i

Long-Reach Passive Optical Networks

By

Huan Song

B. Tech., Electrical Engineering (UESTC) 2001

M.S., Computer Science (UC Davis) 2006

DISSERTATION

Submitted in partial satisfaction of the requirement for the degree of

DOCTOR OF PHILOSOPHY

in

COMPUTER SCIENCE

in the

OFFICE OF GRADUATE STUDIES

of the

UNIVERSITY OF CALIFORNIA, DAVIS

Approved:

Biswanath Mukherjee (Chair)

Dipak Ghosal

Xin Liu

Committee in Charge

2009

ii

Acknowledgements

My deepest gratitude goes to my advisor and committee chair Professor Biswanath Mukherjee. He not

only guided me throughout the course of my graduate study, but also guided my attitude towards career

and life. His extensive knowledge, experience and exceptional ability to find new approaches for

difficult problems were pivotal in this work and my development as a researcher. This work would not

have been completed without his encouragement and patience. I feel privileged to have had the

opportunity to study under his guidance.

I am grateful to Professor Dipak Ghosal and Professor Xin Liu for serving on my PhD dissertation

committee and PhD qualifying examination committee. My understanding of the subject material

benefited immensely through my interactions with them. I thank Professor Norman S. Matloff and

Professor Demet Aksoy for serving on my PhD qualifying examination committee.

I am indebted to Professor Youngil Park of Kookmin University, Seoul, Korea, and Dr. Amitabha

Banerjee, our lab’s alumnus and currently at Sun Microsystems, for their valuable guidance and help

during my PhD research and study. I am thankful to Dr. Byoung-Whi Kim, Ms. Sunhee Yang, and Mr.

Dong-Min Seol at Electronics and Telecommunications Research Institute, Daejeon, Korea for their

guidance and collaboration. I am also grateful to Mr. Davide Cuda of Politecnico di Torino, Italy, for

some nice research collaboration.

I thank the faculty of Computer Science and Electrical and Computer Engineering Departments at

the University of California, Davis, for their superb teaching and guidance. I thank the office staff and

systems support for all their assistance.

I gratefully acknowledge the support of NSF and Electronics and Telecommunications Research

Institute, Daejeon, Korea for the continuous support and funding in this research.

I thank all members of the UC Davis Networks Research Labs for creating a dynamic and

collaborative culture for research and study. Many thanks to Mohammed Al-Baijat, Dragos Andrei, Dr.

Amitabha Banerjee, Marwan Batayneh, Prantik Bhattacharyya, Cieck Cavdar, Joon-Ho Choi, Pulak

iii

Chowdhury, Frederick Clarke, Dr. Anpeng Huang, Sheng Huang, Dr. Yurong Huang, Yali Liu, Avishek

Nag, Dr. Canhui Ou, Vishwanath Ramamurthi, Abu (Sayeem) Reaz, Rajesh Roy, Dr. Suman Sarkar, Dr.

Lei Song, Lei Shi, Ming Xia, Professor Myungsik Yoo, and Yi Zhang for their contribution to our group

and my research, technical expertise and general camaraderie.

Finally, and most importantly, I cordially thank my parents for their education and support, and I

am grateful to my wife for her love and understanding, all through my life.

iv

To my family, for the endless love, support and encouragement.

v

Huan Song

Long-Reach Passive Optical Networks

Abstract

With the advances in optical technology, the span of a broadband access network using Passive Optical

Network (PON) technology can be increased from today’s standard of 20 km to 100 km or higher. Such

an extended-reach PON is known as Long-Reach PON (LR-PON). This technology can enable

broadband access for a large number of customers in the access/metro area, while decreasing capital and

operational expenditures for the network operator. Therefore, it is very desirable to comprehensively

investigate this technology for future broadband access. This dissertation is dedicated to the research of

architecture, management, and reliability of LR-PON.

This dissertation first reviews the evolutionary path of access networks and shows the drivers from

technology and business perspectives for high bandwidth and low cost. A variety of research challenges

in this field is reviewed, from optical components in the physical layer to the control and management

issues in the upper layers. We discuss the requisites for optical sources, optical amplifiers, and optical

receivers in optical access networks with high transmission rate (10 Gbps) and large power attenuation

(due to large split, transmission over 100 km and beyond, and propagation). We analyze the key

topological structures to guarantee physical protection (e.g., tree-and-branch, ring-and-spur). Then,

some relevant demonstrations of Long-Reach optical access networks developed worldwide by

different research institutes are presented.

A major challenge in LR-PON is that the propagation delay (for data as well as control signals)

between the telecom central office (CO) and the end user is increased by a very significant amount. Now,

traditional PON algorithms for scheduling the upstream transmission, such as dynamic bandwidth

allocation (DBA) algorithms, may not be sufficient; actually, they may lead to degraded performance

because of the long delay of the CO-to-Users “control loop”. This challenge motivates us to propose

June 2009

Computer Science

vi

and study a multi-thread polling algorithm to effectively and fairly distribute the upstream bandwidth

dynamically. This algorithm exploits the benefits of having multiple polling processes running

simultaneously and enabling users to send bandwidth requests before receiving acknowledgement from

the CO. We compare the proposed algorithm with traditional DBA, and show its advantage on average

packet delay.

With the increased bandwidth requests from the expanding base of users, LR-PON should utilize

the network resource (e.g., wavelengths, lasers, etc.) more effectively. We propose a new and efficient

protocol to achieve better utilization of tunable lasers, as well as wavelength resources across different

user groups in LR-PON. In order to accommodate downstream bursty traffic and provide Quality of

Service (QoS) in the user-specified Service-Level Agreements (SLA), the protocol integrates our

proposed SLA-aware bandwidth allocation scheme based on flow scheduling. We show the protocol’s

advantage to support incremental upgrade of bandwidth with increasing user bandwidth requests, and to

provide a user with a SLA which guarantees a number of streaming flows with average bandwidth and

maximum delay guarantee (e.g., 5 ms), as well as data flows with average bandwidth specifications.

Since LR-PON serves a lot more users, a network failure may lead to a huge amount of data loss

and negative user experience. Thus, to understand the importance of LR-PON survivability, we propose

hardware-accelerated protection schemes for the LR-PON, incorporating the “ring-and-spur” structure

to achieve fast protection-switching time, and automatic failure location at the Optical Line Terminal

(OLT), which is located at the head end of the optical access network. We design the protection schemes

for multiple network environments, e.g., unidirectional transmission vs. bidirectional transmission, and

1+1 protection vs. 1:1 protection. The numerical examples demonstrate that protection paths can be

established within a few tens of ms after a failure occurs.

This dissertation makes important contributions by studying novel architectures, protocols, and

algorithms of LR-PON that will help to improve the next-generation telecom access networks.

vii

Contents

List of Figures

List of Tables

1 Introduction...........................................................................................................1

1.1 Introduction to Passive Optical Network (PON) ....................................................................1

1.1.1 Existing Broadband Access Solutions ......................................................................2

1.1.2 Long-Reach Passive Optical Network (LR-PON) ....................................................6

1.1.3 Resource Allocation and Bandwidth Scheduling......................................................8

1.1.4 Protection in LR-PON.............................................................................................10

1.2 Research Contributions.........................................................................................................10

1.2.1 A Survey of Research Challenges, Demonstrations, and Bandwidth Assignment

Mechanisms in LR-PON .....................................................................................................11

1.2.2 Multi-Thread Polling: A Dynamic Bandwidth Distribution Scheme in LR-PON ..11

1.2.3 A SLA-Aware Protocol for Efficient Tunable Laser Utilization to Support

Incremental Upgrade in LR-PON........................................................................................12

1.2.4 Protection in LR-PON.............................................................................................12

1.3 Organization .........................................................................................................................12

2 Long-Reach Optical Access Networks: A Survey of Research Challenges,

Demonstrations, and Bandwidth Assignment Mechanisms..................................14

2.1 Introduction ...........................................................................................................................14

2.2 Research Challenges .............................................................................................................17

2.2.1 Signal Power Compensation.................................................................................18

2.2.2 Optical Source .......................................................................................................19

viii

2.2.3 Burst-Mode Receiver ............................................................................................20

2.2.4 Upstream Resource Allocation..............................................................................21

2.2.5 Topology and Protection .......................................................................................21

2.3 Demonstrations of LR-PON.................................................................................................22

2.3.1 PLANET SuperPON...............................................................................................22

2.3.2 Demonstrations from British Telecom (BT) ...........................................................24

2.3.3 Demonstrations from University College Cork, Ireland .........................................27

2.3.4 Demonstrations of “Ring-and-Spur” LR-PON .......................................................30

2.3.5 Other Demonstrations .............................................................................................31

2.4 Dynamic Bandwidth Assignment (DBA) .............................................................................33

2.5 Conclusion............................................................................................................................36

3 Multi-Thread Polling: A Dynamic Bandwidth Distribution Scheme in

Long-Reach PON ......................................................................................................38

3.1 Introduction ..........................................................................................................................38

3.2 Multi-Thread Polling ............................................................................................................42

3.2.1 Applying Previous Approaches in LR-PON ...........................................................42

3.2.2 Idea of Multi-Thread Polling Algorithm.................................................................45

3.2.3 Multi-Thread Polling Algorithm.............................................................................47

3.2.4 Control Frame Design.............................................................................................49

3.2.5 Initiating and Tuning Multiple Threads ..................................................................50

3.2.6 Inter-Thread Scheduling .........................................................................................52

3.2.7 Achieving Fairness among ONUs...........................................................................52

3.2.8 Analysis ..................................................................................................................54

3.3 Illustrative Numerical Results..............................................................................................57

3.4 Conclusion............................................................................................................................62

4 A SLA-Aware Protocol for Efficient Tunable Laser Utilization to Support

Incremental Upgrade in Long-Reach Passive Optical Networks.........................63

4.1 Introduction..........................................................................................................................63

ix

4.2 Dynamic Wavelength Allocation.........................................................................................67

4.3 User-SLA-Aware Bandwidth Allocation .............................................................................70

4.3.1 Flow Classification and User-Defined SLA ...........................................................70

4.3.2 User-SLA-Aware Bandwidth Allocation Algorithm ..............................................70

4.4 Analysis ...............................................................................................................................74

4.5 Ilustrative Numerical Examples ...........................................................................................76

4.6 Conclusion.............................................................................................................................81

5 Protection in Long-Reach Broadband Access Networks ................................82

5.1 Introduction ..........................................................................................................................82

5.2 Hardware-Accelerated Unidirectional Protection.................................................................84

5.2.1 1+1 Unidirectional Protection.................................................................................84

5.2.2 1:1 Unidirectional Protection ..................................................................................89

5.3 Hardware-Accelerated Bidirectional Protection……………………....……………………92

5.4 Illustrative Numerical Examples ..........................................................................................95

5.4.1 Power Budget..........................................................................................................95

5.4.2 Cost and Network Capacity ....................................................................................97

5.4.3 Protection-Switching Time .....................................................................................97

5.5 Conclusion............................................................................................................................99

6 Conclusion .........................................................................................................101

6.1 A Survey of Research Challenges in LR-PON ...................................................................101

6.2 Multi-Thread Polling in LR-PON.......................................................................................102

6.3 A SLA-Aware Protocol in LR-PON...................................................................................103

6.4 Protection in LR-PON........................................................................................................103

x

List of Figures

1.1 Passive Optical Network deployment scenarios (courtesy Dr. Glen Kramer)……………5

1.2 LR-PON simplifies the telecom network…………………………………………………7

2.1 Long-Reach PON (LR-PON) architecture…………………………………...………….18

2.2 ACTS-PLANET architecture and transport system……………………………………...22

2.3 Long-reach optical access network by British Telecom…………………………………24

2.4 Experimental configuration of GPON extended to 135 km via WDM………………….25

2.5 Hybrid DWDM-TDM LR-PON architecture……………………………………………27

2.6 PIEMAN hybrid WDM/TDMA architecture…………………………………………....28

2.7 WE-PON in FTTH topology…………………………………………………………….30

2.8 An example of single-thread polling with stop………………………………………….34

2.9 An example of multi-thread polling……………………………………………………...35

2.10 A two-state DBA protocol for LR-PON………………………………………………....36

3.1 An example of LR-PON…………………………………………………………………39

3.2 An example of single-thread polling…………………………………………………….43

3.3 Achieving fairness in single-thread polling……………………………………………...44

3.4 Idea of multi-thread polling……………………………………………………………...46

3.5 Steps of multi-thread polling…………………………………………………………….48

3.6 Pseudo code for multi-thread polling scheduling………………………………………..54

3.7 Average packet delay with 20-km span………………………………………………….58

3.8 Average packet delay with 100-km span………………………………………………...59

3.9 Comparison of tuning multiple threads……………………………………………….…60

xi

3.10 Average packet delay at different initial cycle time………………………………….….60

3.11 Throughput vs. offered load………………………………………………………..........61

4.1 Dynamic wavelength allocation flowchart………………………………………………69

4.2 Three parts of user-SLA-aware bandwidth allocation…………………………………...72

4.3 “Legal” judgment of a new flow………………………………………………………...73

4.4 Admission control………………….……………………………………………………73

4.5 Illustration of up threshold………………………………….…………………………...76

4.6 Average packet delay for different wavelengths………………………………………...77

4.7 Wavelengths required for network upgrade……………………………………………...78

4.8 Instantaneous delay……………………………………………………….……………..78

4.9 Flow instantaneous delay of a user…………………………….……….………………..80

4.10 Flow instantaneous bandwidth of a user and background traffic…………………….….80

5.1 Operation of a unidirectional protected LR-PON using the “ring-and-spur” architecture.85

5.2 Operations of an access node (with 1+1 protection)…………………………………......87

5.3 OLT locates the failure……………………………………………………….……….....89

5.4 Operation of an access node (with 1:1 protection)…………………………………..…...90

5.5 Operation of a bidirectional protected LR-PON using the “ring-and-spur” architecture..93

5.6 Operation of an access node (bidirectional protection)………………………………......94

5.7 Normalized access node cost and normalized ratio of access node capacity vs. cost……97

5.8 Received signal power of each wavelength at OLT in 1 + 1 protection……………….....98

5.9 Received signal power of each wavelength at OLT in 1:1 protection………..……….....99

xii

List of Tables

2.1 Typical demonstrations of Long-Reach Broadband Access Networks………………….33

3.1 Long-Reach PON demonstrations……………………………………………………....47

4.1 Symbols for the scheduling algorithm…………………………………………………..68

5.1 Power loss of upstream and downstream wavelengths…………………………………96

1

Chapter 1

Introduction

1.1 Introduction to Optical Passive Network (PON)

The past decade has witnessed significant developments in the area of optical networking. Such

advanced technologies as wavelength-division multiplexing (WDM), optical amplification, optical

path routing (using wavelength switching, optical cross-connect), wavelength add-drop multiplexer

(OADM), and high-speed switching have found their way into the wide-area network (WAN),

resulting in a substantial increase of the telecommunications backbone capacity and greatly improved

reliability [1].

At the same time, enterprise networks almost universally converged on 100-Mbps or 1000-Mbps

Fast Ethernet architecture. Some mission-critical local area networks (LANs) even moved to 10-Gbps

rates, courtesy of a new 10-Gigabit Ethernet standard recently adopted by the Institute of Electrical

and Electronics Engineers (IEEE).

An increasing number of households have more than one computer. Home networks allow

multiple computers to share a single printer or a single Internet connection. Most often, a home

2

network is built using a low-cost switch or a hub that can interconnect 4 to 16 devices. Builders of a

new house now offer an option of wiring a new house with a category-t (CAT-5) cable. Older house

have an option of using existing phone wiring, in-house power lines, or an evermore popular wireless

network, based on IEEE 802.11 standard. Different flavors of this standard can provide up to 11 Mbps

bandwidth or up to 54 Mbps bandwidth, with distance being a tradeoff. Whether it is a wireless or

wire-line solution, home networks are essentially miniature LANs providing high-speed

interconnection for multiple devices.

These advances in the backbone, enterprise, and home networks coupled with the tremendous

growth of Internet traffic volume have accentuated the aggravating lag of access network capacity.

The “fist mile” still remains the bottleneck between high-capacity LANs and the backbone network.

1.1.1 Existing Broadband Access Solutions

1.1.1.1 Digital Subscriber Line

The widely deployed access solutions today are digital subscriber line (DSL) and cable modem (CM)

networks. Although they are improvements compared with 56 kbps dial-up lines, they are unable to

provide enough bandwidth for emerging services such as video-on-demand (VoD), interactive gaming,

or two-way video conferencing.

DSL uses the same twisted pair as telephone lines and requires a DSL modem at the customer

premises and digital subscriber in access multiplexer (DSLAM) in the central office (CO). The basic

premise of the DSL technology is to divide the spectrum of the line into several regions with the lower

4- kHz being used by plain old telephone service (POTS) equipment, while the higher frequencies are

allocated for higher-speed digital communications. DSL typically provide 1.5 Mbps of downstream

bandwidth and 128 kbps of upstream bandwidth. While economies of scale have successfully enabled

backbone networks to grow rapidly, the cost of access technologies remains prohibitively high for the

average household. The access network is the bottleneck for providing broadband services such as

video-on-demand, interactive games, and video conferencing to end users.

3

In addition, DSL has a limitation that any of its subscribers must be within 18,000 feet from the

Central Office (CO) because of signal distortions. Typically, DSL providers do not offer services to

customer more than 12,000 feet away. Therefore, only an estimated 60% of the residential subscriber

base in the US can use DSL even if they were willing to pay for it. Although variations of DSL such as

very-high-bit-rate DSL (VDSL), which can support up to 50 Mbps of downstream bandwidth, are

gradually emerging, these technologies have even more severe distance limitations. For example, the

maximum distance that VDSL can be supported over is limited to 1,500 feet. Some other variants of

DSL include G.SHDSL (offering 2.3 Mbps in both directions), ADSL2 (offering 12 Mbps downstream),

and ADSL2+ (offering 25 Mbps downstream).

1.1.1.2 Cable Television

Another alternative for broadband access is through Cable Television (CATV). CATV networks were

originally designed to deliver analog broadcast TV signals to subscriber TV sets. Following this

objective, the CATV networks adopted a tree topology and allocated most of its spectrum for

downstream analog channels. Typically, CATV is built as a hybrid fiber coax (HFC) network with fiber

running between a video head end or a hub to a curbside optical node, and the final drop to the

subscriber being coaxial cable. The coaxial part of the network users’ repeaters (amplifiers) and tap

couplers to split the signal among many subscribers.

Faced with the competition from telecom operators in providing Internet services, cable television

companies responded by integrating data services over their HFC cable networks. This integration

required replacing downstream-only amplifiers used for analog video with bidirectional amplifiers

enabling an upstream data path. Also a medium access protocol had to be deployed to avoid collision of

upstream data transmitted by multiple subscribers concurrently. The main limitation of CATV

architecture is that it is mainly built for delivering broadcast services, so they don't fit well for

distributing access bandwidth. Out of a total cable spectrum width of about 740 MHz, 400 MHz band is

allocated for downstream analog signals, and 300 MHz band is allocated for downstream signals.

4

Upstream communications are left with about a 40 MHz band or about 36 Mbps effective data

throughput per optical node. This very modest upstream capacity is typically shared among 500 to 2000

subscribers. At high load, the network's performance is usually low and cannot satisfy end users’

expectations.

1.1.1.3 Passive Optical Network

Emerging web applications require unprecedented bandwidth, exceeding the capacity of traditional

VDSL or CATV technologies. The explosive demand for bandwidth is leading to new access network

architectures which are bringing the high-capacity optical fiber closer to the residential homes and small

businesses [2]. The FTTx models -- Fiber to the Home (FTTH), Fiber to the Curb (FTTC), Fiber to the

Premises (FTTP), etc. -- offer the potential for unprecedented access bandwidth to end users (up to 100

Mbps per user). These technologies aim at providing fiber directly to the home, or very near the home,

from where technologies such as VDSL or wireless can take over. FTTx solutions are mainly based on

the Passive Optical Network (PON).

PONs use optical fiber to deliver bandwidth in the broadband access network. PONs may support

bandwidth-intensive, integrated, voice, data and video services in the broadband subscriber access

network which technologies such as DSL and CATV cannot support due to lack of bandwidth.

A logical way to deploy optical fiber in the local access network is to use a point-to-point topology,

with dedicated fiber runs from the telecom central office (CO) to each end-user subscriber. While this is

a simple architecture, in most cases it is cost prohibitive due to the fact that it requires significant outside

plant fiber deployment as well as connector termination space in the CO. Considering N subscribers at

an average distance L km from the CO, a point-to-point design requires 2N transceivers and N*L total

fiber length (assuming that a single fiber is used for bi-directional transmission).

5

Fig. 1.1.Passive Optical Network deployment scenarios (courtesy Dr. Glen Kramer) [2].

To reduce fiber deployment, it is possible to use a remote switch (concentrator) close to the

neighborhood. This will reduce the fiber consumption to only L km (assuming negligible distance

between the switch and customers), but will actually increase the number of transceivers to 2N+2, as

there is one more link added to the network. In addition, a curb-switched network architecture requires

electrical power as well as back-up power at the curb switch. Currently, one of the highest costs for

telecom operators is providing and maintaining electrical power in the local loop.

Therefore, it is logical to replace the hardened (environmentally protected) active curb-side switch

with an inexpensive optical splitter. The PON is a technology viewed by many as an attractive solution

to the first-mile problem [2, 3]. A PON minimizes the number of optical transceivers, central office

terminations, and fiber deployment. A PON is a point-to-multipoint (PtMP) optical network with no

6

active elements in the signal’s path from source to destination. The only interior elements used in a PON

are passive optical components, such as optical fiber, splices, and splitters. An access network based on

a single-fiber PON only requires N+1 transceivers and L km of fiber (see Fig. 1.1).

In a PON, transmissions are performed between an Optical Line Terminal (OLT) and Optical

Network Units (ONUs). The OLT resides in the CO and connects the optical access network to the

metropolitan area network (MAN) or wide area network (WAN), also known as backbone or long-haul

network. The ONU is located either at the end-user location (FTTH and FTTB), or at the curb, resulting

in a fiber-to-the-curb (FTTC) architecture.

