Measurement and Modelling of A Free-Space Optical Link and In-Field OFDM Experiment

See
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/254628747
Measurement and Modelling of a Free-Space

Optical Link and In-Field OFDM Experiment
Thesis January 2012
CITATIONS
READS
36
1 author:
Ayman Mostafa
University of British Columbia - Vancouver
9 PUBLICATIONS 32 CITATIONS
SEE PROFILE
Available from: Ayman Mostafa

Retrieved on: 02 September 2016
MEASUREMENT AND MODELLING OF A

FREE-SPACE OPTICAL LINK AND IN-FIELD
OFDM EXPERIMENT
MEASUREMENT AND MODELLING OF A FREE-SPACE

OPTICAL LINK AND IN-FIELD OFDM EXPERIMENT
BY
AYMAN MOSTAFA, B.Sc.
a thesis
submitted to the department of electrical & computer engineering
and the school of graduate studies
of mcmaster university
in partial fulfilment of the requirements
for the degree of
Master of Applied Science
c Copyright by Ayman Mostafa, January 2012

All Rights Reserved
Master of Applied Science (2012)
McMaster University
(Electrical & Computer Engineering)
TITLE:
Hamilton, Ontario, Canada
MEASUREMENT AND MODELLING OF A FREESPACE OPTICAL LINK AND IN-FIELD OFDM EXPERIMENT
AUTHOR:
Ayman Mostafa
B.Sc., (Electrical Engineering)
Alexandria University, Alexandria, Egypt
SUPERVISOR:
Dr. Steve Hranilovic
NUMBER OF PAGES:
xv, 141
ii
To my parents
My wife
My son
Abstract
Free-space optical (FSO) communication is a potential technology for last-mile applications. Key advantages are the unlicensed spectrum, high transmission rates, and
inherent security. Moreover, Radio-over-FSO (RoFSO) allows seamless integration
between the incompatible radio frequency (RF) and optical networks. Such advantages qualify FSO systems to take a front seat in next-generation broadband communication networks. However, the main challenge for FSO systems is the performance
degradation imposed by the atmospheric attenuation and turbulence. To exploit the
advantages of FSO systems, accurate and computationally-efficient channel models
are required.
This thesis represents in-field experimental work related to FSO channel measurement as well as the transmission of orthogonal frequency division multiplexing
(OFDM) over the FSO channel.
A 1.87-km FSO link installed at McMaster University is employed. A high-speed
field-programmable gate array (FPGA)-based digitizer board is used as the underlying hardware platform for interface with the link. A system-on-three-FPGAs is
implemented to act as a universal transceiver for signals composed using MATLAB.
A new technique is developed for conducting the FSO channel measurement. An
optical signal intensity-modulated by a high frequency sinusoid is transmitted. The
iv
received signal undergoes a fast-Fourier transform (FFT) to filter out a large portion
of the interfering noise providing more accurate measurements. Fitting with the lognormal distribution is investigated. A finite-state Markov model is also derived and
its accuracy is verified by the simulation results.
The first realization of an in-field OFDM over FSO transmission system is implemented and tested over the link. The received signal is investigated on the symbol
level and constellation diagrams are visualized. Transmission rates up to 300 Mbps
are achieved with average symbol-error rate (SER) on the order of 106 .
Acknowledgements
This work would not have been possible without the support and contributions of
many individuals.
First of all, I would like to express my gratitude to my thesis supervisor, Dr. Steve
Hranilovic, whose invaluable support, guidance, motivation, logical discussions, and
innovative ideas were key factors to accomplish my work.
I would also like to express my gratitude to Dr. Mohamed Bakr for his guidance
both on academic and non-academic matters throughout the course of my program.
I am indebted to Dr. Ahmed Farid for his continuous help and support since my
first days in Canada.
I would like to thank my lab colleagues, Mohamed El-Shimy, Kasra Asadzadeh,
and Dr. Majid Safari, for many enjoyable conversations we had together.
I offer special thanks to my lab colleague, Danny Vacar, for his valuable contributions in the software issues.
Among the technical staff in the ECE department at McMaster University, Terry
Greenlay deserves a special mention. Without his help, this work would not have
been possible in the given time frame. I would also like to thank Tyler Ackland and
Dan Anthony Manolescu who offered help in practical aspects of the experimental
setup.
vi
I would also like to thank the administrative staff, especially Cheryl Gies, who
facilitated the process entailed in developing this thesis.
I am deeply grateful to my wife for her patience, understanding, and continuous
support especially for taking care of our son almost alone while I was away for many
nights.
Finally, I would like to express my heartily gratitude to my parents for their
endless support and love.
vii
Contents
Abstract
iv
Acknowledgements
vi
1 Introduction
1.1
Wireless Optical Communications . . . . . . . . . . . . . . . . . . . .
1.2
Intensity-Modulation Direct-Detection FSO Channel . . . . . . . . .
1.3
Optical Transmission of Radio Signals
. . . . . . . . . . . . . . . . .
1.4
Thesis Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
1.5
Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
2 Experimental Details
17
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
2.2
Objectives and Requirements . . . . . . . . . . . . . . . . . . . . . .
17
2.3
FSO Link . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
2.3.1
Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22
2.3.2
Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
2.4
FPGA-Based Digitizer Board . . . . . . . . . . . . . . . . . . . . . .
26
2.5
Backbone Firmware . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
viii
2.6
Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
2.7
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
3 FSO Link Measurement and Modelling
37
3.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
3.2
Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
3.2.1
Geometrical Loss . . . . . . . . . . . . . . . . . . . . . . . . .
38
3.2.2
Atmospheric Loss and Turbulence . . . . . . . . . . . . . . . .
39
3.2.3
The Log-Normal Distribution . . . . . . . . . . . . . . . . . .
41
Transceiver Measurement and Modelling . . . . . . . . . . . . . . . .
45
3.3.1
Harmonic Distortion and Nonlinearity . . . . . . . . . . . . .
47
3.3.2
Frequency Response . . . . . . . . . . . . . . . . . . . . . . .
50
3.3.3
Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
3.3.4
Power Loss Estimate . . . . . . . . . . . . . . . . . . . . . . .
56
Channel Measurements . . . . . . . . . . . . . . . . . . . . . . . . . .
57
3.4.1
Measurement Procedure . . . . . . . . . . . . . . . . . . . . .
57
3.4.2
Irradiance Fluctuations . . . . . . . . . . . . . . . . . . . . . .
61
3.4.3
Log-Normal Fitting . . . . . . . . . . . . . . . . . . . . . . . .
68
3.5
Markov Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
3.6
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
82
3.3
3.4
4 In-Field Demonstration of OFDM over FSO
83
4.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
83
4.2
OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
84
4.2.1
84
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ix
4.3
4.4
4.2.2
Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . .
87
4.2.3
Peak-Average Power Ratio . . . . . . . . . . . . . . . . . . . .
88
4.2.4
Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . .
89
4.2.5
OFDM for FSO Systems . . . . . . . . . . . . . . . . . . . . .
93
OFDM over FSO Field Experiments . . . . . . . . . . . . . . . . . .
95
4.3.1
DC-Biased Unclipped OFDM over FSO . . . . . . . . . . . . .
95
4.3.2
DC-Biased Clipped OFDM over FSO . . . . . . . . . . . . . . 103
4.3.3
DVT-B over FSO . . . . . . . . . . . . . . . . . . . . . . . . . 108
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5 Conclusions and Future Directions
113
5.1
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.2
Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
A FSO Link Management Software
117
B Finite-State Markov Model Parameters for K = 64
122
List of Figures
1.1
Block diagram of an intensity-modulation direct-detection FSO channel.
1.2
Radio-over-fiber system. . . . . . . . . . . . . . . . . . . . . . . . . .
1.3
Radio-over-fiber connectivity between micro-cells in cellular networks.
10
1.4
RoFSO along with RoF for back-haul connectivity. . . . . . . . . . .
11
2.1
c
Aerial view of the FSO link location (satellite image 2011
DigitalGlobe Inc. [70]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
2.2
Site views of the experimental FSO link. . . . . . . . . . . . . . . . .
20
2.3
Customized SONAbeam TM -1250-M optical head and PCA box. . . . .
22
2.4
Equivalent circuit of the analog laser driver (modified based on [83]).
24
2.5
Equivalent circuit of the driver for Lasers (2) and (3). . . . . . . . . .
25
2.6
Receiver equivalent circuit (modified based on [83]). . . . . . . . . . .
26
2.7
Architecture of the Triton TM -V5-VXS digitizer board (modified based

on [84]). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8
2.9
28
Block diagram of the backbone firmware and hardware used as a universal analog transceiver. . . . . . . . . . . . . . . . . . . . . . . . . .
31
Block diagram of the experimental setup. . . . . . . . . . . . . . . . .
34
2.10 Experimental setup at McMaster Innovation Park: (a) outdoors, (b)

indoors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xi
35
3.1
Proposed channel model for the FSO link. . . . . . . . . . . . . . . .
46
3.2
Simplified channel model used for harmonic distortion measurements.
47
3.3
Normalized PSD at the receiver used for harmonic distortion measurements at four frequencies: (a) 100 MHz, (b) 200 MHz, (c) 300 MHz,
(d) 400 MHz. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
3.4
Transmitted pulse train used for frequency response measurements. .
51
3.5
Normalized magnitude frequency response of the FSO link obtained by

transmitting the pulse train shown in Fig. 3.4. . . . . . . . . . . . . .
3.6
52
Histogram of thermal plus background noise samples at the receiver

along with Gaussian fit. . . . . . . . . . . . . . . . . . . . . . . . . .
55
3.7
Block diagram of the measurement system at the receiver. . . . . . .
59
3.8
Normalized PSD of the received signal used for channel measurements.
60
3.9
Received irradiance fluctuations through different time scales: (a) 1

msec, (b) 10 msec, (c) 100 msec, (d) 1 sec (samples measured on October 30th , 2011 at 11:32 PM, temperature: 7 C, wind speed: 7.2
km/hr). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
62
3.10 Irradiance fluctuations statistics: (a) normalized autocovariance, (b)

normalized power spectral density (samples measured on October 30th ,
2011 at 11:32 PM, temperature: 7 C, wind speed: 7.2 km/hr). . . . .
64
3.11 Irradiance fluctuations during a 14-hour duration: (a) average intensity, (b) scintillation index (samples measured on October 31st , 2011
starting from 12:00 AM). . . . . . . . . . . . . . . . . . . . . . . . . .
66
3.12 Weather condition corresponding to the irradiance fluctuations in Fig.

3.11: (a) temperature, (b) wind speed. . . . . . . . . . . . . . . . . .
xii
67
3.13 Example (1) of channel envelope histogram with log-normal fit in clear
weather conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
71
72
73
3.17 Example of channel envelope histogram with log-normal fit during recovery from heavy fog condition. . . . . . . . . . . . . . . . . . . . . .
74
3.18 Example of channel envelope histogram with log-normal fit during light
rain condition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
3.19 State diagram of the Markov model for the channel envelope when
K = 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
3.20 The histograms of the channel realizations obtained from the measurements and the realizations generated by the Markov model (samples
measured on November 1st , 2011 at 07:41 AM). . . . . . . . . . . . .
80
3.21 Normalized autocovariance of the channel realizations obtained from

the measurements and generated by the Markov model. . . . . . . . .
81
4.1
Baseband multi-carrier communications system. . . . . . . . . . . . .
84
4.2
Statistics of the amplitude of a typical OFDM symbol with 40000 sub-
4.3
carriers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
OFDM symbol timing recovery by the means of cross-correlation. . .
92
xiii
4.4
Typical cross-correlation between the transmitted waveform of an OFDM

symbol and the received version. . . . . . . . . . . . . . . . . . . . . .
4.5
93
Spectrum of the 20 OFDM channels transmitted in the DC-biased

unclipped OFDM over FSO experiment. . . . . . . . . . . . . . . . .
97
4.6
Block diagram of the implemented OFDM over FSO transmitter. . .
98
4.7
DC-biased unclipped OFDM over FSO: (a) transmitted signal, (b) PSD
of the transmitted signal, (c) received signal, (d) PSD of the received
signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
99
4.8
Block diagram of the implemented OFDM over FSO receiver.
. . . . 100
4.9
Received constellations for DC-biased unclipped OFDM over FSO. . . 101
4.10 Symbol-error rates for DC-biased unclipped OFDM over FSO measured over 9.66 sec transmission time under clear weather conditions:
(a) per subcarrier, (b) per channel. . . . . . . . . . . . . . . . . . . . 102
4.11 DC-biased clipped OFDM over FSO: (a) transmitted signal, (b) PSD
of the transmitted signal, (c) received signal, (d) PSD of the received
signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.12 Received constellations for DC-biased clipped OFDM over FSO. . . . 106
4.13 Symbol-error rates for DC-biased clipped OFDM over FSO measured
over 9.66 sec transmission time under clear weather conditions: (a) per
subcarrier, (b) per channel.
. . . . . . . . . . . . . . . . . . . . . . . 107
4.14 Symbol-error rates for 20 DVB-T channels transmitted over the FSO
link for 9.66 sec during a light-rain condition: (a) per subcarrier, (b)
per channel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
xiv
4.15 Received constellations for higher-order QAMs under clear weather

conditions: (a) 16-QAM, (b) 64-QAM. . . . . . . . . . . . . . . . . . 111
A.1 Snapshot (1) from the STC software: terminal status. . . . . . . . . . 118
A.2 Snapshot (2) from the STC software: lasers power levels and receiver
gain level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
A.3 Snapshot (3) from the STC software: received power. . . . . . . . . . 120
A.4 Snapshot (4) from the STC software: diagnostics. . . . . . . . . . . . 121
xv
Chapter 1
Introduction
1.1
Wireless Optical Communications
Wireless optical communications refer to the transmission of an electrical informationbearing signal by the means of modulating a light source through free space or the
atmosphere [1, 2]. Typical light sources used are light-emitting diodes (LEDs) and
laser diodes. At the receiver, the information signal is recovered by the means of
opto-electrical conversion using a photodetector. Applications for wireless optical
technology include visible light communications (VLC) [3], indoor infrared wireless
communications [4], terrestrial free-space optical (FSO) communications [58], wireless ultra-violet communications [9], ground-to-air links [10], ground-to-satellite links
[11], and inter-satellite links [12].
In next-generation communication networks, both the network size and bandwidth
requirements will continue to expand mainly due to the ever-increasing rates of multimedia (audio/video) interchange. Users expect to be able to communicate at anytime
and anywhere with high data rates, in what is called ubiquitous connectivity. Network
1
M.A.Sc. Thesis - Ayman Mostafa

Chapter 1. Introduction
McMaster University - Electrical Engineering
providers must find means to meet these increasing demands at an affordable cost.
Therefore, a flexible and cost-effective communications platform is required.
Optical fiber and radio frequency (RF) networks are the two primary technologies
used for modern broadband communications. Optical fibers benefit from virtuallyunlimited spectrum and low attenuation levels to achieve high transmission rates (10
Tbps and above) for long distances (several hundreds of kilometers) [13, 14]. These
advantages make optical fibers attractive for backbone connectivity in large-size networks connecting metropolitan area networks (MANs), wide area networks (WANs),
and worldwide networks. A key limitation, however, is that they suffer from inherent
mobility and flexibility problems by being a wired-communication technology, added
to their high infrastructure costs and large deployment times.
On the other hand, RF is the key technology for radio and television (TV) broadcast, cellular phones, and wireless local area networks (WLANs). Although RF directly addresses the mobility and flexibility problems, it is unable to sustain the
transmission rates and distances of optical fiber networks due to inherent limitations on the available spectrum, limitations on the switching speeds of the electronic
devices at the microwave ranges, and high losses in the wireless channel. Current
state-of-the-art data rates for RF networks are capped below 1 Gbps. For example,
the WLAN IEEE 802.11n-2009 amendment supports data rates up to 600 Mbps [15,
p. 247]. In 2010, Samsung demonstrated a WiMAX- 2 mobile transmission based on
the IEEE 802.16m [16] with a data rate of 330 Mbps [17].
The incompatibly in signalling format as well as the mismatch in bandwidth and
transmission rates between the optical and RF channels give rise to the last-mile
bottleneck problem [18]. Last-mile delivery by the means of RF signals instead of

optical fibers limits throughput regardless of the high bandwidth available by the
feeder fiber. In the United States, it was estimated that 95% of the major buildings
in metropolitan areas are only 1.5 km far from a fiber network but do not have access
due to prohibitively high digging and installation costs (US $100,000 - $200,000) per
kilometer [19].
Free-space optical communication links can provide a potential solution to the
last-mile problem in many scenarios, taking advantage of both the flexibility of RF
networks and the high data rates of optical links. Key advantages of FSO links are
their use of unlicensed spectrum, rapid deployment, flexibility, ease of relocation,
immunity to RF and neighboring FSO network interference, inherent security, and
high power efficiency due to narrow laser beamwidths.
On the other hand, FSO communications face many challenges that affect their
reliability, transmission rate, and range. Weather conditions, such as fog, rain, snow,
or dust, impose attenuation and distortion on the propagating optical waves. Even
under clear sky conditions, the non-homogeneity in the atmospheric structure, caused
by spatial and temporal temperature gradients, causes random variations in the refractive index at optical wavelengths. Such non-homogeneity induces fluctuations of
the signal intensity at the receiver termed as scintillation [20]. In addition, FSO links
employ lasers and transmitting assemblies with very small divergence angles making
the link sensitive to misalignment, i.e. pointing, errors that may result from wind
load or buildings sway and vibration [21]. Furthermore, eye-safety regulations impose
limitations on the amount of power transmitted from the lasers [2226].
The basic approach to mitigate the atmospheric loss is to increase the transmitted
power up to the permitted levels specified by eye-safety and device limitations. In

heavy rain or foggy weather conditions, the attenuation can be severely high resulting
in an irretrievable loss of the optical signal and an outage of the link. In such situations, a hybrid RF-FSO system that adds a backup RF connection can recover the
link availability [27]. Atmospheric turbulence is traditionally mitigated by increasing the receiver aperture size where the aperture-averaging effect will significantly
alleviate the scintillation [28]. More complicated techniques employ tracking systems [29], spatial diversity using multiple-aperture systems [30], maximum-likelihood
sequence detection (MLSD) [7], forward error correction codes [3134], and relay
systems [35, 36].
1.2
Intensity-Modulation Direct-Detection FSO Channel
Intensity-modulation direct-detection (IM/DD) is the simplest optical transmission

technique where the output intensity of the laser diode is directly modulated by the
data signal. A block diagram of an IM/DD FSO communication channel is shown
in Fig. 1.1. The data signal s(t) is in electrical domain. It can be an analog or
digital waveform depending on the source. Examples of analog sources are baseband
analog audio or video signals and RF signals. Examples of digital waveforms are onoff-keying (OOK), M -ary pulse-amplitude modulation (PAM), M -ary pulse-position
modulation (PPM), or any baseband discrete-level signalling scheme.
The electronic circuits at the transmitter convert the input signal s(t) into a
current signal x(t) suitable for modulating the laser diode. Signal conditioning may

x(t)
Transmitter
electronics
s(t)
Optical channel fading

h(t)
Laser
diode
Received optical
power I(t)
Data source
(analog or digital)
y(t)
Photodetector
Transimpedance
amplifier
r(t)
Data receiver
Figure 1.1: Block diagram of an intensity-modulation direct-detection FSO channel.

include amplification, attenuation, voltage-to-current conversion, DC shifting, or clipping. The laser diode ideally does linear mapping between the instantaneous electrical input current x(t) and the instantaneous optical output power, i.e. intensitymodulation. The optical wave propagating through the atmosphere experiences fading
represented by h(t).
At the receiver, the photodetector converts the incident optical power I(t) into
current y(t), i.e. direct-detection. The photodetector is typically followed by a transimpedance amplifier that yields a voltage signal r(t) suitable for detection by the
data receiver. A mathematical model that describes the IM/DD FSO communication
system is given by [6]
r(t) = G R h(t) x(t) + w(t),
(1.1)
where G is the transimpedance amplifier gain (V/A), R is the photodetector responsivity (A/W), h(t) is the channel gain representing the fluctuations of the received
optical power, and w(t) is the noise at the receiver usually modelled as additive white
Gaussian noise (AWGN). Without loss of generality, the electro-optical conversion factor of the laser diode at the transmitter has been assumed unity. Further discussion
about the FSO channel model along with an experimental verification is presented in

Sec. 3.3. Using IM/DD systems, as opposed to coherent systems, is preferred because
of design simplicity and the availability of inexpensive components.
The response of the atmospheric channel is unpredictable and time-varying. To
characterize the performance of FSO communication systems, a statistical model
of the FSO channel is required. Useful channel models should be mathematically
tractable and show a good fit with the channel measurements.
Early investigations of the intensity fluctuations of light waves were connected
with astronomy and the optical scintillation of stars. The history of the pioneering
experimental work for measuring the intensity fluctuations and scintillation of optical
waves propagating horizontally in the atmosphere near the earth surface goes back to
1950s and 1960s [37, 38]. Then, since 1970s, many experiments have been conducted
to characterize the FSO channel. Examples of recent experimental work are reported
in [28, 3945]. In [28], a 1.5-km link operating at 1550 nm is used. The theory of
optical scintillation [46] is applied to develop a theoretical probability distribution
function (PDF) from the channel measurements. The fitting parameters with the
log-normal and gamma-gamma distributions are compared between the simulated
and experimental data. Moreover, the aperture-averaging effect on scintillation is
investigated using various receiver aperture sizes. A 1550-nm, 12-km FSO system is
reported in [43]. The link is employed to fit between the log-normal, gamma-gamma,
and exponential distributions and the channel measurements under weak, moderate,
and strong turbulence, respectively. In [45], a 1-km FSO link operating at 1550 nm
with transmission rates up to 1.5 Gbps is used to investigate the effect of fog and
rain on the channel attenuation. A semi-empirical model is derived to relate the
atmospheric attenuation with visibility and rain levels.

