Introduction

1
If a man writes a better book, preaches a better sermon, or makes a
better mouse-trap than his neighbor, the world will make a beaten path to
his door
−RALPH WALDO EMERSON

When Tim Berner-Lee and Robert Cailliau invented the World Wide Web in
1991, nobody realized that the Internet and its applications would pierce human
life as it has now. Today, the World Wide Web has really started to look like,
well, a world wide web. A plethora of applications − Facebook, WhatsApp, and
Skype − have already occupied human time and space, and still innumerable
applications are there on the verge to thrust a colossal amount of data into
today’s network. Clearly, the networks should be enhanced continually to keep
pace with this enormous bandwidth increase.
Currently, the main bandwidth bottleneck lies in the “last mile” of a
telecommunications network that runs from a service provider’s facility to a
home or business. This part of the telecommunications network is also referred
to as an access network. The first step to solve this bandwidth gridlock in access
networks is to use optical technologies as a transmission medium. An optical
fiber, being an excellent communication medium that can supply a high
bandwidth in the order of Tb/s, is ideally suited for high-bandwidth capacities.
2 Chapter 1

Therefore, the trend of the evolution is toward fiber to the home (FTTH) based
access networks, in which an optical fiber directly reaches a customer’s home.
These FTTH networks should be further enhanced to satisfy a continual
bandwidth increase. This evolution requires new architectures and dynamic
bandwidth allocation (DBA) algorithms.
The focus of this chapter is on the design of architectures and DBA algorithms
for next-generation optical access (NGOA) networks. This merges the
knowledge of both architectural designs and DBA algorithms. This chapter
provides a detailed overview and introduction to both.
The architectural background, which comes first, gives an overview of fixed
access networks (section 1.1), optical access network architectures (section 1.2),
and the evolution strategies for NGOA networks (section 1.3). Then, we
introduce DBA algorithms (section 1.4). Thereafter, we present an overview of
this book (section 1.5), and finally, we list all publications obtained during this
PhD research (section 1.6).

1.1 Fixed access network

A telecommunications network can typically be split into four segments: the
core network, aggregation network, access network, and local area network
(LAN), cf. Figure 1.1. The separate network segments can readily be
distinguished by their important differences in traffic, equipment, technologies
and operations. A core network, or backbone network, is the central part of a
telecommunications network that provides various services to customers.
Aggregation networks − also sometimes referred to as metro networks −
aggregate the traffic from several access segments and feed the traffic into the
core network. At the interface between the core and the aggregation network lies
an edge node, which denotes the position of broadband remote access server
(BRAS), at which different Internet service providers (ISPs) are connected.
In this chapter, we focus on access networks, which lie between the
aggregation and LAN network. As we already mentioned, access networks pose
the main bottleneck in providing a fat bandwidth pipe between services and
users. This bottleneck is mainly because of the prime objective of keeping access
networks cost effective, which has forced the operators to use their legacy
equipment as much as possible, resulting in the continual use of the oldest and
the most common transmission medium, which is twisted copper pair [1].
Nevertheless, as the required data-rates have been increasing, the need to
upgrade the capacity of the networks has become imperative. The capacity
upgrade makes the change in the transmission media inevitable, and thus the
networks have evolved by combining a change in the transmission media with a
change in the overlaying technology.
 Internet Internet
Service Service
provider 1 provider 2

Core network

Edge node

Aggre-
gation

Access

LAN

Figure 1.1: High-level overview of a telecommunications network

4 Chapter 1

We will first see in section 1.1.1 as how the increased bandwidth and data-
rates requirements oblige a change in the transmission media. Different options
of transmission media will then be discussed in section 1.1.2. Finally, we will
cover the evolution of the networks in section 1.1.3.

1.1.1 Bandwidth and data-rates requirements

Communication networks have evolved from being a primary source of
speech to becoming a network used for transmitting many other sources of
information like music, pictures, and computer data. The applications − the
source of information − have been requiring increasing data-rates. Looking to the
future, these data-rates are expected to increase at an exponential rate similar to
Moore’s Law – this is also known as Nielsen’s Law1. Today’s high-definition
television (HDTV) requires bit rates between 8 and 15 Mb/s, even with new
compression technology of the Moving Pictures Experts Group’ MPEG4. 3D
immersive HDTV – a technology already being used in some academic and
industrial settings – will require 100 to 300 Mb/s, whereas future evolution like
8k ultra-HDTV (UHDTV) is expected to consume about 160 Mb/s per channel
[2]. Furthermore, 3D versions of UHDTV, combined with time-shifted unicast or
even peer-to-peer applications, will fill bandwidths up to the 400 Mb/s range.
Assuming two parallel independent downstream or download sessions, the
respective (downstream) UHDTV bandwidth will need to be at least 800 Mb/s.
Increased data-rates impel an increase in the bandwidth capacities of the
transmission media by the similar orders of magnitude. Using Shannon’s
equation [3], we can formulate the relation between data-rates and the required
channel capacities as:

C  B log2 (1  SNR) bit/s

Here C is the rate of information transmission per channel, B is the channel

bandwidth, and SNR denotes Signal to Noise ratio.
From Shannon’s equation, it is clear that the increased data-rates entail a
transmission medium with higher capacities. This push toward higher channel
capacities has been one of the main drivers behind the technology evolution in
access networks.

1.1.2 Physical media

There are three important guided transmission media (Figure 1.2): twisted pair

1Nielsen’s law states that the user's connection speed grows by 50 percent per year, or doubles every
21 months.
Introduction 5

(used to connect a home to the telephone network), coaxial cable (used to

connect a home with the TV-distribution network, e.g., Cable TV) and optical
fiber (as of today, used mainly to interconnect large telephone exchanges and
ISP routers). Note that an example of an unguided transmission medium is the
air, which is used for wireless access (out of the scope of this chapter).
 A twisted pair consists of two insulated copper wires that are twisted
together. When the wires are twisted, the waves from different twists
cancel out, so the wires radiate less effectively, and it becomes less
sensitive to external noise sources. This helps it to span longer
transmission distances.
 A coaxial cable consists of a copper wire as the core, surrounded by
an insulating material. The insulator is encased with a conducting
copper shield. Many coaxial cables also have an insulating outer
sheath or jacket. Compared to a twisted pair, a coaxial cable has a
better shielding against external noise sources, and can thus span
longer distances at higher speeds.
 An optical fiber consists of a glass core surrounded by a transparent
cladding material with a lower index of refraction. It also includes a
plastic jacket for the protection of the fiber. The light is guided in the
fiber by the principle of total internal reflection 2.
The main deterministic property of these transmission media is their
bandwidth. Note that the bandwidth of a channel is the range of the frequencies
that it can transmit with reasonable fidelity. Figure 1.3 depicts attenuation3 or
transmission loss at various frequencies in different transmission media. From
this, the typical bandwidth of these media can be concluded as in Table 1.A.

