University of Southern California Dissertations and Theses

Design and analysis of MAC protocols for broadband wired/wireless networks

(USC Thesis Other)

Design and analysis of MAC protocols for broadband wired/wireless networks

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content DESIGN AND ANALYSIS OF MAC PROTOCOLS
FOR. BROADBAND WIRED/WIRELESS NETWORKS
by
Wen-Kuang Kuo
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ELECTRICAL ENGINEERING)
August 2003
Copyright 2003 Wen-Kuang Kuo
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 3116734
INFORMATION TO USERS
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
®
UMI
UMI Microform 3116734
Copyright 2004 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UNIVERSITY OF SOUTHERN CALIFORNIA
THE GRADUATE SCHOOL
UNIVERSITY PARK
LOS ANGELES, CALIFORNIA 90089-1695
This dissertation, written by
W E N - K U A A J ^ KUO
under the direction o f h IS dissertation committee, and
approved by all its members, has been presented to and
accepted by the Director o f Graduate and Professional
Programs, in partial fulfillment o f the requirements fo r the
degree o f
DOCTOR OF PHILOSOPHY
w t
Director
Date A u gu st 1 2 . 2003
Dissertation Committee
Chair
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C o n ten ts
List of Tables iii
List of Figures iv
Abstract vii
1 Introduction 1
1.1 Significance of the R ese arch .................................................................... 1
1.2 Research Issues about DOCSIS and IEEE
802.11 MAC lay er...................................................................................... 5
1.2.1 Issues about DOCSIS MAC la y e r ............................................... 5
1.2.2 Issues about IEEE 802.11 ........................................................... 7
1.3 Contributions of the R esearch................................................................. 7
1.4 Outline of D issertation.............................................................................. 11
2 DOCSIS Standard and Related Work 12
2.1 HFC Networks .......................................................................................... 13
2.2 Overview of the DOCSIS S ta n d a rd ........................................................ 16
2.2.1 DOCSIS and related standards ................................................... 16
2.2.2 MAC operations ........................................................................... 18
2.2.3 QOS S u p p o rt.................................................................................. 32
2.3 Open Research Issues................................................................................. 34
3 A Priority D ynam ic Access Scheme for DOCSIS Cable Networks 37
3.1 Introduction................................................................................................. 37
3.2 Review of Collision Resolution A lgorithm s........................................... 39
3.2.1 Truncated Binary Exponential B ackoff....................................... 39
3.2.2 P-persistence A lg o rith m ................................................................ 41
3.2.3 Pseudo-Bayesian A lg o rith m .......................................................... 42
3.2.4 Extended Distributed Queue Random Access Protocol .... 45
3.2.5 N-ARY TREE A LG O RITH M ...................................................... 47
3.3 A novel priority system for DOCSIS protocol........................................ 48
3.3.1 Motivation and problem description............................................. 48
i
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.3.2 Proposed Priority System ............................................................ 50
3.4 Simulation Results and Discussions....................................................... 57
4 Service Scheduling and B andw idth A llocation for DOCSIS 1.1 64
4.1 Introduction................................................................................................ 64
4.2 DOCSIS v l.l Upstream Scheduling S ervices........................................ 65
4.3 Proposed Schemes...................................................................................... 70
4.3.1 Motivation........................................................................................ 70
4.3.2 New Service Class and Bandwidth Allocation
A lgorithm ....................................................................................... 72
4.3.3 Simulation results and discussion................................................ 79
5 Enhanced C SM A /C A Backoff Scheme for IEEE 802.11 Wireless
LAN System 85
5.1 IEEE 802.11 S ta n d a rd ............................................................................. 86
5.1.1 Introduction..................................................................................... 86
5.1.2 Wireless LAN Architecture............................................................ 88
5.1.3 IEEE 802.11 MAC Layer Protocol................................................ 90
5.2 IEEE 802.11 Backoff Mechanism and Its
A nalysis....................................................................................................... 95
5.2.1 IEEE 802.11 Backoff Mechanism................................................... 95
5.2.2 Markov model and A n aly sis............................................................100
5.3 Proposed Adaptive Backoff Mechanism
and Its Analysis........................................................................ 108
5.3.1 Proposed Adaptive Backoff M echanism......................................... 108
5.3.2 Markov Model and A nalysis............................................................109
5.4 Computer Simulation and Performance
Evaluation.......................................................................................................113
5.5 Conclusion...................................................................................................... 117
6 Conclusion and Future Work 127
6.1 Summary of the Research ..........................................................................127
6.2 Future W ork................................................................................................... 130
Reference List 132
ii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Tables
3.1 Simulation p aram eters............................................................................... 57
3.2 Traffic param eters........................................................................................ 58
4.1 Simulation p aram e te rs............................................................................... 80
5.1 System parameters used in analysis and simulation ............................... 113
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
L ist o f F igures
2.1 The HFC network architecture.................................................................. 14
2.2 Comparison of the DOCSIS protocol stack, the OSI model and the
IEEE 802.14 protocol stack......................................................................... 17
2.3 The logical topology of a cable network..................................................... 19
2.4 The DOCSIS initialization process............................................................. 20
2.5 Upstream bandwidth allocation mapping................................................. 23
3.1 The format of the DOCSIS bandwidth management protocol data
unit (PDU).................................................................................................... 40
3.2 The state machine of CRE in the DOCSIS cable modem................... 41
3.3 Priority scheduler in DOCSIS.................................................................... 55
3.4 The modified allocation MAP...................................................................... 56
3.5 The access delay of a non-priority system................................................. 61
3.6 The access delay of the high priority.......................................................... 61
3.7 The access delay of the medium priority.............................. 62
3.8 The access delay of the low priority............................................................ 62
3.9 The throughput performance of the proposed priority system................... 63
4.1 The UGS scheduling service time diagram................................................ 67
4.2 The rtPS scheduling service time diagram................................................ 68
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.3 The UGPS scheduling service time diagram............................................ 71
4.4 Delay of UGPS and rtPS............................................................................ 81
4.5 The throughput performance of UGPS and rtPS.........................................81
4.6 The throughput performance of the best effort class............................... 82
4.7 The total throughput.................................................................................... 82
5.1 The wireless LAN architecture.................................................................... 89
5.2 The IEEE 802.11 MAC architecture.......................................................... 92
5.3 The IEEE 802.11 super frame structure..................................................... 94
5.4 An example of the backoff procedure for the basic access mode. . . . 98
5.5 A typical example of the RTS/CTS access mode.................................... 99
5.6 The Markov model for the IEEE 802.11 backoff scheme........................... 102
5.7 The Markov model for the proposed backoff scheme..................................110
5.8 The saturation throughput for the 802.11 backoff scheme in the RTS/ CTS
access mode with m = 4.................................................................................119
5.9 The saturation throughput for the 802.11 backoff scheme in the basic
access mode with m = 4.................................................................................119
5.10 The saturation throughput for the proposed backoff scheme in the
RTS/ CTS access mode with m — 4..............................................................120
5.11 The saturation throughput for the proposed backoff scheme in the
basic access mode with m = 4....................................................................... 120
5.12 The saturation throughput for the 802.11 backoff scheme in the RTS/CTS
access mode with W = 16.............................................................................. 121
5.13 The saturation throughput for the 802.11 backoff scheme in the basic
access mode with W = 16.............................................................................. 121
5.14 The saturation throughput for the proposed backoff scheme in the
RTS/ CTS access mode with W = 16........................................................... 122
v
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.15 The saturation throughput for the proposed backoff scheme in the
basic access mode with W = 16................................................................... 122
5.16 The saturation delay for the 802.11 backoff scheme in the RTS/CTS
access mode with m = 4................................................................................ 123
5.17 The saturation delay for the 802.11 backoff scheme in the basic access
mode with rn = 4............................................................................................123
-5.18 The saturation delay for the proposed backoff scheme in the RTS/CTS
access mode with rn — 4................................................................................ 124
5.19 The saturation delay for the proposed backoff scheme in the basic
access mode with m — 4.................................................................................124
5.20 The saturation delay for the 802.11 backoff scheme in the RTS/CTS
access mode with W — 16..............................................................................125
5.21 The saturation delay for the 802.11 backoff scheme in the basic access
mode with W = 16..........................................................................................125
5.22 The saturation delay for the proposed backoff scheme in the RTS/ CTS
access mode with W = 16..............................................................................126
5.23 The saturation delay for the proposed backoff scheme in the basic
access mode with W = 16.............................................................................. 126
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A b stra c t
In this research, we examine the design and analysis of the media access control
(MAC) layer protocols for broadband wired and wireless networking systems.
For wired networks, we focus on the hybrid fiber/coax (HFC) networks used in
digital cable modem services under the Data Over Cable Service Interface Specifi
cations (DOCSIS). The availability of high speed-access enables the delivery of high
quality audio, video and interactive services. To support quality-of-service (QoS)
for such applications, it is important for HFC networks to provide effective medium
access and traffic scheduling mechanisms. We first develop a multilevel priority colli
sion resolution scheme with adaptive contention window adjustment. The proposed
scheme separates and resolves collisions for different classes of traffics, thus achieving
the capability for preemptive priorities. Second, a novel scheduling mechanism and
a new bandwidth allocation scheme are proposed to support multimedia traffic over
DOCSIS-compliant cable networks. The primary goal is to improve the transmission
of real-time variable bit rate (VBR) traffic in terms of throughput and delay.
vii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For wireless networks, we consider the wireless local area networks (Wireless
LANs) specified by the IEEE 802.11 standard. In wireless LANs, the CSMA/CA
(Carrier Sense Multiple Access with Collision Avoidance) protocol supports asyn
chronous data transfer, and adopts an acknowledgement mechanism to confirm suc
cessful transmissions and a handshaking mechanism to reduce collisions. In both
cases, a binary exponential backoff mechanism is used. An enhanced CSMA/CA pro
tocol is presented in this research. The enhanced protocol improves the exponential
backoff scheme by dynamically adjusting the contention window around the optimal
value. Moreover, an analytical model based on the Markov chain is developed to
analyze the system performance in terms of throughput and delay. Numerical results
are presented to show the effect of the proposed backoff mechanism.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C h ap ter 1
Introduction
1.1 Significance of th e R esearch
Recently, the rapid growth of the number of residential Internet users and the in
creased bandwidth requirement of multimedia applications have necessitated the
introduction of an access network that can support the demand of such services.
CATV (Community Antenna Television) networks appear to be one of the very
promising solutions to meet this need. CATV networks were first introduced in the
late 1940s to extend the off-air television broadcast signals to locations where the an
tenna reception was poor. Until mid 1970s, satellite delivery of signals were fed into
Cable TV systems to increase the number of delivered channels in Cable TV beyond
those offered by terrestrial broadcasters. The additional offered services created the
desire for Cable TV systems even at locations with good antenna reception. There
are two major reasons for CATV to be a good candidate for broadband Internet
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
access. First, CATV infrastructures already connect a majority of homes. Second,
the Hybrid Fiber Coax (HFC) used in CATV networks can be used to deliver broad
band services without requiring costly upgrade of existing CATV network systems.
We can thus foresee that HFC networks will play an important role for broadband
access networks in the near future.
CATV networks have been traditionally used to provide analog audio and video
broadcasting from the headend to subscribers. However, cable operators are now
interested in developing high speed packet-based communications systems for CATV
networks that are able to support a . wide variety of services. These services include
the IP telephony service, online video gaming, video conferencing, and other real
time and non-real-time services. To meet these new service requirements, several
industry and academic societies are working on new standards. Among these efforts,
DOCSIS (Data Over Cable Service Interface Specifications) driven by CableLabs
and its vendor companies has been the first set of specifications finalized. SCTE
adopted DOCSIS and submitted it to ITU for approval. Hence, DOCSIS is the first
CATV network specification adopted by the international standard. It has also been
adopted by most major vendors. Thus, DOCSIS is expected to be the most widely
used protocol to provide high speed residential access.
CATV networks are characterized by a tree and branch topology with the broad
casting node at the root and recipients at leaves. The bandwidth is divided into
several channels, most of them dedicated to the downstream transmission (from HE
2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to cable modems) while only a few are for the upstream transmission (from cable
modems to HE). Since all users have to share the upstream channel, it is necessary
to identify a MAC (medium access control) protocol to make efficient use of CATV
networks. MAC protocols are designed to solve the shared medium communication
problem.
CATV systems have to operate in an existing network that was not designed
to deliver the additional services. Besides, network upgrades are very expensive.
Thus, it is important to design a system with the current network architecture while
handling the constraints by a carefully designed MAC protocol. MAC protocols de
ployed for other networks, such as LANs, MANs and WANs, have been investigated
extensively but not for DOCSIS. The main objective of this research is to discuss
the service capabilities of DOCSIS MAC layer protocol and its enhancement. We
briefly review research issues about the MAC layer of CATV networks in the next
section and Chapter 2. In Chapters 3 and 4, we will provide more details about the
MAC-layer protocol enhancement for CATV systems under DOCSIS.
The increasing demand of portable and mobile devices accelerates the develop
ment of wireless packet communication systems. Wireless Local Area Network (wire
less LAN) which provides higher flexibility and convenience than the wired networks
plays a key role in the personal communication systems recently. The IEEE Project
802 introduces an international standard 802.11 for wireless LAN. This standard
defines the detail of both the medium access control (MAC) and the physical layers.
3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The MAC layer supports many important functions of a wireless LAN. One of these
functions is to coordinate multiple access to the wireless medium among a group
of mobile terminals. In the IEEE 802.11 standard, the fundamental mechanism of
accessing a channel is called the distributed coordination function (DCF), which
supports asynchronous data transfer on the best effort basis. The DCF utilizes a bi
nary exponential backoff scheme. To analyze the performance of the DCF protocol,
several papers examined the modelling of the IEEE 802.11 standard. A p-persistent
backoff scheme, instead of the binary exponential backoff scheme, was proposed in
[6 ] and its throughput limit analysis was conducted accordingly. The problem with
the p-persistent backoff scheme is that it is inconsistent with the IEEE 802.11 stan
dard. A Markov model was used in [1 ] to analyze the saturation throughput of the
IEEE 802.11 binary exponential backoff scheme. However, the model in [1 ] does not
consider the busy channel condition.
To achieve a better performance, it is important to understand the limit of the
IEEE 802.11 backoff scheme and modify the protocol accordingly. Moreover, to
gain more insights into the DCF operation and obtain a more precise performance
evaluation of the protocol, the mathematical model used in previous research has
to be refined. In Chapter 5, we discuss issues associated with the current IEEE
802.11 backoff scheme and propose a methodology to improve the performance of
the IEEE 802.11 DCF protocol. Besides, a more accurate Markov model will be
used to analyze the saturation throughput and delay of the proposed scheme.
4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.2 Research Issues about DOCSIS and IEEE
802.11 M AC layer
1 .2.1 Issues a b o u t D O C S IS M A C layer
The DOCSIS standard was developed to facilitate the inter-operability between ca
ble modems and the headend designed by different vendors. However, there are
open vendor-determined issues, such as request minislot allocation and data min
islot scheduling, which significantly influence the performance of the HFC MAC
protocol. We address these issues in this section.
• Allocating request minislots
When a cable modem has a bandwidth requirement and adopts the normal
reservation mode to access the bandwidth, it randomly selects one of the re
quest minislots allocated by the headend and sends its request on that min
islot. The shared request minislot is however subject to collision. Thus, the
objective of a request minislot allocation algorithm is to allocate the right
number of request minislots under various traffic loads so that the request
minislot throughput could be maximized. Some researchers have investigated
this problem with the IEEE 802.14 protocol. For instance, Lin, Huang and Yin
[16] published a statistically optimized minislot allocation (SOMA) algorithm
that maximizes the request minislot throughput by estimating the number of
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
new requests with a time proportional scheme and the number of collided re
quests by looking up a statistical most likelihood number of the request (MLR)
table. Twu and Chen [22] proposed a P-Tree algorithm that also uses the time
proportional scheme to estimate the number of requests. However, this prob
lem has not been well addressed in DOCSIS. This issue will be studied in
Chapter 3.
• Traffic scheduling and bandwidth allocation
The scheduling algorithm highly affects the quality of services (QoS) of each
flow. The headend must schedule data minislots to different flows according to
their QoS parameters. For DOCSIS networks, Rabbat and Siu [17] presented
an efficient scheduling algorithm to multiplex the constant bit rate traffic and
the best effort traffic to support an integrated Service. Alternatively, it can
conduct polling dynamically to allow an idling flow to send a bandwidth request
again, thus reducing delay in the request contention process. Simulation results
show that the minimum bit rate and delay requirements of QoS flows can be
achieved. We will investigate these issues in Chapter 4.
• Performance evaluation
Golmie, Mouveaux and Su [11] compared DOCSIS vl.O with IEEE 802.14 in
terms of contention access, ATM vs. IP transfer, and adequate QoS provision.
The n-ary tree-based collision resolution used in IEEE 802.14 gives lower ac
cess delay and delay variance than the binary exponential backoff used in the
6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
DOCSIS standard. IEEE 802.14 provides a friendly ATM environment and
provides good support of QoS, while DOCSIS offers more efficient Internet
access. However, DOCSIS v l.l might provide better support of QoS.
1.2.2 Issues about IE E E 802.11
According to the IEEE 802.11 DCF protocol, when a station begins its backoff
procedure, the backoff value is selected from the backoff window range using the
uniform distribution. Whenever a collision occurs, the backoff window size is doubled
until it reaches the maximum backoff window size. Since a station uses CW to
control the backoff counter for data transfer, the CW setting strategy will affect
the performance of the CSMA/CA protocol. To optimize the performance of the
