You are on page 1of 96

Chair for Communi

ation Networks
Aa hen University of Te hnology
Prof. Dr.-Ing. B. Walke

Diploma Thesis

Performan e Evaluation of the Wireless


MultiHop Adho ATMLAN BAHAMA
of

Fotios Fitsilis
Matri ulation Number: 195123

Aa hen, 17th De . 1998

Supervised by:
o. Prof. Dr.-Ing. B. Walke
Dipl.-Ing. A. Hetti h, Dipl.-Ing. M. Lott

This publi ation is meant for internal use only.

All rights reserved.

No liabilities with

respe t to its ontent are a epted. No part of it may be reprodu ed, stored in a retrieval
system, or transmitted, in any form or by any means, ele troni , me hani al, photo opying,
re ording, or otherwise, without the prior written permission of the publisher.

I assure, that this work has been done solely by me without any further help from others
ex ept for the o al attendan e by the Chair for Communi ation Networks. The literature
used is listed ompletely in the bibliography.

Aa hen, 17th De . 1998

(Fotios Fitsilis)

ABSTRACT

This diploma thesis presents a performan e evaluation of the DQRUMA MAC proto ol of
the BAHAMA network and its omparison with other existing wireless ATM MAC proto ols.
First, the original proto ol stru ture will be presented and the impa t of important proto ol parameters, su h as
investigated.

Random Ba k O, Slot time

and

Number of Mobiles,

will be

Based on the original DQRUMA proto ol two enhan ements will be pre-

sented.
The rst enhan ement eliminates the ollisions on the

Request A ess (RA) hannel. Sin e


Request Table (RT), whi h is

every terminal owns an entry in a Base Station table alled

updated if a request is transmitted su essfully, this is a omplished by polling the terminals with empty RT elds to send their transmission (Xmt) Requests over the next RA
hannel. The se ond improvement modies the s heduling strategy by taking are of the
sour e hara teristi s. For this, the existing

Nominal Rate method was modied, in order

to be used in a slot-by-slot assignment s heme. In this ontext, a new stru ture was proposed, the so alled

Virtual Frame (VF).

Finally, the performan e of the aforementioned systems was analysed and ompared with
two existing

DSA++

implementations (FDD and TDD) and the

Energy-Burst

proto ol

using dierent simulation s enarios.


Keywords:

W-ATM-LAN, MAC, DQRUMA, polling, BAHAMA, Virtual Frame, DSA,


Energy-Burst, Nominal Rate, S heduling

KURZFASSUNG

Diese Diplomarbeit befat si h mit der Implementierung und Untersu hung des DQRUMA
MAC Protokolls und dem ans hlieeden Verglei h mit existierenden ATM MAC Protokollen.
Zuerst wird die Grundstruktur des DQRUMA Protokolls prsentiert und der Einu wi htiger Protokollparameter untersu ht, darunter

Terminals.

Random Ba k O, Slotzeit und Anzahl der

Basierend auf dieser Struktur wurde das Protokoll weiterentwi kelt. Zwei Mo-

dikationen werden prsentiert.


Die erste Modikation eliminiert die Kollisionen auf dem Kanal, indem die Basis Station
(BS) die Terminals pollt. Dies ges hieht mit Hilfe der

Anforderungstabelle,

die in der BS

verwaltet wird und einen Eintrag fr jedes Terminal im System enthlt, das eine Anforderung erfolgrei h versendet hat. Die zweite Modikation nimmt die Quellen harakteristik
zur Kenntnis.

Dafr wurde die

Nominale Rate

benutzt, die fr das DQRUMA System

modiziert werden mute, um auf ein System mit Slot-fr-Slot Kanalvergabe anwendbar
zu sein. Hier wird eine neue Struktur vorgestellt, der sogenannte

Virtuelle Rahmen.

S hlieli h wurde das originale DQRUMA Protokoll mit dem DSA++ (FDD und TDD
Implementierung) gegenbergestellt. Auerdem wurde es mit dem Energy-Burst Protokoll
vergli hen.
Sti hworte: W-ATM-LAN, MAC, DQRUMA, polling Verfahren, BAHAMA, Virtuelles
Frame, DSA, Energy-Burst, Nominale Rate, S heduling

CONTENTS

Abstra t

Kurzfassung

1 Introdu tion

2 Asyn hronous Transfer Mode (ATM)

2.1

ATM Cell Stru ture

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

2.2

ATM-Swit hing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11

2.3

The ATM referen e model . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11

2.4

Quality of Servi e in an ATM network

. . . . . . . . . . . . . . . . . . . . .

13

2.5

Wireless ATM Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

2.6

The Radio Cell as Virtual ATM Multiplexer . . . . . . . . . . . . . . . . . .