An unmodulated optical wave emitted from the laser at the transmitter side is
usually the typical scenario used in FSO channel measurements as in [28, 43]. Key
advantages of using a unmodulated waves are the simplicity of design and the ability
to use any sampling rate at the receiver based on the required accuracy. However, it is
not always possible to use unmodulated waves as in the case of McMaster University
FSO link where the low cutoff modulation frequency of the band-pass filters at the
transmitter and receiver is around 5 MHz. Using waveforms, such as sinusoids or
square waves, that modulate the transmitted optical wave at frequencies higher than
the cutoff frequency is the obvious solution, however, at the cost of added complexity
and increased sampling rates at the receiver. In [40], pulses with a frequency of 6
MHz are transmitted for channel measurements over a 2.7-km FSO link. The received
samples are bandpass filtered, rectified, and averaged to represent the received optical
intensity.
In any communication system, the received signal is corrupted by noise which sets
a fundamental limit for the reliable transmission range or data rate. Noise comes
form various sources for different communication systems, however, it can be broadly
classified into signal-independent and signal-dependent noise. For FSO systems, examples of signal-independent (additive) noise are the background noise, photodetector
dark current, and thermal noise. Sources of signal-dependent noise are laser intensity
noise, laser phase noise, and photodetector shot noise. In practice, the dominant noise
sources are the background radiation and thermal noise. Background radiation comes
mainly from the sun and black-body radiation and can be reduced using appropriate
spatial and solar filters. Thermal noise is inherent in any electronic system operating
above zero absolute temperature.

To obtain accurate results from the channel measurements, the effect of noise
should be alleviated or, alternatively, taken into account by using suitable mathematical models. In [28], an unmodulated optical signal is transmitted and the measured
samples are corrected by subtracting the mean value of the estimated noise from
all the samples. A more accurate mathematical approach is used in [43]. Noise is
included in the expression of the channel distribution
r = h + w,
(1.2)
where r is the received noisy sample, h is the channel fading realization, and w is
the AWGN with mean w and variance w2 . Since fading and noise are assumed
independent, the sum distribution is given by
pr (r) = ph (h) pw (w),
(1.3)
where px (x) is the PDF of the random variable x and denotes convolution. Parameter estimation techniques are applied to Eq. (1.3) where the noise parameters w
and w2 are assumed to be known.
In this thesis, a different measurement procedure is presented in Chapter 3. Improved noise immunity of the measured samples is obtained by transmitting an optical
wave modulated by a sinusoid and applying fast Fourier transform (FFT) at the receiver to act as a notch filter that minimizes the noise corruption.

Fiber
channel
RF source
E/O
conversion
RF receiver
O/E
conversion
Figure 1.2: Radio-over-fiber system.
1.3
Optical Transmission of Radio Signals
The transmission of RF signals over optical fibers, termed Radio-over-Fiber (RoF),

provides an effective solution for the RF-optical interface problem [4752]. As shown
in Fig. 1.2, the concept is simple. A linear electro-optical conversion of the RF signal
maps it into an optical signal suitable for transmission over the fiber channel. At
the receiver, a reverse mapping is performed using a photodetector to recover the RF
signals in the original radio format. Cellular phone and Digital Video Broadcasting
(DVB) networks are immediate applications of the RoF technology.
The concept of dividing the cellular network coverage area into smaller size cells,
termed micro-cells and pico-cells, connected by the fiber network has been proposed
to increase both the capacity and coverage [51, 52]. Such a technique mitigates the
limited frequency resources by increasing the frequency reuse resulting in more efficient utilization as illustrated in Fig. 1.3. In this scenario, RoF is an ideal candidate
for the feeder network since the large number of small-size cells need a large number of base stations and extensive feeder network that is not feasible by electrical
transmission due to the high losses and limited bandwidth.
By performing electro-optical conversion at the centralized office, the RF signals
can be transmitted on the fiber network to far destinations with very low attenuation

f1
f4
f1
f4
f2
O/E
f3
Base station
Base station
f2
O/E
Fiber
f3
Fiber
f1
f4
Central station
E/O
f2
Fiber network
O/E
f3
Base station
Fiber
Figure 1.3: Radio-over-fiber connectivity between micro-cells in cellular networks.

and without costly repeaters or RF interference. Moreover, using RoF allows the
centralization of the key RF signal processing functions (modulation/demodulation,
multiplexing/demultiplexing) at the central office instead of the base stations which
enables equipment sharing, dynamic allocation of resources, and simplified system
operation and maintenance [47, 50]. Another key advantage is simplifying the design
of the base stations into what is called remote antenna units. Each unit employs
only opto-electrical conversion and amplification to feed the antenna [50]. In microcellular systems, the line-of-sight propagation path is dominant and multi-path effects
are minimized resulting in higher signal qualities. In addition, lower RF power profiles
are needed from both the antenna units and mobile units (cell-phones) resulting in a
reduction in environmental and health effects and an increase in the battery life [47].
The low attenuation and large bandwidth offered by RoF allow sharing of the fiber
network between multiple services, such as cellular, landline, Internet, and DVB, and
10

O/E
Fiber
Landline network
Cellular network
E/O
Fiber network
Cable TV
FSO
link
FSO
link
River, mountain, or
highway
O/E
Internet
Fiber
O/E
Figure 1.4: RoFSO along with RoF for back-haul connectivity.

even between multiple operators resulting in large savings in cost and higher usage
efficiency.
The large installation, operation, and maintenance costs of RoF networks are
justifiable only in large population areas and big cities. However, these costs are not
feasible in rural and low-population areas, behind mountains, or across rivers and
highways. These problems can be solved using the wireless counterpart of the RoF
technology, namely Radio-over-Free-Space-Optics (RoFSO) as shown in Fig. 1.4. A
RoFSO link has all the advantages and drawbacks of the FSO channel. Ideally, it
should allow seamless integration between the RoF feeder network and destinations
where the optical fiber is not available.
Some experimental work related to RoFSO is reported in [29, 5359]. In [54],
11

cellular signals (GSM and CDMA) are transmitted over a 500-m FSO link. Lasers
at wavelengths 850 and 1550 nm are used and performance is evaluated in terms of
optical loss, noise figure, and QPSK constellation curves. In [29, 55, 56], a dense
wavelength division multiplexing (D-WDM) system is developed to transmit four
radio services (3G-WCDMA, ISDB-T, WLAN 802.11a and 802.11g) simultaneously
over the link using four wavelengths around 1550 nm. The system is tested indoors
over a 3-meters range and outdoors over a 1-km distance where an automatic beam
tracking system is used for fine tuning to combat scintillation. Spectrum mask tests
for the transmitted RF services are introduced as the performance metric. In [5759],
a 1-km D-WDM RoFSO link with automatic tracking system is reported. Four RF
services are transmitted simultaneously where adjacent channel leakage ratio (ACLR)
and carrier-to-noise ratio (CNR) of the received channels are mainly considered.
Orthogonal frequency division multiplexing (OFDM) is a very well-established
modulation scheme for modern broadband communication systems. It is currently
applied to digital audio broadcasting (DAB), DVB, digital subscriber line (DSL)
systems, and WLANs [60, 61]. Therefore, the study of OFDM over FSO systems
is appropriate. Many analytical and simulation studies exist for OFDM over FSO
[31, 6264], however, to the best knowledge of the author, no experimental work in
this area has been reported. In this thesis, an experiment is introduced in Chapter 4
as the first realization of OFDM over FSO transmission.
12

1.4
Thesis Contributions
This thesis is the first work to interact with the FSO link that has been deployed
across McMaster University campus in August 2010. It presents the experimental
work that has been conducted for FSO channel measurements and modelling as well
as for OFDM over FSO transmission.
By employing a field-programmable gate array (FPGA)-based digitizer board, a
digital system has been implemented to control the analog transceiver of the FSO
link. The waveform of the transmitted signal is composed on a personal computer
(PC) using numerical computation software environments such as MATLAB [65].
The waveform samples are uploaded to a DDR3 SDRAM on the digitizer board for
transmission by a 2-GSa/s digital-to-analog converter (DAC) that feeds the analog
laser driver. On the receiver side, the electrical signal obtained from the analog
receiver is sampled by a high-speed analog-to-digital converter (ADC) at rates up to
2 GSa/s and buffered on the on-board memory. Buffering sizes up to 3 GBytes at
a time are allowed. Then, the buffered samples are downloaded to a PC for further
processing. Such a buffering system provides seamless integration between the FSO
transceiver and MATLAB which facilitates testing various modulation and signalling
schemes over the FSO link. Although the developed firmware system is simple in
concept, there are many challenges during the design mainly because of the high
sampling and data rates. The most difficult task relates to the design of a finite-state
machine (FSM) that synchronizes and controls the data movement between the three
FPGAs available on the board. In this thesis, the developed system serves as the
backbone hardware for the FSO channel measurements and the OFDM over FSO
experiments.
13

A new measurement system is proposed and implemented for conducting the FSO
channel measurements [66]. The system offers more accurate results than previous
work by improving the noise immunity and increasing the signal-to-noise ratio (SNR)
of the channel samples. By making use of the high sampling rates and large signal
processing capacity provided by the digitizer board, the channel measurements are
conducted using a high-frequency sinusoid instead of the conventional unmodulated
optical wave. The received signal undergoes a notch filter by the use of several FFT
blocks. Results obtained from the channel measurements show a good fit with the
log-normal distribution. Estimated coherence time justifies a slow-fading channel.
Obtained scintillation index values show a weak-turbulence condition. A computationally efficient Markov chain model, which is uniquely defined by a sparse matrix
and a vector, is developed and simulated. The comparison between the simulation
results and channel measurements shows a good fit for the channel distribution as well
as for the autocorrelation justifying the model strength. Such a simple model can be
used for accurately generating different channel realizations for simulation purposes.
A novel in-field OFDM over FSO experiment is conducted to demonstrate the
potential role of FSO channels for OFDM transmission [67]. It is the first realization
of an in-field OFDM over FSO system that supports 300 Mbps transmission rate
over a 1.87-km range. Moreover, the approach used is unique in that it provides
the ability to analyze the received waveforms on the symbol level after demodulation
where constellation diagrams and error rates are obtained.
14

1.5
Thesis Structure
The main focus of this thesis is the experimental work conducted to model the FSO
channel as well as the experiments demonstrating the transmission of OFDM signals
over FSO channels.
In Chapter 2, the experimental details are provided. The technical specifications
of the FSO link are presented including the geographical location, lasers and photodetector specifications, dual-mode operation, signal and power levels, supported
transmission rates, and the management software. The FPGA-based digitizer board
is introduced highlighting the architecture and main features in terms of the computational capabilities, sampling rates, and memory size. The developed data-buffering
system is explained focusing on its ability to interface between MATLAB and the
FSO transceiver. Finally the entire experimental setup is outlined.
In Chapter 3, the results of the channel measurements under different weather
conditions are presented. A review of the atmospheric attenuation and turbulence
as well as the log-normal distribution for modelling the statistical behavior of the
scintillation is presented. A general channel model for the FSO link is proposed and
experimentally verified. The proposed channel measurements technique is developed
and the hardware implementation is outlined. The channel behaviour is investigated
through different time scales and an estimate of the coherence time is obtained. Fitting between the channel histograms obtained from the measurements and the lognormal distribution is justified under clear and foggy weather conditions while it fails
during rain. Finally, a finite-state Markov chain model for the channel is derived and
simulated.
15

In Chapter 4, the in-field experiments of OFDM transmission over the FSO channel are presented. A general review of the advantages and drawbacks of OFDM
modulation are introduced with focus on the practical issues of synchronization and
peak-to-average power ratio. The limitations of the experimental setup for successful OFDM transmission are discussed along with the techniques used to mitigate
their effects. Three in-field OFDM over FSO experiments are explained in detail and
the performance is evaluated in terms of error rates under clear and rainy weather
conditions.
Finally, Chapter 5 presents concluding remarks and directions for future work.
16
Chapter 2
Experimental Details
2.1
Introduction
In this chapter, a comprehensive description of the deployed FSO link at McMaster

University is presented. The accompanying FPGA-based digitizer board used as the
hardware platform along with the implemented backbone firmware are described.
Finally, the entire experimental setup is specified to provide a clear understanding
and better interpretation of the results obtained in the subsequent chapters.
2.2
Objectives and Requirements
The experimental setup has two long term objectives. The first is to study FSO
channel behaviour by conducting channel measurements under a variety of weather
conditions and estimating the channel statistics from the measured samples as shown
in Chapter 3. The second objective is evaluating the performance of FSO communication systems by physical implementation and experimental investigation over the
17

Chapter 2. Experimental Details
link. In this thesis, OFDM over FSO is considered in Chapter 4.

For long term in-field experimental work, a permanent outdoor setup is required
and, therefore, durability is a key factor for selecting the proper system. An FSO
link installed outdoors in a typical Canadian environment should survive in extreme
weather conditions throughout the year. It should cope with severe lightning, intense
cold, snow accumulation, sleet rain, blazing sun, wind, and wide temperature ranges.
Therefore, durability imposes the use of a commercial FSO link. A SONAbeam TM 1250-M provided by fSONA [68] was selected.
To achieve the above-mentioned research goals, the FSO link should support both
analog (RF) and digital waveforms transmission. Moreover, the receiver should support analog reception by providing a linear replica of the photodetector output current
without any non-linear processing. Analog reception is a key requirement for channel
measurements as well as for RoFSO systems. Therefore, the SONAbeam TM -1250-M
link was jointly customized with the manufacturer to add the analog transmission and
reception functionalities in addition to the standard digital operation in the optical
domain provided by the commercial product. More details about the customized link
are given in Sec. 2.3.
For proper selection of the link location, several considerations should be taken
into account. First of all, a clear line-of-sight between the two terminals is mandatory. Power and communications feeders should be available at the installation site.
For rooftop installations, the surface stability and reduced possibility of maintenance
workers passing in front of the link or equipment obscuring the line-of-site are important factors. In addition, the existence of a suitable edge on the rooftop with a nearby
penetration point through the wall is required for running a suitable-length conduit
18

that encloses cables for interconnections between the indoor and outdoor portions of
the installation. Lengths of cables carrying small-level signals should be shortened as
much as possible to minimize the attenuation levels.
A hardware platform should be employed in conjunction with the FSO link in
order to make use of its capabilities in both the analog and digital modes. High-speed
DACs and ADCs are required for driving the analog laser transmitter and capturing
the analog received signal, respectively. High-speed serial input/output ports (I/Os)
with optical connectivity are essential for interface with the digital transmitter and
receiver. Two QuiXilica TM -Triton TM -V5-VXS digitizer boards from Tekmicro [69]
have been used as discussed in Sec. 2.4.
2.3
FSO Link
A customized SONAbeam TM -1250-M FSO link was deployed in August 2010 across
McMaster University campus in west Hamilton area as illustrated in Fig. 2.1 [70].
The link consists of two identical full-duplex transceiver terminals. One terminal is installed on the rooftop of McMaster Innovation Park (MIP) building at Longwood area
southeast to the campus (latitude: 43 150 23.7500 North, longitude: 79 540 2.0600 West).
The other terminal is installed on the rooftop of Brandon-Hall (BH) students residence building at the northern-west backyard of the campus (latitude: 43 150 57.4700 North,
longitude: 79 550 10.9900 West). Other than a possibility for trees growth near MIP
that may block the line-of-sight between the two terminals, both the locations satisfy
the key requirements for a successful FSO link operation. Site views at both locations
are shown in Fig. 2.2. The distance between the two terminals is approximately 1870
meters.
19

c
Figure 2.1: Aerial view of the FSO link location (satellite image 2011
DigitalGlobe
Inc. [70].
Brandon-Hall
McMaster
Innovation
Park
McMaster
Innovation
Park
Figure 2.2: Site views of the experimental FSO link.

20

40 to 60 C
Dust- and water-tight:
IP66 rated [71], NEMA-4 enclosure [72]
Pointing stability
Against up to 120-km/hr wind speed
Dimensions (WHD)
41 41 46 cm3
Mass
Head: 20 kg, PCA box: 8 kg
Housing
Cast aluminum
Laser cooling
Active solid-state cooling
Electromagnetic compatibility FCC 47CFR15 [73]
ICES - 003 [74]
Input voltage
48 VDC
Power consumption
Transceiver: 55 W maximum
Heaters: 200 W maximum
Operating temperature
Environmental seal
Table 2.1: Environmental, mechanical, and electrical specifications of the SONAbeam TM 1250-M link [75].
Table 2.1 shows the main environmental, mechanical, and electrical specifications
of the FSO link. The wide operating temperature range (40 to 60 C), tight sealing
against water and dust, and stability against 120-km/hr wind speed qualify the link
for continuous outdoor operation throughout the year. Heating is necessary in cold
weather to prevent snow and sleet accumulation from blocking the transmitters or
receiver. Active solid-state laser cooling is used to keep the lasers temperatures below
35 C to increase their life times and sustain stable operation.
Each terminal consists of two main parts, the optical head and the power and
control assembly (PCA) box as shown in Fig. 2.3. The optical head represents
the physical interface with the FSO channel. It contains four laser transmitters
surrounding the receiver. The PCA box provides power as well as communication and
control signals from/to the optical head. The PCA box is user-accessible for selecting
the required reception mode. The manufacturer provides the SONAbeam TM Terminal
Controller (STC) software for managing the link [76]. It enables the user to control
21

PCA box
Mounting mast
Mounting yoke
Analog transmitter
(Laser 1)
4
Optical
head
Receiver
Brandon-Hall
Wall-penetration
Conduit
Digital transmitters
(Lasers 2, 3, 4)
Figure 2.3: Customized SONAbeam TM -1250-M optical head and PCA box.
and monitor all the operational parameters from a PC as shown in Appendix A.
2.3.1
Transmitter
Each optical head incorporates four Indium-Gallium-Arsenide-Phosphide (InGaAsP)

multiple-quantum-wells (MQWs) laser diodes for transmission as shown in Fig. 2.3.
Key specifications of the transmitters are summarized in Table 2.2. The 1550-nm
operational wavelength is selected for its advantages from an eye-safety perspective
where 50 times more output power is allowed than in the 850-nm range [77]. Each
laser diode has an independent driver and can operate at one of seven discrete power
levels selected via the STC software. The maximum output power of each laser is
80 mW drops to 60 mW after the lenses and filters. The angle of divergence is
22

Laser type
Peak maximum output power
Average maximum output power
Average power after lens and filter
Wavelength
Divergence angle
Laser safety certificates
InGaAsP MQWs
160 mW
80 mW
60 mW
1550 nm nominal, range 1520 - 1580 nm
2.5 mrad
ANSI Z136.1 [22] & ANSI Z136.6 [23]
IEC60825-1 [24]
21 CFR 1040.10 [25] & 21 CFR 1040.11 [26]
Analog transmitter (Laser 1)

Interface
50 single-ended, SMB connector
Maximum transmitted power
60 mW
Driver input level
500 mV
Digital transmitters (Lasers 2, 3, 4)
Interface
1310 nm SM fiber (Lasers 2, 3, 4)
Single-ended LVPECL (Laser 4)
Fiber input power
11 to 3 dBm
Maximum transmitted power
180 mW (60 3)
Binary data rate
100 Mbps to 1.6 Gbps
Transmission standards
Fast/Gig Ethernet [78]
Fiber Channel [79]
SONET [80] / SDH [81]: OC-1/STM-0,
OC-3/STM-1, OC-12/STM-4, OC-24
SMPTE [82], HDTV
Table 2.2: Specifications of the laser transmitters [83].
2.5 mrad full width at half maximum (FWHM). The optical head satisfies the ANSICLASS 1, IEC-CLASS 1M, and CDRH-CLASS 1M safety standards even when the
four lasers are operating at full power simultaneously.
Analog Transmitter
Laser (1) in Fig. 2.3 allows the analog transmission of an electrical signal provided by
a standard 50-, single-ended, AC-coupled SMB plug. Permitted input signal levels
are between 500 mV. A circuit diagram of the driver equivalent circuit is shown in
23

Vss
Laser diode
RF input
DC Bias
Figure 2.4: Equivalent circuit of the analog laser driver (modified based on [83]).
Fig. 2.4. The DC bias is added at the driver stage to avoid clipping and maintain a
linear mapping between the input electrical signal and output optical power.
Digital Transmitters
Lasers (2), (3), and (4) are used to transmit digital binary data. Lasers (2) and (3)
are fed by a standard 1310-nm single-mode (SM) optical fiber with SC termination.
As illustrated in Fig. 2.5, the input data undergoes opto-electrical conversion, regeneration, and optional re-clocking before driving the laser transmitter. Re-clocking is
dependent on the standard being transmitted and is software-controllable. Transmission rates span from 100 Mbps up to 1.6 Gbps where several standards are supported
including Fast and Gigabit Ethernet, Fiber Channel, and High-Definition Television
(HDTV). Protocol-independent transmission is also supported where re-clocking is
bypassed and raw bits can be transmitted with an arbitrary rate. The bypass mode
is the most suitable mode for investigating binary transmission over the link.
Laser (4) can be configured to allow the same operation as lasers (2) and (3)
by transmitting the data fed from the common fiber input. Alternatively, it can be
24