CSMA/CA protocol, we consider the backoff procedure as a progress to search the
optimal value of CW under a certain traffic condition. The problem is that the CW
resetting scheme in the current IEEE 802.11 standard ignores the traffic load. To
improve the performance of IEEE 802.11 DCF protocol, the backoff scheme should
take the network traffic condition into account. A new design will be proposed in
Chapter 5 based on the above arguments.
1.3 C o n trib u tio n s of th e R esearch
For both HCF and IEEE 802.11 wireless LAN networks, the medium is shared
by all stations and the MAC layer protocols that take care of the coordination of
7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
communication between all stations is indispensable. Design of an effective MAC
layer is critical to the system performance.
With more advanced modulation techniques, future upgrades of this system and
new digital TV broadcasting services, more bandwidth will likely be available to
the upstream channels in HFC networks. In view of this evolution, the protocol
should exploit the medium efficiently over a broad range of transmission speeds.
The protocol has to be flexible enough to control the multiplicity of collision and
support multiple priority traffic classes. Also, a high degree of flexibility is required to
support a large number of users with a mix of applications with different constraints.
Latency should be relatively low (less than a few tens of milliseconds) so that voice
conversation and interactive applications can be supported. One major contribution
of this research is that we propose several enhancements for the DOCSIS protocol
over HFC networks to solve the above problems.
The IEEE 802.11 standard used DCF as the fundamental MAC technique. Most
research evaluates the performance of IEEE 802.11 DCF by means of computer sim
ulation [9], [23] or analytical models with a simplified backoff rule [2], [8]. The other
major contribution of this research is that we succeed in providing a simple Markov
model that accounts for many details of the binary exponential backoff scheme and
allows to compute the saturation throughput and delay of DCF for both basic and
RTS/CTS access modes. The proposed Markov model is more accurate than that
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
in [1 ] since we take account of the busy medium conditions. Furthermore, to en
hance the performance of the primary IEEE 802.11 DCF, we propose a refinement
for the backoff mechanism by dynamically resetting the CW based on the network
traffic load. We also evaluate the performance of the proposed CW resetting scheme
using the new Markov model and validate our theoretical analysis by exhaustive
simulations.
The main contributions of this research are summarized as follows.
• A novel collision resolution scheme is proposed in Chapter 3. This scheme
extends the binary exponential backoff scheme from the single priority to the
multiple priority. This scheme dynamically estimates the number of contending
stations through collision requests and adjust contention windows for different
traffic classes based on their priorities. Through simulation studies, we show
that the proposed scheme enriches and enhances the services of the current
DOCSIS specification.
• To provide services with desired QoS, we add some QoS mechanism to the
headend (HE) to enable it to support delay-sensitive flows and bandwidth re
quirements over CATV networks. We propose a novel scheduling service class
and bandwidth allocation algorithm for the DOCSIS specification in Chapter
4. The primary goal is to improve the transmission efficiency of real time
VBR traffic in terms of throughput and delay under the current DOSIS spec
ification. The performance evaluation of both the DOCSIS specification and
9
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the proposed scheme is illustrated via simulation. It is shown that the pro
posed scheme provides a significant amount of improvement over the existing
DOCSIS QoS scheduling services.
• A discrete-time Markov model is developed based on the scheduling services
of DOCSIS to analyze voice services over DOCSIS-compliant networks. Two
classes of voice services are considered, namely, voice with speech activity
detectors (SAD) and voice without SAD. We use two scheduling service classes
that are well-defined in the DOCSIS specification to transmit these two voice
services separately. Performance metrics, including packet loss probability
and bandwidth utilization, axe computed to demonstrate the efficiency of the
DOCSIS standard.
• For the IEEE 802.11 standard, a dynamic backoff window adjusting scheme
that allows the backoff window size to oscillate around the optimal value is
proposed. Besides, we employ a Markov model by taking the busy channel
condition into account. We show how the proposed scheme affects the behavior
of the backoff mechanism. Moreover, we apply the proposed Markov chain
model to analyze the throughput and delay performance of the new scheme.
Finally, the derived analytical results are verified via comprehensive computer
simulation.
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.4 O utline of D issertatio n
This thesis is organized as follows. Chapter 2 illustrates the MAC protocol of the
DOCSIS standard. The operations of the DOCSIS cable modem is first discussed.
Features and important MAC design issues are also presented. Chapter 3 focuses on
the treatment of the collision resolution problem. Several collision resolution mecha
nisms are reviewed. We show that existing collision resolution mechanisms cannot be
applied to the DOCSIS protocol properly. To solve this problem, we propose a novel
collision resolution mechanism that supports multilevel priority collision resolution
with adaptive contention window adjustment. In Chapter 4, we investigate the QoS
and bandwidth allocation problems for the DOCSIS protocol. A new scheduling
service and a new bandwidth allocation algorithms are proposed. In addition, we
employ the Markov model to evaluate the performance of the DOCSIS protocol.
Chapter 5 describes the detail of the IEEE 802.11 network structure and its MAC
layer operation. The features of the binary exponential backoff scheme is carefully
examined and its performance is analyzed based on a Markov model. To improve the
performance of the IEEE 802.11 MAC layer, we propose some modification of the
primary backoff mechanism. We also present a Markov analysis for the new backoff
design. The performance of both the existing and the proposed backoff schemes are
compared according to theoretical and simulation results. Finally, some concluding
remarks of this research are given in Chapter 6.
11
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
DO CSIS Standard and R elated Work
In this chapter, we review the MAC protocol adopted by DOCSIS HFC networks
and point out some open research problems. DOCSIS was approved by ITU as a
standard and is supported by many industrial vendors. We will first discuss features
of HFC networks in Section 2.1, and then examine basic operations and mecha
nisms of DOCSIS MAC protocols in Section 2.2. The DOCSIS standard treats the
upstream channel as a stream of minislots and has mechanisms for upstream band
width management, downstream in MPEG-2 format, data-link-layer security, and
ranging. Besides, the standard adopts some mechanisms for upstream access modes,
QoS support, and collision resolution which are the main concern of our research
and will be discussed in detail later. Finally, we identify some research issues in
HFC MAC protocols, particularly allocation and scheduling issues, in Section 2.3.
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.1 H F C N etw orks
In a conventional cable TV (CATV) network, the service provider sends analog TV
programs to subscribers via the cable network. Amplifiers must be installed in the
cable network due to the fading signal effect. The amplifiers provide only one-way
capability, accounting for the lack of an upstream channel in the CATV network.
Being adopted by many cable companies, the HFC technology provides upstream
channels in a coaxial cable distribution network. With the availability of upgraded
amplifiers to support two-way amplification and fiber replacement for long distance
transmission, subscribers are able to send data back to the service provider side.
Fig. 2.1 depicts a typical HFC system. A fiber node, capable of serving 500 to
2000 subscribers, receives signals sent from the headend via a fiber. These optical
signals are then translated into electrical signals and sent to amplified tree-and-
branch feeder cables. Subscribers can receive or transmit signals by connecting their
coaxial stations, i.e. set-top boxes or cable modems, to the taps of the network.
With multiple access technologies, all subscribers within a branch can share the
upstream bandwidth to send data back to the headend.
The HFC network possesses the following features that affect the MAC protocol
design.
• Point-to-multipoint downstream and multipoint-to-point upstream
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
jiomes
"'xjpaxial domain
fiber
domain
Coax
amplifiers
headend
A neighborhood
fn: fiber node
Figure 2.1: The HFC network architecture.
It is a point-to-multipoint, tree-and-branch access network in the downstream
direction, but a multipoint-to-point bus-like access network in the upstream
direction. Since collisions may occur among upstream users, the shared up
stream channel needs an efficient scheme to avoid and/ or resolve collisions.
• The inability to detect collisions by stations
Stations can only listen to the downstream traffic, which differs from an Eth
ernet where adaptors can detect when collisions occur. Thus, cable modems
(CM) rely on the headend (HE) to notify them of the results of upstream
transmissions.
• Large propagation delay
The maximum round-trip-delay (RTD) is significantly longer than that of the
Ethernet. Therefore, a channel should be utilized to transmit other data frames
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
during the RTD of a transmitted data frame. In an Ethernet, however, no
other data frames should be transmitted during the RTD of the data frame
if it is to be successful. Furthermore, neutralizing the effect of propagation
delay is of concern to synchronization so that the transmissions from stations
arrive at the right time slots assigned by the headend. Consequently, the MAC
protocol should have a ranging scheme to measure the propagation delay for
each station.
• Asymmetric upstream and downstream bandwidth
The downstream data rate is substantially larger than that of the upstream.
The efficiency of upstream channels is critical.
• Non-uniform user distribution
Most subscribers are distributed over the last few miles of the network. The
propagation time among subscribers in the same cluster to the headend is close
to each other. On the other hand, subscribers in different clusters may a non-
uniform distance distribution. Repeated collisions may occur for a straightfor
ward ranging algorithm that does not consider this factor.
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.2 O verview of th e DOCSIS S tan d ard
2.2.1 D O C S IS an d re la te d sta n d a rd s
To reduce the equipment cost of cable companies, it is desirable to mass-produce
interoperable components and systems. To achieve this goal, several standardization
efforts of cable modem systems have been initiated, including the Multimedia Cable
Network System (MCNS), the IEEE 802.14 Working Group and the European Cable
Communication Association (ECCA). Formed in May 1994 by several vendors, the
IEEE 802.14 Working Group has been working on international standards for data
communications over cables. Due to the delayed progress, four major cable opera
tors, Comcast Cable Communications, Cox Communications, Tele-Communications
Inc., and Time Warner Cable, established MCNS in December 1995 to create the
DOCSIS standard. As to the European cable environment, ECCA started to create
the EuroModem specification in December 1998.
DOCSIS vl.O was approved as a standard by ITU on March 19, 1998, and cur
rently dominates the market. In addition, DOCSIS v l.l, whose major feature is
supporting QoS, was released on July 31, 1999. Broadcom and Terayon are work
ing with MCNS to implement an IEEE 802.14-endorsed advanced PHY technology
into DOCSIS. The emerging standard will be known as DOCSIS 2.0. Currently the
certification program is ready for DOCSIS vl.l-compliant products.
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TC: transmission convergence PMD: physical media dependent
OSI
Data Link layer
Convergence
sublayer
ATM
DOCSIS
Data Link
Layer
Link Security
IEEE 802.14
Data Link
Layer
MAC
Link Security
Media access
arbitration
DOCSIS
Physical
Layer
TC
OSI
Physical layer
TC
IEEE 802.14
Physical
Layer
PMD
PMD
DOCSIS protocol stack OSI model IEEE 802.14 protocol stack
Figure 2.2: Comparison of the DOCSIS protocol stack, the OSI model and the IEEE
802.14 protocol stack.
In contrast, the IEEE 802.14 Working Group was disbanded in March 2000,
and IEEE 802.14a will remain as a draft afterward. The group had a well thought
plan, and its specification is better than that developed by MCNS from a techno
logical viewpoint. However, timing is critical. Since the group was not able to de
velop a specification within a short period of time, its impact becomes quite limited.
Moreover, the EuroModem vl.O was approved by the European Telecommunications
Standard Institute (ETSI) on May 12, 1999. In addition to IEEE 802.14, MCNS,
and ECCA, other standard associations working topics related to cable networks.
They include the Internet Engineering Task Force (IETF) IP over Cable Data Net
work Working Group, the ATM Forum Residential Broadband Working Group, the
Society of Cable Telecommunications Engineers, and ITU.
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Since the development of DOCSIS vl.O followed the development of IEEE 802.14,
it has several good features from IEEE 802.14. including virtual queue, minislot,
downstream MPEG-2 format, security module, piggyback, synchronization proce
dure, and modulation schemes. However, to reduce the implementational complexity,
variable length frames and relatively simple collision resolution schemes are defined
in DOCSIS vl.O. In order to support QoS, six scheduling services are included in
DOCSIS v l.l. Also, segmentation and concatenation of IP traffic are specified to
increase the system throughput in this version. Fig. 2.2 schematically depicts the
protocol stacks of DOCSIS v l.l. Both DOCSIS and IEEE 802.14 provide the ca
pabilities to transport 802.2 logical link control (LLC) protocol data units (PDUs)
over HFC networks. IEEE 802.14a attempts to provide complete support of Asyn
chronous Transfer Mode (ATM), thus making the MAC-CS layer and the ATM layer
necessary. The MAC-CS transforms data passing through the LLC SAP into ATM
PDUs for transmission over the network. The general features of PHY, including
TC and PMD, and MAC layers are described in the following subsections.
2.2.2 M A C operations
2.2.2.1 Initialization
The logical topology of a cable modem network is shown in Fig. 2.3. The down
stream path flows from the headend to all stations and resides in a 6 MHz TV
channel selected by the cable operator. The headend, which is owned and operated
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Network
Upstream
HE
CM CM CM I
Other network
- connections'^,
(vedio, voicej
Downstream
Figure 2.3: The logical topology of a cable network,
by the cable company, is the exclusive transmitter in the down-stream direction.
Therefore, no downstream media access control MAC mechanism is needed. While
the downstream channel is broadcast in nature, each subscriber unit is assigned an
individual address that allows it to filter out any data not addressed to it via unicast,
multicast, or broadcast transmission. Security and data encryption mechanisms are
in place to provide authentication and privacy and to help prevent theft of service
and denial of service attacks.
The upstream paths flow from subscriber units referred to as CMs in this docu
ment towards the headend. Each upstream channel is shared by a number of stations
and is divided in time into individually numbered allocation units called minislots.
The shared nature of the upstream traffic requires a MAC to order transmissions.
Physical requirements of CMs demand isolation of signals in the upstream direction.
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
MAC
Headend
TC PHY
PHY
Cable modem
TC
MAC
Tune D ow nstream
Broadcast messages
Range, response
Adjust range
rational param eters djwnload Ope
-Registration request.
Registration response
Figure 2.4: The DOCSIS initialization process.
This isolation leads to a scenario where upstream transmissions can only be heard
by the headend and not by CMs. Thus, concurrent upstream CM transmissions can
collide, but individual stations are unable to hear transmissions of other stations.
This means that the carrier sense MAC technique (e.g. CSMA) will not work in a
CATV environment.
We investigate MAC operations from the initialization procedure of a startup
station and the normal operation of an initialized CM below.
1. Channel acquisition
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Upon initialization or after recovering from signal loss, a station should ac
quire a downstream channel by scanning the downstream frequency band until
its receiver identifies a valid downstream signal. After achieving physical-level
synchronization, the station can learn the characteristics of the upstream chan
nel from the specific management messages broadcast by the headend. Thus,
the station tunes its transmitter to the upstream frequency band specified in
the messages. Furthermore, when determining that the channel is overcrowded,
the headend may ask some stations to switch to another channel.
2. Ranging
Owing to the large propagation delay in the HFC network, each station must
learn its distance from the headend and compensate for this distance such
that the station and the headend have a consistent system-wide view of time
to synchronize their MAC operations. This process is referred to as ranging.
Later, we further describe the ranging process.
3. Operational parameters download
After performing the ranging process, a station downloads operational parame
ters from the headend. These operational parameters include IP address, secu
rity information, channel con-figuration, class of service configuration, SNMP
MIB object, etc.
4. Registration with the headend
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In DOCSIS, a station sends a registration request, which contains the opera
tional parameters, to the headend. The headend then performs the following
functions:
• Confirms the validity of the operational parameters;
• Builds a profile for the station:
• Assigns a service ID (SID);
• Sends back a registration response to the station.
Fig. 2.4 illustrates the initialization process. After these initialization steps,
the station enters the normal operation.
2.2.2.2 Normal Operation
The MAC is the entity that controls the allocation of bandwidth of the shared up
stream media. As mentioned earlier, only the upstream communication needs MAC
because the headend is the exclusive TV channel selected by the cable operator.
However, downstream transmissions are required for proper operation of the MAC
protocol. Specifically, the MAC protocol operation is split between individual CMs
and the headend controller. Most of the scheduling and processing burden is rele
gated to the headend. During the normal operation, the headend must coordinate
accesses to this shared bandwidth since CMs cannot listen to the upstream channel.
The headend assigns the usage of upstream bandwidth and describes the assignment
in the Bandwidth Allocation Map (MAP).
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Allocation / • * *\
> = Map 1 1 =Timestamp
Downstream PDUs
3 0
Station Contention CM2 Tx CM1 Tx
^ register
^ Upstream Minislots
CMn Tx: minislots dedicated to CM n for Upstrean transmission
Figure 2.5: Upstream bandwidth allocation mapping.
Once the MAP is sent over the downstream channel, CMs can learn the assign
ment from the MAP and proceed according to information conveyed by the MAP.
Basically, some of the upstream minislots are assigned as request minislots, each
of which can accommodate a request packet data unit (PDU). The other minislots
are data minislots where a data PDU may occupy multiple contiguous minislots. To
reduce bandwidth waste due to collisions, CMs first send small request PDUs, which
are subject to collisions, to the headend. The headend then schedules requests and
informs stations, through downstream channels, so that their upstream data PDUs
can be sent collision-free.
Fig. 2.5 illustrates the upstream bandwidth allocation. Upstream bandwidth is
allocated in granular units called minislots. A minislot is the unit of granularity for
23
: Data
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
upstream transmission opportunity. With the minislot mechanism, the upstream
transmission efficiency is increased since the right size of transmission opportunity
is allocated to a specific request, and the bandwidth allocation is thus more flexible.
In DOCSIS, the size of the minislot depends on the adopted modulation scheme
and must be a multiple of 6.25 microsecond. Each minislot has an integer identifier,
called the minislot number, which is assigned by the headend. When the minislot
number counts its maximum value, it wraps back to zero. CMs and the headend
should be able to recognize the minislot number. Notably, concatenated multiple
minislots can be used to transmit a data PDU.
In addition to the normal reservation, DOCSIS also provides isochronous access,
periodic request polling, and immediate access modes. A flow with a constant bit
traffic rate may periodically get data transmission opportunities via the isochronous
access mode, while a flow with a variable bit traffic rate could request bandwidth on-
demand through the periodic request polling mode. Once a DOCSIS CM has short
data that occupies few minislots, the CM may even bypass the request process and
burst its data directly in the immediate access region, if allocated by the headend
when the traffic load is not heavy. These operation modes are about the scheduling
services, we will discuss them in detail later.
2.2.2.3 R anging
Cable networks can have subscriber units as far as 100 miles (for DOCSIS) from
the headend. Typical distances are much less. Signal propagation time over such
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a distance imposes the timing constraint on the MAG design. DOCSIS states that
MAC must handle propagation delays as much as 800 ms in each direction. Such
a delay cause a problem in the upstream since two widely separated units could
start transmission during consecutively numbered upstream minislots yet have their
data arrive at the headend out of order or in a collision state. To help solve this
situation, MAC mandates that upstream timing must be adjusted by each station
in such a manner that if two stations transmit during the same minislot, both their
transmissions will arrive at the headend at the same instance.
To help manage the upstream timing, which has physical layer dependencies, a
new sublayer is inserted between the PHY and the MAC. This layer is known as
the Transmission Convergence (TC) layer that helps hide PHY timing dependencies
from MAC. One job of TC is to coordinate with MAC’s ranging procedure to acquire
and maintain a lock on the upstream timing. A lock means that the TC knows the
exact timing adjustment values needed to account for the propagation delay of its
upstream transmissions. Once a lock is achieved, the MAC protocol can operate
using the notion of minislot numbers, and the TC handles the timing consideration
necessary to deliver numbered minislots to the headend at an appropriate time.
The ranging procedure is stated below.
• Obtain the global timing reference
Once empowered, a CM should listen to the sync message sent periodically by
the headend. The sync message contains a timestamp that records the time