Developments in PON in recent years include Ethernet PON (EPON), ATM-PON (APON, based

on ATM), Broadband-PON (BPON, based on APON, adding support for WDM and higher bandwidth),

Gigabit-PON (GPON, an evolution of BPON, supporting higher rates and multiple layer-2 protocols),

and wavelength-division-multiplexing PON (WDM-PON)

1.1.2 Long-Reach Passive Optical Network (LR-PON)

As broadband uptake increases globally, the services offered are becoming increasingly

bandwidth-intensive. A service that continues to increase bandwidth requirements is high-definition

television (HDTV), which requires in the region of 8 Mbps per channel. For telecommunication

networks to support a single channel of HDTV plus data and telephony services simultaneously, the

minimum bandwidth required is approximately 10 Mbps. Currently, telecom operators face a significant

problem in that the cost of the capital expenditure to deploy traditional networks to support next-

generation, bandwidth-intensive services is higher than the revenues that these services generate. A

passive optical network (PON) offers a different network architecture that enables the essentially

unlimited bandwidth of fiber to the home (FTTH) to be utilized. PONs use a point-to-multipoint

architecture to reduce cost by sharing a significant portion of the network among all customers rather

than each customer having a dedicated connection as in the current point-to-point architecture.

However, the cost reduction offered by a PON is not enough due to the expensive optical equipment that

7

must be installed at each customer premises. Studies carried out in the 1990s estimated that the cost of

installing fiber throughout the United Kingdom (UK) would be in the region of £15 billion [4]; a

massive investment, especially for a private company [5]. FTTH was deferred in favor of incremental

technologies based on DSL, which provided small increases in bandwidth without the requirement for

large capital investment. These technologies reuse the copper infrastructure and use modern encoding

and compression techniques, such as those developed by the Moving Pictures Experts Group (MPEG) to

provide the required performance and quality of service.

Long haul

-100s – 1000s km

-Mesh

Fig. 1.2. LR-PON simplifies the telecom network [1].

8

An alternative architecture was proposed as a more cost effective solution for an optical access

network; the long-reach broadband access using passive optical network technology, Long-Reach PON

(LR-PON). The strength of optical technology is its ability to displace electronics and simplify the

network by combining network tiers. The access and metro networks can be combined into one through

the use of an extended backhaul fiber, possibly 100 km in length to incorporate protection paths and

mechanisms, used with a PON. A general PON architecture consists of a shared fiber that originates

from a local exchange. At a point close to the customer premises, a passive optical splitter is used to

connect each customer to the main fiber. Current standardized PONs support a maximum distance of 20

km with 16, 32, or 64-way splits for Ethernet-PON (EPON), broadband-PON (BPON), or Gigabit-PON

(GPON), respectively. PON architectures with large split sizes are intended to maximize the number of

components shared between all customers. Even though optical amplifiers must be used to increase the

power budget, the distribution section closest to the customer remains passive. Cost savings are

achieved as the synchronous digital hierarchy (SDH) or synchronous optical network (SONET) rings

are replaced with a single backhaul fiber and consolidating the multiple OLTs in traditional PON

deployment at a single CO. As shown in Fig. 1.2, the result is a highly-simplified network. It is

envisioned that only 100 nodes would be required for the entire UK, as opposed to the current 400 nodes

[4].

A number of long-reach optical access network architectures have been developed. Initially, the

networks were based on a single channel, where a single wavelength is shared between all users, using

time-division multiplexing (TDM). These networks were followed by wavelength-division

multiplexing systems that shared a number of wavelengths between groups of users. More recently,

GPON extension systems have been developed that enable a number of existing GPONs to be grouped

and converted into long-reach systems with WDM backhaul systems.

1.1.3 Resource Allocation and Bandwidth Scheduling

In the downstream direction (OLT to ONUs) in a PON, packet frames transmitted by the OLT pass

through a 1:N splitter and reach each ONU. Typical values of N are between 8 and 64. PON is broadcast

9

in the downstream direction. Packets are broadcasted by the OLT and extracted by their destination

ONU based on a Logical Link Identifier (LIID), which the ONU is assigned when it registers with the

network.

In the upstream direction, data frames from any ONU will only reach the OLT and will not reach

any other ONU due to the directional properties of a passive optical combiner. Therefore, in the

upstream direction, the behavior of the PON is similar to that of a point-to-point architecture. However,

unlike in a true point-to-point network, in PON, data frames from different ONUs transmitted

simultaneously may collide. Thus, in the upstream direction, the ONUs needs to employ some

arbitration mechanism to avoid data collisions and fairly share the channel capacity. A contention-based

media-access mechanism (similar to Carrier-Sense Multiple Access / Collision Detection (CSMA/CD))

is difficult to implement because ONUs cannot detect a collision in the fiber from the combiner to the

OLT due to directional properties of the combiner. An OLT could detect a collision and inform ONUs

by sending a jam signal; however, propagation delays in PON (the typical distance from the OLT to the

ONUs is 20 km) greatly reduces the efficiency of such a scheme. To introduce determinism in the frame

delivery in upstream direction, different non-contention schemes have been proposed.

All ONUs are synchronized to a common time reference and each ONU is allocated a time-slot in

which to transmit. Each time-slot is capable of carrying several frames, (e.g., in EPON, several Ethernet

frames are carried). An ONU should buffer frames received from a subscriber until its time-slot arrives.

When its time-slot arrives, the ONU would burst all stored frames at full channel speed. If there are no

frames in the buffer to fill the entire time slot, an idle pattern is transmitted.

Thus, time-slot assignment is a very crucial step. The possible time-slot allocation schemes could

range from a static allocation (fixed time-division multiple access (TDMA)) to a dynamically-adapting

schemes based on instantaneous queue size in every ONU (statistical multiplexing schemes). In the

dynamically-adapting scheme, the OLT can play the role of collecting the queue sizes from the ONUs

and then issuing time slots. Although this approach leads to significant signaling overhead between the

OLT and the ONUs, the centralized intelligence may lead to more efficient use of bandwidth. More

10

advanced bandwidth-allocation schemes are also possible, including schemes utilizing notions of traffic

priority, Quality of Service (QoS), Service-Level-Agreements (SLAs), over-subscription ratios, etc. [6].

1.1.4 Protection in LR-PON

As LR-PON exploits the huge transmission capacity of optical communications technology, and is

oriented for long-range coverage to serve a large number of end users, any network failure may cause a

significant loss of data (and revenue) for the customers and the network operator. So, LR-PON

protection becomes necessary and important.

A protection scheme in an access network must be able to protect against OLT failure and failures

due to fiber cut(s) between OLT and the splitter. Its design should be cost effective at the same time. So

far, four protection schemes have been recommended by International Telecommunication Union (ITU)

for APON interface. Although these schemes are recommended for APON interface, they are applicable

to all FTTH-PON interfaces. Three of the schemes are able to provide protection for OLT failure. Yet,

these schemes are often found not suitable [7] due to their high redundancy causing high cost.

Not much research attention has been paid on the protection of LR-PON. Some approaches include

the “dual homing” proposal [8] from British Telecom (BT) and “scalable extended-reach PON” [9]. The

“dual homing” proposal allows a customer to access the network through another dual router if the

current one fails, thus providing enhanced fault tolerance and superior service to the customer; the

“scalable extended-reach PON” provides protection by rerouting the traffic onto backup fiber in case of

failure.

1.2 Research Contributions

This research makes four important contributions to the architecture, management, and reliability in

LR-PON. We briefly elaborate on these contributions in the following sections.

11

1.2.1 A Survey of Research Challenges, Demonstrations, and

Bandwidth Assignment Mechanisms in LR-PON

First, we review the evolutionary path of access networks and show the drivers from technology and

business perspectives for high bandwidth and low cost. A variety of research challenges in this field is

reviewed, from optical components in the physical layer to the control and management issues in the

upper layers. We discuss the requisites for optical sources, optical amplifiers, and optical receivers in

networks with high transmission rate (10 Gbps) and large power attenuation (due to large split,

transmission over 100 km and beyond, and propagation). We analyze the key topological structures to

guarantee physical protection (e.g., tree-and-branch, ring-and-spur). Then, some relevant

demonstrations of Long-Reach optical access networks developed worldwide by different research

institutes are presented. Finally, Dynamic Bandwidth Allocation (DBA) algorithms that allow to

mitigate the effect of the increased control-plane delay in an extended-reach network are reviewed.

1.2.2 Multi-Thread Polling: A Dynamic Bandwidth Distribution

Scheme in LR-PON

A major challenge in LR-PON is that the propagation delay (for data as well as control signals) between

the telecom central office (CO) and the end user is increased by a very significant amount. Now,

traditional PON algorithms for scheduling the upstream transmission, such as dynamic bandwidth

allocation (DBA) algorithms, may not be sufficient; actually, they may lead to degraded performance

because of the long delay of the CO-to-Users “control loop”. This challenge motivates us to propose

and study a multi-thread polling algorithm to effectively and fairly distribute the upstream bandwidth

dynamically. This algorithm exploits the benefits of having multiple polling processes running

simultaneously and enabling users to send bandwidth requests before receiving acknowledgement from

the CO. We compare the proposed algorithm with traditional DBA, and show its advantage on average

packet delay. We then analyze and optimize key parameters of the algorithm, such as initiating and

tuning multiple threads, inter-thread scheduling, and fairness among users. Numerical results

12

demonstrate the algorithm’s advantage to decrease the average packet delay and improve network

throughput under varying offered loads.

1.2.3 A SLA-Aware Protocol for Efficient Tunable Laser

Utilization to Support Incremental Upgrade in LR-PON

With the increased bandwidth requests from the expanding base of users, LR-PON should utilize the

network resource (e.g., wavelengths, lasers, etc.) more effectively. We propose a new and efficient

protocol to achieve better utilization of tunable lasers, as well as wavelength resources across different

user groups in a LR-PON. In order to accommodate downstream bursty traffic and provide Quality of

Service (QoS) in the user-specified Service-Level Agreement (SLA), the protocol integrates our

proposed SLA-aware bandwidth allocation scheme based on flow scheduling. We show the protocol’s

advantage to support incremental upgrade of bandwidth with increasing user bandwidth requests, and to

provide a user with a SLA which guarantees a number of streaming flows with average bandwidth and

maximum delay guarantee (e.g., 5 ms), as well as data flows with average bandwidth specifications.

1.2.4 Protection in LR-PON

Since LR-PON serves a lot more users, a network failure may lead to a huge amount of data loss and

negative user experience. Thus, to understand the importance of LR-PON survivability, we propose

hardware-accelerated protection schemes for the LR-PON, incorporating the “ring-and-spur” structure

to achieve fast protection-switching time, and automatic failure location at the Optical Line Terminal

(OLT). We design the protection schemes for multiple network environments, e.g., unidirectional

transmission vs. bidirectional transmission, and 1 + 1 protection vs. 1:1 protection. Numerical examples

demonstrate that protection paths can be established within a few tens of ms after a failure occurs.

1.3 Organization

13

Chapter 2 reviews research challenges, related demonstrations, and bandwidth allocation schemes of

LR-PON. This work has been accepted in the IEEE Communications Surveys and Tutorials [10].

Chapter 3 proposes an effective bandwidth allocation scheme to mitigate the upstream packet delay

due to increase control loop in LR-PON. This work has been published in IEEE Journal on Selected

Areas in Communications (JASC) [11] after presentation at the Globecom conference 2007 [12].

Chapter 4 proposes a protocol to effectively utilize the tunable transmitter at the Central Office,

accommodates downstream bursty traffic, and provides Quality of Service (QoS) according to the

user-specified Service-Level Agreements (SLA). This work has been submitted to OSA Journal of

Optical Networking after presentation of various parts at the Optical Fiber Communication Conference

2006 (OFC’06) [13], OFC’07 [14], and the Conference on the Optical Internet 2007 (COIN’07) [15].

Chapter 5 proposes hardware-accelerated protection schemes for the LR-PON to achieve fast

protection-switching time, and automatic failure location at the OLT. This work was presented at

OFC’09 [16] and has been submitted to IEEE/OSA Journal of Lightwave Technology (JLT).

Chapter 6 concludes this dissertation.

14

Chapter 2

Long-Reach Optical Access Networks:

A Survey of Research Challenges,

Demonstrations, and Bandwidth

Assignment Mechanisms

2.1 Introduction

Much of the R&D emphasis in recent years has been on developing high-capacity backbone networks.

Backbone network operators currently provide high-capacity OC-192 (10 Gbps) links, with 40 Gbps

transmission also quite mature now [1]. However, note that the telecom network hierarchy consists of

the backbone network as well as the metro and access networks [1, 17]. Today's access network

technologies such as Digital Subscriber Line (DSL) typically provide 1.5 Mbps of downstream

bandwidth and 128 kbps of upstream bandwidth. While economies of scale have successfully enabled

backbone networks to grow rapidly, the cost of access technologies remains prohibitively high for the

15

average household [18]. The access network is the bottleneck for providing broadband services such as

video-on-demand, interactive games, and video conferencing to end users.

In addition, DSL has a limitation that any of its subscribers must be within 18,000 feet from the

Central Office (CO) because of signal distortions. Typically, DSL providers do not offer services to

customer more than 12,000 feet away. Therefore, only an estimated 60% of the residential subscriber

base in the US can use DSL even if they were willing to pay for it. Although variations of DSL such as

very-high-bit-rate DSL (VDSL), which can support up to 50 Mbps of downstream bandwidth, are

gradually emerging, these technologies have even more severe distance limitations. For example, the

maximum distance that VDSL can be supported over is limited to 1,500 feet. Some other variants of

DSL include G.SHDSL (offering 2.3 Mbps in both directions), ADSL2 (offering 12 Mbps downstream),

and ADSL2+ (offering 25 Mbps downstream).

Another alternative for broadband access is through Cable Television (CATV). CATV networks

provide Internet services by dedicating some Radio Frequency (RF) channels in co-axial cable for data.

However, CATV networks are mainly built for delivering broadcast services, so they don't fit well for

distributing access bandwidth. At high load, the network's performance is usually low and cannot satisfy

end users’ expectations.

Moreover, it is expected that emerging applications such as Internet Protocol TV (IPTV),

video-on-demand (VoD), video file swapping, peer-to-peer applications, real-time network games, etc.

will demand much more bandwidth, and some of these applications may require symmetric bandwidth

as well (equal and high bandwidth in both downstream and upstream directions) [19, 20]. Both DSL and

CATV provide limited and asymmetric bandwidth access (lesser bandwidth in the upstream direction)

with the implicit assumption that present traffic is more downstream-oriented.

Emerging web applications require unprecedented bandwidth, exceeding the capacity of traditional

VDSL or CATV technologies. The explosive demand for bandwidth is leading to new access network

architectures which are bringing the high-capacity optical fiber closer to the residential homes and small

businesses [18]. The FTTx models -- Fiber to the Home (FTTH), Fiber to the Curb (FTTC), Fiber to the

16

Premises (FTTP), etc. -- offer the potential for unprecedented access bandwidth to end users (up to 100

Mbps per user). These technologies aim at providing fiber directly to the home, or very near the home,

from where technologies such as VDSL or wireless can take over. FTTx solutions are mainly based on

the Passive Optical Network (PON). Developments in PON in recent years include Ethernet PON

(EPON), ATM-PON (APON, based on ATM), Broadband-PON (BPON, based on APON, adding

support for WDM and higher bandwidth), Gigabit-PON (GPON, an evolution of BPON, supporting

higher rates and multiple layer-2 protocols), and wavelength-division-multiplexing PON (WDM-PON)

[21, 22, 23, 24]. Recently investigation shows broadband access is penetrating quickly into residential

and business users, e.g., fiber connections account for 36% of all Japanese broadband subscriptions and

31% in Korea [25], and the Australian Government's “BroadbandNow'' project promises to deploy in

five years broadband access nationwide.

PON reduces the network cost by eliminating the power supply (operational cost) along the fiber

path from Central Office (CO) to end users, and by sharing the significant portion of the network cost

among multiple users. However, the cost reduction offered by a PON might not be enough for the future

telecom network, e.g., research shows the realization of fiber access throughout United Kingdom would

cost around ₤15 billion. The economic driver behind technology calls for a new thinking on the

broadband optical access.

An alternative technology, called Long-Reach Passive Optical Network (LR-PON), was proposed

as a more cost-effective solution for the broadband optical access network. LR-PON extends the

coverage span of PONs mentioned above from the traditional 20 km range to 100 km and beyond by

exploiting Optical Amplifier and WDM technologies. A general LR-PON architecture is composed by

an extended shared fiber connecting the CO and the local user exchange, and optical splitter connecting

users to the shared fiber. Compared with traditional PON, LR-PON consolidates the multiple Optical

Line Terminals (OLTs) and the Central Offices (COs) where they are located, thus significantly

reducing the corresponding Operational Expenditure (OpEx) of the network. By providing extended

geographic coverage, LR-PON combines optical access and metro into an integrated system. Thus, cost

17

savings are also achieved by replacing the Synchronous Digital Hierarchy (SDH) or Synchronous

Optical Network (SONET) with a shared optical fiber. In general, the LR-PON can simplify the

network, reducing the number of equipment interfaces, network elements, and even nodes.

Although the idea of extending the reach of a PON has been around for quite a while, it is

emphasized recently because optical access is penetrating quickly into residential and small- business

markets and simplification of the telecom network requires an architecture to combine the metro and

access networks. Figure 1.2 shows how LR-PON simplifies the telecom network. The traditional

telecom network consists of the access network, the metropolitan-area network, and the backbone

network (also called long-haul or core network). However, with the maturing of technologies for

long-reach broadband access, the traditional metro network is getting absorbed in access. As a result, the

telecom network hierarchy can be simplified with the access headend close to the backbone network.

Thus, the network's Capital Expenditure (CapEx) and Operational Expenditure (OpEx) can be

significantly reduced, due to the need for managing fewer control units. But this architecture also brings

with it its own new research challenges, which will be outlined in this chapter.

The rest of this chapter is organized as follows. Section 2.2 discusses the research challenges in the

Long-Reach PON design. Section 2.3 summarizes some relevant Long-Reach PON demonstrations

worldwide by different research institutes. Section 2.4 discusses the Dynamic Bandwidth Allocation

algorithms in Long-Reach PON. Section 2.5 concludes the chapter.

2.2 Research Challenges

18

Fig. 2.1. Long-Reach PON (LR-PON) architecture.

Figure 2.1 shows the general architecture of a LR-PON. The central office (CO) connects the core

network and the access network, and implements layer 2 and layer 3 functions, e.g., resource allocation,

service aggregation, management, and control. The local exchange resides in the local users’ area,

which is close to the end customer equipment: ONU (within 10 km of drop section). The optical signal

propagates across the fiber forming the feeder section (100 km and beyond) with the CO and the local

exchange at its two ends; then the fiber is split and connected to a large number of ONUs. In order to

compensate for the power loss due to long transmission distance and high split size, optical amplifiers

are used at the OLT and the local exchange.

2.2.1 Signal Power Compensation

Optical amplification is indispensable in a LR-PON. Besides amplifying the signal, the amplifiers

introduce two challenges, as indicated below.

1) Optical amplifiers introduce amplified spontaneous emission (ASE) [26]. It is a side effect of the

amplification mechanism, produced by spontaneous emission that has been optically amplified by the

process of stimulated emission in a gain medium. The ASE may have a detrimental effect on system

performance. As the high split of LR-PON would attenuate the signal significantly, the optical signal

power could be sufficiently low at the input of the amplifier. Meanwhile, the ASE noise is

device-dependent, e.g., ASE noise usually accumulates with the length of Erbium-doped fiber (EDF),

Central

Office

(CO)

Local

Exchange

ONU

ONU

ONU

Core

Network

Feeder (up to 100 km and beyond)

Drop (10 km)

Large number of users

Single channel or WDM

19

and noise figure of an ideal Erbium-doped-fiber amplifier (EDFA) is 3 dB [26]. As a result, the

signal-to-noise ratio (SNR) could be reduced significantly. In order to amplify the optical signal while

suppressing the noise, a possible scheme, called dual-stage intermediate amplification, was introduced

[33]. In this scheme, the first stage is composed of a low-noise pre-amplifier, which produces a high

SNR by maintaining its ASE at a low level; and the second stage consists of amplifiers to amplify the

optical signal with enough power, in order to counter the large attenuation in the feeder section (100 km

and beyond).

2) The EDFA features a low noise figure, a high power gain, and a wide working bandwidth, which

enable it to be advantageous in a LR-PON employing WDM. But the relatively slow speed in adjusting

its gain makes it disadvantageous due to the bursty nature of upstream time-division-multiple-access

(TDMA) traffic in a LR-PON, where the optical amplifier needs to adjust its gain fast when packets with

different DC levels pass through it, in order to output packets with uniform signal amplitude. A possible

technology is called gain control by using optical gain clamping or pump power variation. An example

could be an auxiliary wavelength that senses the payload wavelength and is adjusted relative to the

transmitted upstream packet so that the total input power at the EDFA remains constant. Hence, the gain

of the EDFA remains constant for the burst duration. Researchers have also investigated the

semiconductor optical amplifier (SOA) as the amplifier [26, 27]. An SOA can be adjusted faster and

offers the potential for monolithic or hybrid array integration with the other optical components which

makes it more cost competitive. The single-channel SOA amplifier is suitable for the “pay as you grow”

business model [32].

2.2.2 Optical Source

In order to lower the CapEx and OpEx, a standard PON may choose lower-cost uncooled

transmitters in the ONU, because a major investment for an optical access network is the cost associated

with installation of an optical transmitter and receiver in the ONU at the customer premises [18].

However, the uncooled transmitter is temperature dependent which in turn transmits a wavelength with

a possible drift of 20 nm [28]. As no component in a standard PON is wavelength critical, the

20

performance may be unaffected. But in a LR-PON which exploits WDM to satisfy the huge amount of

traffic, the wavelength drift becomes crucial, especially for certain components such as optical filters.