Opto-electrical
conversion
Binary 1310 nm
optical data
Electro-optical
conversion
Regeneration
Re-clocking
1550 nm
To transmitter
telescope
Bypass
(software-controllable)
Figure 2.5: Equivalent circuit of the driver for Lasers (2) and (3).
configured to be independently modulated by a digital electrical signal provided via
a single-ended LVPECL interface using an SMB connector. The operational mode is
configured from the PCA box.
2.3.2
Receiver
The receiver aperture, shown in Fig. 2.3, is 20 centimeters in diameter. Such large
aperture area is the main reason for the reduced scintillation at the receiver as shown
in Chapter 3. The receiver field-of-view is 2.3 mrad. Such small field-of-view added to
two spatial and two spectral solar filters help minimize the background and radiation
noise. The active detecting element is a single Indium-Gallium-Arsenide (InGaAs)
avalanche photodiode (APD) with 200-m length. A block diagram of the receiver
equivalent circuit is shown in Fig. 2.6. The photodetector is followed by a transimpedance amplifier (TIA) with automatic gain control (AGC) configured via the
STC software. The receiver supports dual-mode operation, analog or digital reception
at a time. The operational mode is configured from the PCA box.
In analog mode, the AGC should be disabled and a fixed gain is set by the
25

Vcc
APD bias
AGC (softwfare-controllable)
APD
1310 nm SM fiber
SC connector
TIA
Digital output
100 differential
SMB connectors
Analog (RF) output
Configured from
the PCA box
Electro-optical
conversion
Figure 2.6: Receiver equivalent circuit (modified based on [83]).

user from the STC software. The output voltage is provided directly from the transimpedance amplifier through a standard 100- differential-output via a pair of SMB
connectors. Typical output levels are between 30 mV at the gain level 2000 V/A.
In digital mode, the AGC is enabled and the output of the transimpedance amplifier feeds internal electronics where it undergoes threshold detection and electrooptical conversion to provide the binary output through a standard 1310-nm SM fiber.
Data rates from 100 Mbps up to 1.6 Gbps are supported.
2.4
FPGA-Based Digitizer Board
The architecture of the QuiXilica TM -Triton TM -V5-VXS board is shown in Fig. 2.7
and the main features and specifications are summarized in Table 2.3 [84]. Typical
customers of these high-performance boards are the military agencies. Key applications are communications, radar, sonar, and electronic warfare.
26

The core processing units are three Virtex TM -5 FPGAs from Xilinx [85]. FPGAs
(0) and (2) are XC5VSX95 while FPGA (1) is XC5VFX100. Each FPGA is supported
by two 512-MBytes DDR3 SDRAM memory banks, giving a total storage capacity of
3 GBytes on-board memory. The data bus width is 64 bits with clocking frequencies
between 300 and 400 MHz. Such memory size enables using the board as a large
buffer for the transmitted or received samples. Inter-FPGA connectivity is supported
by four high-speed serial ports (up to 3.75 Gbps each) and 50 differential pairs of
parallel I/O paths between every pair of the three FPGAs. The front-panel highspeed serial interface consists of two SFP+ ports and a single QSFP (quadruple
SFP) port giving a total of six digital I/O channels each running at up to 3.75 Gbps.
This interface enables the implementation of a variety of standard protocols, including
Gigabit Ethernet, Fiber Channel, and 10-Gigabit Ethernet. The analog I/O frontend is supported by a 10-bits, 2.2-GSa/s ADC input and a 12-bits, 4-GSa/s DAC
output. Four on-board clock sources exist in addition to a derivative of the external
clock used for the ADC and DAC.
27

Trigger +
Trigger
DAC out
FPGA 2
XC5VSX95
512 MB
DDR3 SDRAM
512 MB
DDR3 SDRAM
Clock in
Trigger +
Trigger
FPGA 0
XC5VSX95
ADC in +
512 MB
DDR3 SDRAM
512 MB
DDR3 SDRAM
FPGA 1
XC5VFX100
512 MB
DDR3 SDRAM
512 MB
DDR3 SDRAM
ADC in
Figure 2.7: Architecture of the Triton TM -V5-VXS digitizer board (modified based on
[84]).
28

FPGAs
2 XC5VSX95 (FPGAs 0, 2)
1 XC5VFX100 (FPGA 1)
FPGAs features 1536 DSP slices (25 18 multipliers)
(three combined) 4280 integrated block RAMs (36 Kbit)
2 PowerPC 440 RISC cores
up to 550 MHz clock
Serial I/O
2 SPF+ (up to 3.75 Gbps)
1 QSPF (up to 43.75 Gbps)
ADC
10-bits, 0.49 mV LSB resolution
250 mV input signal level
up to 2.2 GSa/s sampling rate
DAC
12-bits, 0.195 mV LSB resolution
400 mV output signal level
up to 4 GSa/s sampling rate
Memory
6512-MB DDR3 SDRAM
64-bits data bus
up to 400 MHz clock (DDR)
Clock sources
100, 125, 156.25, 212.5 MHz
derivative of the external ADC & DAC sampling clock
Table 2.3: Hardware specifications of the Triton TM -V5-VXS board [84].
29

2.5
Backbone Firmware
The capabilities of the Triton TM board allow the real-time implementation of many
coding and modulation schemes for investigation over the FSO link. However, in this
thesis, the approach taken is to use the board as a large buffer for the samples of the
transmitted or received waveforms, while all the necessary digital signal processing is
done off-line on PCs using software packages such as MATLAB.
A block diagram of the firmware used on the three FPGAs illustrating the key
functional blocks along with data paths is shown in Fig. 2.8. A firmware package is provided from Tekmicro that contains black-box cores for the DDR3 SDRAM
controller, a 4:16 demultiplexer (DMUX) for interface with the ADC, and a 16:4 multiplexer (MUX) for interface with the DAC. In addition, a Gigabit Ethernet UDP/IP
core is provided for establishing an Ethernet connection between the board and a PC
for full-duplex data transfer. The data transfer is controlled on the PC using a C++
code that is also provided by the manufacturer. A key challenge of the design is the
distribution of the memory banks among the three FPGAs. These memory banks require an FSM that monitors, controls, and synchronizes the samples movement across
the different clock domains on different FPGAs at rates up to 16 125 MSa/s as well
as synchronizes the data transfer between the on-board memories and the PC.
The analog signal delivered by the transimpedance amplifier of the FSO receiver
is sampled at the rate 2 GSa/s with 10 bits/Sa. The on-board ADC is followed
by a matched 1:4 DMUX to split the 2-GSa/s stream into four parallel 500-MSa/s
differential data paths which are connected to differential I/Os on FPGA (0). The
ADC interface core on the FPGA mainly acts as another 4:16 DMUX that further
splits the data stream into 16 parallel 125-MSa/s paths. The 125 MHz frequency
30

4X500 MSa/s
4:16 DMUX
AT84CS001
10-bit 2 GSa/s
1:4 DMUX
2 GSa/s
AT84AS008 Analog input

10-bit 2 GSa/s
ADC
100
Differential
FSO analog
receiver
16X125 MSa/s
Sampling rate
selector
Analog receiver data path

Analog transmitter data path
Gigabit Ethernet MAC
UDP/IP
DDR3 ADC
memory buffers
ADC buffers
controller
DAC buffers
controller
ADC stream
125 MHz
Oscillator
Ethernet
controller
GTP transceiver
(GigE physical layer)
Optical SFP
1000BASE-LX
(fiber)
DDR3 DAC
memory buffers
DAC stream
16X125 MSa/s
4X500 MSa/s
16:4 MUX
2 GHz
oscillator
MD653D
2 GSa/s
4:1 MUX / DAC
Analog output
50
single-ended
FSO analog
transmitter
Figure 2.8: Block diagram of the backbone firmware and hardware used as a universal
analog transceiver.
31

used is four times less than the maximum allowed clock frequency (550 MHz) for
the Virtex TM -5 family making clock constrains more relaxed. Finally, the required
sampling rate and number of bits per sample are selected before going to the memory
buffer. The ADC buffer controller is basically an FSM that controls the flow of data
between the DDR3 SDRAM controllers across the three FPGAs.
For the DAC path, the samples of the transmitted waveform arrive from the onboard memory buffers in 16 parallel 125-MSa/s paths. A 16:4 MUX compresses the
stream into four 500-MSa/s differential data paths. Finally, the on-board DMUX/DAC
further multiplexes the paths into a single 2-GSa/s stream that drives the analog laser
transmitter of the FSO link.
The Gigabit Ethernet controller on FPGA (2) is responsible for communications
between the board and the connected PC and contains three channels. Two data
channels are used to upload/download samples to/from the DAC/ADC buffers. The
third is a control channel for sending start/stop commands to the board.
For applications that require continuous operation, such as channel measurements
during several hours or days, the system can be left running continuously however
once the on-board memory banks are full, a period of approximately 24 minutes is
needed to dump the entire samples (up to 3 GBytes) into a PC before capturing a
new set of samples.
2.6
Experimental Setup
A block diagram of the entire experimental setup at both locations is shown in Fig.
2.9. Each optical head is connected to a PC via a serial RS-232 port (DB9 connector)
to run the STC software. For redundancy, each optical head has another connection
32

to the McMaster University campus LAN via an Ethernet switch in case the roof PC
fails. For data connections, each terminal is connected to a Triton TM digitizer board
via the PCA box where the DAC and ADC are connected to the analog transmitter
and receiver, respectively while the optical SFP port is connected to the digital data
interface. The digitizer board is connected to the PC via an optical Gigabit Ethernet
link using the optical SFP to establish a UDP/IP connection for data transfer between
the on-board memory and the PC as well as for controlling the board operation.
The input to the ADC and DAC sampling clock is driven by an ultra low-noise 2GHz crystal oscillator from Wenzel [86]. On-board FPGAs are programmed via a
JTAG connection between the PC and the board. A 4-channel, 20-GSa/s digital
oscilloscope is used for visualizing signals. A weather station is fixed on top of the
optical head at MIP since most of the time this side is used as the receiver. The
weather station is provided with a software for logging temperature, pressure, wind
speed, wind direction, and rain level [87]. The entire system at both locations is
managed remotely from the central PCs at the FOCAL lab at ITB building. The
setup at MIP site is shown in Fig. 2.10.
33

FOCAL Lab
Weather
station
GigE
connection
Brandon-Hall
100 M
Ethernet
100 M
Ethernet
Campus Network
10 M
Ethernet
10 M
Ethernet
RS 232
RS 232
Gig Ethernet
(fiber)
Gig Ethernet
(fiber)
2 GHz
Clock
2 GHz
Clock
Optical SFP
Optical SFP
DAC
DAC
Analog
DDR3
SDRAM
McMaster
Innovation
Park
PCA
box
ADC
Optical
SFP
Optical
head
1.87 km
Optical
head
Analog
PCA
box
ADC
Digital
Digital
Optical
SFP
Figure 2.9: Block diagram of the experimental setup.
34
DDR3
SDRAM

(a)
Weather
station
Mounting
mast
Optical head
PCA box
McMaster
Innovation
Park
Outdoors
(b)
Digital scope
Conduit
Indoors
Interface with
PCA box
Oscillator
power supply
FSO link power

supply
2 GHz
oscillator
Digitizer board
power supply and
ventilation chassis
Digitizer
board
McMaster
Innovation
Park
PC
Figure 2.10: Experimental setup at McMaster Innovation Park: (a) outdoors, (b)
indoors.
35

2.7
Conclusions
In this chapter, the details of the experimental FSO link at McMaster University
were presented. The link supports both analog and digital transmission. Technical
specifications of the optical heads along with their main features and functionalities
were described in detail. The high-speed digitizer board used as the hardware platform was briefly outlined. The implemented firmware was explained highlighting its
main function as a large buffer. Such a buffer is used for transmitting arbitrary generated waveforms using the DAC and buffering the received samples over the ADC for
off-line processing. The firmware can be used as a universal transceiver for channel
measurements as well as for testing various modulation schemes over the FSO link.
Finally, the architecture of the overall experimental setup was described. This setup
acts as the foundation for the experimental work explained in the remaining of the
thesis.
In Chapter 3, FSO channel measurements are conducted under a variety of atmospheric conditions. The received samples are used to estimate the statistics of the
fading for proper channel modelling.
36
Chapter 3
FSO Link Measurement and
Modelling
3.1
Introduction
Optical waves propagating through the atmosphere experience attenuation and intensity fluctuations that degrade the performance of FSO communication systems
near the earth surface even for short distances. Attenuation is usually fixed over
relatively long durations and therefore can be predicted or measured and accounted
for. On the other hand, intensity fluctuations are inherently random and occur on
the order of milliseconds resulting in a slow-fading channel. From a communication
systems design perspective, it is always required to model a fading channel using
statistical models. In this chapter, the experimental setup described in Chapter 2
along with a newly developed measurement technique are employed for conducting
the FSO channel measurements and modelling.
In Sec. 3.2, a brief review of the FSO channel and general propagation effects
37

Chapter 3. FSO Link Measurement and Modelling
is introduced along with the log-normal distribution for modelling weak-turbulence
FSO channels. In Sec. 3.3, a general channel model for the FSO link combining
the transmitter, atmospheric channel, and receiver is introduced. Each block of the
proposed channel model is experimentally characterized for proper modelling of the
entire system. In Sec. 3.4, the developed measurement technique is described and the
results from the channel measurements are represented in different weather conditions.
Fitting with the log-normal distribution is also investigated. In Sec. 3.5, a finite-state
Markov model is developed and simulation results are verified against the channel
measurements.
3.2
Channel Model
3.2.1
Geometrical Loss
An unguided electromagnetic wave propagating in free space experiences power loss

at the receiver end due to the geometry of diffraction combined with the limited
effective area of the receiver. The reduction in optical power at the receiver due to
the geometrical loss hg can be estimated by [5]
hg
2
2
Drx
Drx
,
(Dtx + tx L)2
(tx L)2
Dtx << tx L,
(3.1)
where Dtx is the transmitting lens diameter, tx is the transmitted beam divergence
angle, L is the propagation distance, and Drx is the receiving lens diameter.
38

3.2.2
Atmospheric Loss and Turbulence
An optical wave propagating through the atmosphere undergoes attenuation as well

as random fluctuations in the intensity. Atmospheric loss refers to the absorption
and scattering of optical energy by aerosols and molecules with sizes comparable to
the optical wavelength [20, 8890]. Atmospheric loss is usually dominated by aerosol
absorption and Mie scattering [88]. Examples of absorbing aerosols are water droplets,
dust, and organic objects. In addition, the interaction between the optical wave and
the molecules of N2 , O2 , O3 , and CO2 gives rise to the optical energy loss in the
form of heat. Mie scattering is caused by atmospheric particles comparable in size
with the wavelength where the scattered waves mainly go to the forward direction.
Mie scattering decreases with increasing wavelength and it is the main reason for
the red appearance of sunset [20]. In general, atmospheric loss is mainly dependent
on the visibility which is directly affected by the air quality and weather conditions.
Therefore, atmospheric loss is usually considered fixed for relatively long durations
(on the order of hours).
Atmospheric turbulence refers to the inhomogeneity in the atmospheric structure
on a macroscopic level due to temporal and spatial temperature gradients that result in a nonuniform index of refraction along the propagation path. Therefore, the
transmitted optical wave undergoes random refractions resulting in fluctuations of
the signal intensity at the receiver, usually termed as scintillation. Scintillation is
dominant for long paths (beyond 1 km) and those near or parallel to the ground
surface. Scintillation causes a time-varying fading effect at the receiver on the order
of milliseconds justifying a slow-fading channel model for transmission rates at giga
symbols per second.
39

Without loss of generality, it is possible to factor the total atmospheric loss ha
into a fixed loss term hl and a scintillation term hsc , i.e.
ha = hl hsc .
(3.2)
The atmospheric loss hl is usually described by the exponential Beers-Lambert Law

hl = eL/2 ,
(3.3)
where L is the propagation distance (km) and is the atmospheric attenuation coefficient (km1 ). By assuming a small aerosol absorption compared to Mie scattering,
the following formula can be used [91]
3.91
550 nm
q(V )
,
(3.4)
where is the optical wavelength (nm), V is the visibility (km), and q(V ) is the size
distribution of the scattering particles given by
1.6
q(V ) =
1.3
0.585V
V > 50
50 V > 6 .
1
3
(3.5)
6>V
Usually the atmospheric loss is significant for foggy weather or when the water-vapor
density is considerably high, but it can be ignored for clear weather conditions compared to the geometrical loss.
On the other hand, scintillation, as a random phenomena, is characterized using
40

mathematical statistical models. Many of these models have been proposed since
1970s, however none of them can be universally applied due to the non-stationary
nature of the atmospheric turbulence [20]. In the weak-turbulence regime, the lognormal distribution is usually used. It is widely accepted because of the considerable
match with the experimental measurements as well as the mathematical simplicity,
i.e. it is uniquely defined by a single parameter that can be directly related to
the weather measurements. However under strong turbulence conditions, the lognormal distribution usually fails to give good fit with the measurements and doublystochastic distributions are applied [20]. The K-distribution [92, 93] and the lognormally modulated exponential distribution [94] are good candidates that usually
show good agreement with the experimental measurements in the strong-turbulence
regime. The IK distribution [95], the Beckmann distribution [96], and the gammagamma distribution [8] have been shown to fit with the channel measurements under
a wide range of atmospheric conditions. In this thesis, the log-normal distribution is
mainly considered since it has a simple form and the channel measurements show a
weak-turbulence condition with scintillation indices far below unity (on the orders of
102 and 101 ).
3.2.3
The Log-Normal Distribution
The conceptualization of the log-normal distribution is straightforward. It results

from modelling the field of the optical wave at any point in the medium along the
propagation path as the product of field components scattered from a large number of
independent scatterers. By applying the central limit theorem, the logarithm of this
product, i.e. the log-amplitude of the optical intensity, will obey a normal (Gaussian)
41

distribution [90].
Mathematically, the log-normal distribution follows from the Rytov method, i.e.
Rytov approximation, for solving Maxwells wave equation in a random medium [89,
Sec. 5.2.1] and [20, Sec. 5.4 and 5.7.2].
The electric field U (r) of a narrowband wave propagating in a medium with a
random index of refraction n(r) is described by the Maxwells wave equation
2 U (r) + k 2 n2 (r)U (r) [ U (r)] = 0,
(3.6)
where r = x + y + z with x, y, and z are unit vectors along the x, y, and z axes,
respectively, k = 2/ is the wave number, (/x)x + (/y)y + (/z)z is
the gradient operator, and 2 is the Laplacian operator. The last term in Eq. (3.6)
represents polarization which can be ignored for the case of optical waves propagating
in the atmosphere [89].
The index of refraction n(r) can be expressed as the sum of the free-space value,
i.e. unity, plus a random component due to atmospheric turbulence,
n(r) = 1 + na (r).
(3.7)
Therefore, Eq. (3.6) can be simplified to
2 U (r) + k 2 [1 + na (r)]2 U (r) = 0.
(3.8)
The Rytov approach for solving Eq. (3.8) assumes that the electric field U (r) can be
expressed as the multiplication of an unperturbed field and perturbation terms in the
42

form
U (r) = e0 (r)+1 (r)+2 (r)+...... = U0 (r)e1 (r)+2 (r)+...... = U0 (r)e(r) ,
(3.9)
where U0 (r) = e0 (r) is the unperturbed free-space field component while e(r) =
e1 (r) + e2 (r) + ...... represents the perturbation terms, where 1 (r) and 2 (r) are
the first-order and second-order complex phase perturbations, respectively.
By considering only the first-order perturbation term, i.e. (r) = 1 (r), which is
usually valid under weak-turbulence conditions, Eq. (3.9) simplifies to
U (r) = U0 (r)e1 (r) .
(3.10)
The complex phase perturbation 1 can be written as
1 (r) = 1 (r) + jS1 (r),
(3.11)
where 1 (r) denotes the first-order log-amplitude and is Gaussian distributed, while
S1 (r) denotes the first-order log-phase. Therefore, the irradiance of the field at any
point in the turbulent medium along the propagation path is given by
I(r) = |U (r)|2 = |U0 (r)|2 e1 (r)+1 (r) = |U0 (r)|2 e21 (r) .
(3.12)
Since the exponent in Eq. (3.12) is Gaussian distributed, it follows that the logarithm
of the irradiance is Gaussian distributed and, by definition, the irradiance is lognormally distributed. The notation in Eq. (3.12) can be simplified by discarding
the spatial dependence and considering the irradiance at some point, usually at the
43

receiver,
I = A2 e21 ,
(3.13)
where A = |U0 |. Equation (3.13) states that the irradiance intensity is equivalent to
the free-space unperturbed value A2 modulated by the random fluctuation term e21 ,
where the log-amplitude 1 is Gaussian distributed, i.e.
"
#
1
(1 )2
p1 (1 ) =
exp
,
22
2
(3.14)
where = h1 i is the mean of the log-amplitude 1 , 2 = h21 ih1 i2 is the variance,

and hi denotes the expected value. From Equations (3.13) and (3.14), the PDF of
the irradiance fluctuations is given by the log-normal distribution
"
ln
1
pI (I) =
exp
2 2 I
I
A2
2
82
2 #
,
I > 0.
(3.15)
It can be shown that

2
I = A2 e(2 +2 ) ,
(3.16)
2

2
I2 = A4 e(4 +4 ) e4 1 ,
(3.17)
and
where I = hIi is the mean of the irradiance I and I2 = hI 2 i hIi2 is the variance.
The irradiance mean I refers to the fixed term of the atmospheric gain that can
be compensated for by amplifying or attenuating the optical power to obtain a unity
average gain. Therefore, by setting I = 1, then A = 1, = 2 , and Eq. (3.15)
44

simplifies to
"
2 #
ln I + 22
1
pI (I) =
exp
82
2 2 I
I > 0,
hIi = 1.
(3.18)
2
2
2
By defining the log-irradiance variance ln
I = h(ln I) ihln Ii , then from Eq. (3.13),
ln I = 2
(3.19)
and Eq. (3.18) can be rewritten as

"
2 #
2
ln I + 12 ln
1
I
pI (I) =
exp
2
2ln
2ln I I
I
I > 0,
hIi = 1.
(3.20)
2
,
The normalized variance of the irradiance, well known as the scintillation index SI
is given by
2
SI
=
hI 2 i hIi2
,
hIi2
I > 0.
(3.21)
For a normalized irradiance, the scintillation index is equal to the variance. Therefore,
Eq. (3.21) simplifies to
2
2
SI
= hI 2 i 1 = I2 = e(ln I ) 1,
3.3
I > 0,
hIi = 1.
(3.22)
Transceiver Measurement and Modelling
Proper characterization of the FSO link is necessary for a clear interpretation of the
channel measurements as well as for accurate simulation of the underlying communication system. A proposed general channel model for the link is illustrated in Fig.
45

Laser diode
Laser driver
s(t)
hTx(t)
x(t)
fLD[x(t)]
Receiver
I(t)
xLD (t)
R G hRx(t)
y(t)
hg hl hsc(t)
FSO channel fading
r(t)
w(t)
AWGN
Figure 3.1: Proposed channel model for the FSO link.

