xZF Zero-forcing retrieved transmitted signal
ˆ︁
xMMSE MMSE retrieved transmitted signal
xxv
Abbreviations and Symbols
N0System noise
USquare unity matrics
VSquare unity matrics
λSVD channel matrix
RData rate
NNumber of subcarrier
∆BBandwidth of each subcarrier
ηHAverage path loss
ΓEmpirical scaling factor
cSpeed of light
riDistance between AP and MD
Ri,j Hyperbolas distances
θtAngle of transmitter
θrAngle of receiver
τPropagation delay
Aeff Effective area of PD
R0Radiation intensity
PSC Receiver correlation of SC
RSC Energy of SC
SSequence length
MSC Coarse timing matrics of SC
Strain Training symbol
ATraining symbol elements
NgNumber of guard sample
TbPong time
TaPing time
RTT Round trip time
eoffset Offset error
erandom Random error
Distanceest Estimated distance
Distanceact Actual distance
CFθiCorrection factor
mApprox Slope of linear approximation
PdCorrelation definition
PdEnergy definition
MTiming metrics
xxvi
1
Introduction
1.1 Optical Wireless Communication and Positioning
As new bandwidth-demanding Internet of Things (IoT) applications are growing rapidly,
an increasing number of IoT devices will need wireless connectivity to the Internet. The
next generation of IoT devices such as mobile robots, automated guided vehicles, drones,
and wireless endoscopes will produce a large amount of visual information coming from
cameras, lidars, ultrasonic imaging devices, etc., which will increase the demand for
wireless capacity. The wireless spectrum is crowded, however, particularly in unlicensed
radio-frequency (RF) bands. Moreover, the traditional approach to densifying wireless
access points will not be enough. New wireless spectrum will be needed to satisfy future
IoT demands. Further, there are applications in which the use of RF is limited or
not sufficient, such as in industry, medical or aviation scenarios where electromagnetic
interference is considered critical.
The ever increasing demand for mobile communications can be further satisfied by
the availability of a new spectrum and enhanced frequency reuse. Higher frequencies
are increasingly used, while cell size served by base stations and Wi-Fi access points
are accordingly reduced.
While 4G cellular radio uses frequencies at a few GHz, it is expected that 5G will
cover higher frequencies up to 75 GHz. For 6G, even higher frequencies in the Terahertz
and optical spectrum are considered. Optical components are well-established in various
applications and available in the mass market.
The entire electromagnetic spectrum is shown in Fig. 1.1. Optical Wireless
Communication can be divided into three main categories, denoted as free-space-
optical (FSO) communications, optical camera communications (OCC), and wireless
networking with light, which is also called light communications or light fidelity (LiFi).
FSO establishes point-to-point communication over large distances [1]. The FSO
system transmits invisible light from a laser through the atmosphere in a collimated
beam to the receiver, where it is focused onto a highly sensitive photon detector. FSO
1
1. Introduction
Figure 1.1: Electromagnetic spectrum
links have potential for high capacity in the Terabit/s range and the transmission is
through direct line-of-sight (LOS) [2].
OCC systems use commercial LED sources as transmitters and camera systems
as a receiver. The main challenge of OCC deployment is the low frame rate of those
cameras leading to quite limited kbit/s data rates [3].
LiFi is a promising area of research with the potential to provide reliable mobile
communication, for instance in industrial scenarios [4]. LiFi uses light as a medium
for mobile communications and operates in the unregulated optical spectrum without
interfering with radio frequency (RF) based systems. LiFi enables short-range links
in very small cells with a few meters diameter [5]. Further benefits include secure
communication due to spatial confinement of communication inside the the room, and
even more inside the light cone, robustness against electromagnetic interference (EMI),
jamming, and reuse of the existing illumination infrastructure.
Current LiFi systems can support many different applications [6]. Sufficient
coverage in the intended service area can be achieved by deploying a grid of networked
LiFi frontends e.g. in a distributed manner at the ceiling. Moreover, first standards like
G.vlc, IEEE 802.15.13, and 802.11bb become available as signs of growing maturity.
One of the applications is wireless connectivity in manufacturing environments
(see Fig. 5.1) [7]. The specific requirements for mobile communication in this scenario
are a high level of security, robustness against EMI, reliable communication with
moderate data rates (up to 100 Mbit/s with the potential of Gbit/s), and low latency
(few milliseconds) [8]. LiFi is a good candidate to serve these industrial requirements.
However, optical wireless communication (OWC) links depend strongly on the line of
sight (LOS), as even first-order reflections are 20 dB and more in the electrical domain
below the LOS signal [9]. Thus, OWC links are easily broken by shadowing or blocking
from standing or moving objects.
One way to overcome blocking is multiple-input multiple-output (MIMO). MIMO
is state-of-the-art in Wi-Fi and modern cellular networks [10], and it can improve both,
the data rate and reliability of wireless links [11].
This approach supports a higher network capacity, as well as lower latency because
mobility is supported by implementing MIMO algorithms at the lower physical and
2
1.1 Optical Wireless Communication and Positioning
Figure 1.2: Example of a distributed MIMO system for reliable optical wireless
communication (OWC) between the factory network and a mobile robot. The robot
is equipped with an OWC unit providing omnidirectional coverage. Multiple optical
front-ends (OFE) are arranged at the ceiling, like in an illumination infrastructure, and
provide cohesive and overlapping coverage. A centralized digital signal processor (DSP)
connects the distributed OFEs via fixed links to enable MIMO communication. The DSP
is connected to the factory network.
medium access layers in the protocol stack, unlike at the transport and higher layers as
in conventional mobile communication networks [12].
These advanced features aim at new services in the future Industrial Internet-
of-Things (IIoT), where the wired communication links connected to the devices
will be replaced by wireless connections that, however, interfere with each other.
IIoT applications can be divided into monitoring and machine-to-machine (M2M)
communication. For monitoring, machines collect small data packets e.g., device
position, device identifier, and a time stamp. Human interaction is very limited or not
necessary. Therefore, the reliability of the communication link is crucial. By means
of M2M communication, machines interact with each other to accomplish complex
tasks. Delays or any malfunction of the system would lead to a high risk for production
and human safety. Therefore, future IIoT requires very reliable communication links.
Moreover, the number of IoT devices connected to the cloud is estimated at up to 10
devices per square meter [13]. Future IIoT devices are expected to relay data from a
giant number of sensors to the cloud, in order to allow for artificial intelligence (AI)
based decisions in real-time.
In a distributed MIMO (D-MIMO) system, multiple optical front-end (OFE) units
are deployed in a distributed manner to achieve homogeneous coverage in the intended
area. In this way, the mobile device can be connected to multiple OFEs, which will be
essential for high availability [14]. On the other hand, transmitted and received signals
are fed from and into the joint DSP. D-MIMO is a new form of implementing networked
OWC. A recent proof-of-concept explored D-MIMO for OWC in a car manufacturing
cell at BMW’s robotics test lab [15]. Analog OWC baseband signals were transmitted
over twisted-pair cables usually used for Ethernet. Above 70 MHz, EMI was observed
due to FM radio from a nearby TV tower. Unintentionally, the long cables acted as
antennas [15].
3
1. Introduction
Figure 1.3: LiFi positioning integrated with wireless communications for ITS in a smart
factory.
For connecting the central DSP in Fig. 1.2 to the OFEs, there are several fixed home
networking technologies, e.g., power-line communications (PLC), coax cable and plastic
optical fiber (POF). OWC is regarded as an extension of these technologies into the
wireless domain.
In this thesis, the use of POF is studied for distributing OWC signals over short indoor
distances between centralized DSP and distributed OFEs, similar to the well-known
radio-over-fiber (RoF) concept [16]. POF is used as an optical relay technology because
it is simple, low-cost, and inherently robust against EMI like OWC.
Localization services besides reliable wireless communications are essential enablers
for smart manufacturing scenarios. Tools and robots are connected via wireless links
to a local cloud in which the information from numerous sensors and actuators are
collectively processed in order to control the entire workflow. There is also a big
trend towards the use of machine learning, i.e. artificial intelligence in these smart
manufacturing environments. The purpose is to react in real-time to situations and
events, ideally in a proactive manner, based on a previously learned set of methods. For
identifying the most appropriate method from a large database of previously learned
ones, it is essential to know the position of the tools or robots, as otherwise, the search
space might be too large and it is hard to make decisions in real-time.
There are numerous radio-based wireless localization technologies, among which
the global positioning system (GPS) is widely known as a navigation system for
4
1.1 Optical Wireless Communication and Positioning
outdoor environments. However, GPS is neither applicable nor precise enough in
indoor environments since RF signals from satellites experience severe attenuation
when entering into the buildings [17]. Moreover, the accuracy of the estimated position
depends on the available satellites besides weather conditions [18] and accuracy is limited
by multi-path propagation to 1-5 m [19]. For indoor localization, there are Wi-Fi-based
localization techniques [20]. These wireless local area networks (WLAN)s provide
wireless communications in many indoor environments by using unlicensed spectrum
and, thus, it is nearby to consider using them also for localization. WLAN positioning
can be implemented based on the received signal strength indicator (RSSI), together
with location-specific fingerprints, pre-collected from the channel state information
(CSI), which is reported by the mobile devices [21].
The precision reported for Wi-Fi systems is in the order of one or more meters,
using fine timing measurements (FTM), which are specified by the 802.11 standard.
The main reason for this is multi-path being the most relevant propagation mechanism
for radio waves indoors. Multipath results in fading and it is possible if the LOS signal
is received almost simultaneously with the first echo signal, that they cancel each other.
In that case, the next larger echo signal will be assumed as LOS, causing a significant
error. Alternatively, 60 GHz or UWB can be used [22]. These signals have a much
higher bandwidth, however, the signal-to-noise ratio is poor at high frequencies. Current
60 GHz systems require dedicated baseband processing chips, which are normally not
integrated within the current access points and mobile devices. Although LiFi has
lower bandwidth, it can reuse the standard baseband processing which is also used
for power-line communications or Wi-Fi, e.g. based on ITU-T G.9991 and 802.11bb
standards for LiFi. If the LOS is free, precise positioning may be directly integrated
into the existing WLAN access points, which could be extended by optical frontends
distributed at the ceiling.
Today, there is also a need for a wireless communication system for smart
manufacturing, which can provide accurate positioning besides high-speed wireless
communications. The main requirements are: i) it can be easily deployed in large indoor
areas such as a manufacturing hall, ii) uses wide enough bandwidth in an unlicensed
spectrum, and iii) is independent of interference from other rooms.
The coexistence of communication and positioning plays a crucial role in future
generations of communication. Therefore, a joint design and development is a matter
of discussion. Implementation of this coexistence requires shared hardware and other
resources, such as spectrum and base-band processing. Moreover, the convergence
of communication and positioning can provide an opportunity to develop a reliable
and robust system that can increase precision and help reduce the implementation
complexity of artificial intelligence. A wide range of approaches to integrate these
functionalities can be found for RF systems, however, thorough research for LiFi has
not been conducted yet. In this work, at first, an architecture is proposed for reliable
communication, which can be used for positioning as well. Both functionalities use the
same digital signal processing based on the standard G.9991.
5
1. Introduction
In this thesis, a localization system is proposed for LiFi based on the existing
physical layer (PHY) of the ITU-T specification G.9991 [23] in order to provide high-
resolution localization in an industrial environment. The proposed time-of-flight-based
localization technique is designed, implemented, and evaluated both by simulations
and experiments in a laboratory environment, followed by experiments in a conference
room and in an industrial environment, showing the quickly growing maturity of this
approach. The LiFi positioning system leverages information that can be obtained from
preambles already embedded in the existing PHY packet structure for synchronization
and channel estimation while using the frontends and PHY layer signal processing in
the existing LiFi systems based on the ITU recommendation G.9991. The technique
provides significantly improved accuracy compared to Wi-Fi-based solutions and can,
thus, satisfy the requirements of the fourth industrial revolution (Industry 4.0) [24, 25,
15, 26]. The deployment of positioning-enhanced LiFi systems fits nicely to assist the
transition to smart factories which desire reliable and secure wireless communication as
well as accurate positioning toward enhancing the productivity, efficiency, and flexibility
of the entire production process. As a consequence, it contributes to the future economic
growth.
1.2 Structure of the Thesis
Within the scope of this Ph.D. dissertation, a reliable optical wireless communication
and precise positioning technique are presented. The idea of using light for wireless
communication in the automated factory environment was originally presented in [27],
however, this technology was never adapted in the previous factories. Current and
future factories requiring reliable wireless communication at a moderate data rate in
the order of 100 Mb/s, low latency in the order of a millisecond, and robustness against
electromagnetic interference. Optical wireless communication intrinsically has confined
inside single rooms, is less interfered and the spectrum is license-free. Therefore, it
could be a good candidate for reliable communication in industrial manufacturing.
The general idea of this thesis is to develop a reliable communication and positioning
system. Therefore, an all-opticallink is proposed using distributed-MIMO (D-MIMO)
topology for reliable wireless communication and precise positioning.
The concept of D-MIMO for communication and positioning is elaborated in
chapter 2 by considering the related work, use cases, and particular benefits of light for
industrial wireless communications.
Chapter 3 covers fronthaul implementation using POF from the scratch to practical
implementation and first tests. At first, the POF as a medium is introduced, then
the backhaul and fronthaul concepts for LiFi are discussed. Then the initial proof of
concept for concatenating the wired and wireless optical links are described, where the
focus is on implementing an all-optical link by employing plastic optical fiber (POF) as
fronthaul.
Chapter 4 describes the D-MIMO communication system including two transmission
techniques denoted as spatial diversity and spatial multiplexing. Three experimental
6
1.3 Publications
setups are described for investigating the D-MIMO link step by step. At first, the 2
×
2
setup is presented as a proof of the MIMO approach, then the setup is extended to the
4
×
2MIMO system including spatial diversity at the transmission side and equal
gain combining (EGC) as well as selection combining (SC) at the receiver side. Both
combining methods are supporting when using POF as fronthaul. The framework for
the simulation and measurements is then presented. At the end, the multi-user MIMO
concept is tested in the 2
×
4MIMO setup, in combination with spatial multiplexing at
the transmitter side and angular diversity at the receiver side.
Chapter 5 focuses on developing positioning techniques over the optical wireless
medium. At first, various positioning techniques are introduced using an optical wireless
link. Then, the proposed time-of-flight method is elaborated. Three measurement setups
are presented in this chapter. First, the field-of-view characterization of transmitter
and receiver and its effect on ranging measurements are discussed. Secondly, the LiFi
positioning concept for Industry 4.0 is described, and finally, a 4
×
1MISO setup is
presented in both, simulation and measurement. At last, the results of real-time object
tracking through the proposed positioning system are demonstrated.
1.3 Publications
The research results obtained during this Ph.D. study have been submitted and published
in scientific journals and presented at a series of international conferences. This section
lists the peer-reviewed publications, which are the outcome of this Ph.D. thesis.
1.3.1 Publications as main author
1.
Kouhini, S. M., Jarchlo, E. A., Ferreira, R., Khademi, S., Maierbacher, G.,
Siessegger, B., Schulz, Dominic., Hilt, Jonas., Hellwig, P., Jungnickel, V., “Use
of Plastic Optical Fibers for Distributed MIMO in Li-Fi Systems”, Global Li-Fi
Congress 2019.
Previous research on optical wireless communication (OWC) focused mainly on
increasing the data rate for mobile broadband delivery. However, for new applica-
tions such as industrial wireless, reliability and robustness against interference
play a crucial role. This research focuses on the design, characterization, and
real-world testing of novel solutions for OWC taking industrial requirements into
account. Recent experimental works have demonstrated the feasibility of reliable
OWC in a manufacturing environment. Here, networked OWC is proposed and
implemented, which is also known as Li-Fi, by means of a distributed MIMO
approach enabling ultra-reliable and low latency communications (URLLC), which
is an important use case for 5G and beyond mobile networks. For distributing the
MIMO signals, plastic optic fiber (POF) is a promising low-cost solution offering
high data rates, easy deployment and inherent robustness against electromagnetic
interference. As POF may become a main component for Li-Fi systems using
distributed MIMO, commercially available POF solutions are studied and their
7
1. Introduction
usability for this new application to distribute signals between the central unit
and multiple optical frontends is discussed.
In this paper, I contributed by describing the scope of the proposed method,
built the first POF transmitter and receiver laboratory setup as a unidirectional
communication link, and performed initial measurements.
2.
Kouhini, S. M., Mana, S. M., Hellwig, P., Hilt, J., Schulz, D., Paraskevopoulos, A.,
Freund, Ronald and Jungnickel, V., "Performance of Bidirectional LiFi over Plastic
Optical Fiber (POF)." 2020 12th International Symposium on Communication
Systems, Networks and Digital Signal Processing (CSNDSP), IEEE, 2020.
In this work, a bidirectional LiFi link integrated with a plastic optical fiber
(POF) acting as fronthaul is reported to make the system fully robust against
electromagnetic interference in aircraft, industrial wireless, and offices.
I have contributed to implementing the POF link using two specifically developed
printed circuit boards interfacing directly with the DSP and the LiFi frontends.
The fronthaul transports analog LiFi baseband signals transparently over the
POF in a bandwidth of 200 MHz. The compound POF and LiFi link has
been tested in several configurations and achieved unprecedented throughputs of
725 Mbit/s and 901 Mbit/s in forward and reverse link direction, respectively. I
developed the bidirectional LiFi link integrated with a plastic optical fiber (POF)
acting as fronthaul, built the joined wired and wireless link, and performed the
measurement. I proposed to add a fixed gain between the wired and wireless links
to compensate for the path loss happening in POF and wireless link in down-
and uplink, relatively.
3.
Kouhini, S. M., Mana, S. M., Freund, R., Jungnickel, V., Corrêa, C. R. B.,
Tangdiongga, E., Cunha, T., Deng, X., Linnartz, J. P. M., "Distributed MIMO
Experiment Using LiFi Over Plastic Optical Fiber." IEEE Global Communications
Conference-Globecom 2020.
This paper shows the feasibility of a networked LiFi system using a distributed
multiple-input multiple-output (MIMO) link for optical wireless transmission and
a plastic optical fiber (POF) link as a fixed front-haul between distributed optical
front-ends, and a centralized signal processing unit. The concatenation of POF and
optical wireless links yields an easy-to-install all-optical LiFi system which is robust
against both, blockage of individual light beams and electromagnetic interference.
A significant cost-down appears by the use of colored LEDs to feed the POF link
with multiple optical signals, and wavelength division demultiplexing filters. The
spatial crosstalk in the wireless link and the spectral crosstalk over the POF link
can be jointly compensated by the same end-to-end MIMO processing. A common
signal model, which includes the combined effects of both links is provided to
characterize the proposed all-optical LiFi system. The first experimental findings
are reported when using space division multiplexing (SDM), i.e., multiple POFs,
and wavelength-division multiplexing (WDM), i.e., multiple colors in the same
POF, indicating that the performance is mostly limited by the wireless link.
8
1.3 Publications
Moreover, it shows that the positions of mobile users in the wireless link, as
well as gain variations and spectral crosstalk in the front-haul link, influence the
singular values and the achievable data rates of the LiFi system.
In this work, I provide a common signal model for space division multiplexing
POF system, performed the MIMO measurements, and analyzed the data of the
SDM link. I showed that the positions of mobile users in the wireless link, as
well as gain variations, and spectral crosstalk in the front-haul link, influence the
singular values and the achievable data rates of the LiFi system.
4.
Kouhini, S. M., Hellwig, P., Schultz, D., Freund, R., and Jungnickel, V., "Benefits
of MIMO Mode Switching, Angular Diversity and Multiuser Multiplexing for LiFi
," 2021 Optical Fiber Communications Conference and Exhibition (OFC), San
Diego, CA, USA, 2021.
The first real-time experiment with distributed MIMO and multiple users for
LiFi is presented in this paper. MIMO mode switching and angular diversity are
beneficial for robustness. Multiuser multiplexing helps in scenarios where users
have complementary MIMO channels.
In this research, I proposed to use angular diversity in combination with
spatial multiplexing. Moreover, I presented the first real-time experiments with
distributed MIMO and multiple users for LiFi.
5.
Kouhini, S. M., Kottke, C., Ma, Z., Freund, R., Jungnickel, V., Müller, M.,
Behnke, D., Vazquez, M. M., Linnartz, J. P. M. "LiFi Positioning for Industry
4.0." Journal of Selected Topics in Quantum Electronics (JSTQE)). IEEE, 2021.
In this article, a time-of-flight-based indoor positioning system for LiFi is presented
based on the ITU-T recommendation G.9991. Our objective is to realize
positioning by reusing already existing functions of the LiFi communication
protocol which has been adopted by several vendors. Our positioning algorithm is
based on a coarse timing measurement using the frame synchronization preamble,
similar to the ranging, and a fine timing measurement using the channel estimation
preamble. This approach works in various environments and it requires neither
knowledge about the beam characteristics of transmitters and receivers nor the
use of fingerprinting. The new algorithm is validated through both, simulations
and experiments. Results in a 1
m×
1
m×
2
m
area indicate that G.9991-based
positioning can reach an average distance error of a few centimeters in three
dimensions. Considering the common use of lighting in indoor environments and
the availability of a mature optical wireless communication system using G.9991,
the proposed LiFi positioning is a promising new feature that can be added to the
existing protocols and enhance the capabilities of smart lighting systems further
for the benefit of Industry 4.0.
In this collaborative paper, I contributed to developing the concept, developed
and tested the positioning algorithm and partially performed simulations and
measurements.
9
1. Introduction
6.
Kouhini, S. M., Ma, Z., Kottke, C., Mana, M. S., Freund, R., Jungnickel, V.,
"LiFi based Positioning for Indoor Scenarios." 2021 17th International Symposium
on Wireless Communication Systems (ISWCS). IEEE, 2021.
This paper presents a time-of-flight-based positioning approach based on
networked optical wireless communication, which is also known as LiFi. Our
proposed system is based on the ITU-T G.9991 standard and enables positioning
besides communication using the existing physical layer protocol. In this paper,
positioning performance is evaluated in a conference room. Results show that
the physical layer frame in G999.1 enables ranging with an accuracy of below 5
centimeters, besides enabling wireless communications. Moreover, Results show
that LiFi is able to cover a large area suitable for realistic indoor scenarios.
In this paper, I evaluated the LiFi cell characteristics as well as ranging
measurements for different locations. I extended the measurement scenario
from a laboratory setup to a conference room to evaluate the proposed position
performance in a realistic scenario.
7.
Kouhini, S. M., Ma, Z., Kottke, C., Mana, M. S., Freund, R., Jungnickel, V.,
"Object Tracking in an Indoor Scenario: Potential for Centimeter Accuracy with
LiFi." 2022 13th International Symposium on Communication Systems, Networks
and Digital Signal Processing (CSNDSP). IEEE, 2022.
In this paper, positioning is considered as an extra service offered by a LiFi
communication system. Our vision is to detect the required location information
of automatic guided vehicles, intelligent transport systems, and mobile assembling
units in an industrial environment in real-time, basically by using the same
hardware also used for wireless communication. Our proposed system is based on
the time-of-flight (ToF) technique and reuses physical layer mechanisms from the
ITU-T recommendation G.9991 for LiFi. Therefore, an initial analysis of object
tracking has been performed using LiFi and the accuracy of the ranging and
subsequent 3D position detection of the mobile device has been evaluated for two
different LiFi cell layouts in an indoor environment. Although using Matlab, it is
demonstrated for the first time that an object tracking speed of 1 point / second
at an average accuracy of 3 cm is possible. The proposed scheme is considered
promising for real-time implementation and potentially valuable to support future
Internet-of-Things applications.
In this paper, I contributed to optimizing the object tracking aMatlab code for
online operation as well as partially assisted during the measurements. Heat
maps of positioning errors were evaluated at different locations.
8.
Kouhini, Sepideh Mohammadi, Hohmann, J., Mana, S. M., Hellwig, P., Schulz, D.,
Paraskevopoulos, A., Freund, Ronald and Jungnickel, V. "All-Optical Distributed
MIMO for LiFi: Spatial Diversity versus Spatial Multiplexing", in IEEE Access,
2022.
10
1.3 Publications
In this paper, for the first time, an all-optical fixed-wireless LiFi communication
link is presented based on the distributed multiple-input multiple-output (MIMO)
topology. For distributing the wireless signals, plastic optical fibers (POFs) are
used as an analog front-haul. The operation of the distributed MIMO link is
studied in two basic modes, i.e. spatial diversity and spatial multiplexing. For
the diversity mode, a new POF combiner is presented which can support equal
gain as well as selection combining. It is demonstrated that selection combining
is highly effective and enables a similar LiFi performance in up- and downlink,
which is desirable for IoT. For the spatial multiplexing mode, it is observed that
the channel rank and the achievable throughput depend strongly on the user
location.
In this paper, for the first time, I presented an all-optical fixed-wireless LiFi
communication link based on the distributed multiple-input multiple-output
(MIMO) topology. As effective solutions, I studied the benefits of angular diversity
and MIMO mode switching together with multiuser multiplexing and concluded
that a dynamic switching between spatial diversity and spatial multiplexing is a
practical approach.
1.3.2 Publications as contributing author
During my Ph.D. studies, I further co-authored papers on other topics and the
following peer-reviewed papers:
9.
Jungnickel, V., Berenguer, P. W., Mana, S. M., Hinrichs, M., Kouhini, S. M.,
Bober, K. L., and Kottke, C. "LiFi for industrial wireless applications." 2020
Optical Fiber Communications Conference and Exhibition (OFC). IEEE, 2020.
10.
Kouhini, S. M., Alizadeh Jarchlo, E., Khademi, S., Ghassemlooy, Z., and
Alves, L. N., . "A Polynomial Analog Baseband Predistorter for Compensation
of Wireless Amplifier’s Distortion." 2020 12th International Symposium on
Communication Systems, Networks and Digital Signal Processing (CSNDSP).
11.
Mana, S. M., Kouhini, S. M., Hellwig, P., Hilt, J., Berenguer, P. W., and
Jungnickel, V. "Distributed MIMO Experiments for LiFi in a Conference Room."
2020 12th International Symposium on Communication Systems, Networks and
Digital Signal Processing (CSNDSP) . IEEE, 2020.
12.
Elnaz Alizadeh Jarchlo, Sepideh Mohammadi Kouhini, Hossein Doroud,
Elizabeth Eso, Piotr Gawłowicz, Min Zhang, Bernhard Siessegger, Markus
Jung, Zabih Ghassemlooy, Giuseppe Caire, Anatolij Zubow,"Analyzing Interface
Bonding Schemes for VLC with Mobility and Shadowing" 2020 12th International
Symposium on Communication Systems, Networks and Digital Signal Processing
(CSNDSP) . IEEE, 2020.