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
at which the headend transmits the message. Upon receiving a sync message,
the station sets its local clock to the timestamp. When syncing multiple times,
the CM's clock rate can be synchronized to that of the headend. Notably, this
process continues even after initialization.
• Identify the ranging area
The headend also periodically broadcasts a bandwidth allocation MAP and
a ranging invitation to invite all unranged CMs to join the network. A CM
learns the ranging area from the starting minislot number and the ranging
area length described in the message. Notably, the headend must allocate a
ranging area sufficiently large to accommodate the longest propagation delay.
• Send the ranging message
After finding the ranging area, a CM can send its ranging request back to
the headend in the ranging area. If the above ranging message, which is
subject to collision, is successfully received, the headend would evaluate the
timing offset and other miscellaneous parameters that the CM should tune to.
These adjustment parameters are then sent back to the station via the ranging
response.
• Adjust according to the feedback message
A CM is roughly ranged after adjusting its parameters, including timing offset,
power level, frequency offset, and center frequency, according to the offered
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
values in the feedback message. The ranging process is repeated until the
headend determines that no more adjustment is required.
The bandwidth allocation MAP contains not only ranging area information but
also other bandwidth allocation information. In addition, the ranging process is fur
ther divided into initial ranging and station ranging. Initial ranging largely focuses
on obtaining a temporary SID to facilitate further initialization operations, in which
the CM uses the CM maintenance area to perform periodic ranging. Moreover, dur
ing the ranging process, a CM adopts a binary exponential backoff algorithm as a
collision resolution algorithm to resolve collisions.
2.2 .2 .4 A ccess modes
IEEE 802.14a only supports the normal reservation and piggyback reservation modes.
However, in DOCSIS v l.l, the isochronous access, periodic request polling, and im
mediate access modes are also provided. The normal reservation mode prevents
data transmission from excess collisions, and the piggyback reservation mode can
reduce the request access delay. Isochronous access is fulfilled by periodically allo
cating data transmission opportunities, whereas periodic request polling is fulfilled
by periodically allocating request transmission opportunities. Both access modes
are designed for QoS flows. Immediate access is triggered when bandwidth is still
available after satisfying all bandwidth requirements, and this access mode is open
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to both data and request. If the load is light, this access mode could be utilized to
reduce data and request access delays.
2.2.2.5 Transport protocols
Both IEEE 802.14 and DOCSIS support the transport of multimedia traffic. How
ever, the foundations of the two specifications differ. The IEEE 802.14 MAC provides
a fixed length transport that is based on asynchronous transfer mode (ATM). The
DOCSIS MAC provides a variable length transport that is geared towards efficiently
delivering IP traffic encapsulated in Ethernet frames. In both specifications, mech
anisms are provided to encapsulate the respective fixed or variable length protocol
data units (PDUs) into the downstream framing mechanism used in CATV systems.
DOCSIS MAC PDUs are also encapsulated in the payload of an MPEG-2 stream.
MAC PDUs can start anywhere in the MPEG-2 payload and can span MPEG-2
frame boundaries. Unlike IEEE 802.14, DOCSIS does not specify an ATM frame
format. However, to allow for future modifications, the DOCSIS standard does pro
vide a mechanism for the MAC to determine if a frame encapsulates ATM cells.
This allows an existing MAC to ignore ATM data and remain interoperable with
possible future encapsulations. In the upstream, MAC PDUs are packed into minis
lots. Since upstream frames are not required to have a size that is a multiple of the
upstream minislot size, MAC must provide any padding that is necessary.
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2 .2 .2 .6 A d d re ssin g
All subscriber units are assigned a unique 48 bit IEEE 802 MAC address. However,
this address is not the primary means of addressing a station and is only used during
the registration process. During registration, subscriber units are assigned one or
more 14 bit local identifiers. These identifiers describe a mapping between the CM
and the headend controller and are only valid inside the cable modem MAC domain.
The first identifier assigned to a CM is the primary local identifier. It is used for
MAC administration, security, and PHY control. Other local identifiers may be
assigned to support station management, class of service, and multicast groups.
Unlike IEEE 802 MAC addresses, no bit level differentiation exists in the 14 bit
local identifiers. Unicast and multicast identifiers are assigned by the headend from
the same 14 bit address space. Multicast identifiers are also associated with 48 bit
IEEE 802 multicast identifiers. When a new multicast group is formed, the headend
allocates a new local identifier to associate with that group. CMs may join the group
by making requests to the headend. This process will cause the CM to recognize
the multicast identifier and will also allow' that station to receive encryption keys
that allow it to decode the information in the multicast stream. CMs may leave the
multicast group through similar procedures. Once a CM leaves the multicast group,
the headend will cease sending the multicast encryption keys to that unit.
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2 .2 .2 .7 S e c u rity
Security considerations play a key role in cable modem networks. The topic of
security deserves an article in itself and is only briefly described here. The security
models of competing specifications differ but rely on the same set of principles. They
are all designed to protect against theft of service and denial of service caused by
alteration of subscriber unit security components, interception and replay of the data
stream, cloning, masquerading of valid units, and misuse of upstream bandwidth.
The security systems provide the following features.
• Identification and authentication of the sub-scriber
The DOCSIS protocol uses the concept of a secret cookie for authentication.
How that cookie is originally supplied to the CM is part of security policy.
DOCSIS suggests that the MAC address be given to the headend via some side
band channel. Furthermore, DOCSIS outlines the capability to use (X.509v3)
digitally signed certificates generated by the manufacturer. These certificates
are used to authenticate the validity of the subscriber device.
• Confidentiality of user information through encryption
Session keys are used to encrypt or decrypt the payload of all unicast and
multicast PDUs. Initial key values are exchanged during the registration pro
cess and are then continually updated according to the security policy. Each
MAC PDU has an encryption key identifier that aids in switching between
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
security keys. This allows keys to be negotiated and switched on-the-fiy. Each
multicast session has a separate set of keys.
® Anonymity
The registration procedures are designed such that the 48 bit IEEE MAC ad
dress and a CM’s local identifier are never sent together. This helps to provide
anonymity by hiding distance measures that can be derived from monitoring
the ranging process.
The DOCSIS security model can manifest in one of two forms. The first is a full
security specification and the second is a simplified version called Baseline Privacy.
Both provide privacy through data encryption. The encryption algorithm is based
on a , 56 bit DES cipher block chaining algorithm. RSA public key encryption is used
to provide authentication through the use of initial keys stored in the subscriber
unit. These keys are generated and digitally signed by the manufacturer. If one
chooses to use Baseline Privacy, the authentication capabilities are removed. In the
absence of authentication, the system can still ensure that a subscriber unit can only
receive information that it is authorized to access. However, Baseline Privacy is not
able to detect clones or to protect against some of more sophisticated attacks.
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.2.3 Q O S S u p p o rt
To support QoS, DOCSIS defines six QoS services: (1) unsolicited grant service
(UGS); (2) unsolicited grant service with activity detection (UGS-AD); (3) real
time polling service (rtPS); (4) non-real-time polling service (nrtPS); (5) best effort
(BE) service; and (5) committed information rate (CIR) service. We introduce the
six QoS services as follows.
. UGS
The aim of UGS is to reserve guaranteed upstream transmission bandwidth for
traffic flows. This means that the headend must provide fixed size data grants
at periodic intervals to the UGS flows. Upon receiving the request from the
CM, the headend scheduled fixed size grants at periodic intervals to the UGS
flow. The CM needs to send the request only once, it is the responsibility of the
headend to control the timing of the allocated grants that provide the required
delay and delay-jitter bound. UGS can provide deterministic QoS guarantees.
However, the reserved bandwidth may be wasted when a corresponding UGS
flow is inactive.
• UGS-AD
For UGS-AD flows, the headend employs an activity detection algorithm to
examine the flow state. It is assumed that the headend can detect the activity
of UGS-AD flows through the activity detection algorithm. When an UGS-AD
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
flow is active,the headend provides periodic grants to this flow. Once a flow
is changing from an active state to an inactive state, the headend reverts to
provide periodic request polling.
• rtPS
rtPS is intend to reserve the transmission opportunities for real time verbile bit
rate (rt-VBR) applications. Upon receiving the request from the rtPS flow, the
headend polls the flow periodically so that the flow can send their bandwidth
request even when the network traffic is congested. rtPS can provide statistical
QoS guarantees and high network utilization.
• nrtPS
Both nrtPS and rtPS flows are polled through the periodic request polling.
However, nrtPS flows receive few request polling opportunities during network
congestion, while rtPS flows are polled regardless of the network load. The
objective of nrtPS is to reserve non-real-time applications such as FTP. The
headend polls nrtPS flows either on a periodic or non-periodic interval so that
the traffic flows can receive some transmission opportunities during network
congestion. The CM uses both contention request and unicast request oppor
tunities to have the upstream transmission opportunities.
• BE
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The BE service is to provide transmission opportunities for the best effort
traffic. For the BE service, a station must use the normal reservation mode or
the immediate access mode to gain upstream bandwidth.
• CIR
A CIR service can be defined by vendors in a number of different ways. For
example, it could be configured by using the nrtPS service with a reserved
minimum traffic rate.
To meet the QoS requirements, the headend must adopt an admission control
mechanism and a scheduling algorithm among different services to reduce the QoS
violation probability. Each QoS flow matches exactly one QoS service. If a station
has a special bandwidth requirement not specified in the QoS service profile, it
could dynamically request a service by sending a dynamic service addition request
message to the headend. Moreover, after a QoS flow is established, the payload
header suppression mechanism can be adopted to efficiently utilize the bandwidth
by replacing the repetitive portion of payload headers with a payload header index.
2.3 O pen R esearch Issues
DOCSIS was developed to facilitate inter-operability between the CMs and headend
designed by different vendors. However, there are open and vendor-determined is
sues, such as request minislots allocation and data minislots scheduling algorithms.
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
These choices significantly influence the performance of the HFC MAC protocol. We
address these issues below.
• Allocating request minislots
After a CM has a bandwidth requirement and adopts the normal reservation
mode to access bandwidth, it randomly selects one of the request minislots
allocated by the headend and sends its request on that minislot. The shared
request minislot, however, is subject to collision. The objective of allocating
request minislots is to allocate the right number of request minislots under
various traffic loads so that the request minislot throughput could be maxi
mized. Some researchers has investigated this problem for the IEEE 802.14
protocol. For instance, Lin, Huang and Yin [16] has published a statistically
optimized minislot allocation (SOMA) algorithm that maximizes the request
minislot throughput by estimating the number of new requests with a time
proportional scheme and the number of collided requests by looking up a sta
tistical most likelihood number of requests (MLR) table. Twu and Chen [22]
proposed a P-Tree algorithm that also used the time proportional scheme to
estimate the number of requests. However, this problem has not been well
discussed in DOCSIS. This issue will be explored in the chapter 3.
• Traffic scheduling and bandwidth allocation
The scheduling result highly affects the QoS of each flow. Thus, the headend
must schedule data minislots to flows according to their QoS parameters. For
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
DOCSIS networks, Rabbat and Sin [17] presented an efficient scheduling al
gorithm to multiplex the constant bit rate traffic and the best effort traffic to
support Integrated Service. Alternatively, it can conduct polling dynamically
to allow an idling flow to send a bandwidth request again, thus reducing delay
in the request contention process. Simulation results show that the minimum
bit rate and delay requirements of QoS flows are achieved. We will investigate
these issues in Chapter 4.
« Performance evaluation
Golmie, Mouveaux and Su [11] compared DOCSIS vl.O with IEEE 802.14 in
terms of contention access, ATM vs. IP transfer, and adequate QoS provision.
The n-ary tree-based collision resolution used in IEEE 802.14 gives lower access
delay and delay variance than the binary exponential backoff used in DOCSIS.
IEEE 802.14 provides a friendly ATM environment and good support of QoS,
while DOCSIS offers more efficient Internet access. However, DOCSIS v l.l
might provide better QoS support.
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
A Priority D ynam ic A ccess Schem e for
DOCSIS Cable Networks
3.1 In tro d u c tio n
In the DOCSIS protocol, cable modems that have data to transmit have to send
allocation requests to the headend. The size of allocation request is short. Usually,
allocation request contains the local identifier of the cable modem and a count of
the number of minislots needed. If a cable modem has already been granted a right
for upstream transmission due to its previous request, it must wait for the transmit
time to arrive and then send its packets. On the other hand, cable modems that
do not have an existing allocation have to send their requests in contention-based
allocation request minislots. By allocation request minislots, we mean a group of
short-sized minislots in which any cable modem can send a bandwidth request.
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
If only one cable modem sends a request in an allocation request minislot, the
headend will receive that request successfully and respond to it with an allocation
pending message. This message is sent back to the cable modem and will be fol
lowed by an upstream bandwidth grant for that cable modem in a future bandwidth
allocation map. However, if more than one cable modem choose the same allocation
request minislot to send their requests, collision occurs. The contention resolution
algorithm is a set of rules used to resolve collisions in the request minislot contention
process. The main functions provided by the contention resolution algorithms in
clude the following:
1. controlling the transmission of new request transmission rule;
2. providing collision feedback;
3. managing the retransmission process.
In this chapter, we first review the contention resolution algorithm of DOCSIS
and some traditional contention resolution algorithms. The performance and im
plementation issues of these contention resolution algorithms are analyzed. Our
study includes all major DOCSIS proposed for cable MAC protocols, namely, the
truncated binary exponential backoff algorithm, the p-persistence algorithm and the
pseuso-Bayesian algorithm. The performance behavior of these algorithms in the
random access channel is extensively examined. Furthermore, we investigate the
capability of the MAC protocol that is currently defined by DOCSIS to support a
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
preemptive priority access system. Then, we propose a novel priority access scheme,
called the dynamic backoff window scheme, to provide a preemptive priority access
system. Finally, we conduct numerical simulation with the OPNET Common Sim
ulation Framework, to demonstrate the performance improvement of the proposed
algorithm.
3.2 R eview of Collision R esolution A lgorithm s
3.2.1 Truncated Binary E xponential Backoff
Collision resolution schemes for media access in computer networks have been ex
tensively studied. There are two major classes of algorithms. The first class is
ALOHA-based algorithms, including the truncated binary exponential backoff algo
rithm and the p-persistence algorithm. The second class is the splitting algorithm
such as the tree walk and the n-axy tree. DOCSIS adopts a simple contention reso
lution algorithm, known as the truncated binary exponential backoff algorithm. It
is reviewed in this section.
The format of the bandwidth management protocol data unit (PDU) is shown
in Fig. 3.1. This algorithm requires two parameters for operation: the data backoff
start and the data backoff end. These two parameters are used to indicate the initial
and the maximum backoff window size, respectively. They are set by the headend
39
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A llocation
Ranging
Ranging
Data
Data
IEj
IEn
start
backoff
backoff
backoff
tim e
start
end
end
SID IUC O ffset
Figure 3.1: The format of the DOCSIS bandwidth management protocol data unit
(PDU).
and sent as part of the bandwidth allocation map. To begin a request, the cable
modem sets its window to the size of the data backoff start. Then, it selects a random
value within the window range. Once a value is chosen, the cable modem has to
defer for this number of allocation request minislots before it sends its request. If a
cable modem has sent its request but does not receive a response before a timeout
value, it assumes that a collision occurs. In this situation, the cable modem increases
its window size by a factor of 2, as long as that size is less than the data backoff
end. It then retries the request using the new window value. This process continues
for a maximum of sixteen tries. After sixteen tries, the cable modem discards the
request and starts a new request. Note that the headend can make this collision
resolution process look like that of Ethernet by setting the data backoff start to 0
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CRE
Ret uest arrives
'ait for allocation*
m a p y
x
Grant
(cont ;ntion
succe sses)
Timeout (cc intention failc d)
num ber of retries >16
New request: set initial backoff window size to 2D B : 1
select a number w ithin the window range
( Wait for request
^^opportunities Timeout (c ontention failed)
number of retries <16
new window size =
Send the request
(2*old win low size, 2D B E )
Wait for gr
Figure 3.2: The state machine of CRE in the DOCSIS cable modem,
and the data backoff end to 1024. Fig. 3.2 shows the state machine of a collision
resolution engine (CRE) in a DOCSIS cable modem.
3.2.2 P -p ersisten ce A lgorithm
The IEEE 802.14 draft, which is the HFC network protocol proposed by a group of
researchers under IEEE, uses the p-persistence algorithm of asynchronous collision
resolution in the ranging process and synchronous collision resolution in the process
of contending request minislots. In p-persistence, the headend chooses a value p,
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0 < p < 1, and broadcasts it to all cable modems. When a cable modem attempts
to transmit data, it randomly generates a number between zero and one. If this
number is smaller than p, the cable modem transmits. Otherwise, it must defer
transmission. Hence, collision requests are retransmitted with a probability p. This
process is repeated until a request is successfully received at the headend.
One feature of this algorithm is that whether or not to transmit is independent
of previous attempts. The headend and cable modems do not need to store state
information. The efficiency of this algorithm depends on the value of p. Sala, Limb
and Khaunte [19] proved that the optimal value of p is equal to where N is the
number of cable modems involved in the collision resolution process. One way to
optimize this algorithm is to adjust p every time there is a successful transmission.
This is referred to as adjusted p-persistence. Therefore, the value of p in the adjusted
p-persistence algorithm is chosen to be p = where n follows the sequence N ,N —
1. N — 2, • • ■ , 2,1. This means that each time the headend successfully receives a
request, the number of stations involved in collision is decreased and the headend
announces the new p value via the downstream channel.
3.2.3 P seu d o -B ay esian A lgorithm
In real world applications, the exact number of contending cable modems is unknown
and an estimator is required. We review the pseudo-Bayesian estimator proposed
by Rivest [18] for the slotted aloha system in this section.
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
This adjusted p-persistence method differs from conventional Aloha in that new
arrivals are not distinguished from old contenders and, hence, p-persistence is applied
to the first attempt as well as subsequent attempts. The pseudo-Bayesian algorithm
employs the feedback information, which is the result of previous attempts, to esti
mate the current number of contending cable modems. Binary feedback information,
in form of collision/ non-collision, is used. A non-collision state includes a success
ful transmission as well as an empty mini-slot. Usually, the estimator operates for
each contention minislot but not during data transmissions. The estimate N for the
number N of contending cable modems in the (i + l)th contention minislot is given
in terms of the estimate in the i-th contention minislot by
Ni+i = max(A, JY j + A — 1),
where A is the arrival rate of new data. Thus, the estimation of p for the (i + l)th
contention minislot becomes
1
P i+ 1 =
Ni+1
Even though A is unknown, it could be estimated. It needs an accurate estimate
of Na only when the system is near its full load. This is because that there are
many empty slots to be used for transmission at lower data rates. Thus, we may
assume that A has a maximum value of \ requests per contention minislot, which is
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the maximum arrival rate that can be handled by the Aloha channel. The resulting
equations are then simple to evaluate either at the headend or in the cable modems.
The pseudo-Bayesian algorithm can be implemented in either a distributed or
centralized way. In the centralized case, the headend estimates the p value using the