1.1.3 Evolution of fixed access networks

One of the oldest examples of communication networks is a telephone
network [1]. Telephone networks were mainly used to transmit narrowband
voice channels and used twisted pair. Twisted pair is an excellent communication
medium for low bandwidth signals. However, because of the emergence of the
Internet and the World Wide Web and its several applications like video
applications, the need arose to optimize these networks for data communication.
This led to digital subscriber loop (DSL) networks [4], which removed the filter
used by telecommunications operators to restrict the bandwidth of the local loop
to approximately 4 KHz, thus making the entire
2
Total internal reflection is a phenomenon when the light cannot pass through a medium and is
entirely reflected. It happens when a propagating light strikes a medium boundary at an angle larger
than a critical angle, and the refractive index is lower on the other side of the boundary.
3
Attenuation is the reduction in intensity of the light beam (or signal) with respect to distance
travelled through a transmission medium.
6 Chapter 1

twist
Twisted pair length

Outer Outer sheath

conductor

Insulation
Coaxial cable
Inner
conductor

Jacket
Core Cladding
Optical fiber

Angle of Angle of
Light at less than incidence reflection
critical angle is
absorbed in jacket

Figure 1.2: Examples of physical transmission media

30
twisted pair
Attenuation (dB/km)

10
coaxial cable
optical fiber
3

0.3

0.1
1 kHz 1 MHz 1 GHz 1THz 1000 THz
Frequency
Figure 1.3: Attenuation as a function of frequency for different transmission
media

Table 1.A Transmission media and their typical bandwidths

Transmission Media Bandwidth
Twisted pair 10 MHz
Coaxial cable 1 GHz
Optical fiber 10 THz
Introduction 7

capacity of the local loop available to the users. The limiting factor then becomes
the physics of the local loop, which depends on several factors, including its
length, thickness, and general quality.
The first broadband DSL standard was the integrated services digital network
(ISDN) system [5]. Later on, various flavors of DSL technologies (collectively
called xDSL) have been invented for broadband data delivery on twisted copper
pairs. Asymmetric DSL (ADSL) and very high data rate DSL (VDSL) are the
two most common DSL technologies. ADSL provides a downstream data rate of
up to 8 Mb/s and upstream data rate up to 800 Kb/s, over a maximum
transmission distance of 5.5 km [5].
There is a marked relationship between distance and available bandwidth,
when we are using copper. The data rate can be increased by limiting the length
of the copper loop. Thus, newer DSL technologies like VDSL are usually
supported with a fiber deep infrastructure – in which the fiber is extended deeper
(or closer to the users) into the network such as fiber-to-the-curb (FTTC) −
which has a short distance of twisted copper loop. In the 4G broadband project
[6], it was shown that for a copper loop distance of up to 200 m, VDSL can
achieve a data-rate of 1 Gb/s with the use of vectoring, etc.
Figure 1.4 depicts this evolution trend where the digital subscriber line access
multiplexer4 (DSLAM) at the central office (CO) moves toward the users to
minimize the length of the copper loop, thereby supplying higher data rates. This
also leads to a variety of fiber to the X (FTTX) networks [7]-[8], where X can
mean curb, building, and home, etc. Note that FTTX networks use an optical line
terminal (OLT) at the CO and an optical network unit (ONU) at the user’s
premises. Finally, we present a timeline of when different DSL technologies
were adopted in Europe, Figure 1.5. It should be of no surprise that the adoption
of technologies matched with the Neilson’s law of bandwidth evolution.
Another technology that has also been used is hybrid fiber coax (HFC). HFC
networks use a coaxial cable network. Traditionally, coaxial cable has been used
for transmitting one-way broadcasting systems, like Cable TV. Later on because
of the excellent bandwidth of coaxial cable compared to twisted pairs, they were
also used for data communications. Unlike the rest of the world, where DSL
services are more popular than cable modem, in the United States, cable modem
became the dominating form of broadband access because of far more complete
coaxial plant coverage than in other countries 5.

4 A DSLAM is a network device, often located in telephone exchanges, that connects multiple
customer DSL interfaces to a high-speed digital communications channel using multiplexing
techniques.
5 The exceptions to this general trend exist. For example, Belgium has a cable penetration of almost

100%, yet DSL is more dominant. However, in Flanders – with cable operator Telenet – HFC is the
dominant broadband technology. Therefore, there are many regional differences.
8 Chapter 1

copper
Remote
CO node
fiber

Edge DSL
ADSL Node
Agg. network DSLAM modem

VDSL/ Edge DSL

Agg. network DSLAM modem
FTTC Node

Edge ONU DSL

FTTB Node
Agg. network OLT
building modem

Edge
FTTH Node
Agg. network OLT ONU

Figure 1.4: ADSL to FTTH evolution6

10000 FTTH
1000 FTTB
Access Bitrate (Mb/s)

100 VDSL

10 ADSL2+
ADSL
1 ISDN

0.1 Narrowband
Commercially available
0.01 bandwidth in Europe

0.001 Nielson's Law

0.0001
1983 1991 1999 2007 2015 2023

Figure 1.5 Access bit rate evolution of different DSL technologies [source: IEEE
spectrum]

1.2 Optical access network architectures

The potential of optical fiber to deliver high bandwidth has been well
established for many years; however, the recent surge in bandwidth demand,
driven by fast-growing video-on-demand (VOD) services and emerging
applications − network gaming, peer-to-peer downloading − has revitalized the

6 Agg. network denotes aggregation network.

Introduction 9

use of optical fiber for access networks. Currently, optical access networks have
been widely deployed in Asia, North America and Europe. Nevertheless, the first
generation of optical access networks has already reached their bandwidth
saturation, which has impelled innovations in this technology and subsequently
the next generations of optical access networks have been developed. Moreover,
ever-increasing bandwidth requirements, coupled with the innovations in
physical systems, will continue to propel the innovations of optical access
networks. In this section, we first provide a brief discussion of optical
communication line and then discuss optical access network architectures.

1.2.1 Optical communication link

A simple schematic of an optical communication link is presented in Figure
1.6. An optical communication link is composed of an electrical-to-optical
converter, a transmission medium, and an optical-to-electrical converter.
Laser diodes are used as an electrical-to-optical converter. They are thus used
as transmitters of an optical signal. A typical example of a laser diode is
distributed feedback (DFB). These diodes are either directly modulated or
externally modulated, depending upon the desired data rate. Usually, external
modulation is used for higher (10 Gb/s or more) data rates.
An optical fiber is used as the transmission medium. When light propagates
through the optical fiber, it loses its strength and broadens in time. The first
effect is called as attenuation, and the second as dispersion. Attenuation is
measured in dB/km. Because of attenuation, the optical fiber has a limited
length. This length decides the reach or span of the optical system. On the other
hand, dispersion limits the data rate that can be transported over a fiber. It is
measured in ps/km-nm. As the dispersion is a key criterion, typically single
mode fibers (SMF) are used in access networks.
Photo-diodes are used as an optical-to-electrical converter. They are thus used
as receivers of an optical signal. Typically, two kinds of photo-diode are used:
positive-intrinsic-negative (PIN) photodiode and avalanche photo-diode (APD).
These photo-diodes differ in their receiver sensitivity7 and cost. APD has better
receiver sensitivity but a higher cost.
The important characteristic of the optical communication link is the power
budget. The power budget is the allocation of available power (launched into a
given link by a given source) among various losses in order to ensure that
adequate signal strength (power) is available at the receiver. For example, if the
optical communication system has the following parameters: laser transmitting

7
It is the minimum magnitude of input signal required to produce an output signal having a specified
signal-to-noise ratio.
10 Chapter 1

Electrical Optical to
Information Information
to Optical Medium electrical
Source Sink
converter converter

Optical communication System

Transmitter (Tx) = Optical Fiber Receiver (Rx) =

Laser diode Photo-diode

Figure 1.6: Optical communication link

power of 6 dBm8, receiver sensitivity of -30 dBm, losses of 20 dB, then the
power budget is 16 dB (6 dBm-(-30 dBm)-20 dB). Based on this power budget,
Power Budget
the reach of the system can be calculated as . Fiber losses
Fiber Loss
depend upon the used wavelengths.
The losses in the link can be computed based on the insertion loss of the
components. Insertion loss is the loss of signal power resulting from the insertion
of a device in a transmission line and is usually expressed in dB. The insertion
loss should also account for splice losses (losses at the fiber joints, typically 0.1
dB), connector losses (losses due to connectors, typically 0.5 dB), and end-of-life
losses (losses due to the use of components over time).