13.
Kai Lennert Bober, Sreelal Maravanchery Mana, Malte Hinrichs, Sepideh
Mohammadi Kouhini, Christoph Kottke, Dominic Schulz, Ronald Freund,
11
1. Introduction
Volker Jungnickel, "Distributed Multiuser MIMO for LiFi in Industrial Wireless
Applications" in Journal of Lightwave Technology, doi: 10.1109/JLT.2021.3069186.
14.
Jean Paul Linnartz, Carina Ribeiro Barbio Corrêa, Thiago Elias Bitencourt
Cunha, Eduward Tangdiongga, Ton Koonen, Xiong Deng, Matthias Wendt,
Anteneh A Abbo, Pieter J Stobbelaar, Piotr Polak, Marcel Müller, Daniel Behnke,
Marcos Martinez, Santiago Vicent, Taner Metin, Marc Emmelmann, Sepideh
M. Kouhini, Kai Lennert Bober, Christoph Kottke, Volker Jungnickel . "ELIoT:
New Features in LiFi for Next-Generation IoT." 2021 Joint European Conference
on Networks and Communications and 6G Summit (EuCNC/6G Summit). IEEE,
2021.
15.
Mana, S. M., Gabra, K. G. K., Kouhini, S. M., Hellwig, P., Hilt, J., and
Jungnickel, V. "An Efficient Multi-Link Channel Model for LiFi." 2021 IEEE
32nd Annual International Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC). IEEE, 2021.
16.
Jean Paul Linnartz, Carina Ribeiro Barbio Corrêa, Thiago Elias Bitencourt
Cunha, Eduward Tangdiongga, Ton Koonen, Xiong Deng, Matthias Wendt,
Anteneh A Abbo, Pieter J Stobbelaar, Piotr Polak, Marcel Müller, Daniel Behnke,
Marcos Martinez, Santiago Vicent, Taner Metin, Marc Emmelmann, Sepideh
M. Kouhini, Kai Lennert Bober, Christoph Kottke, Volker Jungnickel . "ELIoT:
Enhancing LiFi for a Next Generation Internet of Things." J Wireless Com
Network (2022).
17.
Mana, S. M., Gabra, K. G., Kouhini, S. M., Hinrichs, M., Schulz, D., Hellwig,
P., and Jungnickel, V.. "LIDAR-Assisted Channel Modelling for LiFi in Realistic
Indoor Scenarios." IEEE Access (2022).
18.
Ma, Ziyan, Kouhini M. S, Kottke, C, Freund, R, Jungnickel, V. "LiFi Positioning
and Optimization in an Indoor Factory Environment." Industrial Electronics
Conference (IECON). IEEE (2022).
12
2
Distributed MIMO for
Communications and Positioning
2.1 Distributed MIMO Concept
Multiple-input multiple-out (MIMO) is a system, where a base station with many
frontends simultaneously serves multiple users at the same time and frequency. These
frontends can be implemented in two manners: co-located or distributed.
In co-located MIMO architecture, all the frontends are placed nearby e.g. in an array
in a compact manner and all of them are controlled by the same baseband processing
unit. This architecture has the advantage of low back-haul requirements [28].
In contrast, distributed - MIMO or cell-free MIMO is a system where these
frontends are geographically distributed over the intended cell [29]. These frontends are
controlled via a central unit serving all users in the whole space-time-frequency resources
jointly. This architecture requires a fronthaul link between the central unit and each
distributed unit. Moreover, joint transmission and reception have tight synchronization
requirements.
Distributing wireless access points (AP)s is a well-known approach in cellular radio
networks to increase both, the coverage area and the capacity of a wireless network.
The idea behind this is frequency reuse, i.e. distant APs reuse the same spectrum [30].
However, the capacity is limited by interference, if the nearer AP reuse the same
spectrum. In MIMO systems, parallel signals from multiple frontends can be jointly
processed and interference can be canceled. Therefore, the capacity is high if there is a
minor correlation between the channels. Ideally, channel vectors that contain the links
between multiple APs and MU are orthogonal to each other.
While indoor radio wave propagation is dominated by non-LOS components, light
propagation is mostly based on the LOS. When using MIMO in optical wireless systems,
transmitters, and receivers should be sufficiently separated from each other, or at least
pointed in different directions, if co-located. Distributed MIMO (D-MIMO) systems
13
2. Distributed MIMO for Communications and Positioning
Figure 2.1: Topology of distributed MIMO link setup for LiFi in an indoor scenario.
deploy multiple frontends e.g. at the ceiling and communicate simultaneously with
multiple mobile users in the service area. Due to spatial separation, the correlation
between the channels is reduced.
In [31] it is demonstrated that, by using D-MIMO, improved spatial diversity
and higher data rate are achievable because the channel matrix is well-conditioned.
In [12], [32], [33] and [34], a D-MIMO architecture for LiFi system was proposed and
performance evaluated. Due to the LOS characteristics of light propagation, a D-MIMO
architecture using a coordinator, implemented as a central unit (CU) and multiple
optical frontends (OFE)s, is beneficial for Li-Fi.
In order to operate a D-MIMO link, a fronthaul is required to forward the signal
from the CU to all OFEs. Transmission over the fronthaul can be regarded as relaying.
decode-and-forward (DF) as well as amplify-and-forward (AF) are commonly used for
relaying [31].
In DF, first the received signal is decoded, then re-encoded and forwarded to the
user [35]. Moreover, the baseband processing is split into centralized (higher layer) and
distributed (lower layer) processing. The interface for this and the mapping of signals
e.g. into Ethernet frames is so far only defined for selected split points in advanced
cellular systems like 5G. Even though DF is finally favored in industry, this approach
is rather complex, it increases latency, and this effort pays out only in large numbers of
units.
In amplify-and-forward (AF), the received signal is amplified and forwarded through
the relay. The received signal contains some additional noise, therefore, by amplifying
the noise, the overall performance can be degraded. This is particularly important
in the uplink, where the received wireless signal can be very weak. To optimize the
performance, an automatic gain control (AGC) based on the received wireless signal
strength is required. In [36], a practical method for D-MIMO using AF over POF
is presented, where the AGC is co-located with the central unit (CU) and the POF
attenuation is compensated by electrical amplifiers [36].
Altogether, the distributed MIMO topology breaks into the following parts as shown
in Fig. 2.1:
14
2.1 Distributed MIMO Concept
•Central unit (CU)
•Fronthaul (FH)
•Optical Front-ends (OFE)
•Mobile units (MU)
In the following, each part is elaborated including related functionality.
2.1.1 Central Unit
In [12], the concept for D-MIMO in LiFi systems is laid out showing that this approach
can provide seamless mobility and low latency, besides huge gains at the cell edge. The
CU runs a fundamental channel estimation and feedback protocol to select the strongest
OFEs signals of each MU for joint transmission and reception. Because optical signals
have limited reach, unlike radio waves, implementation has reasonable complexity.
While this is a common approach in cellular radio systems, using the so-called Cloud
Radio Access Network (C-RAN) to centralize all signal processing [37], a minimalistic
version is specified in the IEEE Std 802.15.13-2023 for LiFi in industrial scenarios.
In the following section, it is assumed that the selection of the best serving, i.e.
nearby OFEs is already completed by the CU so that the signal from distant OFEs can
be ignored. This assumption is justified by the LOS-based light propagation, which
implies that the electrical signal reduces by a path-loss exponent of four even in indoor
environments 1.
2.1.2 Fronthaul
In order to transport LiFi signals between CU and each DU, a so-called fronthaul is
used. There are several analog media for distributing the CU signal to each optical
frontend, such as twisted-pair, coax cables, power-line communication (PLC) and POF.
PLC is widely used as a backhaul solution, however, the powerline capacity is shared
among several power outlets, and the diverse powering network also creates multi-path
effects. Moreover, PLC is restricted due to spectrum regulations, as the PLC medium
acts as an antenna, and interference with other services, such as FM radio, shall be
avoided. Coax cables are expensive and, besides satellite television, hardly used in
indoor environments. Twisted-pair cables enable reliable transmission, which is well
known for phone lines and Ethernet cables and have also been tested for LiFi [36].
However, proper shielding is needed to avoid unwanted emissions and make the cable
robust against electromagnetic interference, e.g. from nearby TV towers. An advantage
of electrical media is that powering can be supplied along with the fronthaul. Note
that optical frontends for LiFi can be co-located with the illumination infrastructure,
so that homogeneous coverage similar like from the existing luminaries is achievable,
and the powering is also available. Just the LiFi signals need to be delivered from the
CU to each OFE.
1
This is different for radio links, where the indoor path loss exponent value is between one and two,
typically, due to the dominant non-LOS propagation.
15
2. Distributed MIMO for Communications and Positioning
Here, POF is introduced as a practical solution for certain indoor environments
that are sensitive with respect to electromagnetic interference, such as aircraft, medical,
and manufacturing [36]
2
. Polymethyl methacrylate (PMMA) step index POF known
as a standard POF is an attractive solution for the fronthaul of LiFi systems [32]. POF
is robust against electromagnetic interference, has a small bending radius and is easy
and cheap to install, and allows up to 50 meters distance. The idea is to transport the
baseband signal for LiFi via POF from the CU to the DU in order to distribute the
optical frontends and achieve homogeneous coverage in one room. The work in [32]
contains early results for implementing distributed MIMO in LiFi systems by using
POF as an analog fronthaul.
2.1.3 Optical Front-ends
Current OFEs for LiFi are ceiling-mounted devices containing 4 high-power LEDs
and five photodiodes each equipped with its own trans-impedance amplifier (TIA),
the electrical outputs of which are finally combined. OFEs are in charge of providing
bi-directional communication between the CU and mobile units.
2.1.4 Mobile Units
Current MUs for LiFi are handheld devices, which typically have an OFE pointing
upwards, and the full base-band processing with a USB or Ethernet interface. There
has been recent progress to miniaturize OFEs for LiFi, however, further work is needed
to reduce energy consumption in both, the OFE and the baseband processor. Single
OFEs have a limited field-of-view of 90
◦
degrees full width at half maximum (FWHM).
If two MUs come close to each other, it is difficult to serve them jointly by means of
spatial multiplexing.
In this work, MUs with single and two OFEs are studied. If one OFE is used,
it points upwards. If two OFEs are used, they point upwards but tilted by angle
α
as
α
=
±
45
◦
. This is also denoted as angular diversity (AD). AD is helpful for
providing more than one LOS link to the distributed DUs at the ceiling and enhancing
the coverage area. In a D-MIMO link, AD can also enhance the rank of the MIMO
channel matrix, as looking into different directions can reduce the channel correlation
and allow the use of parallel communication channels between the MU and the OFEs
in the infrastructure [38]. The benefit of equipping LiFi receiver with angular diversity
(AD) mode is investigated theoretically in [38] and [39].
2.2 Related Work
2.2.1 Using light for communication
In the following sub-section, recent related works of using LiFi in positioning and
communication are elaborated. The early studies of using optical wireless as a local
2
While powering could theoretically be provided over the POF, it will not be enough to feed the
optical frontend in a LiFi system which is capable of providing homogeneous coverage in real LiFi cell
size having an area of few square meters.
16
2.2 Related Work
area network and using infrared wireless communication were proposed in [40], [41],
and [42]. Applications of using visible light, white LED, e.g. for transmitting audio
signals were proposed in [3]. For fronthaul implementation, previous works have
used conventional twisted-pair CATx Ethernet cables as a fronthaul acting partly
like antennas that lack the desired robustness against EMI [4], [43], [44]. In some
applications, glass or plastic optical fiber (POF) is more favorable due to its high
bandwidth and inherent robustness against EMI. In this thesis, the implementation of
a POF link as an analog fronthaul and a corresponding integration with a LiFi link
is reported, which is called LiFi over POF. The idea of using POF as a fronthaul has
been proposed in [45].
The first proof-of-concept for using POF as a fronthaul for OWC and the first
experimental setup for a bidirectional single-input single-output OWC over POF link
were presented in [46] and [36], respectively, reporting data rates up to 900 Mbit/s.
In [32], a first SDM over POF link in a star topology together with OWC links is
presented in a 2x2 MIMO setup. The performance is evaluated for various scenarios by
adapting MIMO principles from radio links to LiFi, where it is well known that the
throughput depends critically on the number of non-zero singular values of the channel
matrix, indicating separability, i.e., the ability to remove cross-talk by digital signal
processing without excessive noise enhancement. In a radio link with rich scattering,
an antenna separation of less than half of the carrier wavelength creates independent
separable channels [47, 10]. In contrast, intensity-modulated (IM) optical links must
create separability of signals in other ways, e.g. by spatial separation and different
orientations of the optical transceivers (angular diversity).
In [48] the concept of using wavelength division multiplexing (WDM) for POF
fronthaul implementation has been discussed. Two-channels WDM transmission over
POF has been shown in [49]. In this work, they report a total throughput of 2 Gbit/s
using discrete multi-tone (DMT).
A common objection against the use of OWC is that the link collapses when the
LOS is blocked. The MIMO characterization of indoor OWC link by considering diffuse-
transmission is discussed in [50]. The authors in [51] report preliminary experimental
results of using white LED for a four-channels MIMO system. Using spatial modulation
in combination with the MIMO technique for an indoor system is presented in [52].
The authors claim that by using spatial modulation, the achieved throughput is higher
compared to on-off-keying (OOK) and pulse position modulation (PPM), respectively.
An effective countermeasure proposed in [53] is the emission of light from different
spatially separated locations, similar to common practice in achieving homogeneous
illumination. Such configuration can be considered as a distributed-MIMO (D-MIMO)
link, by assuming that wireless access points in the infrastructure are the inputs and
the mobile devices are the multiple outputs. It has been demonstrated in [15, 26] that
the distributed MIMO architecture can significantly enhance robustness by sending the
same signal from multiple sources. In [34] it is shown, that the D-MIMO setup can
also be used to multiplex parallel data streams for multiple users if they are spatially
17
2. Distributed MIMO for Communications and Positioning
separated. An energy-efficient adaptive MIMO using visible light communication (VLC)
is presented in [54].
2.2.2 Using light for positioning
Numerous optical wireless positioning (OWP) techniques are reported in the literature
considering various methods which can be classified into the following categories: I)
received signal strength (RSS), ii) angle of arrival (AOA), and iii) Time-based algorithms
such as time-of-flight (ToF) and time difference of arrival (TDOA). All these techniques
can be extended into 3D by means of trilateration and triangulation and improved by
means of fingerprinting.
RSS-based positioning systems detect the distance between the transmitter (Tx) and
receiver (Rx) by transferring received signal strength into distance [55]. The accuracy
of the RSS system depends on a precise channel model, which describes how at higher
distances the received signal decreases. However, the RSS technique requires a very
precise and well-calibrated model between detected signal and distance, which varies
for each optical frontend and each propagation environment. Therefore, it can hardly
be used for the mass market.
The AOA technique estimates the receiver position by considering the angle between
the Tx plane and the LOS propagation path [55], [56, 57], [58]. The AOA can be
measured through image transformation or modeling in camera or photo-diode-based
systems, respectively. In camera-based positioning systems, accuracy depends on image
resolution and quantization errors [56, 57]. For photodiode-based systems, precise
channel modeling is required. The AOA of the signal could be fairly inaccurate, thus
yielding a large localization error in the positioning algorithm. The quantization error
in image processing is caused by setting a threshold for determining the light intensity
in pixels and further hampers the performance [55].
Fingerprinting is based on scenario analysis and has two phases [55]. In the offline
phase, environment features are collected in the form of received power, AOA or TDOA,
and stored in a database. In the second/online phase real-time data from the target
are obtained and compared to the database [59] [60]. Fingerprinting requires arduous
data collecting for each intended scenario.
Time-based positioning techniques comprise ToF, TDOA, and two-way ranging
[55]. These techniques are already used in radio frequency (RF) based positioning
systems. The TOF technique considers the speed of light in air and measures the travel
time of the light from the Tx to the receiver to obtain the distance between them [61].
TOF method requires strict synchronization between Tx and Rx. There are hybrid
methods such as TDOA/RSS for not fully synchronized systems [62]. However, these
methods are not applicable in various environments without requiring pre-knowledge of
the beam characteristics of Tx or Rx as well as accumulating fingerprinting data. In
this thesis, the TOF method is adopted for evaluating LiFi positioning through ranging
measurements.
18
2.3 Use Cases for D-MIMO
(a) (b) (c)
Figure 2.2: Example of LiFi systems for reliable wireless communications. LiFi
frontends are connected to the central unit (CU) through POF. (a) LiFi in an aircraft
(©istockphoto.com/ONYXprj). (b) LiFi in manufacturing [65]. (c) LiFi in a conference
room (©Designed by vectorpocket/Freepik) [36].
2.3 Use Cases for D-MIMO
2.3.1 Overview
Several use cases of using OWC in combination with POF link are summarized in
Fig. 2.2. Use in an aircraft (a) takes further advantage of the lightweight of POF, as
compared to heavier copper or aluminum wires. The second use case in Fig. 2.2 (b) is
manufacturing investigated in [63, 64, 65]. In the industrial IoT, OWC can provide
both, positioning and communication. Another application Fig. 2.2 (c) is to distribute
LiFi signals inside homes and offices. Among other options for signal distribution in
home networks (e.g. PLC and PoE), POF is useful to provide a connection through the
walls when considering that it is not conductive and can be collocated in the same ducts
with other electrical cables. Moreover, its performance is not dependent on devices
using electrical energy, and it does not suffer from RF signals also received by electrical
lines.
2.3.2 Distributed MIMO for OWC in Industry
Here the industrial use case is considered a bit more in detail. In an industrial
manufacturing process, any interruption can cause detrimental economic damage to
the manufacturer. For instance, in automotive manufacturing, one vehicle is produced
per minute nowadays [15]. Stopping the production line for one hour can cause an
economic loss of several Million Euros. Therefore, reliable communication plays a vital
role. In industrial use cases with high mobility, ensuring reliability is often not easily
doable. For instance, in robotic manufacturing, as demonstrated in Fig. 1.2, the robot’s
arm moves and rotates fast in 3D space so that the LOS can be blocked rapidly in the
order of 10 milliseconds. D-MIMO can then be used to overcome blockages and provide
reliable connectivity. While one link can be broken, this is unlikely for multiple links
in a distributed MIMO setup [15] where another path can be used to maintain the
communication link even in case of blockage.
19
2. Distributed MIMO for Communications and Positioning
Therefore, D-MIMO is considered as an efficient solution to overcome the
corresponding signal fades of 20 to 30 dB in the electrical domain during sudden
movements and rotations [15]. However, using twisted pairs for distributing OWC
signals needs considerable effort in terms of shielding, interconnecting, and possibly
special cables in order not to be sensitive to EMI. As EMI is regarded as critical for
industry scenarios, POF is considered a reliable low-cost solution to distribute the LiFi
signals from the central unit to the optical front-ends.
20
3
POF for Fronthaul
3.1 POF and LiFi
Standard POFs are made from polymethyl methacrylate (PMMA) as step-index 1-mm
core diameter fibers. POFs are an attractive fronthaul solution for feeding the LiFi
system due to their low cost, high data rates, immunity to electromagnetic interference,
easy installation, and small bending radius. Due to these advantages, POF is already
used in other sectors (automotive, industrial, in-home, etc.). In this thesis, the use of
POF as a feeder link in the LiFi infrastructure was studied to realize the D-MIMO
concept in the optical wireless link [36].
3.2 Backhaul vs. Fronthaul for LiFi
LiFi has the potential to complement RF-based solutions in scenarios where RF is not
permitted, restricted or causes interference such as in an aircraft, in manufacturing [66]
and office scenarios, as shown in Fig. 2.2. However, the distribution of the LiFi
signals is still a matter of debate. The classical approach is backhaul over powerline
communications (PLC). However, bandwidth is limited and shared among all LiFi
access points connected to the same powerline. Powering and backhaul can also be
provided via Power over Ethernet (PoE) in new lighting installations. The main problem
with backhaul is that the full PHY and MAC are integrated into each LiFi access point
so that the classical handover mechanisms can be used only operating in software at
the network and transport layers. The implications may be lost packets and delays,
depending on which handover solution is used.
A modern concept proposed for cellular networks and also for the evolution of Wi-Fi
is distributed MU MIMO. Distributed MU MIMO allows wireless links to be maintained
through multiple optical front-ends, which provides enhanced mobility support as well
as inherent robustness against blockage of the LOS, which can both be handled at the
lower PHY and MAC layers. This is faster and more reliable than at the higher layers.
21
3. POF for Fronthaul
By maintaining the backhaul connection to the network layer, the PHY and MAC layer
functions can be split at various points in the LiFi protocol stack between the central
unit (CU) performing the upper MAC and PHY functions and several distributed
units (DUs) performing the lower functions. Between CU and DU, there is a so-called
fronthaul.
There are simple fronthaul solutions that basically transport the analog base-
band signals from the CU, where all the MAC and PHY layer processing is done, to
the DU which consists of an optical frontend only, besides more complex solutions
(next-generation fronthaul, IEEE Std 1914.1, eCPRI) where the Ethernet protocol
is used as a transport medium while MAC and PHY protocol functions are indeed
split among CU and DU [45]. It has been demonstrated recently that the bandwidth
utilization and performance of both schemes are very similar. Ethernet-based fronthaul
is applicable in shared public networks while analog fronthaul also applies to private
network infrastructures such as the lighting infrastructure which can be reused to
deploy LiFi [67].
While an initial LiFi installation in an industrial scenario used twisted pair as
analog fronthaul [15], the idea here is to replace twisted pair with POF. In [68], POF is
introduced as a communication medium, which has low cost as well as high performance,
lightweight and it is easily deployable in EMI-sensitive areas. These properties make
POF a favorable fronthaul solution for the transport of LiFi signals between CU and
DU.
3.2.1 POF-based transmission of OWC Signals
A main disadvantage of electrical cables in manufacturing is robustness in harsh
EMI [69]. Optical communication is inherently robust and offers high data rates [70,
71]. In contrast to glass fibers, being widely used as a backbone in data centers, besides
local and wide area networks, polymer fibers are mechanically robust, lightweight, and
have potentially lower cost. POFs facilitate fast installation, flexible adaption, and
simple expansion using standardized connectors. A major reason for the low cost is
the relatively large core diameter of the POF, which is in the order of one millimeter.
It allows efficient coupling of visible light into and out of the POF besides the use of
LEDs and large-area silicon photodiodes (LA-PDs), respectively. In fact, LEDs and
LA-PDs are fast enough for current industrial applications. Fast modulation requires
sophisticated LED drivers and trans-impedance amplifiers with a bandwidth of several
hundred MHz. Optical frontends for high-speed POF transmission are available from
previous developments aimed at Gbit/s POF systems based on the IEEE Std 1914.1.
Both aim at ideal impedance matching similar to OWC.
3.2.2 Analog D-MIMO over POF
Figure 3.1 presents the D-MIMO system concept for transmission of OWC signals over
POF which is operated as a relay link in AF mode where analog baseband signals
are transmitted in a transparent manner between DSP and OFE. This scheme is
favorable for low latency and it needs no additional synchronization effort but may
22
3.2 Backhaul vs. Fronthaul for LiFi
Figure 3.1: The analog transmission of OWC signals over POF is investigated. The
diagram shows the downlink only for simplicity, the uplink works similarly.
have implications on the performance.
In Figure 3.1, data is passed into a MIMO DSP generating the OWC waveform.
Next, the waveform is transmitted over POF. Therefore, the signal is passed through the
electrical-to-optical converter (E/O) block, the modulated light then passes through the
POF, which can be up to several ten meters long and introduce multi-path due to modal
dispersion in the fiber, before being received at the optical-to-electrical converter (O/E)
where a small photocurrent is detected proportional to the received optical signal power.
This current is amplified and used to directly modulate the OWC-Tx whose signal is
transmitted over the optical wireless channel. At the OWC-Rx, the signal is retrieved
from the channel and amplified. Due to light propagation in free space, the optical
signal can be attenuated and received via multi-paths. The demodulation process and
data recovery are finally operated in the DSP-Rx where the received signal is impaired
by thermal noise and bandwidth limitation of the POF link before it is directly used
as a baseband signal in the OFE for the OWC link. OFEs and transmission over the
wireless channel add further distortion.
Accordingly, the signal is distorted twice, and the compound channel is used as a
single effective link for communication 1.
1
This is in contrast to the DF relay where the received signal is first decoded and corrected for any
errors before digital data are transmitted using a second waveform over the second link.
23
3. POF for Fronthaul
3.3 Unidirectional SISO Proof of Concept
First, a proof-of-concept for the POF-based OWC link has been set up in the lab. It
consists of unidirectional analog transmission through a POF link followed by an OWC
link.
3.3.1 POF-Link
The schematic for the POF transmission is shown in Fig. 3.2. Electrical-to-optical
and optical-to-electrical conversion is performed inside the LED driver and receiver
units section. The transmitter contains a 650 nm LED, which is driven by a high
bandwidth driver at frequencies up to 250 MHz. It converts the input voltage linearly
into an output current for the LED. Communication works via modulating the signal
onto the LED using intensity modulation (IM) of the LED [72]. The light is passed
through a step-index POF as a transmission medium with various lengths. The receiver
detects the light by direct detection (DD) [72] using an integrated fiber-optical receiver
by using a large area (LA) photodiode, together with the trans-impedance amplifier,
and provides linear conversion of modulated light into a differential analog electrical
signal. The TIA used with the detector must provide low noise over the intended signal
bandwidth above 150 MHz same as the OWC units used in this work. Therefore, the
TIA must be designed properly to take the capacitance and current characteristics of
the photodetector into account. Within the design procedure of the TIA, dynamic
range, cut-off frequency, temperature stability, and isolation from extraneous noise
are taken into account. The amplifier’s feedback loop must be stabilized by a suitable
phase margin compensation and sufficient gain. On the other hand, choosing the TIA
feedback causes a trade-off between gain and useful bandwidth. In addition, the DC
offset cancellation method has been used to avoid output saturation when connecting
the POF link to the OWC units.
Figure 3.2: Block diagram of setup to establish the POF link. It shows the stages for the
LED driver, POF, and receiver units.