To counter the wavelength drift, more expensive cooled transmitters are considered to ensure a stable

wavelength. A possible technology is called reflective ONU (R-ONU) [38], which generates the

upstream signal from the optical carrier feeding from outside (could be the downstream optical carrier or

a shared optical source at the local exchange), using a reflective SOA (RSOA) modulator.

The transmitters in a traditional PON are usually designed for a transmission range which is less

than 20 km. The challenge arises when applying them in a LR-PON where the signal needs to cover a

range of 100 km and beyond.

2.2.3 Burst-Mode Receiver

The different ONU-OLT distances mean different propagation attenuations for signals from ONUs

to the OLT, which in turn result in varied DC levels of bursty packets from the ONUs at the OLT. A

burst-mode receiver is designed for this situation. It includes the automatic gain control (AGC) for

adjusting its 0-1 threshold at the beginning of each received burst, and the clock and data recovery

(CDR) circuit for acquiring phase and frequency locking on an incoming signal. The design of a

burst-mode receiver at 1 Gbps has been addressed in the context of IEEE 802.3ah EPON networks, and

the one at 10 Gbps or 10 G / 1 G hybrid is currently discussed in IEEE 802.3av 10 Gbps EPON. An

LR-PON scales up in speed (10 Gbps and beyond) and number of customers supported (up to 512 users

could share the same channel), challenges might occur in the following aspects. As the optical amplifier

increases the difference of the DC level of upstream signals from different ONUs, the burst-mode

receiver is required to support a wider dynamic range; as the signal power may be attenuated

significantly due to a large split ratio and a long-distance transmission, the burst-mode receiver is

required to have a high sensitivity; and with the increased sharing of the same channel, a more strict

timing control of the guard time between successive ONU transmission slots is required, in order to

achieve a higher bandwidth efficiency. Efforts have been made by researchers, such as a new 10 Gbps

21

burst-mode receiver [40] that uses a multi-stage feed-forward architecture to reduce DC offsets, and a

high sensitivity avalanche photodiode (APD) burst-mode receiver [29] for 10 Gbps TDM-PON systems.

2.2.4 Upstream Resource Allocation

In LR-PON, the end users and the Central Office (CO) (through which users are connected to the rest of

the Internet) are separated by a significant distance, typically 100 km and beyond. Hence, control-plane

delays (ONUs send transmission request to CO, and transmit upstream data upon receiving

acknowledgement from CO) are significant. The delay budget in an access network is approx. 1-2

milliseconds for real-time applications, so various known scheduling algorithms for packet-based

networks are difficult to apply directly. Therefore, efficient remote-scheduling algorithms (e.g., for

dynamic bandwidth allocation) need to be developed which overcome the large CO-user distance,

which support different classes of service, and which are scalable in terms of the number of users

supported as well. We will discuss some research endeavors in Section 2.4.

2.2.5 Topology and Protection

Several candidate network topologies have been proposed for LR-PON. The branch-and-tree topology

has a feeder section of a strand of fibers of 90 km (tree) and is split to multiple users (branches) at the

local exchange; while the ring-and-spur topology has the feeder section composed by a fiber ring and

Optical Add-Drop Multiplexers (OADMs) [1] on the ring, and up and downstream optical signals are

added and dropped through OADMs and split to end users [41]. As LR-PON exploits the huge

transmission capacity of optical technology, and is oriented for long-range coverage to serve a large

number of end users, any network failure may cause a significant loss for customers and the network

operator. So LR-PON protection becomes necessary and important. Various protection schemes for

PONs have been proposed by ITU-T (e.g., G. 983 and G. 984). But protection schemes on emerging

topologies, such as ring-and-spur topology need further investigation. Section 2.3.4 will introduce a

protection scheme for the ring-and-spur topology, and Chapter 5 will propose a novel protection

architecture for LR-PON.

22

2.3 Demonstrations of LR-PON

2.3.1 PLANET SuperPON

The ACTS-PLANET (Advanced Communication Technologies and Services – Photonic Local

AccessNETwork) [30, 31, 32, 33, 34, 35] is an EU-funded project. This project investigated possible

upgrades of a G.983-like APON system in the aspect of network coverage, splitting factor, number of

supported ONUs, and transmission rates. Note that the term SuperPON is not fully accurate since not all

of its components between the OLT and ONUs are passive; but this term has been used quite widely in

the literature to show the SuperPON’s derivation from the traditional PON system with the use of

limited active elements.

Fig. 2.2. ACTS-PLANET architecture and transport system [30 - 35].

The basic architecture is depicted in Fig. 2.2, which was installed in the first quarter of 2000. The

implemented system supports a total of 2048 ONUs and achieves a span of 100 km. The 100-km fiber

span consists of a maximum feeder length of 90 km and a drop section of 10 km. The large splitting

factor is achieved through 2-stage splitting at the local exchange and drop section.

The increased transmission range from 20 km to 100 km and split size from 32 to 2048 increases

the signal attenuation. To compensate for the attenuation, optical amplifiers are located at the feeder

23

section (feeder repeater) and at the intersection between the feeder and the drop sections (amplified

splitter). In the downstream direction, EDFAs are chosen as the amplifier because of their high power

gain, wide working bandwidth, and the continuous wave mode of downstream signal. The cost of

amplifiers is shared among the end users due to the downstream broadcast property, which makes it less

cost-sensitive.

In the upstream direction, due to the possibly unequal distances between the OLT and the ONUs,

optical signal attenuation may not be the same for each ONU. As a result, the power level received at the

OLT may be different (burst mode) for each ONU. PLANET chooses SOAs to amplify the burst-mode

upstream signals. The SOAs are not only placed at the output of the splitter in the local exchange, but

also included in parallel between split stages, in order to reduce the split loss before the signal is

attenuated by a split. Otherwise, if SOAs are only placed after the splitter, the signal-to-noise ratio

(SNR) cannot satisfy the system requirements (SNR > 18.6 dB was required to achieve the PLANET

target performance bit-error rate (BER) of 10–9

, using ON-OFF keying and including a 3-dB optical

power margin [31]). But a side effect, named noise funneling, is introduced due to the placement of

SOAs, because the ASE contribution from the SOAs in parallel is combined and make the effect of ASE

more severe. In order to remedy this effect, the Operation, Administration, and Maintenance ONU

(OAM-ONU) is introduced. It receives and interprets all downstream control information. In this way, it

can calculate information on which ONUs are granted access to the upstream path at a certain instant in

time. Using this information, the protocol switches on an SOA when it is required to amplify the

corresponding upstream signal, with a switching transient lower than 25 ns.

The transport system is based on Asynchronous Transfer Mode (ATM), which offers 2.5 Gbps and

311 Mbps for downstream and upstream transmissions, respectively. A Time-Division Multiple Access

(TDMA) protocol is used to allocate the upstream bandwidth among multiple users. Besides

synchronizing the ONUs’ transmission, the protocol also enables OAM-ONU (shown in Fig. 2.2) to

synchronize the SOAs to set the correct gain to compensate for different power losses of upstream

signals from different ONUs.

24

Other possible architectures for PLANET are discussed in [33], which use different types of

amplifiers. Detailed calculation on the effect of accumulated ASE noise of the amplifiers is included in

[35].

2.3.2 Demonstrations from British Telecom (BT)

Fig. 2.3. Long-reach optical access network by British Telecom [36].

British Telecom has demonstrated its Long-Reach PON, which is characterized by a 1024-way split,

100-km reach, and 10-Gbps transmission rate for upstream and downstream directions [36], as shown in

Fig. 2.3. Compared with the SuperPON discussed earlier, the split size has been halved. This brings a

major benefit by saving the amount of optical amplifiers needed, e.g., 6 optical amplifiers are enough for

up and down streams as opposed to 39 required by SuperPON. The 1024-way split is made up of a

cascade of two N:16 and one N:4 splitters in the drop section.

The system includes a 90-km feeder section between the OLT and the local exchange, and a 10-km

drop section between the local exchange and end users. In order to boost the signal attenuated by large

splitting and long-distance transmission, optical amplifiers are used at the OLT and the local exchange.

In order to lower the CapEx and OpEx of LR-PON, the ONU for the large number of end users has

to be simple and cost efficient. As a result, the transmitter output power in an ONU is constrained

25

because it is a key price factor for the optical transmitter. Besides the constrained signal power, the drop

section has a large power loss of 40.3 dB before the signal is amplified. Hence, a very low signal power

will arrive at the local exchange. To boost the signal to enough power, a two-stage amplifier as

discussed in Section 2.2.1 is implemented, which contains (1) a low-noise pre-amplifier which produces

a high SNR by maintaining its ASE at a low level, and (2) second-stage amplifiers to amplify the optical

signal with enough power.

Other technologies used include forward error correction (FEC) and optical filter. FEC is a coding

technique by which transmission errors can be detected and corrected by encoding the data and

including a number of parity bits. FEC can alleviate the system design requirement by allowing a

relatively higher pre-FEC BER, e.g., a BER of 10-10

required by a LR-PON only needs a pre-FEC BER

of 2.9 x 10-4

. Optical filter can also promote the SNR of the received signal by reducing the ASE noise in

the signal passing through.

Fig. 2.4. Experimental configuration of GPON extended to 135 km via WDM [37].

The above demonstration [36] concerned mainly the physical layer to extend the access network,

e.g., power budget. However, a later demonstration [37] (Fig. 2.4) showed how to integrate the

26

higher-layer issues, e.g., control protocol into LR-PON design. This demonstration extended gigabit

PON (GPON) from 20 km to 135 km, and integrated WDM to increase the system capacity.

The experimental configuration in Fig. 2.4 shows an extended GPON with 40-channel WDM

system. The OLT transmits downstream at a wavelength in the 1490-nm region, according to GPON

standard. The data rate for each downstream channel is 2.488 Gbps. This downstream wavelength will

be converted to a WDM-compatible wavelength of 1552.924 nm by a transponder right after the OLT.

The conversion is necessary because the long-distance transmission exploits a WDM system of 125-km

standard G.652 fiber with an EDFA attached to each end. When arriving at the drop section, the

wavelengths are converted back to 1490-nm range through a bespoke transponder for compliance with

the GPON standard. Then, each wavelength is split to 64 ONUs within the 10-km drop section.

The experiment demonstrates GPON ONUs with different upstream wavelength ranges in the

upstream direction: ~1310 nm or ~1550 nm. These different wavelengths have the same transmission

rate of 1.244 Gbps, and will be converted at the bespoke transponder to a WDM-compatible wavelength

of 1559.412 nm. At the OLT side, an optical filter is placed before the burst-mode receiver to improve

the SNR. Due to the incorporation of the optical filter, the temperature control of the system must be

precise in order to prevent wavelength drift.

The demonstration achieved a BER better than 10-10

in both directions. It also demonstrated that,

by using optical-electrical-optical (OEO) conversion at the transponder and optical amplification, the

GPON can be extended to beyond 100-km range.

27

2.3.3 Demonstrations from University College Cork, Ireland

Fig. 2.5. Hybrid DWDM-TDM LR-PON architecture [38, 39].

The demonstration of a hybrid WDM-TDM LR-PON is reported in [38, 39] by the Photonic System

Group of University College Cork, Ireland. This work supports multiple wavelengths, and each

wavelength pair (up and down stream) can support a PON segment with a long distance (100 km) and a

large split ratio (256 users).

The layout in Fig. 2.5 is divided into four notional locations: 1) customer ONU, 2) street cabinet, 3)

local exchange, and 4) core exchange. The core and local exchange are powered to support signal

amplification on both ends of the long-distance fiber transmission (88 km). The street cabinet contains

cascaded optical splitters to achieve a large split size of 256 in each TDM-PON segment. As WDM

requires precise wavelength control (50 GHz or 100 GHz channel spacing), a WDM centralized source

is placed at the local exchange, which is composed of an array of distributed feedback (DFB) laser

diodes to generate upstream carrier wavelengths. These upstream carrier wavelengths are split and fed

to each ONU through the street cabinet. The customer ONU is colorless which uses a semiconductor

electro-absorption modulator (EAM) to modulate the upstream carrier wavelength generated in the local

exchange, and two SOAs are connected to the input and output of EAM to amplify the upstream

wavelength before and after the modulation.

28

Each up and down stream wavelength has a transmission rate of 10 Gbps. These wavelengths are

achieved by WDM channel allocation, where the C-band is split into two segments: the blue half (1529

– 1541.6 nm) carrying downstream channels and the red half (1547.2 – 1560.1 nm) carrying upstream

channels. The red and blue bands are separated by a guard band of approximately 5 nm. With 100-GHz

channel spacing, 17 pairs of up and down channels can be achieved to support 17 PON segments in the

demonstration.

The demonstration offers a LR-PON with 17 PON segments, each of which supports symmetric

10-Gbps upstream and downstream channels over a 100-km transmission. The system can serve a large

number of users (17 x 256 = 4352 users). Experiments show a low BER of 10-9

in both directions.

Fig. 2.6. PIEMAN hybrid WDM/TDMA architecture [40].

In order to scale the long-reach access network, the authors in [40] cooperated with British

Telecom and other telecom companies, e.g., Alcatel and Siemens, to demonstrate their second-stage

prototype: Photonic Integrated Extended Metro and Access Network (PIEMAN) sponsored by the

Information Society Technologies (IST) Sixth Framework Project. As shown in Fig. 2.6, PIEMAN

consists of a 100-km transmission range with 32 DWDM channels supported, each of which operates at

29

a symmetric 10 Gbps and serves a PON segment. The split ratio for each PON segment is 512, which

means that the maximum number of users supported is 32 x 512 = 16,384.

At the local exchange, the downstream wavelengths are pre-amplified by the EDFA, and then

demultiplexed into the single wavelengths through Arrayed Waveguide Grating (AWG). Then, each

single wavelength is amplified by a single-wavelength EDFA. This configuration shows better

performance and cost efficiency compared with the one which only has a powerful EDFA before the

AWG. For the upstream direction, each upstream wavelength before entering the AWG is amplified by

a single-wavelength EDFA which is stabilized against the transients introduced by bursty traffic. A

common upstream WDM EDFA at the local exchange is not used in order to avoid the transients by

neighboring channels.

Other key issues in PIEMAN include the 10-Gbps burst-mode receiver and colorless ONU. The

design of the 10-Gbps burst-mode receiver is composed of a burst-mode trans-impedance amplifier with

PIN photodiode and a burst-mode post-amplifier to accommodate the high bit rate and the high dynamic

range imposed by the high split and the ASE introduced by the amplifiers. The colorless ONU attempts

to reduce the cost of the ONU by removing the need for an internal wavelength referring and control

function. Instead, the ONU remodulates the downstream optical carrier distributed from the central

office, using the “colorless” reflective modulators that operate across the full upstream bandwidth range

to generate the upstream signal.

30

2.3.4 Demonstrations of “Ring-and-Spur” LR-PON

Fig. 2.7. WE-PON in FTTH topology [41].

Instead of the “tree-and-branch” topology considered in the prior approaches, researchers are also

investigating a “ring-and-spur” topology for LR-PON, as shown in Fig. 2.7, where each PON segment

and OLT are connected through a fiber ring, and each PON segment can exploit traditional FTTX

technology with a topology consisting of several “spurs” served from the “ring”. The ring can cover a

metro area up to 100 km and traditional users in an access area can be served by a spur segment. The

natural advantage of the ring topology is that it provides two-dimensional coverage, as well as failure

protection, e.g., traffic can choose its transmission in an alternate direction to avoid a fiber cut.

An example demonstration is by ETRI, a Korean government-funded research institute, which has

developed a hybrid LR-PON solution, called WE-PON (Wdm-E-PON) [41]. In WE-PON, 16

wavelengths are transmitted on the ring, and they can be added and dropped to local PON segments

31

through the remote node (RN) on the ring. A possible design of the RN includes OADM and optical

amplifiers. As the split ratio of the splitter is 1:32, the system can accommodate 512 users.

Another demonstration [9 42 ], named Scalable Advanced Ring Dense Access Network

Architecture (SARDANA), also implements the “ring-and-spur” topology. Thirty-two wavelengths are

transmitted on the ring with a splitting factor of 1:32 for each wavelength. Over 1000 users are

supported. ONUs are based on 1.25/2.5/5 Gbps-capable RSOA and downstream at 10 Gbps. It also

proposes a protection design in the RN, which enables the RN to receive signals from east or west fiber

connection. So, in case of failure, RN can choose to receive from the direction in which it still connects

with the CO.

2.3.5 Other Demonstrations

Recently, more demonstrations of LR-PON have appeared in the literature [43, 44, 45, 46, 47]. In [43],

the authors proposed a ring-and-spur WDM-TDM access network architecture, named Stanford

University Access (SUCCESS). SUCCESS employs a semipassive configuration of remotes nodes to

enable protection and restoration, and fast tunable lasers at the OLT provide down and up channels

which can be shared by all ONUs to reduce the transceiver count.

Another demonstrator [44], named STARGATE, examines an optically-integrated access-metro

network. STARGATE employs a hybrid structure of a ring-and-spur network and a long-reach star

subnet connecting each remote node. This structure allows ONUs of different PONs to communicate

directly without OEO conversion at any intermediate CO, and it can easily allocate a new channel to an

upgraded user in a pay-as-you-grow manner. The well-designed remote node supports space dimension

and optical bypass, which allow dynamic setup of fine-granularity transparent connections between

ONUs in different WDM PONs in support of emerging applications (e.g., P2P file sharing).

Some research demonstrates possible network configurations in terms of span and capacity, e.g.,

512-ONUs-100 km, 1024-ONUs-100 km [45], and a 85-km LR-PON using a reflective SOA-EA

32

modulator at 7.5 Gbps [46], then a 120-km LR-PON with symmetric up and down data rate of 10 Gbps

[47].

Other endeavors incorporate novel devices or technologies. An SOA-Raman Hybrid Amplifier [48,

49] (SRHA) with an ultra broadband gain (up to 80-nm bandwidth) is introduced, in which the SOA

provides the bulk of the gain while the Raman stage compensates for the SOA's gain tilt, resulting in a

broad relatively flat gain. The test bed incorporating SRHA provides symmetric baseband services at

2.488 Gbps over 60 km (1490 nm down and 1310 nm up) with a 1:64 split and has 3 downstream video

enhancement wavelengths at 1510 nm, 1530 nm, and 1550 nm.

An innovative bi-directional amplification is proposed to offer an amplification scheme of

LR-PON in a cost-effective manner [50]. The bi-directional amplification is realized by using a novel

four-port interleaver to direct the up and downstream wavelengths through the same optical amplifier.

The test shows a power penalty of less than 0.8 dB after 100-km standard single-mode fiber (SSMF)

transmission for all channels at 10 Gbps.

A hybrid WDM-CDM (Code Division Multiplexing)-PON scheme is proposed in [51], trying to

eliminate the optical amplifier through the coding gain of CDM. The test shows a 42-dB loss budget for

transmitting 16 symmetric 2-Gbps channels for 100 km distance, with a 1x32 split at the user side. The

authors in [52, 53] also demonstrate a 35-channel LR-PON over 70-km single-mode fiber, without

using the optical amplifier. It is achieved because the mode control and low front-facet reflectivity of the

Fabry-Perot laser diodes (F-P LDs) at the OLT and ONUs enable the long-reach transmission without

optical amplifier.

The network structure in [54] employs wavelength converters at user premise to convert upstream

wavelength from ONU to OLT before it is transmitted on the long-haul fiber. Thus, the ONU can choose

inexpensive un-cooled lasers to decrease system cost. The demonstration provides 38.8 Mbps per user

to potentially 1280 users over a distance of 120 km.

As a summary, we list the main demonstrations and their characteristics in Table 2.1.

33

Table 2.1. Typical demonstrations of Long-Reach Broadband Access Networks.

Project Protocol Reach (km) Wavelengths Down/up/wavelength

(Gbps)

ONUs

ACTS-PLANET

[30 - 35]

APON 100 1 2.5/0.311 2048

B.T. [36, 37] GPON 135 40 2.5/1.25 2560

WDM-TDM [38, 39] 100 17 10/10 4352

PIEMAN [40] 100 32 10/10 16,384

WE-PON [41]

SARDANA [9, 42]

G&EPON

G&EPON

100

100

16

32

2.5/2.5

10/2.5

512

1024

2.4 Dynamic Bandwidth Assignment (DBA)

As multiple ONUs may share the same upstream channel, DBA is necessary among ONUs. Considering

the LR-PON’s benefits in CapEx and OpEx, as well as its derivation from the traditional PON, the

upstream bandwidth allocation is controlled and implemented by the OLT.

Two kinds of bandwidth allocation mechanisms are used in PON: status-reporting mechanism and

non-status-reporting mechanism. In non-status-reporting, the OLT continuously allocates a small

amount of extra bandwidth to each ONU. If the ONU has no traffic to send, it transmits idle frames

during its excess allocation. Observing a large number of idle frames from the given ONU, the OLT

reduces its bandwidth allocation accordingly; otherwise, OLT increases its bandwidth allocation when

34

observing that the given ONU is not sending idle frames. Although this mechanism has the advantage of

imposing no requirements on an ONU and no need for the control loop between OLT and ONU, there is

no way for the OLT to know how best to assign bandwidth across several ONUs that need more

bandwidth. We then focus on the status-reporting mechanism in the rest of this section.

To support the status-reporting mechanism and DBA arbitration in the OLT, the proposed DBA

algorithms in LR-PON are based on the Multi-Point Control Protocol (MPCP) specified in the IEEE

802.3ah standard. Before explaining the DBA algorithms, we briefly introduce MPCP.

MPCP is not concerned with a particular bandwidth-allocation (or inter-ONU scheduling) scheme;

rather, it is a supporting mechanism that can facilitate implementation of various bandwidth-allocation

algorithms in PON. This protocol relies on two specific messages: GATE and REPORT. Additionally,

MPCP defines REGISTER_REQUEST, REGISTER, and REGISTER_ACK messages used for an

ONU’s registration. A GATE message is sent from the OLT to an ONU, and it is used to assign a

transmission timeslot (bandwidth). A REPORT message is used by an ONU to convey its local

conditions (such as buffer occupancy and the like) to the OLT to help the OLT make intelligent

allocation decisions. Both GATE and REPORT messages are MAC (media access control) control

frames and are processed by the MAC control sublayer.