15

2.7

The proto ol sta k of the ATM air interfa e . . . . . . . . . . . . . . . . . .

17

2.8

2.9

. . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.8.1

Wireless LAN ar hite tures

Centralized Ar hite ture . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.8.2

Distributed (Ad-ho ) Ar hite ture

19

. . . . . . . . . . . . . . . . . . .

Wireless LAN standardization . . . . . . . . . . . . . . . . . . . . . . . . . .

20

2.9.1

20

IEEE Standards

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.9.2

ETSI Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

2.9.3

ATM Forum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

2.10 Wireless ATM resear h . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

2.10.1 European WATM proje ts . . . . . . . . . . . . . . . . . . . . . . . .

20

2.10.1.1

MEDIAN . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.10.1.2

Magi WAND (Wireless ATM Network Demonstrator) . . .

21

2.10.1.3

SAMBA (System for Advan ed Mobile Communi ations)

2.10.1.4

ATMmobil

21

. . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.10.2 WATM resear h in the United States . . . . . . . . . . . . . . . . . .

22

2.10.2.1

SWAN (Seamless Wireless ATM Network) . . . . . . . . . .

22

2.10.2.2

BAHAMA

. . . . . . . . . . . . . . . . . . . . . . . . . . .

22

2.10.2.3

MII (Mobile Information Infrastru ture) . . . . . . . . . . .

22

3 The BAHAMA network

23

3.1

System Ar hite ture

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

3.2

Proto ol Sta ks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

3.3

The BAHAMA VP/VC on ept . . . . . . . . . . . . . . . . . . . . . . . . .

25

3.4

The Homing Algorithm

26

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4 The Medium A ess Control Proto ol DQRUMA


4.1

DQRUMA des ription

29

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

4.2

Proto ol states

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

32

4.3

Slot and Burst Stru ture . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

4.4

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

35

4.4.1

DQRUMA Enhan ements


Collision Resolution

. . . . . . . . . . . . . . . . . . . . . . . . . . .

35

4.4.2

RA Channel Polling

. . . . . . . . . . . . . . . . . . . . . . . . . . .

36

Contents

4.4.3

Ideal DQRUMA

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

4.4.4

Nominal Rate on DQRUMA . . . . . . . . . . . . . . . . . . . . . . .

37

4.4.4.1

Introdu tion of the Nominal Rate (NR)

. . . . . . . . . . .

37

4.4.4.2

The Virtual Frame . . . . . . . . . . . . . . . . . . . . . . .

39

TDD implementation . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

4.4.5
4.5

Other proto ols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

4.5.1

Slotted ALOHA

41

4.5.2

The Dynami Slot Assignment Proto ol DSA++

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . .

4.5.3

FDD DSA++ MAC-Proto ol

. . . . . . . . . . . . . . . . . . . . . .

43

4.5.4

Energy-Burst MAC-Proto ol

. . . . . . . . . . . . . . . . . . . . . .

43

5 BONeS Simulator
5.1

42

45

Building Simulation Models . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

5.1.1

Constru ting Data Stru tures . . . . . . . . . . . . . . . . . . . . . .

45

5.1.2

Building System Models . . . . . . . . . . . . . . . . . . . . . . . . .

45

5.1.3

Running Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

5.1.4

Evaluating Simulation Results . . . . . . . . . . . . . . . . . . . . . .

45

5.1.5

Intera tive Simulation Manager . . . . . . . . . . . . . . . . . . . . .

46

5.1.6

Proje t Editor

46

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.2

Simulation ar hite ture with BONeS

. . . . . . . . . . . . . . . . . . . . . .

46

5.3

Interfa e primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47

5.4

Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

48

5.4.1

PED and System ar hite ture . . . . . . . . . . . . . . . . . . . . . .

48

5.4.2

Terminal stru ture

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

5.4.3

The MAC Layer

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

50

6 Simulation s enarios

55

6.1

DQRUMA slot length

6.2

DQRUMA Simulations at low data rates (2 Mb/s)

6.3

. . . . . . . . . . . . . . .

6.2.1

Simulations with a variable number of terminals

6.2.2

System with dierent load per terminal

6.3.2

56

. . . . . . . . . . .

62
63

. . . . . . . . . . . . . . . . . . . .

65

Original DQRUMA simulations . . . . . . . . . . . . . . . . . . . . .

65

6.3.1.1

. . . . . . . . . . . . . .

65

. . . . . . . . . . . . . . . . . . . . .

66

Impa t of the Ba k O parameter

DQRUMA Polling simulations


6.3.2.1

6.3.3

55

. . . . . . . . . . . . . . . .

Simulations at high data rates (10 Mb/s)


6.3.1

6.4

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Polling simulation with weighted load

. . . . . . . . . . . .