3.3.2 OWC-Link
Figure 3.3 shows two OWC units capable of real-time optical wireless communication
at around 100 Mbit/s over a distance of 10 m. The OWC units have been developed
in a previous project. They consist of a DSP using orthogonal frequency-division
multiplexing (OFDM) over 100 MHz bandwidth with closed-loop adaptive bit loading
to address frequency-selective channel characteristics. All digital signal processing
follows ITU-T recommendation G.9960-2011 in the coax mode. The analog board
24
3.3 Unidirectional SISO Proof of Concept
Figure 3.3: Hardware implementation of POF and OWC link including POF AFEs and
two OWC units with a point-to-point configuration is shown. The POF link has 10 meters
in length and both OWC units are configurable to act as an access point or as a client
which, can operate for distances over 10m.
is custom-designed to allow high optical transmit power and high sensitivity using
off-the-shelf high-power LEDs operating at 860 nm (OSRAM SFH 4715 AS) and LA
silicon photodiodes (PD, Hamamatsu S6968).
The LED driver performs impedance matching for 4 LEDs operated in parallel producing
altogether 2.5 Watts of average optical power emitted into a beam width of 90
◦
full
width at half maximum (FWHM). 5 PDs with individual TIAs and equal gain combining
are used to capture enough light. The LED driver uses a fixed bias current while the
time-varying modulation current amplitude for the data signal is optimized to maximize
the data rate at the intended working distance.
2
Various orientations of the 5 individual
PDs with respect to the optical axis are supported in order to achieve a wider field of
view (FOV) of ≈90◦FWHM.
3.3.3 Result and Discussion
Next, the analog signal path has been characterized by considering analog tests of POF
and OWC links.
3.3.3.1 POF SISO - Link
Frequency response and dynamic range have been measured over the end-to-end POF
link as shown in Figure 3.4. The vector network analyzer (VNA) works as both, signal
source and analyzer. It provides a sine wave like a function generator whose frequency
is swept over time and measures at each frequency the received amplitude and phase.
In order to evaluate the performance transmitter, a high-bandwidth photo receiver has
been used for signal recovery. In this way, the pure bandwidth of the driver and LED
is estimated as shown in Fig. 3.5.
In the next step, the POF is connected to the PD to observe the performance of the
LED and POF together as illustrated in Fig. 3.5. In general, the bandwidth of POF
depends on the length of the fiber due to modal dispersion [73]. For short distances,
2
When modulation amplitude is increased, at first, more signal is received and data rate is increased.
At a certain modulation amplitude, however, the OFDM waveform gets clipped and the data rate will
be reduced, accordingly. The optimal modulation current is found easily by maximizing the data rate.
25
3. POF for Fronthaul
Figure 3.4: Setup to measure the frequency response of a POF link using a vector network
analyzer (VNA).
Figure 3.5: Frequency response for the LED only and the LED including POF for various
fiber lengths, all measured with a wideband reference photoreceiver in the lab. The response
for 10 m POF connected to a first custom-designed POF receiver is shown in Fig. 3.2.
bandwidth is not significantly changed at typical indoor distance up to 10 meters, see
Fig. 3.5. However, at frequencies above 100 MHz, the amplitude response shows a
higher attenuation including a notch attributed to modal dispersion in the POF [73].
Although the notch costs a fraction of the useful signal bandwidth, it is narrow
compared to the overall useful bandwidth. Adaptive OFDM will use both, the bandwidth
below and above the notch so it will not significantly reduce the POF link performance.
3.3.3.2 Analog POF+OWC link
The compound link consists of the combined POF and OWC links as shown in Fig. 3.3
and 3.6. The frequency response of the OWC link is measured in a similar way as in
the previous subsection and shown in Fig. 3.4 by using the OWC units. The blue line
in Fig. 3.7 shows the measured frequency response for the OWC link with a -3 dB and
26
3.3 Unidirectional SISO Proof of Concept
Figure 3.6: Diagram of the combination of the POF and OWC links for further
characterization. This corresponds to a SISO link of Fig. 3.1.
Figure 3.7: Frequency response of the OWC unit alone and integrated with the POF link
with various lengths.
-10 dB bandwidth of around 105 and 167 MHz, respectively. The analog POF+OWC
link is also shown in Fig. 3.7 by the yellow line with a -3 dB and -10 dB bandwidth of
71 MHz and 122 MHz for the POF with the length of 10 meters, respectively. This
indicates that the bandwidth of the POF+OWC link is sufficient for the manufacturing
scenario which requires moderate data rates no more than 100 Mbit/s.
3.3.4 Link-Level Evaluation
Finally, the link performance is characterized. The performance of a communication
link is limited by two main factors, bandwidth
B
and signal-to-noise ratio
SNR
, as this
is well known from channel capacity
C
formulated by the Shannon-Hartley theorem for
the capacity of a communication channel [73]. The highest achievable information rate
that can be communicated without error is obtained with the well-known formula.
C=B·log 2(1 + SNR)bits/s (3.1)
As the SNR depends on the frequency in the practical system, equation 3.1 has to be
applied at each frequency and averaged over all frequencies, finally. Complex modulation
schemes such as OFDM with adaptive bit-loading can exploit a total system bandwidth
that is typically higher than the -3dB bandwidth. Here, the assumption is it can be
better characterized by the -10 dB bandwidth. Based on an assumed
SNR
of 20 dB,
which is in practice limited by the wireless link, it is possible to estimate capacity by
using equation 3.1.
27
3. POF for Fronthaul
The result of capacity estimation for the individual POF and OWC links, as well
as the POF+OWC link, are shown in table 3.1. Current results already indicate
that the capacity is well above 100 Mbit/s and fulfills the requirements for wireless
communication in an industrial environment.
Table 3.1: Expected capacity according to equation 3.1.
Link Bandwidth (-10dB) Capacity
POF 1.5m 244MHz 1.07Gbps
OWC 167MHz 733Mbps
SISO chain with 10m POF 122MHz 535Mbps
3.3.5 Summary
In this section, the concept of using POF in distributed MIMO applications suitable
for industrial environments was proposed and experimentally investigated. This work
was focused on designing the analog communication link based on POF and OWC link
and, examining its main limitations. Experimental results showed that the requirement
of industrial communication can be met by the distributed SISO link. The next section
describes a further improved bidirectional link.
3.4 Bidirectional LiFi over POF
3.4.1 Gain Variation in a Meeting Room
Path loss variations in the wireless channel requires the design of automatic gain
control in the combined LiFi over POF link. In order to study the required dynamics
range, optical wireless channel measurements were performed in a (5
.
7
m×
4
.
5
m×
3
m
)
conference room at Fraunhofer HHI. The channel frequency response was measured
between 1 to 200 MHz by using a vector network analyzer (VNA) for many different
positions of Tx and Rx shown in Fig. 3.8. Raw data were calibrated by using the
apparatus function measured at 80 cm distance with free LOS [74]. Fig. 3.8 shows the
frequency response for various Rx positions in Fig. 3.8 (a).
The path loss increases with distance. The frequency response is flatter, when Rx
is nearer to the Tx due to the stronger LOS. More distant positions show more ripple
and a superimposed low pass at lower frequencies attributed to the diffuse reflections
from the walls and objects in the room. With free LOS, the normalized gain varies
between
−
20
dBel
and
−
40
dBel
, i.e. gain variation is 20
dBel
. If a stable operation is
targeted and the LOS is blocked, gain variations increase to around 35 dB [74]. This
comes close to outdoor RF systems [75]. Depending on the use case, an appropriate
automatic gain control needs to be implemented in the LiFi system.
28
3.4 Bidirectional LiFi over POF
(a)
(b)
Figure 3.8: (a) Overview of the measurement scenario in a conference room and schematic
view of the transmitter (Tx - red) and various receiver (Rx - black) positions placed 215 cm
below the Tx. (b) Measured frequency response for various Rx positions in the meeting
room.
3.4.2 Experimental Setup
A first laboratory setup for LiFi over POF is reported in [65]. This work has further
developed the experimental setup for bidirectional operation and used commercially
available POF OFEs. As a major step, the interface between optical wireless and
POF links needs further optimization. The current setup is intended for typical indoor
scenarios with up to 10 m POF and a few meters of wireless distance.
In the following, the best practice approach for the automatic gain control (AGC)
is described covering the required dynamic range of 40
dBel
taking a blocked LOS
also into account. As shown in Fig. 3.9 (a), starting in the downlink direction at
the central unit (CU), data is fed into the DSP-Tx to generate an adaptive OFDM
waveform following the ITU-T standard G.9991. The analog frontend (AFE) adapts
signals between the DSP and the POF link. A fixed-gain amplifier between the POF
and the OWC link handles the gain adaptation under normal operating conditions to
compensate for the additional path loss introduced by the POF. In the next stage, the
signal goes through the OWC units comprising the OWC-OFE, the optical wireless
channel, and OWC-OFE, AFE, and DSP at the device side.
How to compensate for the dynamics range, if a POF inserts in an optical wireless
channel, where the signal is attenuated and can experience multi-paths due to diffuse
29
3. POF for Fronthaul
(a)
(b)
Figure 3.9: (a) Block diagram of a LiFi over POF link. (b) Experimental setup of the LiFi
over POF link. Wireless distance and POF length are up to 1.5 m and 10 m, respectively.
light propagation? Due to mobility, the path loss is time-variant. Therefore, the signal
needs variable amplification before it can be detected by the digital signal processing
(DSP). Note that there are two contributions to the path gain, a fixed part due to the
insertion of the POF and a variable part in the wireless link.
In the downlink, adaptation can be static to compensate for the POF link. There is an
AGC in the AFE at the device side, which is directly controlled by the DSP through
an serial peripheral interface (SPI).
In the uplink, however, the DSP is located at the CU which also creates the SPI
control packet for the uplink. Moving the AFE to the DU is no option in practice
as the SPI packet needs very high bandwidth and cannot be provided in parallel to
the POF link to set the AGC during the first few
µ
s when an uplink packet is being
received. For practice, a solution was needed where the AFE was placed at the CU.
The solution to this problem leads to an essential condition to be fulfilled in a
combined LiFi over POF link: The additional loss introduced by the POF between
LiFi optical frontend at the DU and AFE at the CU shall be compensated by adding
an amplifier between the OWC Rx and the POF Tx, as shown in Fig. 3.9. The AFE
can compensate for a maximum dynamic range of 54 dB, controlled by the DSP. This
is sufficient to cover the expected gain variations in the wireless link caused by the
mobility of the LiFi device in typical scenarios. In our experimental LiFi over POF
uplink, the dynamic range is controlled by the uplink DSP via the SPI, directly attached
to the AFE like in the downlink. Fig. 3.9 (b) shows the experimental setup including
the developed transceiver POF (TRX-POF) printed circuit board interfacing the DSP
and LiFi front ends.
30
3.4 Bidirectional LiFi over POF
(a)
(b)
(c)
(d)
Figure 3.10: Measured SNR spectra of the LiFi over POF link under different
configurations (a): Downlink SNR for 10 m POF and 50 cm OWC link. (b): Uplink
SNR for 10 m POF and 50 cm OWC link. (c): Downlink SNR for 10 m POF and
150 cm OWC link. (d): Uplink SNR for 10 m POF and 150 cm OWC link. The SNR
performance vs. frequency for the uplink and downlink is not symmetric as explained in
the text.
31
3. POF for Fronthaul
Figure 3.11: Uplink data rate vs. OWC distance. Gain is increased in 3 dB steps.
3.4.3 Performance Results Using LiFi over POF
The LiFi over POF link performance is characterized by measuring data rates and
signal-to-noise ratio (SNR) vs. frequency in both link directions. Metrics are all
obtained from the configuration tool of the real-time DSP. The SNR spectra for the
downlink and uplink for several distances are shown in Fig. 3.10. The downlink is shown
for short and long wireless distances. If the wireless distance is increased, the received
signal is reduced and noise is added. For this reason, there is an overall reduction in
the SNR. Notice that the SNR in the downlink is limited at short wireless distances to
around 25 dB by clipping due to the gain setting in the LED driver.
In the uplink, a higher fixed gain was used due to higher path loss at the wireless
channel, i.e. the SNR at low frequencies is higher and the frequency response has a
more pronounced low-pass behavior. If the wireless distance is increased, the received
signal is reduced, i.e. overall SNR is lower and the noise becomes more visible at higher
frequencies.
In Fig. 3.11 and 3.12, the data rate is plotted vs. distance for various gain settings
between POF and LiFi link, for the uplink and the downlink. In each step, the gain is
increased digitally by 3 dB through the DSP software. Optical attenuation of the POF
is less than 200 dB/km. At 2 and 10 m POF lengths, optical loss is lower than 0.4 and
2 dB, respectively. The graphs show a trade-off between data rate, distance, and gain.
The gain impact is less noticeable in the uplink being designed with a less powerful
driver which keeps the link in a linear operation range to lower power consumption
which is important for battery-powered mobile devices. For that reason, the gain has
no noticeable effect. In the downlink, a powerful driver is used and some clipping is
tolerated. At short distances, the gain has a minor impact most likely because the
receiver is saturated. At larger wireless distances, the path loss is higher and the data
rate can be increased by increasing the gain and allowing for more clipping.
Further measurement results for variable wireless distance are summarized in
table 3.2. The best data rates at 10 m POF and 30 cm OWC distances are 725 Mbit/s
and 901 Mbit/s for downlink and uplink, respectively. Over 150 cm wireless distance,
32
3.5 Conclusions
Figure 3.12: Downlink data rate vs. OWC distance. Gain is increased in 3 dB steps.
Table 3.2: Measured Throughput in down- and uplink of OWC link and distributed POF
plus Optical wireless Link for several distances.
Link Configuration Throughput Downlink-Uplink
1-2 height OWC - 30cm 882 - 959 (Mbit/s)
OWC - 50cm 922 - 938 (Mbit/s)
OWC - 70cm 759 - 800 (Mbit/s)
OWC - 100cm 681 - 743 (Mbit/s)
OWC - 130cm 596 - 671 (Mbit/s)
OWC - 150cm 470 - 611 (Mbit/s)
POF 10m + OWC 30cm 725 - 901 (Mbit/s)
POF 10m + OWC 50cm 537 - 604 (Mbit/s)
POF 10m + OWC 70cm 446 - 435 (Mbit/s)
POF 10m + OWC 100cm 313 - 218 (Mbit/s)
POF 10m + OWC 130cm 239 - 145 (Mbit/s)
POF 10m + OWC 150cm 218 - 115 (Mbit/s)
representing a typical scenario, 115 and 218 Mbit/s are achieved in uplink and downlink,
respectively. These results show that more than 100 Mbit/s can be achieved by the
LiFi over POF approach anywhere in the coverage area.
3.5 Conclusions
In this section, it was highlighted that distributed MIMO is a promising approach
for LiFi which will find many applications e.g. in aircraft, industry, and offices as
it provides a consistent data rate in the area covered by multiple optical frontends
and higher robustness in case the line-of-sight is blocked and against electromagnetic
interference. By using analog LiFi over the POF approach, the LiFi signals can be
distributed from the central unit to the remote optical frontends.
Measurements indicate that 20 and 40
dBel
path loss variation has to be covered
in mobile scenarios with and without line-of-sight, respectively. It is argued that the
33
3. POF for Fronthaul
dynamic gain variation in the wireless link can be fully compensated at the central unit
after the POF link if, in both link directions, an amplifier is introduced at the optical
frontend in order to compensate for the fixed losses in the POF link. A bidirectional
LiFi over POF link was tested in a laboratory setup. Experimental results show that
maximum data rates of 901 Mbit/s and 725 Mbit/s can be obtained in up- and downlink,
respectively, while the minimum rates were 115 and 218 Mbit/s all essentially limited
by the wireless link. These results show that the LiFi over the POF approach is useful
in all intended use cases of LiFi with mobile users in the industrial scenario.
34
4
MIMO Communication
In this chapter, the bidirectional LiFi over POF link is extended to MIMO. The first
objective is to developed a mathematical model of the distributed MIMO link using
concatenated wired and wireless links, where either SDIV or SMUX can be used. Next,
each MIMO mode transmission is evaluated to enable robust high-speed communication
over the LiFi link. The final objective is to assess the proposed setup through simulations
and experiments for various MIMO setups to obtain quantitative results. In the
performance analysis, SNR versus frequency and the achievable throughput are used as
figures of merit.
4.1 MIMO System Model for LiFi over POF
The feeder network can be realized by individual point-to-point connections using one
POF for each OFE and the same wavelength, which is denoted as SDM as shown in
Fig. 4.1. Using SDM, there is no crosstalk in the feeder network.
In this section, the mathematical model for the compound MIMO channel matrix
H(f)
of the D-MIMO link is presented [32]. This model describes the concatenation of
SDM over POF and wireless links in the frequency domain including the propagation
behavior of the parallel signals in the channel. In this model, direct current orthogonal
frequency-division multiplexing (DC-OFDM) is used as a modulation scheme. It is
assumed that each AP is equipped with
NT
light transmitters all working at the
same peak wavelength. The received signal vector from each user at the
f
-th OFDM
sub-carrier is computed by [32]:
y(f) = H(f)x(f) + n(f),(4.1)
where
H(f)
is the
NUNR×NT
end-to-end analog channel matrix,
x(f)
is the
NT×
1
transmit signal vector,
y(f)
is the received signal vector, which contains
NUNR
signals
from
NU
users each having
NR
OFEs, and
n(f)
is the additive white Gaussian noise
vector.
35
4. MIMO Communication
Figure 4.1: SDM approach
The noise vectors
n(f)
can be modeled as complex-valued Gaussian distribution
with zero-mean and variance
σ
as
σ2
n
=
N0B
, where
N0
denotes the noise power spectral
density and
B
is the modulation bandwidth [32]. The power is distributed equally over
all OFDM subcarriers . The overall channel matrix H(
f
)including wired and wireless
channel can be written as:
H(f) = Z(f)G(f)(4.2)
where
Z(f)∈RNUNR×NT
is the wireless MIMO channel and
G(f)∈R(NPD=NT)×NT
denotes the SDM over POF link.
The coefficients of Z(f)are computed by [32]:
Zl,j(f) =HLEDj(f)gTX(ϕj)g(dj,l)gRX(ϕl)HPDl(f),(4.3)
Where
HLEDj(f)
represents the frequency response of the LED plus driver,
gTX(ϕj)
gTX
(
ϕj
)stands for the radiation pattern,
dj,l
is the distance between
j
-th OFE and
the
l
-th user,
ϕj
,
ϕl
are the angles, which the ray radiates from the transmitter and
to the receiver respectively,
g(dj,l)
denotes the wireless link path loss,
gRX(ϕl)
is PD
sensitivity factor at user side, and
HPDl(f)
serves as frequency response of PD besides
trans-impedance amplifier in OWC channel [76, 77].
In general, the POF is a color-dependent medium. However, in the SDM scheme
which is considered here, signals are transmitted via separate POF links all with similar
36
4.1 MIMO System Model for LiFi over POF
wavelengths and, therefore, there is no cross-talk in distributing the signals from CU
to the OFEs. In such case, G(
f
) =
G0
I
NT
is diagonal, where
G0
represents the POF
channel gain.
At the APs, the overlayed signals are forwarded over the wireless channel to the
mobile user.
Hl,i
(
f
)describes each path within H(
f
)served as travelled signal from
each transmitter
i
-th to each
l
-th user, for
j
= 1
, ...,
(
NPD
=
NT
), and corresponding
j-the OWC LED, which can be computed by:
Hl,i(f) =
NT
∑︂
j
Zl,j(f)Gj,i(f)(4.4)
The OWC cross-talk is included by the sum of the transmitted signals. This cross-
talk can be discarded by MIMO pre-processing and the combined channel is written as
Hl,i
(
f
) =
Zl,i
(
f
)
Gi,i
(
f
). Further, for the uplink channel, the signal first goes through
the wireless channel Zand then through the POF channel G.
By canceling the wireless cross-talk, parallel channels are created over SDM POF
and wireless link.
4.1.1 Spatial Diversity
In spatial diversity mode, the same data stream is transmitted from multiple
transmitters. This method increases the reliability of the transmission. In the following,
the mathematical model for both, downlink and uplink including different combining
methods is explained.
4.1.1.1 Downlink
For the downlink performance, consider two users. The received signal for each user
can be expressed as y(1) and y(2). The received signal for user one is calculated by:
y(1) =(︂h(1)
1h(1)
2h(1)
3h(1)
4)︂.
x1
x2
x3
x4
+n(1) (4.5)
The same CU signal transmitted over all OFEs, which is also denoted as spatial
repetition coding, therefore is defined x(1) as:
x(1) =w .d(1) =
1
1
1
1
.d(1) (4.6)
where
w
is 1 due to the repeated signal, and
d(1)
is the data signal. For estimating the
SNR for each user in the downlink, the following formula can be used:
SNR(1) =|h(1).w˜|2.|d(1)|2
|n|2(4.7)
37
4. MIMO Communication
Figure 4.2: Diversity combining was implemented as analog electronics in the combiner
unit, (a) maximum ratio combining with different weight factors, (b) equal gain combining,
which accumulates the receiving signal with equal gain, (c) selection combining, which
monitors the received signal strength and select only the strongest ones.
Where nis the additive white Gaussian noise.
4.1.1.2 Uplink
In the following, the uplink performance is modeled.
y1
y2
y3
y4
=
h1
h2
h3
h4
. x +
n1
n2
n3
n4
(4.8)
where
y
is the received signal,
x
is the input signal and
n
is the noise. In order to
retrieved the sent data (ˆ︁
x), the signal processing can be written as:
ˆ︁
x=
w1
w2
w3
w4
. y (4.9)
Where
w
is defined based on the chosen diversity combination method. For diversity
combining, the following methods are used.
38
4.1 MIMO System Model for LiFi over POF
4.1.1.2.1 Maximum Ratio Combining
Maximum ratio combining (MRC) is considered as an optimal combining solution in
radio systems in the presence of noise as shown in Fig. 4.2 (a). In this method, each
received signal is multiplied by a weighting factor proportional to the channel amplitude
in the respective receive branch and then all weighted signals are summed up. However,
implementation in real-time requires complex signal processing and therefore changes
in the chipset for measurering the channel at each received branch. In the case of MRC,
the weight vector wMRC is defined as eq. 4.10, where each wiexpressed as eq. 4.11.
wMRC =
w1
w2
w3
w4
(4.10)
wi=hi
√︂∑︁i
j=1 |h2
j|
(4.11)
4.1.1.2.2 Equal Gain Combining
This technique applies equal gain to contribution of all received signals in the single-
stream (SS) transmission mode. Fig. 4.2 (b) shows the EGC block diagram, in which
G
is the gain for each received signal with same value. In this method, estimation of the
channel for all received signals is not required and hence, it is simple and cheaper for
practical implementation. In case of EGC,
wi
is equal to 1. The authors in [78] show
that EGC reduces the SNR variations but suffers from noisy channels. In theory, the
MRC scheme can outperform EGC when these weights can be estimated and adjusted
ideally.
4.1.1.2.3 Selection Combining
This technique chooses the received branches with sufficient signal stengths and combines
them, as shown in Fig. 4.2 (c). Therefore, the signals need to be monitored at the
receiver and compared to a threshold. In our implementation, an analog comparator
was used, which checks the received signal power with a fixed, but externally adjustable
reference value. If the signal is above the reference value, the signal is used in the
combining process. For
i
= 0
, .., n
, which
n
is branch of received signal the
wSC
will be
defined as:
wSC =
w1
w2
w3
w4
(4.12)
wi=
1if |yi| ≥ Pthreshold
0otherwise
39
4. MIMO Communication
4.1.2 Spatial Multiplexing
To increase the channel capacity without additional spectral resources, SMUX is used.
In this technique, independent data streams are simultaneously transmitted through
multiple transmitters and the received signal at each receiver corresponds to the
combination of multiple data streams from all the transmitters [79]. Recent research
[80, 81] confirms that this technique can increase the channel capacity without extra
bandwidth or transmit power. However, the detection techniques of this method are
slightly more complex than SDIV.
4.1.2.1 Linear Detection
In this study, measurement modules used a linear detection technique. In general, to
retrieve the transmitted signal linearly the following equation is used:
ˆ︁
x=Wy (4.13)
Below, a brief description of the commonly used zero forcing (ZF) and minimum
mean square error (MMSE) linear detection techniques are presented.
4.1.2.1.1 Zero Forcing
To retrieve the transmitted signal, the left-handed Moore-Penrose pseudo-inverse of the
channel matrix
H
is obtained using a weighting matrix
W
as
W×H
=
I
to reconstruct
the transmitted signal [79] based on the following:
ˆ︁
xZF =W y (4.14)
ZF :W= (HHH)−1HH(4.15)
Therefore, the retrieved transmitted signal through zero-forcing can be computed
as ˆ︁
xZF :
ˆ︁
xZF =x+W n (4.16)
where
HH
is hermitian matrix of MIMO channel matrix
H
,
n
is noise vector of the
system. The drawback of ZF is the enhancement of the noise, which can degrade the
performance. Accordingly, ZF is suitable for conditions where SNR is high.
4.1.2.1.2 Minimum mean square error
MMSE is also a linear equalizer using a modified filtering matrix compared to ZF. The
weight matrix for MMSE can be defined as [79]:
MMSE :W= (HHH+N0I)−1HH(4.17)
Therefore, the transmitted signal is retrieved by MMSE as ˆ︁
xMMSE:
40
4.1 MIMO System Model for LiFi over POF
ˆ︁
xMMSE =W y =WHx +Wn (4.18)
Where
N0
term refers to the noise of the system and
I
is the identity matrix.
This method can overcome the noise enhancement problem of ZF method [79] and is
therefore better suited at low SNR.
4.1.3 Throughput Evaluation
To measure the performance in SMUX mode, it is better to refer to the singular value
decomposition, rather than using ZF and MMSE, because the SVD leads to an upper
bound that is not reached by these schemes. As known from the RF channel, singular
value decomposition (SVD) is an effective mathematical tool for evaluating throughput
in MIMO channels [82, 83]. In this section, the principle of SVD calculation is used
considering that the channel information is available at LiFi transmitters and MU. The
SVD tool converts the MIMO system into parallel channels, where the number of these
channels is equal to the rank of the MIMO channel matrix. At the
f
-th subcarrier
H(f), it can be written:
H(f) = U(f)D(f)V(f)H(4.19)
The
U
and
V
are square unitary matrices and
D
is a quasi-diagonal matrix, which
contains the singular values known as
λ
in its diagonal elements. These can be defined
as:
U(f)∈CNUNR×NUNR(4.20)
V(f)∈CNT×NT(4.21)
D(f)∈RNUNR×NT
+(4.22)
Based on [84, 85],
λ1
(
f
)
≥λ2
(
f
)
≥... ≥λK
(
f
), where
K
=
min
(
NUNR, NT
)
denotes the rank of the channel matrix H(f).
In the SDM feeder link, there is no crosstalk due to using different POF links.