3.1. The received signal r(t) can be expressed as
r(t) = R G hg hl hsc (t) [fLD [s(t) hT x (t)] hRx (t)] + w(t),
(3.23)
where R is the photodetector responsivity, G is the transimpedance amplifier gain, hg

is the geometrical loss, hl is the average atmospheric loss, hsc (t) is the scintillation,
fLD () is the laser transfer function taking non-linearity into consideration, s(t) is the
transmitted signal, hT x (t) is the impulse responses of the laser driver, hRx (t) is the
impulse responses of the receiver, and w(t) is an AWGN. In Subsec. 3.3.1, it is shown
that laser non-linearity is negligible for the signal levels used. Therefore, fLD [x(t)]
b x(t), where b is a constant that is assumed unity without loss of generality, and Eq.
(3.23) can be rewritten as
r(t) = R G hg hl hsc (t) [s(t) hT x (t) hRx (t)] + w(t).
(3.24)
In the following subsections, the blocks in Fig. 3.1 are experimentally quantified. The
channel measurements are presented in Sec. 3.4.
46

x(t)
r(t)
G
Laser diode
Channel loss h
Photo-detector
(responsivity R)
I(t) = h[a + bx(t) +

cx2(t)+ dx3(t)]
Figure 3.2: Simplified channel model used for harmonic distortion measurements.
3.3.1
Harmonic Distortion and Nonlinearity
Linearity is critical when analog waveforms, such as OFDM signals, are transmitted
using the optical links. As opposed to the fiber channel, the FSO atmospheric channel
can be assumed linear. However, the laser transmitter is a source of non-linearity.
To quantify such non-linearity, a third-order memory-less polynomial model is used
along with the simplified channel model illustrated in Fig. 3.2. The received signal
r(t) is expressed as

r(t) = R G h a + bx(t) + cx2 (t) + dx3 (t) ,
(3.25)
where h represents the channel gain, x(t) is the input current to the laser driver,
and a, b, c, and d are the polynomial coefficients to be determined experimentally.
Noise and optical power fluctuations at the receiver are ignored by averaging many
measurements over time. The scaling factor R G h is assumed unity without loss of
generality.
To obtain the polynomial coefficients, a single tone at some frequency f0 is transmitted using the maximum available power (60 mW). Therefore, Eq. (3.25) can be
47

f0 (MHz)
100
200
300
400
a
0.03348
0.02480
0.02478
0.00903
c
0.078322
0.060347
0.052214
0.035498
d
0.015812
0.015692
0.012373
0.007113
Table 3.1: Normalized polynomial coefficients obtained by harmonic distortion measurements at frequencies 100, 200, 300, and 400 MHz.
rewritten as

c
3d
d
c
+ b+
cos [2(f0 )t] +
cos [2(2f0 )t] +
cos [2(3f0 )t]
r(t) = a +
2
4
2
4
(3.26)

when x(t) = cos (2f0 t).

At the receiver, the output signal r(t) is sampled at 5 GSa/s using a digital scope.
The power spectral density (PSD) is estimated from the time samples to obtain the
polynomial coefficients using Eq. (3.26) by equating the corresponding frequency coefficients. Figures 3.3 (a), (b), (c), and (d) show the PSDs of the received waveforms
when the frequencies 100, 200, 300, and 400 MHz are transmitted, respectively. Corresponding values of a, c, and d, normalized with respect to b, the coefficient of the
fundamental frequency f0 , are shown in Table. 3.1. Slight differences between the
coefficients at different frequencies are mainly due to the non-flat frequency response
of the system as will be shown in Subsec. 3.3.2. For all the frequencies tested, the
power level of the second harmonic is at least 22 dB less than the fundamental frequency while the third harmonic is at least 35 dB less. Therefore, non-linearity is
negligible and will not be considered.
48

(a)
(b)
0
f0 = 100 MHz
10
Normalized power spectral density (dB/Hz)
a = 0.033489
b=1
c = 0.078322
d = 0.015812
20
30
40
50
60
70
80
90
100
500
1000
a = 0.0248
b=1
c = 0.060347
d = 0.015692
20
30
40
50
60
70
80
90
100
1500
f0 = 200 MHz
10
Frequency (MHz)
500
(c)
0
f0 = 300 MHz
10
1500
(d)
a = 0.02478
b=1
c = 0.052214
d = 0.012373
20
30
40
50
60
70
80
90
100
1000
Frequency (MHz)
500
1000
a = 0.0090357
b=1
c = 0.035498
d = 0.0071128
20
30
40
50
60
70
80
90
100
1500
Frequency (MHz)
f0 = 400 MHz
10
500
1000
1500
Frequency (MHz)
Figure 3.3: Normalized PSD at the receiver used for harmonic distortion measurements at four frequencies: (a) 100 MHz, (b) 200 MHz, (c) 300 MHz, (d) 400 MHz.
49

3.3.2
Frequency Response
Frequency response is the fundamental characterization of any communication system. For FSO systems, the atmospheric channel, excluding the electronic systems at
the transmitter and receiver, is virtually unlimited in bandwidth. Limitations come
mainly from the electronic components that build up the transmitter and receiver
limiting the useful bandwidth. For the available FSO link, the frequency response
should be determined before installation to characterize the transmitter and receiver
separately. Since that was not possible, frequency response measurements have been
conducted after installation.
The equivalent amplitude frequency response of the analog transmitter and receiver together is obtained by the means of estimating the PSD of the received signal
when a broadband signal within the bandwidth of interest is transmitted. To approximate such a signal, 101 equal amplitude sinusoids with frequencies from DC
up to 1 GHz with 10 MHz step are transmitted simultaneously. The resulting time
signal is the pulse train shown in Fig. 3.4. The received waveform is sampled at 2
GSa/s. The resulting normalized PSD, shown in Fig. 3.5, is estimated by Welchs
periodogram method [97] using 225 (32 M) samples. It is noticed that the system acts
as a low-pass channel with 3-dB cutoff frequency around 200 MHz. A satisfactory
SNR at the receiver can be obtained at up to 400 MHz where attenuation is less than
6 dB. Beyond 400 MHz, the increase in power loss is steep and practically no useful
signal is obtained. In the OFDM over FSO experiments discussed in Chapter 4, the
used bandwidth spans from 20 MHz up to 200 MHz.
50

500
Transmitter input voltage (mV)
450
400
350
300
250
200
150
100
50
0
0.5
1.5
2.5
3.5
Time (sec)
Figure 3.4: Transmitted pulse train used for frequency response measurements.
51

Normalized power spectral density
2
4
6
8
10
12
14
16
18
20
1
10
10
10
Frequency (MHz)
Figure 3.5: Normalized magnitude frequency response of the FSO link obtained by
transmitting the pulse train shown in Fig. 3.4.
52

3.3.3
Noise
In FSO systems, the dominant noise sources are the background radiation and the
thermal noise at the receiver. Being a random phenomena, noise is best characterized
using stochastic models. Background and thermal noise is usually modelled as an
AWGN.
The PDF of an AWGN is given by
"
#
(r w )2
1
exp
,
pr (r) = p
2w2
2w2
(3.27)
where r is the noise sample, w = hri is the mean of r, and w2 = hr2 i hri2 is the
variance.
The Gaussian fit with noise samples can be obtained using the maximum likelihood
method [97] by maximizing the log-likelihood function given by
L(w , w2 ) =
N
X
i=1
(
ln
"
#)
(ri w )2
1
p
exp
,
2w2
2w2
(3.28)
where ri is the ith received noise sample and N is the total number of samples. It
can be shown that
w and
w2 that maximize Eq. (3.28) are given by [98]
w =
N
1 X
ri
N i=1
w2 =
and
N
1 X
(ri
w )2 ,
N i=1
(3.29)
respectively.
To obtain the noise statistics for the available setup, the laser transmitter is turned
off and the receiver output is sampled at 2 GSa/s. Figure 3.6 shows the histogram
of the noise samples, representing the background and thermal noise, along with the
53

Gaussian fit. The histogram is obtained by uniformly dividing the noise samples
range into M bins each with width . The number of samples in each bin nm is scaled
down using the formula
M
X
nm
n
m =
,
N
nm = N,
(3.30)
m=1
to satisfy the unity area-under-curve constraint for PDFs expressed by

M
X
n
m = 1.
(3.31)
m=1
The total number of samples used is N = 225 and the bin width = 0.49 mV which
is the ADC resolution. Using Eq. (3.29), the estimated mean and variance of the
Gaussian distribution are given by
w = 0.62142 mV
and
w2 = 0.62906 V2 ,
(3.32)
respectively.
The root mean square error (RMSE) is usually used for assessing the goodness
of fit between the measured samples and the fitting distribution [97]. The RMSE is
defined by
v
u
M
u1 X
t
RMSE =
(
nm n
m )2 ,
M m=1
(3.33)
where n
m is the number of samples in the mth bin calculated using the fitting distribution. For the noise samples shown in Fig. 3.6, the RMSE equals 0.028116.
To quantify the effects of the background noise and thermal noise separately, the
54

10
Noise samples
Gaussian fit
= 0.62142 mV
= 0.79313 mV
RMSE = 0.028116
1
10
Probability density
10
10
10
10
10
Noise amplitude (mV)
Figure 3.6: Histogram of thermal plus background noise samples at the receiver along
with Gaussian fit.
55

receiver aperture was covered to remove the background noise, however no difference
in the noise statistics was observed. Moreover, no difference was observed between
day and night measurements. Therefore, noise at the receiver comes mainly from the
thermal noise of the electronic circuits and can be modelled as an AWGN.
For the available FSO link, the noise at the receiver depends mainly on the gain
level used by the transimpedance amplifier following the photodetector. It was found
by many measurements that the gain level G = 2000 V/A is the one that maximizes
the SNR of the received signal at the two receivers at both locations. It was also
found that the two receivers do not have the same noise characteristics. The receiver
at BH has approximately 7.5 dB noise power more than its counterpart at MIP at
the gain level G = 2000 V/A. This is the reason why the optical head at BH was
always used as the transmitter while the one at MIP was used as the receiver. All
the experimental work presented in this thesis was conducted by transmitting from
BH and receiving at MIP using the gain level G = 2000 V/A.
3.3.4
Power Loss Estimate
The maximum average optical power transmitted from the analog laser is 60 mW.
At the receiver, the average received optical power for an unmodulated transmission
in clear weather conditions (visibility is more than 24 km [99]) is 12 W. Therefore,
the average total loss in optical power at the receiver is approximately 37 dB. Out of
many possible sources for the loss, geometrical loss hg and the average atmospheric
loss hl can be roughly estimated.
Geometrical loss can be obtained using Eq. (3.1). For the available FSO link,
the laser divergence angle tx = 2.5 mrad, the link distance L = 1.87 km, and the
56

receiver diameter Drx = 0.2 m. Therefore, the resulting geometrical loss hg = 27 dB.
In clear weather conditions, the average atmospheric loss hl estimated using Eq. (3.3)
is approximately 1 dB. Other possible sources of the power loss are alignment loss,
receiver filters and lens loss, and coupling loss.
37
| {zdB}
Total optical loss
3.4
3.4.1
|27{zdB}
1| {z
dB}
dB}
|9 {z
Geometrical loss
Atmospheric loss
Alignment, optics,
hg
hl
and coupling loss
Channel Measurements
Measurement Procedure
Usually an unmodulated optical signal is used for FSO channel measurements as

in [43, 100]. However, in this thesis, an optical signal modulated by a sinusoid is
used instead. At the receiver, an FFT is applied as a notch filter to the received
samples to minimize the noise corruption. A block diagram of the signal processing
system at the receiver is illustrated in Fig. 3.7. A sinusoid of frequency 117.187500
MHz is transmitted through the atmosphere from the analog laser. This frequency
is selected since it is equal to an integer number of the FFT bin widths used. For a
sampling frequency of 2 GSa/s and an FFT of length 512, 30 complete periods of the
117.187500-MHz sinusoid are applied to the FFT input, i.e.
512
30
=
.
9
2 10
117.187500 106
57

As shown in Fig. 3.7, the received signal is sampled at 2 GSa/s and 8192 = 16 512
consecutive samples are buffered at a time. These samples are divided into 16 nonoverlapping groups where each group undergoes a 512-point FFT. Xilinx Radix-4
FFT core [101] is used since it gives the lowest latency for the considered design. The
following 11456 samples are discarded because of the FFT latency time. For each
group, the FFT real and imaginary coefficients corresponding to the bin contains
the 117.187500 MHz frequency are selected to calculate the magnitude frequency
response. Finally, the magnitudes coming from the 16 segments are averaged and
buffered on the on-board memory with an 8-bit resolution. Bit growth during calculations is shown in Fig. 3.7 along the data path. Bit allocation is a compromise
between the dynamic range and resolution. It is carefully chosen to avoid clipping
and keep high resolution at the same time. Using the above measurement parameters
produces channel samples at the rate of 101.8 kSa/s.
The family of Fourier transforms, including the discrete Fourier transform (DFT)
which is efficiently implemented using FFT, is a one-to-one transformation, i.e. there
is a one-to-one correspondence between every transformation pair, and it is always
possible to move from one domain to the other using the forward and inverse transformation functions. Another important feature of Fourier transforms is linearity that
keeps the scaling ratios in both domains. Therefore, obtaining the channel distribution from the Fourier coefficients is exactly equivalent to getting it from the time
samples.
The key advantage of applying FFT is that it helps reject a large portion of
the interfering noise. All the noise in the receiver electrical bandwidth is rejected
except for the noise remaining in the frequency bin of interest. The bin width is
58

30 periods
...
0.5 ns
117.187500 MHz
4.096 s
9.824 s
511 512 513
8192
Discard 11456 ADC samples
19649 19650
Start new cycle
8192 samples buffer (8 bits/Sa)

8 bits
512 FFT
18 bits
Re
2
Re()
512 FFT
Im
Re
Im()
Re()
16
512 FFT
Im
Re
2
Im()
Re()
Im
Im()2
36 bits
35 bits
SQRT()
SQRT()
SQRT()
17 bits
8 bits
Channel state samples

h
at 101.8 kSa/s
Figure 3.7: Block diagram of the measurement system at the receiver.
59

0
10
20 log [abs(FFT)]
20
30
40
50
60
70
50
100
150
200
250
300
350
400
450
500
512FFT bins
Figure 3.8: Normalized PSD of the received signal used for channel measurements.
determined by the FFT length and can be made arbitrary small by increasing the
length to obtain smaller bin width and less noise. However, increasing the FFT
length will decrease the resulting sampling rate (101.8 kSa/s). In addition, from an
implementation perspective, increasing the FFT length will increase the latency time
needed by the FFT core. Therefore, more samples will be discarded and the number
of samples captured for processing will be reduced. As a compromise between the
noise performance and FFT latency, the data samples are divided into 16 segments
where 16 parallel 512-FFT cores are used instead of a single 8192-FFT core. Such
division reduces the overall latency time by a similar factor of 16 [101] allowing more
samples to be processed. Figure 3.8 shows the normalized PSD of the received signal,
before applying the FFT, under clear-weather conditions. When FFT is applied to
the time samples, all the noise power is rejected except for the noise remaining in
the FFT bin surrounding the 117.187500-MHz frequency and, therefore, the resulting
average SNR is approximately 40 dB.
60

3.4.2
Irradiance Fluctuations
Experiments for channel measurements using the technique described in Sec. 3.4.1
have been conducted in October and November 2011. For all the measurements, the
analog laser transmitter was launched at BH, emitting 60 mW optical power and
transmitting a sinusoid with frequency 117.187500 MHz. The receiver was set to the
analog mode. The channel state h is captured at the rate 101.8 kSa/s where each
sample is represented by 8 bits. Using all the available on-board memory (3 GBytes)
permits continuous recording of the channel state for approximately 8 hours and 47
minutes. Then, a 24-minute idle period is required to move the data from the board
to a PC before capturing new samples.
Coherence Time
To get insight about the channel behaviour during short time scales, Figures 3.9 (a),
(b), (c), and (d) show the channel envelope h during periods of 1 msec, 10 msec, 100
msec, and 1 sec, respectively.
A fundamental timing measure of fading wireless channels in general is the coherence time Tc and its frequency reciprocal the coherence bandwidth Bc 1/Tc . The
coherence time specifies the time duration wherein two received signals show high
amplitude correlation. In RF channels, the channel time-variation is due to the relative motion between the transmitter and receiver or the movements of objects in
between resulting in the Doppler spreading effect. In FSO channels, the transmitter
and receiver are usually fixed and the time-variations are due to the atmospheric turbulence along the propagation path. A channel with a large coherence time compared
to the transmitted symbols duration, i.e. Tc >> Ts , is typically termed a slow-fading
61

(b)
150
140
140
130
130
120
120
110
110
Intensity
Intensity
(a)
150
100
100
90
90
80
80
70
70
60
60
50
0.2
0.4
0.6
0.8
50
Time (msec)
140
140
130
130
120
120
110
110
100
90
80
80
70
70
60
60
0.02
0.04
10
0.8
100
90
(d)
150
Intensity
Intensity
(c)
150
50
Time (msec)
0.06
0.08
50
0.1
Time (sec)
0.2
0.4
0.6
Time (sec)
Figure 3.9: Received irradiance fluctuations through different time scales: (a) 1 msec,
(b) 10 msec, (c) 100 msec, (d) 1 sec (samples measured on October 30th , 2011 at 11:32
PM, temperature: 7 C, wind speed: 7.2 km/hr).
62

channel. In such situations, the channel state is considered constant during the transmission of a sequence of symbols.
To estimate the FSO channel coherence time Tc , the autocovariance, i.e. the
autocorrelation of the mean-removed sequence, is calculated. The mean-removed
is given by
channel gain h
N
X
h = h hhi = h 1
hi ,
N i=1
(3.34)
where hi is the ith channel sample and N is the total number of samples used for the
calculation. The autocovariance Rh h is estimated by
Rh h (m) =
PN m h
n+m h
n
N 1m0
Rh h (m)
N + 1 m < 0
n=1
(3.35)
is obtained by
The normalized autocovariance R
hh
(m) = Rh h (m)
R
hh
Rh h (0)
(3.36)
along with the PSD. By considering the coherence time at 5%

Figure 3.10 shows R
hh
of the peak at zero lag, i.e.
(Tc ) = 0.05,
R
hh
(3.37)
the estimated coherence time Tc = 33.57 msec corresponding to a coherence bandwidth Bc 30 Hz. Such coherence time justifies a slow-fading channel model for
transmission rates in the megabit or gigabit per second ranges.
63

(a)
Normalized autocovariance
1.2
1
0.8
0.6
0.4
0.2
0
0.2
100
80
60
40
20
20
40
60
80
100
Time (msec)
(b)
Normalized power spectral density
0
10
20
30
40
50
60
70
80
10
10
10
10
10
Frequency (Hz)
Figure 3.10: Irradiance fluctuations statistics: (a) normalized autocovariance, (b)

normalized power spectral density (samples measured on October 30th , 2011 at 11:32
PM, temperature: 7 C, wind speed: 7.2 km/hr).
64

Scintillation Index
Scintillation index is the normalized variance of the irradiance fluctuations, as given
by Eq. (3.22). For FSO channels, it is a measure of the optical turbulence strength
that causes random fluctuations of the received signal. Usually the turbulence is
considered weak for scintillation index values less than unity [20]. For the sake of
illustration, Figures 3.11 (a) and (b) show the average channel gain hhi and the
2
corresponding scintillation index SI
during a 14-hour period on October 31st , 2011
starting from midnight. Each measurement of average gain or scintillation index is

calculated using N = 32 MSa measurements, i.e., every 5.5 minutes. The scintillation
index can be estimated from the channel measurements through
2
SI
=
hh2 i hhi2
,
hhi2
(3.38)
where
N
1 X
hh i =
(hi hhi)2
N i=1
2
and
N
1 X
hhi =
hi .
N i=1
(3.39)
Figures 3.12 (a) and (b) show the corresponding temperature and wind speed,
respectively. Although the weather was clear, small changes in scintillation index
are expected due to changes in temperature, wind speed and direction, pressure,
and changing particulate density. The measured SI is far below unity justifying a
weak turbulence regime. Notice that at night (between 00:00 and 09:00 hrs) the
temperature was in the range of 4 7 C and the wind speed was less than 2 km/h.
Afterwards, the temperature increased to 11 C as did the wind speed between 8 and
16 km/h. In general, higher temperature and wind speeds correspond to stronger
atmospheric turbulence and higher scintillation index.
65

(a)
130
125
120
Average intensity
115
110
105
100
95
90
85
80
100
200
300
400
500
600
700
800
500
600
700
800
Time (min)
(b)
0.08
0.07
Scintillation index
0.06
0.05
0.04
0.03
0.02
0.01
0
100
200
300
400
Time (min)
Figure 3.11: Irradiance fluctuations during a 14-hour duration: (a) average intensity,
(b) scintillation index (samples measured on October 31st , 2011 starting from 12:00
AM).
66

(a)
12
11
Temperature ( C )
10
9
8
7
6
5
4
3
100
200
300
400
500
600
700
800
500
600
700
800
Time (min)
(b)
18
16
Wind speed (km/hr)
14
12
10
8
6
4
2
0
2
100
200
300
400
Time (min)
Figure 3.12: Weather condition corresponding to the irradiance fluctuations in Fig.