upstream feedback and broadcasts it in the downstream channel. In the distributed
case, each cable modem individually estimates the value of p by monitoring the
feedback on the downstream channel. In both cases, the estimated p value used by
the cable modems lags one round trip time behind the estimation instance so that
both implementations yield identical results. The maximum value of p is set to one
when there are multiple contending cable modems. A maximum value for estimated
p is also set to avoid the transient blocking problem for 1-persistent collisions.
Salles and Gondim [20] proposed a multiple-priority pseudo-Bayesian algorithm
that is modified from the pseudo-Bayesian algorithm. The idea of this algorithms is
that, for each traffic, a permission probability is transmitted by the headend. A bi
nary random generator, which is maintained by each cable modem, issues permission
based on that value. The probability varies dynamically according to the algorithm.
Priority is considered by choosing adequate permission probability values, so that a
higher priority traffic has a higher probability of getting permission than the lower
priority traffic users. When users of different priorities start to contend, the highest
priority traffic has the best chance to get permission due to their high channel access
probability, yielding lower access delay.
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.2.4 E x te n d e d D i s t r i b u te d Q ueue R a n d o m A c c e ss P ro to c o l
Wu and Campbell [24] described a scheme called the Extended Distributed Queue
Random Access Protocol (XDQRAP), which is an extension of the Distributed
Queue Random Access Protocol (DQRAP). DQRAP transfers the concept of dis
tributed queueing used by DQDB into a multiple access protocol suitable for the
tree-branch topology of a typical CATV network.
In DQRAP, all existing tree protocols, including those utilizing minislots use a
single global queue implicitly or directly maintained by all active stations. A second
global queue is used exclusively to resolve collisions in minislots. The addition of
the second queue along with a simple set of rules enables DQRAP to provide close
to ideal performance with just three control minislots.
At any instance, the first global queue contains the count of the number of users
waiting to transmit data while the second global queue contains a count of the sets
of users (if any) that have collided. In practice, the two global queues are coun
ters maintained by each user. Active users also maintain a pointer indicating their
position in the queue. The two queues decouple data transmission and contention
resolution operations so that they proceed in parallel.
When the two global queues for data, transmission and collision resolution are
empty, all newly ready stations simultaneously transmit in a control minislot as well
as a data slot. Statistically, this is the most common condition under a light loading
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
condition and, consequently, immediate access to networks is provided. The con
tention resolution queue acts as a gateway through which users wishing to transmit
must pass to get a . place in the data transmission queue. If the queue is empty,
passage through the gateway is instantaneous.
XDQRAP modifies DQRAP by two aspects. The first one is that the size of
control minislots is extended. The second one is that the immediate access feature is
dropped. XDQRAP increases the size of the control minislots to include destination
and reservation fields. The address field in the control minislot along with the
length field containing the number of slots required alerts the destination to the
source of a multiple slot message. Upon receiving the feedback from the successful
transmission of a control minislot, all users increment their copies of TQ, which is
the counter representing the number of slots required for pending transmission. The
source and destination users calculate the slot number to start transmitting and
reading, respectively, by using the value of TQ before incrementing. XDQRAP is a
reservation protocol. High priority messages can preempt transmission of multiple
slot messages.
The disadvantage of XDQRAP is that it only provides access for two priorities
with a fixed frame format while DOCSIS can support eight priorities and a dynamic
frame foramt. Consequently, XDQRAP is not suitable for the DOCSIS protocol.
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.2.5 N -A R Y T R E E A LG O R ITH M
This algorithm is used in IEEE 802.14 in association with the p-persistence algo
rithm. In this algorithm, each cable modem maintains a counter. The counter value
indicates how many slots the cable modem has to let pass before its next transmis
sion. If the counter is equal to zero, the cable modem is permitted to transmit in
the next slot. Depending on the result of the contention slot, the counter of each
cable modem is adjusted as follows.
• Collision:
— If the cable modem is involved in this contention slot, counter = random[0,
! ] •
— If the cable modem is not involved, counter = counter + (n — 1).
— Note that n is the branching degree of the n-ary tree algorithm.
• No collision (success or idle):
— If the cable modem is involved in this contention minislot, it has success
fully transmitted its message.
— If the cable modem is not involved, counter = counter-1.
One can imagine that there is a virtual stack. The above rule means that the
stations involved in a contention should randomly select a position in the virtual
stack from zero to n -1 . The other cable modems that are already in the stack (i.e.
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
not involved in that contention) lower their positions by one if no collision occurs
and raise their positions by n — 1 if a collision occurs. The n-ary tree algorithm can
be divided into the following two cases.
• Blocking. There are some implementations that block newcomers from joining
the collision resolution. That means newcomers are not allowed to transmit
their requests until the current contention is fully resolved. Newcomers can
only join a new contention when the virtual stack is empty. These are usually
referred to as the tree-search algorithms.
• Non-blocking. In non-bio eking implementations, each newcomer has a counter
value zero. Thus, it can join the collision resolution process any time, even
when the collision resolution is ongoing. These are usually referred to as the
stack algorithms.
3.3 A novel p rio rity system for D O C SIS protocol
3.3.1 M otivation and problem description
In this section, we investigate the ability of the DOCSIS protocol to provide preemp
tive priority access to users. An effective priority mechanism is needed to provide
QoS in CATV networks for services such as video on demand, interactive computer
games, video telephony, and so on. Although the current MCNS DOCSIS proto
col supports a set of upstream flow scheduling services that HE can offer to cable
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
modems, some problems still exist. First, DOCSIS is intentionally support eight
priority traffic class. But DOCSIS protocol does not mention how to support such a
priority system. DOCSIS does not provide a mechanism to give high-priority cable
modems immediate access to the channel, nor does it separate and resolve collisions
in a priority order. Second, The request delay which is the time it takes a cable
modem to send a successful request to headend must be kept as low as possible for
a high priority traffic flow even during a high contention period. The time spent in
the contention process includes that of collisions, retransmissions, etc. During con
tention, DOSCIS treats all cable modems equally irrespective of their traffic priority
and the newcomers of high priority traffic can easily be blocked for more periods
of time. This will result in a large delay for high priority traffic. If a high priority
request is blocked from accessing the channel or suffers a high number of collisions
from a lower priority flow, it cannot have low access delays even we use a preemptive
scheduling algorithm. Priority mechanisms have been implemented in other MAC
protocols, such as DQDB and the token ring. But these priority mechanisms which
are collision-free are not suitable for the contention-based CATV networks. In the
following section, we will present some priority access schemes and examine why
they are not suita,ble for DOCSIS protocol.
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.3.2 P ro p o se d P rio rity S y stem
To solve these problems, we introduce a scheme that supports multi-priority access
system for DOCSIS. In our scheme, higher priority CMs use the contention slots
assigned to them according to their priority for initial access as well as for retrans
missions. A CM with a new request waits for a group of contention slots (i.e.,
window) with a priority that matches its own priority, and transmits the request
with probability 1 when a group with matching priority becomes available. The sta
tion randomly selects a contention slot in the window. As described in Section 2, the
DOCSIS protocol uses binary exponential backoff to resolve collisions. We propose
modifications to the collision resolution backoff scheme by giving a different backoff
value to CMs of high priority. The backoff value in our scheme is equal to the number
of contention slots reserved for high priority CMs. Besides, the backoff value must
be properly selected according to traffic conditions, in order to achieve an optimal
operation,. In particular, the fact that the optimal value of backoff depends on the
number of contending CMs suggests that the binary exponential backoff scheme can
be improved by dynamically selecting the backoff value according to the estimate
of the number of the contending CMs based on the measurements of the channel
activity performed by HE. The details of the dynamic backoff window scheme is
described below.
Consider that n greedy stations (i.e. stations that always have packets to trans
mit) are contending for requested minislots. For a fixed contention window W, the
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
backoff value b(t) is randomly chosen in the range (0, W -l). The b(t) is decremented
in each slot and can be modeled by the following Markov Chain,
P r W i + 1) = - P T {m = k + F r m = °} , 0 < k < W - l (3.1)
Pr{ii(t + 1) = W - 1} = , k = W - 1, (3.2)
where accounts for the fact that after one transmission, the new backoff
value is uniformly chosen in the range (0, W — 1). Pr{b(t) = k + 1} corresponds to
the decrement of the backoff value at each time slot. The steady state probability
Pr{b(t) = k} is given by
lim P r {6(t) = t } =
A station transmits in a slot with probability p0 — ^r^j- The probability ps
that a transmission is successful is given by the probability that exactly one station
transmits on the channel, i.e.
ps = np0(l - p o ) n_1- (3.3)
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
To optimize ps, we set ^ = 0 and assume (1 — po)n — 1 — np0 . Then we obtain
W = 2n — 1. If c is the number of collisions observed by the HE over W requested
minislots, the collision probability pc is given by
^ 7 = [1 - (1 ~Po)n} -n p o l - p 0)n_1- (3-4)
From (3.4), we obtain
. i + yi+c(w + iy/w
n = ------- , (3.5)
where n is the estimate of n based on the number of collisions. From equation we
obtain, W(t + 1) = 2h — 1, where W (t + 1) is the estimate of backoff window based
on the number of collisions.
The evaluation of the estimated W (t + 1) , from the single value of h, is criti
cal since each is computed on the basis of the current measurement of c (collision
numbers), instead of its mean value. In order to provide a smoother behavior of the
estimate, we weight the current estimate n with its previous values, by using the
following weighting function:
h(t + ! ) = (! — a)n(t) + ah(t + 1), (3.6)
where a is a constant value (0 < a < 1), and n is the value used to compute adaptive
backoff window W for each priority. In the above linear weighting equation, all past
values are considered, but the more distant ones have less weight.
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The extended backoff window algorithm for multiple traffic classes is described
as follows.
• CS: total number of contention minislots in each frame
• CSf. contention minislots assigned to class i traffic
• Wj, : estimate of backoff window for class i traffic, assume there are M traffic
class
• G t : the number of guaranteed CS for each priority. This value is set by network
operator to provide a minimum service access
If (CS > E,=i Wi) Then
st = wt(cs - EtU m /E ti m
c s = m + ^
Else
S = C S - £;=i Gi
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CSM = Gm + min(S, W M - GM )
If (S < WM - Gm) Then
CSi = Gi, for i = 1, 2, M — 1
Else
CSm-i = G jf-i + min(S — (Wm ~ Gm), Wm-i — Gm-\)
If (5 - (WM - Gm) < WM -i - G ^ ) Then
CSi = Gi, fori = 1, 2 , M — 1
repeat this procedure until the lowest priority class
The HE scheduling algorithm is not specified in the DOCSIS specifications be
cause they are considered as implementation details. We employ the multiple priority-
queues scheme in this paper where each CM is identified by its priority. The DOCSIS
protocol specifies eight priorities. The multiple priority-queues scheme consists of
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Jiequest CM
HE
Priority 1
CM
CM
Priority 2
CM
CM Priority M
<----
CM
Figure 3.3: Priority scheduler in DOCSIS,
different queues at the HE for different priority CMs. Each time a new request
arrives at the HE, it is queued in the corresponding queue. When the time comes
for the next MAP message, HE computes a horizon of all the events in time order
that can be scheduled and then the requests are served in priority order. Requests
are granted completely or not at all. When a CM has data to transmit, it sends its
request according to the proposed dynamic backoff window scheme and waits for the
allocation of requested bandwidth in a subsequent MAP. If the request is successful,
then the CM will be notified with the minislot number in which it can start trans
mission and the number of minislots assigned to it, otherwise the CM must repeat
its request attempt based on the proposed adaptive backoff window scheme. Fig.
3.3 shows the priority scheduler in DOCSIS environment.
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
MAC Management Message Header
Upstream
UCD Number of elements
Priority i
channel
count for priority i
ID
Allocation start time
ACK time
Ranging Ranging Data backoff Data backoff
backoff backoff start for priority end for priority
start end 1 1
MAP Information Elements
Figure 3.4: The modified allocation MAP.
In order to implement the proposed scheme, we now discuss the modifications
on the current DOSCIS protocol and show that our scheme can be easily supported
by the standard. Reserved fields or user defined element types are used to integrate
our scheme. We point out two main issues in the proposed priority scheme. The
first is the way to provide a station to request a priority level during registration.
We use a idea which is similar to service negotiation during connection setup in
ATM networks. When a station enter the network, after acquiring synchronization
and completes the ranging and power leveling step, it sends a registration request
message to the HE. The DOCSIS protocol supports several user defined classes
of registration request message that allows the station to request different types
of service. We use one class type for requesting a priority service. The HE can
associate a priority level with a station according to the priority value specified
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
in the registration request message. The second issue is to mark the contention
slot with different priorities to perform the priority contention slot described in the
above. The HE marks the contention slots with a priority value as shown in Fig
3.4. For each group of contention slots the HE specifies a priority value, start time
of the group in the next frame and ACK for the contention slots transmitted in the
previous frame. The fields of Data Backoff Start and Data Backoff End are used to
specify the backoff value for each priority group. When a CM enters the network,
it sends a registration request message to the HE. In the DOCSIS protocol, several
user defined classes of registration request are supported to allow user to request
different types of services. We use one class to request the priority service. Thus,
the HE can associate a CM with a priority according to the priority value contained
in the registration request message.
3.4 S im ulation R esu lts an d D iscussions
Table 3.1: Simulation parameters
upstream data rate 2.56 Mbps
number of contention slots/frame 32
number of high priority CMs 10
number of medium priority CMs 20
number of low priority CMs 30
minislot / contention slot size 16 bytes
frame size 500 minislots
number of contention slots guaranteed per priority Gh = 2,Gm = Gi = l
simulation run 60 sec
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.2: Traffic parameters
high priority (ON/ OFF source) mean off time — mean on time = 1 sec,
average bit rate = 80 kbps
medium priority (Poisson source) arrival rate = 50 kbps
low priority (Poisson source) arrival rate = 30 kbps
Recently, Cablelabs and OPNET have jointly developed a common simulation
framework (CSF) for simulation studies related to DOCSIS by using the modeler
simulation tool in OPNET. We modified this model to meet our needs. A summary
of the simulation parameters is shown in Table 3.1. In the simulation, the network
consists of 60 CMs that are divided into three classes of priority. 10 CMs generates
on-off (real time) traffics that are assigned as high priority. 20 CMs generates non-
real time traffic (Poisson source) that are assigned as medium priority. The other 30
CMs also generate non-real-time traffics (Poisson source), which are assigned as low
priority. The traffic parameters are summarized in Table 3.2. We present simulation
results to demonstrate the effectiveness of the priority system based on the access
delay and the throughput measures below. Note that the access delay is the time
it takes a packet to reach the HE from the time the packet arrives at its CM. The
main objective of our experiment is to show that a priority system is needed for the
DOCSIS specifications and to evaluate the performance of the proposed scheme.
First, the mean access delay for the current DOCSIS system that does not employ
any priority scheme is plot in Fig. 3.5 Then, Fig. 3.6, Fig. 3.7 and Fig. 3.8 show the
access delay of high priority , medium priority and low priority CMs in the proposed
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
priority system, respectively. Here, the X-axis represents the simulation time in
seconds and the Y-axis represents access delay time in ms. Observing above figures,
we obtain some important results: First, due to the nature of the binary exponential
backoff scheme, the non-priority system cannot guarantee low delay or delay bound
for any type of traffic. Second, in our proposed priority system, although the low
priority access delay is higher than original DOCSIS system, the high and medium
priority access delay are relatively flat. As expected, the high priority traffic has
lowest access delay, then the medium priority and the low priority has the highest
access delay. Also, the high priority access delay is relatively flat and remains almost
constant even at high traffic loads. This is because the high priority traffic is assigned
contention slots first and has a higher chance to obtain contention slots. The low
priority traffic uses the remaining contention slots and is treated as the best effort
service, thus suffer higher access delay. The throughput performance of a priority
system is shown in Fig 3.9, where the X-axis represents the simulation time in
seconds (totally 60sec) and the Y-axis represents the throughput in kbps. From Fig.
3.9, we see that the high priority traffic throughput is close to 800kbps that is the
total high priority traffic load. The similar observation can be found in medium
priority. This is because the high priority traffic suffers few collisions so that the
HE can successfully receive most of their bandwidth requests. Again, these results
support our conclusion drawn based on the above figures.
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In this chapter, we examined a preemptive priority scheme for transmissions in
the upstream channel of an HFC network by considering a generalized algorithm
and identifying the changes needed in the DOCSIS specification. We proposed a
dynamic backoff window scheme that dynamically adjusts the backoff window sizes
for different priority traffics and employs the scheduling algorithm of multiple priority
queues. Simulation results show that the proposed priority mechanism performs
well and meets the requirements for different traffic types. Efficiency is obtained
by using priority ordering during contention access and scheduling. Finally, results
obtained demonstrate the advantage of using such a priority system for improving
the performance of the DOCSIS protocol. In the future, we will investigate other
traffic scheduling algorithms that meet the QoS requirements for different traffic
types and compare their performances with the scheduling algorithm of multiple
priority queues.
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Simulation
Figure 3.5: The access delay of a non-priority system.
I T
"
c T
ftt
$m W M
r n m m m m m i i u nf
10. ac. ‘ « lr <1 . gUi
Si nulam'ii .:m e Cei)
Figure 3.6: The access delay of the high priority.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Simulation time (sec)
Figure 3.7: The access delay of the medium priority.
Figure 3.8: The access delay of the low priority.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1000
8 0 0
6 0 0
4 0 0
m e d iu m p rio rity
h ig h p rio rity
lo w p rio rity
200
0
S im u la tio n tim e (se c )
Figure 3.9: The throughput performance of the proposed priority system.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C h a p ter 4
Service Scheduling and B andw idth
A llocation for DO CSIS 1*1
4.1 Introduction
In this chapter, we propose a new service scheduling mechanism and a bandwidth
allocation scheme to support multimedia traffic over DOCSIS-compliant cable net
works. The primary goal is to improve the transmission of real-time variable bit
rate (VBR) traffic in terms of throughput and delay under the current DOCSIS
specifications. To investigate the proposed scheme’ s capability of supporting inte
grated services, we combine the transmission of VBR traffic with constant bit rate
(CBR) traffic and non-real-time traffic in simulation. To demonstrate the perfor
mance, we compare the result of the proposed scheme with that of a simple multiple
priority scheme. It is shown via simulation that the proposed method provides a
significant amount of improvement over existing DOCSIS QoS scheduling services.
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Besides, m athem atical analysis of the system performance is also conducted. Since
the simulation model is too difficult to analyze, we only consider two different classes
of voice services in mathematical analysis. We develop an elaborate discrete-time
Markov chain model to obtain the voice packet dropping probability and bandwidth
utilization. Numerical examples are presented to demonstrate the performance of
the system.
4.2 DOCSIS v l . l Upstream Scheduling Services
DOCSIS 1.0 that is the earlier version protocol [5]only supports one QoS class, i.e.
the ” best effort” for data transmission in the upstream direction. In DOCSIS 1.1
[4], QoS is supported in both upstream and downstream traffics by specifying more
classes of service flows with each identifying a particular QoS requirement. Usually,
data packets entering the HFC network are classified into service flows based on a
set of matching traffic criteria. Then, these classified data packets are associated
with a particular QoS level based on the QoS parameters of that particular service