However, the optical wireless link suffers from crosstalk due to several parallel channels
overlapping in the spatial domain. In [32], the achievable throughput estimation
method of [86] was used based on the normalized channel matrix of [82], for throughput
evaluation. The data rate Rcan be expressed as [32]:
R= ∆B
N
∑︂
n=1
K
∑︂
k=1
log2(︃1 + SNR
NTηHΓλ2
k(fn))︃(4.23)
where
N
is the number of subcarriers,
∆B
the bandwidth occupied by each subcarrier,
ηH
the average path loss of H(
f
)and
Γ
= 10 an empirical scaling factor taking into
account the impairments like non-linear distortion (clipping) and imperfect constellation
shaping [87].
41
4. MIMO Communication
4.2 Distributed 2×2MIMO
In the following section, the experimental setup for the SDM link and LiFi link is
described. Furthermore, the D-MIMO measurement scenario is represented.
4.2.1 SDM-over-POF Setup
The experimental link for the single input single output (SISO) SDM-over-POF
transmission is shown in Fig. 4.3. The Avago transceiver (AFBR-59F3Z) is used
as both, transmitter and receiver for the POF link. It contains a 650 nm LED
together with a driver at the transmitter side and a PIN photo-diode together with
a linear trans-impedance amplifier at the receiver side for transparent electro-optical
and optoelectrical conversion, respectively. The transmitters are directly modulated
within the linear region using Direct Current Offset (DCO)-OFDM. The waveform
is generated by Matlab and loaded onto the parallel arbitrary waveform generator
(AWG) Spectrum DN 2.662-08. At the receiver side, the received signal is plugged into
a parallel analog-to-digital converter Spectrum DN 2.445-08 (also denoted as digitizer)
by using a sampling rate of 500 MS/s. The received signals have been analyzed through
offline Matlab processing, as described in [15]. The POF length is selected as 1.5 m for
this setup, where the optical attenuation is less than 0.3 dB (the attenuation factor is
200 dB/km).
4.2.1.1 LiFi Link
This section describes the distributed optical wireless link. The optical frontend
has high optical transmit power using off-the-shelf high-power LEDs operating at
860 nm (OSRAM-SFH 4715 AS) and high sensitivity using large-area silicon PDs (PD,
Hamamatsu S6968). The LED driver performs impedance matching for multiple LEDs
operated in parallel producing around 2.5 Watts of average optical power emitted into
a beam width of 90 ◦full-width-at-half-maximum (FWHM).
The LED driver uses a fixed bias current while the time-varying modulation current
for the data signal is optimized to maximize the data rate at the intended working
distance.
1
The Rx consists of one PD combined with a bootstrapped TIA to capture
the light signal. The LiFi link can operate over several meters of distance and offer
1This means that the gain is increased and some clipping is tolerated.
Figure 4.3: Single input single output (SISO) link block diagram of SDM-over-POF
setup.
42
4.2 Distributed 2×2MIMO
mobility inside the beam while providing an overall 3 dB modulation bandwidth of
around 80 MHz.
4.2.1.2 Measurement Scenarios
In this subsection, the measurement scenarios used for the first D-MIMO experiments
are presented using SDM-over-POF. The experimental setup is illustrated in Fig. 4.4.
To combine the wired and wireless link in the SDM-over-POF setup, fixed gain
amplifiers are used (see Fig 4.1 to equalize the gain of the POF link to unity. In our
measurement, the distance between access points (APs) to users, user to user, and
AP to AP are defined as
d1
,
d2
and
d3
respectively as shown in the Fig. 4.4. This
measurement scenario is an example of the downlink of high bandwidth multi-user
MIMO transmission currently defined for LiFi in IEEE Std 802.15.13-2023. As shown
in Fig. 4.1 (a), each distributed link is operated in a bidirectional manner. But due to
limited space, here, only the downlink is reported.
Two wireless setups are compared. In the first, the APs and users are located at
d1
= 100 cm,
d2
= 70 cm, and
d3
= 70 cm. In a second setup, the users are placed as
close as possible to each other
d2
= 5 cm,
d1
= 50 cm, and
d3
= 35 cm. The detectors
did not collimate and tilt to improve angular selectivity, although this could further
enhance the throughput.
4.2.2 Results
This section reports the results of MIMO measurements. The feeder link is operated in
a transparent mode for SDM, i.e. losses over the POF, in electrical to optical conversion
(e/o) and optical to electrical (o/e) modules are jointly corrected by an appropriate
amplifier in the electrical domain, see [36].
Moreover, this section characterizes the downlink channel properties for the 2
×
2
distributed MIMO system. Fig. 4.5 demonstrates the singular values of D-MIMO over
the SDM channel for two scenarios. In scenario 1, results indicate that two parallel
Figure 4.4: Experimental setup for D-MIMO over POF with SDM approach.
43
4. MIMO Communication
Table 4.1: Throughput evaluation of D-MIMO set up in two different scenarios.
D-MIMO SDM
Scenario 1 586 Mbps
Scenario 2 421 Mbps
Figure 4.5: Down-link normalized singular values for SDM (a),
ξ1S1
and
ξ2S1
for Scenario
1: spatially separated RX
d1
= 100 cm,
d2
= 70 cm,
d3
= 70 cm, and (b) Scenario 2:
ξ1S2
and ξ2S2of co-located RX: d1= 50 cm, d2= 5 cm, d3= 35 cm.
.
data streams are possible and full spatial multiplexing can be used. In scenario 2, the
mobile devices are too near to each other so there is only one strong singular value,
establishing a single spatial degree of freedom, which means in practice, that only one
data stream can be transmitted at the same time.
From the singular values of H(
f
)and, by assuming the average SNR equal to 20 dB,
the achievable data rate has been calculated, according to equation 4.23. Performance
is evaluated for the two scenarios. Throughput results are shown in Table 4.1. As
the POF channels are spatially separated by independent fiber links in the SDM case,
there is no spectral overlap between the channels, which increases the data rate. But
more cables are needed on the ceiling. Most of the throughput comes from only one
data stream in both scenarios 1 and 2. A potential way to overcome this issue is by
using bit-and-power-loading techniques in the spatial domain to designate more power
to some channels. As can be seen from Table 4.1, the SDM link with distant users
(Setup 1) provides a higher capacity compared to the link with nearby users (setup 2).
Therefore, the overall link is mostly limited by the OWC link, not by the POF.
4.2.3 Summary
In this section, a distributed MIMO link is presented for optical wireless transmission
combined with POF as a fronthaul using the SDM technique. Multiple POFs are used
to distribute the signal to the optical frontends. A signal model is proposed to evaluate
the performance of the LiFi link, including characteristics of the fixed and wireless
links. A first experimental setup for D-MIMO is presented. The performance of the
44
4.3 All-Optical Distributed MIMO: Spatial Diversity vs Spatial Multiplexing
link is evaluated for two scenarios in which access points and users are located apart or
close to each other.
The results show that the performance is mostly limited by the wireless link and
it depends on the location of APs and mobile users. By placing APs and users at a
reasonable distance, the throughput is increased. For closely spaced users, diversity
is better used than spatial multiplexing. Future work will focus on combining these
fronthaul techniques into a prototype system and deploying it in a real scenario.
4.3
All-Optical Distributed MIMO: Spatial Diversity vs
Spatial Multiplexing
In the remainder of this section, an adaptive optical D-MIMO architecture for
LiFi is proposed and the basic transmission modes for spatial diversity and spatial
multiplexing are investigated. Therefore, the simulation and measurement environments
for evaluating the SDIV and SMUX are explained. At last, the results retrieved by
simulation, and performed measurements are presented.
4.3.1 Simulation Environment
The simulations have been carried out in Matlab. The physical layer of the ITU-T
recommendation G999.1 for LiFi was implemented. As discussed in the mathematical
model, the properties of the wireless channel depend on the angular-dependent beam
profiles of the transmitter and receiver, and their number. In the simulation, only
LOS propagation is considered for the wireless channel. Reflected light in non-LOS
propagation is usually weak, because most reflections are diffuse, unlike radio waves,
where most reflections are specular. This is included in the channel model that has
been developed and confirmed by measurements in [88].
To assess the properties of the communication link, formulas have been derived for
the SNR-versus-frequency, including various combining techniques, the SVD as well as
throughput. A 4
×
2MIMO cell is evaluated in spatial diversity (SDIV) mode, and a
2
×
4MIMO LiFi cell in SMUX mode, for selected user locations. The simulation
framework is shown in Fig. 4.6.
4.3.2 Experimental Setup
In the experimental setups for both SDIV and SMUX, POF was used to transmit the
signals between CU and distributed optical front ends (D-OFE) as shown in Figs. 4.7
and 4.8. In both setups, the front-haul and the wireless units were used as follows: The
Avago transceiver AFBR-59F3Z was used for the bidirectional fronthaul transport using
POF. On the transmission side, a 650 nm LED was used together with an integrated
driver to convert the electrical into an optical signal. An integrated fiber-optical
receiver with a PIN photo-diode and a trans-impedance amplifier (TIA) was used. In
the wireless link, the OFEs were equipped with off-the-shelf high-power LEDs operating
at 860nm (OSRAM-SFH 4715 AS) and large-area silicon PDs (Hamamatsu S6968) as
45
4. MIMO Communication
Figure 4.6: Simulation Environment for (a) spatial diversity (SDIV), which consists of
four OFEs connected in series pattern and two MUs, (b) Spatial multiplexing (SMUX)
setup with two OFEs and two MUs. Each MUs is equipped with two OFEs, which are
tilted with angle α.
Figure 4.7: SDIV measurement setup: In this setup, the CU distributes the signal equally
and sends it via POF to D-OFEs. On the user side, there are two users with varying
locations.
described in [46]. In these experiments, the same OFEs were employed both in the
wireless infrastructure and at the mobile units.
The SDIV and SMUX modes were realized by using the real-time digital signal
processing in a chipset supporting data transmission over several home networking
media, i.e. coaxial cable, phone line, power line, and POF, according to the ITU-T
G.9960 home networking standard.
4.3.2.1 Spatial Diversity
The SDIV setup shown in Fig. 4.7 represents the 4
×
2 MIMO link. At the CU, the
Wave-2 DCP962P modem from Maxlinear was used in the coaxial cable mode. This
modem is based on two chips, the 88LX5153 yields the digital baseband signals and
the 88LX2730 provides the analog frontends output. The CU is operated as follows. In
the downlink, the output CU signal is replicated
NT
times (number of OFEs) through
Fanout buffers and sent via the POF over the wireless link from each OFE. In the
uplink, it is possible to use EGC or SC. Both are implemented in the analog domain
and hence, act identically on all sub-carriers of the DC-OFDM signal. In SDIV mode,
46
4.3 All-Optical Distributed MIMO: Spatial Diversity vs Spatial Multiplexing
Figure 4.8: SMUX setup: for evaluating SMUX feature in a LiFi link. Each user is
equipped with two OFEs [81].
each MU has one OFE. All MUs are equipped with the same OFE also used in the
D-OFE.
4.3.2.2 Spatial Multiplexing
The experimental setup for the SMUX link is shown in Fig. 4.8. The LiFi cell is
defined as a 2
×
4 MIMO link, and the simulation emulated the same scenario as in the
experiment. The CU was connected to two D-OFEs via POF. Each mobile device was
equipped with two OFEs, which were either pointed in the same or different directions
indicated by the angle
α
. The CU was Maxlinear DCP962, operating in the phone-line
mode, as defined in ITU-T recommendation G.9963 standard. The CU operated as a
MIMO link with two channels connected via POF to each D-OFE. At the MU side,
the OFEs were directly connected to the two channels of the same modem. This
setup was also considered for evaluating the multi-user multiple-input multiple-output
(MU-MIMO) performance. It was observed that after resetting the modem, the link was
operated in the multi-stream (MS) mode if the channel rank was high, and in SS mode
if the channel rank was low. Thereby, the sum data rates in SS and MS modes have
been directly measured or estimated for single user (SU) and MU-MIMO, based on the
frequency-selective SNR measurements performed at both MU modems independently,
following eq. (4.23).
4.3.2.3 MU-MIMO
For one user, here a 2×4 MIMO link is considered, i.e. the wireless infrastructure and
each user has two OFEs. In the SS mode, 2x4 MIMO can offer one data stream that
has additional diversity as it is supported by all 2x4=8 links jointly. Alternatively, in
the MS mode, 2x4 MIMO can offer two data streams in parallel. But then the diversity
order is reduced to 4 links.
MIMO equalizers in the terminals aim to suppress the mutual interference between
the parallel streams. The following techniques are investigated: 1) MIMO mode
47
4. MIMO Communication
switching: If the SNR is poor, or the channel matrix has a low rank, the MIMO
system can switch to the SS mode while at higher SNR and over well-conditioned
MIMO links, it can transmit two data streams. 2) AD: The rank of the MIMO channel
can be improved by spatial separation of the frontends, which is done in the wireless
infrastructure. The same can be reached by AD also at the terminal side. To keep
capacity high, i.e. maintain MS transmission also at the cell edge, each user device is
equipped with two OFEs pointing into different directions, what is denoted as AD. 3)
Multiuser multiplexing (MU-MUX): This mode enables simultaneous communication
to multiple users, which can improve the overall capacity of the communication link.
Furthermore, it can happen that two MIMO users are active in the same service
area. In that case, one stream can be used to serve the first and the other stream
for the second user, if their MIMO channels complement each other. The benefit of
MU-MUX becomes evident as follows: In a link where each mobile device has only one
OFE, the MS mode is applicable only if both users are far from each other. At the
cell edge, if both users are co-located, the channel has ranked one forcing the MIMO
system to operate in SS mode. Still, the devices can share the channel in the time or
frequency domains, but the low rank reduces the capacity.
If users have two OFEs, each device can receive its first and second data stream
from different APs in the infrastructure. In MU-MUX mode, hence, each device receives
its first data stream from its nearest AP and suppresses the alien interference from the
other AP point through its own MIMO equalizer. Both devices can follow the same
approach and thus, they could receive data in parallel also at the cell edge.
Table 4.2: LiFi Cell Layouts, Scenarios, and Cases
SDIV
Layout Scenarios
4×2 MIMO
I: D1= 90cm, D2= 200cm, D3=5cm
II: D1= 90cm, D2= 200cm, D3= 90cm
III: D1= 90cm, D2= 200cm, D3= 270cm
SMUX
Layout Scenarios Cases
2×4 MIMO
I: D1= 70cm, D2=100cm, D3= 5cm without angular diversity
II: D1= 70cm, D2=100cm, D3= 70cm with angular diversity
III: D1= 70cm, D2= 100cm, D3= 140cm
4.3.3 Simulation and Experimental Results
The following section presents the simulation and measurement results of both SDIV
and SMUX modes, corresponding to the setups and scenarios introduced in Table. 4.2.
In both setups,
D1
is the distance between D-OFEs (same for all D-OFEs involved),
D2
the height of D-OFEs above MUs, and
D3
the distance of separation between two
MUs. In each scenario, only
D3
was varied, and the other distances were kept constant.
48
4.3 All-Optical Distributed MIMO: Spatial Diversity vs Spatial Multiplexing
Figure 4.9: Simulated SNR vs. Frequency for up-link, including EGC as well as SC
methods for user 1 (a) and user 2 (b), for the SDIV scenarios I, II, and III.
Figure 4.10: Measured SNR .vs Frequency for uplink, including EGC and SC methods
for user 1 (a) and (b) user 2, for the SDIV scenarios I, II, and III.
4.3.3.1 Spatial Diversity
In this subsection, the spatial diversity results are presented covering both SNR and
throughput.
4.3.3.1.1 Signal-to-Noise -Ratio
At first, based on the model presented in section III-B, the SNR was simulated for
EGC and SC for each scenario of each user, as shown in Fig. 4.9 (a,b). As mentioned
before, the diversity combining was performed only in the uplink, while equal splitting
was used in the downlink. Note that the SNR for both users were identical in the
simulation, due to the symmetry in the scenarios and channel, accordingly.
The SNR of SC always outperforms EGC in all scenarios, e.g., in scenario I
(
D3
= 5
cm
), where users were located close to each other. By using EGC, the received
signals are summed with equal scaling, which leads to lower SNR for each user in
comparison to SC method, where the received signals above a defined threshold are
selected and summed. SC avoids adding the noise from the weak channel, which
49
4. MIMO Communication
Figure 4.11: (a) Estimated throughput for different users locations considering both
EGC and SC. (b) Measured throughput for different users locations considering both EGC
and SC.
leads to a higher overall SNR. The same results for scenario II (
D3
= 90
cm
) and III
(D3= 270cm) were observed by using SC method.
The simulation results were principally confirmed by our experiments in the same
scenarios, as shown in Fig. 4.10 (a,b). In the measurement, the SNR frequency spectrum
was obtained with the monitoring software provided by the digital signal processor
manufacturer [89].
SC leads in the uplink to higher SNR for both users, especially in scenarios II and
III, which are shown by the green and blue dashed lines, respectively. It is also effective
in scenario I (orange line), but the gain is only about 3-5 dB.
There are differences in the SNR values between the simulations and measurements.
These differences are attributed to the empirical gap factor introduced in (21). It mainly
takes the reduced modulation amplitude in real frontends into account to avoid clipping
and non-linear distortion. In addition, there is always a penalty in real implementations,
due to non-ideal coding and transmit signal shaping, which reduces the performance.
Moreover, potential NLOS contributions were not included in the simulation analysis.
Altogether, it was observed that SC outperforms EGC significantly in terms of SNR.
4.3.3.1.2 Throughput
The estimated and measured throughput results for SDIV are shown in Fig. 4.11.
First of all, note that there is a difference between uplink and downlink performance,
particularly when using EGC. This is due to the accumulation of noise when summing
up four received signals. In addition, there are scenario-dependent differences, due to
the different geometrical path gains depending on the user positions. In Fig. 4.11 (a),
the throughput is obtained by simulation from the SNR using
(4.23)
. The observation
is that the simulated throughput for each user is always higher when using SC compared
to EGC. This is also confirmed by the measured throughput of each user obtained from
the chipset software, as shown in Fig. 4.11 (b). SC leads to higher throughput for both
users specifically in scenarios II and scenario III.
50
4.3 All-Optical Distributed MIMO: Spatial Diversity vs Spatial Multiplexing
Figure 4.12: Simulation results for singular values in the SMUX cell layout described in
Section V-C for
D3
=5cm for two users without and with angular diversity for downlink (a)
and uplink (b).
Figure 4.13: Simulation results for singular values in the SMUX cell layout described in
Section V-C for
D3
=140cm for two users without and with angular diversity for downlink
(a) and uplink (b).
4.3.3.2 Spatial Multiplexing
Next, SMUX was considered by first assessing the singular values of the channel matrix
in different scenarios and then by evaluating the throughput.
4.3.3.2.1 Singular Values
Singular values are an indicator for the potential number of data streams that can be
used in parallel for SMUX [90]. Besides, the singular values also show the potential
amplitude gain in these parallel links. In the following, the SVD of the MIMO
channel matrix is evaluated through simulations for different SMUX scenarios following
Table. 4.2.
The SMUX cell layout is shown in Fig. 4.6 (b) and Fig. 4.8, without and with AD.
The singular values from simulations are shown for different spacing between the users,
51
4. MIMO Communication
i.e.
D3
= 5 cm in Fig. 4.12 and
D3
= 140 cm in Fig.4.13. The SVD of user 1 and
user 2 are denoted as
λ1
and
λ2
. In the case of without AD, it is the continuous line,
and with AD it is the dashed line. If both users are nearby (scenario I) and have no
AD, there is only one dominant singular value, i.e. only one of the users (see Fig. 9,
λ1
interprets as user1) can be served at the same time slot, both in downlink and
uplink. If both users are nearby and AD is enabled, there are two dominant singular
values. These singular values have about the same magnitude in the downlink but are
somewhat different in the uplink. This is attributed to non-reciprocity between uplink
and downlink due to the different beam patterns of transmitter and receiver, which
are typical for LiFi. When the distance between the users is larger (Scenario III), see
Fig.4.13, also scenarios without AD can result in two dominant singular values.
Scenario II is similar to scenario III both with and without AD. Therefore, scenario
III in Fig. 10 is only shown. In the downlink (Fig. 10 (a)), there are two relatively equal
singular values, when there are no AD (purple and yellow continuous line), indicating
that the maximum capacity of the channel can be almost realized since the users are
located far from each other and user have both OFEs pointing upwards. Introducing
AD when users are far from each other, leads to unequal singular values, hence lower
capacity. Considering the downlink direction in Fig. 5, OFE 1 of user 1 is not able
to receive a strong signal, since it is looking away. But OFE 2 of user 1 can capture
the strong signal from OFE 2. Therefore,
λ1
, (purple dashed line) has a higher value
than
λ2
(yellow dashed line). However, in the uplink (Fig. 10 (b)), both singular values
without and with AD have strong values. Altogether, AD leads to a more consistent
behavior, specifically when users are close to each other, and uplink direction. Therefore,
AD provides less dependency on the user positions in the cell. Despite being near to
each other (scenario I), multiple data streams can be transmitted, as a benefit from
AD.
4.3.3.2.2 Throughput
Throughput results for SMUX scenarios without and with AD and by using MU-MIMO
were measured and evaluated, respectively. The MS transmission is considered for
every single user (SU). Results are shown in Fig. 4.14. The observation is that the
throughput is higher for all scenarios when using AD. In scenario I, the gain due to AD
is reduced, compared to other scenarios, because the link switched automatically to
SS mode due to the short distance of the users to each other. The MS results without
AD are poor in general. When the channel matrix has reduced rank, it is harder to
support MS transmission mode.
At last, the value of multi-user multiplexing is evaluated. This is a rather
sophisticated mode, in which one stream is assigned to one user and the other stream
to the other user. In MU-MIMO mode, each device receives its own data from the
nearest OFE and suppresses the interference from the other stream which is assigned
to the other device. Note that both devices employ their MIMO equalizers, therefore,
i.e. the other data stream is separated from its own stream, but it is not decoded and
discarded at the interface to the medium access layer. In this way, the highest fractional
52
4.4 Conclusions
Figure 4.14: Measured throughput of spatial multiplexing in downlink and uplink for
three defined scenarios and without and with angular diversity.
.
rates can theoretically combine from both users to evaluate the value of MU-MIMO,
which is shown as green bars in Fig. 4.14. In the downlink, where the performance is
limited by clipping, MU multiplexing does not enhance the throughput in the scenarios
investigated here. But in the uplink, communication is more limited by the noise, and
then the combination of signals from two users becomes efficient. For instance, in
Scenarios I and III (uplink) the throughput increases by combining the best streams.
These results indicate that multi-user multiplexing has always the highest perfor-
mance, which is sometimes also reached by combining the streams from the same user.
Nonetheless, measurement results show that having the option of combining streams for
different users in one MIMO link can lead to the same or eventually higher throughput.
4.4 Conclusions
For the first time, an all-optical distributed wireless communication system was
presented for future IIoT applications and tested by using the ITU-T G.9991 standard.
For implementing the fronthaul, plastic optical fiber was used to distribute the signals
from the central unit to the distributed optical frontends. Next, these signals were
transmitted through the optical wireless link to each mobile user. To analyze the
proposed system, a theoretical signal model was developed, which allows for evaluating
the link performance including both, the wired link concatenated with the wireless link.
The system was considered as a distributed multiple-input multiple-output link that
can be operated in two transmission modes: Spatial diversity and spatial multiplexing.
The performance of these transmission modes was investigated by employing simulation
and measurements in different scenarios.
53
4. MIMO Communication
As observed, the performance of each transmission mode highly depends on the
channel condition, besides the received power. Both are related to the spacing between
the users and the distance to the OFEs. Moreover, for closely-spaced users, SDIV
enables better performance than SMUX without angular diversity. Selection combining
provides impressive diversity gains over equal gain combining, which is attributed to
the spatial selectivity of the optical wireless channel. The SMUX in combination with
angular diversity was considered, which improves the performance depending on the
user location. In low SNR scenarios, specifically in the uplink, multi-user multiplexing
can achieve additional gains.
As an outlook on distributed MIMO communication, future work needs to consider
dynamic switching between these spatial transmission modes, i.e. single- and multi-
stream transmission for single and multiple users, to maximize the performance in each
scenario. Channel estimation for distributed MIMO, the required feedback, and the
control loops are defined in the IEEE Std 802.15.13-2023 for industrial LiFi. These new
features will enable setups with more distributed optical frontends and larger statistics
when users are mobile.
54
5
Positioning
5.1 Positioning for Industry 4.0
Positioning or localization in industrial environments enables modern industrial
applications to support the evolution of smart factories by determining the position of
mobile devices including intelligent transport systems (ITS), smartphones, tablets as
well as (semi-) finished products, and other production resources that are needed in the
factory. In this section, the role of positioning is highlighted for the next generation of
smart manufacturing.
5.1.1 Positioning of Production Resources
Real-time tracking of production resources, tools, and (semi-) finished products is a
feature that can assist production planners, who schedule production machines along
with resources, by displaying the position of resources needed on a factory map and
updating lists with available resources automatically in real-time. These features
enable production planners to schedule in real-time and process engineers to evaluate
production steps and enhance the efficiency of the production process.
5.1.2 Positioning of Transport Systems
Fig. 5.1 shows a transport system in Weidmüller’s factory. It transports parts on
pallets following predefined routes through the mounted-cameras downside of the ITS,
which is also referred to as automated guided vehicle (AGV), and used in factories
for transporting parts. The transport system prohibits crashes which can happen
due to obstacles, however, it is not able to determine a new path on demand since
deviation from pre-defined paths is not possible. The desired manner is to provide
flexible movements which prevent blockages.
By equipping ITSs with positioning functionalities, they are no longer limited to
the fixed route defined with the optical markers (shown in Fig. 5.1), since they receive
positioning information everywhere in the manufacturing hall. As shown in Fig. 5.2,
55
5. Positioning
Figure 5.1: LiFi positioning integrated with wireless communications for intelligent
transport systems (ITS) in a smart factory [91].
the transport system can navigate through the machines along the dashed lines. These
dashed lines represent virtual routes that are defined by position coordinates used for
navigation. This means that the transport system is no longer limited to driving only
to the central collecting points with many stockyards that are placed along the fixed
path.
The transport system can drive directly to a stockyard of a machine. In this scenario,
the path, the source, and the target positions can be defined and changed in a “transport
system navigation software” when machines are transposed. This makes production
more flexible and scalable, which helps to increase productivity. Potential optimization
in productivity is identified through value stream management and reorganization.