3.11: (a) temperature, (b) wind speed.
67

3.4.3
Log-Normal Fitting
Small scintillation index values obtained in Subsec. 3.4.2 justify using the log-normal
distribution, given in Eq. (3.20), for channel modelling.
The log-normal fitting can be obtained using the maximum likelihood method [97]
by maximizing the log-likelihood function
2
L(ln
) =
h
N
X
i=1
ln
1
q
2 2
ln h

2
1 2
ln h + 2 ln h
,
exp
2ln
h
h
> 0,
h
= 1,
hhi
(3.40)
is the normalized channel samples given by

where h
= h .
h
hhi
(3.41)
2
It can be shown that the log-gain variance
ln
that maximizes Eq. (3.40) is given
h
by [98]
2
ln
=
h
N
1 X
(ln hi
ln h )2 ,
N i=1
(3.42)
where
ln h
N
1 X
ln hi .
=
N i=1
(3.43)
The RMSE defined in Eq. (3.33) is used to assess the goodness of fit.
Figures 3.13, 3.14, 3.15, and 3.16 show the histogram of the normalized channel
along with the log-normal fit using maximum likelihood estimation during
samples h
clear weather. Figures 3.17 and 3.18 show the channel during foggy and rainy weather,
respectively. For all the figures, the measurement duration is 165 seconds using 224
samples. It is reasonable to assume unchanged weather conditions throughout the
68

165-sec duration and, therefore, the channel statistics are considered fixed. The
2
average channel gain hhi, scintillation index SI
, log-gain standard deviation
ln h ,
and the RMSE are shown along with the weather parameters on each figure.
In clear weather conditions, the good match between the channel histogram and
the log-normal fit is justified by an RMSE on the order of (102 ) even at the highest
2
scintillation index value (SI
= 0.13). During a heavy fog condition that started on
November 8th , 2011 around 8:50 PM, there was no received optical power and all the
obtained samples were zero. Figures 3.17 (a) and (b) show the channel state during
the recovery from the fog. Low average intensity values are observed however the
log-normal distribution is still a good fit. Figures 3.18 (a) and (b) show the channel
measurements during a light rain condition (less than 2.5 mm/hr). The scintillation
index is still low however the log-normal distribution can no longer fit with the channel
histogram as indicated by the high RMSE value.
It can be seen that all the obtained scintillation index values are relatively small
(less than unity). Even in very hot days during July, scintillation index never exceeded
0.15. Such reduced scintillation index can be related to the relatively large area of
the receiver that alleviate the effect of the atmospheric turbulence on the received
signal due to the aperture-averaging effect [28].
69

st
(a) 12:23 AM
Nov 1 , 2011
10
Experiment
LN fit
Average intensity = 113

SI = 0.024008
= 0.15653
ln
Probability density
10
RMSE = 0.057903
Temperature = 8.8 C
Wind Speed = 3.6 km/h
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
Normalized channel gain
Nov 1st, 2011
(b) 3:36 AM
10
Experiment
LN fit

SI = 0.026899
ln = 0.16481
Probability density
10
RMSE = 0.037888
Temperature = 8.5 C
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
Figure 3.13: Example (1) of channel envelope histogram with log-normal fit in clear
weather conditions.
70

Nov 1st, 2011
(a) 12:51 PM
10
Experiment
LN fit
SI = 0.083532
ln = 0.29591
Probability density
10
RMSE = 0.046188
Temperature = 13.5 C
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
Nov 1st, 2011
(b) 5:30 PM
10
Experiment
LN fit
SI = 0.004824
ln = 0.06954
Probability density
10
RMSE = 0.039233
Wind Speed = 5 km/h
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
weather conditions.
71

st
(a) 11:07 PM
Nov 1 , 2011
10
Experiment
LN fit

SI = 0.13043
ln = 0.38958
Probability density
10
RMSE = 0.064474
Temperature = 8.6 C
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
(b) 4:05 AM
Nov 2nd, 2011
10
Experiment
LN fit

SI = 0.041323
ln = 0.20641
Probability density
10
RMSE = 0.050844
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
weather conditions.
72

(a) 9:18 AM
nd
Nov 2 , 2011
10
Experiment
LN fit

SI = 0.005221
ln = 0.072283
Probability density
10
RMSE = 0.070137
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
Nov 2nd, 2011
(b) 12:52 PM
10
Experiment
LN fit
SI = 0.053346
ln = 0.23445
Probability density
10
RMSE = 0.045982
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
weather conditions.
73

Nov 9th, 2011
(a) 6:03 AM
10
Experiment
LN fit
SI = 0.0072249
ln = 0.085359
Probability density
10
RMSE = 0.096874
Temperature = 9.9 C
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
th
(b) 8:15 AM
Nov 9 , 2011
10
Experiment
LN fit
SI = 0.023173
ln = 0.15239
Probability density
10
RMSE = 0.041002
Rain level = 0
10
10
10
0.5
0.5
1.5
2.5
Figure 3.17: Example of channel envelope histogram with log-normal fit during recovery from heavy fog condition.
74

(a) 10:04 PM
nd
Nov 22 , 2011
10
Experiment
LN fit
SI = 0.014363
= 0.13434
ln
Probability density
10
RMSE = 0.43955
Temperature = 1.8 C
Rain level = 2.4 mm/hr
10
10
10
0.5
0.5
1.5
2.5
Nov 23rd, 2011
(b) 12:34 AM
10
Experiment
LN fit
SI = 0.019024
ln = 0.15001
Probability density
10
RMSE = 0.33407
Temperature = 1.9 C
Rain level = 0.9 mm/hr
10
10
10
0.5
0.5
1.5
2.5
Figure 3.18: Example of channel envelope histogram with log-normal fit during light
rain condition.
75

3.5
Markov Modelling
A finite-state Markov chain can be directly applied to model the time-varying behaviour of discrete fading communication channels [102107]. The main concept is to
divide the full range of the received SNR into a finite number of discrete levels. These
levels are arranged in an ascending order where each level is assigned to a state. Each
state can be independently represented by a binary symmetric channel with a certain
error probability that depends on the average SNR in that state. Finally, Markov
transitions are assumed between the channel states.
In a simpler manner, the envelope of the FSO channel, obtained by the channel
measurements, can be modelled using a finite-state Markov chain.
Let S = {s0 , s1 , s2 , ..., sK1 } denotes a finite set of K states and {Sn }, n =
0, 1, 2, ...., be a constant Markov process with stationary transitions, i.e. the transition
probability is independent of the time index n. Therefore, the transition probability
from state sj to state sk is given by
tj,k = Pr(Sn+1 = sk |Sn = sj ),
n = 0, 1, 2, .....,
j, k {0, 1, 2, ..., K 1}.

(3.44)
A K K state-transition probabilities tj,k , j, k {0, 1, 2, ..., K 1}, define the statetransition probability matrix T. For uniquely defining a K-state Markov model, a
K 1 initial state probability vector is required in addition to the T matrix.
The steady state probability pk defines the probability of being at state sk without
any state information at other time indices, i.e.
pk = Pr(Sn = sk ),
n = 0, 1, 2, .....,
76
k {0, 1, 2, ..., K 1}.
(3.45)

A K 1 steady state probabilities pk , k {0, 1, 2, ..., K 1}, define the steady state
probability vector p. This vector is usually used as the initial state probability vector
required for a complete definition of a Markov chain.
To apply Markov modelling to the FSO channel, the full range of the observed
channel samples is divided into K partitions represented by the finite set of states
S = {s0 , s1 , s2 , ..., sK1 }. For simplicity, uniform partitioning is considered, i.e.
hmax hmin
,
K
(3.46)
and the mapping from the channel samples h to the states sk is obtained by
h sk
for
hmin + k h < hmin + (k + 1),

(3.47)
k {0, 1, 2, ..., K 1}.
The channel gain assigned to every state hk will be defined as the average channel
gain within that state, given by
hk =
[hmin + k] + [hmin + (k + 1)]
= hmin + k + .
2
2
(3.48)
For a sufficiently large total number of channel samples N , the elements of T can
be estimated using
Nj,k
Nj,k
tj,k = Pr(Sn+1 = sk |Sn = sj ) = PK1
=
,
Nj
l=0 Nl,j
j, k {0, 1, 2, ..., K 1},

(3.49)
where Nj,k is the number of observed transitions from state sj to state sk and Nj is
the total number of times observing the channel envelope at state sj . In a similar
77

j
j
j
j
j
j
j
j
=0
=1
=2
=3
=4
=5
=6
=7
k=0 k=1 k=2 k=3 k=4 k=5 k=6 k=7

0.9721 0.0279
0
0
0
0
0
0
0.0007 0.9828 0.0165
0
0
0
0
0
0
0.0033 0.9876 0.0091
0
0
0
0
0
0
0.0076 0.9875 0.0049
0
0
0
0
0
0
0.0131 0.9837 0.0032
0
0
0
0
0
0
0.0170 0.9802 0.0027
0
0
0
0
0
0
0.0229 0.9751 0.0020
0
0
0
0
0
0
0.0237 0.9763
Table 3.2: State-transition probability matrix of the Markov chain model when K = 8.
p0
p1
p2
p3
p4
p5
p6
p7
0.0017397
0.0675381
0.3397615
0.4060296
0.1519208
0.0292364
0.0034790
0.0002945
Table 3.3: Steady state probability vector of the Markov chain model when K = 8.
manner, the elements of p are estimated by
Nk
pk = Pr(Sn = sk ) = PK1
l=0
Nl
Nk
,
N
k {0, 1, 2, ..., K 1},
(3.50)
where N is the total number of samples.

For the sake of illustration, a group of N = 224 channel samples obtained during
a 165-sec measurement period with sampling rate 101.8 kSa/s are used to derive the
Markov model. These samples were measured on November 1st , 2011 at 07:41 AM.
The full range of the samples (hmin = 39, hmax = 201) is divided uniformly into
K = 8 partitions. The entries of T are obtained by Eq. (3.49) and shown in Table
3.2, while the entries of p are obtained by Eq. (3.50) and shown in Table 3.3.
78

t0,1
s0
t 0,0
t1,2
s1
t1,0
t1,1
t 2,3
s2
t2,1
t2,2
t6,7
s7
t 3,2
t7,6
t7,7
Figure 3.19: State diagram of the Markov model for the channel envelope when K = 8.
It can be seen from Table 3.2 that, for any state, the probability of transition
tj,k , j 6= k, is always less than the no-transition probability tj,j (which is higher
than 97%) due to the slow-varying nature of the channel. Moreover, these transitions
happen only between adjacent states, i.e. tj,k = 0, |j k| > 1, which is a typical
property of finite-state Markov models for slow-fading channels [103].
Figure 3.19 illustrates the state diagram of the resulting Markov model when
K = 8.
A more accurate model can be obtained by increasing the number of states. The
full range of the channel samples is divided into K = 64 states. The non-zero elements
of the resulting T and the entries of p are given in Appendix B.
To verify the model strength, the obtained Markov models are simulated to generate two sets of 224 channel realizations to be compared against the channel measurements. Figure 3.20 shows the histograms of the measurements and the samples
generated by the Markov models. The good match between the resulting channel
distributions is noticed. Table 3.4 shows the average channel gain and scintillation
index obtained from the measurements and simulation.
Figure 3.21 shows the ability of the Markov model to generate channel realizations
with autocovariance that fits well with the autocovariance of the channel measurements when K = 64, however it fails when K = 8.
79

10
Experimental
Markov model (K = 64)
2
10
Probability density
10
10
10
10
10
20
40
60
80
100
120
140
160
180
200
220
Channel realization
Figure 3.20: The histograms of the channel realizations obtained from the measurements and the realizations generated by the Markov model (samples measured on
November 1st , 2011 at 07:41 AM).
hhi
2
SI
Experimental Markov model (K = 64) Markov model (K = 8)

104.7207
104.6221
104.5860
0.0298
0.0300
0.0322
Table 3.4: Comparison between the channel statistics obtained from the measurements and the Markov model.
80

1.2
Experimental
Normalized autocovariance
0.8
0.6
0.4
0.2
0.2
100
80
60
40
20
20
40
60
80
100
Time (msec)
Figure 3.21: Normalized autocovariance of the channel realizations obtained from the
measurements and generated by the Markov model.
81

The comparison justifies that the obtained Markov model gives a good approximation to the channel behaviour and can be used for generating channel realizations for
simulation purposes. The Markov model is simple and computationally-efficient since
it is completely defined by a sparse matrix and a vector. Such a model can be used to
accurately generate different FSO channel realizations with correct distribution and
autocorrelation.
3.6
Conclusions
In this chapter, a practical channel model for IM/DD FSO communications links
was proposed and experimentally verified using the experimental FSO link. The
harmonic distortion, frequency response, noise measurements, and channel loss were
measured. For characterizing the turbulence behaviour of the FSO channel, a new
measurement technique was developed to offer more accuracy and immunity against
the noise inherent in any measurement system. Results indicated a low-turbulence
regime and a good fit with the log-normal distribution under clear and foggy weather
conditions. Finally, a finite-state Markov model was obtained for the channel as
a computationally-efficient tool to generate FSO channel realizations with correct
distribution and autocorrelation for simulation purposes.
In Chapter 4, the experimental setup will be employed for an experimental investigation of the reliability of OFDM transmission over weak-turbulence FSO channels
by assessing the performance in terms of symbol-error rate.
82
Chapter 4
In-Field Demonstration of OFDM
over FSO
4.1
Introduction
In this chapter, the reliability of OFDM transmission over FSO channels is experimentally investigated. Reliability is quantified in terms of allowed transmission rates
and resulting symbol-error rate (SER) for a given power budget. A transceiver design
is proposed, implemented, and field-tested over the 1.87-km FSO link. Transmission
rates up to 300 Mbps are achieved. The received signals are analyzed on the symbol level and constellation diagrams. Error rates are presented in clear and rainy
weather conditions. The results demonstrate the potential role of RoFSO systems in
general, and OFDM over FSO particularly, for solving the back-haul and last-mile
connectivity problems.
In Sec. 4.2, a brief review of OFDM systems highlighting their advantages, drawbacks, and feasibility for transmission over FSO channels is introduced. In Sec. 4.3,
83

Chapter 4. In-Field Demonstration of OFDM over FSO
exp(j2pfsc 1)
exp( j2pfsc 1)
Data
symbols
x(t)
Serial to
Prallel
Channel
Parallel to
Serial
Estimated data
symbols
exp(j2pfsc 2)
exp( j2pfsc 2)
.
.
exp(j2pfsc N)
exp( j2pfsc N)
Figure 4.1: Baseband multi-carrier communications system.

three field-experiments over the FSO link are explained in detail where SER is investigated as the performance metric.
4.2
4.2.1
OFDM
Background
Modulation schemes can be broadly classified into two main categories: single-carrier
transmission and multi-carrier transmission. In single-carrier systems, a serial
stream of the information symbols directly modulates a single carrier, usually a sinusoid, whose frequency fc is suitable for the targeted communications channel. On the
other hand, in multi-carrier systems, the data stream is split into lower rate parallel
streams each modulating its own carrier, usually termed a subcarrier, as shown in
Fig. 4.1.
Mathematically, a baseband multi-carrier signal x(t) can be expressed as [18, 108
111]
x(t) =
X
Nsc
X
cki pk (t iTs ) ej2fk t ,
i= k=1
84
(4.1)

where cki is the i th data symbol at the k th subcarrier and is complex in general, Nsc
is the total number of subcarriers, fk is the frequency of the k th subcarrier, Ts is the
symbol period, and pk (t) is the pulse shaping function of the k th subcarrier. Multicarrier transmission divides the wide-band signal into several narrow-band frequency
division multiplexed signals that are transmitted instantaneously. A multi-carrier
system can be regarded as a modulation or multiplexing technique. For multiplexing, subcarriers are modulated by different data symbols from different sources then
transmitted over the same channel targeting corresponding receivers. Multi-carrier
systems waste the bandwidth if the modulated subcarriers have non-overlapping spectra. An efficient bandwidth utilization is achieved by allowing the data signals across
the adjacent subcarriers to overlap in frequency. If the frequency separation between
every two subcarriers satisfies the orthogonality condition,
fk fl =
n
,
Ts
1 l, k Nsc ,
(4.2)
then the overlapped data signals can be resolved at the receiver using a bank of
correlators matched to the subcarriers. Such a modulation or multiplexing technique
is called orthogonal frequency division multiplexing (OFDM) [112, 113].
OFDM owes its popularity to two major discoveries. The first was the realization
that OFDM modulation and demodulation are mathematically equivalent to the inverse discrete Fourier transform (IDFT) and the discrete Fourier transform (DFT),
respectively [114]. The second was the introduction of the FFT as an efficient algorithm for calculating the DFT [115].
The baseband OFDM signal x(t) in Eq. (4.1) is complex in general. Ultimately,
85

the transmitted signal should be real-valued. Therefore, a complex-to-real conversion is necessary. The standard procedure is the utilization of an IQ modulator for
frequency up-conversion,

s(t) = Re x(t)ej2fc t = Re {x(t)} cos (2fc t) Im {x(t)} sin (2fc t),
(4.3)
where s(t) is the real-valued up-converted signal, fc is the carrier frequency, and Re{}
and Im{} are the real and imaginary parts of a complex value, respectively.
Hermitian symmetry is another technique to enforce the baseband signal to be
real-valued before transmission. Such a technique is valid even if no frequency upconversion is required. A sequence of N complex values ck which satisfy the condition
cN k = ck ,
0 k N,
(4.4)
where denotes the complex conjugate, is said to be Hermitian symmetric. If these

complex values are arranged along a frequency axis representing spectral components,
then Hermitian symmetry requires that the coefficients along the lower half and upper half of the frequency axis are complex-conjugate to each other. From Fourier
transform properties, such a sequence has a real-valued IDFT. Hermitian symmetry eliminates the need for IQ modulators and demodulators and enables real-valued
baseband or bandpass transmission of OFDM signals. However, from a bandwidth
perspective, there is no real difference between the IQ and Hermitian symmetry techniques. In both cases, two times the bandwidth of the complex baseband signal are
required for real-valued transmission.
In modern communication modems, the availability of high-speed DACs/ADCs
86

with sampling rates at several giga samples per second at the front end of the transmitters/receivers enable direct up-conversion/down-conversion from/to the digital domain instead of conventional analog IQ modulators/demodulators. Such a concept
greatly simplifies the system and chip design by keeping all the signal processing
in the digital domain and eliminating the normal difficulties associated with analog
components in mixed-signal chip design. The transceiver described in Chapter 2 is an
all-digital direct up-conversion/down-conversion system. In such systems, the use of
Hermitian symmetry for composing real-valued OFDM symbols is very efficient from
a computational perspective.
4.2.2
Advantages
The key advantage of OFDM systems is the ability to remove inter-symbol interference (ISI) by inserting guard intervals between the data symbols. To obtain zero ISI,
the condition tg > td should be satisfied, where tg is the guard interval width and td
is the channel effective delay time. In dispersive channels, different subcarriers are
delayed by different delays. Such frequency-dependent delay destroys the orthogonality between the subcarriers resulting in inter-carrier interference (ICI). To keep the
orthogonality, the guard interval is filled with an exact replica taken from the end of
the OFDM symbol waveform, termed as the cyclic prefix (CP). To deal with residual
ISI, OFDM systems may need at most a one-tap equalizer for each subcarrier [110].
Other advantages of OFDM systems include the ability to eliminate narrow-band
interference by suppressing affected subcarriers. The tightness of the spectral components of OFDM symbols permits efficient utilization of the available spectrum due to
87

the ability to reduce the guard bands between adjacent channels. In addition, the design of the spectral filters at the transmitter is greatly simplified. Moreover, the guard
bands can be totally removed between adjacent channels as long as the orthogonality condition in Eq. (4.2) is maintained between any two subcarriers across different
channels, what is termed orthogonal-band multiplexed OFDM [18]. Orthogonal-band
multiplexed OFDM is typically used in the conducted field experiments described in
Sec. 4.3.
On the other hand, OFDM suffers from two major drawbacks, high peak-toaverage power ratio and sensitivity to frequency and phase offsets.
4.2.3
Peak-Average Power Ratio
The OFDM symbol, seen in the time domain, consists of the summation of complex
sinusoids with different frequencies whose amplitude and phases are determined solely
by the corresponding data symbols being transmitted, as seen from Eq. (4.1). Since
information symbols are random in general, there is a possibility that the peaks of
many subcarriers align at a certain instant and add together resulting in large peaks.
The peak-average power ratio (PAPR) is defined by

max |s(t)|2
max |s(t)|2

=
,
PAPR =
2
E |s(t)|2
0 t Ts ,
(4.5)
where 2 is the average power. The worst case happens when all the Nsc subcarriers
add together at a certain instant giving the maximum nominal value for the PAPR,
i.e. PAPRmax = Nsc . For example, if Nsc = 128, then PAPRmax = 21 dB. In practice,
no power amplifier can keep linear operation across such a wide dynamic range. In
addition, for all-digital direct up-conversion/down-conversion systems, the dynamic
88

range of the DAC/ADC is capped by a limited number of bits. Fortunately such
high peaks occur rarely through the symbol duration. Figure 4.2 (a) shows a typical probability density distribution of the normalized amplitude of a single OFDM
symbol with Nsc = 40000 along with a Gaussian fit. For such a large number of
subcarriers, and assuming random data symbols, the central limit theorem justifies a
Gaussian distribution for the signal amplitude during the symbol period. The cumulative distribution function in Fig. 4.2 (b) shows that the probability Pr {|s(t)| > 2}
is less than 4%. This low probability suggests a simple solution of clipping the peaks,
however, at the cost of reduced SNR at the receiver.
4.2.4
Synchronization
A reliable synchronization scheme between the transmitter and receiver is necessary

for any digital communication system. The receiver should be able to extract the
timing, frequency, and phase information from the transmitted signal [108]. The
relatively long duration of OFDM symbols, compared to single-carrier systems, makes
it less prone to ISI caused by timing offsets, however, it makes the demodulation more
prone to frequency and phase offsets. These offsets can destroy the orthogonality
among the subcarriers resulting in ICI and making the subcarriers unresolvable.
Several levels of synchronization are required for successful OFDM reception
[18, 108, 109]. (1) Carrier phase and frequency synchronization is a fundamental
requirement for any coherent communication system for successful down-conversion
of the received signal. When all-digital direct up-conversion and down-conversion are
used, the sampling frequencies of the DAC and ADC should satisfy the Nyquist sampling theorem, i.e. they should be at least two times the highest frequency component
89