67

. . . . . . . . . . . . . . . . .

69

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

Simulations with Collision Resolution

Proto ol omparison
6.4.1

DQRUMA vs. TDD DSA++

. . . . . . . . . . . . . . . . . . . . . .

75

6.4.2

DQRUMA vs. FDD DSA++

. . . . . . . . . . . . . . . . . . . . . .

78

6.4.3

Nominal Rate (NR) s heduling

. . . . . . . . . . . . . . . . . . . . .

80

6.4.3.1

Impa t on CBR and rt VBR onne tions

. . . . . . . . . .

82

6.4.3.2

Impa t on ABR onne tions

. . . . . . . . . . . . . . . . .

83

7 Con lusion
7.1

Summary

7.2

Outlook

List of Figures

85
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

87

Contents

List of Tables

89

List of Abbreviations

91

Bibliography

93

Contents

CHAPTER 1

Introdu tion

he rapid growth of the tele ommuni ation industry in the past few de ades has reated a new market with an enormous aspe t in the global e onomy.

A whole new

palette of produ ts and servi es has been developed, in order to satisfy our native need

for unlimited worldwide ommuni ation. Mobility is an important keyword in the resear h
and development of tele ommuni ation produ ts. In this ontext, expanding digitalization
has played a signi ant role and led to the expandation of new networks as the ISDN

Integrated Servi es Digital Network), whi h an transmit voi e, text and data over an universal interfa e. Besides ISDN, digital ellular Networks su h as the GSM (Global System
for Mobile Communi ation) and the DCS 1800 (Digital Cellular System) have gained the
(

publi a eptan e servi ing millions of users worldwide.


For domesti areas, the DECT-standard (

Digital European Cordless Tele ommuni ations)

has been developed, whi h will support appli ations within a small range in areas of high

Su h networks have led to the wireless LANs (Lo al Area


Networks) whi h oer a heap, simple and user-friendly alternative to wired networks.
tra density (e.g. O es).

In future developments, ellular networks are expe ted to be upgraded to support Personal
Communi ation Servi es (PCS). This is going to set new requirements to the hardware
and the transmission te hnology, sin e mu h more data, su h as multimedia appli ations,

Asyn hronous Transfer Mode (ATM) has


been designed, whi h an handle many onne tions with dierent Quality of Servi e (QoS)

will have to be transmitted.

This is why the

simultaneously. ATM is not only an alternative for the wired networks. In the past years
it has also gained an important role in wireless networks.
In that ase new requirements have to be met, be ause transmission over the air-interfa e
with a Bit Error Rate (BER) of app.
with a BER of app.

10

13 .

10

is mu h more di ult as in wired networks

Not only the BER, also physi al ee ts in the air, su h as

attenuation, multi-path propagation, di- and refra tion and the Doppler-ee t, are ausing
many problems. For this, new algorithms and te hniques have been invented, whi h an
drop the BER to the a eptable level for data transmission.
At the moment there are two standards, whi h ompete for deployment in wireless LANs,
HIPERLAN and IEEE 802.11. Both are optimized for data transmission. For the wireless
ATM LANs (WATMLANs) many on epts from dierent ompanies are under development.
In this master thesis a ompetitive on ept, the

Broadband Ad Ho ATM Anywhere (BA-

HAMA), has been investigated based upon numerous simulations reated with the simulation tool
network.

BONeS Designer.

BAHAMA is an AT&T proje t for a

entralized Ad-ho

Cellular ar hite ture is applied in this network with one PBS (Portable Base

Station) per radio ell. PBSs an be deployed in an arbitrary topology inside the network
using high-speed onne tions (GB/s) to ea h other. Mobile users with multimedia laptop
omputers will be able to ommuni ate with ea h other at 2-20 Mbps using the PBS ba k-

1. Introdu tion

bone LAN or dire tly (peer-to-peer), if they are lose by.


In a wireless network that supports multimedia tra , the Medium A ess Control (MAC)
proto ol is responsible for sharing the limited bandwidth in an e ient and fair manner.
In the BAHAMA network a new MAC proto ol has been used, the DQRUMA-proto ol

Distributed-Queueing Request update Multiple A ess). There are two possible proto ol
implementation forms. In this thesis the FDD (Frequen y-Division Duplex) implementa-

tion is going to be dis ussed. The FDD mode uses two dierent frequen ies, one for the
uplink (mobile-to-base) and one for the downlink (base-to-mobile) transmission.

In an-

other implementation, in order to a hieve dynami sharing of the uplink and downlink

Time-Division Duplex)

bandwidth, a TDD (

slot allo ation s heme is used. Here are the

uplink and downlink hannels time multiplexed on a single frequen y hannel.

This is

a hieved by treating the PBS like a super mobile.