flow. QoS may be guaranteed by shaping, policing, and/or prioritizing data packets
at both CM and HE.
The QoS parameters include upstream service scheduling class parameters and
some additional traffic parameters. The parameters defined by the service schedul
ing classes consist of bandwidth and delay guarantees. The bandwidth scheduling
algorithm implemented at HE is responsible for allocating the upstream bandwidth
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for a , particular upstream service flow according to the QoS parameters to meet the
service requirements.
To meet QoS requirements of growing real time applications, DOCSIS defines
five upstream services includeing UGS (Unsolicited Grant Service), UGS-AD (Un
solicited Grant Service with Activity Detection), rtPS (Real Time Polling Service) ,
nrtPS (Non Real Time Polling Service), and BE (Best Effort). Except for BE, these
services avoid request contention by an unsolicited grant or polling. Unsolicited
grants, which are opportunities for collision-free data transmission periodically is
sued by the CMTS, allow CMs to transmit their PDUs without bandwidth requests.
The polling service provides collision-free request opportunities so that packet access
delay could be guaranteed.
The UGS scheduling service is intended to support CBR traffic (such as audio
streams)transmission over the upstream channel. The HE provides fixed size data
grants at periodic intervals to the UGS flows after the CM specifies the QoS parame
ters during service flow registration. Since the bandwidth is reserved without request
contention, the UGS can guarantee both bandwidth and data access delay. The QoS
parameters for the UGS are: the Nominal Grant Interval specified according to the
packet inter-arrival time of the flow. The Unsolicited Grant Size represents packet
size, and the Tolerated Grant Jitter is the maximum variance of packet access delay
the flow can tolerate. In DOCSIS, the purpose of UGS is defined as ”to reserve
specific upstream transmission opportunity for specific real-time traffic flows”. For
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
IMM i = Map message | | = Data packet
►
Time axis
MM2 !
E 0 0 0
Incoming
p a c k e t s 0 0 0
Figure 4.1: The UGS scheduling service time diagram,
example, a 64 kbps voice application (CBR) can be transmitted using a UGS service
flow with 10 ms nominal grant interval and 80 bytes grant size. We show an example
of the UGS scheduling service execution in Fig. 4.1.
The video stream is not suitable to be served by the UGS service since it is
VBR and bursty. Usually, it is impossible to know the bandwidth requirement for
the UGS grant interval during the call connection time due to the VBR nature of
the video stream. If a video stream is served with its peak data rate-based UGS
allocation, this may waste the network bandwidth. However, if a video stream
is allocated with the average bit rate CBR bandwidth, it may suffer undesirable
delay, jitter and also result in packet loss. Therefore, DOCSIS defines the real-time
polling service class (rtPS) to serve real-time VBR traffic. Here, the definition of
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
jMM| = Map message j |= Data packet |j = Request message
Time axis
MMl MM2 MM3
t t t t
/ \
I \
I
/ I
I
I !
CM
□ 0
Incoming [2]
packets
L IJ
Figure 4.2: The rtPS scheduling service time diagram.
the rtPS service class is ”to reserve upstream transmission opportunities for real
time traffic flows. These service flows receive periodic transmission opportunities
regardless of network congestion, but these service flows release their transmission
opportunities to other service flows when inactive.” The idea of rtPS service class is
as follows. A service flow with the rtPS scheduling class is allocated periodic request
opportunities in which upstream bandwidth requests can be transmitted. Basically,
each CM gets polled with request to find out about the instantaneous upstream
bandwidth requirements for its data. A scheduling diagram of the rtPS service class
is shown in Fig. 4.2. Note that the CM transmits the first unit of data during
the second MAP interval, since it waits for HE to allocate the upstream bandwidth
based on its previous request.
68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The HE polls an rtPS service flow by allocating request opportunities. The CM
uses these request opportunities to send bandwidth requests to HE to ask data trans
mission. For example, a video stream with an average bit rate of 6.4 Mbps can be
transmitted in an rtPS service flow with request opportunities granted every 10 ms.
In this case, the CM will request data transmission opportunity for 8000 bytes on
average. The disadvantage of the rtPS service class is that an CM may experience
a long delay since each CM has to wait for the HE to poll for the bandwidth re
quest even with many packet arriving already. This high delay nature of the rtPS
service can be improved if one can eliminate such a time lag. In fact, this problem
can be solved by employing the piggyback mechanism that piggybacks bandwidth
requests within packets transmitted to HE. However, DOCSIS 1.1 does not adopt
the piggyback procedure as of today.
For some audio compression technologies such as ITU G.728, silence suppression
is supported, so there is no data during silence. To efficiently utilize the band
width, DOCSIS defines Unsolicited Grant Service with Activity Detection (UGS-
AD), which is with a combination of UGS and rtPS. Non real time polling service
(nrtPS) closely resembles rtPS, but has a longer polling interval, around 1 sec or
more. Therefore, nrtPS flow can use both unicast request opportunities and broad
cast request opportunities. The conventional service provided in the previous version
of DOCSIS is best effort (BE). For this service, the CM generally uses contention
request opportunities for bandwidth demands.
69
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.3 Proposed Schem es
4 .3.1 M o tiv atio n
From the discussion of last section, it is apparent that the UGS service can provide
low latency for real time traffic. UGS service is more suitable for the CBR traf
fic as it can provide fixed upstream bandwidth which is reserved during call-setup
phase. However, for real-time VBR traffic, the UGS service will under-utilize the
network bandwidth and also the fixed bandwidth allocation cannot keep up with the
unexpected increase in the instantaneous bandwidth requirement in case of sudden
increase in the bandwidth. On the other hand, the rtPS service has higher band
width utilization for real-time VBR traffic but results in higher delay. For rtPS
scheduling, each data packet must wait for a request opportunity in order for the
CM to request upstream bandwidth, and then must wait until the HE responds to
the upstream bandwidth request.
Therefore, we would like to develop a new service class, which provides low
delay close to that of UGS while has channel utilization close to that of rtPS. To
meet these intend features, it is important to eliminate the lengthy delay associated
with the CM waiting for request opportunities,issuing the requests for transmission
to the HE and subsequent waiting by the CM for the data grants, as the current
rtPS service. Piggybacking requests for the VBR portion of the video data on top
of the unsolicited data transmissions by the CM would eliminate the above delay
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
iMMl = Map message I I = Data packet Q - Data packet with
piggyback request
Time axis
MM3 MM4 MM2
Incoming1 —1
packets j ~ 2 ~ !
U J
Figure 4.3: The UGPS scheduling service time diagram.
and should increase channel utilization as well. Based on the above observation
and discussion, we propose a new schedulingThe idea of piggybacking bandwidth
requests for the VBR portion associated with the UGS data transmissions by CM
would eliminate the long delay and also increase bandwidth utilization [3]. For
this, we adopt a new scheduling service class called the unsolicited grant piggyback
request service (UGPS) in the DOCSIS protocol. A timing diagram of the UGPS
service execution by a CM-HE pair is shown in Fig. 4.3.
71
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4 .3 .2 N e w Service C lass and B a n d w id th A llo catio n
A lg o r ith m
The UGPS service reserves a fraction (explained below) of the VBR traffic’s average
bit rate requirement. This reserved part is allocated to the service flow by HE at
periodic time periods and is similar to the UGS service. This part is called the service
unsolicited allocation. The bandwidth requests of the remaining VBR portion of the
traffic are then piggybacked with the packet transmitted in the reserved bandwidth.
The HE will process these piggybacked requests and issue the corresponding data
grants indicated by MAP messages. Whenever more upstream bandwidth is needed,
the CM piggybacks a request in the data grant slot and the HE responds by allocating
an appropriately sized upstream data transmission opportunity.
The idea of UGPS can be further illustrated by the time diagram shown in Fig.
4.3, The first data unit is bigger than the available unsolicited grant. Here, a portion
of packet (la) is transmitted along with a piggybacked request for the size of the
remaining fraction (lb) and (2). In the next MAP message (i.e., MM2), the HE al
locates the unsolicited grant and an additional transmission opportunity in response
to the piggybacked request. The fractions (lb) and (2) are then transmitted in the
unsolicited grant, and (3) are transmitted in the requested grant. To summarize, HE
allocates data grants of a unsolicited grant size with a nominal grant interval between
two successive data grants. Whenever there is more upstream bandwidth needed,
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CM piggybacks a request in the d ata grant slot, and HE responds by allocating an
appropriately sized upstream data transmission capacity.
An example of transm itting a VBR M PEG stream w ith a 6.4 Mbps average bit
rate by using the UGPS follows. The following allocation of service could be used:
• 3.2 Mbits/s using unsolicited grants, at a 20 ms nominal grant interval with an
unsolicited grant size of 8000 bytes,and piggybacked requests for the remaining
VBR portion of data.
Measurement-based provisioning of the unsolicited allocation. It is obvious that
by reserving a bandwidth less than the average bit rate over the unsolicited por
tion of the UGPS, high utilization of the unsolicited portion of the channel can be
assured. Elimination of request opportunities further improves channel utilization.
Meanwhile, low latency can be guaranteed by using piggyback requests for the VBR
portion of the stream instead of requesting bandwidth from the HE during the nom
inal polling intervals. Another question still remains. That is, what should be the
unsolicited portion of UGPS allocation to support high bandwidth utilization and
low latency for real-time VBR traffic? Now, we deal with the problem of deter
mining the size of the unsolicited allocation. We propose a scheme in which HE
dynamically adjusts the unsolicited allocation according to periodic measurements
of two variables: 1) unsolicited bandwidth allocation, and 2) additional amount of
traffic granted by HE in response to the piggybacked requests from CM.
73
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In order to estim ate the dynamic bandw idth requirements of a VBR stream more
accurately, the unsolicited portion of bandw idth is allocated based on the following
equation:
unsolicitedjallocation(n + 1)
= unsolicitedjallocation(n) — unused Jbytes{n)
+piggybackcrequest(n), (4.1)
where n is the index of a MAP cycle, theunusedJyytes{n) is the average unused
bandwidth of the unsolicited allocation portion in the previous N MAP cycles, and
piggybackjrequest(n) is the average value of the piggyback request portion trans
mitted over the previous N MAP cycles. The reason for this estimation strategy
is that real-time VBR data, such as coded MPEG video streams, exhibit strong
long-range dependence and are bursty over multiple time periods. This is so called
” self-similar” traffic. Since there is strong correlation in a VBR source, we use the
previous history to predict the future. The bandwidth reservation in current and
previous scheduling cycles represents the prediction for the next scheduling cycle.
Second, The value of unsolicited jallocation(n + 1) in equation 4.1 may be equal to
zero. In this case, the unsolicited-allocation{n + 1) is set to one time slot. The
reason is that if the and the CM is not allocated any bandwidth in this cycle, then it
must contend for the channel or wait HE to poll it. To avoid the contention and/or
74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
polling delay, we allocate it one time slot. W hen the next bursty traffic comes, it
can piggyback its request.
In each MAP, HE computes the total bandwidth requirement of all VBR streams.
If the sum of unsolicited-allocation (i.e. the left hand side of equation 4.1 ) of
all VBR streams exceeds the channel capacity, we propose a bandwidth allocation
algorithm below.
Initially, the total residual bandwidth, initiaLResBW, is equal to the upstream
bandwidth. The HE computes the initial average residual bandwidth,
initial-Avg-ResBW, which is equal to the initiaLResBW divided by the total
number of VBR streams (K ). If the unsolicitecLallocation is more than
initiaLAvg-ResBW for a VBR, stream, the HE only allocates Avg-ResBW to it.
Otherwise, the unsolicitecLallocation is allocated. In the latter case, the excess
bandwidth allocated to each stream i is computed as the initiaLAvg-ResBW minus
unsolicited-allocation. These excess bandwidths for all streams are added to obtain
a new value of residual bandwidth (ResBW) available with the HE. The Avg-ResBW
is then computed as the ResBW divided by the number of VBR streams (i.e., the
streams which have higher unsolicited-allocation than Avg-ResBW). In the second
iteration, the Avg-ResBW is allocated to each VBR stream in the same way. This
process runs repeatedly until ResBW equals to zero. The detailed algorithm is given
below.
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
F o r (each VBR stream i, i = 1, 2. , K) {
If (end of each MAP cycle) {
measure average unusedJbytes (of previous N MAP cycle)
measure average piggyback crequest (of previous N MAP cycle)
unsolicitedMllocation(n + 1)
= unsolicited jallocation{n) — unused Jbytes{n) + piggyback jrequest(n)
I f (unsolicited-allocation(n + 1) < = 0)
unsolicitedjillocation{n) ~ 1
76
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
}
If (E*Li unsolicited-allocation(i) > channel capacity) {
Step 1. k — K
Set S = 1,2,..., K
Step 2. initiaLResBW = max upstream bandwidth available for VBR streams
initiaLAvg-ResBW = initiaLResBW/ K
For (each VBR stream i) {
allocated-BW(i) = Min(unsolicited-allocation(i), initial-Avg-ResBW)
If (alloca,ted-BW(i) < initiaLAvg-ResBW) {
S = S - i
k = k — 1
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Step 3. ResBW = "}2ies initial.Avg-ResBW — allocated-BW(i)
If (ResBW + 0) {
Avg-ResBW = ResBW/k
For (VBR stream in S) {
available-BW (i) = allocated-BW (i) + Avg-ResBW
allocated-BW (i) = Min(unsolicited-allocation(i), available-BW(i))
If (v,nsolicited-allocation(i) < available-BW (i)) {
S = S - i
78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ResBW = Y^ies initial-Avg-ResBW — allocated-BW(i)
If (ResBW ± 0)
Repeat this process (step 3) until ResBW = 0
4.3.3 S im u latio n re su lts a n d discussion
We have conducted simulations by using the DOCSIS module of the OPNET com
mon simulation framework (CSF) that was jointly developed by CableLabs and
79
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.1: Simulation parameters
upstream data rate 2.56 Mbps
Minislot size 16 bytes
Frame size 200 minislots
CBR arrival rate 60 kbps
VBR average arrival rate 80 kbps
VBR peak arrival rate 120 kbps
ABR peak arrival rate 40 kbps
N (MAP cycle) 5
Simulation run 60 sec
OPNET. We have modified this DOCSIS model to meet the requirements of our ex
periments. We have compared our proposed scheme with the multi-priority scheme
[21]. In the case of multi-priority scheme, UGS (CBR), rtPS (real time VBR) and
Best Effort (non real-time traffic) are assigned priorities in a decreasing order. This
means that the HE first allocates bandwidth to UGS followed by rtPS. The Best
Effort traffic is served at last. The same priority is used in the proposed scheme.
The main difference is that the proposed scheme uses the bandwidth allocation al
gorithm (described in Section III) along with the new service class (i.e. UGPS) that
replaces rtPS in DOCSIS. The cable network is compsed of 10 CMs that transmit
CBR traffic, 20 CMs that transmit VBR traffic, and 20 CMs that transmit ABR
traffic. The simulation parameters are listed in Table 4.1.
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Simulation time (sec)
Figure 4.4: Delay of UGPS and rtPS.
1600
1400
H 1200
C 1000
< g .
i
600 f
400
200
0
4— rtPS
• — U GPS .............. ....... .........
I 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58
Simulation time(sec)
Figure 4.5: The throughput performance of UGPS and rtPS.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
— —
400
150
100
50
0
1 3 5 7 9 11 13 15 17 19 2 ! 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59
Simulation time (sec)
Figure 4,6: The throughput performance of the best effort class.
3 0 0 0 ....
•Proposed Schem e
2 5 0 0 !
• m ulti-p rio rity
2000
< § . 1 500
X$
e
1000
!
5 0 0 '•
1 3 5 7 9 11 13 15 17 19 21 2 3 25 2 7 2 9 31 33 3 5 3 7 3 9 4 1 4 3 45 47 4 9 51 5 3 5 5 5 7 59
Simulation time (sec)
Figure 4.7: The total throughput.
8 2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Fig. 4.4 compares the average access delay of the proposed UGPS and DOCSIS
rtPS. As expected, the proposed service class has lower access delay (about 13-14
ms) than that of rtPS (about 25-30 ms). This is because the proposed service class
uses piggyback instead of using polling, thus the average access delay is significantly
reduced. Fig 4.5 compares the throughput of UGPS and rtPS. The proposed UGPS
has a higher throughput than DOCSIS rtPS. The reason is that the dynamic band
width estimation scheme combined with the information of the piggyback request
helps the HE to estimate more accurate bandwidth demand of CMs. Thus, our new
scheduling service and dynamic bandwidth allocation scheme has a better perfor
mance than the existing DOCSIS standard. Fig. 4.6 shows that the multi-priority
scheme has a higher throughput than the proposed scheme for the best effort service
class. This is because the proposed scheme allocates more bandwidth to VBR traf
fic, thus reducing the transmission rate of the best effort service. However, Fig4.7
shows that the proposed scheme can achieve a higher overall throughput and higher
bandwidth utilization than the multi-priority scheme.
The QoS service classes and their characteristics of DOCSIS 1.1 protocol were
reviewed in this work. To improve the QoS performance of the DOCSIS protocol,
we proposed a novel scheduling service class and a bandwidth allocation scheme to
support real-time VBR traffic transmission. The goal of the proposed scheme is to
achieve higher bandwidth utilization of cable networks and lower latency for the
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
video stream. It was demonstrated by simulation results that the proposed scheme
achieved a better performance in terms of a higher throughput and lower latency
than the current DOCSIS specification
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C hapter 5
E n h an ced C S M A /C A Backoff Schem e for
IE E E 802.11 W ireless LA N S ystem
In this chapter, we focus on the design and analysis of protocols for the MAC
(Medium Access Control) layer of another important shared-medium communication
system using multiple access; namely, the IEEE 802.11 Wireless Local Area Network
(wireless LAN). The IEEE 802.11 standard is first introduced in Section 5.1. Details
of the MAC layer, the backoff mechanism of IEEE 802.11 and its analysis will be
provided in Section 5.2. It is worthwhile to point out that, although the backoff
mechanism used in IEEE 802.11 is similar to that of DOCSIS, some modification
is needed to meet the requirement of the wireless environment. Then, an enhanced
CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance) backoff scheme
is proposed and studied in Section 5.3.
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In wireless LANs, the CSMA/CA protocol supports asynchronous data transfer,
and adopts an acknowledgement mechanism to confirm successful transmissions and
a handshaking mechanism to reduce the number of collisions. In both cases, a
binary exponential backoff mechanism is used. The proposed protocol improves the
exponential backoff scheme by dynamically adjusting the contention window (CW)
around the optimal value. Moreover, an analytical model based on the Markov chain
is developed to analyze the system performance in terms of throughput and delay.
Theoretical and simulation results are presented in Section 5.4 to show the effect of
the proposed backoff mechanism.
5.1 IEEE 802.11 Standard
5.1.1 Introduction
The emergence of portable terminals is accelerating the deployment of wireless net
works, which will play an important role in personal communications systems. A
wireless local area network (Wireless LAN) is a way to connect portable computers
over radio or infrared wireless links in a small area such as an office or a home envi
ronment. Wireless LANs are much flexible to install than wired LANs. To speed up
the design of wireless LANs, the IEEE 802.11 study group proposed an international
standard [12] for wireless LANs. The standard defines detail functions for both the
Media Access Control (MAC) layer and the Physical (PHY) Layer.
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The initial IEEE 802.11 protocol was designed to provide a data rate up to 2
Mbps. Later, the IEEE approved the 802.11b addendum to the original 802.11
specification to allow a data rate up to 11 Mbps [14], The intent of the IEEE
802.11b standard is to support mobility and high data rate multimedia applications
such as video teleconferencing, streaming video, and voice over IP (VoIP). However,
the bandwidth of IEEE 802.lib-based networks is still limited so that it is difficult
to support multiple media applications simultaneously. To support the demand for
mobility, low latency, and high data rate communications, several new wireless LAN
standards have been proposed. The 802.11a standard specifies an operation in the 5
GHz band, utilizes orthogonal frequency division multiplexing (OFDM) in the PHY
layer and provides for data rates ranging from 6 to 54 Mbps [13]. These data rates
are capable of supporting high traffic applications in a mobile, multi-user wireless
LAN environment. It is worthwhile to point out that the MAC layer protocol is
identical for each member of the IEEE 802.11 specification family.
The wireless medium is a shared resource, and users access the channel through
a multiple access scheme. The MAC layer protocol is needed to coordinate commu
nications among users to achieve efficient network management. Besides, the MAC
protocol must provide fairness and robustness to users. There are many proposed
multiple access schemes for wireless LANs, such as the carrier sense multiple ac
cess (CSMA), the polling, and the time division multiple access (TDMA) schemes.
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The IEEE 802.11 standard utilizes the CSMA protocol, which is a member of the
ALOHA family protocol to realize multiple access.
5.1.2 W ireless L A N A rc h ite c tu re
IEEE 802.11 Wireless LANs have two configurations: infra-structured and ad-hoc
wireless LANs. In a typical ad-hoc wireless LAN, stations establish peer-to-peer
communication among themselves independently in their service area. Note that
ad-hoc networks presume a non-fully connected network topology. Infra-structured
wireless LANs establish the communication between stations with the help of an
infrastructure such as a wired or wireless backbone.