5.2 Wireless Positioning Techniques
Numerous wireless positioning approaches are reported to address diverse applications
and environments. There are either passive or active techniques, and the used
transmission medium and technologies can be different, leading to different accuracy
[92]. In the following subsections, an overview of the most common wireless positioning
techniques is given.
56
5.2 Wireless Positioning Techniques
Stockyard
near the machines
Stockyard near the fixed transport path
Fixed path defined with optical marker
Flexible virtual transport path through the machines
Figure 5.2: Top view illustration of the factory floor, there is a fixed path for the transport
system, stockyards along the fixed path, and alternative paths through the machine’s own
stockyards [91].
5.2.1 RF-Based Positioning
The wide range of indoor applications are interested in estimating the position of mobile
devices with high accuracy. Several methods and technologies have been proposed
to meet indoor positioning requirements. In the following, time-based geometric
positioning algorithms based on RF technologies are described.
Time of flight (TOF) is defined as the time that the signal takes to travel from
a first point to a second point. By knowing the TOF, the distance between points
can be calculated by multiplication of the TOF with the propagation speed of the
signal [92]. Another method is TDOA) which has two variants. In the first method,
one mobile device (MD) acts as a receiver with an unknown position and multiple
APs act as transmitters with known positions. All APs are synchronized, and the MD
measures the differences in the ToFs of all transmitted signals. This approach is similar
to the GPS [92]. The second variant of TDOA uses one MD as a transmitter and
multiple synchronized APs as a receiver with known positions [93]. The accuracy of
both methods is dependent on the synchronization between the APs and the MD, as
well as the achievable precision of the TOA estimation which can be limited due to the
bandwidth and other factors. Differential time difference of arrival (DTDOA) has a
reference node (RN) in addition to TDOA. The RN enables proper synchronization
among the APs [92].
In the round-trip-time (RTT) or two-way ranging technique, each AP and the MD
act in a bidirectional manner. The AP sends an initial packet (ping) at the time of
transmission
tot
and the MD receives it at a local clock called the time of arrival
toa
.
After a defined processing time
tproc
at the MD, it sends out a second packet (pong) in
the reverse link direction [94]. The time of arrival
toa
of the second packet is recorded
57
5. Positioning
Figure 5.3: Scenario for radio-based positioning with moving intelligent transport
systems(ITS) between separate buildings indicating the problems of radio-based signaling
and handover due to interference from other rooms, reflections at huge machines, walls
and surrounding objects [96].
by the AP. The combination of all times allows the computation of the round-trip time
(RTT), then TOF, and finally distance between AP and MD.
These techniques are used in most time-based RF positioning systems, especially
for outdoor environments where the propagation is mostly based on the line-of-sight
(LOS) and few specular reflections e.g., from large buildings [95]. In indoor scenarios,
however, they become inaccurate due to the physical effects of attenuation and multi-
path propagation due to reflections. In industrial environments, ITS move inside and
between buildings. However, RF waves pass through walls, inside which the light has a
different speed. These phenomena cause inaccuracy in the delay estimation due to the
variety of materials in buildings.
Moreover, multipath propagation due to reflections leads to fading effects. Due to
the fading, non-LOS signals can be stronger than the LOS as shown in Fig. 5.3 [96].
Additionally, EMI around ITS stemming from other radio-based systems sharing the
same frequency band makes current WiFi-based positioning techniques insufficient for
smart manufacturing environments because the accuracy is around 1 meter at 5 GHz,
depending on the channel bandwidth and distance [96], [97]. Altogether, the complex
features of radio propagation in indoor environments lead to extra delays for time-based
algorithms, and thus to reduced accuracy.
5.2.2 Camera-based Positioning
Commercially available cameras have a limited frame rate, much lower than what would
be needed to see any light communication (LC) data if that LC data is modulated at
frequencies that do not cause visible flicker to humans. Nonetheless, two key properties
allow the use of cameras in a positioning system:
•
Cameras involve a line scanning mechanism as shown in Fig. 5.4 and 5.5. This
translates temporal light modulation into a spatially resolved fluctuation of the
58
5.2 Wireless Positioning Techniques
Figure 5.4: Line scanning (yellow) in a camera picture of a ceiling with light source [91].
Figure 5.5: Signal fragments being stitched together from multiple frames to recover the
lamp identifier
intensity of the light on the picture, as illustrated in Fig. 3. This can be used
to identify the lamp. The light modulation frequencies are in the range of 1 to
8 kHz. This avoids visible flicker and allows the use of low-resolution cameras
and legacy cameras that do not allow exposure and focus control optimized for
positioning.
•
Cameras, particularly high-resolution ones, can accurately estimate the angle
toward light sources. If the lens properties are known, and if the positions of the
light sources are known, one can use the pixel location of the light sources in the
image to calculate back from which position the picture was taken.
For instance, the company Signify has developed such a system, referred to as
Interact Indoor Navigation [98, 99]. It uses white (visible) light fixtures the output
of which is modulated to provide an unidirectional, low data rate communications
link between ceiling-mounted luminaries and a smartphone or tablet equipped with a
camera, accelerometers, and gyroscopes (standard features) running an iOS or Android
mobile operating systems. As modulation frequencies are low, standard illumination
LEDs in the lighting fixtures can directly be modulated by the LED DC driver itself,
59
5. Positioning
Figure 5.6: Autonomous lamps emit unique codes. These are picked up by the phone
camera. An Internet link (Wi-Fi 5G,...) to a Signify–Philips database allows the phone
app to translate received codes and camera-measured angles into positions. A customer
can use these data in a proprietary application [91].
to deliver a 16-bit identifier code, unique per laminar, to mobile smart devices running
a Signify-developed software application.
The unique fixture identification codes, as compiled from multiple camera frames as
shown in Fig. 5.6, are used together with the angle-of-arrival functionality provided by
the smart device camera. This enables not only position and location information but
also estimates of the direction in which the smartphone user/device is facing. As the
indoor positioning algorithm runs locally on the receiving mobile device, there is no
limit to the number of users of the system. The Interact Indoor Navigation system is
supported by luminaries from dozens of suppliers that are certified under the Yellow Dot
program and can provide data analysis of movement patterns for retail applications.
In the following, some performance benchmarks are given. If today’s smartphone
cameras are used in a calibrated lab setting where light sources are located at exactly
known locations, a positioning accuracy of a few cm is achievable. However, when
applied in an industrial setting such as a large warehouse or factory site, further error
mechanisms occur. For instance, typical luminaries such as trunk lighting are not
point sources but have a large emitting surface. Another restriction is the desired
compatibility with low–resolution VGA cameras. A typical commercial installation is
in a 100 m long by 40 m wide hall with a 7.7 m ceiling height. Yellow Dot-certified
trunk lighting was installed using a 1.5 m pitch for the location codes and a 3.8 m pitch
from one line of trunk lighting to another. In-field measurements were performed using
an Apple iPhone 7+ located at a height of 1.2 m above floor level.
In the first test, the indoor position of the iPhone was determined directly under
and a lighting track and in between two tracks. Relative to a reference laser range finder
60
5.2 Wireless Positioning Techniques
system, and typical VLC location accuracy errors were around 30 cm but never larger
than about 40 cm. The track lighting and luminaire pitch were typical of that employed
for general illumination purposes, such that no special measures are required in the
field. The system is designed to cope with personal smart devices moving at normal
walking speed (1 m/s). Besides requiring more accurate installation, setting minimum
specs for smartphones used, etc., improved accuracy can also be achieved by combining
the lighting position fixes with other modalities, such as radio, accelerometers, and
dead reckoning.
5.2.3 LiFi-based Positioning
Introducing LiFi as a complementary technology besides RF, it can overcome some of
the challenges mentioned above. The advantages of LiFi-based positioning for smart
manufacturing are primarily due to the better propagation characteristics of the light,
resulting in improved accuracy and resilience against interference. Unlike RF, light
propagates mostly through the line-of-sight (LOS) while multi-path plays a negligible
role also because most reflections are diffuse and not specular [100]. Accordingly, there
is no fading and the very high probability that a significant signal is due to a free LOS
between Tx and Rx. Moreover, light does not propagate through walls, thus there
is no change in the speed of light, and it cannot be interfered by EMI. In all these
points, where the use of RF for positioning is problematic, LiFi has clear advantages as
a potential solution. This is exemplified in the same scenario mentioned above, but
using LiFi, in Fig. 5.7.
Challenges for LiFi-based positioning systems are the required coverage and energy
efficiency in particular for mobile devices [96]. To enable localization in an industry
environment, full coverage is required. This is a challenge because of the limited area
that LiFi front-ends can cover. LiFi positioning will require significantly more front-ends
compared to an RF-based system. Furthermore, a dedicated backhaul/fronthaul at
the ceiling is required to connect each front end. This can be overcome by reusing the
lighting infrastructure. Energy efficiency can be an issue since the mobile units generally
work on battery. Because multiple frontends are needed to estimate the position in 3D,
the coverage area of optical frontends needs a significant overlap. In order to realize
this, wide transmitter beams are needed which causes lower received optical power.
This is an additional reason why a power-efficient design is needed for LiFi. The use of
power-efficient modulation, such as On-Off Keying with frequency-domain equalization
(OOK-FDE) [101] is promising. In the rest, some well-known positioning techniques
are described using LiFi and some related works.
5.2.3.1 Received Signal Strength Indicator
The RSSI-based technique is the most widely studied technique for indoor positioning
using LiFi in the literature. In this technique, light waves are transmitted in a clear
medium. At the MD, the received light is converted into an electric signal. The distance
between the transmitter and MD is calculated from the known signal power, received
signal power, and the attenuation model [102]. The RSSI technique depends on the
61
5. Positioning
Figure 5.7: Improved scenario using LiFi for integrating positioning with wireless
communications solving problems from radio-based scenario regarding interference and
multi-path propagation and simplifying handovers for moving, ITS [96].
received light intensity at the receiver and a pre-calculated light intensity at a given
calibration distance.
In academic work about the RSSI technique, LiFi transmitters are often assumed
to have Lambertian beam patterns. This allows an injective mapping of RSSI to
distance. Industry is rather skeptical about this because luminaries are designed
to provide homogeneous illumination, which favors patterns that allows less RSSI
differentiation by the light level. However, it is not injective in the mathematical sense,
i.e. two distances may result in the same RSSI [103]. When used for RSSI positioning,
Lambertian beams may give optimal results. Moreover, RSSI detectors may miss
information from weak light sources at higher angles, as side emission patterns may be
cut off to avoid glare. The receive pattern is often considered as a plane photodiode,
however, it is sometimes combined with an optical concentrator leading to a more
directional pattern [104].
In a real device, there are shading effects at the edges of the FOV. Besides the
distance, obviously, the RSSI depends on the tilt of the transmitter and receiver with
respect to the LOS. Moreover, the RSSI technique is vulnerable to shadowing [105],
or to dust and aging of the light source. Establishing a clear relation between the
distance and received signal strength is difficult to implement outside a laboratory,
where conditions cannot be well-defined. In [106], RSSI positioning with differential
detection is reported. The method reduces positioning instability caused by light
intensity fluctuations by using a second detector with a known position.
While 1D accuracy is up to 4 cm, the accuracy for 2 and 3 dimensions (2D-3D)
was studied in [107], leading to 10 cm and 22 cm error for 2D and 3D, respectively.
The main disadvantage of the RSSI positioning for LiFi is that a precise model of
all components in the system is required. While this might be applicable to a single
62
5.2 Wireless Positioning Techniques
vendor, it can hardly be considered for a mass market where many vendors implement
different devices that adhere to the same standard.
5.2.3.2 Finger Printing/Scene Analysis
Fingerprinting can also be called scenario analysis as it usually uses features or
fingerprints of the environment or object to identify its location. More specifically,
relative location is determined by comparing real-time monitored feature data to
various features of the environment that have been collected and stored. In the indoor
environment, fingerprinting usually estimates the positioning by collecting the signal
power based on the received signal strength indicator (RSSI).
There are two distinct phases for the scenario analysis, i.e. offline and online. In the
offline phase, every RSSI (location-related data) in the indoor environment is detected
and collected. The online phase refers to the real-time monitoring of each localization
element, i.e. RSSI, in the measured environment and the estimation of its approximated
relative position in relation to the data collected in the previous offline phase. For the
feature comparison method, probabilistic methods like k-nearest-neighbor or artificial
neural networks are usually used [108, 109]. The major disadvantage of fingerprinting
is the laborious collection of RSSI values during the scene analysis [110].
5.2.3.3 Angle of Arrival
The AOA technique determines the position based on the estimated angle-of-arrival of
the LED’s light at the MD. The advantages are that angle is constant over time [111], no
synchronization is required between LEDs [112], and only two and three measurements
are required for 2D and 3D positioning, respectively. The major challenge is designing
the receiver, which needs to provide a wide field of view (FOV) and detect signals
from multiple transmitters [111]. In [113] and [114] the receiver is implemented using
pyramids and cubes, respectively.
Another solution is a receiver with angular diversity aperture [115], [111]. As
discussed in the 5.1.3.2 section, Signify developed and rolled out a versatile camera-
based interact system that in fact estimates AoA from illuminated pixels in a camera
image. The theoretical accuracy depends, among other things, on the resolution of the
image, the angle of view of the lens, the emitter size, the quality of the lens, and the
relative positions of lamps and smartphones.
5.2.3.4 Phase Difference of Arrival
The phase difference of arrival (PDOA) approach estimates the distance between LEDs
and MDs based on phase differences between the signals. Each signal arriving at the
photodetector of MD, experiences a different time delay depending on the distance
value. Therefore, different delay times can be estimated from phase differences of
sinusoids by which the optical signal is modulated [116].
For PDOA, one obvious advantage is that it can be combined with other positioning
techniques such as RSSI and ToF/TDOA where it can significantly improve the
63
5. Positioning
Figure 5.8: LiFi - based TDOA localization.
accuracy [117], [118, 119]. In [120] a differential PDOA method with sub-decimeter
accuracy is demonstrated. However, the PDOA technique is limited to cases in which
the transmitter separation remains small. For large areas, the actual phase difference
of the signals cannot be computed uniquely, as phase is restricted to the range [0
,
2
π
),
and longer delays would create ambiguities (phase wrapping) [117].
5.2.3.5 Time of Flight (TOF)
The TOF technique for LiFi, which is outlined thoroughly for the first time in [91], is
very similar to RF-based TOF systems. Considering that the speed of light in the air
is constant, the distance from the mobile target to reference points will be proportional
to the travel time of the light. TOF is also named one-way-ranging. The AP sends
a packet and records the time of transmission (TOT). At MD, the packet arrives at
the time of arrival (TOA). By assuming AP and MD are synchronized, TOF can be
calculated as:
TOF =TOA −TOT, (5.1)
The distance between AP and MD can be estimated as:
r=c×TOF, (5.2)
where
c
is equal to the speed of light. By means of trilateration, the 2D dimension of
MD position is given by considering MD located at (
x0, y0
)and APs at (
xi, yi
)and
i= 1,2,3as follows:
(x0−xi)2+ (y0−yi)2=r2
i(5.3)
where
ri
is the distance between each AP and MD. In [61] a TOF-based localization
is presented for smartphones. The AP transmits both, sound and light waves. The
microphone is used to detect the acoustic signal and the camera is used for time
synchronization. The technique yielded an accuracy of 10 to 20
cm
. However, acoustic
waves are audible and the flicker of the light was noticeable.
64
5.3 Approach and Implementation
5.2.3.6 Round trip Time of Flight (RTTOF)
For LiFi, it is proposed to use RTTOF which is ranging based on TOF in a bidirectional
system. Similar to RF, each station is equipped with a transceiver (LED+Photo-
diode (PD)). The first station sends a packet to the second station, then the second
station sends a packet back towards the first station. Also, the rest of the process is
similar to RF-based ToF [121].
5.2.3.7 Time Difference of Arrival (TDOA)
By transmitting multiple packets from multiple APs to a MD at the same time, the
MD position can be estimated by exploiting the different signal delays. This requires
synchronization between all stations. The difference in the distance between each
station and the MDs can be calculated by multiplying the TDOA by the speed of light.
Based on the constant difference in distance, a hyperbola of possible positions can be
determined.
TDOA measurements from at least three stations must be obtained in a 2-D
environment. Hence, two distance differences
R2
–
R1
and
R3
-
R1
determine two
hyperbolas, and their intersection yields the position of the receiver. Suppose that
the coordinates of the receiver are (
x, y
)and let the coordination of any two LEDs be
(xi, yi)and (xj, yj). The hyperbolas Ri,j can be calculated following [112].
Ri,j =√︂(xi−x)2(yi−y)2−√︂(xj−x)2(yj−y)2(5.4)
where the intersections of hyperbolas correspond to the receiver position. Similar to
RF, synchronization of all transmitting and receiving front-ends is required, which is a
challenge in practice. A simulation study of the TDOA based positioning in [122] used
the TDOA of received pilot signals. Results indicate an average accuracy of 3
.
6
cm
in a
5
×
5
×
3
m3
room. Simulations in [118] using TDOA yield an accuracy of 1
cm
. A low
complex differential TDOA is presented in [119] within average positioning accuracy of
9.2cm.
There are other techniques using TDOA in hybrid ultrasound-light devices like
cricket sensors for distance measurement [123, 124]. In [125] a double-way distance
measurement using TDOA between optical and ultrasound signals is proposed. Unlike
conventional cricket-based systems, the method is able to measure the distance both at
the mobile node and base station.
5.3 Approach and Implementation
In this work, an advanced positioning technique is proposed for Industry 4.0 scenarios.
The wireless propagation times between a MD and multiple ceiling-mounted OFEs
are measured. Signals follow the G.9991 PHY frame structure. The idea of TOF is
extended by measuring the phase versus frequency to improve the precision well below
one sample interval. For measuring the 2D and 3D positions, finally, a multi-lateration
algorithm is used. This approach is implemented in Matlab and the performance
65
5. Positioning
Figure 5.9: System architecture. (a) block diagram of OFDM transmitter and receiver
for positioning integrated with wireless communications. (b) PHY frame structure, (c)
Time flow chart for RTTOF measurements.
validated through measurements. In the following, the technique and the evaluation
framework are explained in detail.
Each subsection introduces the key building blocks and aspects covered by the
simulation chain, followed by the measurement setup.
5.3.1 Scenario
Full LiFi coverage and access to at least 3 transmitter units at each location is the
mathematical requirement to extract the precise position in 2D at a random location.
This implies that there is a need for sufficient overlap between the signals from all
required LiFi transmitters in the intended coverage area[126]. This condition must
be met when deploying the OFEs of the LiFi system at the ceiling for the entire area
in which the mobile units can move. It is assumed that each OFE contains a single
transmitter (Tx) and a single receiver (Rx). This means that depending on the FOV of
each unit, a relatively dense grid of OFEs may be needed.
The ITU-T recommendation G.9991 uses DC-OFDM for LiFi with up to 200 MHz
bandwidth. Therefore, our simulation chain utilizes different functions from a DC-
OFDM transmitter and receiver. The wireless channel depends on the geometrical
properties of the intended coverage area, the room geometry, as well as the numbers
and placement of OFEs and MD which are all configurable to investigate different
variants of the scenario. Fig. 5.9 (a) shows an example of the scenario for 4 Txs (LED
1 – LED 4) and 1 Rx with an intended coverage area of 6
m×
5
m×
10
m
. From the
geometrical properties, the LOS channels from each Tx/Rx pair are calculated by using
Lambertian beam characteristics for each OFE. In our simulated channel, only LOS
propagation is considered.
66
5.3 Approach and Implementation
5.3.2 Digital Signal Processing
The digital signal processing (DSP) is based on OFDM transmissions from multiple
Txs units to a single Rx unit. Using adaptive OFDM for optical communication is well
described in [127]. The block diagram of the simulation for a single transmission link
is shown in Fig. 5.9 (a). The transmitter acts as an OFDM waveform generator to
send information in the form of bits. When using LiFi for wireless communications,
these bits are fed (from right to left in Fig. 5.9 (a)) into the OFDM transmitter DSP,
starting with a forward error correction (FEC).
After serial/parallel conversion (S/P), a training sequence (TS), which is defined in
the frequency domain, is added for the purpose of channel estimation. After taking the
inverse discrete Fourier transform (IDFT) and adding the cyclic prefix (CP), the signal
is sent out by the LED. In our simulation chain, DCO-OFDM modulation is used. In
order to get a real-valued signal, following the G.9991 standard, the OFDM signal is
upconverted to a low intermediate frequency (frequency up-shift), and a DC is added
to modulate the LED around a certain bias point.
The packet structure based on G.9991 is shown in 5.9 (b). The figure is taken from
the IEEE Std802.15.13-2023 in which the same PHY is also used. The PHY frame
contains an initial preamble for automatic gain control and synchronization, followed by
the TS used for channel estimation which is used to detect the PHY header. The header
may be followed by additional channel estimation (ACE) symbols which are defined and
utilized here for multiple-input multiple-output (MIMO) channel estimation. These
MIMO pilots are introduced in IEEE Std 802.15.13-2023 and they were added in this
section in order to estimate the channel from multiple OFEs to the MD. Arbitrary
data could be added finally in the payload, and they can be used e.g. for bit error rate
(BER) or SNR measurements.
No LED and PD impairments have been considered in the simulation, but their
impact is included in the experimental results. The LOS channel is calculated from the
given Tx/Rx geometry which results in an attenuation and a delay.
On the receiver side, noise is considered by adding a sequence of independent and
identically distributed (i.i.d.) noise samples that simulate additive white Gaussian
noise (AWGN) with a predefined noise power. Frame synchronization and channel
estimation are first applied. Further standard functions like channel equalization,
demapping of QAM symbols, and FEC decoding are possible but they are not used for
positioning. The actual estimation of the Rx position is based on combining a coarse
and fine delay estimation to get the relative distance between each Tx and the Rx with
sub-sample precision. For multi-lateration, the known positions have taken all Tx units
into account.
5.3.3 MIMO Pilots
On the transmitter side, 4-QAM modulated pilot subcarriers are added for channel
estimation. To enable the simultaneous detection of multiple Tx channels, orthogonal
sub-carrier combs are used for each Tx unit. For example, for the case of 4 Tx units,
subcarriers 1, 5, 9, ... are used for Tx 1, subcarriers 2, 6, 10, .. . for Tx 2, and
67
5. Positioning
so on. These orthogonal sequences allow simultaneous channel sounding of all Txs
by filtering the subcarrier signals which correspond to one Tx at the receiver side.
The sparse channel information for each Tx can be interpolated at the Rx side to
estimate the complete channel frequency response [128]. An important prerequisite for
orthogonality is precise clock synchronization of all Txs. This is assumed to be ideal in
the simulation environment, while further investigations on implementation are needed
in a real system.
5.3.4 Channel Model
This section describes the channel model between multiple Txs and a single Rx unit.
The dominating losses in the optical channel for indoor scenarios originate from the
fact that the optical signal is distributed in the whole coverage area of one OFE but
only a small fraction therefrom is captured by the small area of the PD at the Rx.
Since the Tx and Rx may be tilted by angles
θt
and
θr
with respect to the LOS, the
channel impulse response computed as [129]:
hLOS(t) = δ(t−τ)1
d2R0(θt)Aeff (θr)(5.5)
where
τ
=
d/c
is the delay due to free-space propagation over a distance
d
at the speed
of light
c
,
Aeff
is the effective area of the photodiode including the light concentrator,
and
R0
is the radiant intensity. At the Rx, the received signal
y
(
t
)is composed of the
transmit signal affected by the linear channel response and additive white Gaussian
noise n(t)as shown below [130]:
y(t) = hLOS(t)x(t) + n(t).(5.6)
As already mentioned above, transmission over the LOS channel is considered only.
Together with a pre-defined noise signal, this yields a certain signal-to-noise ratio (SNR).
In the simulation, the LED, the driver electronics, and the detector are considered to
have flat frequency responses, both, in amplitude and phase. In practice, a similar
scenario can be realized by calibrating out a so-called apparatus function.
5.3.5 Frame Synchronization
In OFDM-based systems, synchronization plays a crucial role. The light-induced
current at the PD is used to detect the coarse timing information needed for the
TOF. The first step in the measurement of the TOF is to establish a reliable time
synchronization, i.e. estimate the beginning of the transmitted data packet. Using
OFDM, time synchronization is also used for removing the cyclic prefix (CP). It ensures
that inter-symbol interference (ISI) between consecutive OFDM symbols is avoided.
The Schmidl Cox (SC) algorithm is a popular method for an autocorrelation-based
synchronization [131]. It has reduced performance but can be implemented with
significantly lower complexity compared to the optimal cross-correlation approach. The
SC preamble consists of at least two consecutive parts in the time domain which are
68
5.3 Approach and Implementation
identical. At the receiver, a correlation term called
PSC
is calculated by sliding a
window with a length of
S/
2samples over the received signal. For real-valued signals,
the correlation factor is given by [131].
PSC(d)=ΣS/2
m=1xd+mxd+m+S/2(5.7)
with
x
denoting the actual sample value at the running time index
m
,
d
the delay
between the original and the delayed version of the received signal, and
S
the total
sequence length. In addition, an energy term RSC is calculated as
RSC(d)=ΣS/2
m=1|xd+mxd+m+S/2|2(5.8)
and used for normalization. The coarse timing metric called MSC is given by:
MSC(d) = |PSC(d)|2
(RSC(d))2(5.9)
The SC method yields a plateau when detecting the start of the frame, as a result of
correlated equal parts in the periodic preamble. In case only noise or random data are
received, the result is zero. The main drawback of the basic SC scheme is the plateau,
rather than a peak, introducing some uncertainty for the frame start estimation. To
reduce the uncertainty and improve the coarse timing estimation, Minn’s modified
preamble has been introduced in [132] where the samples of the training symbol
Strain
have the form of (5.10) [132]:
Strain = [AA −A−A](5.10)
with
A
denoting samples of length
L
=
N/
4. The training symbol has four parts with
equal lengths. The first two parts are identical and the last two are the negative of the
first two. Similar to SC, the time metric is calculated as shown in (5.9), however, the
definition of the correlation term Pdand energy Pdare defined as (5.11) and (5.12):
Pd= Σ1
k=0ΣL−1
m=0r∗(d+ 2LK +m)r(d+ 2LK +m+L)(5.11)
Rd= Σ1
k=0ΣL−1
m=0|r(d+ 2LK +m+L)|2(5.12)
with
r
denoting the received sample vector given by
r
(
k
) =
x
(
k−L
), where
Ng≤k≤
N−
1,
Ng
number of guard samples. If
Pd
and
Rd
are given, the time metric
M
can
be calculated and the start of each frame can be detected. Note that although Minn’s
preamble yields a peak, it has a rising and falling edge. Moreover, the peak can be
deformed due to the noise and eventual multi-path in the channel. Accordingly, there
is some variance in the estimation of the frame start. For this reason, a so-called fine
timing measurement is applied based on the MIMO channel estimation described below.