After a des ription of the ATM basi features, a short introdu tion in the wireless networks
ar hite tures is going to be presented. Se tion 2.9 oers an overview of the ongoing standardization eorts on WATMLANS. The BAHAMA network Ar hite ture will be des ribed
next, followed by a detailed presentation of the DQRUMA proto ol and its modi ations.
Chapter 5 des ribes the fun tionality of the Simulation tool

BONeS Designer.

Then the

simulation results from various tra s enarios will be dis ussed in order to estimate the
proto ols performan e. The results will be ompared with those from the

ATMLAN

on ept, so that both systems an dire tly be onfronted.

ends with a synopsis of the most important issues presented.

Philips Ad-Ho

The master thesis

CHAPTER 2

Asyn hronous Transfer Mode (ATM)

TM is the abbreviation for

Asyn hronous Transfer Mode and denominates a new asyn-

hronous transmission and swit hing pro edure with spe i multiplex operations. It

was developed a ording to the I-300 series of the ITU re ommendations

Tele ommuni ation Union

International

in order to mat h the in reasing requirements aused by new

servi es and more powerful omputers.

In ATM networks, data streams are split into pa kets of onstant length, alled ATM ells.
These ells onsist of a 5 byte header and 48 bytes of payload. Sin e the ells are transmitted asyn hronously, they do not o upy dedi ated time slots. On a physi al hannel,
ells belonging to dierent onne tions are transmitted nested in time a ording to their
order of arrival.

The onne tions are virtual, and ell sequen e is preserved.

When no

onne tion needs apa ity, the ATM multiplexer stus empty ells into the data stream
(see Figure 2.1). In this way, only the a tually required apa ity of the physi al hannel
is assigned to ea h onne tion, and it is possible to rea t on the dynami ommuni ation
properties of onne tions with time-variant request for bandwidth (see Table 2.1).
method alled

statisti al multiplexing

This

allows to transmit more data ompared to a syn-

hronous multiplexing method for a physi al hannel with the same apa ity. This gain is
a hieved by exploiting the statisti al properties of the tra .

Servi e

Bit rate

Burstiness

Spee h, Telephony

64 kbit/s



25 ms

Video onferen e (low quality)

128 kbit/s

50-250 ms

Video onferen e (high quality)

1-10 Mbit/s

50-250 ms

Data

0.1-30 Mbit/s

2 - 200

Delay

 Ratio of maximum ompared to mean bit rate


 maximum sustainable end-to-end delay
 without e ho ompensation

Table 2.1:

Chara teristi s of typi al ATM servi es [26

ATM networks use virtual onne tions (Virtual Channel Conne tion, VCC) for ell transmission. When establishing a onne tion between terminals, the ATM network determines
a route omprising the used swit hes.
of ea h swit h.

This route is stored in the internal routing table

In this manner, a virtual hannel through the swit hes is dened, and

the ATM ells of the orresponding virtual onne tion an be transmitted. Swit hing of
ells then is performed by using the routing parameters in luded in the ATM ell header.
The routing parameters are interpreted as pointers to orresponding entries in the network
nodes routing tables.

2. Asyn hronous Transfer Mode (ATM)

10

Connection A
A

Connection B
B

Connection C
C

A
B
C

2.1

Cell stream after ATM Multiplexer

Figure 2.1:

Connection A
Connection B
Connection C
empty cell

ATM Time Division Multiplex

ATM Cell Stru ture

The ell header stru ture of an ATM ell depends on whether the ell is transmitted within
the network (between ATM swit hes only) or between network and user (see Figure 2.2).
Two interfa es are dened, respe tively [25:

 Inter-Network Interfa e (INI)


 User-Network Interfa e (UNI)
BISDN NNI

BISDN UNI
GFC

VPI

VPI

VCI

VPI

VCI
VCI

VPI

VCI

VCI
PT

CLP

VCI

HEC

PT
HEC

CLP

Cell loss priority

PT

Payload type

GFC

Generic flow control

VCI

Virtual channel identifier

HEC

Header error control

VPI

Virtual path identifier

UNI

Usernetwork interface

NNI

Networknode interface

Figure 2.2:

CLP

Byte

4
5

ATM Cell Stru ture at UNI and INI [25

GFC Generi Flow Control, 4 bit (UNI only)


This eld is used for a ess ontrol of terminals at the UNI.

VCI Virtual Channel Identier, 2 byte


The referen e on the virtual hannel serves to dierentiate between simultaneous
onne tions . VCIs are only valid between two network nodes.

VPI Virtual Path Identier, 8 or 12 bit


VPIs denote a bun h of hannels. Several routes of one dire tion, ea h onsisting of
several virtual hannels, an be separated.

In this way, hannels belonging to the

same bun h an be identied and routed together qui kly by an ATM swit h.