An 802.11 network typically consists of Basic Service Sets (BSS) that are inter
connected with a Distribution system (DS). Fig. 5.1 show's such a wireless LAN
architecture. Each BSS consists of mobile stations that are controlled by a single
coordination function to be defined in Section 5.1.3. The geographical area covered
by BSS is called the basic service area (BSA), which is analogous to a cell in the
cellular communication systems. All stations in a BSS can communicate with other
stations in the same BSS directly.
The IEEE 802.11 standard specifies an ad-hoc architecture for a wireless LAN.
An ad-hoc wdreless LAN is characterized by (i) the lack of an access point (AP),
(ii) no functionality to support mobility, and (iii) the support for data transfer only
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Backbone Network (Distribution System
Server
| Workstation
Workstation
Mobile Station Mobile Stktii
Basic Service Set A
Basic Service Set B
Figure 5.1: The wireless LAN architecture.
between stations that belong to the same BSS. Therefore, an ad-hoc wireless LAN is
simply an independent BSS (IBSS) where a station communicates directly with one
or more other stations in the same BSS without sending traffic through a centralized
AP.
Besides the ad-hoc architecture, IEEE 802.11 defines another type of network
configuration known as the infrastructure network. The objective of an infrastruc
ture network is to support the range extension and specific services for users. Sta
tions in a BSS gain access to the DS and to stations in remote BSSs through an
AP. The AP is similar to the base station in cellular systems. The AP is an entity
that implements both 802.11 and DS MAC protocols, and can communicate with
stations in the same BSS to which it belongs and to other APs that are connected
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to the DS. The DS is a backbone network that takes care of MAC service data units
(MSDU) transport.
According to the IEEE 802.11 specification, DS is implemented independently
and could be the Ethernet LAN, the token bus LAN, the token ring LAN, or the
fiber distributed data interface (FDDI) metropolitan area network (MAN). Before a
station can access the wireless channel, it has to be associated with an AP. A station
can only connect to one AP at any given time. The APs support range extension
by connecting to the DS. Multiple BSSs that are integrated together with their APs
connecting to the DS form a Extended Service Set (ESS). The upper layer protocol
views ESS as one large BSS. A network of interconnected BSSs is shown in Fig. 5.1,
in which stations can roam without loss of connectivity between different BSSs that
forms an ESS.
5.1.3 IEEE 802.11 M AC Layer P rotocol
The IEEE 802.11 MAC layer has several functions that include: channel allocation,
protocol data unit (PDU) addressing, frame formatting, error checking, fragmenta
tion and reassembly. Three types of frames are defined: management, control and
data. Stations use the management frame to associate and disassociate with AP,
maintain timing and synchronization, authenticate and de-authenticate with AP.
Control frames including acknowledgements are used for handshaking. Data frames
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
are used to transmit data unit and can be combined with polling and acknowledge
ments.
The basic access method in the 802.11 MAC protocol is the Distributed Coordina
tion Function (DCF), which adopts the Carrier Sense Multiple Access with Collision
Avoidance (CSMA/CA) protocol. In addition to DCF, the 802.11 MAC protocol
also adopts an optional access method known as the Point Coordination Function
(PCF). The PCF is an access method that is similar to polling and uses a point
coordinator located at the AP to determine which station has the right to transmit.
Ad-hoc wireless LANs support only the DCF since there is no AP in these systems.
The DCF can be used to transmit non-real time traffic since it is contention-based.
Time-bounded services such as the real time traffic are implemented by the PCF as
connection-based data transfers. The PCF attempts to provide bounded delay and
delay jitter. Since the PCF is optional, support for time bounded services is also
optional. We will describe the DCF and the PCF in detail below.
5.1.3.1 Distributed Coordination Function (DCF)
The DCF is used to support asynchronous data transfer. According to the 802.11
specification, the DCF operates solely or coexists with the PCF in the infrastruc
ture network, and operates solely in an ad-hoc network. Fig. 5.2 shows the MAC
architecture. The DCF is on top of the PHY layer. In the DCF mode, each station
with queued packets for transmission must contend for access to the media and must
re-contend for all subsequent packets when the current packet is transmitted. The
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Provide contention-free
services
Provide contention
services and
basis of PCF
I
Point coordination
function (PCF)
MAC
layer
Distributed coordination funct
’
ion (DCF)
Physical layer convergence procedure
(PLCP) sublayer
Physical
layer
Physical medium dependent
(PMD) sublayer
Figure 5.2: The IEEE 802.11 MAC architecture.
DCF is based on the CSMA/CA protocol. Note that the wireless terminal cannot
transmit while listen simultaneously. Therefore, CSMA/CD (collision detection) is
not used. The IEEE 802.11 supports two carrier sense mechanisms: the physical
carrier sensing and the virtual carrier sensing. The physical carrier sensing is per
formed by the PHY layer which detects the channel activity using the relative signal
strength from other users or analyzes detected packets to decide the presence of
other users.
The virtual carrier sensing is implemented by sending the MPDU duration in
formation in the header of RTS (request to send), CTS (clear to send) and data
packets. The duration field records the amount of time after the end of the current
packet the media will be used to complete the successful transmission of the data or
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
management packets. Stations in the BSS use the information in the duration field
to adjust their NAV (network allocation vector). The NAV indicates the amount
of time that must elapse until the current transmission is complete and the channel
can be sensed again as idle. The channel is marked as busy if either the physical or
the virtual carrier sensing indicates the channel is busy.
The IEEE 802.11 standard defines the interframe space (IFS) which is imperative
idle periods on the wireless media. IFS can be used to implement priority access
to the channel. Three IFS intervals are defined: SIFS (short IFS), PIFS (point
coordination function IFS), DIFS (DCF-IFS). The SIFS interval is the smallest,
then followed by PIFS and DIFS. Stations only required waiting for SIFS interval
has the highest priority to access the channel.
5.1.3.2 Point Coordination Function (PCF)
The PCF is an optional function that provides contention-free access mode. In
IEEE 802.11, DCF and PCF coexist with a superframe (SF) that is denoted as
the Contention Free Repetition Interval (CFRI). As shown in Fig. 5.3, each CFRI
includes the following two periods: (1) the contention free period (CFP) in which
the PCF protocol is used and (2) the contention period (CP) in which the DCF
protocol is used.
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Super Frame
Contention Free Repetition Interval
Super Frame
lllliku U U M
~ Y"
PCF
Contention Free
Period
~ Y ~
DCF
~Y~
PCF DCF
Contention Period
Figure 5.3: The IEEE 802.11 superframe structure.
The PCF protocol is based on a polling scheme controlled by the Point Coordi
nator (PC). A station with data to transmit in the CFP has to register with the PC
in the CP first. The PC maintains a list for registered stations and polls each station
according to the polling list. Each polled station is allowed to transmit one MPDU.
The destination has to acknowledge the packet right after it successfully receives
the packet. The PC controls the medium by broadcasting a Beacon signal which
contains the CF parameters. The Beacon signal will be issued at the rate specified
in the CF repetition rate. Once a non-PC station receives the Beacon, it will set
its NAV to the maximum duration of the CFP specified in the CF parameters. The
NAV is used for preventing a station from accessing the medium during CFP. Once
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
all stations in the polling list have been polled, the PC will send a CF-end signal to
indicate the end of the CF period and enter the contention period.
5.2 IE E E 802.11 Backoff M echanism an d Its
A nalysis
In this section, we present the IEEE 802.11 backoff mechanism, which plays a fun
damental role in the channel access scheme used in the 802.11 standard. In order
to understand the performance of the IEEE 802.11 backoff mechanism, we propose
a Markov model for analysis, which is modified from the work in [1]. We compute
the saturation throughput and delay based on the proposed Mrkov model. To verify
the correctness of the model, we use OPNET in computer simulation and compare
results obtained from theory and simulation. A deeper understanding of the 802.11
backoff mechanism can be obtained accordingly.
5.2.1 IEEE 802.11 Backoff M echanism
In the IEEE 802.11 standard, the DCF has two operational modes: the basic access
mode and the RTS/CTS access mode. We will discuss them below.
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.2.1.1 Basic Access Mode
In the basic access mode of the DCF operation, a station has to sense the status
of the channel before initiating a transmission to determine if another station is
transmitting. The station continues with its transmission if the channel is found
to be idle for an interval that exceeds the DIFS. If the channel is busy, the station
has to wait until the channel becomes idle for a DIFS interval, then generates a
random backoff period for an additional deferral time before transmitting. The
random backoff procedure is to minimize the number of collisions during contention.
This backoff period is used to initialize the backoff timer. The backoff timer is
decremented only when the medium is idle. It is frozen when the medium is busy.
After a busy period, the decrement of the backoff timer resumes after the medium
has been idle longer than a DIFS interval. A station transmits its data only when
the backoff timer reaches zero. The backoff interval is chosen to be
Backoff-Time — INTiCW x Random(0,1)) x SlotJTime, (5.1)
where CW means the contention window, which is an integer between CWmin and
CWmax, and where CWmin and CWmax are system parameters, Random(0,1) is
an uniformly distributed pseudo-random number between 0 and 1, and Slot-Time is
the transmission time of one packet.
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The contention window takes an initial value of CWmin for each packet queued
for transmission. To reduce the probability of collisions, stations double the size of
the CW until CW reaches CWmax after each unsuccessful transmission. The CW
remains at the value of CWmax for the remaining retries.
If the packet is successfully received by the destination, the receiver has to send
an acknowledgement after an SIFS interval immediately following the reception of
the packet. Note that the receiver transmits the acknowledgement without sensing
the status of the channel. If an acknowledgement is not received, the transmitter
assumes that the packet is lost and schedules a retransmission immediately. An
example of the basic access mode is shown in Fig. 5.4
5.2.1.2 R TS/C TS Access Mode
The DCF provides another transmitting scheme using the short Request-To-Send
(RTS) and Clear-To-Send (CTS) packets. In the basic access method, a packet
could be corrupted due to stations that are hidden from the transmitter. The hid
den terminal phenomenon results in (i) an increased collision probability and (ii) the
transmission time wasted due to collision. This wasted time is significant since the
transmitter does not know packet collision until time out to wait for the acknowl
edgement. The RTS/CTS scheme tries to reserve the channel for the time duration
needed to transfer the packet.
97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
DIFS DIFS
station A
station B U -
station E
Differ
D itfe i
- I . -
station C U
station D
= selected backoff = remaining backoff
Figure 5.4: An example of the backoff procedure for the basic access mode.
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
source
station
IRTS
data
SIFS SIFS
destination
station
other
station
CTS
ACK
t
DIFS
NAV(RTS)
NAV(CTS)
NAV(data)
contention
window
defer access
Figure 5.5: A typical example of the RTS/ CTS access mode.
When using the RTS/CTS mechanism, the transmitter sends an RTS packet
after the channel has been idle for an interval exceeding DIFS. The receiver responds
with a CTS packet which can be transmitted after the channel has been idle for an
interval exceeding SIFS. After the successful exchange of RTS and CTS packets,
the transmitter transmits the packet after waiting for a SIFS interval. If CTS is
not received within a predetermined interval, the RTS is retransmitted following the
backoff rules as discussed in the basic access mode. The RTS/ CTS access mode
is illustrated in Fig. 5.5. To increase the probability of successful transmission,
the RTS/ CTS mode utilizes the virtual carrier sensing mechanism. However, this
increases the overhead of exchanging RTS and CTS packets.
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.2.2 M ark o v m odel a n d A nalysis
By modifying the Markov models given in Bianchi’s work [1] and Ziouva’s work [25],
we propose a new Markov model to analyze the performance of the 802.11 DCF
protocol in this section. We assume a fixed number of stations, each of which always
has queued packets for transmission, i.e. the transmission queue of each station is
assumed to be nonempty.
Consider a fixed number of contending stations. In the saturation condition,
each station always has packets waiting for transmission, after the completion of
each successful transmission. Let b(t) be the stochastic process representing the
random backoff window value for a given station at slot time t and s(t) the stochastic
process representing the backoff stage i for the same station at time t. Thus, the
two-dimensional process, {s(t),b(t)}, is a discrete-time Markov chain. Furthermore,
it is clear from the backoff algorithm described above that the probability p for
transmitted packets to collide as well as the probability pb for the channel to be in
a busy status are independent of the backoff procedure.
The discrete-time Markov chain model corresponding to the above problem is
shown in Fig. 5.6. To analyze this Markov model, we have to find the steady state
probability for a station to be in state {i, k}. Let A* = limt_to o Pr{s(f) = i, b(t) = k}
100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
be the stationary probability of the Markov model, where 0 < i < m , 0 < k < W— l.
In the steady state, the following system of equations holds
bi,o — p h - 1,0 ;
ho = plbo.o, 0 < i < m.
(5.2)
(5.3)
Besides, we have the following relations based on the regularities of the Markov
chain:
& i,k
Wt - k
1 ~ P h
1 - P b ^ j = 0 ° 3 , 0
Wt
if i = 0
if 0 < i < m
(5.4)
1 ^pb (pm—1,0 ~b fcm,o) If ^ ^
Let Wi — 2lW and W = CWmin + 1. We have
m Wi-l
1 = E E hk- (5.5)
j= 0 fc=0
By solving (5.5), we obtain
6 2 ( l - 2 p ) ( l - p ) ( l - p » )
0 ,0 pW[ 1 - (2p)™ ) + (1 - 2p)(W + 1)' 1 J
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(1-P)/W ,
1-p,
1 -p , 1-p,
0, W0-l
PA V
1-P, 1-P, 1-P,
i, W,-l
PAV,
1-P, 1-P, 1-P, 1-P,
m , 0
P b P b
PA V ,
Figure 5.6: The Markov model for the IEEE 802.11 backoff scheme.
102
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
We can compute bhk if the values of W, m, p, pb are given. Usually, the values of
W and m are known, we only have to calculate p and pb- Let r be the transmission
probability. Since a station transmits only when the backoff counter becomes zero
{i.e. the station is at any of the b^o states), we have
r = t k o = ^ L- (5-7)
i=0 1 - P
A packet collides when more than one station transmits at the same time slot so
that the probability p of packet collision is given by
p = l - { l - r ) n~1. (5.8)
The channel is sensed busy when at least one station transmits at that time slot.
Hence, the probability for a channel to be busy is given by
pb = 1 - (1 _ r ) n. (5.9)
Equations (5.6), (5.7), (5.8) and (5.9) are a nonlinear system of equations in terms
of t and p . and we can solve this system numerically. Once t , p and pb are obtained,
we can compute the stationary distribution of this Markov chain.
103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.2.2.1 Saturation Throughput Analysis
Here, we assume that each transmission is a renewal process whether it is successful
or not. Based on this assumption, we can compute the throughput of the CSMA/CA
protocol during a single renewal interval between two consecutive transmissions. The
throughput S is defined as
g E[payload information transmitted in a slot time] _ ^
E[time between two consecutive transmissions]
It can be rewritten as
g =
-E'M + PsTs + (1 — Ps)Tc ’ 1 ' '
where we have the following definitions:
1. E[P]: the average payload length;
2. Ts: the average time that the channel is sensed busy with a successful trans
mission;
3. Tc: the average time that the channel is sensed busy with a collision;
4. ps: the probability that a transmission is successful;
5. E[p\: the mean value of tp, where is the number of consecutive idle time
slots before a transmission takes place due to the backoff mechanism.
To simplify the analysis, we assume that all packets have the same size. As a
result, E[P] = P is a fixed value. According the IEEE 802.11 specification, the
104
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
values of Tc and Ts depends on the channel access mode and defined in the following
equations. That is, we have
Ts = H + P + S + SIFS + ACK + 8 + DIFS, (5.12)
Tc = H + P + 8 + DIFS. (5.13)
for the basic access mode and
Ts = RTS + < 5 + SIFS + CTS + £ + SIFS + H +
P + 8 + SIFS + A C K + 8 + DIFS, (5.14)
Tc = RTS + < 5 + DIFS, (5.15)
for the RTS/CTS mode, where H is the sum of the PHY header and the MAC
header for the frame header while 8 denotes the propagation delay. A transmission
is said to be successful if only one of the n stations transmits, given that there is at
least one transmission. Thus, we have
n r ( l — r ) n 1
= w W - ( “ 6 >
Then, the mean number E[tp\ of consecutive idle slots before a transmission takes
place is given by
E[ip] = — - 1. (5.17)
Pb
105
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Plugging Ta, Tc, ps and E[ip] given above into Equation (5.11), we obtain the through
put of the CSMA/CA protocol for both the basic and the RTS/CTS modes.
5.2.2.2 Saturation Delay Analysis
The packet delay D is defined as the duration from the generation of the packet and
its successful transmission. The mean packet delay can be expressed as
E[D] = E{NC }{E[B] + T C + T0) + (E[B\ + Ta), (5.18)
where E[NC ] is the average number of collisions of a frame until its successful trans
mission and E[B) is the average backoff delay in a station that includes both the
counter decrementing time and the counter frozen time. Since ps is the probability
for a transmission to be successful, the average number of retransmission is l/p s.
Thus, we have
E[NC ] = 1 - 1 . (5.19)
Ps
The quantity E[B) depends on the value of the backoff counter and the duration
that the counter freezes when the channel is sensed busy.
106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
If the counter of a station is at state we need a time interval of k slots for the
counter to reach state 0 without considering the time interval in which the counter
is stopped. The average value of this time interval is computed as
m W i - 1
E[x] = E E khk- (5.20)
i = 0 fc=0
This value can be computed using (5.2), (5.3) and (5.4). Furthermore, let us consider
the time in which the counter freezes and denote it by F. When a counter freezes,
it remains stopped for the duration of one transmission. This duration depends
on whether the transmission is successful or not. Before computing E[F]) we have
to find E [ N Fr) first, which is the average number of times that a station detects
transmission from other stations before its counter reaches 0. Let E[p) be the mean
number of consecutive idle slots before a transmission proceeds. Then, we have
- L (5-21)
E[F] = E[NFt} [ PsTs + (1 - P s ) T c\. (5.22)
Therefore, we get
E[B] = E[x) + E [ N Fr][PsT s + (1 - P s ) T c). (5.23)
107
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The value T0 in Equation (5.18) is the time that a station has to wait when
its frame transmission collides before sensing the channel again. It depends on the
access mode. That is, we have
By substituting (5.19), (5.23) and (5.24) into (5.18), we can compute the packet
delay of the CSMA/CA protocol. Note that the values Ts and Tc in (5.18) have
already been derived in Section 5.2.2.1.
5.3 P ro p o sed A d ap tiv e Backoff M echanism
an d Its A nalysis
5.3.1 P roposed A daptive Backoff M echanism
From the discussion in Section 5.2, we know that a station uses CW to control the
backoff counter for data transfer. It is clear that the performance of the CSMA/CA
protocol is affected by the way to set up the CW value. To optimize the perfor
mance of the CSMA/CA protocol, it is important to search the optimal value of
CW dynamically. The problem with the current IEEE 802.11 standard is that its
CW resetting scheme does not attempt to optimize the CW value. Instead, the
station always resets the initial CW value to CWmin.
S IF S + ACKtimeout for ACK CSMA/CA;
(5.24)
S IF S + CTStimeout for RTS/CTS CSMA/CA.
108
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Let us consider a situation where there exists severe traffic congestion in the
system. Following the IEEE 802.11 backoff scheme, every station has to start from
the minimum backoff window size CWmin for each packet transmission. This tends
to result in transmission failure, since the backoff window size is too small for each
station to transmit the packet under a heavy traffic load and the collision probability
is high. To derive a more effective algorithm, we let each station adjust its backoff
window size using an indicator that can reflect the network traffic load. This will
achieve better bandwidth utilization and decrease the collision probability.
Here, we propose a dynamic CW resetting scheme. Our method can be easily
implemented in the current IEEE 802.11 standard without complex calculation and
run-time estimation as done in [6]. The proposed scheme is described as follows:
• After a successful transmission, the CW value w is set to max[ta/2, CW m in +
1],
• Whenever a transmission fails, the CW value w is set to min[2w, CW max + 1].
5.3.2 M arkov M o d el and A nalysis
In this section, we use a Markov model to analyze the performance of the proposed
adaptive backoff window scheme. Here, we assume a fixed number of stations, each
of which always has queued packets for transmission, i.e. the transmission queue of
each station is assumed to be nonempty. The other assumptions and notions are the
same as those in Section 5.2
109
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(1-P)/W,
I -p .
1 -p ,
PA V ,
1-P, 1-P,
1-P,
1-Ph
1-P, 1-P,
PA V ,
1-P,
0 , W 0- l
1 -p , 1 - p ,
o o
l~ P b (i, Wr l '
Figure 5.7: The Markov model for the proposed backoff scheme.
1 1 0
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The discrete-time Markov chain model corresponding to the problem described
above is shown in Fig. 5.7. To analyze this model, we have to find the steady state
probability for a station to be in state { ? ', k}. Let = lim^oo Pr{s(i) = i, b(t) = k}
be the stationary probability of the Markov model, where 0 < 2 < m, 0 < k < W i~ 1.
In the steady state, the following system of equations hold
bifi = xlb0fi, 0 < i < to,
(5.25)
(5.26)
where x = Besides, we have the following relations based on the regularities of
the Markov chain:
bi,k —
Wt - k
W,