Fig. 2.8. An example of single-thread polling with stop.

35

Fig. 2.9. An example of multi-thread polling [55].

The proposed DBA algorithms work in conjunction with MPCP. The OLT has to first receive all

ONU REPORT messages before it imparts GATE messages to ONUs to notify them about their

allocated timeslot. As a result, the upstream channel will remain idle between the last packet from the

last ONU transmission in a polling cycle k and the first packet from the first ONU transmission in

polling cycle k + 1, as shown in Fig. 2.8. Here, the polling cycle is the time period (or cycle) in which

ONUs transmit sequentially. This idle time equals to the Round-Trip Time (RTT) which is needed for

the control messages to propagate along the path ONU � OLT � ONU. In a traditional PON, this idle

time is negligible because its RTT is only 0.1 ms with 10-km span. LR-PON increases the RTT to 1 ms

with 100 km of OLT-ONU distance, which results in 10x the idle time in a traditional PON.

In order to combat the detrimental effect of the increased RTT, the work in [55] proposed the

Multi-Thread Polling algorithm, in which several polling processes (threads) are running in parallel, and

each of the threads is compatible with the proposed DBA algorithms in a traditional PON. Figure 2.9

shows an example of Multi-Thread Polling (two threads are shown in the example) and compares it with

traditional DBA algorithms (so-called one-thread polling with stop). As shown in Fig. 2.9, the idle time

is eliminated because, when ONUs wait for GATE messages from OLT in the current thread which

incurs idle time in one-thread polling, they can transmit their upstream packets which are scheduled in

another thread simultaneously. Note that ONU1 and ONU2 in the illustration may not be the exact ones

36

shown in our mentioned demonstrations. They apply to ONUs which share the same upstream

wavelength.

Fig. 2.10. A two-state DBA protocol for LR-PON [56].

The research in [56] also proposed a two-state DBA protocol for LR-PON. The two states include a

normal-state cycle shown as yellow color in Fig. 2.10, and a virtual-state cycle shown as blue color in

Fig. 2.10. The normal cycle implements DBA through status-reporting mechanism, and leaves an idle

timeslot between successive transmission slots. Accordingly, the idle timeslots will constitute virtual

polling cycles, during which the ONUs can transmit data by means of a prediction method to estimate

their bandwidth requirement, as shown in Fig. 2.10.

2.5 Conclusion

Long-Reach Passive Optical Network (LR-PON) exploits the huge transmission capacity of optical

technology, and is oriented toward long-range transmission and a large user base. LR-PON is anchored

at a Central Office (CO), so that all higher-layer networking functions can now be located further

upstream in the “network cloud". The OLTs of the traditional PON (which used to sit approx. 10-20 km

from the end user) can now be replaced at the local exchange by some elementary hardware, which

contains a small amount of compact low-power physical-layer repeater equipment, such as optical

amplifiers and Optical Add-Drop Multiplexer (OADM). As a result, the telecom network hierarchy can

be simplified with the access headend closer to the backbone network. Thus, the network's Capital

37

Expenditure (CapEx) and Operational Expenditure (OpEx) can be significantly reduced.

In this chapter, we discussed the research challenges related to LR-PON, from the physical layer,

e.g., power attenuation, to the upper layer, e.g., bandwidth assignment. We reviewed some existing

LR-PON demonstrations: PLANET SuperPON initiated the research on LR-PON and extended the

network reach based on the APON model; BT’s demonstrations increased the network capacity up to 10

Gbps per channel and incorporated WDM based on a GPON model; and PIEMAN further exploited the

huge optical transmission capacity to support up to 16,384 users. WE-PON offered a novel

ring-and-spur topology to provide a better two-dimensional geographical coverage and protection for

PON traffic by using a ring. Several other demonstrations were also discussed. The impact of increased

RTT on higher-layer control was also discussed. Dynamic bandwidth allocation (DBA) algorithms, e.g.,

multi-thread polling and two-state DBA, have been proposed to remedy the impact of long propagation

delay by utilizing the idle time between transmission cycles.

A number of new technologies and designs are developing the LR-PON towards a higher-capacity

system – from a single wavelength to WDM, from less than Gbps transmission speed, e.g., 311 Mbps in

PLANET to 10 Gbps symmetric transmission, e.g., PIEMAN. GPON and EPON protocols will be

naturally inherited and be adjusted for the LR-PON, since they are mature standards in PONs today.

LR-GPON may be preferred by US and some European countries because many of their national

carriers choose GPON as first-mile solution, e.g., BT and Verizon. LR-EPON may be a choice of the

Asian countries since EPON is preferred there. Introducing LR-PON into practical use will be a gradual

process. As a first step, network operators need the LR-PON to cover the “green field” without

establishing a new OLT [57]. Then, with the consolidation of OLTs and the merging of access and

metro, we are expecting ubiquitous LR-PONs.

38

Chapter 3

Multi-Thread Polling: A Dynamic

Bandwidth Distribution Scheme in

Long-Reach PON

3.1 Introduction

In recent years, an increasing number of advanced components have been deployed for the broadband

access network, e.g., optical amplifiers which extend the span of the optical access network from 20 km

to 100 km and beyond. These solutions are referred to as SuperPON or Long-Reach PON (LR-PON) in

the literature [30, 31, 32, 33, 34, 35]. It is expected that the cost of these advanced components can drop

as they mature and higher volume production is required. Hence, it is now worthwhile to investigate

their benefits in broadband access network design.

39

(a) A Long-Reach PON with ring-and-spur structure.

(b) Logical connectivity of OLT and ONUs on the same wavelength.

Fig. 3.1. An example of LR-PON.

Another reason to investigate the new broadband optical network architectures employing these

advanced components is the advances in wavelength-division multiplexing (WDM) technology which

enables more wavelengths to be multiplexed on a fiber, with each wavelength operating (soon) at a

transmission rate of 40-100 Gbps. Using this technology in a traditional optical access network, whose

span is 10~20 km, the carrier can distribute far more bandwidth than end users’ requests due to the

limited number of users in the coverage area, i.e., a WDM-PON covering traditional PON distances may

have a lot of spare capacity. To make full use of optical capacity and serve more users in an optical

40

access network, it is worthwhile to increase the network span to cover more users, so there exists a

strong case to adopt the Long-Reach PON to increase its reach up to 100 km and beyond.

The LR-PON can simplify the network, reducing the number of equipment interfaces, network

elements, and even nodes. The Optical Line Terminal (OLT) of the traditional PON can be replaced at

the local exchange by some elementary hardware, which contains a small amount of compact

low-power physical-layer repeater equipment, such as optical amplifiers. The PON headend and all

higher-layer networking functions can now be located further upstream in the “network cloud”. As a

result, a network operator may need a lot fewer major central offices, thereby reducing capital

expenditure (CapEx) as well as their maintenance cost, namely operational expenditure (OpEx). Since

the LR-PON can cover a large metropolitan area by spanning a distance of approx. 100 km, it can

combine optical access and metro into an integrated system, where the architecture is converged and the

overhead at the interface between access and metro could be reduced significantly.

Compared with a traditional optical access network, LR-PON covers a wider geography over a

two-dimensional area. The distance from the carrier side (i.e., from the Optical Line Terminal (OLT)) to

the user side (i.e., to the Optical Network Unit (ONU)), can be up to 100 km. Figure 3.1 shows an

example of a LR-PON that can be potentially deployed in the Greater Sacramento region. Physically,

the long-reach metro fiber ring and the distributive PON segments in different cities form a

ring-and-spur structure, while the logical connectivity of OLT and ONUs sharing the same wavelength

corresponds to the tree-and-branch structure in a traditional PON. In this deployment, the ring topology

is good for network resilience and bidirectional transmission; also it provides two-dimensional coverage

(vs. the linear coverage of a traditional PON). At each node of the ring, an Optical Add-Drop

Multiplexer (OADM) is used to “add and drop” a wavelength to the end users and to compensate for the

signal power loss along the long-range transmission. Although power supply is needed in an OADM,

the term “Long-Reach PON” is not fully accurate since its components are not truly passive; but this

term, particularly SuperPON, has been used quite widely in the literature to show its derivation from the

traditional PON system with the use of limited active elements.

41

Related works on LR-PON (or Super-PON) are listed in the reference and several prototypes have

been demonstrated in recent years, such as ACTS-PLANET [30 - 35]; SuperPON by British Telecom

[36, 37]; Hybrid PON [38, 39] by University College, Cork; and WDM-Ethernet-PON (WE-PON) [41]

by Electronics and Telecommunication Research Institute (ETRI), a Korean government-funded

research institute. Table 3.1 summaries the main characteristic of these demonstrations.

Table 3.1. Long-Reach PON Demonstrations.

Project Protocol Reach (km) #Wavelengths Down/Up (Gbps) #ONUs

Planet APON 100 > 1 2.5/0.311 2048

B.T. GPON 135 40 2.5/1.25 2560

Univ.

Cork

100 17 10/10 4252

WE-PON G&EPON 100 16 2.5/2.5 512

All transmissions in a LR-PON are performed between the carrier side (i.e., OLT) to the user side

(i.e., to the ONUs). Therefore, it is a point-to-multipoint network in the downstream direction, and it is a

multipoint-to-point network in the upstream direction, as in a traditional PON. The downstream

transmission on any particular wavelength is broadcast to all users served by that wavelength, so it has

fewer challenges; but the upstream transmission needs an efficient multiple-access protocol, which is

particularly challenging now because of the much longer OLT-ONU propagation delay.

As multiple users share the upstream channel resource (e.g., users in Dixon city share lambda 16 in

Fig. 3.1), intelligent scheduling is needed to avoid collision of data transmissions from ONUs to OLT.

The reader is referred to [2, 58] for a discussion of various multiple-access schemes and their

applicability to PON-based access networks. Time-division multiple access (TDMA) is an approach

which has the advantage of simplicity. However, in [58], the authors show the limitations of a fixed

42

bandwidth allocation scheme (e.g., TDM), viz. lack of statistical multiplexing for accommodating the

bursty user traffic. To achieve an effective resource assignment scheme for ONUs, many

multiple-access schemes using a dynamic bandwidth allocation (DBA) algorithm have been proposed,

such as Interleaved Polling with Adaptive Cycle Time (IPACT) [59]. In these schemes, the OLT polls

each ONU for how many bytes of data each ONU has for transmission. After receiving the responses

from ONUs, the OLT uses its DBA scheme to assign the appropriate bandwidth (timeslot) to each ONU.

Since the polling of each ONU is interleaved, where the next ONU is polled before the transmission

from the previous one has arrived at the OLT, these schemes provide statistical multiplexing for ONUs

and result in efficient upstream channel utilization.

LR-PON extends the OLT-ONU distance from 20 km to 100 km, which increases the Round-Trip

Time (RTT) to 1 ms. When an ONU requests upstream bandwidth, it will take a much longer time to

receive the acknowledgement from the OLT before it can transmit its upstream packets. In this study,

we first analyze the packet delay due to the increased RTT in the LR-PON if the traditional OLT-ONU

signaling protocols are used; and then we propose a new approach, called multi-thread polling

algorithm, to achieve better performance.

Section 3.2 analyzes previous approaches to schedule upstream traffic in a PON, e.g., IPACT; then

it describes the multi-thread polling algorithm. Section 3.3 presents illustrative numerical examples.

Section 3.4 concludes the study.

3.2 Multi-Thread Polling

3.2.1 Applying Previous Approaches in LR-PON

To avoid collision of upstream transmissions in a PON, two control messages are used [59]: “Request”

and “Gate”. The Request message is sent to OLT from ONUs, which contains the number bytes an ONU

wants to transmit, the ONU id, and other necessary information. The Gate message is sent to ONUs

43

from OLT, which contains the number of bytes an ONU can transmit as granted by the OLT, target

ONU id, and other necessary information.

In such a scheme, every ONU, before sending its data, will send a Request message to the OLT. As

the OLT knows exactly the number of bytes each ONU wants to send, it implements the DBA scheme

and allocates the bandwidth according to the requested bandwidth and sends back to each ONU a Gate

message. This way, during temporary overload situations, the OLT can “control” the problem; also it

can avoid collisions among ONU transmissions by assigning each ONU an appropriate transmission

instant. The ONU will send the amount of bytes indicated in the Gate message as soon as it receives it.

At the end of a data transmission from an ONU, a new Request from the ONU will be piggybacked to

the OLT. It will be used in the following round of bandwidth allocation. Figure 2 shows an example of

this process (called single-thread polling), where a “thread” is a communication channel between ONUs

and the OLT. For simplicity of illustration, only two ONUs are shown, and they may be at different

distances from the OLT.

Fig. 3.2. An example of single-thread polling.

Assuming that an incoming packet arrives at ONU2, as shown in Fig. 2, the delay of the packet is

equal to [59]:

d = dpoll + dgrant +dqueue

where

t

ONU1

ONU2

OLT

request Grant data transmission packet

dpoll dgrant dqueue

44

dpoll: time between packet arrival and the next “Request” sent by that ONU. On average, dpoll = C/2,

where C is called cycle time and it is the time interval between successive Requests from the

same ONU.

dgrant: time interval between ONU’s Request for a transmission window till the grant from OLT is

received.

dqueue: queuing delay after appropriate grant from OLT arrives.

Fig. 3.3. Achieving fairness in single-thread polling.

Applying this to LR-PON, we can find the impact of increased RTT on packet delay. When RTT

increases, dgrant will increase, which results in more packets being buffered in the ONU between

successive Requests from that ONU. As a result, the number of bytes transmitted by each ONU in one

scheduling cycle, shown in Fig. 3.2 as gray area, will be increased, which in turn increases dqueue, C, and

dpoll.

If a more comprehensive evaluation is taken to incorporate the fairness issue among users, we can

see more clearly the impact of increased RTT. At the end of each scheduling cycle C, the OLT collects

Requests of all ONUs, and then calculates bandwidth allocation to each ONU using some “fair”

algorithm, based on the set of Requests. In such a way, spare bandwidth saved from lightly-loaded

ONUs can be reallocated to some heavily-loaded ONUs, and malicious Requests (e.g., from malicious

users who send overly larger Requests, trying to monopolize bandwidth) will be truncated. With this

mechanism, an idle time exists in the upstream channel between consecutive scheduling cycles, as

t

ONU1

OLT

Idle Time Scheduling Cycle

data transmission Grant Request

ONU2

45

shown in Fig. 3.3, which approximately equals the RTT between OLT and ONU. In traditional PON,

this idle time is negligible because its RTT is only 0.1 ms with 10-km span. LR-PON increases the RTT

to 1 ms with 100 km of OLT-ONU distance, which results in 10x the idle time in traditional PON.

Besides the status-reporting mechanism discussed above, non-status-reporting mechanism is also

used in PON. In non-status-reporting, the OLT continuously allocates a small amount of extra

bandwidth to each ONU. If the ONU has no traffic to send, it transmits idle frames during its excess

allocation. Observing a large number of idle frames from the given ONU, the OLT reduces its

bandwidth allocation accordingly; otherwise, OLT increases its bandwidth allocation when observing

the given ONU is not sending idle frames. Although this mechanism has the advantage of imposing no

requirements on an ONU and no need for the control loop between OLT and ONU, there is no way for

the OLT to know how best to assign bandwidth across several ONUs that need more bandwidth. We will

focus our discussion on the status-reporting mechanism in this chapter.

Some enhancement of single-thread polling has been proposed. For example, Ref. [60] proposed a

modified grant table generation algorithm to have an OLT employ some early allocation mechanism in

which a light-loaded ONU can be scheduled instantly without waiting for the end of the scheduling

cycle, but this scheme might lose efficiency at high network load. Ref. [61] predicts and schedules

constant-bit-rate (CBR) traffic to transmit during the idle time, but it works on a more detailed traffic

classification and a certain traffic pattern, e.g., the percentage of CBR traffic.

3.2.2 Idea of Multi-Thread Polling Algorithm

To achieve better performance (in terms of lower packet delay and guaranteed fairness) in a LR-PON,

an idea is to allow an ONU to send its Request before the previous Gate message is received, as shown

in Fig. 3.4, thereby creating a new “thread” of signaling between ONUs and the OLT. Again, for

simplicity of illustration, only two ONUs and two threads are shown in Fig. 3.4.

46

Fig. 3.4. Idea of multi-thread polling.

Compared with Fig. 3.2, there are two “polling processes” (threads) running in parallel in Fig. 3.4

(denoted by red and black). Let the black “polling process” be that of the traditional (single-thread)

PON. Assuming that a packet arrives at ONU2 at a time as shown in Fig. 3.4, the “Request” message

will be sent in the second “polling process” (denoted by red), instead of waiting until the end of data

transmission of the first “polling process”. It is straightforward to note, by comparing Figs. 3.2 and 3.4,

the three components of packet delay dm_grant, dm_queue and dm_poll (we add an ‘m’ to denote ‘multiple

threads’) are now significantly decreased.

The multiple-thread polling can also eliminate the idle time, whereas keeping the fairness, because

the transmission of Gate messages is interleaved with upstream data transmission in another polling

process. As in Fig. 3.4, after the upstream transmission in “gray” thread, ONUs don’t have to wait for

Gate messages from OLT in current “gray” thread; instead, they start their upstream transmission in

“red” thread, according to Gate messages in “red” thread received before.

In this study, we call the proposed scheme “multi-thread polling” because it has parallel “polling

processes” running simultaneously, compared with one-thread polling in previous work. The number of

threads is not limited to two. It can be increased depending on the network environment, such as

hardware processing time, required delay bound, etc. We will show the effect of the number of threads

on the packet delay.

t dm_poll dm_grant dm_queue

request Grant data transmission packet

ONU1

ONU2

OLT

C1inst C2

ins

47

3.2.3 Multi-Thread Polling Algorithm

In this subsection, we give a high-level description of our proposed algorithm. For simplicity of

illustration, we consider a system of an OLT and two ONUs: ONU1 and ONU2, and two threads: thread

1 and thread 2. For more ONUs and threads, the same logic can be applied.

OLT maintains a polling table, shown in Fig. 3.5. In the table, each ONU has an entry which

records the ONU’s RTT and its most recent requests in each thread (T1 and T2).

1. Consider time t0, when OLT allocates bandwidth in thread T1, and OLT knows the requested

bytes for each ONU in threads T1 and T2 from the polling table. At time t0, the OLT sends a Gate

message to ONU1, allowing it to send 5000 bytes as indicated in the polling table.

2. When ONU1 receives the Gate message from the OLT, it starts transmitting its data up to the

size of the granted window, i.e., up to 5000 bytes. As ONU1 keeps receiving data from end users

during the time interval it waits for the acknowledgement from the OLT, ONU1 will generate its

own Request, containing this aggregated data size till the Request is generated. This Request is

piggybacked to the data transmitted to the OLT. In our example, the new Request is 4500 bytes,

as shown in Fig. 3.5(a).

OLT

ONU1

ONU2

TX

RX

TX

RX

TX

RX

5000

5000 bytes 4500

5000

1000400020002

500480050001

RTTT2T1ONU

1000400020002

500480050001

RTTT2T1ONU

1000400020002

500480045001

RTTT2T1ONU

1000400020002

500480045001

RTTT2T1ONU

t05000 bytes 4500

(a)

48

OLT

ONU1

ONU2

TX

RX

TX

RX

TX

RX

5000

5000 bytes 4500

5000

2000

3500

3500

2000

1000400020002

500480045001

RTTT2T1ONU

1000400020002

500480045001

RTTT2T1ONU

1000400035002

500480045001

RTTT2T1ONU

1000400035002

500480045001

RTTT2T1ONU

5000 bytes 4500

2000 bytes

2000 bytes

(b)

OLT

ONU1

ONU2

TX

RX

TX

RX

TX

RX

5000

5000 bytes 4500

5000

45005000 bytes

2000

2000

4800

4800

30004800 bytes

30004800 bytes

1000400035002

500480045001

RTTT2T1ONU

1000400035002

500480045001

RTTT2T1ONU

1000400020002

500300045001

RTTT2T1ONU

1000400020002

500300045001

RTTT2T1ONU

3500

2000 bytes

2000 bytes

3500

(c)

1000400035002

500300045001

RTTT2T1ONU

1000400035002

500300045001

RTTT2T1ONU

1000250035002

500300030001

RTTT2T1ONU

1000250035002

500300030001

RTTT2T1ONU

4000 bytes

4000

2500

4000 bytes 2500OLT

ONU1

ONU2

TX

RX

TX

RX

TX

RX

5000

5000 bytes 4500

5000

45005000 bytes

2000

2000

4800

4800

3000

30004800 bytes3500

2000 bytes

2000 bytes3500

4800 bytes

(d)

Fig. 3.5. Steps of multi-thread polling.

49

3. OLT knows exactly when the last bit of ONU1’s transmission will arrive even at the time it

sends Gate to ONU1, because the OLT knows the RTT and granted transmission window size

for ONU1. Then, based on the last bit’s arrival time from ONU1 and the RTT for ONU2, the

OLT can schedule a Gate message to ONU2 such that the first bit from ONU2 will arrive with a

small guard interval after the last bit from ONU1. In the example, ONU2 is granted to send 2000

bytes, as shown in Fig. 3.5(b).

4. Before the new Request from ONU1 arrives, OLT schedules the Gate message of thread T2 to

ONU1, shown by a red arrow in Fig. 3.5(c). Similar to Step 3, the OLT schedules the Gate

message to ONU1 such that the arrival of the last bit from ONU2 and the first bit from ONU1 are

separated by a guard interval. Here, the OLT allows ONU1 to send its requested 4800 bytes in

thread T2, which was registered in the polling table by previous Request of ONU1 in T2. Upon

receiving the grant, ONU1 transmits data up to its granted window, and piggybacks its new

Request to update the polling table in the OLT. Note that, if OLT still schedules the bandwidth

to ONU1 in thread T1, the upstream bandwidth can not be fully utilized. Even if the OLT sends

the Gate message as soon as the new Request of thread 1 from ONU1 arrives, e.g., 4500 bytes,

there is still a wider interval than the guard interval between the last bit of the previous ONU2

transmission and the first bit of ONU1 transmission.