69
5. Positioning
Table 5.1: System Parameters
Bandwidth Cyclic Prefix OFDM Symbol Subcarrier
(MHz) (Count) (Count) (Count)
100 32 27512
ADC Speed DAC Speed Fram Synchronization Training Sequence
(MS/s) (MS/s) (Count) (Count)
500 625 128 100
5.3.6 MIMO Channel Estimation
Our simulation environment supports MIMO pilots which are orthogonal and can
be sent in parallel by all transmitting OFEs. After propagating through the MISO
channel, these sequences are superimposed at the receiving MD. Orthogonality is
enabled through a MIMO pilot design described in subsection A.3.
After filtering the individual complex-valued channel frequency responses (CFR)
for each link, the estimated channel state information contains information about the
received signal strength from each OFE as well as its individual delay. Based on the
phase information in the CFR, where phase increases linearly with the OFDM subcarrier
index with a common slope depending on the delay, the fine timing information on
the TOF from each OFE to the MD can be obtained. The system parameters are
summarized in Table 5.1.
5.3.7 TOF Estimation
The precise knowledge of the transmission channel or in more detail the time delay
between one ceiling unit and the mobile unit is required to calculate the relative distance
between both. The delay is calculated in a two-step approach, first with a coarse timing
estimation based on the frame start detection in the OFDM modem and second by the
fine timing estimation based on the phase versus frequency in the CFR for the link to
each OFE. As shown in Fig. 5.9 (b), ToF is extracted by considering the synchronization
header. From the preamble and the channel estimation, the coarse and fine timestamps
are retrieved sequentially which is described in the following.
A coarse timestamp can be obtained at the transmission and reception of frames.
For increased accuracy, the additional phase information of the signal is retrieved from
the physical layer, e.g., based on MIMO channel estimation. The reuse of mechanisms
originally designed for communications purposes reduces the complexity and allows
easier deployment of positioning due to the integration into a single system. Using this
combined approach, the following conditions must be met:
•
Full LiFi coverage in the target area with at least 3 visible Tx for the Rx unit(s)
at all times
•
Coarse timing information obtained by detecting the frame start of all received
packets
•Fine timing information obtained from the CFR for each link
70
5.3 Approach and Implementation
(a)
(b)
Figure 5.10: (a): Block diagram of point-to-point distance measurement, The electrical
synchronization cable keeps a constant time offset between Tx and Rx, and uses the same
clocks, to simplify the measurement. In reality, the positioning will be implemented as a
bidirectional RTToF protocol so that the cable is not needed.
(b): Expanded setup using four LiFi Txs and one Rx to perform 3D localization.
5.3.8 RTTOF
RTTOF is necessary when APs are not well synchronized. This means that RTTOF
allows deployment of positioning also without using the MIMO approach, i.e., by using
conventional APs which have the full PHY and MAC layer implemented. Note that
the MAC can be split into real-time (lower MAC, for sharing the medium access) and
non-real-time (upper MAC, e.g. for authentication and security handshake. An active
ping-pong protocol can then be leveraged for ranging and estimation of the distance
between the user and APs by subtracting the difference between the local clocks (see Fig.
5.9 (c)). The communication protocol must therefore support the exchange of ranging
frames and the transport of control information besides data. Here, the implementation
of the ping-pong protocol is described between the two stations required for RTTOF
71
5. Positioning
Figure 5.11: Simulation results for combining coarse and fine timing for final distance
measurement. The SNR condition is good at this point, which leads to estimating the
distance without error.
in our simulation environment. The time flow chart of our RTTOF implementation is
shown in Fig. 5.9 (c).
For simplicity, it assumes a reciprocal behavior for the channel in downlink and
uplink. The authors are aware of the fact that this is not perfectly given in real
systems which can use different LED and PD beam characteristics. However, since our
algorithms mainly focus on the delays and not on the amplitudes, such simplification
may be justified.
The RTTOF handshake starts with the first station sending a ping packet to
the receiver, at the time when the packet leaves the transmitter. Upon reception of
the initial packet, the second station will generate the pong packet during a known
processing time denoted as
Tb
. After the pong packet arrives back at the first station,
the timer will record the received time as
Ta
. In the last step, the
RTT
and the distance
are estimated based on the following (5.13) and (5.14) :
RTT =Ta−Tb= (T4−T1)−(T3−T2)(5.13)
Distance = (Tf×C) = RTT ×C/2(5.14)
5.4 Measurement Setups and Results
The measurement block diagram and the laboratory setup are shown in Fig. 5.10. For
the transmitter, the synchronization, and the channel estimation at the receiver, the
same Matlab implementation is used as described for the simulation. Basically, the
72
5.4 Measurement Setups and Results
Figure 5.12: Simulation results for MSEs of x-, y- and z- dimensions vs. SNR.
waveform is generated by Matlab, converted to analog, transmitted and received with
two OFEs, converted to digital and then fed into the Matlab receiver algorithm.
To evaluate the real distance of the wireless channel, all intrinsic systems delays
need to be compensated. These originate from the digital-to-analog converter (DAC)
and analog-to-digital converter (ADC), various electrical wires, and the OFEs.
Compensation is performed by using an electrical reference signal together with an
initial calibration performed at a known distance. Further, the clock of the DAC and
ADC are synchronized as shown in 5.10 (b). The OFDM frame consists of a frame
synchronization sequence, a training sequence for the channel estimation and payload
data (see Fig.5.9 (a) and (b)). After the digital generation of the signal, sample values
are uploaded into the memory of the arbitrary waveform generator, output by DAC,
amplified by the driver and electrical-to-optical (E/O) converted by the LED operating
at 850 nm. At the receiver side, the signal is optical-to-electrical (O/E) converted
by the photodiode, amplified, converted to digital samples by the ADC and further
processed in the Matlab DSP.
Therefore, the signals are fed into the fast Fourier transform (FFT), by using the
coarse trigger timing obtained from the frame synchronization algorithm. After the
FFT, magnitude, and phase information are extracted for each OFDM subcarrier pilot.
Through a linear approximation of the phase-versus-frequency response, the fine timing
is determined from the slope of the phase information versus frequency.
Only pilots above a certain SNR level are considered for slope estimation. In the last
step, the distance is calculated by considering the reference information from calibrating
the system delays.
73
5. Positioning
Figure 5.13: 1-D distance measurement (blue) with single Tx and Rx OFE and after
using 10x averaging window (red).
5.4.1 Laboratory
In this section first, our simulation results are presented and then validated through
measurements. All results are obtained using the ToF technique.
5.4.1.1 Simulation Analysis
Simulation results are shown to verify the accuracy of the distance estimation under
ideal conditions by combining the coarse and fine timing information obtained by
processing the OFDM frame synchronization and the channel estimation based on the
MIMO preamble, respectively. Moreover, the effect of variable SNR on the positioning
accuracy is investigated. All results are obtained for the scenario with four Txs and
one Rx shown in Fig. 5.9 (a). Fig. 5.11 shows the distances calculated from coarse and
fine timing estimation for Tx number two.
After combining both curves and applying an additional fixed calibration factor,
the estimated distance (y-axis) is almost identical to the distance obtained from the
geometry (x-axis). Note that there is a small constant error between true and estimated
distance, which can be compensated by further improved calibration. The curves for
the other three Txs are very similar and show the same level of accuracy.
These results demonstrate that in the simulation environment, considering
bandwidth higher than 50 MHz, a positioning accuracy below 1 cm is possible. Moreover,
Fig. 5.12 evaluates the mean square error (MSE) versus SNR for each axis. The x- and
y-axis values are retrieved from equation 5.3. Then, the z-axis value is calculated by
knowing (
xi, yi, zi
)of each
Txi
, and the estimated ToF to each
Txi
. The results of
each axis show the typical 1/SNR behavior of the MSE and indicate a robust distance
74
5.4 Measurement Setups and Results
Figure 5.14: Influence of signal bandwidth on normalized distance.
estimation even at low SNR. All the distances between the TRxs and the Rx are very
similar. The x- and y- values are retrieved individually from the differences between the
measured distance values and thus are much more susceptible to measurement errors
compared to the z-value where, in principle, the four measured distances are averaged.
5.4.1.2 Measurements Results
In this section, measurement results are shown to validate thepositioning concept.
5.4.1.2.1 1D
First, a simplified unidirectional LiFi transmission between a single Tx and a single
Rx is used to estimate the distance between both OFEs. Fig. 5.13 shows the results
for a measurement with a 1.5 m distance between Tx and Rx, with the blue curve
showing the actual distance and the red one the result after 10 times averaging. It can
be observed that deviations are in the order of 1-2 cm only, with even smaller values
being possible due to averaging. Note that the small constant offset from the expected
1.5 m distance can be reduced by the calibration routine described below.
To demonstrate the influence of the signal bandwidth on the distance accuracy,
measurements with variable signal bandwidths have been performed. Fig. 5.14 shows
the normalized deviation of the distance with 10x averaging for signal bandwidths from
10 to 250 MHz.
There are two main observations: First, the distance errors become smaller at higher
signal bandwidths. Second, there is no further improvement above 100 MHz. The first
observation can be attributed to the better phase resolution at higher bandwidths, as
expected. The second observation is due to the transmission characteristics of our OFE
which has a low-pass characteristic, see Fig. 5.15 (a). Above 100 MHz, attenuation
75
5. Positioning
Figure 5.15: (a) Frequency response of the LiFi signal and (b) channel phase response
with linear regression.
increases due to the limits of the LED and the driver which results in reduced SNR
and increases the errors in the phase estimation. The linear approximation of the
slope should discriminate the results at higher frequencies as they are less reliable, see
Fig. 5.15 (b).
By considering Fig. 5.13 and Fig. 5.14, our initial simulation and measurement
results show the feasibility of the concept and that it is possible to achieve 1D positioning
errors below 1 cm using realistic LED-based OFEs.
5.4.1.2.2 3D
In the next step, the system is extended to multiple Txs, to evaluate true 3D positioning
of the receiver as shown in Fig. 5.10 (b). In this setup, Txs are located at the coordinates
(1,0.45,2) m, (0.35,0.45,2) m, (0.35,1,2) m, and (1,1,2) m. The Rx is moving to the
edges from the center of the room.
Error =√︁(xest −xact)2+ (yest −yact)2+ (zest −zact)2(5.15)
The 3D position is estimated by trilateration as described in 5.2.3.5. The
experimental results of the estimated 3D position through 40 iterations and the real
receiver location for selected places in the room are summarized in Fig. 5.16. The
difference between estimated and real 3D receiver positions are very small. The deviation
is slightly higher towards the edges of the room. As discussed below, this is related to
the initial calibration, which is more precise in the center of the coverage area.
Fig. 5.17 shows the resulting average MSEs, for each the x-,y-, and z-axis and
taking 40 independent measurements into account at each Rx location . The x-axis
shows generally higher MSEs. Similar to the simulation results, the errors of the z-axis
are the smallest, with one exception. In Fig. 5.18, the total 3D error, using (5.15), is
shown for three Rx positions over 40 measurements. There are two observations: First,
76
5.4 Measurement Setups and Results
Figure 5.16: Estimated position through 40-iterations (o) and real position (
∗
), and Txs
position (∇)in 3D view.
there is a different offset error for each position, with Rx(1,0.72,0) as the highest and
Rx(0.67,0.725,0) as the lowest. The offset error (
eoffset
) is attributed to the calibration
which has been done only in 1D here at a a particular location. Note that the offset
error reduces in the center between the four transmitters and it is increased at the
edges. Second, besides the offset, there is a variation around the offset, which is called
the random error (
erandom
), which is attributed to the noise. As the signal-to-noise
ratio varies as a function of distance, the random error depends on the receiver position.
These two 3D error vectors can be added as given in (5.16):
e=eoffset +erandom (5.16)
While the first term can be compensated by more careful calibration, the second
term can be reduced by averaging.
5.4.1.3 Summary
In this section, an indoor positioning system was presented for networked optical
wireless communications denoted as LiFi. The proposed positioning system is based
on time-of-flight measurements between multiple optical front-ends deployed at the
ceiling and a mobile device moving inside the overlapping coverage area. The advanced
positioning algorithm leverages already existing algorithms (frame synchronization,
channel estimation) that were previously designed for the communication capability
of LiFi being available when following the ITU-T recommendation G.9991. The
77
5. Positioning
Figure 5.17: Mean square error (MSE) of receiver for each axis after 40x averaging.
Figure 5.18: X, Y and Z axis error detection in receiver position [127].
application of this technique requires strict synchronization between the optical frontends
in the LiFi infrastructure which can best be realized using a distributed multiple-
input multiple-output (D-MIMO) architecture. The distances to all frontends can
be simultaneously estimated by using orthogonal pilot sequences which can be added
to G.9991 as additional channel estimation symbols. It was demonstrated that the
advanced positioning algorithm can reach precision below 1 cm with realistic optical
frontends when using a 1D positioning algorithm to determine the receiver coordinates.
In the experimental setup of 1
m×
1
m×
2
m
, the mean 3D distance errors of x-axis and
y-axis, and z-axis are less than 8cm at the edges.
78
5.4 Measurement Setups and Results
Figure 5.19: The transmitter and receiver are tilted in various directions.
5.4.2 Conference Room
The following trials have been used to further mature the positioning approach, to
improve its accuracy and demonstrate its potential in increasingly realistic application
scenarios.
As already mentioned, at least three valid signals are required to estimate the
3D position. Therefore, deployment of multiple Txs with sufficient spatial overlap is
essential to provide enough coverage in the intended area for both, communication and
positioning. In general, sufficient optical power at the receiver from at least three Txs
is required. One key parameter besides the Tx spacing is the FOV at the Tx and at
the Rx. Therefore, it is essential to investigate the maximum distance between Tx and
Rx, as well as the angle between Tx and Rx. These characteristics are examined for
our LiFi Txs and Rx as shown in Fig .5.19.
In the following, two measurement setups are explained. These experiments are
performed in a conference room with the dimension of 5
.
8
×
4
.
6
×
3
.
1
m3
as shown
in Fig. 5.19 and Fig. 5.20. OFEs are equipped with off-the-shelf LEDs operating at
860 nm (OSRAM SFH 4715 AS) and the Rx uses a large area photodiode (Hamamatsu
S6968) including an optical concentrator to enhance the effective coverage area. The
description of OFEs is elaborated in [46] and[32]. The DAC is a Spectrum DN 2.662-08,
and ADC is Spectrum DN 2.445-08, both running at 500 MS/s.
5.4.2.1 Tx and Rx FOV Characterization
At first, the FOV of a Tx and Rx are evaluated and optimized for the optical setup
as discussed in section 5.4.2. In Fig. 5.19 the setup is shown. Tx and Rx are placed
directly in front of each other to have full LOS at a distance
D
, where
D
= 3
,
4
,
5
m
.
At first, the Tx angle is tilted by 0
◦
, 30
◦
, 45
◦
, 60
◦
, 70
◦
, 80
◦
, 90
◦
with respect to
the vertical axis, while Rx angle is fixed to zero until the distance recognition failed.
Afterward, the Tx angle is kept to zero and Rx angle adjusted to 0
◦
, 30
◦
, 45
◦
, 60
◦
,
70 ◦, 80 ◦, 90 ◦.
79
5. Positioning
Figure 5.20: Measurement setup including 4Txs and one Rx. Rx is placed in three
different locations of the transmitter cell.
This is repeated for three distances and at various combinations of Tx and Rx
angles. In Fig. 5.21, the corresponding distance error (mean square error (MSE) of
distance) is shown for 3, 4, and 5 m distance including different Rx and Tx angles. It
is observed that tilting the Tx and Rx to higher angles leads to increasing the MSE
of the distance estimation. As expected, a larger the distance the higher is the error.
Both effects are directly related to the received power at the Rx. Lower power levels
result in reduced signal-to-noise ratios leading to imprecise ToF estimation until the
point where no distance estimation is possible anymore.
For instance, in the case of Tx tilt angles beyond 70
◦
, no valid distance estimation is
possible. For the Rx tilt angles case, beyond 70
◦
the system is still operating, however,
the MSE value is too large. This can be attributed to the concentrator infront of the
photodiode, which collects only multipath signals having longer delays at these large
FOVs. At last, Tx and Rx angles changed simultaneously and distance recognition
is evaluated similarly to the previous step. The results are shown in Fig. 5.22. This
investigation has shown, that a combination of Tx and Rx tilt angles
>
0
◦
results in a
stronger increase of the MSE, as expected. As a rule of thumb, tilt angles up to 50°are
possible with the optical frontends used in this study.
5.4.2.2 Ranging
In the next step, a ranging measurement was applied between four Tx units and a single
Rx unit in a conference room scenario as shown in Fig. 5.20. The Tx units are deployed
in such a way, that for every position inside the rectangle, at least three valid signals
can be received at the Rx. The Rx is located at three different locations, all at the
same height. The first location (location 1 in Fig. 5.20) is directly below
Tx3
, at the
edge of the room. The second location, (location 2 in Fig. 5.20) is between
Txs3,4
and
the last location (location 3 in Fig. 5.20) is in the middle of the room. The distances
80
5.4 Measurement Setups and Results
Figure 5.21: MSE of measured distance and actual distance based on Tx and Rx angle
rotation for D:3,4,5 m (D: distance between Tx and Rx) in log scale. Tx angle rotation
reaches beyond 70◦due to the large FOV of the photodiode.
towards each Tx were estimated for these locations and a distance error was calculated
using Eq. 5.17.
RMSED=1
n
n
∑︂
i=1 √︁(Distanceest −Distanceact)2(5.17)
The measurement was repeated several times at each location, with
n
denoting the
number of iterations in Eq. 5.17, where
Distanceest
and
Distanceact
are estimated and
compared to the actual distance between Tx and Rx.
Fig. 5.23 shows the MSE for each Rx location including measurement error. At
location 1, with Rx directly below
Tx1
, the MSE for
Tx1
is the smallest and for
Tx4
the highest. This is as expected since
Tx4
is the furthest away and has the largest angle
relative to Rx. At the second location, the Rx is between
Tx2
and
Tx4
, therefore, the
MSE has lower values at these two distances. For the third location, with Rx placed in
the middle of the Txs, the error towards each Tx has a similar value. The overall error
for all measurements is very small (
<
1 cm) which shows the robustness of the system.
These results further indicate that a larger grid of the Tx units will be possible since
only three Txs are required for the 3D-position estimation.
5.4.2.3 Summary
Altogether, the conference room results show that the precision of distance measurement
is in the range of below 5 centimeters, indicating that LiFi positioning is a promising
solution for indoor localization applications. In the next step, the 3D positing of the
Rx will be evaluated after enabling the system to track a moving object in a real
environment.
81
5. Positioning
Figure 5.22: MSE of measured distance and actual distance based on the simultaneous
rotation of Tx and Rx for D: 3, 4, 5 m.
Figure 5.23: MSE of estimated distance when Rx is placed in three different locations.
5.4.3 Object Tracking
5.4.3.1 Scenario and setup
A distributed multiple-input single-output (D-MISO) architecture is used now in a
factory scenario for indoor positioning and object tracking, as shown in Fig. 5.24. Our
test setup includes four transmitters (Tx)s and one MU as a receiver (Rx). The signal
structure and the channel estimation are based on ITU-T recommendation G.9991 [133,
91] as described in the digital processing subsection and in Fig. 5.24 (a)-(c).
The number of required transmitters for positioning depends on the area covered
by each LiFi transmitter. Following e.g., [134], at least three valid signals are necessary
to detect the position of the MU. Therefore, arranging transmitters in a way to always
provide these three signals is essential. The coverage of each LiFi transmitter is defined
by its FOV and its height over the floor.
82
5.4 Measurement Setups and Results
Figure 5.24: System architecture: (a) Transmitter signal, (b) Received signal (c) Packet
Structure, (d) Measurement setup showing LiFi Txs overlap and thirteen location indexes of
mobile user (MU).
Figure 5.25: (a) Ranging root-mean-square-error (RMSE) for 2 m Txs setup (b) Ranging
RMSE of 1.5 m Tx setup, (c) x, y, z RMSE of Rx for 2 m Txs setup, (d) x, y, z RMSE of Rx for
Txs setup of 1.5 m.
83
5. Positioning
Figure 5.26: (a-d) Ranging RMSE heatmap for each individual Tx in the 1.5 m cell configuration,
each color indicates the corresponding error interval (e) online tracking results for a mobile object.
The object follows an arbitrary path, and its location is detected as shown by circles.
Figure 5.27: (a-c) RMSE heatmap for each x, y, and z direction considering 1.5 m cell
arrangement.
The LiFi transmitter and receiver are the same modules that have been used in
section 5.3.1. The four LiFi Txs consist of a custom-designed LED driver and off-the-
shelf LEDs operating under eye-safe conditions at 860 nm (OSRAM SFH 4715 AS)
[91]. Each Tx is driven by an independent signal, with a signal structure as discussed
above. The channel estimation sequences of each Tx are orthogonal, which allows the
simultaneous estimation of the frequency responses used to calculate the fine timing
information [91]. Five parallel channels of the DAC (Spectrum DN 2.662-08, 625 MS/s)
were utilized one for a reference signal and four for the LiFi Tx frontends. The reference
signal is required to allow time synchronization of the ADC and the DAC.
Two LiFi cell configurations have been evaluated, where the Txs are arranged in
a square layout with 1.5 m and 2 m spacing. The positioning accuracy of the MU
has been evaluated at thirteen locations marked in Fig. 5.24 d). At each location,
n
= 40 measurements have been performed. From these sequences, the root mean
squared errors (RMSE) for the distances between all Txs and the MU and for the actual
3-dimension (3D) position have been calculated.
84
5.4 Measurement Setups and Results
5.4.3.2 Results and Discussion
It indicated that a 2 m
×
2 m grid between the LiFi transmitters can be regarded as the
upper limit for our position system to provide centimeter-ranging accuracy.
The corresponding 3D RMSE of each x, y, and z axis for the 2 m and 1.5 m cell
are shown in Fig. 5.26 (c) and (d). It is calculated based on (5.18), (5.19) and (5.20),
where
xest
and
xact
are estimated and actual value of x-coordination, similar for y and
z-coordination.
xRMSE =1
n×√︁(xest −xact)2(5.18)
yRMSE =1
n×√︁(yest −yact)2(5.19)
zRMSE =1
n×√︁(zest −zact)2.(5.20)
Fig. 5.25 (c) shows that the highest error belongs to location 2 in the corner of the
cell. In Fig. 5.25 (d), the average RMSEs of each x, y, and z-axis for all MU locations
are 2, 3, and 1 cm. For the 2 m cell configuration in Fig. 5.25 (a, c), similar observations
are made, however, the overall RMSEs are generally higher, (see Fig. 5.25 (a)) since the
signal-to-noise ratio is lower. At least for one position (location 2) only two Tx signals
with errors below 10 cm can be detected, which leads to less accurate positioning.
The heatmaps corresponding to the ranging errors of each transmitter are shown in
Fig. 5.26 (a-d). It is observed that the ranging RMSE increases at a larger distance
from each Tx, which indicates that each Tx can cover most of the intended cell, and a
combination of 3 Tx allows full coverage. Fig. 5.26 (e) illustrates the measured location
of an object moved to a number of locations and successfully tracked with the system
as shown by the circles. Despite the relatively low data processing speed of DAC and
ADC, Matlab processing has been optimized so that 1 measurement / second could
be realized, which limits mobility. This allows a first real-time positioning experience,
however, still with limited mobility. Real-time implementation of Field Programmed
Gate Arrays (FPGA) remains as task of future development.
Fig. 5.27 demonstrates the RMSE of each x, y, and z-direction for the 1.5 m cell
arrangement. It shows that the x and y-direction have slightly higher RMSE compared
to the z-axis. It can be observed that the lowest and highest RMSE in the x and y-axis
are between 0.7 cm and 5 cm leading to an average accuracy of below 3 cm.
5.4.3.3 Summary
In this section, object tracking in an indoor scenario has been investigated. The
proposed approach has been validated through experiments in two different LiFi cell
layouts. Our results indicate that the average precision of positioning in each dimension
can be below 3 centimeters. Although implemented in Matlab, the demo system allows
continuous tracking of a mobile object’s position at 1 measurement per second. LiFi
positioning is considered a valuable new feature that has the potential for centimeter
accuracy, is implementable with limited efforts in future chipsets, and notably promising
for many indoor applications, such as in the industrial Internet of Things.
85
5. Positioning
Figure 5.28: (a) Block diagram of OFDM transmitter and receiver for proposed LiFi positioning
system. (b) Measurements setup in the factory hall. Red dots show measurement points.
5.4.4 Post Optimization in a factory
The final experiment was performed at Weidmueller’s factory as presented in Fig. 5.28
(a,b). The location is inside the hallway between two machine shop floors, where
typically AGV move around to transport materials.
5.4.4.1 Measurement Scenario and Setup
Two LiFi neighboring cells have been set up, as shown in Fig. 5.28 (b), with 6 Txs
placed in a rectangular plane with around 1.5 m distance, covering around 4.5 m
2
. The
actual positions of the transmitters have been initially estimated with a laser range
meter, with the origin (0,0) located in the upper right corner below
Tx1
. Within the
coverage area, 5
×
7was defined, i.e., 35 points for evaluating the position estimation
of Rx. Before starting the evaluation, a calibration for each transmitter was performed
at two points in the area, to estimate the additional delay due to the cables, devices
and optical frontends.
The 6 Txs consist of a custom-designed LED driver and the off-the-shelf LEDs
(OSRAM SFH 4715 AS). Each Tx is driven by a separate channel of a digital-to-analog
converter (DAC) (Spectrum DN 2.662-08, 625 MS/s) with a sample rate of 625 MS/s.
The signals of all Txs are received superimposed at the Rx. The Rx consists of 5
large-area silicon photo-diodes (Hamamatsu S6968) with 150
mm2
photosensitive area
and a custom design trans-impedance amplifier. The signal is then connected to a
channel of a 14-bit, 8-channel ADC (Spectrum DN 2.445-08) clocked at 500 MS/s. In
addition, one electrical channel between the DAC and ADC is used for a reference
signal, to allow synchronization between all Txs and the Rx.
5.4.4.2 Ranging and 3-D Evaluation
The ranging and 3-D evaluation was done in two steps, the initial calibration and the
actual measurements with and without optimization. Fig. 5.29 (a) shows the flowchart
of the positioning measurements without and with optimization, and Fig. 5.29 (b) for
the calibration.