yz^{bo,o + bifi) if i = 0
p6«-i,o+(l-p)6i+i,o if 0 < i < m
i~Pb
+ bm,o) if i — m
(5.27)
Let Wi = 2% W and W = C W m in + 1. Then, we have
m W i — 1
1 = E E h t-
i= 0 fc=0
(5.28)
111
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
By solving (5.28), we obtain
, = _________ 2(l-2ar)(l-g)(l - pb )________ , ,
0 ,0 W[ 1 — (2x)m+1](l — x) + (1 — 2x )(l — xm+1)
We can com pute bitk if the values of W. m, p, pb are given. Usually, the values of
W and m are known, we only have to calculate p and p & . Let r be the transmission
probability. Since a station transmits only when the backoff counter becomes zero
(i.e. the station is in any of states 6^0), we have
^ (1 — xm+1)boQ
T = W bi,o = ------:------------- ■ (5.30)
f^o 1 ~ x
A packet collides when more than one station transmits at the same time slot so
that the probability p of packet collision is given by
p = 1 — (1 — r ) n_1. (5.31)
The channel is sensed busy when at least one station transmits at that time slot.
Hence, the probability for a channel to be busy is given by
pb = l - ( l - r ) U (5.32)
Following the same derivation procedure given in Section 5.2, we can determine the
stationary distribution of the proposed Markov chain. That is, the saturation delay
112
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 5.1: System parameters used in analysis and simulation
Packet payload 1024 bytes
MAC header 34 bytes
PHY header 16 bytes
ACK 14 bytes+PHY header
RTS 20 bytes+PHY header
CTS 14 bytes+PHY header
Channel bit rate 1 Mbps
Propagation delay 1 [ I S
Slot time 50 ( j s
SIFS 28 [is
DIFS 128 [is
ACK.Timeout 300 [is
CTS-Timeout 300 [is
and throughput can be derived using the same formulas given in Section 5.2.2.1 and
Section 5.2.2.2
5.4 Com puter Sim ulation and Perform ance
Evaluation
This chapter serves two purposes. First, we would like to validate the theoretical
performance analysis for the two backoff schemes derived in Sections 5.2 and 5.3.
Thus, we compare analytical results with simulation results conducted with OPNET.
Second, we would like to demonstrate the advantage of the proposed backoff scheme
by comparing its saturation throughput and access delay with those of the original
backoff scheme given in IEEE 802.11.
Parameters used in our analysis and simulation are summarized in Table 5.1.
113
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Let us first examine the saturation throughput behavior of the RTS/CTS and
the basic access modes for the IEEE 802.11 backoff scheme and the the proposed
backoff scheme. They are given in Figs. 5.8, 5.9, 5.10 and 5.11.
In these four figures, the X-axis is the number of stations while the Y-axis denotes
the saturation throughput. Parameters m and W stand for the number of the backoff
stage and the initial backoff window size, respectively. In these figures, we plot results
with the same number of the backoff stage, i.e. m = 4, but with a different initial
backoff window size i.e. W = 16,32,64. We have the following observations.
1. The analytical model is quite accurate. It is consistent with the general trend
of simulation results and the error is less than 1%.
2. The proposed scheme has a higher saturation throughput since it can decrease
the chance of collision by adaptively adjusting the contention window.
3. The saturation throughput depends on the initial backoff window size. From
the figures, we see that W = 16 has a much poorer throughput than W =
64. This is because a small backoff window increases the collision probability,
especially when the number of stations is large.
4. When the number of stations increases, the saturation throughput of the
basic mode decreases significantly since the probability of collision becomes
larger. In contrast, saturation throughput degradation is less significant for
the RTS/CTS mode when the number of stations increases.
114
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5. T he basic access mode has a higher throughput when we choose a larger initial
backoff window size. However, for the RTS/C TS access mode, the throughput
is less improved even w ith a larger initial backoff window size. This can be
explained by the fact th a t a larger backoff window can decrease the probabil
ity of collision and the number of retransmissions for the basic access mode.
In contrast, the RTS/CTS access mode by itself can avoid collision and the
associated waste of the bandwidth when collision occurs.
Figs. 5.12, 5.13, 5.14 and 5.15 plot the saturation throughput for the IEEE 802.11
backoff scheme and the proposed backoff scheme with the basic and the RTS/CTS
access modes. In these figures, the X-axis is the number of stations and Y-axis is
the saturation throughput. In these figures, the initial backoff window sizes are the
same (i.e. W = 16) while the number of the backoff stage varies (i.e. m = 3,5, 7).
From these figures, we show the effectiveness of the proposed scheme since fewer
collisions occur in both access modes. Several observations are given below.
1. The case m — 3 has a poorer throughput than cases with m — 5 or 7 in both
access modes. This is because a smaller backoff stage number increases the
collision probability especially when the number of stations is large.
2. The basic mode improves its throughput when a larger number of stations are
active. For the RTS/CTS access modes, the improvement is less obvious.
3. The RTS/ CTS access mode has better performance than the basic access mode,
since the RTS/CTS access mode waits less bandwidth when collision occurs.
115
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figs. 5.16, 5.17, 5.18 and 5.19 show the saturation delay for the IEEE 802.11
and th e proposed backoff schemes in the RTS/C TS and the basic access modes. In
these figures, the X-axis is the number of stations while the Y-axis is the saturation
delay in the unit of msec. In all figures, we plot results with the same number of
the backoff stage [i.e. m = 4) but with different initial backoff window sizes with
W = 16,32, 64. We have the following observations.
1. These figures demonstrate that the proposed backoff scheme has a lower access
delay compared to the original IEEE 802.11 backoff scheme in both access
modes. This is due to the fact that the number of retransmissions is reduced
in the proposed scheme.
2. The RTS/CTS access mode has a lower access delay that the basic access mode
since a smaller amount of overhead is needed for transmission.
3. When the number of stations attempting to transmit increases, the number of
collisions also increases. Hence, the access delay becomes longer.
4. When the number of stations reaches a certain value, the collision probability
increases and stations have to choose a higher stage number. Choosing a
higher stage number increases delay. In these figures, a steeper slope means
that stations choose a larger contention window.
Figs. 5.20, 5.21, 5.22 and 5.23 show the saturation delay for the IEEE 802.11
backoff scheme and the proposed backoff scheme in the basic and the RTS/ CTS
116
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
access modes. In these figures, the X-axis is the number of stations while the Y-axis
is the saturation delay in the unit of msec. These figures depict the relationship
between the access delay and a different number of backoff stages (i.e. m — 3,5, 7)
w ith the same initial backoff window size W = 16. We see from these figures th at
stations suffer higher delay when the backoff procedure chooses a higher number of
backoff stages. When the number of collisions increases, stations will choose a higher
number of backoff statges, which corresponds to a larger backoff window size and
increases access delay.
5.5 Conclusion
In this research, we proposed an enhanced backoff scheme for the IEEE 802.11 DCF
protocol, which is effective yet easily implemented in the current standard. A new
Markov model was presented to compute the theoretical saturation throughput and
access delay for both the basic and the RTS/CTS access modes. Based on the
analytic model, we demonstrated the performance enhancement of the proposed
contention window selection scheme. Besides, the relationship between the perfor
mance improvement, the number of backoff stages and the initial backoff window
size was investigated. We observed that the performance of the basic access mode
strongly depends on these parameters while that of the RTS/ CTS access mode is less
affected by the variation of these parameters. Finally, the RTS/ CTS access mode
117
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
has a higher throughput and a lower access delay than the basic access mode
the saturated traffic condition.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.87
w=64
0.85
0.84
w=64 w=16
w=32
802.11 original theoretical
802.11 original simulation
0.79
30 40 50 10 20
Number of stations
Figure 5,8: The saturation throughput for the 802.11 backoff scheme in the
RTS/ CTS access mode with m = 4.
— 802.11 original theoretical
•• 802.11 original simulation
W=64
=
p
W=32
0 .55
50 10 20 30 40
Number of stations
Figure 5.9: The saturation throughput for the 802.11 backoff scheme in the basic
access mode with m — 4.
119
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.87
w=64
V.-32
0.85
w=16
w=64
w=16
‘g - 0.83
w=32
Proposed scheme theoretical
Proposed scheme simulation
0.82
0 .8 1
10 20 30 40 50
Number of stations
Figure 5.10: The saturation throughput for the proposed backoff scheme in the
ETS/CTS access mode with m — 4.
W=64
W=32
W=16
0 .6 5
i.55
-Proposed scheme theoretical
Proposed scheme simulation
10 20 4 0 30 50
Number of stations
Figure 5.11: The saturation throughput for the proposed backoff scheme in the basic
access mode with m = 4.
120
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0 .8 /
0.86
m=7
0.85
m=5
a ° -8 4
o
3
a 0.83 ni=3
in:
m=7
m=3
— 802-11 original theoretical
••802.11 original simulation
0.79
40 50 1 0 20 30
Number of stations
Figure 5.12: The saturation throughput for the 802.11 backoff scheme in the
RTS/CTS access mode with W = 16.
0.85
802.11 original theoretical
802.11 original simulation
£
f a
S’ .
O
3
.75
m=7
&
® 0.65
1"
c: 0.6
m=5
'.55
m=3
50 10 20 30 40
Number of stations
Figure 5.13: The saturation throughput for the 802.11 backoff scheme in the basic
access mode with W — 16.
1 2 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.87
m=7
0.85
O
S3 0.84
£
§ 0.83
& ° -8 2
m=3
m=3 'x * - jn=5
Proposed scheme theoretical
Proposed scheme simulation
0.81
20 30 40 50 10
Number of stations
Figure 5.14: The saturation throughput for the proposed backoff scheme in the
RTS/CTS access mode with W = 16.
0.85
m=7
0.75
0.65
m=5
0.55
Proposed scheme theoretical
Proposed scheme simulation
10 20 30 40 50
Number of stations
Figure 5.15: The saturation throughput for the proposed backoff scheme in the basic
access mode with W — 16.
1 2 2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
- 802.1 i original theoretical
’ 802.11 original simulation
1000
w=64
o
S3 600
Q -
2 L
S T
400
w=32
200
w=16
40 50 2 0 30 10
Number of stations
Figure 5.16: The saturation delay for the 802.11 backoff scheme in the RTS/CTS
access mode with m — 4.
802.11 original theoretical
802.11 original simulation. •
1400
0 0
» 1200
w=64
I 800
w=32 600
400
w=16
200
20 30 40 50 10
Number of stations
Figure 5.17: The saturation delay for the 802.11 backoff scheme in the basic access
mode with m = 4.
123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Proposed scheme theoretical
Proposed scheme simulation 1000
Ej 800
p
o'
£ 3
Cu 600
S T
* <
400
w=64
w=32
200
w= 16
10 20 30 40 50
Number of stations
Figure 5.18: The saturation delay for the proposed backoff scheme in the RTS/ CTS
access mode with m = 4.
-Proposed scheme theoretical
• Proposed scheme simulation
1400
1200
0 0
P
j=f 1000
3
o'
£ 3
C a
n s
ST 600
800
w=32
400
w=16
200
10 40 50 20 30
Number of stations
Figure 5.19: The saturation delay for the proposed backoff scheme in the basic access
mode with m = 4.
124
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
•802.11 original theoretical
■802.11 original simulation.®
1 2 0 0
o o
g . 1 0 0 0
m=7
800
C .
600
400
m=5
200
m=3
40 50 10 20 30
Num ber o f stations
Figure 5.20: The saturation delay for the 802.11 backoff scheme in the RTS/CTS
access mode with W = 16.
2000
802.11 original theoretical
*802.11 original simulation
1800
1600
1400
m=7
a 1 2 0 0
o’
? 1000
G u
gr 8oo
600
m=5
400
200
, m=3
40 10 20 30 50
N um ber o f stations
Figure 5.21: The saturation delay for the 802.11 backoff scheme in the basic access
mode with W = 16.
125
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
■Proposed scheme theoretical
Proposed scheme simulation
1200
GO
g 1000
m=7
© ‘ 800
Cl
C t >
S' 8 0 0
400
m=5
200
m=3
10 20 30 40 50
Number of stations
Figure 5.22: The saturation delay for the proposed backoff scheme in the RTS/CTS
access mode with W — 16.
2000
Proposed scheme theoretical
Proposed scheme simulation
1800
1600
g - 1400
o
1 :3 1200
Q -
0 >
S' 1000
8 00
m=7
600
m=5
400
200
, m=3
10 20 30 40 50
Number of stations
Figure 5.23: The saturation delay for the proposed backoff scheme in the basic access
mode with W = 16.
126
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Conclusion and Future Work
6.1 Summary of the Research
In this dissertation, we first considered the improvement of the DOCSIS MAC pro
tocol for QoS provision while allowing services to share the upstream channel over a
HFC system according to their need. In Chapter 3, we studied the collision resolu
tion problem for the DOCSIS system. We proposed a novel dynamic backoff window
protocol. Unlike the DOCSIS specification that does not support any priority mech
anism for prioritized traffic, our scheme can support multiple priority traffic. CMs
send their bandwidth request to the HE to ask for permission to transmit data by
the binary backoff procedure. To resolve collisions and increase the successful trans
mission probability for requests, the HE dynamically changes the backoff window
size adapting to the number of contending stations for each priority traffic. After
127
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
receiving the bandwidth requests from CMs, the HE grants the transmission accord
ing to the multiple priority queue scheduling discipline. Finally, CMs send their
data, upon receiving the grant. The proposed scheme is a simple reservation scheme
that put all computational burden at the HE while keeping the CM execution un
changed. To make the proposed scheme work properly, we supplemented some data
in the bandwidth allocation map message. As a result, the proposed scheme can
be easily implemented in the DOCSIS specification with simple modifications. The
experimental study demonstrates that our protocol is more efficient than the origi
nal DOCSIS MAC protocol since the contention slots are appropriately assigned for
different priority traffic classes.
To design a QoS mechanism for the HFC system that meets the requirements of
a wide range of applications, we investigated the traffic scheduling and bandwidth
allocation problems for the DOCSIS specification in Chapter 4. We proposed a new
scheduling service class and the corresponding bandwidth allocation algorithm for
real time VBR traffic. Our goal is to support applications like MPEG video, video-
on-demand, on-line video game, etc. The novel scheduling service class employs
the idea of the UGS service class and the piggyback mechanism. The bandwidth
request of the average data rate of a VBR traffic is allocated periodically which is
similar to the UGS service. The bandwidth request of the remaining VBR portion
is piggybacked to the HE. We used the feature of real time VBR traffic, that is
the traffic pattern has long range dependence, to predict the bandwidth allocation
128
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
of the UGS part. We studied the performance of the proposed scheme by simula
tion. Simulation results show that the proposed scheme does not only increase the
bandwidth utilization but also decreases the access delay for real time VBR traffic.
Then, we discussed the QoS performance of voice services in DOCSIS networks. In
our proposed scheme, voice data without SAD are served by the UGS class while
voice data with SAD are served by UGS-AD. To facilitate the performance analysis,
a three-state Markov chain was proposed to represent the state transition of voice
with SAD traffic. The packet dropping probability and the bandwidth utilization
efficiency were derived based on the Markov model. We conducted numerical exper
iments to show the performance of voice traffic over the DOCSIS cable networks.
We studied the backoff mechanism for the IEEE 802.11 wireless LAN system in
Chapter 5. The drawback of the current 802.11 DCF is that it does not consider
the previous collision history. Besides, the adjustment of the CW dose not reflect
the variation of the network traffic load. An enhancement of the backoff mecha
nism was proposed to fix this problem. That is, the proposed scheme adjusted the
backoff window size according to the network traffic load. To show the performance
enhancement, we adopted the Markov model to analyze the backoff mechanism and
obtained the theoretical saturation throughput and delay for the proposed backoff
scheme. Simulation studies were also conducted to verify the correctness of theoret
ical derivation.
129
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.2 Future W ork
There are some possible extensions of our current research as stated below.
« It was mentioned in [10] that cable networks are vulnerable due to a variety of
radio-frequency (RF) impairments. These RF impairments can result in signif
icant packet loss for both downstream and upstream traffics. The transmission
control protocol (TCP) will degrade dramatically in performance due to packet
loss. Therefore, the performance analysis of TCP under different HFC network
conditions is an important topic. Besides, to enhance the performance of TCP
in the HFC networks, modification of TCP is necessary.
• The IEEE 802.11 standard has provided basic real time support by introduc
ing the PCF. Based on the polling scheme, the PCF is a contention-free access
mode. The PCF might be enhanced using improved scheduling and/or signal
ing schemes. The IEEE 802.11 standard does not define the detail operation of
the PCF protocol for both infrastructure and ad-hoc networks. To support di
verse applications on wireless LAN (e.g. video-on-demand, video conferences,
VOIP), the design of the PCF protocol to meet different QoS requirements for
real time applications is an interesting problem.
• New enhancements are being added to the IEEE 802.11 MAC layer to support
QoS required by real time applications. Two main schemes for improving
the performance of the IEEE 802.11 MAC layer are being considered by the
130
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
802.11 Working Group E. The first one is an extension to the current DCF
contention-based access scheme to provide service differentiation via priorities.
The other one is a modification to the existing PCF for more efficient polling
schemes. These two modifications are not yet finalized, and a large amount
of research effort is required to understand the new IEEE 802.11 QoS-aware
MAC behavior.
® The 802.11 standard introduced the Wired Equivalent Privacy (WEP) pro
tocol to bring the security level of wireless systems closer to that of wired
networks. The primary goal of WEP is to protect the confidentiality of user
data from eavesdropping. Unfortunately, WEP has several weaknesses, such
as key stream reuse, key management, inappropriate use of CRC, and so on.
Understanding important security principles and designing an enhanced WEP
algorithm are important to achieve a higher security performance of the IEEE
802.11 system.
131
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
R eference List
[1 ] G. Bianchi. Performance Analysis of the IEEE 802.11 Distributed Coordination
Function. IEEE Journal on Selected Areas in Communication. 18:535-547,
March 2000.
[2 ] G. Bianchi, L. Fratta, and M. Qliveri. Performance Analysis of IEEE 802.11
CSMA/CA Medium Access Control Protocol. IEEE PIMRC Proc., pages 407-
411, Oct 1996.
[3 ] D. Bushmitch, S. Mukherjee, S. Narayanan, M. Ratty, and Q. Shi. Supporting
MPEG Video Transport on DOCSIS-Compliant Cable Networks. IEEE Journal
on Selected Areas in Communication, 18:1581-1595, 2000.
[4 ] Cable Television Laboratories, Inc. Data Over Cable Service Interface Specifi
cations Radio Frequency Interface Specification 1.1. July 1999.
[5 ] Cable Television Laboratories, Inc. Data Over Cable Service Interface Specifi
cations Radio Frequency Interface Specification 1.0. May 1997.
[6 ] F. Cali, M. Conti, and E. Gregori. Dynamic Tunning of the IEEE 802.11
Protocol to Achieve a Theoretical Throughput Limit. IEEE/ACM Trans, on
Networking, 8:37-51, Dec 2000.
[7 ] Y. C. Chen, S. C. Hsu, T. L. Hsu, and W. S. Hsieh. The Strategies of Traffic
Control for Multimedia Data Transmission with QoS Guarantees over CATV
Network. IEEE Trans. Consumer Electronics, 45:107-117, 1999.
[8 ] H. S. Chhaya and S. Gupta. Performance Modeling of Asynchronous Data
Transfer Methods of IEEE 802.11 MAC Protocol. Wireless Networks, 3:217-
234, 1997.
[9 ] B. P. Crow. Performance Evaluation of the IEEE 802.11 Wireless Local Access
Network. M.S. Thesis, Dept. Electrical and Computer Eng., Univ. Arizona,
Tucson, AZ, 1996.
[10] C. Eldering, N. Himayat, and F. Gardner. CATV Return Path Characterization
for Reliable Communications. IEEE Commun. Mag., 33:62-69, Aug 1995.
132
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[11] N. Golmie. F. Mouveaux, and D. Su. A Comparison of MAC Protocols for
Hybrid Fiber/Coa,x Networks: IEEE 802.14 vs. MCNS. IEEE ICC Proc., pages
266 -272, Jun 1999.
[12] Institute of Electrical and Electronics Engineers. IEEE Std 802-11: Wireless
LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications.
Aug 1997.
[13] Institute of Electrical and Electronics Engineers. IEEE Std 802-1 la: Wireless
LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications:
High-Speed Physical Layer in the 5 GHz Band. Sep 1999.
[14] Institute of Electrical and Electronics Engineers. IEEE Std 802-llb: Wireless
LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications:
Higher-Speed Physical Layer Extension in the 2.4 GHz Band. Sep 1999.
[15] F. Koperda, B. Lin, and J. Collins. A Proposal to Use XDQRAP for 802.14.
Contribution to the IEEE 802.14 WG, IEEE 802.14-95/068, Oct, 1995.
[16] Y. D. Lin, C. Y. Huang, and W. M. Yin. Allocation and Scheduling Algorithm
for IEEE 802.14 and MCNS in Hybrid Fiber Coaxial Networks. IEEE Trans.
Broadcasting, 44:427-435, Dec 1998.
[17] R. Rabbat, and K. Y. Siu. QoS Support for Intergrated Services over CATV.
IEEE Commun. Magazine, 37:64-68, Jan 1999.
[18] R. L. Rivest. Network Control by Bayesian Broadcast. M IT Lab for Computer
Science, Cambridge, Massachusetts, MIT/LCS/TM-287, 1985.
[19] D. Sala, J. O. Limb, and S. U. Khaunte. Adaptive Control Mechanism for Cable
Modem MAC Protocols. IEEE INFOCOM Proc., pages 1392 -1399, 1998.
[20] R. M. Salles and P. R. D. L. Gondim. Performance of Priority-Based Mul
tiple Access with Dynamic Permission (PBMA.DP) for Multimedia Wireless
Networks. IEEE ICC Proc., pages 161-165, July 1998.
[21] Silberschatz, Galvin, and Gagne. Operating System Concepts. Jon Wiley &
Sons, 6th edition, 2002.
[22] D. C. Twu and K. C. Chen. Dynamic Control Scheme for the IEEE 802.14 Draft
Protocol Over CATV/HFC Networks. IEEE Commun. Letters, 2:197-199, Jul
1998.
[23] J. Weinmiller, M. Schlager, A. Festag, and A. Wolisz. Performance Study of
Access Control in Wireless LANs IEEE 802.11 DFWMAC and ETSI RES 20
HIPERLAN. Mobile Networks and Application, 2:55-67, 1997.
133
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[24] C. Wu and G. Campbell. Extended DQRAP (XDQRAP): A Cable TV Protocol
Function as a Distribued Switch. Illinois Institute of Technology, Computer
Science Department, Technical Report 94-2, 1994.
[25] E. Ziouva and T. Antonakpoulos. CSMA/CA Preformance under High Traf
fic Conditions: Throughout and Delay Analysis. Computer Communications,
25:313-321, 2002.
134
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Dynamic radio resource management for 2G and 3G wireless systems