PT Payload Type, 3 bit


This eld des ribes the type of information ontained in the payload eld as user

2.2. ATM-Swit hing

11

or signalling information.
(OAM) or

The latter is employed for

Operations and Maintenan e

Resour e Management (RM). In this ase, an ATM swit h has to evaluate

both the ell header and the signalling information ontained in the payload eld.

CLP Cell Loss Priority, 1 bit


This bit serves to mark ells with low priorities, whi h are dis arded rst depending
on the network onditions.

HEC Header Error Control, 1 byte


In ontrast to the payload, the ATM ell header is prote ted by an error ontrol
me hanism des ribed in ITU-T Re ommendation I.432. Dete tion of two-bit-errors
and orre tion of one-bit-errors is possible. The HEC eld is he ked in ea h ATM
swit h. In addition, the HEC is used to dete t the beginning of an ATM ell.

2.2

ATM-Swit hing

Swit hing is performed for virtual onne tions. The ATM ell header ontains a VCI and
a VPI, whi h are evaluated by the a tual ATM swit h to determine the subsequent swit h.
Ea h ATM swit h only knows its next.

In order to set up routing tables and establish

virtual onne tions, the omplete sour e and destination address of the virtual onne tion
is sent on e. Arriving at a network node, VCI and VPI are extra ted and a ording to the
routing table, a new VPI and, depending on the type of swit h, a new VCI are determined.
The new identiers are entered in the ell header, then the ell is moved to the exit of the
swit h.

VC switch
3
1

VCI 1
VCI 2

VPI 1

VPI 4

VCI 3
VCI 4

VCI 3
VCI 4

VPI 2

VPI 5

VCI 5
VCI 6

VCI 1
VCI 2

VCI 5
VCI 6

VPI 3

VPI 6

VCI 1
VCI 2

VCI 1
VCI 2

VP switch

Figure 2.3:

VPI 2

VCI 4

VPI 1

VPI 3

VCI 3

VPI 4

VPI 5

VCI 1
VCI 2

VP switch

Swit hing of Virtual Channels/Virtual Paths [25

Denoting a virtual onne tion with a VPI and a VCI mirrors the two types of network
nodes (see Figure 2.3):

 ATM Swit h

ATM ross- onne ts swit h the hannel bun hes in the orresponding dire tions depending on the VPI. The VCI remains un hanged.

 ATM Cross-Conne ts

In ATM swit hes, both VPI and VCI are altered.

2.3

The ATM referen e model

Based on the ISO-OSI ommuni ations referen e model, a spe i referen e model onsisting of four

layers

was dened for ATM (see Figure 2.4): the physi al layer, the ATM

2. Asyn hronous Transfer Mode (ATM)

12

ATM Adaption Layer (AAL) and a layer representing the fun tions of the higher
Three dierent proto ol planes were introdu ed: user plane, ontrol plane and man-

layer, the
layers.

agement plane.

The management plane fullls two fun tions: plane management of the

whole system as well as administration of the single layers (layer management).

Management plane
User plane
Higher layers

Control plane
Higher layers

ATM adaptation layer (AAL)


Plane management

ATM layer

Layer management
Physical layer

Figure 2.4:
Physi al layer

ATM Referen e Model

The physi al layer on erns the ommon ISO-OSI layer 1 fun tions as bit

syn hronization, hannel oding and the supervising fun tions.

Its fun tionality is

determined by the underlaying transmission medium.

ATM layer

The ATM layer orresponds to ISO-OSI layer 3. It ontains the ATM spe i

fun tions for transport of ATM ells:

 Multiplexing and demultiplexing of ATM ells belonging to dierent onne tions


 Generation and Extra tion of ATM ell header information.
 Priority handling to avoid ATM ell loss and to minimize the waiting time
 Fun tions to avoid buer overows (Congestion Control)
 Conne tion-spe i supervision of data rates a ording to the agreed Quality of
Servi e (QoS), (Usage Parameter Control, UPC)

 Generi ow ontrol at the UNI


ATM adaption layer

The ATM adaption layer (AAL) is on erned with servi e-spe i

requirements and therefore orresponds to ISO-OSI layer 4. It provides servi es whi h


are realized using the servi es of the underlaying ATM layer. The AAL layer onsists
of two

sublayers:

 The Segmentation And Reassembly

(SAR) sublayer performs the segmentation

of data streams and

 the Convergen e Sublayer

(CS) ompensates for delay variations o urring for

syn hronous servi es in the ATM network and allows dete tion of lost ells for
data servi es.

2.4. Quality of Servi e in an ATM network

13

The following AAL types have been dened [25:

Type

Des ription

AAL 1

Allows syn hronous data transfer with

onstant bit rates by

ompensating delay variations at the re eiving terminal


AAL 2

Allows syn hronous data transfer with

variable bit rates by

ompensating delay variations at the re eiving terminal


AAL 3/4

For data servi es and non real-time-oriented servi es;


allows dete tion of lost pa kets and ells ; Type 3 is
onne tion-oriented, whereas Type 4 is onne tionless

AAL 5

Dire t a ess to the ATM layer

S-AAL

Redu ed Type 3 for signalling tra

Table 2.2:
2.4
The

ATM Adaption Layer Types [25

Quality of Servi e in an ATM network

Quality of Servi e

(QoS) in ATM networks des ribes the quality of a onne tion

the network should be able to guarantee.

During onne tion establishment, the QoS is

negotiated with a tra ontra t between network and user. The ATM network must be
able to meet the requirements of dierent servi es. Quality of Servi e is hara terized using

Network Performan e Parameters,

whi h are measurable at the UNI. QoS requirements

are des ribed by the network parameters listed below [21:

CER

Cell Error Ratio:


Ratio of in orre t transmitted ells to su essfully transferred ells in a population of
interest

CLR

Cell Loss Ratio:


Ratio of lost ells to transmitted ells in a population of interest

CMR

Cell Misinsertion Rate:

Total number of misinserted ells observed during a spe ied time divided by the
time interval duration

CTD

Mean Cell Transfer Delay:


Arithmeti al average of end to end ell transmission delays

maxCTD

Maximum Cell Transfer Delay:

ATM ells whi h have ex eeded their maxCTD (also alled due date) are onsidered
lost

CDV

Cell Delay Variation:


Variation of end to end ell transmission delay

The following parameters spe ify the tra hara teristi s:

2. Asyn hronous Transfer Mode (ATM)

14

PCR

Peak Cell Rate:


The maximum number of transitted ells per se ond

SCR

Sustainable Cell Rate:


Upper boundary of the mean data rate

MCR

Minimum Cell Rate:

The minimum number of transmitted ells per se ond

MBS

Maximum Burst Size:

The maximum number of ells transmitted in a burst


To meet the properties of dierent sour e tra hara teristi s, four

servi e ategories were

dened. They all have dierent requirements on erning the QoS (see Table 2.3) [26:

CBR Constant Bit Rate

onstant bit rate, equal to the

The tra sour es of this servi e ategory send with a

maximum bit rate negotiated during onne tion establishment. Realtime transmission and omplian e with the maxCTD are demanded. (example: spee h transmission)

VBR Variable Bit Rate


VBR tra sour es send with

variable bit rates.

A dierentiation between

(RT) oriented servi es with a dened maxCTD and

non Real-Time

Real-Time

(nRT) servi es

for less time riti al appli ations is made. (example: video transmission)

ABR Available Bit Rate


Within this servi e ategory, tra sour es adjust their bit rate dynami ally to the
resour es being available for transmission.
real-time behaviour.

There are

no

requirements to delay or

Only the CLR is dened as network parameter.

(example:

le-transmission)

UBR Unspe ied Bit Rate


This servi e ategory oers no servi e guarantee at all. Neither CLR nor maxCTD
are dened (

Best Eort ).

parameter

ATM servi e ategory


CBR

VBR(RT)

CLR

VBR(NRT)

ABR

UBR

dened

CTD

n/a

dened

maxCTD
CDV

n/a

dened

n/a

dened

Table 2.3:

n/a

ATM servi e ategories and their QoS parameters

Supervision and ontrolling of the QoS

tra management

requires spe ial fun tions in

the terminal equipment and in the a ess and swit hing nodes. Before a eption of a new
onne tion-request, the

Conne tion Admission Control

(CAC) he ks whether a new on-

ne tion an be admitted. The CAC only works orre tly if a onne tion's tra does not
ex eed the parameters of the tra ontra t.
In ATM multiplexers and ATM swit hing nodes ells from dierent onne tions ompete
for the transmission lines behind the multiplexer.

The sequen e of the waiting ells is

2.5. Wireless ATM Networks

15

determined by a ontrol unit, alled s heduler, obeying a spe i strategy. The s heduler's
task is to distribute the apa ity for transmission onsidering the QoS requirements of ea h
onne tion.

2.5

Wireless ATM Networks

This hapter presents the ATM air interfa e between

T erminals (WT) and the ATM ore network.

W ireless (respe tively mobile) ATM

Subsequently, the stru ture of the proto ol

sta k is presented. Finally, some wireless LAN ar hite tures are des ribed.

Figure 2.5 illustrates an example of a stru ture for a mobile ommuni ation network. Base
stations form a ess points to the ATM ore network, onsisting of one or more transmitand-re eive units (

Base Station Trans eiver, BST) and one Base Station Controller (BSC).