86
5.4 Measurement Setups and Results
Figure 5.29: (a) Flowchart for evaluating regular (green boxes) and optimized (orange boxes)
distance as well as 3-D location information. (b) Flow chart for the initial calibration measurement.
(c) Ranging error and its linear correction factor as a function of the angle. (d) Spatial relationship
between Tx and Rx pair.
The regular measurement without the optimization is shown in Fig. 5.29 (a) by
the green boxes. At first, the distances
Di
,
i
= 1
,
2
,
3
,
4
,
5
,
6, between each of the six
Txs and the Rx have been obtained by combining coarse and fine timing estimation as
discussed above. Using only the ranging information from the 3 Txs with the highest
SNR together with the known position of the Tx (
Xi, Yi, Zi
), the coordinates of Rx
(X0, Y0, Z0)can be calculated:
D2
i= (X0−Xi)2+ (Y0−Yi)2+ (Z0−Zi)2(5.21)
Because of the same height of all Txs in our configuration, only
X0
and
Y0
can
be obtained from the Eq. 5.21. Hence, the next step is to calculate the Z-direction
information (Z0) using the estimated X0and Y0:
Z0=1
6(Zi−√︂D2
i−(X0−Xi)2−(Y0−Yi)2)(5.22)
The ranging and 3-D information are now compared to the actual distances and
positions, which have been acquired in advance with a laser range meter to calculate
the errors. Before discussing these results, the optimization is explained in the next
section.
5.4.4.3 Optimization
In order to estimate the location of the Rx more accurately, the relation of the ranging
error and the tilting angle was investigated. As mentioned in our previous work [134], a
higher tilted angle between Tx and Rx showed a higher ranging error, regardless of the
overall distance. In Fig. 5.29 (c) this relation is shown for a single Tx by the blue curve.
87
5. Positioning
Figure 5.30: (a) Sector segmentation of 35 marked reference points based on
T x4
.(b)
Relationship between ranging error and angle for four sectors for
T x4
. Blue curves: actual
values from calibration measurement; red curve: linear approximation; yellow curve: ranging error
with applied correction factor.
The ranging error is defined as
DE
=
Dactual −Di
with
Dactual
the actual distance,
Di
the estimated distance, and the angle θis defined as shown in Fig. 5.29 (d).
One would expect that the error varies around zero and increases with higher angles,
due to the lower SNR at higher distances. Instead, an approximately linear behavior
can be observed, which allows the introduction of a calibration factor. The linear
behavior is attributed to the different propagation properties of LED modes at different
angles. However, further investigations are required to verify this.
The first step of the proposed optimization method consists of an initial calibration,
the framework is shown in Fig. 5.29 (b). By the measurements of 35 calibration points
in the LiFi cell, the ranging error vs. angle behavior for each Tx is estimated.
The angle
θ
between each Tx and Rx at different locations, as shown in Fig. 5.29
(d), is known at the calibration points. Now a linear approximation for the ranging
error
DE
=
Dactual −Di
vs. the angle
θ
is performed as shown by the red curve in Fig.
5.29 (c).
This linear approximation allows the introduction of a correction factor
CFθi
for
each Tx. The actual value of the
CFθi
at any measurement point can be calculated by
multiplying the estimated angle
θi
with the slope of the linear approximation (
mApprox
),
i.e., the slope of the red curve in the figure, as follows (5.23):
CFθi=θi×mApprox +const. (5.23)
Note that the linear approximation needs to be calculated only once for each Tx
and can be used for optimizing any number of further positioning measurements.
Future positioning measurements with optimization would then apply the correction
factor by the steps shown in Fig. 5.29 (a): First, the usual ranging is performed to
88
5.4 Measurement Setups and Results
Figure 5.31: (a) Ranging error of regular measurement for
T x4
for each position. (b) Ranging
error information of corrected optimized measurement of T x4for each position.
each visible Tx, followed by a rough 3-D estimation of the Rx. Second, the angle
θi
between Tx and Rx is calculated from the 3-D estimation, and the correction factor
based on
θi
is applied based on the initial ranging information, which in turn provides
a corrected ranging information and further a corrected 3-D position:
Dnew =Di+CFθi(5.24)
For a better linear approximation, the calibration points were sectored depending on
their position. Fig. 5.30 (a) shows the used segmentation for
Tx4
. As can be seen, four
sectors have been established with each covering 45
◦
of the LiFi cell. The calibration
points were allocated to each sector accordingly, with some of the points being used
in neighboring sectors. This is done for each of the other five Txs as well. Fig. 5.30
(b) shows the results for all four
Tx4
sectors. The blue curve shows the relationship
between the ranging error
DE
and angle
θ
from the initial calibration measurement and
the red curve is the linear approximation of the error. It can be observed that for some
sectors, e,g., in sector 2, the linear approximation is a good fit to the original points,
while in Sector 1, only a limited approximation can be achieved. The yellow curve in
Fig 5.30 (b) shows the resulting new ranging error after applying the correction factor,
in the case of the calibration measurement.
5.4.4.4 Results and Discussion
Now the results of the regular and the optimized measurements are compared in terms
of ranging error and 3-D error. At first, the results of the regular measurement are
discussed.
5.4.4.4.1 Result without Optimization
The ranging errors for the regular measurement are shown in Fig. 5.31 (a) for
Tx4
for
each of the 35 investigated positions of Fig. 5.30 (a). In these figures, the maximum
and the minimum of the distance error over 50 measurements are shown, as well as
89
5. Positioning
Figure 5.32: (a) Original ranging RMSE heatmap of
T x4
. (b) Optimized ranging RMSE
heatmap of T x4.
Table 5.2: Regular and Optimized Ranging Measurement
Status
Txs Tx1 Tx2 Tx3 Tx4 Tx5 Tx6
RMSE RMSE RMSE RMSE RMSE RMSE
(m) (m) (m) (m) (m) (m)
Before
Correction 0.042 0.068 0.029 0.041 0.044 0.026
After
Correction 0.038 0.042 0.022 0.011 0.02 0.014
the average by the red square. It can be observed that the variation of the error is
relatively high when the Rx is the furthest away, for instance, at the points on the
far side close to
Tx1
,
Tx3
and
Tx5
. This is as expected since the SNR is low due to
the higher distances to
Tx4
. On the other hand, points close to
Tx4
show much lower
variations of the error, e.g., at points 20, 24, and 25. The corresponding heat-map of
the ranging Root Mean Square Error (RMSE) for the regular measurement of
Tx4
is
shown in Fig. 5.32 (a). The distribution shows lower values close to
Tx4
and increasing
values towards the corners. The maximum value of the ranging measurement RMSE is
about 12 cm corresponding to Fig. 5.31 (a).
The results for all Txs are summarized in Table 5.2 in the first row. For all Txs the
RMSE is between 2.6 and 6.8 cm, with
Tx2
the highest and
Tx6
the lowest. Overall
the RMSE is very similar for all Txs.
With the ranging information available, the 3-D position can be calculated and
compared to the real ones. Fig. 5.33 (a), (b), and (c) and show the heatmap of the
positioning RMSE in X, Y, and Z-direction. The performance varies slightly in the
different sectors of the LiFi cell. The X direction shows higher errors in the middle
of the cell, while the Y and Z directions show high errors in the top left corner. The
small areas with very high errors (> 15 cm) can be attributed to poor SNR conditions
at these points. The overall RMSEs for X, Y, and Z directions are 7 cm, 6 cm, and
3 cm as shown in Table 5.3.
90
5.4 Measurement Setups and Results
a) b)
d) e) f)
0.00
0.03
0.06
0.09
0.12
0.15
0.18
0.21
0.24 00.51.5 1.0
1
2
3
Cell X-Axis
Cell Y-Axis
Original RMSE in X-direction [m]
0.00
0.02
0.05
0.07
0.09
0.11
0.14
0.16
0.18
0
0.51.5 1.0
Cell X-Axis
1
2
3
Cell Y-Axis
Original RMSE in Y-direction [m] 0.000
0.008
0.015
0.023
0.030
0.038
0.045
0.053
0.060
1.5
Original RMSE in Z-direction [m]
0
0.5
1.0
Cell X-Axis
1
2
3
Cell Y-Axis
0.00
0.03
0.06
0.09
0.12
0.15
0.18
0.21
0.24
Corrected RMSE in X-direction [m]
00.51.5 1.0
1
2
3
Cell Y-Axis
Cell X-Axis
0.00
0.02
0.05
0.07
0.09
0.11
0.14
0.16
0.18
0
0.5
1.5 1.0
1
2
3
Cell Y-Axis
Cell X-Axis
Corrected RMSE in Y-direction [m]
0.000
0.008
0.015
0.023
0.030
0.038
0.045
0.053
0.060
1.5
Corrected RMSE in Z-direction [m]
0
0.51.0
Cell X-Axis
1
2
3
Cell Y-Axis
c)
Tx4
Tx1
Tx2
Tx3
Tx5
Tx6
Tx2 Tx1
Tx4 Tx3
Tx6 Tx5
Tx2 Tx1
Tx4
Tx4
Tx3
Tx3
Tx3 Tx3
Tx6
Tx6 Tx6 Tx6
Tx5
Tx5 Tx5 Tx5
Tx4
Tx4
Tx1
Tx1
Tx1
Tx2 Tx2 Tx2
Figure 5.33: (a-c) Regular RMSE in X, Y, and Z-direction. (d-f) Optimized RMSE in each X,
Y, and Z-direction.
5.4.4.4.2 Results with Optimization
The Fig. 5.31 (b) shows the behavior of the ranging error after optimization, again
exemplary for
Tx4
. Comparing with the initial ranging errors in Fig. 5.31 (a), it can
be seen that the optimization yields a remarkable reduction of the average ranging
error, especially at higher distances from
Tx4
, e.g., at points 26 and 31. Note that the
error spread is not affected by the optimization, due to the way how the correction
factor is applied.
In Fig. 5.32 (b), the corresponding heatmap of the ranging RMSE of
Tx4
is shown.
There are two interesting observations. First, the overall RMSE is clearly decreased and
second, the high RMSEs especially at the corners could be reduced significantly. For the
other five Txs, a similar behavior can be observed as shown in Table 5.2. For
Tx2,4,5,6
,
an improvement of about 50% can be observed and for
Tx1,3
, the improvement is about
15%. The different optimization outcomes, e.g., for Tx1 and Tx4, can be attributed to
the different behavior of the LEDs and the quality of the linear approximation.
Using the corrected ranging estimation, the 3-D position can be calculated and
compared to the actual value. The RMSEs of the corrected position information for X,
Y, and Z-direction are shown in Fig. 5.33 (d), (e), and (f). It can be observed that the
areas with higher RMSEs, i.e., green, yellow, and red, are significantly smaller after
optimization. Furthermore, the areas with very high errors, e.g., in the top left corner
for Y and Z and the middle left area for X direction, could be minimized as well.
In Table 5.3 the RMSE in the X, Y, and Z-directions and the total mean RMSE
before and after the optimization are shown. The RMSE for each direction could be
reduced by about 2 cm, down to 5, 4, and 1 cm in X, Y, and Z directions, respectively.
The mean RMSE over all three directions could be reduced from 5.3 down to 3.3 cm.
91
5. Positioning
Table 5.3: Regular and Optimized 3-D Measurement
Status
Direction X-axis Y-axis Z-axis Mean
RMSE RMSE RMSE RMSE
(m) (m) (m) (m)
Before Correction 0.07 0.06 0.03 0.053
After Correction 0.05 0.04 0.01 0.033
5.4.4.5 Summary
In this section, a LiFi-based positioning system was tested with and without optimization
in a real factory environment. The results demonstrate an average accuracy of about
7, 6, and 3 cm in the X, Y, and Z-direction, respectively. To improve the accuracy of
the proposed positioning system, a correction factor is introduced, based on the tilt
angle. For the 3-D information, the accuracy could be optimized to about 5, 4, and
1 cm for the X, Y, and Z-direction, respectively, and the overall RMSE of the position
information is reduced from 5.3 to 3.3 cm.
5.5 Requirements for Chipset Integration
Currently, ITU-T G.9991 Recommendation is approved by ITU-T for LiFi applications.
It allows building LiFi systems by reusing existing chipsets designed for home networking
applications, creating this way an early mass market for LiFi. LiFi chipsets that support
positioning must fulfill a number of new characteristics. This section gives a short
description of these properties, their availability in current chipsets, and a possible
roadmap for standardization.
5.5.1 Chipset Description and Current Capabilities
The high-level architecture of the digital baseband G.9991 chipset is shown in [135] (page
9). The baseband chip decodes the frames coming from the channel and injects frames
in the channel through an analog frontend chip that performs the signal adaptation to
the medium.
The positioning techniques explained in this section are supposed to run on the
digital baseband by leveraging the capabilities the chip offers. While the OFDM engine
is partly implemented using a hardware accelerator, the positioning algorithms may
run in the embedded microprocessor that has access to the internal registers of the chip
through a set of application programming interfaces (API). The techniques that have
been described in this research can already be partially implemented using commercially
available chipsets. In particular, the RTTOF mechanism which is future plan for the
LiFi positioning system, can be implemented using the time tagging capabilities of the
chipset. The chipset can make use of some of the functionalities and the framing of the
standard that allows for refining the procedure explained in our presented framework.
For example, ITU-T G.9991 defines a bidirectional frame exchange between two nodes.
In this frame exchange, a transmitter sends a BMSG frame to the receiver and the
92
5.5 Requirements for Chipset Integration
receiver immediately transmits a BACK frame to the transmitter as soon as the first
frame is received.
By using a BMSG/BACK exchange,
Tb
=(
T3−T2
)(Fig. 5.9 (c)) becomes
a well-known fixed number by the construction of the chipset following G.9991
Recommendation. Therefore, equation 5.13 is simplified, i.e. only
T1
and
T4
need to be
measured. This
T1
and
T4
information can be extracted from the network time reference
(NTR) fields of the BMSG and BACK frames. This means that for positioning,
T3
and
T4
information do not need to be communicated to the transmitter. This makes the
system more flexible (no additional protocol needed) and reduces the latency in the
positioning measurement (avoiding the time lost during the exchange of information).
However, the limitation of this method comes from the fact that the NTR field is
limited to a precision of 10 ns. That means following (5.14), the precision that can be
achieved with the current chipsets is limited to some meters.
5.5.2 Future Chipset Development
After the experiments performed in the framework of this research, it can be realized
that chipsets need to include several improvements in the next evolution in terms of
both, hard- and software. Only in this way, obtaining the high precision of positioning
predicted in this research can be realized also in practice. The main two functionalities
that will be needed are briefly described in the following.
5.5.3 Increase the Precision of NTR Field
The current version of the ITU-T G.9991 Recommendation specifies an accuracy of 10
ns for the NTR counter. To obtain this, an internal clock of 100 MHz is needed. The
clock frequency can be escalated in future implementations up to, e.g., 1 GHz. This
would automatically increase the precision of the positioning system by a factor of ten.
Accompanying the increase of the clock frequency in the chipset, a new amendment to
the standard will be needed to specify a higher accuracy of the NTR field.
5.5.4
Provide Better Access to Lower Layer Information of Upper
Layers
To fine-tune the positioning information, as described in this work, additional
information on the CFR and the SNR needs to be conveyed from the lower layers of
the chips (i.e. the PHY) to the upper layers where the positioning information can
be computed. In particular, the positioning algorithm needs access to the complex-
valued channel frequency response (CFR). In the current chipset implementation, this
information exists since it is necessary for the correct decoding of the payload of
the physical frames. Therefore, it is necessary to perform an accurate estimation of
the channel frequency response (for frequency-domain equalization to be used by the
OFDM engine). While the amplitude information is accessible, e.g. to compute the bit
loading, the phase-versus-frequency information which is required for the fine timing is
not accessible to the upper layers. In next-generation chipsets, a new API has to be
93
5. Positioning
developed to provide such information on the complex-valued CFR to the upper layers.
This evolution implies a change in the product but not necessarily a change in the
ITU-T recommendation. The exact information to be exchanged between the physical
layer and the upper layer is still to be defined and depends on the exact algorithms to
be utilized.
5.6 Conclusions
In this chapter, an indoor positioning system is demonstrated for networked optical
wireless communications denoted as LiFi. The proposed positioning system is based
on time-of-flight measurements between multiple optical front-ends deployed at the
ceiling and a mobile device moving inside the overlapping coverage area. The advanced
positioning algorithm leverages already existing algorithms (frame synchronization for
coarse timing, channel estimation for fine timing) that were previously designed for the
communication capability of LiFi following the ITU-T recommendation G.9991. The
use of the standard for positioning requires strict synchronization between the optical
frontends in the LiFi infrastructure, which can best be realized using a distributed
multiple-input multiple-output (D-MIMO) architecture. The distances to all frontends
can be simultaneously estimated by introducing orthogonal pilot sequences which can
be added to G.9991 as additional channel estimation symbols. It is shown that the
advanced positioning algorithm can reach 1-D precision below 1
cm
with realistic optical
frontends. By using trilateration to determine the receiver coordinates, the mean 3D
distance errors of the x-axis, y-axis, and z-axis are around 3
cm
. Implementation of
the positioning is not possible with the current G.9991 chipsets due to previous design
choices. In future generations of G.9991 chipsets, the described positioning technique
can be achieved by i) synchronizing the OFEs in the infrastructure, ii) implementing
the MIMO pilots, and iii) enabling access to complex-valued channel frequency response
measurements for higher layers.
94
6
Conclusions
The main focus of this Ph.D. thesis was on integrating optical wireless communication
and positioning by using the distributed multiple - input multiple - output (MIMO)
topology. This work specifically targets on enhancing the robustness and reliability
of optical wireless communications and providing an additional time-of-flight-based
positioning capability jointly.
After introducing the state of the art and summarizing the new contributions in
this thesis in Chapter 1, the basic concept of distributed MIMO, with a central unit
connected to multiple distributed units via fronthaul links for jointly serving multiple
mobile units in the service area, was introduced in Chapter 2.
In Chapter 3, the use of plastic optical fiber (POF) as a front-haul link was
investigated [46]. First, the analog link properties were evaluated and in the next
stage, the concatenation of POF link and the optical wireless link and their limitations
were examined. In an early stage, a primary test-bed for the single-input-single-output
analog POF transceiver was designed and developed. The frequency response and link
properties were evaluated experimentally. Next, the SISO POF link was combined with
the wireless link. The performance of the combined wired and wireless link was evaluated
experimentally and the results showed that data rate requirements of industrial wireless
communication can be met. This initial link was unidirectional performance and
vulnerable to shadowing and blockage. Therefore, bidirectional operation and MIMO
were introduced to improve the reliability and robustness of the system. To reach
bidirectional operation, the amplify-and-forward approach was selected. A major
problem was to overcome the substantial gain variation in the combined fixed-wireless
link. Therefore, a best-practice approach was developed: Introduce a fixed gain amplifier
to compensate for the losses in the POF link and place an automatic gain control (AGC)
at the central unit to compensate for the variable part of the path loss in the optical
wireless channel. In this way, a bidirectional LiFi over POF link was setup in the
laboratory and tested. Experimental results showed maximum data rates of 901 Mbit/s
and 725 Mbit/s in the up-and downlink, respectively, while the minimum rates were 115
and 218 Mbit/s, all essentially limited by the wireless link. These results show that the
95
6. Conclusions
LiFi over the POF approach is practical for the intended use cases of LiFi with mobile
users [36]. To realize a distributed MIMO link for optical wireless communication,
next, the bidirectional setup was extended to multiple distributed units connected via
POF used as a fronthaul to the central unit. To facilitate comparison between theory
and measurements, a common signal model was proposed to evaluate the performance
of the LiFi link, including characteristics of the fixed and wireless links and a first
experimental setup for D-MIMO for LiFi was presented. The performance of the link
was evaluated for two scenarios in which access points and users are located apart or
close to each other. The results show that the performance is mostly limited by the
wireless link and depends on the locations of distributed optical frontends (D-OFE)s
and mobile users. By placing both at a reasonable distance, the throughput can be
increased. For closely spaced users, diversity is better than spatial multiplexing.
In Chapter 4, the experimental system was finalized for functional tests of MIMO
communication. System performance was analyzed by using both, a theoretical
MIMO signal model including the combined wired link and wireless links, and the
experimental link. A distributed MIMO link can be operated in several transmission
modes: Spatial diversity, spatial multiplexing, and multi-user multiplexing. Simulations
and measurements were conducted to assess the performance of these transmission
modes across various scenarios. In general, the performance depends critically on the
location of the mobile users. For closely spaced users, spatial diversity offers better
performance than spatial multiplexing without angular diversity. Selection combining
provides impressive gains over equal gain combining, which is attributed to the spatial
selectivity of the optical wireless channel. It has considered spatial multiplexing in
combination with angular diversity, which improves the performance for different
locations of the users. In scenarios with low SNR, particularly in the uplink direction, it
was noted that multi-user multiplexing can yield additional advantages. As an outlook,
future work needs to focus on dynamic switching between these spatial modes, i.e.
single- and multi-stream transmission for single and multiple users, to maximize the
performance in each scenario. Moreover, larger scenarios with more optical front-ends
and users should be considered, and sufficient statistics of the scenarios.
Besides the use of distributed MIMO for LiFi as a communication link, in Chapter
5, the use of the same system for indoor positioning was investigated. At first, wireless
positioning was reviewed including radio-, camera-, and LiFi-based approaches. For LiFi,
there are power-, angle- and time-of-flight-based algorithms. The proposed positioning
system measures the time-of-flight (ToF) between multiple optical frontends deployed
at the ceiling and a mobile device moving inside the overlapping coverage areas of
these optical frontends. The advanced positioning algorithm leverages already existing
waveform following the ITU-T G.9991 recommendation, namely frame synchronization
for coarse timing and channel estimation for fine timing that were previously developed
for the communication capability of LiFi [91, 134]. The application of the ToF
technique in 3-D requires strict synchronization between the optical frontends in the
LiFi infrastructure which is best realized using the distributed MIMO architecture. The
distances to all frontends can be simultaneously estimated by using orthogonal pilot
96
sequences which can be added to G.9991 as additional channel estimation symbols. To
detect the receiver position using the ToF technique, the following steps are required: 1)
At least three signals are received at the Rx in parallel, one from each Tx. 2) ToF and
relative distance are determined by considering coarse and fine-timing information. 3)
The x, y, and z coordination of the receiver are calculated using the trilateration method
taking the known Tx position into account. It was demonstrated that an appropriate
transmitter arrangement is needed to ensure three sufficiently strong signals at the
receiver in the whole coverage area. Transmitter parameters such as power, field of
view (FOV), radiation pattern, and the FOV at the receiver are important to identify
the optimal distance between the transmitters. Initially, a SISO setup was setup and the
proposed ToF scheme implemented. By varying the distance between the transmitter
and receiver, as well as the angle, the mean square error (RMSE) of the estimated
distance was calculated indicating the below 1 cm precision is possible in 1-D. Next, a
4×1 MISO setup was used to evaluate 3-D positioning in a laboratory setup. It was
demonstrated that the advanced positioning algorithm can reach high precision with
real optical frontends by using a 3D trilateration algorithm. In the experimental setup
of 1
m×
1
m×
2
m
, the mean 3D distance errors of the x-axis and y-axis, and z-axis are
less than 8
cm
at the edges, with significantly better lower values between the optical
frontends. Further, object tracking in an indoor scenario has been investigated for two
different cell layouts, where a tighter arrangement allows in principle better accuracy,
but limits the coverage area and requires a higher number of ceiling OFEs, which is not
cost-effective. Therefore, an optimum distance between the transmitters at the ceiling
needs to be determined. The system was further developed towards continuous tracking
of the mobile objects and determining the 3D position in an indoor environment.
Although implemented in Matlab, the demo system allows continuous tracking of a
mobile object’s position at 1 measurement per second. A positioning optimization
technique was introduced to counteract a systematic error. It was observed that an
increased tilt angle between the transmitter and receiver results in higher-ranging
errors, which can be modeled as a linear behavior. The optimization method measures
the linear relation between error and tilt angle and uses then a linear correction to
reduce the ranging error for any number of positioning measurements. Results show
that for the 3-D information, the accuracy could be optimized to about 5, 4, and
1 cm for the X, Y, and Z-direction, respectively, and the overall RMSE of the position
information is reduced from 5.3 to 3.3 cm. The discussed LiFi positioning approach can
be implemented using existing LiFi standards for communication and can be considered
a potentially valuable tool with great promise for smart factories. Implementation
in current G.9991 chipsets limits the precision due to previous design choices. In
future generations of G.9991 chipsets, the described positioning technique can be
realized by using well-understood technologies with reasonable effort. In summary, LiFi
positioning presents a valuable novel capability with the potential to achieve centimeter-
level accuracy. Its integration is feasible with minimal effort promising for various
applications in the industrial Internet of Things, where precise indoor positioning is a
highly demanded essential feature.
97
6. Conclusions
While this research was ongoing, joint communication and sensing (JCAS) developed
into a hot topic towards the development of the 6G mobile systems. This thesis has
shown that LiFi is well suitable for both capabilities when choosing the distributed
MIMO topology. While for communication, LiFi can reuse well-known transmission
schemes also applied in mobile radio systems, in particular the dynamic MIMO mode
switching concept. For positioning, LiFi can realize 10x-100x improved precision
compared to radio-based positioning schemes, due to the dominant LOS-based
propagation. Both results illustrate that LiFi is a good candidate to densify Wi-
Fi networks and add unprecedented features, such as very high data rates in dense
cells and centimeter accuracy for indoor positioning. As future work, there are still
several steps to integrate positioning and communication using the LiFi technique that
need to be assessed. Therefore the suggested changes need to be taken into account for
the G.9991 chipset. Moreover, integrating the angle of diversity to the user side is an
essential enabler to provide enough coverage and mobility for joint communication and
positioning.
98
References
[1]
Harald Haas, Jaafar Elmirghani, and Ian White. Optical wireless communication.
2020.
[2]
In Keun Son and Shiwen Mao. “A survey of free space optical networks”. In:
Digital communications and networks 3.2 (2017), pp. 67–77.
[3]
Nirzhar Saha et al. “Survey on optical camera communications: challenges and
opportunities”. In: Iet Optoelectronics 9.5 (2015), pp. 172–183.
[4]
Murat Uysal and Hatef Nouri. “Optical wireless communications—An emerging
technology”. In: 2014 16th international conference on transparent optical
networks (ICTON). IEEE. 2014, pp. 1–7.
[5]
Alejandro Aragón-Zavala. Indoor wireless communications: From theory to
implementation. John Wiley & Sons, 2017.