(a)
10
Transmitted samples
Gaussian fit
1
Probability density function
10
10
10
10
= 0.0015808
= 1.2528
6
Normalized transmitted signal amplitude
(b)
100
Cumulative distribution function (%)
90
80
70
60
50
40
30
20
10
0
5
Normalized amplitude /
Figure 4.2: Statistics of the amplitude of a typical OFDM symbol with 40000 subcarriers.
90

in the transmitted and received signal. In practice, the sampling frequency should
be higher than the factor of two to alleviate the effect of clock jitter, i.e. frequency
deviation from pure periodicity, on the transmitted or received signals. Although
the sampling clock frequencies at the transmitter and receiver need not be equal,
both should be locked to some reference phase and frequency to preserve the timing,
frequency, and phase information. The simplest way to satisfy such condition is the
use of the same sampling frequency at the transmitter and receiver while the receiver
sampling clock being phase-locked to its counterpart at the transmitter.
In the available setup, the sampling clock sources at the transmitter and receiver
are operating in the free-running mode. However, since they have ultra low-noise and
since over-sampling is used by a factor of five, the sampling clock jitter is ignored.
(2) Window synchronization is required to align the start and end of every OFDM
symbol to avoid inter-symbol interference before sampling and applying the FFT.
Window synchronization is relatively simple for OFDM due to longer symbol duration
and the insertion of guard intervals.
For the available setup, to mitigate the absence of timing information at the receiver, a conceptually simple method is used to recover the timing information using
cross-correlation, as shown in Fig. 4.3. A single OFDM symbol is transmitted and
repeated continuously. At the receiver, the incoming signal is sampled continuously
and buffered for off-line processing on a PC. The duration of the OFDM symbol is determined from the number of samples assuming exactly matched sampling frequencies
at the transmitter and receiver. Since the transmitted signal is a periodic repetition
of the same symbol, capturing the correct number of samples at any instant will ensure the inclusion of a full OFDM symbol, however the start and end of the symbol
91

Transmitted samples
0.5 ns
Total symbol duration (OFDM + guard interval)
. m 1
m
m
m+1
m+1
.
.
N
N
3
2
. m 1
3
m+1
. m 1
Timing error < 0.25 nsec
Received samples
Figure 4.3: OFDM symbol timing recovery by the means of cross-correlation.

are unknown. The received time samples are cross-correlated with two replicas of
the known transmitted samples, which is equivalent to circular cross-correlation.
Although such a technique is very expensive from a computational perspective, it
guarantees perfect OFDM symbol timing with negligible timing error less than half
the sampling period. Figure 4.4 shows a typical result of the cross-correlation of
an OFDM symbol with duration 262.144 sec and 40000 subcarriers. The required
timing adjustment is the time shift corresponding to the peak value of the resulting
cross-correlation, i.e. when the transmitted and received sequences coincide in time.
(3) Subcarrier recovery is required to recover the phase information from the subcarriers after applying the FFT. Phase ambiguity causes the rotation of the received
symbols constellations as shown in Fig. 4.9. Frequency-dependent phase rotation is
caused by dispersive channels and/or the transmitter or receiver non-ideal frequency
response and its mitigation requires channel estimation.
For the OFDM over FSO experiments discussed in Sec. 4.3, the phase ambiguity of
the subcarriers is mitigated using differential-quadrate phase shift keying (DQPSK)
to modulate the subcarriers instead of QPSK. Therefore, information symbols are
encoded in the phase differences instead of the absolute phases. The major drawback
of DQPSK is the increase in SER due to error propagation.
92

9
12
x 10
10
Crosscorelation
4
600
400
200
200
400
600
Time shift (sec)
Figure 4.4: Typical cross-correlation between the transmitted waveform of an OFDM

symbol and the received version.
4.2.5
OFDM for FSO Systems
Because of design simplicity and the availability of relatively inexpensive components,

almost all the existing FSO systems are IM/DD. In such systems, the message signal
is transmitted by the means of directly modulating the intensity of a laser source. By
definition, optical intensity cannot be negative, while the transmitted signal is bipolar
in general. Therefore, in IM/DD RoFSO systems, a non-negativity constraint should
be applied on the RF waveform before the electro-optical conversion.
For the case of OFDM signals, the simplest way to force the RF signal into nonnegative values is the addition of a fixed DC bias , i.e.
sDC (t) = s(t) + ,
93
sDC (t) 0,
(4.6)

where sDC (t) is the DC-biased unclipped OFDM signal and s(t) is the real-valued
OFDM signal given by Eq. 4.3.
Because of the large PAPR, the value of K may be prohibitively high. This
problem is solved by applying the DC bias after clipping the OFDM signal, i.e.
sDC (t) = s(t) + ,
sDC (t) 0,
(4.7)
where sDC (t) is the DC-biased clipped OFDM signal and s(t) is the real-valued clipped
OFDM signal given by
|A|
s(t) > |A|
s(t) =
s(t) |A| s(t) > |A| ,
|A| |A| > s(t)
(4.8)
where A is the clipping threshold.

Other techniques exist to satisfy the non-negativity constraint without DC biasing.
In asymmetrically-clipped optical OFDM [116, 117], only odd subcarriers are loaded
with symbols information while the even subcarriers are suppressed to zero, resulting
in an asymmetric time domain signal. Since all the information can be recovered
from the positive portion, the negative portion is suppressed to zero and only the
positive values are transmitted. Such scheme is efficient from a power perspective
however at the cost of reduced SNR, increased PAPR, and loss of half the transmission
bandwidth.
94

4.3
OFDM over FSO Field Experiments
To prove the feasibility of OFDM transmission over FSO channels, many in-field
experiments have been conducted in August, September, and November 2011. Three
experiments are outlined in this section. The first two have been conducted in clear
weather conditions to highlight the improvement in performance achieved by clipping
the OFDM signal. The third experiment is intended to simulate the transmission of
20 terrestrial digital video broadcasting (DVB-T) channels in light rain conditions to
assess the performance against the error rate requirements for the well-established
DVB-T standard.
4.3.1
DC-Biased Unclipped OFDM over FSO
An example of a DC-biased unclipped OFDM over FSO experiment is comprehensively described in this section. Transmission specifications as well as reception parameters and weather conditions are given in Table 4.1.
The experiment was started on November 6th , 2011 at 01:07 PM under clear
weather conditions (visibility: 16 km, temperature: 16.7 C, wind speed: 14 km/hr).
Twenty orthogonal-band multiplexed OFDM channels are transmitted over the link.
Each channel is composed of 2000 subcarriers that occupy a 7.81-MHz bandwidth.
The total transmission bandwidth is 156.25 MHz starting from 20 MHz as illustrated
in Fig. 4.5.
Figure 4.6 shows a block diagram of the transmitter. On a PC using MATLAB,
an OFDM symbol with duration 256 sec is composed. Using a 2-GHz sampling
frequency, 512000 time samples are required. By arbitrarily using the same number
of samples in the frequency domain, the sampling frequency axis is divided into
95

No. of channels
No. of subcarriers / channel
Channel bandwidth
Spacing between channels
Modulation
Total data rate
OFDM symbol duration
Cyclic prefix duration
Total symbol duration
Total transmission bandwidth
Sampling rate
IFFT length
No. of received OFDM symbols
Transmission duration
Transmitter
Average transmitted AC power / channel
Average total transmitted AC power
Laser DC current
Average laser AC current
Average transmitted optical power / channel
Average total transmitted optical power
Receiver
Average received optical power / channel
Average total received optical power
Average TIA output power /channel
Average total TIA output power
Average SER
Environmental conditions
Time & date
Visibility
Temperature
Wind speed (average)
Absolute pressure
Humidity
20
2000
7.81 MHz
0 MHz
DQPSK (2 bits/symbol)
152.59 M symbol/s (305.18 Mbps)
256.00 sec
6.14 sec
262.14 sec
156.25 MHz (20 - 176.25 MHz)
2 GSa/sec
512000
20 36864
9.66 sec
6.41 W (21.93 dBm)
128.21 W (8.92 dBm)
167 mA
54.6 mA
3 mW (4.77 dBm)
60 mW (17.78 dBm)
0.29 W (35.38 dBm)
5.90 W (22.29 dBm)
4.84 nW (53.15 dBm)
96.78 nW (40.14 dBm)
1.34 104
1:07 PM - 3:05 PM, Nov 6th , 2011
16 km [99]
16.7 C
14 km/hr
1006.5 hPa
42%
Table 4.1: Transmission specifications and reception parameters of the DC-biased

unclipped OFDM over FSO experiment.
96

2000
subcarriers
Ch. 1
20 MHz
Ch. 2
Ch. 3
Ch. 20
176.25 MHz
27.81 MHz
Figure 4.5: Spectrum of the 20 OFDM channels transmitted in the DC-biased unclipped OFDM over FSO experiment.
512000 segments, each with 3.91 kHz bandwidth. Each channel has 2000 DQPSK
symbols which are the complex frequency coefficients that modulate the subcarriers.
These coefficients are arranged along the frequency axis starting from 20 MHz up
to 176.25 MHz, then Hermitian symmetry is applied. A 512000-IFFT is applied to
the entire frequency coefficients to obtain the real-valued time samples. Then 12288
time samples are inserted for the cyclic prefix. The samples are rescaled to occupy
the full range of the on-board DAC (2048 to 2047), i.e. max {|s(t)|} = DACmax ,
where s(t) is the real-valued unclipped OFDM signal before adding the DC bias and
DACmax = 400mV is the full range of the DAC as given in Chapter 2. Finally,
the time samples are downloaded to the board. Figures 4.7 (a) and (b) show the time
signal and associated PSD, respectively.
On the board, the memory is read continuously to feed the DAC which is sampled
at 2 GSa/s. The DAC output is connected to the analog laser driver where DC bias
is added before transmission over the link.
The block diagram of the receiver is shown in Fig. 4.8. After opto-electrical
conversion, the signal is sampled at the rate 2 GSa/s and the resulting samples are
stored on the on-board memory. For this experiment, 6 3 = 18 GBytes samples
were buffered, corresponding to 9.66 sec effective transmission time. Every 3 GBytes
samples represent 1.61 sec continuous reception before entering a 24-min idle period
to download the samples to the PC before capturing a new set of samples. The
97

PC (Matlab)
2000 DQPSK
Ch. 1
symbols
15.26 Mbps Serial to
parallel
2000 DQPSK
Ch. 2
symbols
parallel
.
.
.
.
Ch. 20
2000 DQPSK
symbols
parallel
Digitizer board
512000
IFFT
Cyclic prefix
insertion
3 GBytes
memory
FSO transmitter
DAC
Analog driver
& laser
2 GHz
DC bias
Telescope
cN-k = ck*
(Hermitian symmetry)
Figure 4.6: Block diagram of the implemented OFDM over FSO transmitter.
9.66-sec time period represents 20 36864 = 737280 OFDM symbols across the 20
channels, or equivalently, a total transmission of 1.51 109 DQPSK symbols. Timing
information is recovered by the means of the cross-correlation described in Subsec.
4.2.4.
Figures 4.7 (c) and (d) show the received time samples and their PSD. It can be
seen that lower frequency channels have, in general, higher SNR due to the bandwidth
limitations of the system which acts as a low-pass filter as was shown in Chapter 3.
The cyclic prefix is removed from the adjusted time samples before they undergo a
512000-FFT. Finally, the 2000 coefficients of every channel are obtained for DQPSK
demodulation. The received constellations of four channels are shown in Fig. 4.9
illustrating the effect of the atmospheric turbulence and noise on the constellations.
In addition, the phase ambiguity is evident which is largely mitigated by the use of
98

(a)
(b)
0
400
Transmitted waveform (mV)
300
200
100
0
100
200
300
400
10
20
30
40
50
60
70
80
100
Time (sec)
200
(c)
500
400
500
10
Received waveform (mV)
400
(d)
15
10
15
300
Frequency (MHz)
5
10
15
20
25
30
35
40
Time (sec)
100
200
300
Frequency (MHz)
Figure 4.7: DC-biased unclipped OFDM over FSO: (a) transmitted signal, (b) PSD
of the transmitted signal, (c) received signal, (d) PSD of the received signal.
99

FSO receiver
Telescope
Photodetector &
trans-impedance
amplifier
Digitizer board
ADC
Memory
PC (Matlab)
Time samples
Symbol timing
recovery
Cyclic prefix
removal
Transmitted
samples
2 GHz
Ch. 1
15.26 Mbps Parallel
to serial
DQPSK
demodulator
Ch. 2
15.26 Mbps Parallel
to serial
DQPSK
demodulator
Ch. 20
15.26 Mbps Parallel
to serial
DQPSK
demodulator
2000
coefficients
2000
coefficients
.
512000
.
FFT
.
.
2000
coefficients
Figure 4.8: Block diagram of the implemented OFDM over FSO receiver.
DQPSK.
Figures 4.10 (a) and (b) show the SER per subcarrier and per channel, respectively.
Higher frequency channels have higher error rates due to the low-pass effect of the
system. The average SER across all the channels is 1.34 104 . The ripples in the
SER in Fig. 4.10 (a) can be related to the corresponding ripples in the received PSD
in Fig. 4.7 (d). Such ripples are due to the non-flat frequency response of the system.
It can be seen that some subcarriers have relatively high error rates, which suggests
adaptive power loading of each subcarrier to improve the performance.
For the above transmission system, there is no difference between visualizing the
transmitted signal as 20 orthogonal-band multiplexed OFDM symbols or just considering it a single broadband OFDM symbol with 40000 subcarriers.
100

Channel 7 (66.9 ~ 74.7 MHz)

1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
Imaginary
Imaginary
Channel 1 (20 ~ 27.8 MHz)

1
0
0.2
0
0.2
0.4
0.4
0.6
0.6
0.8
0.8
1
1
0.5
0.5
1
1
0.5
Real
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0.2
0.2
0.4
0.6
0.6
0.8
0.8
0
0.4
0.5
0.5
Channel 20 (168.4 ~ 176.2 MHz)

1
Imaginary
Imaginary
Channel 13 (113.8 ~ 121.6 MHz)

1
1
1
Real
0.5
1
1
Real
0.5
0.5
Real
Figure 4.9: Received constellations for DC-biased unclipped OFDM over FSO.
101

(a)
Symbol error rate / subcarrier
10
10
10
10
10
20
40
60
80
100
120
140
160
Frequency (MHz)
(b)
10
Symbol error rate / channel
SER per channel

Average SER
10
10
10
12
14
16
18
20
OFDM channels
Figure 4.10: Symbol-error rates for DC-biased unclipped OFDM over FSO measured
over 9.66 sec transmission time under clear weather conditions: (a) per subcarrier,
(b) per channel.
102

4.3.2
DC-Biased Clipped OFDM over FSO
In the previous experiment, the signal applied to the DAC was not clipped which
limited the electrical power applied to the laser driver and relatively high error rates
were obtained. In this experiment, the OFDM signal is clipped at |A| = 2 using
Eq. (4.8), where is the standard deviation of the unclipped signal. The clipped
signal is rescaled to occupy the full range of the DAC (2 = 400 mV). Clipping
and rescaling permits 8 dB more electrical power to be applied to the laser driver
via the DAC. Transmission specifications for this experiment are highlighted in Table
4.2. The experiment was started on November 6th , 2011 at 01:07 PM under clear
weather conditions (visibility: 16 km, temperature: 15.4 C, wind speed: 9 km/hr).
The effective transmission time is 9.66 sec. The transmission bandwidth, number of
channels, and the OFDM symbol parameters are the same as in Table 4.1.
Figure 4.11 (a) shows the time waveform of the clipped signal with the same
vertical scale used in Fig. 4.7 (a). Clipping noise is obvious in the PSD in Fig. 4.11
(b) as compared to the unclipped version in Fig. 4.7 (b).
The received constellation diagrams and resulting SER are shown in Fig. 4.12 and
4.13, respectively. The average SER is 1.23 106 , two orders in magnitude better
than the unclipped transmission mainly due to the increase in the average received
power. It can be seen that 15 channels have SER less than 106 .
103

Transmitter
Clipping threshold
Average transmitted AC power / channel
Average total transmitted AC power
Average laser AC current
Receiver
Average TIA output power /channel
Average total TIA output power
Average SER
Time & date
Visibility
Temperature
Absolute pressure
Humidity
2
36.8 W (14.34 dBm)
736 W (1.33 dBm)
133 mA
26.59 nW (45.75 dBm)
531.80 nW (32.74 dBm)
1.23 106
3:54 PM - 5:51 PM, Nov 6th , 2011
16 km [99]
15.4 C
9 km/hr
1005.8 hPa
44%
Table 4.2: Parameters and conditions of the DC-biased clipped OFDM over FSO
experiment (transmission bandwidth, number of channels, and OFDM specifications
are given in Table 4.1.)
104

(a)
(b)
0
400
Transmitted waveform (mV)
300
200
100
0
100
200
300
400
0.5
1.5
5
10
15
20
25
30
35
40
45
2.5
100
Time (sec)
200
(c)
500
400
500
15
Received waveform (mV)
400
(d)
20
10
5
0
5
10
15
20
300
Frequency (MHz)
0.5
1.5
5
10
15
20
25
30
35
40
2.5
Time (sec)
100
200
300
Frequency (MHz)
Figure 4.11: DC-biased clipped OFDM over FSO: (a) transmitted signal, (b) PSD of
the transmitted signal, (c) received signal, (d) PSD of the received signal.
105

Channel 7 (66.9 ~ 74.7 MHz)

1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
Imaginary
Imaginary
Channel 1 (20 ~ 27.8 MHz)

1
0
0.2
0
0.2
0.4
0.4
0.6
0.6
0.8
0.8
1
1
0.5
0.5
1
1
0.5
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0.2
0.2
0.4
0.6
0.6
0.8
0.8
0
0.4
0.5
0.5
Channel 20 (168.4 ~ 176.2 MHz)

1
Imaginary
Imaginary
Channel 13 (113.8 ~ 121.6 MHz)

1
1
1
Real
Real
0.5
1
1
0.5
0.5
Real
Real
Figure 4.12: Received constellations for DC-biased clipped OFDM over FSO.
106

(a)
10
10
10
10
10
20
40
60
80
100
120
140
160
Frequency (MHz)
(b)
10
SER per channel

Average SER
10
10
10
10
12
14
16
18
20
OFDM channels
Figure 4.13: Symbol-error rates for DC-biased clipped OFDM over FSO measured
over 9.66 sec transmission time under clear weather conditions: (a) per subcarrier,
(b) per channel.
107

4.3.3
DVT-B over FSO
Digital video broadcasting - terrestrial (DVB-T) refers to the European standard for
terrestrial digital video broadcasting that was first launched in 1997 [118]. In DVB-T,
MPEG-compressed digital video and audio are transmitted using coded OFDM. Two
modes were defined for the DVB-T standard depending on the number of subcarriers
in the OFDM symbol, the 2k mode and 8k mode. The 2k mode has 1705 subcarriers
with 4 kHz separation while the 8k mode has 6817 subcarriers each is 1 kHz apart.
Typical modulation schemes for the subcarriers are QPSK, 16-QAM, and 64-QAM.
DVB-T uses a concatenated forward error correction (FEC) coding system. The
internal code is a convolutional code with rates from 1/2 to 7/8, while the external
one is a Reed-Solomon RS (204, 188) code. The FEC limit for this coding system is
2 104 which results in a decoded probability of error of 1011 [119].
In this experiment, the signals transmitted over the FSO link are intended to represent the outputs of the DVB-T modulator. Transmission specifications and weather
conditions are outlined in Table 4.3. The experiment was started on November 9th ,
2011 at 05:12 PM during a light-rain weather condition (visibility: 13 km, temperature: 14.2 C, wind speed: 8 km/hr, rain level: 1.5 mm/hr). The effective transmission
time is 9.66 sec and the clipping threshold for the OFDM signal is 2.
Figure 4.14 shows the SER across the 20 DVB-T channels. The FEC coding limit
is shown on the same figure for comparison. It is obvious that 19 channels out of 20
satisfy the 2 104 error threshold.
Because of the limited power budget, the attempt to transmit 20 channels using
higher-order constellations was not successful even under clear weather conditions
since the obtained constellation diagrams were completely unresolvable. However,
108