5. Then, OLT schedules Gate message to ONU2 in thread T2, as shown in Fig. 3.5(d). Similar

calculation as explained in Step 3 applies. The new Gate message allows ONU2 to transmit its

requested 4000 bytes in T2. Upon receiving the Gate message, ONU2 sends 4000 bytes and

piggybacks the new Request. When OLT receives the Request (2500 bytes), it updates the

polling table.

3.2.4 Control Frame Design

Multi-Point Control Protocol (MPCP) specified in IEEE 802.3ah defines control messages between a

master unit (OLT) and slave units (ONUs) connected to a point-to-multi-point segment to allow

50

efficient transmission of data [2]. GATE (Grant) and REPORT (Request) messages mentioned above

are specified in MPCP control messages, which are 64-byte medium-access control (MAC) frames.

Besides the information of source, destination, timestamp, etc., MPCP reserves 44-byte

“opcode-specific fields” for specific MPCP functions.

Multi-thread polling scheme uses the reserved 44 bytes to carry the information (REPORT: 2-byte

requested window size and 1-byte thread number; GATE: 2-byte granted window size, 2-byte grant start

time, and 1-byte thread number). The remaining bytes can be reserved for window requesting and

granting for multiple priority queues if Class of Service (CoS) is addressed in future research.

3.2.5 Initiating and Tuning Multiple Threads

We define the following symbols:

Tn: thread n, n is the index of threads, Nn ≤≤1 ;

Cninst

: instantaneous cycle time of thread n, e.g., C1inst

in Fig. 3.4; and

Max(RTT): maximum round-trip time of ONUs.

When OLT initiates multiple threads, the initial cycle time of each thread is set to value t. The

relation of t and the total number of threads N is:

+=

t

TRTTMaxN

process)( (3.1)

where Tprocess is the Request processing time at the OLT. The reason to choose maximum RTT of ONUs

is that multiple threads can be initiated during the RTT interval in order to gain the benefit of “Request

before Gate” of multi-thread polling.

After initiation, instantaneous cycle time of each thread may change due to the dynamics of the

upstream traffic. This fluctuation might impose a negative effect on the packet delay. Imagine a

situation where two threads T1 and T2 are used in multi-thread polling algorithm. If C1inst

>> C2inst

,

51

threads T1 and T2 are almost merged together to be one thread, which actually degrades the multi-thread

polling to single-thread polling.

To prevent the degrading of multi-thread polling due to the fluctuation of the cycle time of the

thread, the multi-thread polling algorithm tunes its threads during its operation. Defining a tuning

threshold Thtune, if the adjacent threads, e.g., Tn-1 and Tn and ratio tuneinst

n

inst

n ThC

C>−1

, Tn-1 and Tn will be

“tuned”: i.e., Cn-1inst

will be decreased and Cninst

will be increased. This can be achieved by moving some

bandwidth ∆ allocated to ONUs in thread Tn-1 to thread Tn, such that tuneinst

n

inst

n ThC

C<

∆+

∆−−1. In order to

calculate the range of ∆, we compute ∆ in the following two border conditions:

=∆+

∆−

=∆+

∆−

−

−

threads) e succeesivof cycle (equal C

C

threshold) of (border ThC

C

inst

n

inst

n

tuneinst

n

inst

n

11

1

(3.2)

So, we have the range of ∆ as:

21

11

inst

n

inst

n

tune

tune

inst

n

inst

n CC

Th

ThCC −<∆<

+

×− −− (3.3)

Applying the “tuning” method in the multi-thread polling algorithm, a justification has to be

considered. Recall when we compared the instantaneous cycle time of the currently-scheduled thread Tn

with its predecessor thread Tn-1, where thread Tn-1 has been scheduled and its Gate messages have been

sent out by OLT. As a result, it is not practical to reassign the allocated bandwidth in thread Tn-1. The

modification adopted in the multi-thread polling algorithm is to have the OLT record ∆ (negative value

for thread Tn-1 and positive value for thread Tn) and tune threads Tn-1 and Tn in the next round.

Although packets in this moved bandwidth may encounter increased delay, in the long run, the

overall system performance will be improved, as will be shown in the numerical results.

52

3.2.6 Inter-Thread Scheduling

In each thread, OLT does bandwidth allocation, and distributes Gates to all ONUs. In multi-thread

polling, OLT can make use of not only the information of Requests in the current thread, but also the one

in subsequent threads before the time the OLT calculates bandwidth allocation. For example, consider

that three threads T1, T2, and T3 are used in multi-thread polling. Before OLT calculates bandwidth

allocation in T1, Requests in T2 have arrived, which report the latest information of ONUs’ packet

queues. This information will be counted in the bandwidth allocation in T1. Thus, packets arriving at

ONUs in T2 will not be queued until Gates of T2 are received; instead, they can be transmitted in T1. So,

the average packet delay can be further improved.

3.2.7 Achieving Fairness among ONUs

Let us define:

Ri, n: requested bandwidth for ONUi in thread n, Nn ≤≤1 , Ii ≤≤1 ;

Bi, n: granted bandwidth for ONUi in thread n; and

TMax_C: maximum cycle time of a thread.

Below, we discuss the bandwidth allocation in two scenarios: normally loaded and heavily loaded.

When the aggregated bandwidth request is less than the maximum cycle time, i.e, CMax

i

ni TR _, ≤∑ ,

the network is normally loaded and each ONU is allocated bandwidth as it requests, i.e., nini RB ,, = .

When the aggregated bandwidth request is more than the maximum cycle time, i.e.,

CMax

i

ni TR _, >∑ , the network is temporarily heavily loaded. In this situation, bandwidth requests of some

heavily-loaded ONUs may not be fully granted, in order to provide the minimum guaranteed bandwidth

(denoted by BMin) of other ONUs in condition of heavily network load,

where IBNTB gCMaxMin /)( _ ×−= , I is the number of ONUs, and Bg is the guard time separating two

53

consecutive transmission windows of ONUs. It is straightforward to note in the above formula that MinB

is a qualified fair share bandwidth quota of each ONU. IPACT [59] used limited bandwidth scheme to

allocate bandwidth to ONUi as follows:

≥

<=

MinniMin

Minnini

niBR if B

BR if RB

,

,,

, (3.4)

Due to the bursty nature of Internet traffic, even when the network is temporarily heavily loaded,

there are still some ONUs which might be lightly loaded while some other heavily-loaded ONUs might

have more traffic to transmit thanMinB . The aggregation of spare capacity saved from lightly-loaded

ONUs forms a total excess bandwidth ∑∈

−=Mi

niMinexcess RBB )( ,, where M is the set of lightly-loaded

ONUs whose Minni BR ≤,

. This excess capacity can be distributed fairly among highly-loaded ONUs,

and we exploit the excess bandwidth as follows:

Mini

excess

iMinni BR if BBB ≥+=,

∑∈

×=Qi

niniexcess

excess

i RRBB ,, / (3.5)

where Q is the set of heavily-loaded ONUs for whichMinni BR >,

.

To capture the various properties of the multi-thread polling algorithm, e.g., tuning threads,

inter-thread scheduling and fair scheduling, Fig. 3.6 shows the corresponding pseudo code.

54

Fig. 3.6. Pseudo code for multi-thread polling scheduling.

3.2.8 Analysis

We continue the analysis in Section 3.2.1 to investigate the benefits of multi-thread polling. We assume:

Check ∆ for current thread n;

)( _, CMax

i

ni TRif <∆+∑ {

For (k = n; k MOD N ≠ n; k++){

// Requests in thread k and after have not arrived.

break;

Rifi

ki )0( , =∑

// If valid requests in thread k, then calculate and update requests in thread k and n.

){( _,, CMax

i

ni

i

ki TRRif ≤∆++∑∑

;0

;

,

,,

=

=+

ki

kini

R

RR

}

else{

;_,, CMax

i

ni

i

kiexcess TRRR −∆++= ∑∑

;

;/

;/

,,,

,,,

break

RRRR

RRRR

i

kiexcesskiki

i

kiexcesskini

∑

∑

×=−

×=+

}

}

}

fair scheduling;

55

Cn: average cycle time for each thread, Nn ≤≤1 ;

C: average cycle time for traditional single-thread polling; and

tg: guard time between successive threads, typical value = 2 µs.

Comparing Figs. 3.2 and 3.4, it is found that in the time line the multiple threads actually split the

traditional thread. As a result, average cycle time C1 to CN of each thread constitutes the average cycle

time C in single-thread polling:

g

N

i

i NtCC +=∑=1

(3.6)

To compare with single-thread polling, we present the packet delay dm in multi-thread polling as:

dm = dm_poll + dm_grant + dm_queue (3.7)

where

dm_poll: time between packet arrival and the next “Request” sent by that ONU;

dm_grant: time interval between ONU’s Request for a transmission window till the grant from OLT

is received; and

dm_queue: queuing delay after appropriate grant from OLT arrives.

Then, we analyze the three components of delay as follows.

�dm_poll

It is obvious that the more threads used, the less time a packet will wait for its Request being

transmitted. An extreme case is that the ONU transmits a Request immediately after a packet arrival. On

average,

dm_poll = dpoll / N = C / 2N (3.8)

�dm_grant

56

In our discussion, we consider two scenarios: (1) normal case where the OLT grants an ONU

bandwidth as it requested, i.e., no overload situation; and (2) overload situation, where a Request might

have to wait for multiple cycle times before Gate is issued.

In the first case, a Request in thread n is granted to transmit in the next cycle of thread n:

dm_grant = mg

n

i

i

N

ni

i RTTNtCC ∆+−++∑∑−

=+=

2/1

11

= C - Cn – RTT/2 + ∆m (3.9)

where ∆ m < Cn is the time interval when the ONU which sent the Request has to wait for its time

window in the cycle of thread n.

Considering the inter-thread scheduling in our proposed algorithm, dm_grant can be optimized by

granting transmission in a thread before thread n. But RTT is indispensable for the propagation of

Request and Gate messages. So, we have the lower bound of dm_grant:

Min(dm_grant) = RTT + tp (3.10)

where tp is the processing time at OLT.

Based on the discussion above, we have:

RTT + tp ≤ dm_grant < C - Cn – RTT/2 + ∆m (3.11)

In the second case, an ONU’s transmission encounters longer delay due to overloaded network

traffic, so:

dm_grant = kC- Cn – RTT/2 + ∆m (3.12)

where k will be increased if network load is overloaded.

In both cases above, dm_grant can be further decreased if inter-thread scheduling is adopted to have a

Request be granted in earliest available thread.

�dm_queue

57

In multi-thread polling, each thread has less data to transmit in its cycle compared with

single-thread polling. On average,

dm_queue = dqueue / N (3.13)

Since PON is sensitive to DBA complexity, our proposed multi-thread polling algorithm tries to

balance between efficiency and complexity. Although multiple threads require more computation,

these threads are implemented independently and do not need control information exchanged between

them in most cases. In this way, the complexity grows linearly relative to single-thread polling.

Generally, N threads mean N times more computation than single-thread polling. The number of

threads is also constrained. The reason is two-fold. Larger number of threads results in smaller cycle

time and decreased packet delay, while it decreases the available bandwidth for data transmission by

using more bandwidth for control messages and guard band. In our experiments, the optimal number of

threads is three as we will see below.

3.3 Illustrative Numerical Results

We simulate a LR-PON with M ONUs. We let single or multiple wavelengths to be used to carry the

upstream traffic, depending on the aggregated user requests. To make the problem clear and focused on

packet delay, we assume that the M ONUs share the same upstream channel (one upstream wavelength).

If more users join the network, a new upstream channel can be assigned to them and the same polling

scheme is implemented on the new channel independently.

From the access side, packets arrive at an ONU from a user connected to the ONU. Packets are

buffered in the ONU until the ONU is allowed to transmit them to the OLT. In our model, we consider

RD to be the data rate of the access link from a user to an ONU, and RU to be the date rate of the upstream

channel from ONU to OLT. Note that, if DU RMR ×> , the bandwidth utilization problem does not

exist, because the system capacity is greater than the aggregated load from all ONUs. (This should be

58

the steady-state case, but during temporary overload situations due to bursty traffic, the load may be

higher than capacity.) So, we choose M = 16 and RU and RD to be 1000 Mbps and 100 Mbps,

respectively, as in previous studies [59].

We generate packets in the form of Ethernet frames (64 to 1518 bytes) and packets arrive at each

ONU from the end user. To reflect the property of Internet traffic, the generated user traffic is

self-similar by aggregating multiple sub-streams [59], each consisting of alternating Pareto-distributed

on/off periods, with a Hurst Parameter of 0.8. The buffer size at each ONU is limited to 10 Mbytes.

To study the performance of our multi-thread polling algorithm, we simulate the LR-PON in two

different scenarios, where ONUs are 20 km and 100 km from the OLT, respectively. The initial interval

of threads is set to be 0.3 ms for the 100-km scenario and 0.1 ms for 20-km scenario; the tuning

threshold Thtune is set to be 5 in the simulation. To highlight the benefit of multi-thread polling on the

average packet delay, we drop the constraints on single-thread polling by allowing it to ignore the

fairness issue, and thereby the idle time (shown in Fig. 3.3) is not counted, as shown in Fig. 3.2.

0.1

1

10

100

1000

0.1 0.3 0.5 0.6 0.7 0.8 0.9 0.95 1.02 1.15

Load (Normalized)

Avera

ge p

acket

dela

y (

ms)

Single Thread Multi-Thread

Fig. 3.7. Average packet delay with 20-km span.

Figure 3.7 shows the average packet delay when ONUs are 20 km from the OLT. This (20 km) is

the maximum distance of traditional EPON [58]. The load is normalized to the upstream channel rate

(1 Gbps). When load is not very high, such as 0-0.8, single-thread polling has good performance to keep

59

the average packet delay less than 3 ms, because of small RTT. But when load increases to be more than

0.8, the average packet delay increases sharply. For example, the delay is 11.84 ms at load 0.9. The

reason for high delay at the high load is the increased number of bytes each ONU has to transmit in each

polling cycle C, which increase dqueue, dgrant, and dpoll. To make matters worse, the bursty nature of

self-similar traffic may generate high-volume instantaneous traffic, which significantly increases the

average packet delay. Meanwhile, the multi-thread polling algorithm has less delay even at high load.

For example, it has average packet delay of 2.987 ms at load 0.9. This is because, by using multiple

threads, each ONU needs to send fewer bytes of data in each thread. Actually, the large polling cycle

time C of single-thread polling at high load is divided into multiple smaller cycle times of threads by

multi-thread polling.

When load is not very high, e.g., 0-0.8, there is not much difference between single-thread polling

and multi-thread polling, because the RTT at 20 km is only 0.2 ms and the polling cycle C of

single-thread polling will not increase too much at light load. When load is more than 1, the 10-Mbyte

ONU buffer will overflow, which produces the same packet delay for single-thread and multi-thread

polling algorithms.

1

10

100

1000

0.1 0.3 0.5 0.6 0.7 0.8 0.9 0.95 1.02 1.15

Load (Normalized)

Avera

ge p

acket

dela

y (

ms) Single Thread Multi-Thread

Fig. 3.8. Average packet delay with 100-km span.

Figure 3.8 shows the average packet delay when ONUs are 100 km from the OLT. This distance

(100 km) is the expected coverage distance of a LR-PON. As RTT increases to 1 ms (for 100 km),

60

multi-thread polling shows better performance on average packet delay even at light loads. This is

because multi-thread polling decreases dm_poll and dm_queue, although ONUs do not have much data to

send in each cycle time.

1

10

100

1000

0.1 0.3 0.5 0.6 0.7 0.8 0.9 0.95 1.02 1.15

Load (Normalized)

Avera

ge p

acket

dela

y (

ms)

with without

Fig. 3.9. Comparison of tuning multiple threads.

2

3

4

0.1 0.2 0.3 0.4 0.5Initial Cycle Time of Thread (ms)

Avera

ge p

acket

dela

y

(ms)

Fig. 3.10. Average packet delay at different initial cycle time.

Figure 3.9 shows the effect of tuning the multiple threads. At light load, e.g., 0.1-0.4, there is not

much improvement, because the possibility of increasing the thread interval to overrun the tuning

threshold is very low at light load. When the load grows, the tuning scheme shows its benefit of

decreasing the average packet delay.

61

Figure 3.10 shows the average packet delay when choosing different initial cycle time of thread at

a load of 0.8 with 100-km OLT-ONU distance. The best performance is achieved when setting the initial

thread interval to 0.2 ms. The reason is two-fold. First, smaller initial thread interval means more threads

are running in parallel, which results in smaller cycle time and decreased packet delay. Second, more

threads mean that more bandwidth is used for control messages and guard band. As a tradeoff, a good

compromise is achieved at initial thread interval of 0.2 ms. But considering the computational

complexity of practical hardware and to save the bandwidth from control messages, we choose this

value as 0.3 ms in our LR-PON simulation.

0.00.10.20.30.40.50.60.70.80.91.0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1Load (Normalized)

Th

rou

gh

pu

t

Single Thread Multi-Thread

Fig. 3.11. Throughput vs. offered load.

Figure 3.11 compares the improvement in throughput when considering the fairness among ONUs.

In such a way, single-thread polling has to introduce “idle time” between its successive scheduling

cycles. Multi-thread polling achieves a maximum throughput of 96%, compared with single-thread

polling’s maximum throughput of 76%. As 1-ms idle time (RTT of 100 km OLT-ONU distance) exists

62

between consecutive 4-ms maximum cycles, the upper bound of throughput for single-thread polling is

79%, without considering the guard time and OLT’s computation time.

3.4 Conclusion

In this study, we addressed the problem of dynamic bandwidth allocation in a Long-Reach PON

(LR-PON). We proposed a multi-thread polling algorithm to remedy the effect of the long CO-to-Users

“control loop”. Moreover, we achieved insights on the algorithm by analyzing its key parameters, such

as initiating and tuning multiple threads, inter-thread scheduling, and fairness among users. Numerical

results show that, by setting the proper initial thread interval and tuning threshold, the average upstream

packet delay is decreased, especially at high traffic load; and the network throughput is increased.

63

Chapter 4

A SLA-Aware Protocol for Efficient

Tunable Laser Utilization to Support

Incremental Upgrade in Long-Reach

Passive Optical Networks

4.1 Introduction

Emerging web applications require high bandwidth, such as Internet Protocol TV (IPTV),

video-on-demand (VoD), video file swapping, peer-to-peer applications, real-time network games, etc.,

which significantly exceed the capacity of traditional Digital Subscriber Line (DSL) or Cable Television

(CATV) access technologies. The explosive demand for bandwidth is leading to new access network

architectures which are bringing the high capacity of the optical fiber closer to the residential homes and

small busineses [19]. The FTTx models -- Fiber to the Home (FTTH), Fiber to the Curb (FTTC), Fiber

64

to the Premises (FTTP), etc. -- offer the potential for unprecedented access bandwidth to end users (up to

100 Mbps per user). These technologies aim at providing fiber directly to the home, or very near the

home, from where technologies such as DSL or wireless can take over. FTTx solutions are mainly based

on the Passive Optical Network (PON). Developments in PON in recent years include Ethernet PON

(EPON), ATM-PON (APON, based on ATM), Broadband-PON (BPON, based on APON, adding

support for WDM and higher bandwidth), Gigabit-PON (GPON, an evolution of BPON, supporting

higher rates and multiple layer-2 protocols), and wavelength-division-multiplexing PON (WDM-PON)

[21].

PON reduces the network cost by eliminating the power supply (operational cost) along the fiber

path from the telecom Central Office (CO or equivalent), to end users, and by sharing the significant

portion of the network cost among multiple users. However, the cost reduction offered by a PON might

not be enough for the future telecom network, e.g., research shows the realization of fiber access

throughout United Kingdom would cost around ₤15 billion [ 62 ]. The economic driver behind

technology calls for a new thinking on the broadband optical access.

An alternative technology, called Long-Reach Passive Optical Network (LR-PON) [30-35], has

been proposed as a more cost-effective solution for the broadband optical access network. LR-PON

extends the coverage span of PONs mentioned above from the traditional 20-km range to 100 km and

beyond by exploiting Optical Amplifier and WDM technologies. A general LR-PON architecture is

achieved by an extended shared fiber connecting the CO and the local user exchange, and optical splitter

connecting users to the shared fiber. Compared with a traditional PON, LR-PON consolidates multiple

Optical Line Terminals (OLTs) and the Central Offices (COs) where they are located, thus significantly

reducing the corresponding Operational Expenditure (OpEx) of the network. By providing extended

geographic coverage, LR-PON combines optical access and metro into an integrated system. Thus, cost

savings are also achieved by replacing the Synchronous Digital Hierarchy (SDH) or Synchronous

Optical Network (SONET) (which are prevalent technologies in metro networks) with a shared optical

65

fiber. In general, the LR-PON can simplify the network, reducing the number of equipment interfaces,

network elements, and even nodes.

Although the idea of extending the reach of a PON has been around for a while, it is emphasized

recently because the optical access is penetrating quickly into residential and small-business markets,

and the simplification of the telecom network requires an architecture to combine the metro and access

networks. The traditional telecom network consists of the access network, the metropolitan-area

network, and the backbone network (also called long-haul or core network). However, with the

maturing of technologies for long-reach broadband access, the traditional metro network is getting

absorbed in access. As a result, the telecom network hierarchy can be simplified with the access headend

close to the backbone network.