[6]
L. Grobe et al. “High-speed Visible Light Communication Systems”. In: IEEE
Communications Magazine 51.12 (Dec. 2013), pp. 60–66. issn: 1558-1896. doi:
10.1109/MCOM.2013.6685758.
[7]
Dominic Schulz et al. “Use cases for optical wireless communication”. In: Optical
Fiber Communication Conference. Optical Society of America. 2018, M1F–3.
[8] S Van Der Gaast. “Report by Stratix and TU/e”. In: no. December (2017).
[9]
Sreelal Maravanchery Mana et al. “Experiments in non-line-of-sight Li-Fi
channels”. In: 2019 Global LIFI Congress (GLC). IEEE. 2019, pp. 1–6.
[10]
Ezio Biglieri et al. MIMO wireless communications. Cambridge university press,
2007.
[11]
Volker Jungnickel. “Optical wireless in 5G”. In: IEEE Summer Topicals, tutorial.
2016.
[12]
Kai Lennert Bober et al. “Distributed Multiuser MIMO for LiFi in Industrial
Wireless Applications”. In: Journal of Lightwave Technology 39.11 (2021),
pp. 3420–3433.
[13]
Mohammad Aazam, Sherali Zeadally, and Khaled A Harras. “Deploying
fog computing in industrial internet of things and industry 4.0”. In: IEEE
Transactions on Industrial Informatics 14.10 (2018), pp. 4674–4682.
[14]
László Lovász. “On the Shannon capacity of a graph”. In: IEEE Transactions
on Information theory 25.1 (1979), pp. 1–7.
99
REFERENCES
[15]
Pablo Wilke Berenguer et al. “Optical wireless MIMO experiments in an
industrial environment”. In: IEEE Journal on Selected Areas in Communications
36.1 (2017), pp. 185–193.
[16]
AM Zin et al. “An overview of radio-over-fiber network technology”. In:
International Conference On Photonics 2010. IEEE. 2010, pp. 1–3.
[17]
Wenjun Gu et al. “Impact of multipath reflections on the performance of indoor
visible light positioning systems”. In: Journal of Lightwave Technology 34.10
(2016), pp. 2578–2587.
[18]
Mohamad Yassin and Elias Rachid. “A survey of positioning techniques and
location based services in wireless networks”. In: 2015 IEEE International
Conference on Signal Processing, Informatics, Communication and Energy
Systems (SPICES). IEEE. 2015, pp. 1–5.
[19]
Davide Dardari, Pau Closas, and Petar M Djurić. “Indoor tracking: Theory,
methods, and technologies”. In: IEEE Transactions on Vehicular Technology
64.4 (2015), pp. 1263–1278.
[20]
Hui Liu et al. “Survey of wireless indoor positioning techniques and systems”.
In: IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications
and Reviews) 37.6 (2007), pp. 1067–1080.
[21]
Chen Chen et al. “High accuracy indoor localization: A WiFi-based approach”. In:
2016 IEEE International Conference on Acoustics, Speech and Signal Processing
(ICASSP). IEEE. 2016, pp. 6245–6249.
[22]
Guowei Shi and Ying Ming. “Survey of indoor positioning systems based on
ultra-wideband (UWB) technology”. In: Wireless Communications, Networking
and Applications. Springer, 2016, pp. 1269–1278.
[23]
High-speed indoor visible light communication transceiver – System architecture,
physical layer and data link layer specification. 88. ITU-T G999.1. 2019. url:
https://www.itu.int/rec/T-REC-G.9991-201903-I/en.
[24]
Andreja Rojko. “Industry 4.0 concept: background and overview”. In: Interna-
tional Journal of Interactive Mobile Technologies (iJIM) 11.5 (2017), pp. 77–90.
[25]
Roberto Sabella et al. “Optical transport for Industry 4.0”. In: Journal of Optical
Communications and Networking 12.8 (2020), pp. 264–276.
[26]
Pablo Wilke Berenguer et al. “Real-time optical wireless mobile communication
with high physical layer reliability”. In: Journal of Lightwave Technology 37.6
(2019), pp. 1638–1646.
[27]
Arthur Lessard and Mario Gerla. “Wireless communications in the automated
factory environment”. In: IEEE Network 2.3 (1988), pp. 64–69.
[28]
Hien Quoc Ngo et al. “Cell-free massive MIMO versus small cells”. In: IEEE
Transactions on Wireless Communications 16.3 (2017), pp. 1834–1850.
100
REFERENCES
[29]
Huaiyu Dai. “Distributed versus co-located MIMO systems with correlated
fading and shadowing”. In: 2006 IEEE International Conference on Acoustics
Speech and Signal Processing Proceedings. Vol. 4. IEEE. 2006, pp. IV–IV.
[30]
Ibrahim Can Sezgin et al. “Evaluation of distributed mimo communication
using a low-complexity sigma-delta-over-fiber testbed”. In: 2019 IEEE MTT-S
International Microwave Symposium (IMS). IEEE. 2019, pp. 754–757.
[31]
Shichuan Ma, Yaoqing Yang, and Hamid Sharif. “Distributed MIMO technologies
in cooperative wireless networks”. In: IEEE Communications Magazine 49.5
(2011), pp. 78–82.
[32]
Sepideh Mohammadi Kouhini et al. “Distributed MIMO Experiment Using
LIFI Over Plastic Optical Fiber”. In: IEEE Global Communications Conference-
Globecom 2020. 2020.
[33]
Pablo Wilke Berenguer et al. “Distributed 8
×
6 MIMO experiments for
optical wireless communications”. In: 2017 European Conference on Optical
Communication (ECOC). IEEE. 2017, pp. 1–3.
[34]
Sreelal Maravanchery Mana et al. “Distributed MIMO experiments for LiFi in a
conference room”. In: 2020 12th International Symposium on Communication
Systems, Networks and Digital Signal Processing (CSNDSP). IEEE. 2020, pp. 1–
5.
[35]
Kyu-Sung Hwang, Young-Chai Ko, and Mohamed-Slim Alouini. “Outage
probability of cooperative diversity systems with opportunistic relaying based
on decode-and-forward”. In: IEEE Transactions on Wireless Communications
7.12 (2008), pp. 5100–5107.
[36]
Sepideh Mohammadi Kouhini et al. “Performance of bidirectional LiFi over
plastic optical fiber (POF)”. In: 2020 12th International Symposium on
Communication Systems, Networks and Digital Signal Processing (CSNDSP).
IEEE. 2020, pp. 1–6.
[37]
Aleksandra Checko et al. “Cloud RAN for mobile networks—A technology
overview”. In: IEEE Communications surveys & tutorials 17.1 (2014), pp. 405–
426.
[38]
Asanka Nuwanpriya, Siu-Wai Ho, and Chung Shue Chen. “Angle diversity
receiver for indoor MIMO visible light communications”. In: 2014 IEEE Globecom
Workshops (GC Wkshps). IEEE. 2014, pp. 444–449.
[39]
Zhe Chen, Nikola Serafimovski, and Harald Haas. “Angle diversity for an indoor
cellular visible light communication system”. In: 2014 IEEE 79th Vehicular
Technology Conference (VTC Spring). IEEE. 2014, pp. 1–5.
[40]
Fritz R Gfeller and Urs Bapst. “Wireless in-house data communication via diffuse
infrared radiation”. In: Proceedings of the IEEE 67.11 (1979), pp. 1474–1486.
[41]
Joseph M Kahn and John R Barry. “Wireless infrared communications”. In:
Proceedings of the IEEE 85.2 (1997), pp. 265–298.
101
REFERENCES
[42]
David Wisely and Ian Neild. “A 100 Mbit/s tracked optical wireless telepoint”.
In: Proceedings of 8th International Symposium on Personal, Indoor and Mobile
Radio Communications-PIMRC’97. Vol. 3. IEEE. 1997, pp. 964–968.
[43]
Syed Hassan Raza Naqvi et al. “On the transport capability of LAN cables in
all-analog MIMO-RoC fronthaul”. In: 2017 IEEE Wireless Communications and
Networking Conference (WCNC). IEEE. 2017, pp. 1–6.
[44]
Diogo Acatauassu et al. “An efficient fronthaul scheme based on coaxial cables for
5g centralized radio access networks”. In: IEEE Transactions on Communications
69.2 (2020), pp. 1343–1357.
[45]
Nathan J. Gomes et al. “Invited Papers”. English. In: Optical Fiber Technology
26.Part A (2015), pp. 50–58.
[46]
Sepideh Mohammadi Kouhini et al. “Use of plastic optical fibers for distributed
mimo in li-fi systems”. In: 2019 Global LIFI Congress (GLC). IEEE. 2019,
pp. 1–5.
[47]
Volker Jungnickel, Volker Pohl, and Clemens Von Helmolt. “Capacity of MIMO
systems with closely spaced antennas”. In: IEEE Communications Letters 7.8
(2003), pp. 361–363.
[48]
Carina Ribeiro Barbio Correa et al. “WDM over POF for D-MIMO LiFi system”.
In: 1st Optical Wireless Communication Conference. 2020.
[49]
Carina Ribeiro Barbio Correa et al. “Luminaire-Free Gigabits per second LiFi
Transmission employing WDM-over-POF”. In: 2020 European Conference on
Optical Communications (ECOC). IEEE. 2020, pp. 1–4.
[50]
Yazan A Alqudah and Mohsen Kavehrad. “MIMO characterization of indoor
wireless optical link using a diffuse-transmission configuration”. In: IEEE
Transactions on Communications 51.9 (2003), pp. 1554–1560.
[51]
Dominic O’Brien. “Multi-input multi-output (MIMO) indoor optical wireless
communications”. In: 2009 Conference Record of the Forty-Third Asilomar
Conference on Signals, Systems and Computers. IEEE. 2009, pp. 1636–1639.
[52]
Raed Mesleh et al. “Indoor MIMO optical wireless communication using spatial
modulation”. In: 2010 IEEE International Conference on Communications.
IEEE. 2010, pp. 1–5.
[53]
V Jungnickel et al. “Optical Wireless Communication for Backhaul and Access”.
In: 2015 European Conference on Optical Communication. IEEE. 2015, pp. 1–3.
[54]
Iman Tavakkolnia et al. “Energy-efficient adaptive MIMO-VLC technique
for indoor LiFi applications”. In: 2018 25th International Conference on
Telecommunications (ICT). IEEE. 2018, pp. 331–335.
[55]
Yuan Zhuang et al. “A survey of positioning systems using visible LED lights”.
In: IEEE Communications Surveys & Tutorials 20.3 (2018), pp. 1963–1988.
102
REFERENCES
[56]
You Li et al. “Self-contained indoor pedestrian navigation using smartphone
sensors and magnetic features”. In: IEEE Sensors Journal 16.19 (2016), pp. 7173–
7182.
[57] Priyanka Aggarwal. MEMS-based integrated navigation. Artech House, 2010.
[58]
Gregary B Prince and Thomas DC Little. “A two phase hybrid RSS/AoA
algorithm for indoor device localization using visible light”. In: 2012 IEEE
Global Communications Conference (GLOBECOM). IEEE. 2012, pp. 3347–3352.
[59]
Adli Hasan et al. “Fingerprint-based visible light positioning using multiple
photodiode receiver”. In: 2020 IEEE Sensors Applications Symposium (SAS).
IEEE. 2020, pp. 1–6.
[60]
Yousef Almadani et al. “Visible Light Communications for Industrial Applica-
tions—Challenges and Potentials”. In: Electronics 9.12 (2020), p. 2157.
[61]
Takayuki Akiyama, Masanori Sugimoto, and Hiromichi Hashizume. “Time-of-
arrival-based smartphone localization using visible light communication”. In:
2017 International Conference on Indoor Positioning and Indoor Navigation
(IPIN). IEEE. 2017, pp. 1–7.
[62]
Ertan Kazikli and Sinan Gezici. “Hybrid TDOA/RSS based localization for
visible light systems”. In: Digital Signal Processing 86 (2019), pp. 19–28.
[63]
Pablo Wilke Berenguer et al. “Optical Wireless MIMO Experiments in an
Industrial Environment”. In: IEEE Journal on Selected Areas in Communications
36.1 (2017), pp. 185–193.
[64]
P. W. Berenguer et al. “Real-Time Optical Wireless Mobile Communication
With High Physical Layer Reliability”. In: Journal of Lightwave Technology 37.6
(Mar. 2019), pp. 1638–1646. issn: 1558-2213. doi:
10.1109/JLT.2019.2893520
.
[65]
S. M. Kouhini et al. “Use of Plastic Optical Fibers for Distributed MIMO
in Li-Fi Systems”. In: 2019 Global LIFI Congress. June 2019, pp. 1–5. doi:
10.1109/GLC.2019.8864130.
[66]
Signify.
http://https://www.signify.com/global/innovation/trulifi
.
Accessed: 2019.
[67]
M. Hinrichs et al. “Analog vs. next-generation digital fronthaul: How to minimize
optical bandwidth utilization”. In: 2017 International Conference on Optical
Network Design and Modeling (ONDM). 2017.
[68]
Olaf Ziemann et al. POF-polymer Optical Fibers for Data Communication.
Springer Science & Business Media, 2002.
[69]
Xiaopeng Yang et al. “Core Technology for Achieving Plastic Optical Fibers in
an All-Optical Network”. In: 2017 16th International Symposium on Distributed
Computing and Applications to Business, Engineering and Science (DCABES).
IEEE. 2017, pp. 127–130.
103
REFERENCES
[70]
Olaf Ziemann, Hans Poisel, and Juri Vinogradov. “Potential of high speed, short
distance optical data communication on large diameter optical fibers”. In: 2006
1st Electronic Systemintegration Technology Conference. Vol. 1. IEEE. 2006,
pp. 409–414.
[71]
Sven Loquai et al. “42-Gb/s transmission over large-core 1-mm PMMA graded-
index polymer optical fiber”. In: IEEE Photonics Technology Letters 25.6 (2013),
pp. 602–605.
[72]
Jan Willem Wallenburg. “Polymer optical fibre: characterization and analog
transmission”. In: Graduation report, The Department of Electrical Engineering
of the Eindhoven University of Technology (1994).
[73]
Evgenii E Narimanov and Partha Mitra. “The channel capacity of a fiber optics
communication system: Perturbation theory”. In: Journal of lightwave technology
20.3 (2002), p. 530.
[74]
S. M. Mana et al. “Experiments in Non-Line-of-Sight Li-Fi Channels”. In: 2019
Global LIFI Congress. June 2019, pp. 1–6. doi:10.1109/GLC.2019.8864131.
[75]
Stephan Jaeckel, Lars Thiele, and Volker Jungnickel. “Interference-limited MIMO
Measurements”. In: 2010 IEEE 71st Vehicular Technology Conference. IEEE.
2010, pp. 1–5.
[76]
Thilo Fath and Harald Haas. “Performance comparison of MIMO techniques for
optical wireless communications in indoor environments”. In: IEEE Transactions
on Communications 61.2 (2012), pp. 733–742.
[77]
John R Barry. Wireless infrared communications. Vol. 280. Springer Science &
Business Media, 1994.
[78]
Pablo Wilke Berenguer et al. “Analog antenna diversity for reliable optical
wireless communication systems”. In: 2018 Global LIFI Congress (GLC). IEEE.
2018, pp. 1–5.
[79]
Asma Mohamed, Abdel Halim Zekry, and Reem Ibrahim. “FPGA Imple-
mentation of Sphere Detector for Spatial Multiplexing MIMO System”. In:
International Journal of Electronics and Telecommunications 65 (2019).
[80]
Yinzhou Tan and Harald Haas. “Coherent LiFi System with Spatial Multiplex-
ing”. In: IEEE Transactions on Communications (2021).
[81]
SM Kouhini et al. “Benefits of MIMO Mode Switching, Angular Diversity and
Multiuser Multiplexing for LiFi”. In: Optical Fiber Communication Conference.
Optical Society of America. 2021, F1A–7.
[82]
Volker Jungnickel et al. “Capacity measurements in a cooperative MIMO net-
work”. In: IEEE transactions on vehicular technology 58.5 (2008), pp. 2392–2405.
[83]
Wael Abu Shehab and Zouhair Al-qudah. “Singular value decomposition:
Principles and applications in multiple input multiple output communication
system”. In: Int. J. Comput. Networks Commun 9.1 (2017), pp. 13–21.
104
REFERENCES
[84]
Ezio Biglieri et al. MIMO Wireless Communications. Cambridge University
Press, 2007. doi:10.1017/CBO9780511618420.
[85]
David Tse and Pramod Viswanath. Fundamentals of Wireless Communication.
Cambridge university press, 2005.
[86]
Gerard J Foschini and Michael J Gans. “On Limits of Wireless Communications
in a Fading Environment when using Multiple Antennas”. In: Wireless Personal
Communications 6.3 (1998), pp. 311–335.
[87]
Shokoufeh Mardanikorani, Xiong Deng, and Jean-Paul MG Linnartz. “Sub-
carrier loading strategies for DCO-OFDM LED communication”. In: IEEE
Transactions on Communications 68.2 (2019), pp. 1101–1117.
[88]
Sreelal M Mana et al. “LIDAR-Assisted Channel Modelling for LiFi in Realistic
Indoor Scenarios”. In: IEEE Access (2022).
[89]
Wave-2 Coax and Phone Line Networking Evaluation. 88. Maxlinear. 2019. url:
https://www.mouser.de/datasheet/2/146/wave-2-coaxial-and-phone-
line-networking-evaluatio-1815047.pdf.
[90]
David Tse and Pramod Viswanath. “MIMO I: spatial multiplexing and channel
modeling”. In: Fundamentals of Wireless Communication (2005), pp. 290–331.
[91]
Sepideh Mohammadi Kouhini et al. “LiFi positioning for industry 4.0”. In: IEEE
Journal of Selected Topics in Quantum Electronics 27.6 (2021), pp. 1–15.
[92]
Ahmed Makki et al. “Survey of WiFi positioning using time-based techniques”.
In: Computer Networks 88 (2015), pp. 218–233.
[93]
Stefan Schwalowsky et al. “System integration of an IEEE 802.11 based TDoA
localization system”. In: 2010 IEEE International Symposium on Precision Clock
Synchronization for Measurement, Control and Communication. IEEE. 2010,
pp. 55–60.
[94]
André Günther and Christian Hoene. “Measuring round trip times to determine
the distance between WLAN nodes”. In: International conference on research in
networking. Springer. 2005, pp. 768–779.
[95]
Volker Jungnickel et al. “Spatial degrees of freedom in small cells: Measurements
with large antenna arrays”. In: 2013 IEEE 77th Vehicular Technology Conference
(VTC Spring). IEEE. 2013, pp. 1–6.
[96]
Marcel Müller et al. “Leverage LiFi in Smart Manufacturing”. In: 2020 IEEE
Global Communications Conference (GLOBECOM 2020). 2020.
[97]
Super-Resolution Time of Arrival Estimation. Interact. 2022. url:
https :
//www.mwrf.com/technologies/systems/article/21214458/mathworks-
algorithms- to- antennas- threedimension- indoor- positioning- using-
ieee-80211az-waveforms.
[98] Indoor Navigation and Location Analytic. Interact. 2020.
[99]
YellowDot Program. Interact. 2020. url:
https://www.interact-lighting.
com/global/indoor-navigation/yellowdot.
105
REFERENCES
[100]
Mohammadreza Aminikashani, Wenjun Gu, and Mohsen Kavehrad. “Indoor
positioning with OFDM visible light communications”. In: 2016 13th IEEE
annual consumer communications & networking conference (CCNC). IEEE.
2016, pp. 505–510.
[101]
Malte Hinrichs et al. “A Physical Layer for Low Power Optical Wireless
Communications”. In: IEEE Transactions on Green Communications and
Networking (2020).
[102]
Ziwei Ye et al. “A linearly attenuated lighting for visible light positioning system
based on RSSI”. In: Optics Communications 432 (2019), pp. 32–38.
[103]
Hongming Yang et al. “Uniform illumination rendering using an array of LEDs:
a signal processing perspective”. In: IEEE transactions on signal processing 57.3
(2008), pp. 1044–1057.
[104]
Mohammed Abd Elkarim, Nazmi Azzam Mohammed, and Moustafa Hussein Aly.
“Exploring the performance of indoor localization systems based on VLC-RSSI,
including the effect of NLOS components using two light-emitting diode lighting
systems”. In: Optical Engineering 54.10 (2015), p. 105110.
[105]
Karel Heurtefeux and Fabrice Valois. “Is rssi a good choice for localization
in wireless sensor network?” In: 2012 IEEE 26th international conference on
advanced information networking and applications. IEEE. 2012, pp. 732–739.
[106]
Huichao Lv et al. “High accuracy VLC indoor positioning system with differential
detection”. In: IEEE Photonics Journal 9.3 (2017), pp. 1–13.
[107]
David Plets et al. “On the impact of LED power uncertainty on the accuracy of
2D and 3D Visible Light Positioning”. In: Optik 195 (2019), p. 163027.
[108]
Huy Q Tran and Cheolkeun Ha. “Fingerprint-based indoor positioning system
using visible light communication—a novel method for multipath reflections”.
In: Electronics 8.1 (2019), p. 63.
[109]
H. Zhang et al. “High-Precision Indoor Visible Light Positioning Using Deep
Neural Network Based on the Bayesian Regularization With Sparse Training
Point”. In: IEEE Photonics Journal 11.3 (2019), pp. 1–10. doi:
10.1109/JPHOT.
2019.2912156.
[110]
Rhowel M Dellosa, Arnel C Fajardo, and Ruji P Medina. “Modified fingerprinting
localization technique of indoor positioning system based on coordinates”. In:
Indonesian Journal of Electrical Engineering and Computer Science 15.3 (2019),
pp. 1345–1355.
[111]
Stefanie Cincotta et al. “Visible light positioning using an aperture and a
quadrant photodiode”. In: 2017 IEEE Globecom Workshops (GC Wkshps). IEEE.
2017, pp. 1–6.
[112]
Trong-Hop Do and Myungsik Yoo. “An in-depth survey of visible light
communication based positioning systems”. In: Sensors 16.5 (2016), p. 678.
106
REFERENCES
[113]
Se-Hoon Yang et al. “Three-dimensional visible light indoor localization using
AOA and RSS with multiple optical receivers”. In: Journal of Lightwave
Technology 32.14 (2014), pp. 2480–2485.
[114]
Ling Wei et al. “Cubic-receiver-based indoor optical wireless location system”.
In: IEEE Photonics Journal 8.1 (2016), pp. 1–7.
[115]
Thomas Q Wang, Cuiwei He, and Jean Armstrong. “Angular diversity for
indoor MIMO optical wireless communications”. In: 2015 IEEE International
Conference on Communications (ICC). IEEE. 2015, pp. 5066–5071.
[116]
Ayesha Naz et al. “PDOA based indoor positioning using visible light communi-
cation”. In: IEEE Access 6 (2018), pp. 7557–7564.
[117]
Hui Chen et al. “A Joint TDOA-PDOA Localization Approach Using Particle
Swarm Optimization”. In: IEEE Wireless Communications Letters (2020).
[118]
S. Jung, S. Hann, and C. Park. “TDOA-based optical wireless indoor localization
using LED ceiling lamps”. In: IEEE Transactions on Consumer Electronics 57.4
(2011), pp. 1592–1597. doi:10.1109/TCE.2011.6131130.
[119]
Pengfei Du et al. “Demonstration of a low-complexity indoor visible light
positioning system using an enhanced TDOA scheme”. In: IEEE Photonics
Journal 10.4 (2018), pp. 1–10.
[120]
Sheng Zhang et al. “Experimental demonstration of indoor sub-decimeter
accuracy VLP system using differential PDOA”. In: IEEE Photonics Technology
Letters 30.19 (2018), pp. 1703–1706.
[121]
F. Zafari, A. Gkelias, and K. K. Leung. “A Survey of Indoor Localization Systems
and Technologies”. In: IEEE Communications Surveys Tutorials 21.3 (2019),
pp. 2568–2599. doi:10.1109/COMST.2019.2911558.
[122]
Trong-Hop Do, Junho Hwang, and Myungsik Yoo. “TDoA based indoor visible
light positioning systems”. In: 2013 Fifth international conference on ubiquitous
and future networks (ICUFN). IEEE. 2013, pp. 456–458.
[123]
Nissanka B Priyantha, Anit Chakraborty, and Hari Balakrishnan. “The cricket
location-support system”. In: Proceedings of the 6th annual international
conference on Mobile computing and networking. 2000, pp. 32–43.
[124]
Adam Smith et al. “Tracking moving devices with the cricket location system”. In:
Proceedings of the 2nd international conference on Mobile systems, applications,
and services. 2004, pp. 190–202.
[125]
Jose Rabadan et al. “Hybrid visible light and ultrasound-based sensor for distance
estimation”. In: Sensors 17.2 (2017), p. 330.
[126]
Jorge Torres-Solis, Tiago H Falk, and Tom Chau. A review of indoor localization
technologies: towards navigational assistance for topographical disorientation.
INTECH Open Access Publisher, 2010.
[127]
Jean Armstrong. “OFDM for optical communications”. In: Journal of lightwave
technology 27.3 (2009), pp. 189–204.
107
REFERENCES
[128]
Stefan Schiffermüller and Volker Jungnickel. “Practical Channel Interpolation
for OFDMA.” In: GLOBECOM. 2006.
[129]
Pablo Rafael Andreas Wilke Berenguer. “Physical layer reliability aspects in
industrial optical wireless communication”. In: (2019).
[130]
Alan V Oppenheim, John R Buck, and Ronald W Schafer. Discrete-time signal
processing. Vol. 2. Upper Saddle River, NJ: Prentice Hall, 2001.
[131]
Timothy M Schmidl and Donald C Cox. “Robust frequency and timing
synchronization for OFDM”. In: IEEE transactions on communications 45.12
(1997), pp. 1613–1621.
[132]
Hlaing Minn, Mao Zeng, and Vijay K Bhargava. “On timing offset estimation
for OFDM systems”. In: IEEE Communications letters 4.7 (2000), pp. 242–244.
[133]
Unified High-Speed Wireline-Based Home. “Networking Transceivers—System
Architecture and Physical Layer Specification”. In: ITU-T Rec. G 9960 (2019).
[134]
Sepideh Mohammadi Kouhini et al. “LiFi based Positioning for Indoor Scenarios”.
In: 2021 17th International Symposium on Wireless Communication Systems
(ISWCS). IEEE. 2021, pp. 1–5.
[135]
88LX5152, 88LX5153 Wave-2 G.hn Digital Baseband (DBB) Processor. DBB
Wave-2. Maxlinear. 2020. url:
https://https://www.maxlinear.com/ds/
88lx5152_88lx5153.pdf.
108