The BSC onne ts the base station to the ATM network and exe utes the proto ols.

ATM network
WT1

Base Station BS1


TRX
ATM switch

BSC

WT2
TRX

Base Station BS2


WT3

TRX

BSC
ATM mobility switch

ATM air interface

Figure 2.5:

Ar hite ture of a ellular mobile ATM network [26

The mobile network shown above allows wireless a ess to the ATM ore network in spe i lo ations like in buildings or even outdoors. Handover fun tions enable the terminals
to move freely within the area overed by the base stations.

The te hni al data of the air interfa e is based on re ommendations of the wireless ATM
group (RES10 WAG) of the

European Tele ommuni ations Standards Institute (ETSI) and

are summarized in Table 2.4.

2.6

The Radio Cell as Virtual ATM Multiplexer

Typi al operational areas of ellular mobile ATM networks are

W ireless Lo al Area N etworks

(W-LANs). Provision of the same servi es oered by terminals with xed ATM a ess is
desired for WT's.

This means that WTs and terminals with xed ATM a ess should

employ the same AAL layer servi es.

2. Asyn hronous Transfer Mode (ATM)

16

frequen y band

5.2 GHz

multiplex method

FDM

a ess methods

TDMA

duplex method

TDD

bandwidth on a frequen y hannel

23.5 MHz

gross data rate of one hannel

50000 ATM ells/s

Table 2.4:

Te hni al data of the ATM air interfa e [5

virtual ATM
Multiplexer

ATM Core Network

ATM

Application

AAL

Mobile ATM Terminal

Base
station

fixed ATM Terminal

Application

AAL

ATM

AAL

Mobile ATM Terminal


ATM

Application

virtual connections

Figure 2.6:

Merging a ellular ATM radio network into an ATM ore network [26

Figure 2.6 illustrates the end-to-end hara teristi of AAL transport proto ols. Data transmission on the ATM air interfa e is based on ATM ells within the ATM layer, therefore
ee ts of the air interfa e are not visible for servi e users of the ATM layer (the AAL instan es). Regarded from a servi e user's point of view, all WTs in a radio ell maintaining
virtual onne tions via the base station seem to be onne ted to an ATM multiplexer by
wire.

This approa h leads to a model of the ATM radio ell with its entral base station and
its WTs as a distributed, virtual ATM multiplexer with the radio interfa e being lo ated
inside the multiplexer.

The ATM multiplexer oers the WTs a modied UNI, denoted W-UNI (see Figure 2.7).
Modi ations of the UNI on ern the

ontrol and management plane, in order to be able to

manage the ee ts of mobility, announ e hanges to the user and enable him to interfere.
Data transparen e demands the

user plane to remain un hanged.

Between the virtual ATM multiplexer and the ATM ore network, a modied NNI (MNNI) is dened (see Figures 2.9 and 2.10). In addition to the NNI properties, the M-NNI
allows handover and management of handover network resour es.

The virtual ATM multiplexer forms a distributed queueing network with queues within
the WTs (for the uplink ATM ells) and the base station (for the downlink ATM ells) and
a ontrol unit, alled

s heduler.

the sequen e of ATM ells.[26

The s heduler is part of the base station and determines

2.7. The proto ol sta k of the ATM air interfa e

2.7

17

The proto ol sta k of the ATM air interfa e

In ontrast to the ore network, the usage of the radio hannel as transmission medium
of the virtual ATM multiplexer needs the onsideration of radio link spe i aspe ts in
proto ol design:

Radio Propagation:
Channel a ess:

E.g. shadowing, reexions, dira tion and multi-path propagation

Coordination of medium a ess to preserve the ATM ell transmission

sequen e determined by the s heduler

Error prote tion:

Sin e the reliability of the radio hannel is smaller than in wired net-

works, methods for error prote tion are vital to meet the guaranteed QoS

Even the use of opti sbers for transmission of ATM ells an not ompletely suppress
errors in ells. Therefore the AAL layer uses, depending on the servi e type, an end-to-end
error orre tion me hanism. Within the ATM layer, the HEC is employed to prote t the
routing information of the ATM ell against transmission errors.

The real-time oriented CBR and VBR servi es use the AAL proto ols Type 1 and Type

Proto ol Data Unit (PDU) is equipped with a rundete tion of lost or misinserted ATM ells. An optional
te hnique allows orre tion of bit errors. If the bit error

2 respe tively (see Figure 2.2). Ea h


ning number and a parity bit for

Foward Error Corre tion

(FEC)

probability in the ATM layer is higher than the ability to orre t errors in the AAL layer,
whi h is the ase for the radio link, the user's QoS an not be guaranteed any more.