Figure 3.1 shows an example of a LR-PON that can be potentially deployed in the Greater

Sacramento region. Physically, the long-reach metro fiber ring and the distributive PON segments in

different cities form a ring-and-spur structure, while the logical connectivity of OLT and ONUs sharing

the same wavelength corresponds to the tree-and-branch structure in a traditional PON. The OLTs of the

traditional PON (which used to sit approximate 20 km from the end user) can now be replaced at the

access node by some elementary hardware, which contains a small amount of compact low-power

physical-layer repeater equipment, such as optical amplifiers and Optical Add-Drop Multiplexers

(OADMs). An OADM is used to drop downstream wavelengths and add upstream wavelengths to/from

its connected target PON segment. The OADM can be reconfigurable in a more practical setting (i.e., a

ROADM) to support dynamism in network configurations and accommodate traffic variations. In

cooperation with the dynamism in the access node, the ONUs in each PON segment can be

wavelength-agnostic by using Reflective Semiconductor Optical Amplifiers (RSOA). The downstream

wavelengths are remodulated with the upstream signal at an ONU and then added to the metro ring

through the ROADM.

66

Although a limited amount of low-power components is needed, the term LR-PON has been used

quite widely in the literature to show its derivation from the traditional PON system with the use of

limited active elements.

In LR-PON, when the supported bandwidth in each PON segment needs to be increased (because

of users demanding more services/bandwidth), more wavelength(s) need to be added to each user group.

A traditional approach is to add an additional wavelength to the PON segment. The disadvantage of this

approach is that wavelengths will be dedicated to each PON segment. The utilization of the added

wavelengths may be low, because the addition of services in a PON segment is gradual, and the

cumulative user bandwidth request is not expected to increase on the order of the channel bit-rate (i.e.,

from 1 Gbps to 2 Gbps). Rather, addition of new services and demands for additional bandwidth from

users are expected to occur incrementally (in small steps, e.g., in the range of 10 Mbps per user). In the

traditional approach, the service provider may have to invest a lot in supporting many wavelengths

although the aggregated demand of all user groups may not be very high.

To address the drawbacks of the above wavelength allocation approach, we exploit the dynamism

of ROADM and wavelength-agnostic ONU to enable some added wavelengths which can be shared

among the various PON segments. In particular, the CO uses a tunable transmitter to tune to the added

wavelengths and to let the wavelengths target for different PON segments in a TDM manner. We

propose a wavelength allocation protocol to allow dynamic wavelength allocation under a realistic

traffic model (i.e., bursty traffic), and accommodate smooth upgrade of a LR-PON for added

wavelengths. We demonstrate how the network economics improves in terms of the number of tunable

lasers required to support a given user bandwidth request.

Since we have provided capacity to users, we then address the problem of how to provide

guaranteed Quality of Service (QoS) by utilizing the dynamism of LR-PON and optimizing the

transmission of traffic flows. Traffic flows carrying services over the access network from CO to the

ONUs are mainly classified into two categories: (i) Streaming Flows, such as Video on Demand (VoD)

and video conferencing, and (ii) Best Effort (BE) flows, such as file downloading and web browsing [9].

67

A streaming flow has strict delay and bandwidth requirements, more than the ones of the BE flow. To

support these different flow characteristics and differentiate the user’s requests, a Service-Level

Agreement (SLA) is defined. The SLA of a user specifies (i) a number of streaming flows, each of

which has guaranteed average bandwidth and maximal packet delay, and (ii) BE flows with aggregated

average bandwidth. So, now the problem is how to provide QoS according to the user SLA. We propose

a user-SLA-aware bandwidth allocation scheme based on flow scheduling by effectively utilizing the

dynamism of sharing wavelengths among the PON segments. The scheme guarantees the user SLA and

prevents an user’s illegal flow requests from deteriorating network performance, especially at heavy

traffic load.

The study is organized as follows. Section 4.2 describes our proposed dynamic wavelength

allocation protocol to upgrade the LR-PON capacity. Section 4.3 describes the user-SLA-aware

bandwidth allocation scheme based on flow scheduling. Section 4.4 analyzes the algorithm and

discusses some key properties. Section 4.5 presents illustrative numerical results. Section 4.6 concludes

the study.

4.2 Dynamic Wavelength Allocation

In this section, we discuss our dynamic wavelength allocation scheme in LR-PON. Please refer to the

notations of symbols in Table 4.1.

We assume the downstream traffic from CO to user j of PON segment i will be buffered in input

queue qij at the CO, where Ni ≤≤0 and Mj ≤≤0 . In each scheduling cycle, the scheduler may

allocate extra tunable wavelengths to a PON segment by activating the tunable lasers in case its traffic is

heavily loaded. Similarly, some tunable lasers will be deactivated to free extra wavelengths from

lightly-loaded PON segments.

68

Table 4.1: Symbols for the scheduling algorithm.

Symbol Meaning Typical Value

C Capacity of a wavelength channel 1 Gbps

N Number of PON segments 4

M Number of users in a PON segment 32

reqij Current size of input queue dedicated to user j in PON segment i 0 – 10 Mbytes

aij Allocated bandwidth to input queue qij 0 – 10 Mbytes

Ttuning Tuning time of the tunable laser 10 ns

R Number of wavelengths that can be received by an ONU 3

Rcurr_i Number of wavelengths currently received by PON segment i 0 – 3

LTh Lower buffer threshold for dropping extra wavelength 0.5 Mbits

UTh Upper buffer threshold for adding an extra wavelength 5 Mbits

T Number of tunable lasers in the OLT 1 - 10

Tcurr Number of tunable lasers currently in use in the OLT 1 - 10

TCycle Cycle time of scheduling 5 ms

69

Assign Bandwidth a i j for

all allocated λ’s, update

req i j to remaining requests,

and calculate UTh.∑ j (req i j) >UTh

∑j (req i j) < LTh

Tunable laser used

by user group i?

Free the

tunable laser

Y

Y

Y

N

N

N

Scheduling

done

Start

End

(T-TCurr>0)

and

(R-RCurr>0)

An additional λassigned

by tuning laser, assign aij

for this λ, update reqij to

remaining requests, and

update UTh.

Y

N

Fig. 4.1. Dynamic wavelength allocation flowchart.

Figure 4.1 shows the scheduling process in each scheduling cycle. For each PON segment i, the

scheduler first checks if there is a tunable laser used by this segment. It may be freed by comparing the

aggregated request of the users in PON segment i with LTh. Then, the scheduler allocates bandwidth

across all the allocated wavelengths for each user in segment i. After that, if the remaining aggregated

bandwidth requests for the segment i is still high, e.g., more than UTh, extra tunable wavelengths will be

assigned as available until the remaining request is less than UTh or there are no more available tunable

transmitters (at CO) or receivers (at ONU). In general, when traffic in a PON segment is overloaded, the

corresponding buffer size in the CO will increase to trigger the added wavelengths through the

scheduler. When traffic in a PON segment is lightly loaded, the corresponding buffer size in the CO will

decrease to release the added wavelengths gradually.

When assigning the allocated wavelengths to each PON segment, the scheduler should then

allocate bandwidth to each ONU in the PON segment, satisfying the QoS according to SLA. The

70

protocol in the flowchart (Fig. 4.1) does not describe the bandwidth allocation in detail, and leaves it to

the following section.

4.3 User-SLA-Aware Bandwidth Allocation

4.3.1 Flow Classification and User-Defined SLA

The access service bundle may have three types of service flows: voice, streaming video, and data.

Voice flow uses pulse-code modulation at constant bit rate. The streaming flow is a collection of various

kinds of video streaming with coding schemes such as MPEG2 [63] or H.264 [64]. Important properties

of these flows are burstiness and stringent QoS requirements. Another type is data flow. Although data

flows are bursty, delay and jitter are less important issues. These are Best Effort (BE) flows and are

given lower priority than the streaming flows.

Based on the above service flow classification, it is not sufficient to describe user SLA only by

average guaranteed bandwidth as has traditionally been the case. A more detailed scheme of

guaranteeing certain number of flows of specified average bandwidth is a desirable solution. Therefore,

we define the user SLA as a number of streaming flows provided with guaranteed average bandwidth

and maximal packet delay for each of them, and BE flows provided with total average bandwidth. As an

example, a user SLA may guarantee two streaming flows with average 20 Mbps (MPEG2 video) each

and maximal packet delay less than 5 msec, as well as 20 Mbps bandwidth for aggregated BE flows. A

user can specify its own SLA profile in its contract with the carrier. To meet such an user SLA, we

propose the SLA-aware QoS algorithm in the following subsection.

4.3.2 User-SLA-Aware Bandwidth Allocation Algorithm

In LR-PON, the input queue qij for buffering downstream traffic of user j in PON segment i consists of

subqueues, each of which is dedicated to a downstream data flow for user j. The incoming traffic flows

71

from the backbone network or generated at the CO will be buffered in these queues and then they will be

transmitted through the allocated wavelengths to different PON segments.

The user-SLA-aware bandwidth allocation is based on flow scheduling. It can be implemented

independently for each PON segment. To help understand the algorithm, we define the following

symbols:

sijk: kth streaming flow of user j in PON segment i;

req(sijk): size of subqueue for the kth streaming flow of user j in PON segment i;

eijk: kth BE flow of user j in PON segment i;

req(eijk): size of subqueue of the kth BE flow of user j in PON segment i;

rinst(s), rinst(e): instantaneous rate of a streaming or BE flow, which is calculated as

rinst = total bits of the flow transmitted during the last cycle / TCycle;

l(s), l(e): Boolean variable, indicating whether a streaming or BE flow by user SLA is legal;

SLA (j): SLA of user j;

NS(SLA(j)): number of streaming flows of SLA(j); and

BWBE(SLA(i)): aggregated average bandwidth of BE flows of SLA(i).

For each subgroup, the algorithm maintains two tables: a SLA table which records all the users’

SLAs in the subgroup; and a flow table which records each flow’s flow ID, instant rate, type (streaming

or BE), status (legal or not according to user SLA), user who requested the flow, and routing

(transmitted by which allocated wavelength).

The user-SLA-aware bandwidth allocation is composed of three parts: admission control, cycle

time scheduling, and termination control, which are triggered by the generation of a flow, scheduling

cycle, and termination of a flow, respectively, as shown in Fig. 4.2. We will explain the three parts one

by one.

72

Fig. 4.2. Three parts of user-SLA-aware bandwidth allocation.

4.3.2.1 Admission Control

When a new flow, e.g., sjk or ejk, arrives at the CO, the admission control is triggered. It identifies the

flow (including flow ID, type, legal, etc.) and registers it in the flow table. An important content of the

identification is “legal” judgment. We show the algorithm for admission control in Fig. 4.3. The new

streaming flow is always legal as long as its number is smaller than the guaranteed number of streaming

flows in the SLA of user j, and the new data flow is legal when the aggregated rate of data flows of user

j is smaller than the one in the SLA of user j.

After the new flow’s status is identified, admission control then takes the status as the input.

Admission control accepts a legal flow and registers it in the flow table in all cases, while it rejects an

illegal flow if traffic of PON segment i is heavily loaded (Congestion (i) = true), as shown in Fig. 4.4.

73

Fig. 4.3. “Legal” judgment of a new flow.

Fig. 4.4. Admission control.

if (Thj ij Ureq >∑ )

Congestion (i) = true;

else Congestion (i) = false;

if (ijks )

if (k > NS(SLA(j)))

)( Ijksl = false;

else )( Ijkjsl = true;

if (Ijke )

if ( ))(()( jSLABWer BEIjk

k

inst >∑ )

)( Ijkel = false;

else )( Ijkel = true;

74

4.3.2.2 Cycle Time Scheduling

The cycle time scheduling is triggered during the execution of dynamic wavelength allocation (in

Section 4.2) to assign bandwidth to each user j in PON segment i across the allocated wavelengths

during each scheduling cycle, as shown in the following steps:

1. For all legal streams, allocate bandwidth as much as their requests: req(sijk);

2. Distribute bandwidth to the legal BE flows fairly in PON segment i;

3. If there is remaining bandwidth, distribute it to the illegal streams and BE flows fairly in PON

segment i; and

4. Calculate instantaneous rate of each flow and update them in the flow table.

4.3.2.3 Termination Control

When an existing flow is over, the termination control is triggered. It updates the flow table by deleting

the flow’s entry, and evaluates again the “legal status” of each illegal flow according to Fig. 4.3.

4.4 Analysis

An important challenge in the proposed algorithm is how to determine the threshold values (i.e., UTh and

LTh) to add the new wavelength or release the added wavelength. This approach is important because it

adjusts the size of the queues in the OLT by adding to or releasing extra bandwidth, which in turn

determines the delay of buffered packets. If we can control the maximal queue size of the streaming

flows by using a threshold, we can control the maximum packet delay of the streaming flows.

The analysis is based on the worst-case consideration, where the allocated wavelengths for PON

segment i are heavily loaded and downstream packets are heavily accumulated in the queues in the CO,

75

before the new wavelength is allocated in the next scheduling cycle. Since the legal streaming flows take

the highest priority in bandwidth allocation, the buffered legal streaming packets in the current cycle

will take all the network resources (even entire allocated wavelengths, including newly allocated ones in

the next scheduling cycle) to transmit. Based on the analysis, we assume the worst-case scenario where

each legal streaming flow reaches its peak rate right after the scheduling cycle. As a result, the

accumulated legal streaming flows during the current scheduling cycle must be transmitted within the

maximum packet-delay time period by allocating additional wavelengths. Assuming Max

ijkf to be the

peak rate of legal streaming flow k of user j in PON segment i, which is also known as the flow link rate,

C is the transmission rate of one wavelength (e.g., 1 Gbps), TCycle is the scheduling cycle, Rcurr_i is the

current wavelength used by PON segment i, and MaxD is the maximal packet delay, the up threshold

value (UTh) must satisfy the following inequality:

( )1)( _

,

_ +××≤+×−∑ icuurMaxTh

kj

Cycleicurr

Max

ijk R C D U T CRf

(4.1)

ThTh U L ×= α (4.2)

The above equation assumes that the congestion will be mitigated by adding an additional

wavelength. Figure 4.5 shows how the up threshold adjusts the dedicated queue size. For example,

assuming 40 input streams with link rate at 50 Mbps each, TCycle = 5 ms, DMax = 5 ms, and Rcurr_i = 1, the

UTh should be less than 5 Mbits. )1,0(∈α is a network operator-specified parameter to adjust the

frequency of adding and releasing additional wavelengths. Generally, when α approaches 1, the

frequency will increase, and vice verse.

76

Fig. 4.5. Illustration of up threshold.

4.5 Illustrative Numerical Examples

We evaluate a LR-PON with illustrative values shown in Table 4.1. We consider Ethernet frames with

packet-size distribution between 64 and 1518 bytes (including Ethernet headers). Each user requests for

a service flow whose inter-arrival time and duration are exponentially distributed. We generate these

bursty flows by aggregating multiple sub-streams (the stream here is different from the streaming

defined in this work), each consisting of alternating Pareto-distributed on/off periods, with a Hurst

Parameter of 0.8 [65]. For each ONU, the basic user SLA defines one streaming flow with 20 Mbps

average bandwidth and 5 ms maximum packet delay, and 10 Mbps for the aggregated average rate of BE

traffic. We set the link rate of each streaming flow as 40 Mbps. If each user requests all the services in its

SLA, the saturated traffic load is 96.01000/32)20110( =××+ normalized to the channel

capacity (1 Gbps). In order to upgrade the LR-PON with premium services, we provide a user with a

premium SLA as two streaming flows, and 20 Mbps for BE traffic. Correspondingly, the normalized

saturated traffic load is 1.92. By proportionally changing the arrival rate of streaming and BE flows, we

can generate traffic at different normalized loads, e.g., 0.8 to 2.0 (load more than 1.92 can be achieved

77

by adding some illegal traffic flows), which corresponds to the average bandwidth request per user from

25 to 125 Mbps.

Figure 4.6 shows the average packet queuing delay vs. per-user bandwidth request. The number of

tunable lasers available at the CO is varied from 1 to 10, and this is compared to the fixed allocation

scheme, in which each user group has 2 or 3 dedicated wavelengths allocated to it (shown as 4 fixed and

8 fixed in Fig. 4.6). In Fig. 4.6, we limit showing the average delay to cases where the delay was below

0.6 ms. At the same traffic load, the LR-PON with our proposed dynamic wavelength allocation scheme

has a lower average delay with lesser number of used wavelengths, compared to the fixed allocation

scheme. For example, fixed allocation of two dedicated wavelengths per user group has average delay of

0.433 ms for user bandwidth request of 62.5 Mbps, while the proposed scheme using one tunable laser at

the OLT has average delay of 0.397 ms. As the number of available tunable lasers is increased, higher

user bandwidth requests may be supported while limiting the queuing delay to less than 0.6 ms. Because

the traffic is very bursty, extra wavelengths are required even at lower load to limit the average delay.

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0.55

0.6

25 50 75 100 125

Average User Bandwidth Request (Mbps)

Aver

age

Pac

ket

Del

ay (

ms)

1

2

3

4

5

6

7

8

9

10

4 (Fixed)

8 (Fixed)

Fig. 4.6. Average packet delay for different wavelengths.

78

0

2

4

6

8

10

12

25 50 75 100 125

Average User Bandwidth Request (Mbps)

Nu

mber

of

extr

a W

avel

eng

ths

# Wavelengths (Tunable) # Wavelengths (Dedicated)

Fig. 4.7. Wavelengths required for network upgrade.

0

5

10

15

20

25

30

0 0.8 1.6 2.4 3.2 4 4.8 5.6 6.4 7.2 8 8.8 9.6 10.4 11.2 12 12.8

Time (sec)

Dela

y (

mse

c)

Streaming flow

BE flow

Uniform

Fig.4.8. Instantaneous delay.

79

Figure 4.7 shows how wavelengths may be incrementally added to support increased user

bandwidth requests. We compare the number of wavelengths required by the proposed scheme with the

fixed allocation scheme. The proposed scheme requires less number of tunable wavelengths to be added

to meet the user bandwidth request. Moreover, the granularity of the number of tunable wavelengths

added is always small, namely one tunable wavelength for each upgrade, compared to 4 wavelengths in

fixed allocation. This figure also shows when the service provider may add extra tunable lasers to meet

the increased per-user average bandwidth request.

Figure 4.8 shows the flow-scheduling algorithm’s effect on controlling the maximum packet delay.

DMax is set to 5 msec and normalized load is 0.85. The simulation shows the streaming flows with delay

less than 5 msec, at the cost of increased BE flow delay, which is still tolerable. For comparison, the

other strategy (uniform) does not classify the flows, so it does not discriminate the delay between the BE

and streaming flows. As a result, the uniform traffic has packet delay greater than the streaming flow,

e.g., its delay is more than 5 ms at time 4.9 s.

To show the proposed algorithm’s effect on meeting the user-specified SLA, we investigate the

heavily-loaded case where each user requests all the services in the premium SLA. As we discussed

before, the normalized load is 1.92. Under this heavy load, we show the performance of different flows

(legal and illegal, both streaming and BE). Note that, in our setting, we assume the streaming flow has a

relatively longer life with 46.7 / 20 = 2.335 s on the average. The system will stabilize after 2.335 s after

initiating all the traffic flows. To examine the performance of illegal streaming flows, we randomly pick

a user and manually insert one more streaming flow (average 20 Mbps, ever lasting) besides the user’s

SLA at staring time. Figure 4.9 shows the instantaneous packet delay of the legal stream, the illegal

stream, and the BE flow of the user; and Fig. 4.10 shows instantaneous bandwidth of the legal stream,

BE flow of the user, and aggregated traffic of the 32 users in the same PON segment. Note that the

volume of aggregated traffic shown in the graph is 1/32 of actual traffic for better visual effect. The

aggregated traffic in Fig. 4.10 shows two bursty time intervals: 3.1-3.4 seconds and 3.9-4.3 seconds, in

which the BE traffic and illegal streaming flows suffer longer delay and decreased bandwidth, as shown

80

0

5

10

15

20

25

2.77 2.85 2.93 3.01 3.09 3.17 3.25 3.33 3.41 3.49 3.57 3.65 3.73 3.81 3.89 3.97 4.05 4.13 4.21 4.29

Time (second)

De

lay (

mse

c)

legal stream 1legal stream 2illegal streamBE flow

Fig.4.9. Flow instantaneous delay of a user.

0

20

40

60

80

100

120

2.77 2.87 2.97 3.07 3.17 3.27 3.37 3.47 3.57 3.67 3.77 3.87 3.97 4.07 4.17 4.27 4.37 4.47

time (sec)

insta

nt ra

te (

Mbp

s)

aggregated traf f ic/32

BE flow

streaming f low

Fig. 4.10. Flow instantaneous bandwidth of a user and background traffic.

in Fig. 4.9. But the average bandwidth of BE traffic for a long duration is 19 Mbps, which is very close

to 20 Mbps as indicated in the user SLA, as shown in Fig. 4.10. Even during bursty periods, the legal

81

streaming flows still get guaranteed delay (< 5 msec) (Fig. 4.9) and can even maintain peak rate (40

Mbps) at the first bursty period with a long-run average bandwidth of 20 Mbps (Fig. 4.10).

4.6 Conclusion

We proposed a new and efficient protocol to achieve better utilization of tunable lasers across

different user groups in a LR-PON. The protocol integrates the user-SLA-aware bandwidth allocation

algorithm to accommodate downstream bursty traffic and provide Quality of Service (QoS) in the user

SLA. Illustrative numerical results demonstrate the protocol’s advantage to support incremental

upgrade of bandwidth with increasing user bandwidth requests, and provide a user with a SLA which

guarantees a number of streaming flows with average bandwidth and maximum delay guarantee (i.e., 5

ms), as well as data flows with average bandwidth specifications.

82

Chapter 5

Protection in Long-Reach Broadband

Access Networks

5.1 Introduction

In recent years, an increasing number of advanced components have been deployed for the optical

access network, e.g., optical amplifiers which extend the optical access network’s reach from 20 km to

100 km and beyond, and such networks are referred to as SuperPON or Long-Reach PON (LR-PON) in

the literature [30-35, 66, 67]. Advances in wavelength-division multiplexing (WDM) technology

enable LR-PON to multiplex more wavelengths on a fiber, with each wavelength operating (soon) at a

transmission rate of 10-40 Gbps, or higher. Hence, compared with a traditional PON whose span is

10~20 km, LR-PON can serve many more users in a much larger geographical area. The multiple

Optical Line Terminals (OLTs) in a traditional PON deployment can be consolidated at a single, more

powerful OLT in a LR-PON, which can save huge Capital Expenditure (CapEx) and Operational

Expenditure (OpEx) for the network operator.