No. of channels
No. of subcarriers / channel
Channel bandwidth
Spacing between channels
Modulation
Data symbols / OFDM symbol / channel
OFDM symbol duration
Guard interval duration
Clipping threshold
Total transmission bandwidth
Average total received optical power
Average SER
Time & date
Visibility
Temperature
Absolute pressure
Humidity
Rain level
20
2048
7.61 MHz
1.53 MHz
DQPSK
1706
224 sec
28 sec
2
182.86 MHz (20 - 202.86 MHz)
4.60 W (22.29 dBm)
4.82 105
5:12 PM - 7:11 PM, Nov 9th , 2011
13 km [99]
14.2 C
8 km/hr
989.6 hPa
97%
1.5 mm/hr
Table 4.3: Specifications of the DVB-T over FSO experiment.
109

(a)
10
10
10
10
10
20
40
60
80
100
120
140
160
180
200
Frequency (MHz)
(b)
10
10
10
10
10
SER per channel

Average SER
FEC coding limit
10
10
12
14
16
18
20
DVBT channels
Figure 4.14: Symbol-error rates for 20 DVB-T channels transmitted over the FSO
link for 9.66 sec during a light-rain condition: (a) per subcarrier, (b) per channel.
110

(b)
1
0.5
0.5
Imaginary
Imaginary
(a)
1
0.5
0.5
1
1
0.5
0.5
1
1
0.5
0.5
Real
Real
Figure 4.15: Received constellations for higher-order QAMs under clear weather conditions: (a) 16-QAM, (b) 64-QAM.
transmitting higher-order constellations was possible by limiting the number of channels and locating them at the lower frequency region where attenuation has its lowest
levels. Figure 4.15 (a) shows the received constellations of the first channel out
of four 16-QAM DVB-T channels. All the power was allocated to these channels
(transmitted on September 14th , 2011 at 06:23 PM under clear weather conditions,
temperature: 21.1 C, wind speed: 5 km/hr). The clipping threshold for the OFDM
signal is 2. The received constellations diagram of the 64-QAM transmission is
shown in Fig. 4.15 (b). It was obtained by allowing only two channels to share all the
laser power (transmitted on September 14th , 2011 at 06:55 PM). The use of 16-QAM
seems promising and low SER can be obtained however it would be hard to maintain
reliable transmission using the 64-QAM unless more optical power at the transmitter
is available.
111

4.4
Conclusions
In this chapter, it was shown by in-field experiments that OFDM over FSO is a promising technology that permits a transparent interface between fiber and RF/coaxial networks. To demonstrate its feasibility, several experiments have been conducted over
the 1.87-km FSO link in different weather conditions. Composed DQPSK-OFDM
symbols were transmitted and the received signals were analyzed on the symbol level
to obtain the constellation diagrams and SER. Uncoded transmission at 300 Mbps
was achieved with average SER on the order of 106 in clear weather. Good performance was also measured during light rain with SER on the order of 105 , however,
the link was unusable during heavy rain or fog. These in-field experiments demonstrate the potential role and limits of OFDM over FSO for back-haul connectivity
and last-mile bottleneck problems.
112
Chapter 5
Conclusions and Future Directions
5.1
Conclusions
Three main contributions have been presented in this thesis. The first one is the development of a hardware system that simplifies the interface between signal processing
softwares such as MATLAB and the analog transceiver of McMaster University FSO
link. An FPGA-based digitizer board with 3 GBytes memory and 2 GSa/s DAC
and ADC is employed as the main hardware platform. A system-on-three-FPGAs
is implemented to use the board as a large high-speed samples buffer. This buffer
enables the transmission of MATLAB-composed waveforms over the FSO link as well
as buffering the received samples before downloading to a PC for off-line processing.
The transmitted waveforms can represent a variety of modulation, signalling, and
coding schemes for investigation over the link. In this thesis, the developed system
was employed in FSO channel measurements and OFDM over FSO transmission.
The second contribution is the realization of an FSO channel measurement system that uses a new approach. The conventional measurement systems transmit a
113

Chapter 5. Conclusions and Future Work
continuos wave optical signal where time samples are captured at the receiver representing the channel realizations. Such systems suffer from reduced SNR due to noise
corruption. In this thesis, an optical wave modulated by a high-frequency (around
100 MHz) sinusoid is transmitted. At the receiver, the incoming signal is sampled at
2 GSa/s. The captured samples undergo FFT that acts as a notch filter around the
transmitted frequency to reduce the accompanying noise. This technique provides
more immunity to the captured samples against corrupting noise. The SNR of the
received samples is approximately 40 dB under clear weather conditions. Using the
developed system, channel measurements are conducted under a variety of weather
conditions. The widely-accepted log-normal distribution for FSO channels in weakturbulence regimes is experimentally verified and good fitting is obtained with the
measurements. Obtained scintillation index values are much less than unity due to
the aperture-averaging effect of the receiver. The coherence time is found to be on the
order of tens of milliseconds showing a slow-fading channel for megabit and gigabit per
second transmissions. A finite-state Markov chain model is derived from the channel
measurements. Numerical simulation results verify the model strength by showing a
good match between the simulated and measured distribution and autocorrelation.
This model can be efficiently used to generate a variety of FSO channel realizations
with accurate scintillation index and autocorrelation for simulation purposes.
The third contribution is the implementation of the first realization of an OFDM
over FSO transmission. In all related previous work, RF signals representing commercial services were being transmitted, and the performance was mainly investigated in
terms of the RF parameters such as carrier-to-noise ratio and channels leakage ratio.
In this work, composed OFDM symbols are transmitted over the FSO link and the
114

received signals are analyzed on the symbol level to obtain the constellation diagrams
and SER. Transmission rates up to 300 Mbps are achieved with average SER on the
order of 106 without coding. These results show the potential role of OFDM over
FSO for integration with RoF networks to fix the back-haul and last-mile connectivity
problems.
5.2
Future Directions
The existing setup along with the developed hardware and software systems are a
major step toward the experimental verification of many theoretical results in FSO
communication systems. A variety of developed modulation, signalling, and coding
techniques are waiting for experimental investigation over a real world FSO link. The
implemented buffering system makes such a step straightforward. Research areas
such as M -ary OOK and M -ary PPM, nonuniform signalling and capacity-achieving
distributions, and coded versus un-coded performance have not been experimentally
investigated at the gigabit per second rate.
The developed FSO channel measurement system will be run continuously throughout the year to produce a large database of channel measurements under different
weather conditions. Such database can be used to build comprehensive statistical
models and empirical formulas that directly relate the weather parameters with the
channel statistics and the performance of the communication link.
The derived finite-state Markov model for the FSO channel is the first step toward
comprehensive simulations using such accurate, simple-to-derive, and computationallyefficient channel model.
115

The addition of a mechanical shutter that covers a portion of the receiver decreasing its effective area may increase the obtained scintillation index values allowing the
investigation of the aperture-averaging effect and channel distributions other than the
log-normal. The gamma-gamma distribution is a good candidate.
At the receiver, there is a gap between the received signal levels (30 mV) and
the available ADC range (250 mV). Such a gap wastes the 3 most significant bits
(MSBs) of the ADC which stick to zero and only 7 bits (6 LSBs + the sign bit) are
effectively used. The insertion of a variable-gain, ultra low-noise RF amplifier at the
receiver between the transimpedance stage and the ADC will allow better utilization
of the valuable ADC range to obtain more accurate results by increasing the number
of levels that represent the analog received signal.
The establishment of a synchronization scheme between the transmitter and receiver is a key requirement for real-time reception. Real-time operation will enable
longterm investigation of the performance of various communication systems over the
FSO channel providing more accurate results and performance measures than the
current buffer-then-process-off-line system.
The developed OFDM over FSO transmission system is a major step for investigating such potential technology. Much work exists to improve the results by investigating higher-order QAM constellations and applying adaptive power loading, adaptive
bit loading, and coded transmission. Low-density parity-check (LDPC) codes and
punctured codes are good candidates to start with since much theoretical work has
been developed in this area.
116
Appendix A
FSO Link Management Software
The SONAbeam TM Terminal Controller (STC) is a software provided by fSONA for
managing the FSO link from a connected PC. The STC software can be run on a PC
whose COM port is connected to the terminal serial console port within the PCA box
via a standard RS-232 interface. Alternatively, the terminal can be controlled over
an Ethernet network through the terminal Ethernet LAN Port. Any PC on the same
network can control the terminal by running the STC software and configuring the
IP address.
Figures A.1, A.2, A.3, and A.4 show snapshots of the software interface window.
The terminal status, such as the power supply voltages and currents, connectivity,
temperature, and humidity, are monitored. Lasers power levels and currents, received power levels, clock recovery and synchronization, and AGC can be adjusted or
monitored from the STC software.
117

Appendix A. FSO Link Management Software
Clock and data recovery

status (digital receiver)
Power supply voltages

and currents status
Lasers currents and

temperatures status
Terminal received
power (W)
Figure A.1: Snapshot (1) from the STC software: terminal status.
118

Lasers power levels
Analog receiver
gain level
Receiver mode
Transmission standard
(digital transmitter)
Figure A.2: Snapshot (2) from the STC software: lasers power levels and receiver
gain level.
119

Terminal received
power (W)
Terminal received
power chart
Figure A.3: Snapshot (3) from the STC software: received power.
120

Figure A.4: Snapshot (4) from the STC software: diagnostics.
121
Appendix B
Finite-State Markov Model
Parameters for K = 64
122

Appendix B. Finite-State Markov Model Parameters for K = 64
t1,0
t2,1
t3,2
t4,3
t5,4
t6,5
t7,6
t8,7
t9,8
t10,9
t11,10
t12,11
t13,12
t14,13
t15,14
t16,15
t17,16
t18,17
t19,18
t20,19
t21,20
t22,21
t23,22
t24,23
t25,24
t26,25
t27,26
t28,27
t29,28
t30,29
t31,30
0.0048232
0.055928
0.021183
0.035108
0.035061
0.043403
0.028469
0.052103
0.026921
0.04988
0.031324
0.055821
0.03288
0.056187
0.036287
0.036555
0.058629
0.038563
0.061132
0.040291
0.063771
0.042925
0.064878
0.045894
0.069203
0.047306
0.071864
0.048793
0.074413
0.051135
0.051116
t0,0
t1,1
t2,2
t3,3
t4,4
t5,5
t6,6
t7,7
t8,8
t9,9
t10,10
t11,11
t12,12
t13,13
t14,14
t15,15
t16,16
t17,17
t18,18
t19,19
t20,20
t21,21
t22,22
t23,23
t24,24
t25,25
t26,26
t27,27
t28,28
t29,29
t30,30
t31,31
0.99222
0.95498
0.89038
0.92586
0.87068
0.91551
0.87457
0.91751
0.87294
0.91898
0.87567
0.91541
0.87681
0.91568
0.87369
0.91567
0.91517
0.87365
0.91499
0.87105
0.91424
0.86678
0.91093
0.86807
0.90822
0.8626
0.90677
0.85737
0.90618
0.85555
0.90389
0.9032
t0,1
t1,2
t2,3
t3,4
t4,5
t5,6
t6,7
t7,8
t8,9
t9,10
t10,11
t11,12
t12,13
t13,14
t14,15
t15,16
t16,17
t17,18
t18,19
t19,20
t20,21
t21,22
t22,23
t23,24
t24,25
t25,26
t26,27
t27,28
t28,29
t29,30
t30,31
t31,32
0.007772
0.040193
0.053691
0.052957
0.094207
0.049434
0.082027
0.054018
0.074954
0.054096
0.074449
0.053264
0.067374
0.051436
0.070119
0.048041
0.048277
0.067725
0.046439
0.067816
0.04547
0.069449
0.046142
0.067048
0.045884
0.068197
0.045922
0.070768
0.045024
0.070037
0.044971
0.045671
Table B.1: Non-zero elements of the state transitions probability matrix of the Markov
chain model when K = 64.
123

t32,31
t33,32
t34,33
t35,34
t36,35
t37,36
t38,37
t39,38
t40,39
t41,40
t42,41
t43,42
t44,43
t45,44
t46,45
t47,46
t48,47
t49,48
t50,49
t51,50
t52,51
t53,52
t54,53
t55,54
t56,55
t57,56
t58,57
t59,58
t60,59
t61,60
t62,61
t63,62
0.077736
0.053725
0.078841
0.055247
0.080915
0.058585
0.082963
0.058135
0.084356
0.059736
0.081321
0.059005
0.083027
0.058284
0.084045
0.059872
0.067996
0.088201
0.067868
0.08182
0.061986
0.11172
0.06264
0.08858
0.063969
0.10155
0.067114
0.075472
0.082495
0.10938
0.074627
0.073569
t32,32
t33,33
t34,34
t35,35
t36,36
t37,37
t38,38
t39,39
t40,40
t41,41
t42,42
t43,43
t44,44
t45,45
t46,46
t47,47
t48,48
t49,49
t50,50
t51,51
t52,52
t53,53
t54,54
t55,55
t56,56
t57,57
t58,58
t59,59
t60,60
t61,61
t62,62
t63,63
0.85281
0.90084
0.85203
0.89947
0.84483
0.89649
0.8466
0.89695
0.84534
0.89579
0.85024
0.89757
0.84549
0.8997
0.84409
0.89304
0.88557
0.84284
0.89467
0.84262
0.88545
0.81696
0.89869
0.82577
0.89666
0.85614
0.87025
0.81402
0.84708
0.79688
0.85821
0.92643
t32,33
t33,34
t34,35
t35,36
t36,37
t37,38
t38,39
t39,40
t40,41
t41,42
t42,43
t43,44
t44,45
t45,46
t46,47
t47,48
t48,49
t49,50
t50,51
t51,52
t52,53
t53,54
t54,55
t55,56
t56,57
t57,58
t58,59
t59,60
t60,61
t61,62
t62,63
0.069457
0.045425
0.06913
0.04527
0.074259
0.044915
0.070439
0.04488
0.0703
0.044422
0.068439
0.043399
0.071488
0.04197
0.071864
0.047087
0.046432
0.068956
0.037365
0.075558
0.052568
0.071325
0.038671
0.085652
0.039366
0.042313
0.06264
0.11051
0.070423
0.09375
0.067164
Table B.2: Non-zero elements of the state transitions probability matrix of the Markov
chain model when K = 64 (continue).
124

p0
p1
p2
p3
p4
p5
p6
p7
p8
p9
p10
p11
p12
p13
p14
p15
p16
p17
p18
p19
p20
p21
p22
p23
p24
p25
p26
p27
p28
p29
p30
p31
2.3007 105
3.7074 105
2.6643 105
6.7532 105
0.00010186
0.0002737
0.00031173
0.00089818
0.0009312
0.0025926
0.0028117
0.0066829
0.0063768
0.013067
0.011962
0.023114
0.030377
0.025012
0.043925
0.033368
0.056163
0.040046
0.06479
0.04608
0.067317
0.044634
0.064347
0.041117
0.059633
0.036082
0.04942
0.043479
p32
p33
p34
p35
p36
p37
p38
p39
p40
p41
p42
p43
p44
p45
p46
p47
p48
p49
p50
p51
p52
p53
p54
p55
p56
p57
p58
p59
p60
p61
p62
p63
0.025551
0.033041
0.01904
0.023829
0.013332
0.0169
0.0091473
0.01108
0.0058944
0.0069357
0.0037894
0.0043963
0.0022987
0.0028205
0.0014092
0.0016924
0.001172
0.00061631
0.00062531
0.00028557
0.00034809
0.00016379
0.0001865
8.142 105
0.00010902
4.226 105
2.6643 105
2.2113 105
2.9624 105
1.9073 105
2.3961 105
2.1875 105
Table B.3: Steady state probability vector of the Markov chain model when K = 64.
125
Bibliography
[1] S. Hranilovic, Wireless Optical Communication Systems. Springer, 2005.
[2] R. Ramirez-Iniguez, S. M. Idrus, and Z. Sun, Optical Wireless Communications.
Taylor & Francis Group, LLC, 2008.
[3] R. D. Dupuis and M. R. Krames, History, development, and applications of
high-brightness visible light-emitting diodes, IEEE Journal of Lightwave Technology, vol. 26, no. 9, pp. 11541171, May 2008.
[4] J. R. Barry, Wireless Infrared Communications.
Kluwer Academic Press,
Boston, MA, 1994.

[5] H. Willebrand and B. S. Ghuman, Free Space Optics: Enabling Optical Connectivity in Todays Networks. Sams Publishing, 2002.
[6] X. Zhu and J. M. Kahn, Free-space optical communication through atmospheric turbulence channels, IEEE Transactions on Communications, vol. 50,
no. 8, pp. 12931300, August 2002.
[7] , Markov chain model in maximum-likelihood sequence detection for freespace optical communication through atmospheric turbulence channels, IEEE
Transactions on Communications, vol. 51, no. 3, pp. 509516, March 2003.
126

Bibliography
[8] M. A. Al-Habash, L. C. Andrews, and R. L. Phillips, Mathematical model for

the irradiance probability density function of a laser beam propagating through
turbulent media, Optical Engineering, vol. 40, pp. 15541562, August 2001.
[9] Z. Xu and B. M. Sadler, Ultraviolet communications: potential and stateof-the-art, IEEE Communications Magazine, vol. 46, no. 5, pp. 6773, May
2008.
[10] A. Harris, J. J. Sluss, H. H. Refai, and P. G. LoPresti, Alignment and tracking
of a free space optical communications link to a UAV, in Proceedings of the
24th Digital Avionics Systems Conference (DASC), October 2005, pp. 1.C.21
to 1.C.29.
[11] D. Giggenbach, J. Horwath, and B. Epple, Optical satellite downlinks to optical ground stations and high-altitude platforms, in The 16th IST Mobile and
Wireless Communications Summit, July 2007, pp. 14.
[12] C. C. Chen and C. S. Gardner, Impact of random pointing and tracking errors
on the design of coherent and incoherent optical intersatellite communication
links, IEEE Transactions on Communications, vol. 37, no. 3, pp. 252260,
March 1989.
[13] A. Sano, E. Yamada, H. Masuda, E. Yamazaki, T. Kobayashi, E. Yoshida,
Y. Miyamoto, S. Matsuoka, R. Kudo, K. Ishihara, Y. Takatori, M. Mizoguchi,
K. Okada, K. Hagimoto, H. Yamazaki, S. Kamei, and H. Ishii, 13.4-Tb/s (134
111-Gb/s/ch) no-guard-interval coherent OFDM transmission over 3600 km
of SMF with 19-ps average PMD, in 34th European Conference on Optical
Communication, ECOC 2008, Brussels, Belgium, September 2008, pp. 12.
127

Bibliography
[14] J. Yu, X. Zhou, M. F. Huang, Y. Shao, D. Qian, T. Wang, M. Cvijetic, P. Magill, L. Nelson, M. Birk, S. Ten, H. B. Matthew, and S. K. Mishra, 17 Tb/s (161
114 Gb/s) polmux-RZ-8PSK transmission over 662 km of ultra-low loss fiber
using C-band EDFA amplification and digital coherent detection, in 34th European Conference on Optical Communication, ECOC 2008, Brussels, Belgium,
September 2008, pp. 12.
[15] IEEE Std 802.11nTM -2009.
IEEE 802.11 Working Group for WLAN
Standards, October 2009. [Online]. Available:
http://standards.ieee.org/
getieee802/download/802.11n-2009.pdf
[16] IEEE Std 802.16mTM -2011.
IEEE 802.16 Working Group on Broadband
Wireless Access Standards, May 2011. [Online]. Available: http://standards.

ieee.org/getieee802/download/802.16m-2011.pdf
[17] [Online].
Available:
http://www.samsung.com/global/business/
telecommunication/newsView.do?b2b bbs msg id=378

[18] W. Shieh and I. Djordjevic, OFDM for Optical Communications.
Academic
Press, 2009.
[19] H. A. Willebrand and B. Ghuman, Fiber optics without fiber, IEEE Spectrum,
vol. 29, no. 8, pp. 4145, August 2001.
[20] L. C. Andrews and R. L. Phillips, Laser Beam Propagation through Random
Media, 2nd ed. Bellingham, Washington: SPIE, 2005.
128

Bibliography
[21] A. A. Farid and S. Hranilovic, Outage capacity optimization for free-space optical links with pointing errors, IEEE Journal of Lightwave Technology, vol. 25,
no. 7, pp. 17021710, July 2007.
[22] Safe Use of Lasers, ANSI Z136.1.
American National Standards Institute,
ANSI, 2007.
[23] Safe Use of Lasers Outdoors, ANSI Z136.6.
American National Standards
Institute, ANSI, 2005.

[24] Safety of Laser Products, IEC 60825-1.
International Electrotechnical Com-
mission, IEC, 2001.