PDF

Contributions to image and video coding for reliable and secure communications

PDF

Design and analysis of server scheduling for video -on -demand systems

PDF

Design and analysis of OFDMA and SC -SDMA for broadband multiuser communications

PDF

Contributions to coding techniques for wireless multimedia communication

PDF

Call admission control and resource allocation for quality of service support in wireless multimedia networks

PDF

Design and performance analysis of low complexity encoding algorithm for H.264 /AVC

PDF

Adaptive video transmission over wireless fading channel

PDF

Design and performance analysis of a new multiuser OFDM transceiver system

PDF

Contributions to efficient vector quantization and frequency assignment design and implementation

PDF

A template-based standard-cell asynchronous design methodology

PDF

Intelligent image content analysis: Tools, techniques and applications

PDF

Intelligent systems for video analysis and access over the Internet

PDF

Error resilient techniques for robust video transmission

PDF

Efficient acoustic noise suppression for audio signals

PDF

Blind multiuser receivers for DS -CDMA in frequency-selective fading channels

PDF

Direction -of -arrival and delay estimation for DS -CDMA systems with multiple receive antennas

PDF

Color processing and rate control for storage and transmission of digital image and video

PDF

Code assignment and call admission control for OVSF-CDMA systems

PDF

Algorithms and architectures for robust video transmission

Asset Metadata

Creator Kuo, Wen-Kuang (author)

Core Title Design and analysis of MAC protocols for broadband wired/wireless networks

School Graduate School

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag engineering, electronics and electrical,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Kuo, C.-C. Jay (committee chair), [illegible] (committee member), Zimmermann, Roger (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-640437

Unique identifier UC11340086

Identifier 3116734.pdf (filename),usctheses-c16-640437 (legacy record id)

Legacy Identifier 3116734.pdf

Dmrecord 640437

Document Type Dissertation

Rights Kuo, Wen-Kuang

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA