Sunteți pe pagina 1din 74

國 立 中 央 大 學

通 訊 工 程 研 究 所
碩 士 論 文

正交分頻多重接取系統下之跨層式排程及
資源分配演算法設計

研 究 生:林瑤亭

指導教授:陳永芳 博士

中華民國九十六年六月
國立中央大學圖書館
碩博士論文電子檔授權書
(95 年 7 月最新修正版)

本授權書所授權之論文全文電子檔(不包含紙本、詳備註 1 說明),為本人於國立
中央大學,撰寫之碩/博士學位論文。(以下請擇一勾選)

( )同意 (立即開放)
( )同意 (一年後開放),原因是:
(ˇ )同意 (二年後開放),原因是:投稿審定中
( )不同意,原因是:

以非專屬、無償授權國立中央大學圖書館與國家圖書館,基於推動「資源共享、

互惠合作」之理念,於回饋社會與學術研究之目的,得不限地域、時間與次數,

以紙本、微縮、光碟及其它各種方法將上列論文收錄、重製、公開陳列、與發行,

或再授權他人以各種方法重製與利用,並得將數位化之上列論文與論文電子檔以

上載網路方式,提供讀者基於個人非營利性質之線上檢索、閱覽、下載或列印。

研究生簽名: 學號:945203013

論文名稱: 正交分頻多重接取系統下之跨層式排程及資源分配演算法

設計

指導教授姓名:陳 永 芳 教授 `

系所 :通 訊 工 程 學系 所 †博士班 „碩士班

日期:民國 96 年 7 月 18 日

備註:
1. 本授權書之授權範圍僅限電子檔,紙本論文部分依著作權法第 15 條第 3 款之規定,採推
定原則即預設同意圖書館得公開上架閱覽,如您有申請專利或投稿等考量,不同意紙本
上 架 陳 列 , 須 另 行 加 填 聲 明 書 , 詳 細 說 明 與 紙 本 聲 明 書 請 至
http://thesis.lib.ncu.edu.tw/paper.htm 查閱下載。
2. 本授權書請填寫並親筆簽名後,裝訂於各紙本論文封面後之次頁(全文電子檔內之授權
書簽名,可用電腦打字代替)。
3. 請加印一份單張之授權書,填寫並親筆簽名後,於辦理離校時交圖書館(以統一代轉寄
給國家圖書館)。
4. 讀者基於個人非營利性質之線上檢索、閱覽、下載或列印上列論文,應依著作權法相關
規定辦理。
正交分頻多重接取系統下之跨層式排程及
資源分配演算法設計

指導老師:陳永芳 博士 研究生:林瑤亭

論文摘要

在本文中,我們討論兩個主題。第一個部份,我們提出了一個階層式
排程演算法設計。IEEE 802.16 的標準提供了四種不同的服務,分別是非

請求的頻寬分配(UGS)、即時輪詢服務(rtPS)、非即時輪詢服務(nrtPS)、
盡力傳送(BE)。每個服務的封包傳輸優先權皆有不同的限制。在排程演算
法中,不僅根據媒體存取層(MAC layer)不同連線的服務質量所要求的保
證傳輸頻寬和最大延遲時間來設計排程的方式,其中我們計算出每個框架
(frame)中不同連線需要傳輸的資料量,藉以選擇使用者給實體層做傳遞
之用;同時,也在排程演算法中加入不同用戶的實體層通道狀況。如此,
媒體存取層在選擇使用者時,同時可以顧及服務質量和用戶的通道狀況,
於是整個系統可以用此達到最好的傳輸效能和服務要求。這個演算法跟近
兩年發表過的三個排程演算法做比較,在模擬結果可以看出我們提出的演
算法效能較好。第二個部份,我們計算出分配子通道的最佳解,利用這樣
的最佳解設計出演算法,藉由這樣的演算法,我們可以根據媒體存取層所
選擇的用戶限制下,達到最高的資料傳輸量。結合這兩個部份,於是我們
的演算法可以在滿足媒體存取層的不同傳輸要求下,又能提高整體的總傳
輸速率,由我們的模擬繪圖可以證明我們的設計確實達到我們所想要的結
果。

I
A Cross Layer Scheduling and Resource Allocation

Algorithm for OFDMA Wireless Networks

Advisor: Dr. Yung-Fang Chen Student: Iao-Tin Lin

Abstract

As multimedia communications develop rapidly in the recent years,


in order to maximize the capacity of wireless networks, scheduling plays
an important role in supplying quality of service (QoS) requirements to
broadband wireless communications. The IEEE 802.16 standard provides
four different scheduling services: Unsolicited Grant Service (UGS),
real-time Polling Service (rtPS), non-real-time Polling Service (nrtPS),
and Best Effort (BE). In this paper, we formulate the optimal problem by
maximizing the average utility function of all active users and then
propose a cross-layer algorithm to achieve the higher throughput by
allocating resources dynamically. Our scheme is aimed at designing
jointly dynamical subchannel assignment (DSA)and capacity planning
(CP) solutions. The simulation focuses on IEEE 802.16 wireless systems
working in Orthogonal Frequency Division Multiple Accesses (OFDMA)
and Point-to-Multipoint (PMP) mode, with one single cell serving many
Mobile Stations (MSs) in downlink transmission. Finally, the numerical
result shows that the cross-layer algorithm approaches a higher
throughput and improves extremely the performance of BE and nrtPS
services while supporting the quality of rtPS services at acceptable levels.

II
致謝

首先誠摯的感謝指導教授陳永芳博士,老師悉心的教導使我得以一

窺領域的深奧,藉由不時的討論並指點我正確的方向,使我在這兩年

中獲益匪淺,並給於學生相當大的空間去發揮個人所長,學習如何思

考及解決問題。老師也盡力提供實驗室的硬體設備,使我們能無後顧

之憂的跑模擬做研究。

兩年裡的日子,實驗室裡共同的生活點滴,學術上的討論、言不及

義的閒扯、趕作業的革命情感。感謝眾位學長姐、同學、學弟妹的相

互扶持,你/妳們的陪伴讓兩年的研究生活變得絢麗多彩。

感謝明彥、錫明、良媖學姐們不厭其煩的指出我研究中的缺失,且

總能在我迷惘時為我解惑;謝謝博班韋丞學長不時給予我研究上及做

人處事上的建議;感謝煜偉、育維、易凡同學,感謝你們不管是課業

上或是生活上的相互幫忙砥礪,恭喜我們順利走過這兩年。實驗室的

世原、正毅、克軒學弟們當然也不能忘記,謝謝你們幫忙處理雜事及

帶來實驗室的歡笑。

老公世傑在背後的默默支持更是我前進的動力,沒有世傑的體諒、包

容,相信這兩年的生活將是很不一樣的光景。

最後,謹以此文獻給我摯愛的雙親。

III
Contents
論文摘要..................................................................................................... I
Abstract.................................................................................................... II
致謝..........................................................................................................III
Contents .................................................................................................. IV
List of Figures......................................................................................... VI
List of Table...........................................................................................VII
Chapter 1 Introduction.............................................................................1
Chapter 2 System Model ..........................................................................5
2.1 Traffic Models for QOS Architecture at the MAC layer ..............5
2.2 Wireless Networks configuration................................................11
2.2.1 Subcarrier allocations in Wimax OFDMA ........................11
2.2.2 Two-dimensional subchannel-mapping structure ..............20
2.2.3 Specifications of the entire system ....................................21
Chapter 3 .................................................................................................22
Cross-Layer Resource Allocation Algorithm .......................................22
3.1 Concepts of the present priority algorithm .................................22
3.2 Best Channel First (BCF) scheduling algorithm with Best
Channel First (BCF) subchannel allocation policy......................24
3.3 Proportional Fair (PF) scheduling algorithm with Best Channel
First (BCF) subchannel allocation policy ....................................29
3.4 Priority function (PRF) for scheduling algorithm with Best
Channel First (BCF) subchannel allocation.................................32
3.5 Modified capacity priority algorithm..........................................37
3.6 Subchannel allocation algorithm for OFDMA systems..............44
3.6.1 Problem formulation ..........................................................44

IV
Chapter 4 Simulation Results ................................................................48
4.1 Simulation models for multi-user OFDMA................................48
4.1.1 Parameters of MSs Generation Model...............................48
4.1.2 Channel Models of IEEE 802.16 OFDMA systems ..........48
4.1.3 AMC Design at the PHY....................................................50
4.1.4 Physical Layer parameters .................................................52
4.1.5 Transmission Power ...........................................................52
4.2 Simulation results for multi-user OFDMA ...............................54
Chapter 5 Conclusions............................................................................62
Reference..................................................................................................64

V
List of Figures
Figure 2.1 near Real-Time Video Traffic Model. ......................................6
Figure 2-2 Non Real-Time FTP Traffic Model ..........................................8
Figure 2.3 Best Effort HTTP Traffic Model. ..............................................9
Figure 2.4 Cluster Structure for Downlink PUSC. ...................................16
Figure 2.5 Downlink Frame Structure in IEEE 802.16 ............................20
Figure 3.1 Flowchart of Best Channel First (BCF) scheduling algorithm
with Best Channel First (BCF) subchannel allocation policy 28
Figure 3.2 Proportional Fair (PF) scheduling algorithm with Best Channel
First (BCF) subchannel allocation policy...............................31
Figure 3.3 Priority Function (PRF) scheduling algorithm with Best
Channel First (BCF) subchannel allocation policy ................36
Figure 3.4 A Brief Topology of Our Cross Layer System ........................37
Figure 4.1 Total Throughput of The System versus Total Number of
Mobiles ...................................................................................54
Figure 4.3 Average Transmission Rate to Minimum Data Rate Ratio for
Video Streaming versus Total Number of Mobiles. ...............57
Figure 4.4 Empty Ratio for Video Streaming versus Total Number of
Mobiles. ..................................................................................58
Figure 4.5 Average Data Rate of FTP Service versus Total Number of
Mobiles. ..................................................................................59
Figure 4.6 Average Transmission Rate per Mobile per Sec of FTP Service
versus Total Number of Mobiles. ...........................................60
Figure 4.7 Average Data Rate of HTTP Service versus Total Number of
Mobiles. ..................................................................................61

VI
List of Table
Table 2-1 near Real-Time Video Traffic Model Parameters.......................7
Table 2-2 FTP Characteristics.....................................................................8
Table 2-3 HTTP Traffic Model Characteristics ........................................10
Table 2-4 2048-FFT OFDMA Downlink Carrier Allocations- PUSC......12
Table 2-5 1024-FFT OFDMA Downlink Carrier Allocations- PUSC......13
Table 2-6 512-FFT OFDMA Downlink Carrier Allocations- PUSC........14
Table 2-7 128-FFT OFDMA Downlink Carrier Allocations- PUSC........15
Table 2-8 MCS Modes in The IEEE 802.16 Standard..............................21
Table 4.1 SUI 2 Channel Model Parameters ............................................49
Table 4.2 Required SNR for Different Modulations ................................51
Table 4.3 Physical Layer Parameters........................................................52

VII
Chapter 1 Introduction
In the recent years, the number of users who are familiar with
accesses to high-quality and high-data-rate multimedia applications, such
as peer-to-peer applications, internet accesses, IP telephony, and video
conferences, grows rapidly all over the world. Some solutions are
proposed to improve the performance of high-speed communication with
distinct Quality of Service (QoS) requirements in Broadband Wireless
Access (BWA) systems. The IEEE Project 802 working group 16, also
referred to as IEEE 802.16, is developing standard for fixed [1] and
mobile [2] BWA systems, which assume OFDM-based technology to be
the radio interface basis. However, the IEEE 802.16 standard formulates
a set of specifications for different classes of QOS architectures, but
reserves scheduling and connection admission control (CAC) strategies
for developers. Although many wire-line scheduling algorithms are
obtainable already from the published articles, the mechanisms could not
be really applied to wireless networks. In wireless networks, channel
quality affected by multipath fading and Doppler spreads is time varying.
If the wireless channel of a connection meets serious fading, it is wasteful
to assign any bandwidth to the connection; in wireline networks, the
designers only care about traffic services and queuing conditions. In [3],
there was an overview of the fundamental definitions and classifications
for scheduling techniques and we can find the differences between
wireless and wireline scheduling mechanisms. In [17], the article has
presented some important resource-allocation problems and then
described the proposed solutions for IEEE 802.16 wireless networks.

1
OFDM techniques divide the total transmit channel into many
orthogonal subchannels, which are composed of subcarriers. The
resources of OFDM systems are displayed by two-dimensional matrix,
where the dimensions are subchannels and OFDM symbols in the time
domain, respectively. The OFDMA system in the IEEE 802.16 defines
three subcarrier allocation schemes, which are named FUSC, PUSC, and
AMC modes, individually. The subcarrier allocation schemes specify
how to distribute the subcarriers for a certain subchannel in OFDMA
symbols. Among the allocation schemes, the number of data subcarriers
grouped into a subchannel is identical, but the subcarriers are distributed
in totally different way according to different allocation algorithms. More
details of FUSC and PUSC modes are shown in [1-2]. Furthermore, those
subchannels can be allocated to different users in non-overlapping
manners in order to provide a flexible multiuser access scheme and
exploit multiuser diversity. In [4], the author provided several
two-dimensional mapping approaches implementing the OFDMA mode
of the IEEE 802.16 standard. In this thesis, we use PUSC mode to
perform our subcarrier allocation strategy. A tradeoff between fairness
and efficiency needs to be considered in wireless communications. A fair
system may decrease total throughput and bandwidth efficiency because
of some users with serious propagation loss or bad fading channel
conditions. However, an efficient system may only consider the users
with good channel conditions. Finding the balance between fair and
efficient resource allocations is a very crucial work for wireless
systems/networks. In [5] and [6], the authors have provided a theoretical
framework for cross-layer optimization for OFDM wireless systems.

2
They have proved that OFDM networks/systems with dynamic subcarrier
assignment (DSA) and adaptive power allocation (APA) can significantly
improve the performance. And then they have focused on an effective and
practical algorithm for efficient and fair resource allocation depending on
the theoretical framework established in [5]. However, in these studies,
they didn’t include different classes of QoS services in their simulations.
All the services in wireless networks are managed at the medium
access control (MAC) layer. We can review the detail of the MAC
mechanisms defined by IEEE 802.16 standard in [7-8]. In [8], multimedia
applications have been simulated by following the IEEE 802.16 MAC
protocol and the authors have discussed the trade off between average
delay and throughput with respect to frame duration’s size in duplex
transmission. The IEEE 802.16 standard provides four different
scheduling services: Unsolicited Grant Service (UGS), real-time Polling
Service (rtPS), non-real-time Polling Service (nrtPS), and Best Effort
(BE). In [9], the scheduler assigned strict priority from high to low in the
order of rtPS, nrtPS, and BE, and guaranteed QOS requirements for high
rate multimedia data in OFDM systems. On the other hand, conditionally
high priority scheduling and cross-layer simulations have been
investigated in [10] and [11], respectively. They have proposed that the
performance of non-strict requirement scheduling services, e.g. BE and
nrtPS services, is obviously improved by giving conditional high priority
to strict requirement scheduling service, e.g. the rtPS service.
In [10], the authors have proposed a packet scheduler and a CAC
algorithm and shown that the scheduler and the CAC scheme would
greatly improve the performance of BE and nrtPS services while

3
maintaining the quality of video service at an acceptable level in OFDMA
systems. In [11], the authors have developed a low complexity cross-layer
scheduling algorithm at the MAC layer for multiple connections. The
scheduler assigned a priority of each connection depending on its channel
quality in an OFDM system. However, the dynamic subcarrier
assignment and adaptive power allocation schemes have not been
proposed in these studies.
In this thesis, we propose a cross-layer-based scheduler at the MAC
layer with diverse QoS requirements for multiple connections, where
each connection employs a near optimal subchannel allocation at the
physical layer (PHY). We compute a priority for each connection and
update it dynamically depending on channel quality and QoS-service
satisfaction. Furthermore, we offer a resource allocation unit (RAU)
mapping algorithm to allocate the resource more efficiently and fairly in
OFDMA systems. In Chapter 2, we describe the system model under our
consideration and specify the characteristics of the four QoS services. In
Chapter 3, we investigate a proposed cross-layer algorithm, for
scheduling blocks and 2D mapping blocks, respectively. In Chapter 4, we
describe the simulation environment and parameter values of the system
model and present the performance of the developed algorithm compared
with the other algorithms specified in chapter 3. Finally, Chapter 5
concludes this thesis.

4
Chapter 2 System Model

2.1 Traffic Models for QOS Architecture at the


MAC layer
The IEEE 802.16 standard defines four difference-scheduling services
to meet the QoS requirements of multimedia applications according to
packet arrival patterns (e.g., fixed/variable size of a data packet or
periodic/aperiodic packet time interval) and QoS service requirements
(e.g., applications with maximum delay requirements or minimum
guaranteed bandwidth). In the following, we describe each scheduling
service supporting a specific class of multimedia applications. In this
thesis, we use the traffic models defined in [12] and [13] and only
simulate the rtPS, nrtPS, and BE services in our system.

Unsolicited Grant Service (UGS):


UGS generates fixed-size data packets at periodic intervals to
support real-time applications with fixed throughput, strict latency, and
tolerated jitter. The UGS services apply to T1/E1 and voice over IP (VoIP)
without silence suppression. A UGS connection never requests bandwidth
to the base station. Therefore, the BS computes the minimum amount of
data transported for the UGS connection to satisfy its requirement at the
setup duration. In this thesis, we do not simulate the conditions of UGS
connections.

Real-Time polling service (rtPS):


Real-Time polling service (rtPS) generates variable-size data packets

5
at periodic intervals to provide real-time applications with less strictness
on delay than UGS, such as VoIP with silence suppression and Moving
Picture Expert Group (MPEG) video conferencing. The maximum
latency (i.e. bounded waiting time for a packet at the MAC layer) and the
minimum traffic data rates are the key QoS requirements for the rtPS
service. Because of the variable size of packets, rtPS connections need to
report their current bandwidth requirement to the BS. A packet of the
rtPS connection will be dropped when exceeding the maximum delay.

Traffic Model of rtPS service for video streaming:

Figure 2.1 near Real-Time Video Traffic Model.


A video streaming session is defined as the entire video simulation
time. The session time is divided into buffering windows, which have a
regular interval T and are composed of a fixed number of packets. A
buffering window consists of eight packets with variable sizes and arrives
at the base station every 100 ms. The size of packet is distributed as
truncated Pareto with a mean of 50 bytes and a maximum of 125 bytes. In
our simulation, we follow the parameters in [10] to set the packet size
with a mean of 500 bytes and a maximum of 1400 bytes, which has video
data rate of 300 kb/s. Packet coding delay is the inter-arrival time
between two packets in a buffering window. The distribution of the

6
packet coding delay is determined as a mean of 6 ms and a maximum of
12.5 ms according to truncated Pareto.
¾ Table 2-1 near Real-Time Video Traffic Model Parameters
Information Packet size Interval time between two
types packets in a frame
Distribution Truncated Pareto Truncated Pareto
(mean= 500 B, (mean= 6ms, Max=12.5 ms)
Max=1400B)
α α
αk αk
Distribution f x = α +1 , k ≤ x < m =
f x α +1 , k ≤ x < m
x x
parameters α α
⎛k⎞ ⎛k⎞
f x =⎜ ⎟ ,x = m
f x =⎜
⎝m⎠ ⎟ ,x = m
⎝m⎠
α = 1 .2 , k = 2 0 , m = 1 4 2 0 α = 1 .2 , k = 2 .5 m s , m = 1 5 m s

Note: Subtract k from the Note: Subtract k from the


generated r.v. to get generated r.v. to get Interval
packet size (bytes) time (ms)
Number of packets per buffering widows: 8
Buffing window size: 100 ms

Nonreal-time polling service (nrtPS):


Nonreal-time polling service (nrtPS) guarantees a minimum traffic
rate and has no delay requirement. The nrtPS connections are suitable for
time-insensitive and bandwidth-intensive applications, such as File
Transfer Protocol (FTP).

Traffic Model of nrtPS service for File Transfer Protocol (FTP):

7
Packet calls

Reading Reading time


Packets of file1 Packets of file2 Packets of file3
time time

Figure 2-2 Non Real-Time FTP Traffic Model


In FTP applications, a session time alternates between a file
transmission and the reading time. The distribution of one FTP file size is
according to a lognormal distribution, which is given as a mean of 2 M
bytes and a maximum of 5 M bytes. On the other hand, the reading time,
i.e. the time interval between two download files, is an exponential
distribution with a mean of 60s. Every file is divided into fixed sizes of
packets, with 76% using 1500 bytes and 24% using 576 bytes.
¾ Table 2-2 FTP Characteristics
Component Distribution Parameters pdf
File size Truncated Mean= 2 MB −(ln x − µ )
1
fx = ], x ≥ 0
exp[
2πσ x 2σ 2
lognormal Maximum= 5
σ = 0.35, µ = 14.45
MB

f x = λ e− λ x , x ≥ 0
Reading Exponential Mean=60 sex
λ = 0.017
time

Note for truncated lognormal:

A r.v. generated according to the truncated pdf will be discarded and another
one is regenerated when the r.v. is outside the availd interval.

8
Best Effort Service (BE):
Best Effort service (BE), which is used for electrical mail (E-mail)
and Hypertext Transport Protocol (HTTP), does not guarantee any delay
or throughput. Typically, the BE connections can not use the bandwidth
of the system until the connections supported by the above three services
are allocated.

Traffic Model of BE service for HTTP:


A session time

A packet call A packet call A packet call

Nd

Reading
Reading time
time time
Main object Parsing time
Time interval between
embedded objects
main and embedded
objects

Figure 2.3 Best Effort HTTP Traffic Model.


The session of BE service alternates between packet-download time
and reading time. In Figure 2-3, the packet downloads are defined as a
packet call which consists of a main object and Nd embedded objects. At
beginning, a web-browser sends a main object containing the initial
HTML page, and then parses the HTML page for additional embedded
files such as graphs or buttons. Both main and embedded objects are
distributed as truncated lognormal with different means of 10710 and
7758 bytes, and the same maximum values of 2 MB. The number of
embedded objects per packet call (per page) is determined as a truncated
Pareto with a mean of 5.64 and a maximum of 53. A parsing time
between main and embedded objects is exponential distribution with a
mean of 0.13 sec. Finally, the reading time alternates between two packet

9
calls distributed as exponential distribution with a mean of 30 sec.

¾ Table 2-3 HTTP Traffic Model Characteristics


Component Distribution Parameter pdf
Main object Mean=10710
bytes 1 ⎡ − (ln x − µ )2 ⎤
fx = exp ⎢ ⎥, x ≥ 0
Std. dev=25032
2 πσx ⎢⎣ 2σ 2 ⎥⎦
σ = 1.37, µ = 8.35
Min.=100 bytes
Truncated Max.=2 M bytes
⎡ − (ln x − µ)2⎤
Embedded Lognormal Mean=7758 bytes fx =
1
exp⎢ ⎥, x ≥ 0
2πσx ⎢⎣ 2σ2 ⎥⎦
objects Std dev=126168
σ = 2.36, µ = 6.17
Min.=50bytes
Max.=2 M bytes
α
Number of Truncated Mean= 5.64 αk
f x = α +1 , k ≤ x < m
embedded Pareto Max.=53 x
α
⎛ ⎞
k
object per fx= ⎜ ⎟ ,x = m
⎝ ⎠
m
call
α = 1.1, k = 2, m = 55

− λx
Parsing time Exponential Mean=0.13 sec f x = λe ,x ≥ 0

λ = 7.69

Reading time Mean=30 sec − λx


f x = λe ,x ≥ 0

λ = 0.033

10
2.2 Wireless Networks configuration
2.2.1 Subcarrier allocations in Wimax OFDMA
There are N FFT subcarriers allocated in one OFDMA symbol, which
consists two sets of subcarrier blocks, guard tones and used
subcarriers Nused . The standards define three different types of subcarrier
allocations, which are named Fully Usage of Subchannel (FUSC),
Partially Usage of Subchannel (PUSC), and AMC to distribute the used
subcarriers. Among the three subcarrier-allocation schemes, FUSC and
PUSC are kinds of distributed subcarrier permutations, and AMC is a
kind of adjacent subcarrier permutation. The number of OFDMA symbols
that the subcarriers within one subchannel are distributed over is different
between uplink and downlink transmissions. For downlink, we can select
any one of the three subcarrier allocations to generate subchannels. In the
FUSC and AMC mode, the algorithm distributes the subcarriers
belonging to the same subchannel over one OFDM symbol but over two
OFDM symbols in the PUSC mode. On the other hand, for uplink, it is
only one subcarrier permutation, i.e. PUSC mode, whose subchannels
consisting of subcarriers distributed over three OFDMA symbols.
Because the downlink loads in OFDMA wireless networks are
exactly higher than the uplink ones, we focus on the downlink
transmission. But, our scheme can be used to the uplink as well. In this
thesis, we consider the downlink PUSC mode of a single cell, which
consists of multiple mobile stations (MSs) connecting to a base station
(BS) or relay station. Because we use the PUSC subcarrier permutation,
we only develop the algorithm of PUSC permutation in IEEE 802.16e
OFDMA mode and leave the other permutation schemes aside in this

11
thesis.

z Symbol structure for Downlink PUSC:


In the downlink PUSC, the standard defines four FFT sizes, N FFT i.e.
2048-FFT, 1024-FFT, 512-FFT and 128-FFT, which is specified in the
following tables.
¾ Table 2-4 2048-FFT OFDMA Downlink Carrier Allocations- PUSC
Parameter Values Comments
Number of DC subcarriers 1 Index 1024
Number of Guard subcarriers, left 184 -
Number of Guard subcarriers, right 183 -
Number of Used Subcarriers (Nused) Number of all
1681
including all possible allocated pilots
subcarriers
and the DC subcarrier
used within a

symbol.
renumbering sequence 6, 108, 37, 81, 31, 100, Used to
42, 116, 32, 107, 30, 93, renumber
54, 78,10, 75, 50, 111, clusters before
58, 106, 23, 105, 16, allocation to
117, 39, 95, 7,115, 25, subchannels.
119, 53, 71, 22, 98, 28,
79, 17, 63, 27, 72,
29,86, 5, 101, 49, 104,
9, 68, 1, 73, 36, 74, 43,
62, 20, 84, 52, 64, 34,
60, 66, 48, 97, 21, 91,
40, 102, 56, 92, 47,90,
33, 114, 18, 70, 15, 110,
51, 118, 46, 83, 45, 76,
57,99, 35, 67, 55, 85,
59, 113, 11, 82, 38, 88,
19, 77, 3, 87,12, 89, 26,

12
65, 41, 109, 44, 69, 8,
61, 13, 96, 14, 103,
2,80, 24, 112, 4, 94, 0

Number of carriers per cluster 14 -


Number of clusters 120 -
Number of data subcarriers in each 24 -
symbol
per subchannel
Number of subchannels 60 -
PermutationBase12 6,9,4,8,10,11,5,2,7,3,1,0 -
(for 12 subchannels)
PermutationBase8 7,4,0,2,1,5,3,6 -
(for 8 subchannels)

¾ Table 2-5 1024-FFT OFDMA Downlink Carrier Allocations- PUSC


Parameter Values Comments
Number of DC subcarriers 1 Index 512
Number of Guard subcarriers, left 92 -
Number of Guard subcarriers, right 91 -
Number of Used Subcarriers (Nused) Number of all
841
including all possible allocated pilots
subcarriers
and the DC subcarrier
used within a

symbol.
renumbering sequence 6, 48, 37, 21, 31, 40, Used to
42, 56, 32, 47, 30, 33, renumber
54, 18,10, 15, 50, 51, clusters before
58, 46, 23, 45, 16, 57, allocation to
39, 35, 7, 55,25, 59, 53, subchannels.
11, 22, 38, 28, 19, 17,
3, 27, 12, 29, 26,5, 41,
49, 44, 9, 8, 1, 13, 36,
14, 43, 2, 20, 24, 52,4,
34, 0

13
Number of carriers per cluster 14 -
Number of clusters 60 -
Number of data subcarriers in each 24 -
symbol per subchannel
Number of subchannels 30 -
PermutationBase6 (for 6 3,2,0,4,5,1 -
subchannels)
PermutationBase4 (for 4 3,0,2,1 -
subchannels)

¾ Table 2-6 512-FFT OFDMA Downlink Carrier Allocations- PUSC


Parameter Values Comments
Number of DC subcarriers 1 Index 256
Number of Guard subcarriers, left 46 -
Number of Guard subcarriers, right 45 -
Number of Used Subcarriers (Nused) Number of all
421
including all possible allocated pilots
subcarriers
and the DC subcarrier
used within a

symbol.
renumbering sequence Used to
12, 13, 26, 9, 5, 15, 21,
renumber
6, 28, 4, 2, 7, 10, 18, clusters before

29,17, 16, 3, 20, 24, 14, allocation to


subchannels.
8, 23, 1, 25, 27, 22, 19,

11, 0

Number of carriers per cluster 14 -


Number of clusters 30 -
Number of data subcarriers in each 24 -
symbol
per subchannel
Number of subchannels 15 -
PermutationBase5 -
4,2,3,1,0

14
(For 5 subchannels)

¾ Table 2-7 128-FFT OFDMA Downlink Carrier Allocations- PUSC


Parameter Values Comments
Number of DC subcarriers 1 Index 64
Number of Guard subcarriers, left 22 -
Number of Guard subcarriers, right 21 -
Number of Used Subcarriers (Nused) Number of all
85
including all possible allocated pilots
subcarriers
and the DC subcarrier
used within a

symbol.
renumbering sequence Used to
2, 3, 1, 5, 0, 4
renumber
clusters before
allocation to
subchannels.

Number of carriers per cluster 14 -


Number of clusters 6 -
Number of data subcarriers in each 24 -
symbol
per subchannel
Number of subchannels 3 -

z Downlink PUSC subcarrier-allocation algorithm:


Each of the four-downlink PUSC permutations has the same number
of data subcarriers in one subchannel, which is equal to 48.
We specify the downlink PUSC subcarrier-allocation algorithm step by
step as the following:

15
Step 1: Dividing the subcarriers into physical clusters
First, we define a cluster structure as Figure 2.4:
According to the cluster structure and the symbol indexes, we divide the
used subcarriers into physical clusters. The number of clusters, NClusters , in
an OFDM symbol varies with the FFT sizes. Every 14 consecutive used
subcarriers are separated into a cluster and a physical index is given to the
cluster.

even symbols

odd symbols

data subcarrier
pilot subcarrier

Figure 2.4 Cluster Structure for Downlink PUSC.

Step 2: renumbering the physical clusters into logical clusters


After assigning a physical number to each cluster, we renumber the
physical indexes of clusters into logical cluster indexes by:
Logical index= renumbering sequence ((physical index +13*IDcell)
mod NClusters ),
where IDcell is the base station index.

Step 3: dividing the clusters into groups


All FFT-size allocation algorithms divide the clusters into six groups,
but the size of each group varies with the FFT sizes. We specify the
group sizes versus FFT sizes as following:

16
For 2048-FFT size:
Group0: cluster 0-23, Group1: cluster 24-39, Group2: cluster 40-63,
Group3:cluster6 4-79, Group4 cluster 80-103 and Group5: cluster
104-119

For 1024-FFT size:


Group0: cluster 0-11, Group1: cluster 12-19, Group2: cluster 20-31,
Group3: cluster 32-49, Group4 cluster 40-51 and Group5: cluster 52-59

For 512-FFT size:


Group0: cluster 0-9, Group1: null, Group2: cluster 10-19,
Group3: null, Group4 cluster 20-29 and Group5: null

For 128-FFT size:


Group0: cluster 0-1, Group1: null, Group2: cluster 2-3,
Group3: null, Group4 cluster 4-5 and Group5: null

Step 4: Allocating subcarriers to subchannel


In PUSC subcarrier allocation, pilot subcarriers within each cluster are
firstly allocated into OFDMA symbols and then the remaining data
subcarriers within those symbols are allocated group by group as
following procedure:
Subcarrier (κ , s) = Nsubchannel ⋅ nκ + {Ps [nκ mod Nsubchannel ] + IDcell}mod Nsubchannel (2.1)

where
Subcarrier (κ , s ) is the permutated subcarrier index of the present group
corresponding subcarrier κ in subchannel s, where κ is

17
running index 0…24 and s is running index0… NTS ,
where NTS is the total subchannels in one OFDMA
symbol.
N subchannel is the number of subchannels in each group,
N subcarrier is the number of subcarriers for one subchannel
allocating in one OFDMA symbol, i.e. N subcarrier = 24 ,
IDcell is identifies of the particular BS segment,
Ps [ j ] means the jth element obtained by rotating {Permutation-
Base} cyclically to the left s times,
nκ is defined as nκ = (κ + 13 ⋅ s ) mod N subcarrier

For 2048-FFT size:


From Table 2-4, the algorithm uses (2.1) to partition 24 data subcarriers
into each subchannel per OFDMA symbol with PermutationBase12, i.e.
N subchannel =12, for even numbered major groups, and PermutationBase8, i.e.

N subchannel =8, for odd numbered major groups.

For 1024-FFT size:


From Table 2-5, the algorithm uses (2.1) to partition 24 data subcarriers
into each subchannel per OFDMA symbol with PermutationBase6, i.e.
N subchannel =6, for even numbered major groups, and PermutationBase4, i.e.

N subchannel =4, for odd numbered major groups.

For 512-FFT size:


From Table 2-6, the algorithm uses (2.1) to partition 24 data subcarriers
into each subchannel per OFDMA symbol only with PermutationBase5,

18
i.e. N subchannel =5, for even numbered major groups.
For 128-FFT size:
From Table 2-5, the algorithm uses (2.1) to partition 24 data subcarriers
into each subchannel of each OFDMA symbol.

For example:
We set the parameters of subchannel as IDcell=2, FFT size=1024,
S = 3 ∈ subchannel index belongs Group0, N subchannel = 6 , N subcarrier = 24 .

We permute the subcarriers of 3-th subchannel over one OFDM symbol


by:
Subcarrier (κ , s) = Nsubchannel ⋅ nκ + {Ps [nκ mod Nsubchannel ] + IDcell}mod Nsubchannel , κ ∈{0,1,.., Nsubcarrier −1}

Take subcarrier (0, 3) for example:

Subcarrier (0,3) = 6 ⋅ nκ + {Ps [nκ mod 6] + IDcell}mod6


= 6 ⋅[((0 +13*3)mod 24)] + {P3[((0 + 13*3) mod 24)mod6] + 2}mod6
= 6 ⋅15 + {P3[3] + 2}mod6
= 90 + {40 + 2}mod6
= 90 + 6 = 96

Thus, the subcarrier index 0 of the 3-th subchannel would be mapping


into the 96-th data subcarrier index in Group0.
The details of the other subcarrier allocation algorithms are specified
in IEEE 802.16 standards, and omitted here.

19
2.2.2 Two-dimensional subchannel-mapping structure
OFDMA
symbol FEC
time Block RAU with 3 FEC blocks

Subchannel 1 FCH

RAU
Subchannel 2

Subchannel 3
DL
Data Burst
Preamble

Subchannel 4 MAP

UL
Subchannel MAP
Ns time
k k+1 k+2 k+3 k+4 k+5 k+6 k+7 k+8 k+9 ………………
Downlink Frame
frequency

Figure 2.5 Downlink Frame Structure in IEEE 802.16


Following the PUSC subcarrier allocation in the IEEE 802.16
standard, we employ a forward error control (FEC) block as a basic unit
for resource allocations. One FEC block (i.e. one subchannel in
OFDMA-based system) consisted of the subcarriers over two consecutive
OFDMA symbols. We define the number of several consecutive FEC
blocks as a “resource allocation unit (RAU)”. In this thesis, an RAU is set
to be the size of one-subchannel. The example of a downlink IEEE
802.16 frame structure is shown in Figure 2.5. Furthermore, we use
PUSC allocation with the sizes of 1024-FFT in the simulation. Every two
OFDMA symbols have 30 subchannels with 48 data subcarriers in each
subchannel for a downlink transmission mode in Wimax systems. A
frame structure is grouped into several data bursts that are composed of
consecutive RAUs in the frequency domain and FEC blocks in the time
domain. The data burst is built in the form of rectangularity with one type
of modulation and coding mode and assigned to only one user.

20
2.2.3 Specifications of the entire system
In this thesis, each MS can support one connection for convenience,
and each connection only belongs to one type of the three typical Internet
services. All connections communicated with the base station use the
time-division duplex (TDD) mode in the system. The performance of the
IEEE 802.16 wireless system is assessed in Point-to-Multipoint (PMP)
mode. We apply a buffer for each connection, which is connected with
the BS to request bandwidth, and operate a first-input-first-output (FIFO)
mode to transmit packets. In the time domain, the transmission time is
divided by a frame time consisting of fixed OFDMA symbols.
Adaptive modulation and coding (AMC) scheme, with a pair of one
specific error-correcting coding and one modulation, is available for each
subchannel in the frame duration under the above framework. We assume
six modulation and coding schemes (MCS) with M-ary quadrate
amplitude modulation (MQAM) and convolutional codes (CC) in Table
2-8. The channel quality of each MS can be either reported from MS or
estimated by BS itself. Suppose that the base station knows the channel
gains of all users’ subcarriers. When the algorithm is performed, the BS
selects an appropriate MCS for each data burst.
¾ Table 2-8 MCS Modes in The IEEE 802.16 Standard
MCS mode 1 2 3 4 5 6
Modulation QPSK QPSK 16QAM 16QAM 64QAM 64QAM
CC coding rate 1/2 2/3 1/2 3/4 2/3 3/4
FEC Block 96 144 192 288 384 432
size(bits)
Rn (bits/symbol) 1.0 1.5 2.0 3.0 4.0 4.5

21
Chapter 3
Cross-Layer Resource Allocation Algorithm
3.1 Concepts of the present priority algorithm
For the cross-layer resource allocation algorithm design, we only care
about three viewpoints, which are, maximum total throughput, fair
transmission among users, and satisfaction with the QoS requirements of
each user. However, it’s difficult to achieve all benefits in the
above-mentioned viewpoints. For example, suppose that one user with
bad fading channel quality asks for strict QoS requirement. If the base
station chooses the user to be transmitted, the total performance of the
system will degrade seriously; if the base ignores the user’s request, the
unfair transmission will occurs. In the following section, we will
introduce three scheduling algorithms, which have been proposed in the
recent years, and we illustrate the pros and cons of each algorithm. We
use the three different scheduling algorithms to be compared with our
proposed mechanism in this thesis.
The best channel first (BCF) scheduling with best channel first (BCF)
allocation algorithm, which is specified in section 3.2. This scheme
selects those unassigned RAUs with the best average received SNR to
one user. The algorithm may have an advantage of better total throughput,
but may not satisfy the QoS requirement of each connection on the
contrary. This algorithm is proposed in [10].
The second algorithm is the proportional fair (PF) scheduling with the
best channel first (BCF) allocation algorithm. This PF algorithm has been
designed to achieve equal average data rate between users. Because this

22
scheme only cares about impartial transmissions, it may cause low total
throughput and un-satisfaction with QoS requirements. The detail
algorithm is specified in section 3.3.
The third cross-layer scheduling algorithm is first proposed in [11].
The authors have designed a low complexity priority function to assign
the priority to each connection. In the algorithm, they have considered
about the QoS requirements with each mobile’s channel quality to figure
out a priority factor for each user. The method can look after both side of
performance and diverse QoS requirements. The more detail description
of the method is in section 3.4.

23
3.2 Best Channel First (BCF) scheduling
algorithm with Best Channel First (BCF)
subchannel allocation policy

I. Estimate of Data Volume in Video Buffer


The video packets have the strictest requirement in time latency
and minimum data rate. We consider received video packets are stored in
a video buffer at each video mobile. Delay will occur when the data in the
buffer is empty. Therefore, the base needs to estimate the remaining
video data at each buffer to avoid delay conditions. The scheduling
algorithm is published in [10].
First, the authors have defined Bi (t ) to be the remaining data (in bits)
th
in the video buffer of the i mobile at time frame t. The value of Bi (t )
will be updated before they perform the scheduling algorithm at each
frame time by:
Bi (t ) = max{Bi (t − 1) + Di (t − 1) − RirtpS iT f , 0} (3.1)
where RirtpS is the video playback rate of connection i, which the video
mobile informs the base at the initial connection setup time. The traffic
data, Di (t − 1) , is the data that are successfully transmitted to the mobile i at
frame time t-1. Each frame time duration is denoted by T f .

II. Transmission Priority of Traffic


Transitionally, most schedulers would give highest priority to rtPS
service, which is ahead of nrtPS, and BE services. However, the authors
brought up a new concept that the rtPS packets are treated as the same
priority level as nrtPS and BE services until the deadline condition of

24
rtPS service is achieved.
The priority mechanism has only two priority levels, that is, high and
low. The nrtPS and BE services are always set in low priority level; On
the other hand, the rtPS packets have low priority in default , and become
high priority level when the video buffers belonging to users are going to
be empty.
For any mobile i, the authors set two levels of threshold values to
decide which priority level of the mobile is and how many packets will be
transmitted for this mobile. First, the two threshold levels have been
denoted QTH 1 and QTH 2 (for this thesis, we set QTH 1 = 0.1 and QTH 2 = 0.05 ).
At each frame, the base decides the priorities for those video mobiles
according to the following three conditions:

Bi (t )
(1) > QTH 1 (3.2)
RirtPS iT f

The priority of all video packets for mobile i remains low.


Let M i (t ) be data in bits that will be transmitted to the mobile i at the
present frame and set default value of M i (t ) to be the size of one packet.
We define one packet containing 576 bytes for convenience.
(2) Q ≥ Bi (t ) > Q (3.3)
TH 1 TH 2
RirtPS iT f

The base gives high priority with the first one packet to the mobile i.
Thus, we can define M i (t ) = 576*8 (in bits).
(3) Q ≥ Bi (t ) (3.4)
TH 2
RirtPS iT f

25
The base station gives high priority for the first two packets to the video
mobile i. At the same way, we assign M i (t ) = 576*8* 2 (in bits).
Every packet, which is given high priority, will be mapped as the
amount of resource allocation units (RAUs). In this thesis, one RAU
represents one PUSC subchannel in Wimax OFDMA network. We define
N F to be the total number of RAUs within a frame structure. Ri (t ) is the

number of RAUs that are assigned to high-priority packets of mobile i at


frame time t.

III. Best Channel first subchannel allocation


After the base has separated packets in transmission queues for all
mobiles into two groups, i.e. low-priority and high-priority levels, the
scheduler allocates the RAUs depending on the downlink channel quality
of each mobile and performs the resource allocation from the
high-priority group to the low-priority group in order.
If there are still some available RAUs after allocating for high-priority
packets, the allocation mechanism will distribute the available RAUs to
low-priority packets until all of the RAUs in a frame are all assigned.

Performing by iterations in each group, the base choose one mobile i*


with the best average SNR of unsigned subchannels over frame t, that is:

i* = arg max{SNRi (t )} (3.5)


i

where SNRi (t ) represents the average SNR of the total available


subchannels over frame t.
After choosing a mobile i* to be transmitted at the MAC, the base

26
continues allocating RAU one by one for mobile i* until the
transmission data at the present frame duration is larger than M i (t ) .
*

The allocation algorithm chooses the best average-SNR from the


available RAUs for mobile i* as:
(3.6)
k * = arg max{SNRi* (κ , t )}
κ
where
κ is the index of RAU
SNRi* (κ , t ) is the average SNR value of the κ th RAU at frame t for
mobile i*.
The system operation of the BCF scheduling with BCF allocation
scheme is illustrated in Figure 3.1.

27
nrtPS connections
rtPS connections
Be connections

B (t )
i
> QTH 1
R ⋅Tf
i
rtPS
else

High Priority Low Priority


connections connections
Choose a new i*
Choose a new i*
i = arg max{SNRi (t )}
*
i i*
High-priority allocation
allocate until i*
M i* ( t ) ≤ 0
By
k = arg max{SNRi* (κ , t )}
*
κ

If
(1) all RAUs over this frame
are assigned else
or i * = arg max{SNRi (t )}
i
(2)no high-priority packets
waiting for tansmission Low-priority allocation
allocate until

no high-priority packets
M i* ( t ) ≤ 0
waiting for tansmission By
else
k = arg max{SNRi* (κ , t )}
*
κ

If all RAUs over this frame If all RAUs over this frame
are assigned are assigned

End

Figure 3.1 Flowchart of Best Channel First (BCF) scheduling algorithm


with Best Channel First (BCF) subchannel allocation policy

28
3.3 Proportional Fair (PF) scheduling algorithm
with Best Channel First (BCF) subchannel
allocation policy

I. Estimate of Data Volume in Video Buffer and


Transmission Priority of Connections
The Proportional Fair (PF) algorithm proposed in [10] has the same
concept of the high-priority and the low-priority assignment as BCF
scheduling describing in section 3.2. That is briefly specified as the
following:
At each frame, the base decides the priorities for video mobiles
according to the following three conditions:
Bi (t )
(1) > QTH 1 (3.7)
RirtPS ⋅ T f

The priority of all video packets for mobile i remains in low priority.
Let M i (t ) be data in bits that need to be transmitted for mobile i and set a
default value of M i (t ) to be one-packet size, which we define one packet
containing 576 bytes for convenience.
(2) Q ≥ Bi (t ) (3.8)
> QTH 2
RirtPS ⋅ T f
TH 1

The base gives high priority with the first one packet to mobile i.
Thus, we can define M i (t ) = 576*8 (in bits).

(3) Q ≥ Bi (t ) (3.9)
RirtPS ⋅ T f
TH 2

The base station gives high priority with the first two packets to the video

29
mobile i. As the same way, we assign M i (t ) = 576*8* 2 (in bits).
II. Best Channel first subchannel allocation

Proportional Fair (PF) scheduling algorithm is designed to achieve fair


transmission between any mobile regardless of its channel quality.
Consider ri (t ) to be the instantaneously transmitted data rate of mobile i
at frame t and Ri (t ) to be the estimate of the average data rate of mobile i
from initial time to the present frame t. Thus, the average rate is updated
frame by frame as:
Ri (t + 1) = (1 − δ ) ⋅ Ri (t ) + δ ⋅ ri (t ) (3.10)
In this thesis, we set the update-coefficient δ = 0.001 .
Performing as the mechanism in section 3.3, we use the following
formula to choose a mobile i* for transmission. However, not similar to
the BCF scheduling considering maximizing the total throughput, the PF
scheduling is an algorithm, which cares about the fair transmission for
every user connected on the network. The base selects a suitable mobile
i* to transmit as the following:

ri (t ) (3.11)
i* = arg max{ }
i Ri (t )
*
After selecting a fitted mobile i , the mapping block allocates the
traffic data belonging for mobile i* until M i (t ) ≤ 0 , and then continues to
*

seek for a new mobile i* if the present frame has any available RAUs
unallocated. The base also uses best subchannel first algorithm to allocate
subchannel for a selected mobile i* by:

κ * = arg max{SNRi (κ , t )} *
κ
(3.12)

30
Where κ is the available subchannel index of mobile i* .

nrtPS connections
rtPS connections
Be connections

B (t )
i
> QTH 1
R ⋅ Tf
i
rtPS
else

High Priority Low Priority


connections connections
Choose a new i*

ri (t ) Choose a new i*
i* = arg max{ }
i Ri (t ) i*
High-priority allocation
allocate until i*
M i* ( t ) ≤ 0
By
k = arg max{SNRi* (κ , t )}
*
κ

ri (t )
If i* = arg max{ }
(1) all RAUs over this frame i Ri (t )
are assigned else
or
(2)no high-priority packets
waiting for tansmission Low-priority allocation
allocate until

no high-priority packets
M i* ( t ) ≤ 0
waiting for tansmission By
else
k * = arg max{SNRi* (κ , t )}
κ

If all RAUs over this frame If all RAUs over this frame
are assigned are assigned

End

Figure 3.2 Proportional Fair (PF) scheduling algorithm with Best Channel
First (BCF) subchannel allocation policy

31
3.4 Priority function (PRF) for scheduling
algorithm with Best Channel First (BCF)
subchannel allocation
The scheduling algorithm was published in [11]. Initially, the
mechanism was designed for OFDM systems. The authors introduced a
low-complexity priority function (PRF) for multiple connections with
diverse QoS requirements. They defined the PRF for each connection and
updated it dynamically depending on the wireless channel quality and
QoS satisfaction. In this thesis, we will modify the PRF to be fitted for
OFDMA system. At the MAC layer, the scheduler simply assigns the
order of mobiles per frame by deciding the priority of each connection as
following:

i* = arg max{φi (t )} (3.13)


i

where φi (t ) is the PRF function for connection i, at frame t, and we will


specify below. If any connections have the same value of φi (t ) , the
scheduler will randomly select one of them. In the following, we
introduce how to update the PRF φi (t ) for those connections with
different QoS requirements:
I. For rtPS connections:

⎧ Ri (t ) 1
⎪ β rt ⋅ R ⋅ F (t ) , if Fi (t ) ≥ 1, Ri (t ) ≠ 0
⎪⎪ N i

φi (t ) = ⎨ β rt , if Fi (t ) < 1, Ri (t ) ≠ 0
⎪0
(3.14)
, if Ri (t ) = 0

⎪⎩

where,

32
β rt is the rtPS-class coefficient in the range of [0,1]

Fi (t ) is the rtPS-rate satisfaction indicator, defined as

B (t ) (3.15)
Fi (t ) = i

R ⋅ Tf
i
rtPS

with X i (t ) being the estimated video data in mobile i at frame time t.


The definition is the same as Bi (t ) = max{Bi (t − 1) + Di (t − 1) − RirtpS iT f , 0} in
section 3.2.
Ri (t ) is the AMC level according to average SNR of the remaining

available RAUs of mobile i at frame t.


RN is the maximum AMC level in the system, and we set RN = 4.5 in

this thesis.
Depending on the PRF for rtPS connections, if Fi (t ) < 1 , that is, the
packets of connection i need to be sent immediately in order to avoid
playback delay and packet drop and the PRF will be β rt , which the
highest priority is given. However, if Fi (t ) ≥ 1 , the video data in mobile i
are enough to be played in next frame duration. Then, the priority is
1
affected on the value of. F (t ) and the channel quality factor :
i

Ri (t )
∈ [0,1]
RN
(3.16)
The channel factor is the normalized channel quality for mobile i. The
rtPS connections, which satisfy their QoS requirements, have higher
priority depending on high-received SNR. Therefore, the values of PRF
for rtPS connections are φi (t ) ∈ [0, β rt ] . Finally, we can notice that the
priority of each rtPS connection is according to its channel quality and
estimated volume of rtPS data that are stored in the buffer of the mobile
station.

33
II. For nrtPS connections:
The PRF for nrtPS has the same definition as

⎧ Ri (t ) 1
⎪ β nrt ⋅ ⋅ , if Fi (t ) ≥ 1, Ri (t ) ≠ 0
RN Fi (t )
⎪⎪ (3.17)
φi (t ) = ⎨ β nrt , if Fi (t ) < 1, Ri (t ) ≠ 0
⎪0 , if Ri (t ) = 0

⎪⎩

When an nrtPS connection sets up the connection initially, it will ask


for a minimum transmission rate to the base. That is, the average data rate
of an nrtPS connection should be greater than the minimum required rate.
Thus, we define the average data rate for nrtPS connection of mobile i
as χ i (t ) . At the iteration, we update χ i (t ) as
ci (t ) (3.18)
χi (t ) = χ i (t − 1) ⋅ (1 − δ ) + δ ⋅
Tf

where the updated factor is δ = 0.001 and ci (t ) is the traffic data that are
transmitted to mobile i successfully at frame t. We can define the rate
satisfaction indictor as
χ i (t ) (3.19)
Fi (t ) =
RinrtPS

The PRF for each nrtPS connection also has nrtPS-class coefficients rate
Ri (t )
β nrt ∈ [0,1] satisfaction indictor Fi (t ) , and channel quality factor.
RN
If Fi (t ) ≥ 1 , the average transmission rate is satisfied, and its priority
1
depends on the effect of channel quality, and β nrt .
Fi (t )

If Fi (t ) < 1 , the scheduler will give the connection a highest priority in


nrtPS class to satisfy the requirement as soon as possible.
Thus, the boundary of the PRF in this class is φi (t ) ∈ [0, β nrt ]

34
III. For BE connections:
The BE service has no delay requirements and no minimum data rate
guarantee, thus the PRF for a BE connection may simply be
Ri (t )
φi (t ) = β BE ⋅ (3.20)
RN

where β BE is the BE-class coefficient and has a value in the range of [0,1].
Because the normalized channel quality and the BE-class coefficient are
both smaller than 1, the priority function of a BE connection φi (t ) is
in [0, β BE ] .

IV. Coefficient setting:


If multiple connections have the same value of φi (t ) , the scheduler will
randomly choose one of them. In each iteration, the base select one user
i* for transmission at the MAC, and the mapping block will use the best
channel first (BCF) scheme to distribute RAUs to mobile i* until the data
volume for mobile i* in this frame has achieved one-packet size or no
packet belonging to mobile i* waits for transmission in the buffer at the
base station. The different classes of those three service coefficients play
important roles in giving discriminating priorities for each connection. If
we want the system to be strict priority assignment, i.e., the priority order
to be rtPS>nrtPS>BE classes, we can set the service-class coefficients to
be β rtPS > β nrtPS > β BE . In other words, we can change the priority order
between those services. In this thesis, we set the service-class coefficients
under the constraint:
β rtPS = 1.0 > β nrtPS = 0.8 > β BE = 0.6 . (3.21)

35
rtPS connections Be connections nrtPS connections

Ri (t )
φi (t ) = β BE ⋅
RN

⎧ Ri (t) 1 ⎧
⎪βrt ⋅ R ⋅ F(t) , if Fi (t) ≥1, Ri (t) ≠ 0
Ri (t) 1
⎪βnrt ⋅ R ⋅ F (t) , if Fi (t) ≥ 1, Ri (t) ≠ 0
⎪⎪ N i
⎪⎪ N i
φi (t) = ⎨βrt , if Fi (t) <1, Ri (t) ≠ 0 φi (t) = ⎨βnrt , if Fi (t) < 1, Ri (t) ≠ 0
⎪0 ,if Ri (t) = 0 ⎪0
⎪ , if Ri (t) = 0

⎪⎩ Calculate PRF ⎪⎩
φi (t )

Select a new
Choose a user i* i*
i* = arg max{φi (t )}
i

If else

1.
If the total data that transmit to mobile
Mapping Block i* is smaller than one-packet size.
Best channel First (BCF) OR
2. No packets in buffer for mobile i* at
the base

If else

If all RAUs over this frame


have been assigned

If all RAUs over this frame


have been assigned

End

Figure 3.3 Priority Function (PRF) scheduling algorithm with Best


Channel First (BCF) subchannel allocation policy

36
3.5 Modified capacity priority algorithm

required data per Connection Channel Quality


FEC block indexes
rtPS connections 1 2 3 4 5 6 7 8 9 10
i∈A 1
rtPS data Data
2D Mapper 2
i ∈B burst
3
nrtPS connections Active 4
i∈A Scheduler data slot Mapping 5
Connection nrtPS data
List i ∈B & 6
power assignment 7
BE connections BE data i ∈ ActiveList
Frame Structure

subchannel

Connection Mapping Turning

A: The connections in highest channel quality area, that is, the average SNR of the connections in the highest level of modulation
B: The connections that are within active List but not belong to A, that is, A ∩ B = ∅

Figure 3.4 A Brief Topology of Our Cross Layer System

In the above section 3.1, we address that there is a trade-off between


achievement of total throughput and QoS requirements. If we want to
increase the transmission data rate, the users with highest average
received SNR values should be selected; if we want to achieve all the
users’ QoS requirements, we should transmit the traffic data regardless of
users’ channel conditions but may decrease the performance.
In this section, we modify the above algorithms and develop our
cross-layer scheduler to improve the network performance and satisfy
diverse requirements, respectively.
Figure 3.4 illustrates a brief topology of the cross layer system in this
thesis. At the AMC, each connection with either rtPS, nrtPS or BE
service asks the base for some bandwidth. Then, the base assigns
differentiable connection identification (CID) index for this connection.
37
The information of a connection will be written into the active list only if
the buffer for the connection at the base is not empty. In our model, the
scheduler decides the priority of a connection according to the channel
quality, the service type, and the required transmission data (in bits) per
frame. After the scheduling block, one user would be selected from our
algorithm and the following is the mapping block at the PHY layer. The
user’s transmission data would be translated directly to some RAUs,
determined by the mapping block. By estimating the average SNR
measurements of each mobile, the mapping block uses our allocation
algorithm to map suitable RAUs for the selected mobile.

I. QoS-satisfied factor ϕR (i,τ ) for each connection


In order to share the resource more fairly and to satisfy the QoS service
requirements, the base should know how many data each connection
needs to transmit at the present frame, or it would not achieve its QoS
requirement. For example, the PRF scheduler designed in section 3.4
gives the same service-class factor β to each QoS-unsatisfied connection.
Therefore, if there are two connections with the same service-class factor,
and the connection A has more data needing to be transmitted than the
connection B, according to the PRF algorithm, the base would randomly
choose one user within the connection having the same service-class
factor. At the result, the PRF algorithm would uncertainly choose the
connection A. Thus, we design a QoS-satisfied factor ϕR (i,τ ) to estimate
the average required traffic data for connection i at frame period τ .

For an rtPS connection, the mobile stores the received packets and

38
then plays the packets in order. When video data are getting empty, the
continuous playback of video streaming begins to lag. Therefore, the BS
needs to transmit sufficient packets to avoid the buffer being empty. The
scheduler estimates the amount of video data Cv (i,τ ) in the buffer of each
mobile at frameτ by:

Cv (i,τ ) = Cv (i,τ − 1) + Dv (i,τ − 1) − Rv (i ) ⋅ TF (3.22)

The QoS-satisfied factor ϕR (i,τ ) for rtPS connection i is:


ϕ R (i,τ ) = max{Rv (i ) ⋅ 2TF − Cv (i,τ ), 0}
(3.23)
where
Rv (i ) is the minimum data rate of rtPS connection i.

TF is the length of frame time in the system.

For an nrtPS connection i, its average transmission rate should be


greater than the minimum reserved rate R f (i ) . We compute the

QoS-satisfied factor ϕR (i,τ ) per frame by:

R f (i ) − C f (i, µi )
ϕ R (i,τ ) = + ψ f (i ) (3.24)
N f (i,τ )

where
R f (i ) is the minimum transmitted data rate for ith nrtPS connection.

µi is past time in sec from the connection set up

i.e. if the initial connection-set time is si and then µi = floor (t − si )

C f (i, µi ) is the traffic data that has been transmitted successfully to the

39
µi sec.
th
mobile i within

ψ f (i ) the accumulated data from the initial setting time to µi − 1 sec,

should be transmitted in order to satisfy QoS requirement for connection


i.
After all RAUs in τ frame block are all assigned, we update C f (i, µi )

by:
Cf (i, µi ) = Cf (i, µi ) + D(i,τ ) i ∈ index in ActiveList (3.25)

where D(i,τ ) is the actually transmitted bits for connection i to its mobile
in τ frame block.
⎧ ⎡ t − si ⎤ ⎡ 1 ⎤
⎪min( N f (i,τ −1) −1,1) if mod(⎢ ⎥ , ⎢ ⎥) ≠ 0
⎪ ⎢ TF ⎥ ⎢TF ⎥
N f (i,τ ) = ⎨
⎪⎡ 1 ⎤ ⎡ t − si ⎤ ⎡ 1 ⎤
⎪⎢ T ⎥ if mod(⎢ ⎥ , ⎢ ⎥) = 0
⎩⎢ F ⎥ ⎢ TF ⎥ ⎢TF ⎥

i ∈ index of nrtPS connections (3.26)

where ⎡⎢ x ⎤⎥ represents the smallest integer equal to or smaller than x. be

⎡t − s ⎤ ⎡ 1 ⎤
if mod( ⎢ i ⎥ , ⎢ ⎥ ) = 0 for i ∈ index of nrtPS connections occurs, the
⎢ TF ⎥ ⎢ TF ⎥

base updates the value in C f (i, µi ) and ψ f (i ) by:

ψ f (i ) = ψ f (i) + [ R f (i ) − C f (i, µi )] (3.27)

C f (i, µi + 1) = 0 (3.28)

For each BE connection, guaranteeing no throughput or delay causes


serious latency and unfair resource management to the users. In this thesis,
we design a scheme to notice the base not to forget the feeble BE service.
Th (i ) denotes the deadline for the base to transmit all BE data to the

40
connection i. The QoS-satisfied factor ϕR (i,τ ) for connection i at the
frame τ is:
Rh (i,τ )
ϕ R (i,τ ) = (3.29)
N h (i,τ )

⎧⎡ Th (i) ⎤
⎪ if t = 0
, where Nh (i,τ ) = ⎨⎢⎢ TF ⎥⎥ (11)

⎩min( Nh (i,τ − 1) − 1,1) if t > 0

Rh (i,τ + 1) = Rh (i,τ ) − D (i,τ ) (3.30)

Rh (i,τ ) is the remaining data waiting for transmission at the frame τ .

II. Hierarchical Scheduling Set


After the base computes the QoS-satisfied factor ϕR (i,τ ) for each
connection selected into the Active List. We design a hierarchical
scheduling algorithm to choose suitable users for transmission. We
propose a hierarchical cross-layer scheduling algorithm to improve the
efficiency of the system. We divide the connections on the active list into
three parts:

Part I

A set of the rtPS and the nrtPS connections whose average SNRs are
higher than the threshold of best MCS level, i.e. 64QAM 3/4, are grouped
into part I

Part II
A group of rtPS connections except those in part I.

Part III

41
The remaining connections except from those in part Part II and I.

III. Scheduler Design


To improve the total throughput, each user’s channel quality and the
traffic volume of each connection are necessary to be considered. The
algorithm consists of two steps, priority scheduling and 2D resource
allocation. First, the priority assignment function, which is updated
dynamically depending on the wireless channel quality and required data
volume, is defined as:
i* = arg max{ϕ R (i,τ ) ⋅ Ω j (i,τ )}
i

(3.31)
where Ω j (i,τ ) is the channel quality coefficient and expressed as follows:

ζ j (τ )
Ω j (i , τ ) = (3.32)
1 M
∑ ζ l (τ )
M − 1 l =1
l≠ j

The channel quality coefficient quantifies the channel condition by the


average capacity ζ j of all unassigned RAUs to the mobile j.
That is,

1 N (3.33)
ζ j (τ ) =
N
∑ log
k =1
2 (η j (k ,τ ) + 1)

where
N is the number of all unassigned RAUs in the frame block τ .
η j (k ,τ ) is the average SNR value of k-th RAUs of the mobile j.

By introducing the parameter Ω j (i,τ ) , the channel conditions will

42
significantly affect the priority decision, especially in the case of high
SNR.
The key idea of the 2D mapping algorithm is to allocate bits and the
required power efficiently and maximize the total throughput at the PHY.
After the proposed scheduler has selected a suitable connection i* for the
mapping block, the mapping block will choose a RAU reserved for the
connection i*. According to the IEEE 802.16 standard, we assume that a
maximum of 30 subchannels can be used simultaneously in a cell. The
mapping algorithm is shown as follows:
We use one-order stage approach, which is proved in section 3.6 to
assign RAUs to users by

M
1
κ = arg max{ci (κ ) −
*
[∑ ca (κ )]} (3.34)
κ M − 1 a =1
a ≠i

Where
κ is RAU index in frame structure
ci (κ ) is average capacity of RAU index k belonging to user i

43
3.6 Subchannel allocation algorithm for
OFDMA systems

3.6.1 Problem formulation

In order to obtain the cross-layer maximum throughput, we extend


the orthogonal subcarriers to the infinite-horizon case in frequency
domain. There are M users in a single cell with an entire bandwidth B.
Consider all system bandwidth divided into M non-overlapping frequency
sets. We assign a frequency set R j for user i. Then, we assign G
consecutive subchannels as an allocation unit, which is one RAU. Then
1 lG
Ci (k ) = ∑ log2 (1 + pi (a) ρi (a))
G a =1+ (l −1)G
(3.35)

Let Ci (k ) be the average capacity for the i-th user at the k-th RAU
with average transmission power pi (a) and the average signal-to-noise
ratio ρi (a) , where a denotes a subchannel index. Therefore, we can
express the transmission rate of user i as
ri = ∫ Ci (k )dk (3.36)
Ri
In order to maximize system throughput, we denote the utility
function U (r ) = r and U (ri ) = ri to be a utility for user i. We obtain a
formula from the following theorem for the optimal cross-layer
subchannel assignment, and prove it.
Theorem: For a system with M users, if the RAU assignment is
optimal for all users, i.e. Ki* , i ∈{1, 2,..., M } is optimal, and then we
formulate an N-order stage approach as:
N N M N N
1
∑ Ca (ka* ) ≥
a =1
[∑
M − 1 a =1

b =1
Cb (ka* ) + ∑∑ Cc (ka* )]
a =1 c =1
b ≠ all a∈N c≠a
(3.37)

44
Proof of theorem:
Case 1: one-order approach:

For system with two users


Assume that if the base chooses an optimal RAU index κ for user 1, the
*

throughput of the system would get to maximum value. However, if any


other index κ is chosen, the performance will decrease. Therefore, we
define ∆r1 = C1 (κ )∆κ as the decrease value of data rate for user 1,
where ∆k → 0 . Thus, the utility function can formulate as:
U1 (r1* − ∆κ ⋅ C1 (κ )) + U2 (r2* + ∆κ ⋅ C2 (κ )) ≤ U1 (r1* ) + U2 (r2* ) (3.38)
which is equal to
U2 (r2* + ∆κ ⋅ C2 (κ )) + U2 (r2* ) ≤ U1 (r1* ) −U1 (r1* − ∆κ ⋅ C1 (κ ))

dividing by ∆κ , we obtain
U2 (r2* +∆κ ⋅ C2 (κ )) U2 (r2* ) U1(r1* ) U1(r1* −∆κ ⋅ C1(κ ))
− ≤ −
∆κ ∆κ ∆κ ∆κ

according to ∆γ i (κ ) = ∆κ ⋅ Ci (κ ) , ∆κ → 0 ⇒ ∆ri → 0 ,i=1,2


U2 (r2* + ∆r2 ) −U2 (r2* )
lim c2 (k)
∆r2 →0 ∆r2
U1(r1* ) −U1(r1* −∆r1)
≤ lim c1(κ )
∆r1 →0 ∆r1

since Ui (ri ) = ri ⇒Ui '(ri ) = 1, we define that κ will satisfy c2 (k) ≤ c1(κ )
* * *

For system with three users


1
{U1 (r1* − ∆κ ⋅ C1 (κ )) + U2 (r2* + ∆κ ⋅ C2 (κ )) + U3 (r3* )} +
2
1
{U1 (r1* − ∆κ ⋅ C1 (κ )) + U2 (r2* ) + U3 (r3* + ∆κ ⋅ C3 (κ ))}
2
≤ U1 (r1* ) + U2 (r2* ) + U3 (r3* )
which is equal to
[U2 (r2* +∆κ ⋅ C2 (κ)) −U2 (r2*)] +[U3(r3* +∆κ ⋅ C3 (κ)) −U3 (r3*)]
≤ 2[U1(r1*) −U1(r1* −∆κ ⋅ C1(κ))]

45
The same way in system with two users, we obtain that κ satisfy
*

1
c1 (κ ) ≥ [c2 (κ ) + c3 (κ )]
2 (3.39)
It is the same way to approach one-order stage with M user, and then we
have a general formula of one-order stage when assigning a slot to user i:
M
1
ci (κ * ) ≥ [∑ ca (κ * )]
M − 1 a =1
a ≠i
(3.40)

Case 2: two-order approach:

For system with three users

Let κ1 and κ 2 be optimal granularity indexes for user 1and user 2. If we


* *

select another κ1 , κ 2 that will decrease the total data rate of system. Thus,
1
{U1 (r1* − ∆κ1 ⋅ C1 (κ1 )) + U2 (r2* − ∆κ2 ⋅ C2 (κ2 ) + ∆κ1 ⋅ C2 (κ1 ))
4
+ U3 (r3* + ∆κ2 ⋅ C3 (κ2 ))} +
1
{U1 (r1* − ∆κ1 ⋅ C1 (κ1 )) + U2 (r2* − ∆κ2 ⋅ C2 (κ2 ))
4
+ U3 (r3* + ∆κ1 ⋅ C3 (κ1 ) + ∆κ2 ⋅ C3 (κ2 ))} +
1
{U1 (r1* − ∆κ1 ⋅ C1 (κ1 ) + ∆κ2 ⋅ C1 (κ2 ))) +
4
U2 (r2* − ∆κ2 ⋅ C2 (κ2 ) + ∆κ1 ⋅ C2 (κ1 )) + U3 (r3* )} +
1
{U1 (r1* − ∆κ1 ⋅ C1 (κ1 ) + ∆κ2 ⋅ C1 (κ2 ))) +
4
U2 (r2* − ∆κ2 ⋅ C2 (κ2 )) + U3 (r3* + ∆κ1 ⋅ C3 (κ1 ))}
≤ U1 (r1* ) + U2 (r2* ) + U3 (r3* )
(3.41)

which is equal to

46
U1 (r1* ) −U1 (r1* − ∆r11 ) U2 (r2* ) −U2 (r2* − ∆r22 )
4 c1 (κ1 ) + 4 c2 (κ2 )
∆r11∆κ2 ∆κ1∆r22
U3 (r3* + ∆r31 + ∆r32 ) −U3 (r3* + ∆r31 )
≥ c3 (κ2 ) +
∆κ1∆r32
U3 (r3* + ∆r31 ) −U3 (r3* ) U (r* + ∆r32 ) −U3 (r3* )
2 c3 (κ1 ) + 3 3 c3 (κ2 ) +
∆r31∆κ2 ∆κ1∆r32
U1 (r1* − ∆r11 + ∆r12 ) −U1 (r1* − ∆r11 )
2 c1 (κ2 ) +
∆κ1∆r12
U2 (r2* − ∆r22 + ∆r21 ) −U2 (r1* − ∆r22 )
2 c2 (κ1 )
∆r21∆κ2

∆rij = ∆κ j ci (κ j )
where we define
The above inequality can be simplified
c1(κ1 ) c2 (κ2 ) c3 (κ2 ) c3 (κ1 ) c1(κ2 ) c2 (κ1 )
4 +4 ≥2 +2 +2 +2
∆κ2 ∆κ1 ∆κ1 ∆κ2 ∆κ1 ∆κ2

where we define ∆κi = 1 for the digital system because the minimum
interval of granularity index is one.
That is,
1
c1(κ1 ) + c2 (κ2 ) ≥ [c3 (κ2 ) + c3 (κ1 ) + c1(κ2 ) + c2 (κ1 )]
2 (3.42)

where κ1 and κ 2 satisfies the above inequality.


* *

By the similar method, we approach a two-order stage with M users as


general formula:
1 2 M 2 2
c1(κ ) + c2 (κ ) ≥
* *
[∑∑ cb (κa ) + ∑∑cb (κa* )]
*

M −1 a=1 b=1
1 2
a=1 b=1
b≠1,2 b≠a
(3.43)
Case 3: N-order approach:
N N M N N
1
∑ Ca (ka* ) ≥
a =1
[∑
M − 1 a =1

b =1
Cb (ka* ) + ∑∑ Cc (ka* )]
a =1 c =1
(3.44)
b ≠ all a∈N c≠a

The approach is similar with the above proof.

47
Chapter 4 Simulation Results

4.1 Simulation models for multi-user OFDMA

4.1.1 Parameters of MSs Generation Model


We perform a single cell to be hexagonal in shape with the length of
each side being one km. M users connect with the BS which is located at
the center of the cell. Each MS is randomly distributed in service area of
the cell. We consider two types of mobility for users that are pedestrian
and vehicular modes. The probabilities of the two mobility types are 0.5
identically at initialization. The moving direction is changed with
probability 0.05 per 10 sec. The moving speeds of the vehicular-type and
the pedestrian-type users are both generated as uniformly distribution.
The distribution of the pedestrian-type is with a mean of 3 (km/h) and a
variance of 0.25 (km/h), and that of the vehicular-type is with a mean of
60 (km/h) and a variance of 100 (km/h). Finally, we change the moving
speed of all MSs every 10 sec. We follow [10] to set all the parameters of
the MSs.

4.1.2 Channel Models of IEEE 802.16 OFDMA


systems
In this thesis, we evaluate the path loss and the fast fading channel in
a Multipath environment.

48
Path Loss:
The path loss is defined in [14] and modeled by
PL(d ) = 128.1 + 37.6 log10 (d ) [dB] (4.1)

where d is the distance between the BS and a MS in kilometer. We


update the path loss every frame time.

Fast Fading:
The short-term fading gain is generated by the Jakes’ fading model
in [15] to be a single path. The transmission of the system works at the
carrier frequency 2 GHz. We update the short-term fading gain every
frame time. The moving speed of each MS will be illustrated in section
IV-E.

Multipath:
The Stanford University Interim (SUI) channel models with six
different scenarios (from SUI1 to SUI 6) are widely used for simulating
the multi-path environments of IEEE 802.16 systems [16]. We select SUI
2 channel model to simulate the multi-path environment with the
parameters in the omni-antenna situation in Table 4.1. The perfect
channel estimation is performed in the system.
¾ Table 4.1 SUI 2 Channel Model Parameters
Tap1 Tap2 Tap3 Unit
Delay 0 0.5 1 µs
Power (omni. ant.) 0 -12 -15 dB

49
4.1.3 AMC Design at the PHY
We consider the downlink of a single cell consisting of M users. The
channel conditions are assumed to be known at the BS. x j [γ ] denotes a
set of data symbols and ρ j [γ ] is the transmission power at the kth
Η j [γ ]
subcarrier of the jth user in a frame. is an estimate of the channel
transfer function combined with path loss and fast fading for any mobile i
of subcarrier k. The receive signal can be written as
R j [γ ] = ρ j [γ ]⋅ | H j [γ ] | ⋅x j [γ ] + η n (4.2)

ηn denotes the additive white Gaussian noise with zero mean and
B
variance , σ 2 = N 0 where B is assumed to be total available bandwidth
N
with N subcarriers per OFDMA symbol and N0 is the noise
power spectral density. Therefore, we can compute the signal-to-noise
ratio (SNR) for γ th subcarrier of jth user by

2
ρ j [γ ] H j [γ ]
ζ j [γ ] = (4.3)
B
N0
N

We define the SNR per slot to be the average SNR of the subcarriers
in κ th subchannel. That is

Ns ζ j (γ )
ζ j (κ ) = ∑ (4.4)
γ =1 Ns

where Ns is the total data subcarriers in the κ th subchannel.


We use the below formulas for FFT and IFFT to make sure that the value
of the computed SNR are identical in both the time domain and the
frequency domain.

50
1 N
X (k) = ∑
N j=1
x( j)ωN ( j−1)(k −1) (4.5)

N
1
x( j ) =
N
∑ X (k )ω
k =1
N
− ( j −1)( k −1)
(4.6)

where ωN =exp ( −2 π i ) / N

The BS selects the MCS modes (see Table I) of a MS using the


average SNR valued from the BS. The SNR requirement depending on
the different modulation types for a BER less than 10-6 is defined from
the figures 4 to 5 in [10]. We separate the entire SNR range over
transmission channels into N+1 (N=6) non-overlapping consecutive
intervals with boundary threshold defined in Table 4.2 and denoted as
{ζ n }nn ==1N which represents SNR threshold value from high to low according

to n=6 to n=1 for modulation modes.


The modulation mode n is selected when the average SNR of a specific
subchannel ζ ∈ [ζ n , ζ n +1 ) for n = 1,..., N -1. When ζ < ζ 1 occurs, no data
will be sent because of the critical fading channel condition. On the other
hand, when ζ ≥ ζ N occurs, the MCS selector will choose mode N to be
transmitted.
¾ Table 4.2 Required SNR for Different Modulations
Modulation and Coding Required SNR
QPSK 1/2 6 dB
QPSK 3/4 9 dB
16 QAM 1/2 12 dB
16 QAM 3/4 18 dB
64 QAM2/3 21 dB
64 QAM3/4 30 dB

51
4.1.4 Physical Layer parameters
We consider only a single-hexagonal cell for convenience. The
length of each side of one cell is 1 km. A base is located at the center in
the cell. In Table 4.3, we show the physical layer parameter values in the
simulation of the system.
¾ Table 4.3 Physical Layer Parameters
Physical Layer Parameter Value
Bandwidth (M Hz) 7
Number of data subcarriers per subchannel per 24
OFDM symbol
Number of data subcarriers per subchannel 48
Number of subchannels per OFDM symbol 30
Ratio of cyclic prefix time to useful time per 1
OFDM symbol 4
OFDM symbol per frame 30
Thermal noise density (dBm/Hz) -164
Total transmit power of a base (dB) 15
Number of subcarrier in one symbol 1024
Frame time (ms) 4.8
Central frequency (G Hz) 2

4.1.5 Transmission Power


In the simulation, we set 15watt for the total transmission power of a
base station. Among the total power, 80% is used for data subcarriers and
20 % is allotted to the pilot subcarriers in an OFDMA symbol. At the

52
initialization, the transmission power of a data subcarrier is set to (80/Nd)
watt, where Nd is the number of the total data subcarriers in a frame
structure.

53
4.2 Simulation results for multi-user OFDMA

7
x 10
2
Proposed scheduling with BCF allocation
1.9 Proposed scheduling with proposed allocation
BCF scheduling with BCF allocation
1.8 PF scheduling with BCF allocation
PRF scheduling with BCF allocation
1.7
Total Throughput (bps)

1.6

1.5

1.4

1.3

1.2

1.1

0.9
28 32 36 40 44
Number of Mobiles

Figure 4.1 Total Throughput of The System versus Total Number of


Mobiles

In this chapter, we present the simulation results and specify the


efficacy of the proposed cross-layer scheduling and mapping algorithm.
We compare the results of the algorithms illustrated in section 3.2,
3.3 and 3.4 with our proposed algorithm. In the result of figure 4.1, it
shows the total throughput over the wireless network versus different
numbers of mobiles in the system. By observing this figure, our proposed
scheduling with the proposed allocation algorithm has the maximum
throughput compared with the other algorithms. As the users increasing,
the total throughput of our proposed algorithm is getting larger. The
throughput of the BCF scheduling algorithm, which choose users with
54
best channel conditions is a little lower than the ours. Regarding to the
other mechanisms such as PRF and PF algorithms, the performance is not
better than the above two algorithms.
Although the PRF algorithm considers with channel quality and QoS
requirement for each user, its performance is still not better than ours.
That is because when users figure out the same priority function, the PRF
scheduler would randomly choose one user to transmit without
considering their channel conditions at that situation. When the user with
bad channel condition is selected prior to the one with good channel
condition, the transmission rate would be getting decreasing.
Furthermore, the PF algorithm cares about the fair transmission
toward the users, but it disregard of users’ different channel quality. That
leads the poorest transmission rate among those algorithms.
Thus, if we map resource for the users who have better average SNR,
and then the performance of the system will apparently improve as shown
in figure 4.1.
In the figures 4.2 to 4.4, the average data rate of video streaming and
empty ratio of video buffers versus the different users connected on the
system is shown. In this thesis, we set all video connections having equal
minimum transmission rates of 300 kbps for convenience. The average
transmission rate of each algorithm is shown in figure 4.2. the BCF
algorithm has the best data rate with users than the others. In Figure 4.3,
we show how much average percentage of video service to achieve its
QoS requirement varied by different number of mobiles. We can find that
the percentage of all video service under diverse algorithms are equal or
larger than 100% so that all of them satisfy the QoS requirements in our

55
simulation with the number of mobiles varied from 28 to 44. However,
we can observe the empty ratio under all algorithms in this paper in figure
4.4. Our proposed algorithm provide lowest empty ratio compared with
the other algorithms.
Let us examine the results between figures 4.2 to 4.4. We can see
that the BCF scheduling algorithm has the best throughput performance
of video service in figure 4.2 but does not have the lowest empty ratio of
video buffers in figure 4.4. Therefore, the system may provide longer
transmission delay for video mobiles by using the BCF scheduler and
send enough transmission data at the same time. One the other hand, the
transmission for video buffers is good enough under our proposed
algorithm. By using our proposed schedulers, the system can provide
shortest transmission delay. As the number of mobiles in the system is
getting lower, the performance of transmission delay is relatively
decreased.

56
5
x 10
3.35 Proposed scheduling with BCF allocation
Proposed scheduling with proposed allocation
3.3 BCF scheduling with BCF allocation
PF scheduling with BCF allocation
Average Data Rate of Video Service (bps)

3.25 PRF scheduling with BCF allocation

3.2

3.15

3.1

3.05

2.95
28 32 36 40 44
Number of Mobiles

Figure 4.2 Average Data Rate of Video Streaming Service versus


Total Number of Mobiles
Average Data Rate to Minimum Data Rate Ratio for Video Service (%)

112
Proposed scheduling with BCF allocation
Proposed scheduling with proposed allocation
110 BCF scheduling with BCF allocation
PF scheduling with BCF allocation
PRF scheduling with BCF allocation
108

106

104

102

100

28 32 36 40 44
Number of Mobiles

Figure 4.3 Average Transmission Rate to Minimum Data Rate Ratio for
Video Streaming versus Total Number of Mobiles.
57
-1
10
Empty Radio of Video Buffer

-2
10

-3
10
Proposed scheduling with BCF allocation
Proposed scheduling with proposed allocation
BCF scheduling with BCF allocation
PF scheduling with BCF allocation
PRF scheduling with BCF allocation

-4
10
28 32 36 40 44
Number of Mobiles

Figure 4.4 Empty Ratio for Video Streaming versus Total Number of
Mobiles.

Referring to FTP service, we display the results in Figures 4.5 to 4.6.


The average data rate of FTP service is also presented in Figure 4.5
versus diverse mobiles. We can find that the average data rates are in
order of [Ours>PRF>BCF>PF] from high to low. The first two
algorithms consider with the QoS requirement of FTP service, and
therefore they have higher data rate than those without thinking of QoS
requirement. Our algorithm scheme not only tests if one connection
satisfy the request or not, but further estimates how much data does each
connection needing with introducing channel conditions as the PRF
algorithm. Thus, our scheme has more opportunity to map FTP-service
data more efficiently and fairly.

58
Thus, the average throughput of FTP service for all algorithms is
[ours>PRF>BCF>PF] in sequence. We can summary that the FTP
performance would improved when the scheduler think of the QoS
requirement as designing.

6
x 10
16
Proposed scheduling with BCF allocation
Proposed scheduling with proposed allocation
14 BCF scheduling with BCF allocation
Average Data Rate of FTP Service per Sec (bps)

PF scheduling with BCF allocation


PRF scheduling with BCF allocation
12

10

0
28 32 36 40 44
Number of Mobiles

Figure 4.5 Average Data Rate of FTP Service versus Total Number of
Mobiles.

59
5
x 10 Proposed scheduling with BCF allocation
7
Average Data Rate of FTP Service per Mobile per Sec (bps) Proposed scheduling with proposed allocation
BCF scheduling with BCF allocation
6 PF scheduling with BCF allocation
PRF scheduling with BCF allocation

0
28 32 36 40 44
Number of Mobiles

Figure 4.6 Average Transmission Rate per Mobile per Sec of FTP Service
versus Total Number of Mobiles.

Figure 4.7 depicts the HTTP throughput versus diverse users. As


shown in Figure 4.6, the BCF scheme has best capacity of HTTP service
comparing with the others. When there are too many users connecting to
the base, our scheme reserves bandwidth for video and FTP connections
so that the traffic throughput of HTTP connections would get decreasing.
On the other hand, due to disregard of all traffic classes’ requirements
under the PR and BCF algorithms, the two algorithms provides
remarkably high HTTP throughput by sacrificing the transmit bandwidth
of FTP service as shown in Figure 4.5.
Finally, our proposed scheme contributes distinctly high HTTP data
rate compared with the PRF scheme in all number of mobiles. Among the

60
algorithms described in this thesis, only the PRF and our schemes have
considered of the HTTP (FTP) mobile requests. Since the proposed
scheduler has given a delay threshold time for the BE-service mobiles,
the performance improvement is much greater than under the PRF
scheme.
5
x 10
Proposed scheduling with BCF allocation
18
Proposed scheduling with proposed allocation
Average Data Rate of HTTP Service per Sec (bps)

16 BCF scheduling with BCF allocation


PF scheduling with BCF allocation
14 PRF scheduling with BCF allocation

12

10

0
28 32 36 40 44
Number of Mobiles

Figure 4.7 Average Data Rate of HTTP Service versus Total Number of
Mobiles.

61
Chapter 5 Conclusions
The OFDMA systems with the cross-layer resource allocation
technique, which support multiple service-class mobiles, have been
considered. For the scheduling algorithm at the MAC, we have proposed
a new scheduler. The main contribution of the proposed scheduling
algorithm is a new priority assignment, which combines each mobile’s
channel capacity with the amount of data in bits that each mobile needs to
transmit. If some mobiles have the same QoS-satisfied factor, the
scheduler chooses one user with the best channel quality for the PHY
layer.
For resource allocation algorithm at the PHY, we have offered a
utility-based function to compute the optimal allocation for the purpose of
maximum total throughput.
The simulation results show that the proposed scheduler extremely
improves the performance of HTTP and FTP services while holding the
quality of video service at an acceptable level. While the users are too
many to be served at the same time, the system automatically reserves
bandwidth for the service with stricter requests such as video and FTP but
still maintains the performance of BE service as high as possible.
Furthermore, by observing the results of the simulation, the scheduler
algorithm affects the performance of the system more than the allocation
algorithm does at the PHY.
In summary, the proposed cross-layer resource allocation is expected
to improve the transmission rate and the fairness among users with
maintaining all traffic classes at a acceptable level
In this thesis, we do not design a suitable Adaptive Power

62
Assignment (APA) algorithm for the OFDMA wireless systems. Thus,
the performance might be improved by adding the APA scheme with our
proposed algorithm.

63
Reference
[1] IEEE 802.16-2004, IEEE Standard for Local and Metropolitan Area
Networks Part 16: Air Interface for Fixed Broadband Wireless Access
Systems, IEEE 802.16-2004, Oct. ,2004
[2] IEEE Standard for Local and metropolitan area networks Part 16: Air
Interface for Fixed and Mobile Broadband Wireless Access Systems
Amendment 2: Physical and Medium Access Control Layers for
Combined Fixed and Mobile Operation in Licensed Bands and
Corrigendum, IEEE 802.16e-2005, Feb. ,2006.
[3] H. Fattah and C. Leung, “An overview of scheduling algorithms in
wireless multimedia networks,” IEEE Wireless Communications, vol.
9, no.5, pp.76-83 Oct. 2002.
[4] Y. Ben-Shimol, I. Kitroser, and Y. Dinitz, “Two-dimensional mapping
for wireless OFDMA systems,” IEEE Trans. Broadcasting, vol. 52, no.
3, Sept. 2006.
[5] G. Song; Y. (G.) Li, “Cross-layer optimization for OFDM wireless
networks—Part I: Theoretical framework,” IEEE Trans. Wireless
Commun., vol. 4, no. 2, pp. 614–624, Mar. 2005.
[6] G. Song; Y. (G.) Li, “Cross-layer optimization for OFDM wireless
networks—Part II: Theoretical framework,” IEEE Trans. Wireless
Commun., vol. 4, no. 2, pp. 625–634, Mar. 2005
[7] C. Cicconetti, C. Eklund, L. Lenzini, and E. Mingozzi, “Quality of
Service Support in IEEE 802.16 Networks,” IEEE Network Magazine,
vol. 20, no. 2, Mar. 2006.

64
[8] C. Cicconetti, C. Eklund, L. Lenzini, and E. Mingozzi,, “Performance
Evaluation of the IEEE 802.16 MAC for QoS Support,” IEEE Trans.
Mobile Computing,, vol. 6, no. 1, Jan. 2007
[9] L. Xiao, S. Zhou, and Y. Yao, “QoS-oriented scgeduling algorithm for
mobile multimedia in OFDM,” in Proc. IEEE PIMRC, Sept. 2003,
pp.545-549.
[10] W. S. Jeon and D. G. Jeong, “Combined Connection Admission
Control and Packet Transmission Scheduling for Mobile Internet
Services,” IEEE Trans. Veh. Technol., vol. 55, no. 5, Sept. 2006
[11] Q. Liu, X. Wang, and G. B. Giannakis, “A Cross-Layer Scheduling
Algorithm with QOS Support in Wireless Network,” IEEE Trans. Veh.
Technol, vol. 55, no. 3, MAY. 2006
[12] IEEE C802.20-04/22, Traffic Models for IEEE 802.20 MBWA System
Simulation (Ver. 02), Jan. 2004
[13] 3GPP2, C50-20010507-004R2, 1xEV-DV Evaluation Methodology
(Rev. 26), May 2001
[14] 3GPP, Physical Layer Aspects of UTRA High Speed Downlink Packet
Access (Release 4), Mar. 2001. 3G TR25.848 V4.0.0
[15] W.C. Jakes, Microwave Mobile Communication. Piscataway, NJ:
IEEE Press, 1994
[16] V. Erceg, K.V.S. Hari, M.S. Smith, D.S. Baum et al, “Channel
Models for Fixed Wireless Applications”, IEEE 802.16.3c-01/29r1, 23
Feb. 2001
[17] S. H. Ali, K.D. Lee and V. C. M. Leung, “Dynamic Resource
Allocation in OFDMA wireless metropolitan area networks,” IEEE
Wireless Communications, vol.14, no.5, pp.6-13 Feb. 2007.

65

S-ar putea să vă placă și