[25] Performance Standards for Light-Emitting Products, Laser Products, 21 CFR
1040.10. Center for Devices and Radiological Health, CDRH.
[26] Performance Standards for Light-Emitting Products, Specific-Purpose Laser
Products, 21 CFR 1040.11.
Center for Devices and Radiological Health,
CDRH.
[27] H. Wu, B. Hamzeh, and M. Kavehrad, Achieving carrier class availability of
FSO link via a complementary RF link, in Conference Record of the ThirtyEighth Asilomar Conference on Signals, Systems and Computers, vol. 2, November 2004, pp. 1483 1487.
[28] F. S. Vetelino, C. Young, L. Andrews, and J. Recolons, Aperture averaging effects on the probability density of irradiance fluctuations in moderate-to-strong
turbulence and spherical waves in the atmosphere, Journal of Applied Optics,
vol. 46, no. 11, pp. 20992108, April 2007.
129

Bibliography
[29] K. Tsukamoto, T. Higashino, S. Komaki, K. Kazaura, K. Wakamori, T. Suzuki,

and M. Matsumoto, A new loss model and system design method for radio on
free space optical link, in International Topical Meeting on Microwave Photonics, MWP 09, October 2009, pp. 14.
[30] S. M. Haas, Capacity of and Coding for Multiple-Aperture, Wireless, Optical
Communications. PhD thesis, Department of Electrical Engineering and Computer Science, MIT, 2003.
[31] J. A. Anguita, I. B. Djordjevic, M. A. Neifeld, and B. V. Vasic, Shannon capacities and error-correction codes for optical atmospheric turbulent channels,
Journal of Optical Networking, vol. 4, no. 9, pp. 586601, 2005.
[32] W. Gappmair and M. Flohberger, Error performance of coded FSO links in
turbulent atmosphere modeled by gamma-gamma distributions, IEEE Transactions on Wireless Communications, vol. 8, no. 5, pp. 22092213, May 2009.
[33] I. B. Djordjevic, O. Milenkovic, and B. Vasic, Generalized low-density paritycheck codes for optical communication systems, IEEE Journal of Lightwave
Technology, vol. 23, no. 5, pp. 19391946, May 2005.
[34] I. B. Djordjevic and B. Vasic, Nonbinary LDPC codes for optical communication systems, IEEE Photonics Technology Letters, vol. 17, no. 10, October
2005.
[35] M. Safari and M. Uysal, Relay-assisted free-space optical communication,
IEEE Transactions on Wireless Communications, vol. 7, no. 12, pp. 54415449,
December 2008.
130

Bibliography
[36] S. Kazemlou, S. Hranilovic, and S. Kumar, All-optical multihop free-space optical communication systems, Journal of Lightwave Technology, vol. 29, no. 18,
pp. 26632669, September 2011.
[37] A. S. Gurvich, V. I. Tatarskii, and L. P. Tsvang, DAN SSSR, vol. 123, 1958.
[38] M. E. Gracheva and A. S. Gurvich, Strong fluctuations in the intensity of light
propagated through the atmosphere close to the earth, Radiofiz., vol. 8, no. 4,
pp. 717724, 1965.
[39] I. I. Kim and E. Korevaar, Availability of free space optics (FSO) and hybrid
FSO/RF systems, Proceedings of the SPIE, vol. 4530, pp. 8495, 2001.
[40] M. Gebhart, E. Leitgeb, and J. Bregenzer, Atmospheric effects on optical
wireless links, in Proceedings of the 7th International Conference on Telecommunications, ConTEL 2003, vol. 2, Zagreb, Croatia, June 2003, pp. 395 401.
[41] M. S. Awan, C. Capsoni, E. Leitgeb, R. Nebuloni, Marzuki, F. Nadeem, and
M. S. Khan, FSO-relevant new measurement results under moderate continental fog conditions at Graz and Milan, in Advanced Satellite Mobile Systems,
2008. ASMS 2008. 4th, Bologna, August 2008, pp. 116 121.
[42] S. A. Zabidi, W. A. Khateeb, M. R. Islam, and A. Naji, The effect of weather on
free space optics communication (FSO) under tropical weather conditions and
a proposed setup for measurement, in International Conference on Computer
and Communication Engineering (ICCCE 2010), Kuala Lumpur, Malaysia,
May 2010, pp. 15.
131

Bibliography
[43] A. Khatoon, W. G. Cowley, and N. Letzepis, Channel measurement and estimation for free space optical communications, in AusCTW 2011, January February 2011.
[44] O. Fiser, J. Svoboda, Z. Chladova, V. Schejbal, and J. Pesek, FSO link attenuation measurement and modelling on Milesovka Hill, in Proceedings of the
5th European Conference on Antennas and Propagation (EUCAP), April 2011,
pp. 2518 2521.
[45] F. S. Marzano, S. Mori, F. Frezza, P. Nocito, G. M. T. Beleffi, G. Incerti,
E. Restuccia, and F. Consalvi, Free-space optical high-speed link in the urban
area of southern Rome: preliminary experimental set up and channel modelling, in Proceedings of the 5th European Conference on Antennas and Propagation (EUCAP), April 2011, pp. 2737 2741.
[46] L. C. Andrews, R. L. Phillips, C. Y. Hopen, and M. A. Al-Habash, Theory of
optical scintillation, Journal of Optical Society of America A, vol. 16, no. 6,
pp. 14171429, June 1999.
[47] H. Al-Raweshidy, Radio over Fiber Technologies for Mobile Communications
Networks, H. Al-Raweshidy and S. Komaki, Eds. Artech House, Inc, 2002.
[48] W. S. C. Chang, RF photonic technology in optical fiber links, W. S. C. Chang,
Ed. Camridge, 2002.
[49] K. Tsukamoto, Radio-on-fiber (RoF) technology and its applications, IEICE
Technical Report, OSCS2002, vol. 102, no. 358, pp. 2228, October 2002.
132

Bibliography
[50] A. Ngoma, Radio-over-Fibre Technology for Broadband Wireless Communication Systems.
PhD thesis, faculty of Electrical Engineering of the Eindhoven
University of Technology, 2005.

[51] S. Kajiya, K. Tsukamoto, and S. Komaki, Proposal of fiber-optic radio highway
networks using CDMA method, in Fourth IEEE International Conference on
Universal Personal Communications ICUPC 95, vol. E79-C, no. 1, November
1995, pp. 496500.
[52] B. J. Koshy and P. M. Shankar, Spread-spectrum techniques for fiber-fed microcellular networks, IEEE Transactions on Vehicular Technology, vol. 48,
no. 3, pp. 847857, May 1999.
[53] H. H. Refai, J. J. Sluss, and H. H. Refai, The transmission of multiple RF
signals in free space optics using wavelength division multiplexing, Proceedings
of the SPIE, vol. 5793, pp. 136143, May 2005.
[54] G. Katz, S. Arnon, P. Goldgeier, Y. Hauptman, and N. Atias, Cellular over
optical wireless networks, IEE Proceedings OptoElectronics, vol. 153, no. 4,
August 2006.
[55] K. Tsukamoto, H. Onodera, K.-H. Kim, T. Nakamura, T. Higashino, Y. Aburakawa, S. Komaki, K. Wakamori, K. Takahashi, T. Suzuki, K. Kazaura, A. M.
Shah, K. Omae, M. Matsumoto, and T. Satou, Link design of radio on free
space optic system for heterogeneous wireless services, in International Topical
Meeting on Microwave Photonics, Gold Coast, Queensland, Australia, September - October 2008, pp. 232235.
133

Bibliography
[56] T. Higashino, K. Tsukamoto, S. Komaki, K. Kazaura, K. Wakamori, and

M. Matsumoto, A demonstrative link design of ROFSO and its optimum performance - indoor short range experiment and a new model of optical scintillation, in Innovations for Digital Inclusions, K-IDI 2009. ITU-T Kaleidoscope,
August 2009, pp. 17.
[57] P. T. Dat, A. M. Shah, K. Kazaura, K. Wakamori, T. Suzuki, K. Omae,
M. Matsumoto, Y. Aburakawa, K. Takahashi, T. Nakamura, T. Higashino,
K. Tsukamoto, and S. Komaki, Investigation of suitability of RF signal transmission over FSO links, in International Symposium on High Capacity Optical
Networks and Enabling Technologies, HONET 2007, November 2007, pp. 16.
[58] K. Kazaura, P. Dat, A. Bekkali, A. Shah, T. Suzuki, K. Wakamori,
M. Matsumoto, T. Nakamura, K. Takahashi, T. Higashino, Y. Aburakawa,
K. Tsukamoto, and S. Komaki, Experimental evaluation of a radio-on-FSO
communication system for multiple RF signal transmission, Proceedings of the
SPIE, vol. 7199, January 2009.
[59] P. T. Dat, A. Bekkali, K. Kazaura, K. Wakamori, and M. Matsumoto, A universal platform for ubiquitous wireless communications using radio over FSO
system, Journal of Lightwave Technology, vol. 28, no. 16, pp. 22582267, August 2010.
[60] R. V. Nee and R. Prasad, OFDM wireless multimedia communications. Boston:
Artech House, 2000.
[61] Y. Wu and B. Caron, Digital television terrestrial broadcasting, IEEE Communications Magazine, vol. 32, no. 5, pp. 4652, May 1994.
134

Bibliography
[62] I. B. Djordjevic, B. Vasic, and M. A. Neifeld, LDPC coded OFDM over the
atmospheric turbulence channel, Optics Express, vol. 15, no. 10, pp. 63366350,
May 2007.
[63] A. Bekkali, C. B. Naila, K. Kazaura, K. Wakamori, and M. Matsumoto, Transmission analysis of OFDM-based wireless services over turbulent radio-on-FSO
links modeled by gammagamma distribution, IEEE Photonics Journal, vol. 2,
no. 3, pp. 510520, June 2010.
[64] I. B. Djordjevic and H. G. Batshon, LDPC-coded OFDM for heterogeneous
access optical networks, IEEE Photonics Journal, vol. 2, no. 4, pp. 611619,
August 2010.
[65] The MathWorks, Inc. (2011) MATLAB and Simulink for Technical Computing.
[Online]. Available: http://www.mathworks.com/products/matlab/
[66] A. Mostafa and S. Hranilovic, Channel measurement and Markov modelling
of an urban free-space optical link, IEEE Journal of Optical Communications
and Networks, (in preparation).
[67] , In-field demonstration of OFDM-over-FSO, IEEE Photonics Technology Letters, (submitted).
[68] fSONA Networks, Inc. (2011) fSONA: free Space Optical Networking
Architecture. [Online]. Available: http://www.fsona.com/
[69] TEK Microsystems, Inc., TekMicro. [Online]. Available:
tekmicro.com/
135
http://www.

Bibliography
[70] DigitalGlobe, Inc. (2011) Satellite Imagery and Geospatial Information

Products. [Online]. Available: http://www.digitalglobe.com/
[71] Degrees of Protection Provided by Enclosures (IP Code).
IEC60529, Interna-
tional Electrotechnical Commission, IEC.

[72] NEMA Enclosure Types. NEMA-4, National Electrical Manufacturers Association, NEMA.
[73] Code of Federal Regulations, Title 47, Part 15. FCC 47CFR15, Federal Communications Commission (FCC), 1997.
[74] Digital Apparatus, Spectrum Management and Telecommunications Policy.
ICES-003 Interference-Causing Equipment Standard, ICES, 2004.
[75] SONAbeamTM 1250-M datasheet. [Online]. Available: http://www.fsona.
com/prod/SONAbeam M.pdf
[76] SONAbeamTM Terminal Controller Software Users Guide. fSONA.
[77] SONAbeamTM Safety. [Online]. Available: http://www.fsona.com/product.
php?sec=safety
[78] IEEE 802.3. Ethernet Working Group, Institute of Electrical and Electronics
Engineers, IEEE.
[79] Fibre Channel Physical and Signaling Interface (FC-PH).
T11 (previously
known as X3T9.3), InterNational Committee for Information Technology Standards, (INCITS), 1993.
136

Bibliography
[80] Synchronous Optical Networking, SONET.
ANSI T1.105, American National
Standards Institute, ANSI.

[81] Synchronous Digital Hierarchy, SDH.
ITU G.707, G.783, G.784, and G.803,
European Telecommunications Standards Institute, ETSI.

[82] SMPTE Timecode.
Society of Motion Picture and Television Engineers,
SMPTE.
[83] Engineering Specification - Custom 1250-M for McMaster University,
ENGSPEC-90-0272. fSONA.
[84] Tekmicro, Triton-V5-VXS datasheet. [Online]. Available:
http://www.
tekmicro.com/PDFs/Tekmicro Triton 2channel ADC DAC.2010.07.pdf

[85] Xilinx, Inc. (2011). [Online]. Available:
http://www.xilinx.com/support/
documentation/virtex-5.htm
[86] Wenzel Associates, Inc. (2011). [Online]. Available: http://www.wenzel.com/
[87] Tycon Power Systems, Inc. (2011) EasyWeather Wireless Data Logging
Weather Station Software. [Online]. Available: http://proweatherstation.com/
easyweather/easyweather.htm
[88] W. Pratt, Laser Communication Systems. J. Wiley & Sons, New York, 1969.
[89] S. Karp, R. M. Gagliardi, S. E. Moran, and L. B. Stotts, Optical Channels:
Fibers, Clouds, Water, and the Atmosphere. Plenum Press, 1988.
[90] J. W. Strohbehn, Ed., Laser Beam Propagation in the Atmosphere. SpringerVerlag, 1978.
137

Bibliography
[91] I. Kim, J. Koontz, H. Hakakha, P. Adhikari, R. Stieger, C. Moursund, M. Barclay, A. Stanford, R. Ruigrok, J. Schuster, and E. Korevaar, Measurement of
scintillation and link margin for the TerraLink laser communication system,
Proceedings of the SPIE, pp. 100118, January 1998.
[92] E. Jakeman and P. N. Pusey, The significance of K-distributions in scattering
experiments, Physical review Letters, vol. 40, pp. 546550, February 1978.
[93] G. Parry and P. N. Pusey, K distributions in atmospheric propagation of laser
light, Journal of Optical Society of America, vol. 69, no. 5, pp. 796798, May
1979.
[94] J. H. Churnside and R. J. Hill, Probability density of irradiance scintillations
for strong path-integrated refractive turbulence, Journal of Optical Society of
America A, vol. 4, pp. 727733, 1987.
[95] L. C. Andrews and R. L. Phillips, I-K distribution as a universal propagation
model of laser beams in atmospheric turbulence, Journal of Optical Society of
America A, vol. 2, no. 2, pp. 160163, February 1985.
[96] R. J. Hill and R. G. Frehlich, Probability distribution of irradiance for the
onset of strong scintillation, Journal of Optical Society of America A, vol. 14,
pp. 15301540, July 1997.
[97] S. M. Kay, Modern Spectral Estimation: Theory And Application.
Prentice
Hall, Englewood Cliffs, New Jersy, 1988.

[98] Nist/sematech e-handbook of statistical methods. [Online]. Available:
http://www.itl.nist.gov/div898/handbook/
138

Bibliography
[99] The weather network. [Online]. Available: http://www.theweathernetwork.

com/weather/caon0289
[100] F. S. Vetelino, B. Clare, K. Corbett, C. Young, K. Grant, and L. Andrews,
Characterizing the propagation path in moderate to strong optical turbulence, Journal of Applied Optics, vol. 45, no. 15, pp. 35343543, May 2006.
[101] Xilinx
LogiCORE
IP
2011. [Online]. Available:
Fast
Fourier
Transform
v7.1,
March
http://www.xilinx.com/support/documentation/
ip documentation/xfft ds260.pdf
[102] W. Turin, Digital Transmission Systems: Performance, Analysis, and Modelling, 2nd ed. McGraw-Hill, NY, 1999.
[103] H. S. Wang and N. Moayeri, Finite-state Markov channel - a useful model for
radio communication channels, IEEE Transactions on Vehicular Technology,
vol. 44, no. 1, pp. 163171, February 1995.
[104] Q. Zhang and S. Kassam, Finite-state Markov model for Rayleigh fading channels, IEEE Transactions on Communications, vol. 47, no. 11, pp. 16881692,
November 1999.
[105] C. D. Iskander and P. Mathiopoulos, Finite-state Markov modeling of diversity
Nakagami channels, in Proceedings of the 7th Canadian Workshop on Information Theory, 2001.
[106] V. Bhaskar, Finite-state Markov model for lognormal, chi-square (central), chisquare (non-central), and K-distributions, International Journal of Wireless
Information Networks, vol. 14, no. 4, pp. 237250, October 2007.
139

Bibliography
[107] X. Zhu and J. M. Kahn, Markov chain model in maximum-likelihood sequence

detection for free-space optical communication through atmospheric turbulence
channels, IEEE Transactions on Communications, vol. 51, no. 3, pp. 509516,
March 2003.
[108] A. R. Bahai, B. R. Saltzberg, and M. Ergen, Multi-Carrier Digital Communications; Theory and Applications of OFDM, 2nd ed. Springer Science + Business
Media, Inc., 2004.
[109] R. Prasad, OFDM for Wireless Communications Systems. Artech House, 2004.
[110] S. Hara and R. Prasad, Multicarrier Techniques for 4G Mobile Communications.
Artech House, 2003.
[111] H. Liu and G. Li, OFDM-Based Broadband Wireless Networks; Design and
Optimization. John Wiley & Sons, Inc., 2005.
[112] RW Chang, Synthesis of band-limited orthogonal signals for multichannel data
transmission, Bell Systems Technical Journal, vol. 45, pp. 17751796, December 1966.
[113] , Orthogonal frequency division multiplexing, U.S. Patent no. 3488445,
1970.
[114] S. B. Weinsten and P. M. Ebert, Data transmission by frequency-division multiplexing using the discrete Fourier transform, IEEE Transactions on Communication Technology, vol. 19, no. 5, pp. 628634, October 1971.
[115] J. W. Cooley and J. W. Tukey, An algorithm for the machine calculation
140

Bibliography
of complex Fourier series, Mathematics of Computation, vol. 19, no. 90, pp.
297301, April 1965.
[116] J. Armstrong and A. J. Lowery, Power efficient optical OFDM, IEEE Electronics Letters, vol. 42, no. 6, pp. 370372, March 2006.
[117] J. Armstrong and B. J. C. Schmidt, Comparison of asymmetrically clipped optical OFDM and DC-biased optical OFDM in AWGN, IEEE Communications
Letters, vol. 12, no. 5, pp. 343345, May 2008.
[118] Digital Video Broadcasting (DVB); Framing structure, channel coding and modulation for digital terrestrial television. EN 300 744 V1.1.2, European Telecommunications Standards Institute (ETSI), August 1997.
[119] U. Ladebusch and C. A. Liss, Terrestrial DVB (DVB-T): A broadcast technology for stationary portable and mobile use, Proceedings of the IEEE, vol. 94,
pp. 183193, January 2006.
141

Measurement and Modelling of A Free-Space Optical Link and In-Field OFDM Experiment

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Measurement and Modelling of A Free-Space Optical Link and In-Field OFDM Experiment

Uploaded by

Copyright:

Available Formats

See

Measurement and Modelling of a Free-Space

Available from: Ayman Mostafa

MEASUREMENT AND MODELLING OF A

MEASUREMENT AND MODELLING OF A FREE-SPACE

c Copyright by Ayman Mostafa, January 2012

Master of Applied Science (2012)

(Electrical & Computer Engineering)

Hamilton, Ontario, Canada

Dr. Steve Hranilovic

Wireless Optical Communications . . . . . . . . . . . . . . . . . . . .

Intensity-Modulation Direct-Detection FSO Channel . . . . . . . . .

Optical Transmission of Radio Signals

Objectives and Requirements . . . . . . . . . . . . . . . . . . . . . .

FPGA-Based Digitizer Board . . . . . . . . . . . . . . . . . . . . . .

3 FSO Link Measurement and Modelling

Atmospheric Loss and Turbulence . . . . . . . . . . . . . . . .

The Log-Normal Distribution . . . . . . . . . . . . . . . . . .

Transceiver Measurement and Modelling . . . . . . . . . . . . . . . .

Harmonic Distortion and Nonlinearity . . . . . . . . . . . . .

Power Loss Estimate . . . . . . . . . . . . . . . . . . . . . . .

4 In-Field Demonstration of OFDM over FSO

Peak-Average Power Ratio . . . . . . . . . . . . . . . . . . . .

OFDM for FSO Systems . . . . . . . . . . . . . . . . . . . . .

OFDM over FSO Field Experiments . . . . . . . . . . . . . . . . . .

DC-Biased Unclipped OFDM over FSO . . . . . . . . . . . . .

DC-Biased Clipped OFDM over FSO . . . . . . . . . . . . . . 103

DVT-B over FSO . . . . . . . . . . . . . . . . . . . . . . . . . 108

5 Conclusions and Future Directions

Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

A FSO Link Management Software

B Finite-State Markov Model Parameters for K = 64

Block diagram of an intensity-modulation direct-detection FSO channel.

Radio-over-fiber connectivity between micro-cells in cellular networks.

RoFSO along with RoF for back-haul connectivity. . . . . . . . . . .

Site views of the experimental FSO link. . . . . . . . . . . . . . . . .

Customized SONAbeam TM -1250-M optical head and PCA box. . . . .

Equivalent circuit of the analog laser driver (modified based on [83]).

Equivalent circuit of the driver for Lasers (2) and (3). . . . . . . . . .

Receiver equivalent circuit (modified based on [83]). . . . . . . . . . .

Architecture of the Triton TM -V5-VXS digitizer board (modified based

Block diagram of the experimental setup. . . . . . . . . . . . . . . . .

2.10 Experimental setup at McMaster Innovation Park: (a) outdoors, (b)

Proposed channel model for the FSO link. . . . . . . . . . . . . . . .

Simplified channel model used for harmonic distortion measurements.

Transmitted pulse train used for frequency response measurements. .

Normalized magnitude frequency response of the FSO link obtained by

Histogram of thermal plus background noise samples at the receiver

Block diagram of the measurement system at the receiver. . . . . . .

Normalized PSD of the received signal used for channel measurements.

Received irradiance fluctuations through different time scales: (a) 1

3.10 Irradiance fluctuations statistics: (a) normalized autocovariance, (b)

3.12 Weather condition corresponding to the irradiance fluctuations in Fig.

3.21 Normalized autocovariance of the channel realizations obtained from

Baseband multi-carrier communications system. . . . . . . . . . . . .

Statistics of the amplitude of a typical OFDM symbol with 40000 sub-

OFDM symbol timing recovery by the means of cross-correlation. . .

Typical cross-correlation between the transmitted waveform of an OFDM

Spectrum of the 20 OFDM channels transmitted in the DC-biased

Block diagram of the implemented OFDM over FSO transmitter. . .

Block diagram of the implemented OFDM over FSO receiver.

Received constellations for DC-biased unclipped OFDM over FSO. . . 101

4.15 Received constellations for higher-order QAMs under clear weather