83

Figure 3.1 shows an example of a LR-PON which can be potentially deployed in the Greater

Sacramento region, California. The ring topology is suitable for network resilience, and it provides

two-dimensional coverage (vs. the linear coverage of a traditional PON). At each node of the ring, an

Optical Add-Drop Multiplexer (OADM) is used to “add and drop” a wavelength to a set of end users and

compensate for the signal power loss along the long-range transmission. Depending on users’

bandwidth requests, multiple wavelengths can be “dropped and added” through multiple ODAMs, e.g.,

City of Davis in Fig. 3.1. End users remodulate the “dropped” wavelength with their upstream signals

and “add” it onto the ring. The transmission direction of downstream and upstream wavelengths on the

ring can be the same or opposite, depending on the structure of the OADM and the OLT, and the overall

network architecture. Note that the term LR-PON is not fully accurate since not all of its components

between the OLT and ONUs are passive; but this term has been used quite widely in the literature to

show the LR-PON’s derivation from the traditional PON with the use of limited active elements.

As LR-PON exploits the huge transmission capacity of optical communications technology, and is

oriented for long-range coverage to serve a large number of end users, any network failure may cause a

significant loss of data (and revenue) for the customers and the network operator. So, LR-PON

protection becomes necessary and important. Various protection schemes for PONs have been proposed

by ITU-T (e.g., G. 983 [68] and G. 984 [69]). But protection schemes on emerging topologies, such as

ring-and-spur topology need further investigation. We propose a hardware-accelerated protection

scheme for the ring-and-spur LR-PON, in order to locate the failure and establish the protection paths

promptly in the optical layer with hardware control.

The rest of the chapter is organized as follows. Section 5.2 describes our proposed

hardware-accelerated protection designs for the LR-PON exploiting the unidirectional transmissions,

including 1 + 1 and 1:1 protections. Section 5.3 describes the protection design for the LR-PON with

bidirectional transmissions. Section 5.4 shows the illustrative numerical examples, including the power

budget, cost vs. performance, and the protection switching time.

84

5.2 Hardware-Accelerated Unidirectional Protection

Generally, there are two types of survivability architectures: 1+1 architecture and 1:1 architecture. The

1+1 architecture uses two overlaid PONs. The traffic is bridged onto both a working PON and a

protection PON. Upon receiving a signal at the OLT, the traffic is selected based on signal quality. With

this approach, fast protection can be achieved. However, in this architecture, no extra traffic can be

supported. Compared with no-protection case, it requires double the bandwidth.

In the 1:1 architecture, normally, the traffic is transmitted over the working PON. Once a failure

occurs, the traffic is switched on to the protection PON. This approach is relatively slower than the 1+1

architecture. However, compared with the 1+1 architecture, it can either significantly reduce the spare

capacity requirement or carry extra low-priority traffic depending on the network design.

In this section, we proposed two hardware-accelerated protection designs: 1+1 unidirectional

protection and 1:1 unidirectional design. In the 1+1 unidirectional protection design, a backup fiber ring

is used as a backup resource; while in the 1:1 unidirectional protection design, the backup fiber ring is

used also to carry preemptable network traffic in normal circumstance.

5.2.1 1+1 Unidirectional Protection

OLT

1

2

4

3

Primary

Backup

Pi

Po

Bi

Boλj

Access Node

Up/down wavelength

Optical Amplifier

Pi

Po

Bo

Bi

OLT

1

2

4

3

Primary

Backup

Pi

Po

Bi

Boλj

Access Node

Up/down wavelength

Optical Amplifier

Pi

Po

Bo

Bi

(a) Normal operation.

85

OLT

1

2

4

3

Primary

Backup

Pi

Po

Bi

Boλj

Access Node

Down wavelength

Optical Amplifier

Pi

Po

Bo

Bi

Up wavelength

(b) Protection operation.

Fig. 5.1. Operation of a unidirectional protected LR-PON using the “ring-and-spur” architecture.

Figure 5.1 shows our proposed protection scheme for the ring-and-spur LR-PON (unidirectional

access ring). The LR-PON connects to the CO with an OLT, linking the access network to the backbone

network. As shown in Fig. 5.1(a), two fiber rings connect the OLT and access nodes which add and drop

wavelength(s) (note that the PON segments connected to the access nodes are not shown here because

our protection scheme only includes the OLT and access nodes). The primary ring carries network

traffic in normal operation, and the backup ring provides survivability in case of a failure. In this

illustration, in normal state, the primary ring transmits on downstream wavelengths from the OLT in one

direction, e.g., counter clockwise. A set of wavelengths are assigned to each access node. Downstream

traffic (wavelengths) destined for node i is dropped at node i (e.g., λi is dropped at access node i), and

then node i reuses these same wavelengths (after remodulation at the reflective ONU) to send its

upstream traffic along the primary ring back to the OLT. This is possible by using an Optical Add-Drop

Multiplexer (OADM) at the access node, where the OADM can be reconfigurable as well (i.e., a

ROADM) to support dynamism in network configurations and to accommodate traffic variations. In

normal operation, the backup ring transmits the downstream wavelengths from the OLT in the opposite

direction, e.g., clockwise, but this signal is ignored. Each access node bypasses the downstream

86

wavelengths on the backup ring and does not add upstream wavelengths onto the backup ring. (Note that

a complementary research problem which is not addressed here is the design of a corresponding and

efficient dynamic bandwidth allocation (DBA) algorithm which takes into account the respective

propagation delays between the OLT and ONUs and vice versa.)

Figure 5.1(b) shows the operation of the “ring-and-spur” LR-PON after a failure (e.g., fiber cut)

between nodes 2 and 3. Now, downstream traffic is sent to nodes 1 and 2 via the primary ring and to

nodes 3 and 4 via the backup ring. Upstream traffic is sent from nodes 1 and 2 to the OLT on the backup

ring, while nodes 3 and 4 use the primary ring to send data to the OLT. Below, we outline the hardware

support needed to quickly detect failures and to reroute traffic as necessary.

OADMPi Po

Bo

DET

DET

I

Biλj

DET Optical detector

II

Control Signal

λj

Filter

(a) Access node in normal operation.

87

OADMPi Po

Bo

DET

DET

I

Bi

DET Optical detector

II

Control Signal

λj λj

Filter

(b) Access nodes 1 and 2 in protection.

OADMPi Po

Bo

DET

DET

I

Bi

DET Optical detector

II

Control Signal

λj λj

Filter

(c) Access nodes 3 and 4 in protection.

Fig. 5.2. Operations of an access node (with 1+1 protection).

Figure 5.2(a) shows the structure of an access node in normal operation. It contains two 2x2 optical

switches, a 3-way optical switch, an OADM, an optical detector array, and an optical filter. The optical

detector detects a network failure by monitoring the signal strength on its attached port, and outputs 0

for normal state and 1 for failure. The output (0 or 1) controls the 2x2 optical switches and the 3-way

optical switch, as shown by dashed lines. In normal state, the detectors detect normal traffic signal on its

88

attached ports Pi and Bi. Its output 0 keeps the two 2x2 optical switches in “bar” state and sets the 3-way

optical switch. Thus, downstream wavelengths on the primary ring arrive at port Pi, “drop and add”

through an OADM, output at port Po, and transmit to the port Pi of the next access node. Downstream

wavelengths on backup ring arrive at port Bi and are bypassed to the next access node through port Bo.

When a failure occurs (see Fig. 5.1(b)), the access nodes detect signal loss at their input ports (Pi

for nodes 3 and 4, and Bi for nodes 1 and 2). At nodes 1 and 2, the detector on port Bi outputs 1 to set the

3-way switch, while detector on port Pi outputs 0 to keep the two 2x2 optical switches in “bar” state, as

shown in Fig. 5.2(b). Thus, downstream wavelengths on the primary ring are dropped and upstream

wavelengths are added onto the backup ring. At nodes 3 and 4, detector on port Pi outputs 1 to set the

two 2x2 optical switches in “cross” state and also sets the 3-way switch (Fig. 5.2(c)). Thus, downstream

wavelengths on the backup ring are dropped and upstream wavelengths are added onto the primary ring.

As described above, the protection paths can be established under hardware control right after the

failure is detected. It is also necessary to maintain the paths once they are established, because the valid

optical signal will reappear at the nodes not neighboring to the failure location, which could trigger a

new control signal to change the status of the 2x2 switch and the 3-way switch. For example, protected

upstream wavelengths from access node 3 arrive at port Pi of node 4, and the detector would output 0 to

set the two 2x2 switches back to “bar” state. To avoid this situation, two optical filters are placed before

the detector array. The filters are designed to bypass the wavelengths dropped and added locally. Thus,

the protected wavelengths from the other nodes cannot pass through the filters to interrupt the

established protection paths.

After the failure is fixed, the detectors detect valid signals at both ports Pi and Bi. Thus, each node

is reconfigured back to the status of normal operation, as shown in Fig. 5.2(a).

Figure 5.3 shows how the OLT can automatically locate a failure and switch to the valid upstream

wavelengths on the primary or backup rings under hardware control. Assuming that node i is served by

wavelength λi (totally N nodes) and a failure occurs between nodes M-1 and M, the detectors array of the

primary ring will detect valid upstream wavelengths λM-λN, while detectors of the backup ring will

89

detect valid upstream wavelengths λ1-λM-1. Based on this information, the control unit locates the failure,

and then configures the corresponding 3-way switches to choose the valid upstream wavelengths for the

OLT receivers.

Primary

Upstream

Backup Upstream

AWG

1××××N

λ1λ2…λNDET P1

DET P2

DET PN

AWG

1××××N

λ1λ2…λNDET B1

DET B2

DET BN

Control

Fig. 5.3. OLT locates the failure.

5.2.2 1:1 Unidirectional Protection

The 1:1 unidirectional protection design also protects the traffic in the ring-and-spur LR-PON, but

makes use of the bandwidth on the backup fiber ring in the normal situation. In case of a failure, it

switches the traffic on the primary ring to the backup ring. In this way, this design can significantly

reduce the spare capacity requirement.

90

OADMPi Po

Bo

DET

DET

I

Biλj

DET Optical detector

II

Control Signal

λj

OADM

IIIIII

Filter

(a) Access node in normal operation.

OADMPi Po

Bo

DET

DET

I

Biλj

II

λj

OADM

III

DET Optical detector Control Signal Filter

(b) Access node in protection.

Fig. 5.4. Operation of an access node (with 1:1 protection).

The ring-and-spur LR-PON has the same architecture as shown in Fig. 5.1(a), with the addition of

carrying upstream and downstream traffic on the backup ring in normal circumstances. In order to

91

support this function, we propose the corresponding access node design, as shown in Fig. 5.4(a). In this

design, an additional OADM is connected to the backup ring in order to add and drop wavelengths. In

order to utilize the bandwidth resources effectively, the wavelengths used in the primary ring are reused

in the backup ring to carry the PON traffic in the opposite directions. In an access node, the same

downstream wavelengths are dropped from the primary and backup ring (e.g., λj), and split to end user

ONUs respectively. The user ONU remodulates the received downstream wavelength with its upstream

signal and transmits it to the access node (e.g., the two upstream λj in Fig. 5.4(a)). The upstream

wavelengths are connected to a 2x2 switch (e.g., switch III in Fig. 5.4(a)). The two output ports of the

switch are connected with the two OADMs.

In normal circumstances, the primary ring carries the high-priority traffic, and it drops and adds

traffic at each access nodes. The backup ring carries the low priority-traffic, and it drops and adds traffic

at each access nodes. Wavelengths on the primary and backup rings works independently and do not

interfere with each other.

In case of a failure, protection switching will be triggered in each access node by detecting the loss

of signal. Assuming that a fiber cut occurs between access nodes 2 and 3 (as shown in Fig. 5.1 (b)), the

optical detector attached to port Bi of access nodes 1 and 2 will detect the signal loss and move the 2x2

switch to the cross state. In this way, downstream wavelengths are dropped from the primary fiber and

are added onto the backup fiber. Note that the downstream wavelengths will not be dropped from the

backup fiber because of the signal loss. In access nodes 3 and 4, the optical detector attached to the port

Pi will detect the signal loss and set the 2x2 switch to the cross state, and downstream wavelengths are

dropped from the backup fiber and are added onto the primary fiber. The optical filter before the

detector only allows the corresponding up and down wavelength (e.g., λj) to pass through, in order to

avoid the case that the protected upstream wavelength sets the 2x2 switch back to the bar state, as

discussed in the 1+1 unidirectional protection case.

In this design, the OLT can also automatically locate a failure and switch to the valid upstream

wavelengths on the primary or backup rings under hardware control, using the design shown in Fig. 5.3.

92

After the OLT locates the failure, it will drop the low-priority traffic on the backup ring and duplicate

the high-priority traffic (originally on the primary ring) on both primary and backup rings by having the

transmitters of the backup ring to transmit high-priority traffic.

After the failure is fixed, the detectors detect valid signals at both ports Pi and Bi. Thus, each node

is reconfigured back to the status of normal operation, as shown in Fig. 5.4(a). The OLT also detects

valid signal on all wavelengths and configures itself back to normal state to transmit high-priority traffic

on the primary ring and low-priority traffic on the backup ring.

5.3 Hardware-Accelerated Bidirectional Protection

In this section, we assume that the LR-PON can be managed more intelligently. The communication

between the OLT and the access nodes can choose its direction (clockwise or counter-clockwise) to

achieve high performance, such as decreased transmission hops. Figure 5.5(a) shows the normal

operation of a bidirectional LR-PON, where access nodes 1, 2, and 3 receive downstream wavelengths

from the primary ring in counter-clockwise direction and add upstream wavelengths onto the backup

ring in clockwise direction, and access nodes 4, 5, and 6 receive downstream wavelengths from the

primary ring in clockwise direction and add upstream wavelengths onto the backup ring in

counter-clockwise direction.

93

OLT1

3

6

4

Primary

Backup

Po

Pi

Bo

Bi

Access Node

52Pi

Po

Bi

Bo

λj

(a) Normal operation.

OLT

1

3

6

4

Primary

Backup

Po

Pi

Bo

Bi

Access Node

52Pi

Po

Bi

Bo

λj

(b) Protection operation.

Fig. 5.5. Operation of a bidirectional protected LR-PON using the “ring-and-spur” architecture.

94

Bidirectional couplerDET Optical detector Control Signal

Unidirectional couplerUnidirectional coupler

OADMPiPo

BiBo

DET

λjλj

Optical filter

(a) Access node in normal operation.

OADMPiPo

BiBo

DET

λjλj

Bidirectional couplerDET Optical detector Control Signal

Unidirectional couplerUnidirectional couplerOptical filter

(b) Access nodes 2 & 3 in protection.

Fig. 5.6. Operation of an access node (bidirectional protection).

At each access node, an optical detector detects normal optical signal at input port Pi, and it outputs

control signal 0 to set the three 3-way switch, as shown in Fig. 5.6(a). As a result, the downstream

wavelengths enter each access node at port Pi, transmit through the OADM, and leave the access node at

95

port Po; the upstream wavelengths enter each access node at port Bi, where new upstream wavelength is

added, and leave the access node at port Bo.

When a failure occurs, e.g., a fiber cut between access nodes 1 and 2 as shown in Fig. 5.5(b),

signal loss at port Pi of access nodes 2 and 3 is detected, which sets their three 3-way switches, as shown

in Fig. 5.6(b). As a result, downstream wavelengths are fed to the corresponding OADM through ports

Po of access nodes 2 and 3; and upstream wavelength is added and transmitted to port Bi. Access nodes

1, 4, 5, and 6 maintain the normal state, because their optical detector detects normal signal at Pi.

The optical filter before the detector only allows the corresponding up and down wavelength (e.g.,

λj) to pass through, in order to avoid the case that the protected traffic passes through the port Pi and

configures the access node back to the normal state, similar to the discussion for 1+1 unidirectional

protection.

5.4 Illustrative Numerical Examples

In this section, we present illustrative numerical examples on the power budget, cost, spare capacity

usage, and protection-switching time of our hardware-accelerated LR-PON protection designs, focusing

on the 1+1 and 1:1 unidirectional protection schemes and their comparison.

5.4.1 Power Budget

The typical power losses of the optical components in our designs are listed below.

– Add and drop loss of OADM: 6.5 dB for 64 WDM channels

– Multiplex and demultiplex loss in OADM: 4 dB

– 3-way switch loss: 1 dB

– 2x2 optical switch loss : 1 dB

– Propagation loss: 0.2 dB/km

96

For 1 + 1 unidirectional protection, the downstream wavelengths passing through an access node

will encounter 6 dB power loss (multiplex and demultiplex loss + two 2x2 optical switch loss); and the

dropped downstream wavelength at the targeted access node will suffer 11.4 dB power loss (add and

drop loss + multiplex and demultiplex loss + one 2x2 optical switch loss).

For 1:1 unidirectional protection, the downstream wavelengths passing through an access node will

encounter 4 dB power loss (multiplex and demultiplex loss); and the dropped downstream wavelength

at the targeted access node will suffer 10.4 dB power loss (add and drop loss + multiplex and

demultiplex loss).

Table 5.1. Power loss of upstream and downstream wavelengths.

Downstream Upstream

Pass through Dropped Pass through Added

1 + 1 6 dB 11.5 dB 6 dB 12.5 dB

1:1 4 dB 10.5 dB 4 dB 11.5 dB

Table 5.1 shows the power losses for upstream and downstream wavelengths at the access node in

1 + 1 and 1:1 protection schemes.

Assuming N access nodes and a 100-km fiber ring, the downstream wavelength which suffers the

highest power loss is the one which is dropped last. The highest power loss for downstream is:

6(N-1) + 11.5 + 0.2 x 100 = 6N + 25.5 dB for 1 + 1 protection, and

4(N-1) + 10.5 + 0.2 x 100 = 4N + 26.5 dB for 1:1 protection.

The highest power loss for upstream is

6(N-1) + 12.5 + 0.2 x 100 = 6N + 26.5 dB for 1 + 1 protection, and

4(N-1) + 11.5 + 0.2 x 100 = 4N + 27.5 dB for 1:1 protection.

97

Since a power loss of over 25 dB requires using an optical amplifier, we can calculate the number

of amplifiers needed. For example, two amplifiers are needed if there are four access nodes along the

fiber ring in both protection schemes. The 1:1 protection saves (2N – 1) dB at each access nodes. As a

result, for every 13 nodes, the 1:1 protection saves an optical amplifier compared with 1 + 1 protection.

When more access nodes are deployed, the 1:1 protection needs fewer amplifiers.

5.4.2 Cost and Network Capacity

0

0.2

0.4

0.6

0.8

1

1.2

Cost Capacity/Cost

1+1 protection 1:1 protection

Fig. 5.7. Normalized access node cost and normalized ratio of access node capacity vs. cost.

Figure 5.7 shows the cost comparison of the access node in 1 + 1 and 1:1 scenarios. The cost of a node in

1:1 protection is higher because it incorporates another OADM, but the cost does not double since the

node in 1:1 protection saves one 2 X 2 switch and one 3-way switch. The node in 1:1 protection provides

two times the capacity than the node in 1 + 1 protection in normal operation by utilizing the backup fiber

ring. Figure 5.7 also shows the relationship between capacity and cost. As a result, 1:1 protection

achieves a better performance.

5.4.3 Protection-Switching Time

We consider a ring-and-spur LR-PON as shown in Fig. 5.1(a). In our investigation, four access nodes

are connected to the OLT through the primary and backup fiber rings. Each access node adds and drops

98

one wavelength. The distance between the OLT and it neighboring access nodes and the distance

between adjacent access nodes are 20 km each. Two different scenarios (1 + 1 unidirectional protection

and 1:1 unidirectional protection) are simulated, using the VPI simulator.

We simulate a fiber cut between access node 4 and the OLT at time between 27 ms and 28 ms.

Figure 5.8 shows the received signal power from each of the four wavelengths at the OLT (y-axis, in mw)

vs. time (x-axis, in ms) for the 1 + 1 unidirectional protection. The numerical result shows that, when a

failure occurs the protection paths can be established and the received signal can be valid again

(between 41 ms to 42 ms) i.e., within 14 ms after the failure occurs. The delay time of the mechanical

optical switches takes a main part of about 10 ms. The propagation delay for 100-km fiber is around 0.5

ms (light propagates at 2 x 108 m/s in the fiber). The time to detect the signal loss and the time to

establish the control signal are below 1 ms, according to the hardware settings.

00.00050.0010.00150.0020.00250.0030.003524 26 28 30 32 34 36 38 40 42 44 46Time (ms)OLT_Rx (mw) 00.00050.0010.00150.0020.00250.0030.0035

24 26 28 30 32 34 36 38 40 42 44 46Time (ms)OLT_Rx (mw)

(a) (b)

00.00050.0010.00150.0020.00250.0030.003524 26 28 30 32 34 36 38 40 42 44 46Time (ms)OLT_Rx (mw) 00.00050.0010.00150.0020.00250.0030.0035

24 26 28 30 32 34 36 38 40 42 44 46Time (ms)OLT_Rx (mw)

(c) (d)

Fig. 5.8. Received signal power of each wavelength at OLT in 1 + 1 protection.

99

Figure 5.9 shows the received signal power from each of the four wavelengths at the OLT (y-axis,

in mw) vs. time (x-axis, in ms) for 1:1 unidirectional protection. The numerical result shows that the

protection paths can be established (between 47 ms to 48 ms) i.e., within 20 ms after the failure occurs.

The 1:1 protection switching takes a longer time (6 ms) than the 1 + 1 scenario, because the OLT needs

the extra time to transit the network traffic from the primary ring to the backup ring.