1.2.2 Optical access network architectures

In a typical optical access network architecture, an ONU at a subscriber’s end
is connected to an OLT at the CO, through an optical distribution network
(ODN) composed of different fiber segments and remote nodes (RNs), cf. Figure
1.7. In the uplink (toward core) direction, an OLT is connected to an aggregation
network. Real architectural implementations could include several stages of RNs,
allowing the topology to be scalable with the number of connected users. As an
example, we consider two splitting stages of RNs.
Optical access networks can use different multipoint topologies. The most

8 dBm (sometimes dBmW or Decibel-milliwatts) is an abbreviation for the power ratio in decibels
(dB) of the measured power referenced to one milliwatt (mW).
Introduction 11

Access network

Central Office Street cabinet Building Home

Feeder Distribution Last Mile

Fibre Fibre Fibre

OLT
Remote node Remote node
ONU
Core Aggregation
network

…
…

Figure 1.7: Optical access network

common architecture, however, is based on a tree topology, with the OLT as the
root of the tree and the ONUs as the leaves. The first RN (RN1) in an optical
access network is connected to the OLT via a feeder fiber (FF). Through the
distribution fiber (DF), each output port of RN1 goes to a second RN (RN2).
Each output port of the RN2 is then connected to one ONU by the last mile fiber
(LMF).
There are three main categories of FTTH networks (Figure 1.8), home run,
active and passive [9]. The simplest FTTH architecture is home run or point-to-
point (P2P) fiber architecture, which offers a dedicated fiber from the CO to each
user. In case of an active optical network (AON), a switch or router is installed
between the CO and the user, and from this point, a dedicated fiber reaches each
user. On the other hand, passive optical networks (PONs) use passive
splitters/combiners instead of the active switch. Both AONs and PONs are point-
to-multipoint (P2MP) networks. We will now discuss these architectures in more
detail.

1.2.2.a Point to point architectures

P2P architectures provide individual fibers between the CO and each user.
This architecture thus requires many fibers, shooting up its first installation cost.
Many fiber terminals (as many as homes) also increase the required floor space
at the CO. Conversely, this architecture offers high capacity and a high flexibility
to upgrade each user individually.

1.2.2.b Active optical networks

Current AONs use an active switch at the RN. The active switch may be pure
open systems interconnection (OSI) layer 2, or layer 2 with some layer 3 features
(e.g., ICMP snooping), or a layer 3 device (i.e., IP routers). Because of different
technical backgrounds, different practical and geographical constraints, AON
12 Chapter 1

Point to Point

CO

Active Optical Network

CO

Active
Switch

Passive Optical Network

CO

Passive
Splitter

Figure 1.8: FTTH architectures

networks differ from each other. In an active star deployment, a single fiber is
connected from an aggregation switch at the CO to an active switch at the RN,
from where individual fibers run to each ONU. Compared to P2P architecture, it
reduces the number of fibers. However, the active switch needs powering and
maintenance. It also needs to survive a wider range of temperatures than in-door
equipment.
A typical advantage of an AON technology is that the ONUs can talk amongst
each other. This provides great opportunities for maintaining locality of traffic
Introduction 13

meaning that the local traffic remains local. Another advantage is the ease of
service provisioning by utilizing Virtual-LANs and/or per service IP address
spaces. Virtual-LANs are a means to create logical topologies on top of the
actual network topology; for example, a logical topology can be created that
includes only the customers that subscribe to a particular service such as IPTV.
AONs are mostly deployed in Europe, while Sweden in particular is
completely dominated by AON. Several tier 2 and 3 network operators in Europe
also deploy AON, e.g., in Denmark, Norway, the Netherlands, and Austria. Hong
Kong also deploys a very large AON, and there are deployments in the Middle
East and North America as well, but worldwide PON architectures are the
dominant technology of choice amongst deployed FTTH schemes. Currently,
about 70-80% deployment is of PON, out of which Gigabit PON (GPON) is the
dominant technology [8].

1.2.2.c Passive optical networks

PONs use passive components at the RN. The passive components do not
require any electrical powering, have a high reliability, and are typically easy to
handle. The most common passive component that is used by the currently
deployed PONs is a power splitter. In the downstream direction, the power
splitter broadcasts content to all users, and the users select their respective
content. Thus, the power splitter based architectures are also referred to as
broadcast and select. The most common examples of broadcast and select
architectures are Ethernet PON (EPON) [10] and Gigabit PON (GPON) [11].
They use two wavelengths, one for the downstream (OLT to ONUs) and the
other for the upstream (ONUs to OLT) direction. In the upstream direction, a
mechanism is required to avoid the collision between the data of different users.
A commonly used mechanism is time division multiple access (TDMA). We will
discuss these mechanisms in more detail in section 1.4. When TDMA is used in
the upstream direction, the power splitter based PONs are also referred to as
TDMA-PONs.
A power splitter is a very simple and an inexpensive component. It however
imposes an insertion loss of at least 3 dB per 1:2 split, as per 1:2 split half of the
power is lost, i.e. 10 log102 = 3 dB. Thus a power splitter of 1:32 fan out imposes
an insertion loss of log232  3 dB. Another disadvantage of a power splitter
stems from its broadcasting nature, which makes the content of a user available
to all users. This may raise security concerns, which require good encryption,
especially for business users.
The evolution of PONs aims at increasing the capacity of the PONs, which
can be done by increasing the line rate (the rate in b/s at which the optical
communication line accepts bits) of the PONs or by using multiple wavelengths
per PON, or by combining both approaches. Up to now, many generations of
14 Chapter 1

PON systems have been developed. The first generation 9 of PON consists of
EPON and GPON. From an architectural perspective, EPON (offers up to 1 Gb/s
in downstream direction) and GPON (offers up to 2.5 Gb/s in the downstream
direction) use a power splitter at the RN and use a TDMA based medium access
control (MAC) protocol.
Further evolution of PON, referred to as next generation PON1 (NG-PON1),
supports coexistence with GPON/EPON on the same ODN. The coexistence
enables seamless upgrade of individual customers to NG-PON on a live ODN
without disrupting services of other customers. XG-PON (successor of GPON)
and 10G-EPON (successor of EPON) are the important examples of NG-PON1.
NG-PON2 may bring revolutionary change and may be disruptive with no
requirement in terms of coexistence with G-PON/EPON on the same ODN. The
important examples of these technologies are WDM- and TWDM-PON.
We now discuss some important PON technologies in more detail.

Higher line rate PONs

By increasing the line rates, the total capacity of the PONs can be increased
[12]. Thus, higher line rate PONs like 10G-EPON (10.3 Gb/s downstream and
1.25 Gb/s upstream), XG-PON (10 Gb/s downstream and 2.5 Gb/s upstream),
and XLG-PON10 (40 Gb/s downstream and 10 Gb/s upstream), have been
envisioned to increase the capacity of first generation of PONs like EPON and
GPON.
Increased line rates require advanced transceivers, faster processing, and a
higher resistivity to dispersion. These requirements impose significant cost and
power consumption penalties at both the OLT and the ONUs. Note that as the
ONU is not shared, an increase in the cost of the ONUs hurts operators most.
Thus, architectures should be designed to keep the ONU as cost-effective as
possible.

Wavelength division multiplexed PON (WDM-PON)

WDM-PON increases the bandwidth of the PON by employing wavelength-
division multiplexing (WDM) so that multiple wavelengths may be supported in
either or both upstream and downstream directions. As an example, it uses
multiple wavelengths and each user gets a different pair (for downstream and
upstream) of wavelengths. The simplest solution is to use a colored transceiver at
ONU, in which every ONU gets a transceiver of a different color (or wavelength)
[13]. Consequently, the vendors have to stock an inventory of different colored
transceivers. In addition, colored transceivers shrink the cost reduction that can

9 The first generation of PON also consists of variants like ATM based PON (APON) and Broadband
PON (BPON). However, they are omitted, as they are obsolete today.
10
XLG-PON is the successor of XG-PON. Note that XL is the Roman notation of 40.
Introduction 15

be attained by mass production. Thus, colorless transceivers are preferred at the

ONUs. There are many approaches for colorless transceivers [14]. One approach
is to use a tunable transceiver, which use a tunable transmitter and a broadband
receiver (so that it can scan a broader spectrum). The tunable transmitter is
typically more expensive compared to a fixed one. In addition, as the receiver is
broadband, it adds to the cost. Another disadvantage of WDM-PON is its
inflexibility [15] in resource allocation. As each user gets one wavelength, there
is no possibility of a flexible resource use: even if a user does not use its
wavelength, it cannot be reconfigured to another user.
WDM-PON has also some advantages compared to TDMA-PON. WDM-PON
avoids the complexity of a TDMA based bandwidth allocation. As each user gets
a dedicated wavelength in WDM-PON, it is assured a high quality of service
(QoS) without complexities and inefficiencies of TDMA. It also removes the
need of burst mode transceivers, i.e., the transceivers that can change their
transmission and reception levels in very short time-scales (ns). Note that,
TDMA-PON requires a burst mode receiver at the OLT and a burst mode
transmitter at the ONU. The burst mode at a receiver is used to adjust the
receiver to the different levels of the received signals. In WDM-PON, as signals
from only one user is received, there is no requirement of a burst mode receiver
at the OLT. In addition, the burst mode at the transmitter is only required for
TDMA-PON − which requires a user should transmit only at its own time
interval − so that it could enable a faster on/off of a transmitter.
WDM-PON does not need to use a power splitter because every user gets its
own wavelength. A passive component that looks quite promising for WDM-
PON is an arrayed waveguide grating (AWG). An AWG splits the content
according to the input wavelengths and typically feeds one wavelength per
output port. If there is one user per output port of an AWG, every user receives
its dedicated wavelength. The main advantage of an AWG is its low (3 to 4 dB)
and fan-out independent insertion loss. In addition, the architecture becomes
quite secure as every user gets only its wavelength. The main disadvantage of an
AWG, however, is that it needs to replace already deployed power splitters,
hindering migration, which is a key criterion in the choice of a new technology
for the network operators. The alternative is to use WDM-PON over a power
splitter. The question here is what is preferred more, the advantages of an AWG
or an easy migration enabled by a power splitter.
A typical example of WDM-PON is shown in Figure 1.9. We assume 32
wavelengths, and each transmitter (Tx) and receiver (Rx) at 1 Gb/s. We assume
an AWG at the RN, which (de)multiplex (MUX/DEMUX) these wavelengths.
For this configuration, the transmitter is assumed tunable, so that it can transmit
on any one of the 32 wavelengths, and the receiver is assumed broadband to
receive any one of the 32 wavelengths. For simplicity, we only show
16 Chapter 1

OLT ONU 1 TX
TX
λ1 λ1
(1 G)
RX
RX
(1 G)

TX
fiber TX
(1 G)
RX
(1 G) λ32 λ32 RX
ONU 32
MUX DEMUX MUX/ DEMUX

Figure 1.9: WDM-PON architecture

downstream channels. Further, the upstream and downstream wavelengths can

be assumed in different wavelength bands.

TWDM-PONs
Time and wavelength division multiplexed PONs (TWDM-PONs) [16]
combine both approaches of using higher line rates and of using WDM to
increase the scalability of the PONs. WDM is used to increase the overall
capacity and each wavelength at a higher line rate is shared among multiple users
using TDMA. The ONU needs a tunable transmitter and a broadband receiver;
the frequency range of a broadband receiver is typically narrower than in the
case of WDM-PON as it uses a fewer wavelengths. The ONU also needs a
tunable filter (TF), as multiple wavelengths are available at its input.
TWDM-PONs can use a combination of an AWG and a power splitter at the
RN or just a pure power splitter. More advanced system concepts, e.g., with a
wavelength selective switch (WSS11) [17]-[22] at the RN, have also been
envisioned for TWDM-PON. It will also be shown that TWDM-PON can use
such complex systems due to its high sharing granularity without affecting the
overall cost per user significantly.
A typical configuration of TWDM-PON has been shown in Figure 1.10. We
assume four wavelengths at the OLT with a upstream line rate of 2.5 Gb/s and a
downstream line rate of 10 Gb/s, and a split of 1:512. Further, we assume a
power splitter at the RN.

Other advanced PONs

Orthogonal frequency division multiplexed PON (OFDM-PON), ultra-dense
WDM-PON (U-DWDM-PON) and optical code division multiplexing PON
(OCDM-PON) have also been widely considered in research.
OFDM-PON uses OFDM modulation to increase the robustness of data
transmission on the physical layer. The OLT sends the OFDM broadcast data at

11 WSS is an active switch that can route wavelengths in a flexible manner.

Introduction 17

OLT ONU 1 TX
TX
(10 G)
λ1
RX RX

Splitter
Power
(2.5 G)
TF
TX
fiber
TX
(10 G)
RX
(2.5 G)
λ4 RX
MUX ONU N
MUX/DEMUX /DEMUX TF

Figure 1.10 TWDM-PON architecture

the aggregate bit rate, e.g., 10 or 40 Gb/s, to a passive power splitter from where
the data is transmitted to each ONU. The ONU receives the whole 10 or 40 Gb/s
data stream and selects its part of the data. This can be done by choosing specific
subcarriers (frequency slots) since OFDM allows access to the subcarrier level. It
is also possible to mix OFDM with the time domain, such that the ONU receives
specific subcarriers at specific time slots. Compared to TDMA-PON, OFDM-
PON has a better tolerance with respect to transmission impairments, and
therefore possesses the potential to achieve higher reaches, particularly at 40G or
more. An OFDM-PON for ultra-high capacity converged wireline-wireless
access networks was studied in the framework of the European FP7-project
ACCORDANCE [23].
Another attractive trend for PONs is U-DWDM-PON, in which the frequency
spacing between different channels is reduced to 12.5 GHz 12 or even lower. An
advantage of such narrow spaced channels is that a large number of channels can
be obtained in the same frequency spacing. This increases the capacity of the
PON. The detection of ultra dense channels, however, requires coherent lasers,
and optical heterodyne or homodyne reception [24]. Coherent detection increases
the power budget due to the improved receiver sensitivity; however, it adds cost
and power consumption due to the complex receiver designs.
OCDM-PON uses orthogonal codes, in a similar fashion as is done in the
wide-band code-division multiple access (WCDMA) technology widely used in
mobile communications (e.g., in 3G systems), to address capacity upgrade in
PONs.
Optical coding (OC) can be used for dynamic bandwidth allocation (DBA) in
PONs. OC techniques hold great promise for enabling enhanced real-time DBA
algorithms in PONs. However, the design of OC when the number of users
grows is an open issue. Currently, coders/decoders are in their early stages of

12Note that WDM-PON can be distinguished into two types based on frequency spacing. Frequency
spacing is more than 200 GHz for coarse WDM-PON and is lower than 200 GHz (typically 100/50
GHz) for dense WDM.
18 Chapter 1

development.
Table 1.B gives an overview of the important optical access technologies and
their related standard.

1.3 Evolution strategies for next generation optical

access networks
Optical access networks should be evolved to meet the ever-increasing
bandwidth requirements. This evolution should optimize the network operator’s
investment and would depend on the innovations in the system concepts. For
example, OFDM-PON requires heavy digital signal processing, which is
currently costly and power consuming. With the innovations to follow, however,
OFDM-PON may become a reality. Currently, the optimal NGOA solutions [25]
are actively investigated, e.g., this was recently done in the European FP7-
project OASE (optical access seamless evolution) – where a time frame of up to
2020 was considered – and in the standardization forum FSAN (full service
access network).

1.3.1 Design requirements

The NGOA architectures should also support various other design
characteristics: long reach, reliability, energy efficiency, and open access. The
primary objective of any design feature is to minimize the total cost of
ownership, and to make the network profitable for the operators.

Table 1.B: Important optical access technologies and their related standards
Standard Ethernet EPON GPON 10G- XG-PON TWDM-
PtP EPON PON
Architecture Point to Point to Point to Point to Point to Point to
Point Multi- Multi-point Multi-point Multi-point Multi-point
point
Downstream 100 Mb/s 1.25 1.25 Gb/s 10.3 Gb/s 10 Gb/s 40 Gb/s
data rate 1 Gb/s Gb/s 2.5 Gb/s
Upstream 100 Mb/s 1.25 155 Mb/s 1.25 Gb/s 2.5 Gb/s 10 Gb/s
data rate 1 Gb/s Gb/s 1.25 Gb/s 10.3 Gb/s 10 Gb/s
2.5 Gb/s
Coverage 10 km 10-20 km 20 km 10-20 km 20 km 40 km
(without
amplifiers)
Split ratio not 1:16 1:32, 1:64, 1:16,1:32 1:32, 1:64, 1:256
applic. 1:128 1:128
Introduction 19

1.3.1.a Long reach

Long reach is considered for a higher node consolidation, where a central
access node (CAN) replaces many active COs. The reduction in active network
sites minimizes the operational expenditures (OpEx) for an operator. It also
minimizes the need of aggregation, reducing the number of aggregation switches
and its associated capital expenditures. The evolution of the network toward
longer reach is highlighted in Figure 1.11, where the OLT is moving toward the
network core. The full advantages of a long reach can be tapped by combining it
with a high fan out. A high customer fan out allows a higher sharing of fiber,
components, and rack space of the OLT, thus minimizing the cost and power
consumption of the network per user. The combination of a high fan out (>256)
and a long reach (> 20 km) is a challenging requirement. If a power splitter is
used at the RN, the fan out of the RN increases the insertion loss. This requires
novel RN designs. Note that such designs can be more complex than a power
splitter or an AWG because the complexity of the RN design does not
significantly increase the cost per user due to a high sharing granularity. We
propose the design of such RNs in [14].

1.3.1.b Resilience
Resilience or network protection is a key requirement in the NG-PONs [26]-
[29]. NG-PON faces more challenges in achieving a high reliability performance
than the conventional PON as it has longer fiber lengths with a higher fiber cut
probability, there are more customers on a single PON segment, and it includes
components with a higher complexity (tunability etc) and thus with a poorer
reliability performance.
The network operators have to pay a penalty for the service outage and thus
unprotected architectures have a significant cost of penalty. Protection also
involves duplicating facilities, which comes at a cost and thus the network
protection should only be done when the extra cost in protection compensates for
the penalties. A good strategy is to provide full protection only to business users
20 Chapter 1

Remote
CO node
fiber

Edge
Agg. network OLT ONU
FTTH Node

Edge
Agg. netw. OLT ONU
FTTH Node

Edge
OLT ONU
FTTH Node

Figure 1.11: Evolution of fixed access networks

who seek reliable services or otherwise impel a heavy penalty, but only a partial
protection to residential users who prefer a low cost of service. Another driver of
protected architectures is the fear of negative publicity (negative release on press,
newspapers, TV) that can happen if a large network outage occurs. Thus,
network operators are often more worried about a big failure disconnecting all
clients for 1 hour at the same time than for multiple small failures throughout the
year disconnecting every client for 1 hour on an average.
This confronts us with two major challenges. The first challenge is to design
reliable architectures that provide end-to-end protection for business users and
partial protection to residential users. When designing partially protected
architectures, it is also vital to identify the most unreliable elements. The second
challenge is to propose a new reliability metric that captures the fear of network
operators about large failures.
In this chapter, we propose four protection schemes that vary in the degree of
protection for the business and residential user (chapter 6). We also introduce a
new parameter referred to as failure impact (FI). In the end, we analyze the
protection schemes based on protection coverage, unavailability, FI and cost.

1.3.1.c Energy efficiency

An important criterion for the evaluation of NGOA networks is energy
efficiency. Energy efficiency is important for minimizing the operational
expenditures of network providers as well as for reducing environmental impacts
by minimizing carbon emissions. In current FTTH-based telecommunication
networks, the access segment (including customer premises equipment)
consumes a major fraction (67%) of the energy. Thus, significant energy savings
can be procured by decreasing the energy consumption of access networks.
The ONUs consume the major portion (90%) of the energy consumed in
optical access networks. Thus, several low power modes have been actively
considered at the ONUs, e.g., sleep and doze modes [31]-[32]. Low power
modes have the potential to save a significant amount of energy at the ONUs. In
Introduction 21

these modes, whenever there is no traffic to receive or send, non-essential

functionalities are turned off at the ONU’s end.
The energy consumption at the OLT could be reduced by using transceivers at
a high utilization, leading to higher aggregation levels at the OLT and
minimizing traffic burst. This can be accomplished in many ways. One approach
is to use a high sharing at the RN. The second approach is to share a transceiver
among many PON segments at the OLT. This does not enforce increasing split
ratios in the RNs. The third approach can be applied to multi-wavelength PON
systems, such as TWDM-PON, in which the number of transceivers used could
be made dependent on the network load. All these approaches have significant
potentials to reduce energy consumption. Note that the utility of these
approaches is also linked to how flexible architectures are. The issue of
flexibility is covered in section 1.3.1.e.
In this chapter, we propose a novel algorithm for enabling sleep modes in
EPON (chapter 4). There is a tradeoff between sleep efficiency and QoS,
therefore a major challenge is to maximize the sleep efficiency while delivering
required QoS to the users. We also investigate the effect of increased transceiver
utilization on the energy consumption of the NGOA technologies (appendix A).
Lastly, we investigate the sleep mode potential of TWDM-PON in appendix B.

1.3.1.d Open access

Open access refers to the sharing of a network infrastructure among different
network entities in a non-discriminatory way [33]-[34]. Open access allows
different network entities to participate over a common physical infrastructure.
This has many advantages. First, each network entity does not have to make
huge investments in deploying fiber in the field, and thus has a lower barrier to
enter in the market. This encourages competition and therefore supports new –
open access related – business models to make FTTH networks an economically
viable solution.
Open access ideally refers to the coexistence of network entities with distinct
functionalities (Figure 1.12 model A):
 Physical infrastructure provider (PIP) − responsible for installation of the
physical infrastructure (implying trenches, conduits, ducts, fiber, housing).
 Network provider (NP) − responsible for all active equipment between the
users and the CO.
 Service provider (SP) – supply of services (telephony, IPTV, broadband
Internet, mobile backhauling).
In reality, however, one entity can also act to provide two (or even three)
functional roles. For example, an entity can act as both SP and NP e.g., (Figure
1.12 model B, E and F). This means that we can have multiple scenarios ranging
from pure open access to incumbency consisting of a vertically integrated
operator (Figure 1.12).
22 Chapter 1

Open access can be offered at different layers (Figure 1.13) depending on how
a user selects a specific network entity, e.g., by the selection of a fiber,
wavelength, or a packet field (Ethernet address, VLAN tag, MPLS, IP). This
classifies open access as fiber, wavelength, and bit-stream open access. Fiber and
wavelength open access open PIP-NP interface, and bit-stream open access
opens NP-SP interface. While bit-stream open access can be offered using
logical layers, fiber and wavelength open access require new architectures. We
design such architectures to support open access on fiber and wavelength layer,
and evaluate their cost (chapter 7).
Open Access to Incumbency Models

SP SP SP SP SP SP SP SP SP SP SP SP SP SP SP
SP SP SP
SP & NP & & &
NP NP NP NP
NP NP NP & NP
SP Vertically Vertically
Integrated Integrated
PIP & NP Vertically Operator Operator
Integrated
PIP PIP PIP PIP Operator with bit
stream access ”All-in-a-box”

A B C D E F G H

Figure 1.12: Different open access models

Layer Shared elements

Fibre 1(Physical Layer Dark fibre

PIP-NP

Open Access)
Wavelength Optical Layer (Coarse WDM or
Dense WDM)

Bit stream 2 (Data Link Dark fibre and link-layer

Layer Open electronics at each end.
NP-SP

Access) E.g., Ethernet-based VLAN, or

ATM-based PVCs.
3 (Network Layer Basic network service provided.
Open Access) E.g., IP Layer service over cable to
support MPLS-based VPN.

Figure 1.13: Different open access layers

1.3.1.e Flexibility
The architecture may share the resources (i.e., wavelengths and time slots) in a
static manner or may provision a more dynamic sharing of the network
resources. The architectural capability by which all users dynamically share the
network resources is referred to as flexibility [19]-[21]. In TDMA-PON, the
Introduction 23

network resource is one wavelength and all users dynamically share this
wavelength. Thus, the TDMA-PON architecture is fully flexible. In TWDM-
PON, the network resource is a set of wavelengths, and if all users have access to
all wavelengths at all time instances, then the architecture is fully flexible,
whereas if all users have access to just one wavelength, then the architecture is
inflexible or static.
Flexibility increases the gains of statistical multiplexing, which increases
system efficiency, leading to several performance improvements, like in the
delay performance, congestion avoidance, etc. The other important advantages
that can be accrued because of flexibility are network planning, migration, and
energy efficiency. This is because the resources can be flexibly reconfigured
according to the network load, user’s demand, etc. For the same performance, a
flexible architecture requires a lower capacity (which can translate to fewer
transceivers) compared to a static architecture, especially at a light load. Thus,
some of the capacity can be ideally switched off, minimizing energy
consumption. An illustrative example is provided in Figure 1.14 to demonstrate
this. The left part of Figure 1.14 shows that over time, users who require almost
no services (e.g., business users during the nighttime hours) can be reconnected
to a particular wavelength. In this scenario, the wavelength can be reallocated as
shown in the right-hand part of Figure 1.14 and this provides a mean to turn off
some of the OLT transceivers as well as line cards, to enhance energy savings at
the OLT over time. This might help the operator to build its network greener.
The same scheme also depicts the possibility of dynamic allocation of
wavelengths amongst users according to their traffic needs.
Even if the flexible architectures appear as the only way to go, there are
challenges and problems associated with designing a flexible architecture
(chapter 8). For example, a power splitter is a fully flexible component, but
increases insertion loss and security concerns. Thus, there is a need to quantify
the advantages and disadvantages of flexibility in order to design the architecture
embedded with the optimal degree of flexibility.

1.3.2 Evolution trend

Optical access networks are being evolved toward higher capacity, while
maximizing the investment of network operators. The first generation of NGOA
networks were EPON, GPON, and Ethernet PtP. EPON is widely deployed in
Asia and North America, whereas GPON is deployed in Europe. Ethernet PtP is
deployed in Sweden. Later on, these systems (especially EPON and GPON) were
upgraded to higher line rate PONs like 10G-EPON and XG-PON. These
24 Chapter 1

Turn off

OLT OLT

Figure 1.14: Illustration of an energy efficiency scenario for flexible TWDM-

PONs
generations of PONs are usually referred to as NG-PON. As the bandwidth
requirement continues to grow, there are large research efforts to innovate
technology for the future, in this regard, the FSAN standardization forum and the
OASE (optical access seamless evolution) research project are important to
mention. FSAN has recently chosen TWDM-PON as a primary candidate of
NGPON-2 and WDM-PON as the secondary candidate of NGPON-2. WDM-
PON was chosen for scenarios with high QoS requirements, as WDM-PON does
not suffer from typical QoS performance issues of TDMA-PONs. The OASE
project also advocated WDM- and TWDM-PON. The other PONs like 40G
TDMA-PON, OFDM-PON, and UDWDM-PON will also be feasible in the
future. The development of these technologies is purely linked to the
development of their associated systems.
Figure 1.15 presents the timeline of optical access technologies as forecasted
by the OASE project. The time denotes by when the technologies can be
standardized.

1.4 Medium access control (MAC) protocols and

dynamic bandwidth allocation (DBA) algorithms
The bandwidth allocation to users in an efficient and a fair way is an important
challenge, especially in a logical point-to-multipoint system. Hence, MAC
protocols are employed for facilitating the bandwidth distribution among users.
MAC protocols specify the exchange of control messages between the OLT and
the ONU, and DBA algorithms are built on top of it to distribute the bandwidth
among users. MAC protocols are different for different technologies. Even
TDMA-PONs, like EPON and GPON, have some important differences. For
example, EPON does not allow the fragmentation of packets, whereas GPON
does. This creates some important differences in the design of the MAC
protocols, and consecutively, the DBA algorithms. We will further point out
these differences in section 1.4.2. WDM-PON gives a dedicated wavelength per
Introduction 25

user, and thus it poses a fewer challenges in the design of a MAC protocol. In
WDM-PON, the MAC protocols can still be used for enabling sleep modes, so
that the ONU does not need to be awake at all times for receiving packets.

GPON
TWDM-PON 40G TDM-PON UDWDM-PON
EPON

2005 2010 2015 2020 2025

XG-PON
WDM-PON OFDM-PON
10G-EPON

Figure 1.15: The timeline of optical access technologies

TWDM-PON combines the bandwidth-scheduling problem of TDMA-PON
with an additional dimension of wavelength allocation. The joint scheduling
problems are generally much more complex than the separated time and
wavelength allocation problems. On the other hand, OFDM-PON requires a
MAC protocol for allocating sub-channels among users. In CDMA-PON,
multiple ONUs can transmit simultaneously using OCDMA. The number of
ONUs that can transmit simultaneously, however, is limited due to the multiple
access interference, and thus the transmission of some ONUs is delayed. A MAC
protocol is required to minimize this performance degradation. The challenges
for other PON architectures can be derived from these basic challenges. As much
of the work in this chapter is focused on EPON, we will describe the EPON
based MAC protocol in more detail in section 1.4.1.
Together with the MAC protocols, suitable DBA algorithms are also required.
An optimal design of DBA algorithms is needed to satisfy QoS bounds, to
maintain energy efficiency, and to satisfy the service level agreements (SLAs) of
different SPs in the case of multiple SPs per PON. We will discuss the
challenges of a DBA algorithm in section 1.4.2. Like for MAC protocols, we
focus only on the DBAs for TDMA-based PONs.

1.4.1 Medium access control protocol

In this section, we describe the MAC protocol employed for EPON.
Conventional LAN mechanisms like carrier sense multiple access with collision
detection (CSMA/CD) or CSMA with collision avoidance (CSMA/CA) are
inefficient because the users will not be able to detect the collision that takes
place after the power splitter and thus the OLT has to detect the collision and
convey this information to the users. Nevertheless, because of the long
propagation distances (typically up to 20 km) between the users and the OLT,
these schemes will deteriorate the performance. Hence, a solution is to employ
centralized TDMA based MAC protocols.
An EPON employs a multi-point control protocol (MPCP) to arbitrate the
26 Chapter 1

access to the shared medium in order to avoid data collisions in the upstream
direction. The MPCP arbitration mechanism is developed by the IEEE 802.3ah
Task Force. The MPCP provides mechanisms for auto-discovery, ranging, and
registration for newly added ONUs, and a signaling infrastructure for
coordinating data transmissions from the ONUs to the OLT. MPCP uses two
types of control messages to facilitate bandwidth allocation: REPORT and
GATE.
Each ONU holds Ethernet frames in its queue, and reports its queue depth
using a REPORT message. A REPORT message can report the queue size of up
to 13 queues. Upon receiving a REPORT message, the OLT uses a DBA
algorithm to calculate the transmission slots of the ONU. Note that MPCP
provides a framework for the implementation of different DBA algorithms, but
does not specify any particular DBA algorithm. After executing the DBA
algorithm, the OLT transmits a GATE message to the ONU, which contains the
allocated grant size of the ONU. Each GATE message can support up to four
transmission grants. Each transmission grant contains the transmission start time
and transmission length of the corresponding ONU. Each ONU updates its local
clock using the timestamp contained in each received transmission grant, and
maintains its synchronization. Each ONU then transmits its backlogged traffic
according to the granted transmission slot and its REPORT message. Note that
an ONU can append the REPORT message before or after the data transmission.
These steps are also depicted in Figure 1.16 for a three ONU system, assuming
that the ONU transmits its REPORT after data. The OLT first issues a GATE
message, and then the ONU transmits the allocated bandwidth and REPORTs its
new bandwidth arrival. After receiving the REPORT message, the OLT
consequently updates its table, which is then supplied to a DBA algorithm to
calculate the next grant for the ONU. Note that there is a guard time between two
successive ONU’s transmission.
EPON does not allow packet fragmentation, and thus some bandwidth may go
waste if the granted size is not what was requested. For example, if the ONU
requests 600 bytes for two packets waiting in its queue of size 400 bytes and 200
bytes, but the OLT grants 500 bytes, then only the first packet will be
transmitted. This means that 100 bytes will go waste, as the packet of size 200
bytes cannot be fragmented. These unused time slots (e.g., 100 bytes in this case)
are referred to as unused slot remainders (USRs). Some additional points to be
mentioned are that the ONU does not necessarily have to reply to a GATE
message. Thus if the ONU is in sleep, it may ignore any GATE message
transmitted during its sleep period. MPCP however limits the maximum non-
response time of an ONU to 50 ms. If the ONU does not response in this period,
it is deregistered from the network. The time to register back in the network can
be 10 s long.
Introduction 27

1.4.2 Dynamic bandwidth allocation algorithms

The primary challenge in the design of DBA algorithms is to ensure a high
QoS, Figure 1.17. QoS requires bounds on average/maximum delay,
average/maximum delay variation, fairness, and high throughput [35]. Together
with a high QoS delivery, several DBA algorithms have focused on energy
efficiency.
We will first discuss DBA algorithms in the context of TDMA-based PONs,
specifically EPON.
 ONU Bytes RTT ONU Bytes RTT ONU Bytes RTT

1 6000 120 1 550 120 1 550 120

2 3200 170 2 3200 170 2 570 170
3 1800 200 3 1800 200 3 1800 200

6000 3200 1800

Tx
OLT
Rx 6000 Bytes 550 3200 Bytes 570

6000 Bytes 550 guard band

Tx
ONU1
Rx 6000

3200 Bytes 570

Tx
ONU2
Rx 3200

Tx 1800 Bytes 2700

ONU3
Rx 1800

6000 : GATE message 6000 Bytes : DATA 550 : REPORT

Figure 1.16: REPORT and GATE exchange in EPON

 Average
/Maximum
Required for mainly
voice applications and Delay Required for voice and
emerging video Bound video applications
applications

Delay
Variation QoS Fairness
Bound

Ensures low packet Fairness must be for

loss rate; important for High all traffic classes and
all traffic classes users
Throughput Involves SLA and
Interoperability issues

Figure 1.17: QoS requirements for DBA algorithms

In EPON, because of the broadcasting in the downstream direction, the

downstream traffic can be transmitted on a first-come-first-serve basis. This
approach, however, has a disadvantage. Because the ONUs do not know when
their packets would arrive, they have to be awake at all times, and thus there is
no opportunity for the ONUs to sleep. Thus, the DBA algorithms must transmit
the downstream traffic in an energy efficient way. We will explore this problem
in detail in Chapter 4.
On the other hand, for the upstream traffic, a fixed bandwidth allocation can
be adopted in which the ONUs transmit in their fixed time slots. It, however,
shows a very poor performance as the access traffic is highly bursty [36]-[37]
and much of the bandwidth assigned to one ONU can be left unutilized. This
scheme will thus degrade channel utilization and increase delay.
Thus, the DBA algorithms employ centralized polling mechanisms in which
the OLT arbitrates the bandwidth allocation by polling the ONUs in a round
robin manner. To increase the bandwidth efficiency in these protocols, the
polling is interleaved, i.e., the next ONU is polled before the transmission from
the previous one has arrived. An example of polling in EPON is the exchange of
GATE and REPORT messages between the OLT and the ONUs. The OLT
request the bandwidth demand by issuing a GATE message to the ONU, and the
30 Chapter 1

ONU reports the bandwidth (queue size) by transmitting back a REPORT

message. Based on this information, the OLT determines bandwidth allocation
for each ONU.
The simplest approach for the bandwidth allocation is referred to as gated
scheme, which means that the OLT grants whatever an ONU requests. This
approach, however, has serious security concerns. A malicious user can hog the
entire bandwidth by requesting for very large grant sizes, influencing the
performance of other ONUs. Thus, this scheme is not fair, and does not assure
any QoS. For example, a heavily loaded ONU can prevent another ONU from
getting its fair share. Therefore, other schemes like limited are proposed in which
an OLT grants up to some maximum transmission window. This assures that
every ONU accesses bandwidth within a definite time. This is also very
important for serving priority traffic. The priority traffic like voice is very
sensitive to delay and thus it is required that the priority packets are transmitted
within a definite time. The limited scheme assures that the cycle length − the
maximum time interval within which every ONU is polled at least once − is
limited, allowing priority traffic to be transmitted within a definite time. The
cycle length should be chosen according to the delay bound of the priority traffic.
For example, if the maximum delay tolerance of voice traffic is 2 ms, then the
cycle length should be 2 ms.
In addition, note that the bandwidth is allocated by a centralized mechanism in
which the OLT assigns bandwidth to every ONU. Thus, the bandwidth is
allocated in rounds, where the length of each round depends on the round trip
time (RTT) between the OLT and the ONU. This is of no consequence for the
first generation PONs, in which the distance between the OLT and the ONU is
up to a maximum of 20 km, and thus the RTT is limited up to 200 µs.
Nevertheless, as the distances between the OLT and the ONUs are increasing,
the RTT becomes a critical problem, and the DBA algorithms face new
challenges. For example, when the distance between the OLT and the ONU is
100 km, the RTT is 1 ms, which is even more than the maximum allowed delay
bound at the ONU interface for the voice traffic [35]. We focus on these
problems in Chapter 2 and Chapter 3.
GPON has fewer challenges in the resource allocation than EPON. This is
because the packets in GPON can be fragmented and thus there is no wastage
corresponding to the USRs formation. On the other hand in EPON, the
maximum USR of up to 1518 B (the maximum size of an Ethernet packet) can
be formed. The percentage of USRs formed will be higher if the ONUs are
issued smaller transmission slots, because even then a USR of 1518 B can be
formed but over a smaller allocated bandwidth. Thus in EPON, the maximum
transmission slot per ONU should be sufficiently long. This enlarges the
maximum cycle length, and as the delay is proportional to the cycle lengths, the
Introduction 31

DBAs for EPON experience a higher delay compared to GPON’s DBA.

In case of next generation PON technologies, there are new challenges. For
example in TWDM-PON, the bandwidth scheduling is a two dimensional
problem, in which in addition to the typical dimension of the time domain, a new
problem dimension is “which wavelengths should be allocated to which ONU
groups”. When selecting the wavelength and the ONUs, many selection criteria
can be adopted, like maintaining the same load on each wavelength, or allocating
the wavelength which has earliest finished transmission. Another interesting
approach is to use the number of wavelengths according to the network load
(chapter 8, and appendix B). This presents interesting avenues for minimizing
energy as some transceivers can be switched off at a light load. Nevertheless,
when the wavelengths are switched off, the load per remaining wavelengths
increases, degrading the delay performance. Hence, there is a trade-off between
the delay performance and energy efficiency. Even so, a typical delay analysis
for the PONs shows that the delay on a wavelength does not increase
significantly if the normalized13 load does not increase beyond 0.7. Thus, the
number of wavelengths can be chosen appropriately. For example, if the number
of wavelengths is four and the load per wavelength is 0.2, then at least two
wavelengths can be switched off, leading to a normalized load of 0.4 per
remaining two wavelengths. Another challenge in this dynamic wavelength
allocation is to confront with the penalties of tuning time (appendix C). When an
ONU is switched from one wavelength to another, it introduces another overhead
of tuning time. If this tuning time is long (in ms), a lot of bandwidth would be
wasted if the ONUs hop onto another wavelength per cycle (about 2 ms). Thus, a
long tuning time prohibits almost any wavelength switching. Conversely, if the
tuning time is short (in ns), the wavelengths can be steered almost in every cycle.

13 This is the load per channel capacity. For example, a load of 500 Mb/s on a 1 Gb/s wavelength,
translates to a normalized load of 0.5.
32 Chapter 1

References
[1] A. S. Tanenbaum, Computer Networks, New Jersey: Prentice Hall, 2003.
[2] 249025-ICT OASE Project, D2.1: Requirements for European Next-
Generation Optical Access Networks.
[3] S. Haykins, Communication Systems, New Jersey: Prentice Hall, 2003.
[4] DSL Forum ([Link]
[5] C. Lam, Passive optical networks: principles and practice, Academic
Press, 2011.
[6] P. Odling, T. Magesacher, S. Host, P.O. Borjesson, M. Berg, and E.
Areizaga, “The fourth generation broadband concept,” IEEE
Communications Magazine, vol. 47, no. 1, pp. 62 – 69, 2009.
[7] P. E. Green, “Fiber-to-the-Home: the next big broadband thing,” IEEE
Communications Magazine, vol. 42, no. 9, pp. 100-106, 2004.
[8] Fiber to the Home Council, [Link]
[9] T. Koonen, “Fiber to the home/fiber to the premises, what, when and
where ?,” Proceedings of the IEEE, vol. 94, no. 5, pp. 911-934, 2006.
[10] IEEE Stds. 802.3ah, Ethernet First Mile (EFM); 802.3av, 10GEPON.
[11] ITU-T Recs. G.983, BPON; G.984, GPON.
[12] E. Harstead, D. Van Veen, and P. Vetter, “Technologies for NGPON2:
why I think 40 G TDM PON (XLG-PON) is the clear winner,” in
Optical Fiber Communication (OFC), 2012.
[13] K. Grobe, M. Roppelt, A. Autenrieth, J. P. Elbers, and M. Eiselt, “Cost
and energy consumption analysis of advanced WDM-PONs,” IEEE
Communications Magazine, vol. 49, no. 2, pp. s25-s32, 2011.
[14] A. Banerjee, Y. Park, F. Clarke, H. Song, S. Yang, G. Kramer, K. Kim,
and B. Mukherjee, “Wavelength-division-multiplexed passive optical
network (WDM-PON) technologies for broadband access: a review
[Invited],” Journal of optical networking, vol. 4, no. 11, pp. 737-758,
2005.
[15] A. Dixit, B. Lannoo, G. Das, D. Colle, M. Pickavet, and P. Demeester,
“Flexible TDMA/WDMA passive optical network: energy efficient next-
generation optical access solution,” Optical Switching and Networking,
vol. 10, no. 4, pp. 491-506, 2013.
[16] Y. Luo, X. Zhou, F. Effenberger, X. Yan, G. Peng, Y. Qian, and Y. Ma,
“Time and wavelength division multiplexed passive optical network
Introduction 33

(TWDM-PON) for next generation PON stage 2 (NG-PON2),” IEEE

Journal of Lightwave Technology, vol. 31, no. 4, pp. 587 – 593, 2013.
[17] B. Lannoo, G. Das, A. Dixit, D. Colle, M. Pickavet, and P. Demeester,
“Novel hybrid WDM/TDM PON architectures to manage flexibility in
optical access networks,” Telecommunications System, vol. 54, no. 2, pp.
147-165, 2013.
[18] G. Das, B. Lannoo, A. Dixit, D. Colle, M. Pickavet, and P. Demeester,
“Flexible hybrid WDM/TDM PON architectures using wavelength
selective switches,” Optical Switching and Networking, vol. 9, no. 2, pp.
156–169, 2012.
[19] A. Dixit, B. Lannoo, G. Das, D. Colle, M. Pickavet, and P. Demeester,
“Evaluation of flexibility in hybrid WDM/TDM PONs,” in IEEE
International Symposium on Access Spaces (ISAS), 2011.
[20] B. Lannoo, G. Das, A. Dixit, D. Colle, M. Pickavet, and P. Demeester,
“Flexible hybrid WDM/TDM PON: architectures, MAC Protocols and
techno-economic evaluation,” in Asia Communications and Photonics
Conference and Exhibition (ACP), 2010.
[21] A. Dixit, B. Lannoo, G. Das, D. Colle, M. Pickavet, and P. Demeester,
“Flexibility evaluation of hybrid WDM/TDM PONs,” in IEEE Advanced
Networks and Telecommunications System (ANTS), 2011.
[22] A. Dixit, B. Lannoo, D. Colle, M. Pickavet, and P. Demeester,
“Wavelength switched hybrid WDM/TDMA PON: a flexible next-
generation optical access solution,” in IEEE International Conference on
Transparent Optical Networks (ICTON), 2012.
[23] K. Kanonakis, I. Tomkos, T. Pfeiffer, J. Prat, P. Kourtessis,
“ACCORDANCE: a novel OFDMA-PON paradigm for ultra-high
capacity converged wireline-wireless access networks,” in IEEE
International Conference on Transparent Optical Networks (ICTON),
2010.
[24] J. M. Fabrega, L. Vilabru, and J. Prat, “Experimental demonstration of
heterodyne phase-locked loop for optical homodyne PSK receivers in
PONs,” in IEEE International Conference on Transparent Optical
Networks (ICTON), 2008.
[25] L. Kazovsky, S. Wong, S. Yen, V. Gudla, P. Afshar, and D. Larrabeiti,
“Next-generation optical access networks: goals, challenges and research
opportunities,” in Acreo/KTH Seminar, 2010.
[26] M. Mahloo, A. Dixit, J. Chen, C. M. Machuca, B. Lannoo, D. Colle, L.
Wosinska, “Towards reliable hybrid WDM/TDM passive optical
34 Chapter 1

networks,” IEEE Communications Magazine, vol. 52, no. 2, pp. S14 -

S23, 2014.
[27] A. Dixit, J. Chen, M. Mahloo, B. Lannoo, D. Colle, and M. Pickavet,
“Efficient protection schemes for hybrid WDM/TDM passive optical
networks,” in International Conference of Communications (ICC), 2012.
[28] A. Dixit, B. Lannoo, D. Colle, M. Pickavet, and P. Demeester, “Trade-
off between end-to-end reliable and cost-effective TDMA/WDM passive
optical networks,” in IEEE International Congress on Ultra Modern
Telecommunications and Control Systems and Workshops (ICUMT),
2012.
[29] A. Dixit, B. Lannoo, D. Colle, M. Pickavet, and P. Demeester,
“Protection strategies for next generation passive optical networks -2,” in
Optical Networking and Design Modeling (ONDM), 2014.
[30] IEEE 802.3av, “Power saving adhoc report,” 2008.
[31] A. Dixit, B. Lannoo, G. Das, D. Colle, M. Pickavet, and P. Demeester,
“Evaluation of ONU power saving modes in next generation optical
access networks,” in European Conference on Optical Communication
(ECOC), 2012.
[32] A. Dixit, B. Lannoo, D. Colle, M. Pickavet, and P. Demeester, “ONU
power saving modes in next generation optical access networks:
progress, efficiency and challenges,” Optics express, vol. 20, no. 26, pp.
B52-B63, 2012.
[33] A. Dixit, M. Van der Wee, B. Lannoo, D. Colle, S. Verbrugge, M.
Pickavet and P. Demeester, “Fiber and wavelength open access in
WDM-and TWDM passive optical networks,” IEEE Network Magazine,
accepted.
[34] B. Lannoo, A. Dixit, D. Colle, and M. Pickavet, “Architectural solutions
for open access in next generation passive optical networks (NG-
PONs),” in Asia Communications and Photonics Conference and
Exhibition (ACP), 2012.
[35] A. Dixit, B. Lannoo, G. Das, D. Colle, M. Pickavet, and P. Demeester,
“Dynamic bandwidth allocation with SLA awareness for QoS in
Ethernet passive optical networks,” IEEE/OSA Journal of Optical
Communications and Networking, vol. 5, no. 3, pp. 240-253, 2013.
[36] W. Willinger, M. Taqqu, R. Sherman, and D. Wilson., “Self-similarity
through high variability: statistical analysis of Ethernet LAN traffic at
the source level,” IEEE/ACM Transactions on Networking, vol. 5, no. 1,
pp. 71-86, 1997.
Introduction 35

[37] W.E Leland, M. Taqqu, W. Willinger, and D. Wilson., “On the self-
similar nature of Ethernet traffic,” IEEE/ACM Transactions on
Networking, vol. 2, no. 1, pp. 1-15, 1994.