00.00050.0010.00150.0020.00250.0030.003526 28 30 32 34 36 38 40 42 44 46 48Time (ms)OLT_Rx1 (mv) 00.00050.0010.00150.0020.00250.0030.0035

26 28 30 32 34 36 38 40 42 44 46 48Time (ms)OLT_Rx2 (mv)

(a) (b)

00.00050.0010.00150.0020.00250.0030.003526 28 30 32 34 36 38 40 42 44 46 48Time (ms)OLT_Rx3 (mv)

00.00050.0010.00150.0020.00250.0030.003526 28 30 32 34 36 38 40 42 44 46 48Time (ms)OLT_Rx4 (mv)

(c) (d)

Fig. 5.9. Received signal power of each wavelength at OLT in 1:1 protection.

5.5 Conclusion

We proposed several hardware-accelerated protection schemes for the ring-and-spur LR-PON. These

schemes can locate the failure and establish the protection paths promptly in the optical layer under

100

hardware control. The 1:1 unidirectional protection scheme doubles the network capacity by utilizing

the backup resource for transmitting lower-priority traffic, and achieves better capacity vs. cost ratio

compared with 1 + 1 unidirectional protection, while the 1 + 1 protection achieves a more prompt

protection switching (14 ms) compared with 1:1 protection (20 ms), where these numbers are based on

our chosen example, because of simplified control and traffic transition in the OLT.

101

Chapter 6

Conclusion

Long-Reach Passive Optical Network (LR-PON) exploits the huge transmission capacity of optical

technology, and is oriented toward long-range transmission and a large user base. LR-PON is anchored

at a Central Office (CO), so that all higher-layer networking functions can now be located further

upstream in the “network cloud". The OLTs of the traditional PON (which used to sit approx. 10-20 km

from the end user) can now be replaced at the local exchange by some elementary hardware, which

contains a small amount of compact low-power physical-layer repeater equipment, such as optical

amplifiers and Optical Add-Drop Multiplexer (OADM). As a result, the telecom network hierarchy can

be simplified with the access headend closer to the backbone network. Thus, the network's Capital

Expenditure (CapEx) and Operational Expenditure (OpEx) can be significantly reduced.

6.1 A Survey of Research in LR-PON

In this part, we discussed the research challenges related to LR-PON, from the physical layer, e.g.,

power attenuation, to the upper layer, e.g., bandwidth assignment. We reviewed some existing

LR-PON demonstrations: PLANET SuperPON initiated the research on LR-PON and extended the

network reach based on the APON model; BT’s demonstrations increased the network capacity up to 10

102

Gbps per channel and incorporated WDM based on a GPON model; and PIEMAN further exploited the

huge optical transmission capacity to support up to 16,384 users. WE-PON offered a novel

ring-and-spur topology to provide a better two-dimensional geographical coverage and protection for

PON traffic by using a ring. Several other demonstrations were also discussed. The impact of increased

RTT on higher-layer control was also discussed. Dynamic bandwidth allocation (DBA) algorithms, e.g.,

multi-thread polling and two-state DBA, have been proposed to remedy the impact of long propagation

delay by utilizing the idle time between transmission cycles.

A number of new technologies and designs are developing the LR-PON towards a higher-capacity

system – from a single wavelength to WDM, from less than Gbps transmission speed, e.g., 311 Mbps in

PLANET to 10 Gbps symmetric transmission, e.g., PIEMAN. GPON and EPON protocols will be

naturally inherited and be adjusted for the LR-PON, since they are mature standards in PONs today.

LR-GPON may be preferred by US and some European countries because many of their national

carriers choose GPON as first-mile solution, e.g., BT and Verizon. LR-EPON may be a choice of the

Asian countries since EPON is preferred there. Introducing LR-PON into practical use will be a gradual

process. As a first step, network operators need the LR-PON to cover the “green field” without

establishing a new OLT. Then, with the consolidation of OLTs and the merging of access and metro, we

are expecting ubiquitous LR-PONs.

6.2 Multi-Thread Polling in LR-PON

In this research, we addressed the problem of dynamic bandwidth allocation in a Long-Reach PON

(LR-PON). We proposed a multi-thread polling algorithm to remedy the effect of the long CO-to-Users

“control loop”. Moreover, we achieved insights on the algorithm by analyzing its key parameters, such

as initiating and tuning multiple threads, inter-thread scheduling, and fairness among users. Numerical

results show that, by setting the proper initial thread interval and tuning threshold, the average upstream

packet delay is decreased, especially at high traffic load; and the network throughput is increased.

103

6.3 A SLA-Aware Protocol in LR-PON

In this study, we proposed a new and efficient protocol to achieve better utilization of tunable lasers

across different user groups in a LR-PON. The protocol integrates the user-SLA-aware bandwidth

allocation algorithm to accommodate downstream bursty traffic and provide Quality of Service (QoS) in

the user SLA. Illustrative numerical results demonstrated the protocol’s advantage to support

incremental upgrade of bandwidth with increasing user bandwidth requests, and to provide a user with a

SLA which guarantees a number of streaming flows with average bandwidth and maximum delay

guarantee (e.g., 5 ms), as well as data flows with average bandwidth specifications.

6.4 Protection in LR-PON

In this research, we proposed several hardware-accelerated protection schemes for the ring-and-spur

LR-PON. These schemes can locate the failure and establish the protection paths promptly in the optical

layer under hardware control. The 1:1 unidirectional protection scheme doubles the network capacity by

utilizing the backup resource for transmitting lower-priority traffic, and achieves better capacity vs. cost

ratio compared with 1 + 1 unidirectional protection, while the 1 + 1 protection achieves a more prompt

protection switching (14 ms) compared with 1:1 protection (20 ms), where these numbers are based on

our chosen example, because of simplified control and traffic transition in the OLT.

Bibliography

104

Bibliography

[1] B. Mukherjee, Optical WDM Networks, Springer, Feb. 2006.

[2] G. Kramer, Etherenet Passive Opitcal Networks, McGraw-Hill Professional, 2005.

[3] B. Ling, “PON architecture futureproofs FTTH,” Journal of Lightware Tech., vol. 16, no. 10, pp.

104-107, 1999.

[4] D. B. Payne and R. P. Davey, “The Future of Fiber Access Systems,” BT Tech. Journal, vol. 20, pp.

104-14, 2002.

[5] D. W. Faulkner et al., “Optical Networks for Local Loop Applications,” Journal of Lightwave Tech.,

vol. 7, no. 11, pp. 1741-1751, 1989.

[6] G. Kramer, B. Mukherjee, S. Dixit, Y. Ye, and R. Hirth, “Supporting differentiated class of service

in Ethernet passive optical networks,” OSA Journal of Optical Networking, vol. 1, no. 8, pp.

280-298, 2002.

[7] D. J. Xu, W. Yen, and E. Ho, “Proposal of a New Protection Mechanism for ATM PON interface,”

IEEE International Conference on Communications (ICC 2001), vol. 7, pp. 2160-2165, June 2001.

[8] D. K. Hunter, Z. Lu, and T. H. Gilfedder, “Protection of long-reach PON traffic through router

database synchronization,” Journal of Optical Networking, vol. 6, no. 5, pp. 535-549, 2007.

[9] J. A. Lazaro, J. Prat, P. Chanclou, G. M. T. Beleffi, A. Teixeira, I. Tomkos, R. Soila, and V.

Koratzinos, “Scalable Extended Reach PON,” Proc. IEEE/OSA Optical Fiber Communication

Conference (OFC’08), Mar. 2008.

[10] H. Song, B.-W. Kim, and B. Mukherjee, “Long-Reach Optical Access Networks: A Survey of

Research Challenges, Demonstrations, and Bandwidth Assignment Mechanisms,” IEEE

Communications Survey and Tutorials, accepted, 2009.

[ 11 ] H. Song, B.-W. Kim, and B. Mukherjee, “Multi-Thread Polling: A Dynamic Bandwidth

Distribution Scheme in Long-Reach PON,” IEEE Journal on Selected Areas in Communication,

vol. 27, no. 2, San Francisco, USA, Feb. 2009.

[ 12 ] H. Song, B.-W. Kim, and B. Mukherjee, “Multi-Thread Polling: A Dynamic Bandwidth

Distribution Scheme in Long-Reach PON,” Proc., IEEE Globecom, pp. 2450-2454, Nov. 2007.

Bibliography

105

[13] H. Song, B. Mukherjee, Y. Park and S. Yang, “Shared-wavelength WDM-PON access network for

Supporting Downstream Traffic with QoS,” Proc., IEEE/OSA Optical Fiber Communication

Conference (OFC 2006), Anaheim, USA, Mar. 2007.

[14] H. Song, A. Banerjee, and B. Mukherjee, “A Protocol for Efficient Tunable Laser Utilization to

Support Incremental Upgrade in a WDM-PON,” Proc., IEEE/OSA Optical Fiber Communication

Conference (OFC 2007), Anaheim, USA, Mar. 2007.

[15] H. Song, A. Banerjee, and B. Mukherjee, “Shared-Wavelength WDM-PON Access Network ---

Bursty Traffic Accommodation and User-defined SLA Support,” Proc., COIN’06, Korea, 2006.

[16] H. Song, B.-W. Kim, and D.-M. Seoul, “Hardware-Accelerated Protection in Long-Reach PON,”

Proc., IEEE/OSA Optical Fiber Communication Conference (OFC 2009), Mar. 2009.

[17] O. Gerstel, “Optical Networking: A Practical Perspective,” IEEE Hot Interconnects, Aug. 2000.

[18] NSF Workshop Report, “Residential Broadband Revisited: Research Challenges in Residential

Networks, Broadband Access, and Applications," Oct. 2003. http://cairo.cs.uiuc.edu/nsfbroadband/

[19] M. Wegleitner, “Maximizing the Impact of Optical Technology," Keynote Address, IEEE/OSA

Optical Fiber Communication Conference (OFC 2007), Anaheim, CA, Mar. 2007.

[ 20 ] Gartner Report, “One Gigabit or Bust Initiative, A Broadband Vision for California,”

http://www.cenic.org/GB/gartner/, 2003.

[21] 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]," OSA Journal of Optical Networking, vol. 4, no. 11, pp.

737-758, Nov. 2005.

[22] R. Lauder, “Technology and Economics for Coarse Wavelength Multiplexing Workshop," Proc.,

IEEE/OSA Optical Fiber Communication Conference (OFC'04), Los Angeles, CA, 2004.

[23] J. George, “Designing Passive Optical networks for cost effective triple play support," Proc., FTTH

Conference, 2004.

[24] G. Agrawal, Fiber-Optic Communication Systems, John Wiley and Sons, Inc., New York, 2002.

[25] M. H. Kang, “Development of Broadband Convergence Network and Services in Korea," Proc.,

IEEE/OSA Optical Fiber Communication Conference (OFC 2008), San Diego, CA, Feb. 2008

(Invited Talk).

[26] G. Keiser, Optical Fiber Communications, McGraw-Hill, 2000.

Bibliography

106

[27] K.-I. Suzuki, Y. Fukada, D. Nesset, and R. Davey, “Amplified gigabit PON systems [Invited],”

Journal of Optical Networking, vol. 6, no. 5, pp. 422-433, May 2007.

[28] D. P. Shea and J. E. Mitchell, “Long-Reach Optical Access Technologies,” IEEE Network, vol. 21,

no. 5, pp. 5-11, September, 2007.

[29] T. Nakanishi, K.-I. Suzuki, Y. Fukada, N. Yoshimoto, M. Nakamura, K. Kato, K. Nishimura, Y.

Ohtomo, and M. Tsubokawa, “High sensitivity APD burst-mode receiver for 10Gbit/s TDM-PON

system,” IEICE Electronics Express, vol. 4, no. 10, pp. 588-592, 2007.

[30] I. Van de Voorde, C. M. Martin, J. Vandewege, and X. Z. Qiu, “The superPON demonstrator: an

exploration of possible evolution paths for optical access networks,” IEEE Communication

Magazine, vol. 38, no. 2, pp. 74-82, Feb. 2000.

[31] D. J. G. Mestdagh and C. M. Martin, “The Super-PON concept and its technical challenges,”

Broadband Communications, pp. 333-345, Apr. 1996.

[32] C. Martin, “Realization of a SuperPON Demonstrator,” Proc., NOC, pp. 118-93, 1997.

[33] M. O. van Deventer, J. D. Angelopoulos, H. Binsma, A. J. Boot, P. Crahay, E. Jaunart, P. J. Peters,

A. J. Phillips, X. Z. Qiu, J. D. Senior, M. Valvo, J. Vandewege, P. Vetter, and I.

van de Voorde, “Architecture for 100 km 2048 split bidirectional SuperPONs from

ACTS-PLANET,” Proc., SPIE, vol. 2919, pp. 245-251, Nov. 1996.

[34] I. Van de Voorde, M. O. van Deventer, P. J. M. Peters, P. Crahay, E. Jaunart, A. J. Phillips, J. M.

Senior, X. Z. Qiu, J. Vandewege, J. J. M. Binsma, and P. J. Vetter, “Network topologies for

SuperPON,” Proc., IEEE/OSA Optical Fiber Communication Conference (OFC’97), 1997.

[35] M. O. van Deventer, Y. M. van Dam, P. J. M. Peters, F. Vermaerke, and A. J. Phillips, “Evolution

Phases to an Ultra-Broadband Access Network: Results from ACTS-PLANET,” IEEE

Communication Magazine, vol. 35, no. 12, pp. 72-77, Dec. 1997.

[36] D. P. Shea and J. E. Mitchell, “A 10 Gb/s 1024-Way Split 100-km Long Reach Optical Access

Network,” Journal of Lightwave Technology, vol. 25, no. 3, pp. 685-693, Mar. 2007.

[37] R. P. Davey, P. Healey, I. Hope, P. Watkinson, D. B. Payne, O. Marmur, J. Ruhmann, and Y.

Zuiderveld, “DWDM Reach Extension of a GPON to 135 km,” Proc., IEEE/OSA Optical Fiber

Communication Conference (OFC’05), Mar. 2005.

[38] G. Talli and P. D. Townsend, “Hybrid DWDM-TDM Long-Reach PON for Next-Generation

Optical Access,” Journal of Lightwave Technology, vol. 24, no. 7, pp. 2827-2834, July 2006.

Bibliography

107

[39] G. Talli and P. D. Townsend, “Feasibility demonstration of 100 km reach DWDM SuperPON with

upstream bitrates of 2.5 Gb/s and 10 Gb/s,” Proc., IEEE/OSA Optical Fiber Communication

Conference (OFC’05), Mar. 2005.

[40] G. Talli, C. W. Chow, P. Townsend, R. Davey, T. D. Ridder, X.-Z. Qiu, P. Ossieur, H.-G. Krimmel,

D. Smith, I. Lealman, A. Poustie, S. Randel, and H. Rohde, “Integrated Metro and Access Network:

PIEMAN,” Proc., 12th European Conf. Networks and Opt. Commun., Kista, Sweden, June 2007.

[41] “Wdm-E-PON (WE-PON),” Working Documents, ETRI, 2007.

[42] J. A. Lazaro, I. R. Martinez, V. Polo, C. Arellano, and J. Prat, “Hybrid Dual-fiber-Ring with

Single-fiber-Trees Dense Access Network Architecture using RSOA-ONU," Proc. IEEE/OSA

Optical Fiber Communication Conference (OFC’07), Mar. 2007.

[43] F.-T. An, K.-S. Kim, D. Gutierrez, S. Yam, E. Hu, K. Shrikhande, and L. Kazovsky, “SUCCESS: A

Next-Generation Hybrid WDM/TDM Optical Access Network Architecture,” Journal of

Lightwave Technology, vol. 22, no. 11, pp. 2557-2569, Nov. 2004.

[44] M. Maier and M. Herzog, "STARGATE: The Next Evolutionary Step toward Unleashing the

Potential of WDM EPONs," IEEE Communications Magazine, vol. 45, no. 5, pp. 50-56, May 2007.

[45] M. Rasztovits-Wiech, A. Stadler, S. Gianordoli, A. Siemens, and K. Kloppe, “10/2.5 Gbps

Demonstration in Extra-Large PON Prototype (invited)," Proc., ECOC, 2007.

[46] I. T. Monroy, F. ohman, K. Yvind, R. Kjaer, C. Peucheret, A. M. J. Koonen, and P. Jeppesen, “85

km Long Reach PON System Using a Reflective SOA-EA Modulator and Distributed Raman Fiber

Amplification,” Proc., Lasers and Electro-Optics Society (LEOS), 2006.

[47] R. Kjaer, I. T. Monroy, L. K. Oxenloewe, P. Jeppesen, and B. Palsdottir, “Bi-directional 120 km

Long-Reach PON Link Based on Distributed Raman Amplification," Proc., Photonics in

Switching, 2007.

[48] H. H. Lee, K. C. Reichmann, P. P. Iannone, X. Zhou, and B. Pálsdóttir, “A hybrid-amplified PON

with 75-nm downstream bandwidth, 60 km reach, 1:64 split, and multiple video services," Proc.

IEEE/OSA Optical Fiber Communication Conference (OFC 2007), Mar. 2007.

[49] P. P. Iannone and K. C. Reichmann, “Hybrid SOA-Raman Amplifiers for Fiber-to-the-Home and

Metro Networks,” Proc., IEEE/OSA Optical Fiber Communication Conference (OFC 2008), Feb.

2008.

Bibliography

108

[ 50 ] M.-F. Huang, J. Yu, J. Chen, G.-K. Chang, and C. Sien, “A cost-effective WDM-PON

Configuration Employing Innovative Bi-directional Amplification," Proc., IEEE/OSA Optical

Fiber Communication Conference (OFC’07), Mar. 2007.

[51] H. Iwamura, G. C. Gupta, M. Kashima, H. Tamai, R. Watanabe, T. Ushikubo, and T. Kamijoh,

“42dB Loss Budget Hybrid DWDM-CDM-PON without Optical Amplifier," Proc., IEEE/OSA

Optical Fiber Communication Conference (OFC’07), Mar. 2007.

[52] S.-M. Lee, S.-G. Mun, and C.-H. Lee, “Consolidation of a Metro Network into an Access Network

based on Long-reach DWDM-PON,” Proc., IEEE/OSA Optical Fiber Communication Conference

(OFC 2007), 2007.

[53] S.-M. Lee, S.-G. Mun, M.-H. Kim, and C.-H. Lee, “Demonstration of a Long-Reach DWDM-PON

for Consolidation of Metro and Access Networks,” Journal of Lightwave Technology, vol. 25, no.

1, pp. 271-277, Jan. 2007.

[ 54 ] D. P. Shea and J. E. Mitchell, “Experimental Upstream Demonstration of a Long-Reach

Wavelength-Converting PON with DWDM Backhaul," Proc., IEEE/OSA Optical Fiber

Communication Conference (OFC’07), Mar. 2007.

[55] H. Song, A. Banerjee, B. W. Kim, and B. Mukherjee, “Multi-Thread Polling: A Dynamic

Bandwidth Distribution Scheme in Long-Reach PON,” Proc., IEEE Globecom, pp. 2450-2454,

Nov. 2007.

[56] C.-H. Chang, N. M. Alvarez, P. Kourtessis, and J. M. Senior, “Dynamic Bandwidth assignment for

Multi-service access in long-reach GPON,” Proc., ECOC, 2007.

[57] P. P. Iannone and K. C. Reichmann, “Strategic and Tactical Uses for Extended PON,” Invited

Talk, IEEE/OSA Optical Fiber Communication Conference (OFC 2008), Feb. 2008.

[58] G. Kramer, B. Mukherjee, and G. Pesavento, “Ethernet PON (ePON): Design and Analysis of an

Optical Access Network,” Photonic Network Communications, vol. 3, no. 3, pp. 307-319, July

2001.

[59] G. Kramer, B. Mukherjee, and G. Pesavento, “IPACT: A Dynamic Protocol for an Ethernet PON

(EPON),” IEEE Communications Magazine, vol. 40, no. 2, pp. 74-80, February 2002.

[60] C. Assi, Y. Ye, S. Dixit, and M. Ali, “Dynamic Bandwidth Allocation for Quality-of-Service Over

Ethernet PONs,” IEEE Journal on Selected Areas in Communication, vol. 21, no. 9, pp. 1467-1477,

November 2003.

Bibliography

109

[61] A. Shami, X. Bai, C. Assi, and N. Ghani, “Jitter Performance in Ethernet Passive Optical

Networks,” Journal of Lightware Technology, vol. 23, no. 4, pp. 1745-1753, April 2005.

[ 62 ] D. Meyer, “BT Openreach unsure of fibre demand,” ZDNet, September 2007,

http://news.zdnet.co.uk/communications/0,1000000085,39289708,00.htm

[63] P. N. Tudor, “MPEG-2 Video Compression,” MPEG-2 - What it is and What it isn't, IEEE

Colloquium on, pp. 2/1-2/8, January 1995.

[64] T. Wiegand, G. J. Sullivan, G. Bjǿntegaard, and A. Luthra, “Overview of the H.264/AVC Video

Coding Standard,” IEEE Transaction on Circuits and Systems for Video Technology, vol. 13, no. 7,

pp. 560-576, July 2003.

[65] W. E. Leland, M. S. Taqqu, W. Willinger, and D. V. Wilson, “On the self-similar nature of Ethernet

traffic,” IEEE/ACM Transactions on Networking, vol. 2, no. 1, pp. 1-15, Jan. 1994.

[66] I. Tomkos, “Metropolitan area optical networks: Needs and requirements,” Invited paper, IEEE

Circuits and Devices Magazine, July 2003.

[67] S. Azodolmolky and I. Tomkos, “A Techno-Economic Model for FTTH Deployments,” Journal of

Telecommunications Management, vol. 1, no. 3, pp. 294-310, July 2008.

[68] ITU-T, Broadband optical access systems based on passive optical networks (PON), 2005,

http://www.itu.int/rec/T-REC-G.983.1/en.

[ 69 ] ITU-T, Gigabit-capable Passive Optical Networks (GPON): General characteristics, 2003,

http://www.itu.int/rec/T-REC-G.984.1/en.