Automati Repeat Request (ARQ) proto ol in the topmost sublayer (Servi e Spe i Convergen e Sublayer (SSCS), whi h is based
on fun tions for dete tion of bit errors and ell delays of the underlaying Common Part
Convergen e Sublayer (CPCS) and Segmentation And Reassembly (SAR).

The AAL proto ols Type 3/4 and Type 5 use an

An e ient exe ution of these ARQ proto ols an be rea hed for a pa ket loss ratio of

10

1.

To mat h this pa ket loss ratio, the bit error ratio for pa kets of 1 kbyte length

has to be smaller than

10

7.

With the bit error ratio of a radio link prote ted by FEC

te hniques usually being mu h higher, an e ient exe ution of ARQ proto ols is no longer
possible. Therefore, an additional ARQ proto ol is deployed in the

Logi al Link Control

(LLC) sublayer straight at the air interfa e, ensuring the ATM layer transparen e for the
AAL layer.
Figure 2.7 illustrates the resulting wireless proto ol sta k onsisting of
and

Data Link Control

(W-PHY)

(W-DLC) layer. The W-PHY layer onsiders the hara teristi s of

the radio link, whereas the W-DLC layer omprises the


sublayer and the

Physi al

Logi al Link Control

Medium A ess Control

(W-MAC)

(W-LLC) sublayer, whi h ontrols the logi al han-

nels and ontains fun tions for error prote tion . The MAC layer proto ol is dis ussed in
detail in Chapter 4.

The onne tion-oriented fun tionality of the error prote tion performed by the ARQ proto ols requires the LLC sublayer to reside above the s heduler's multiplexing fun tion.
The ATM referen e model does not assume a onne tion oriented error prote tion, i.e., no
LLC sublayer is provided. For this reason, the remaining fun tionality of the DLC layer

2. Asyn hronous Transfer Mode (ATM)

18

AAL
ATM

AAL
WDLC

User Process

WDLC

User Process

WLLC
WMAC

WPHY

ATM

WLLC
WMAC

WPHY

PHY

Base Station with


ATM Switch

WUNI

Wireless Terminal

Figure 2.7:

ATM
PHY

ATM Core Network

UNI

Proto ol Sta k of the ATM Air Interfa e

( ontrolling of the QoS, multiplexing) has been a ounted to the bottom of the ATM layer.
In this thesis an ideal hannel is used, that is error rate equal zero. Hen e, ells an only
be lost due to buer overows and are not transmitted again.

2.8

Wireless LAN ar hite tures

Wireless ATMLAN are divided into two ar hite tural models. These are the
and the

distributed (Ad-ho ) model.

entralized

The se ond possibility oers many advan es for using

into LANs, be ause of lower ost and greater system s alability.

2.8.1 Centralized Ar hite ture

WT
Base

WT

Station

WT
Distributed
ATM-Switch

WT

WT

Base

Base

Station

Station

WT

WT
Base
Station

WT

Figure 2.8:
The

Centralized WATMLAN Ar hite ture based on a distributed ATM swit h

Centralized Ar hite ture pla es every Wireless Terminal (WT) in the range of one Base

Station (BS), whi h oordinates the data transfer in its ell. The BS must also have ATM
fun tionality for generating a onne tion with the ATM swit h. The swit hes ar hite ture
may be of both types, namely entralized or distributed (see Figure 2.8 and 2.9) .

2.8. Wireless LAN ar hite tures

W-ATM
Terminal

W-ATM
TA

19

W-ATM

W-ATM

Radio

Radio

Port

W-ATM
RRAL

ATM
WUNI

ATM
TRX

W-ATM
TA

W-ATM
Terminal

Port

Philips Distributed
Mobility Enabled

ATM
TRX

W-ATM
RRAL

ATM
WUNI

ATM Switch
ATM
TRX

ATM
Host

ATM
UNI

Figure 2.9:

ATM
TRX

ATM
TRX

ATM
Network

ATM
MNNI

ATM
NNI

Centralized WATMLAN Ar hite ture

2.8.2 Distributed (Ad-ho ) Ar hite ture

W-ATM TA
with
ATM TRX

W-ATM
Terminal

Adhoc W-ATM
RRAL

W-ATM TA
with
ATM TRX

W-ATM
Terminal

ATM
WUNI

ATM
WUNI
Mobility
ATM
Host

ATM

Enabled

Network

ATM

W-ATM TA
with
ATM TRX

Switch

ATM
UNI

ATM
NNI

Figure 2.10:

ATM
MNNI
Distributed WATMLAN ar hite ture

Using this ar hite tural model an autonomous WLAN an be built. All WTs have then
equal rights for using the shared transport medium. The only dieren e between the two
ar hite tural models onsists on a modi ation of the 'R' RAL Interfa e, whi h is now
named:

Ad-ho 'R' RAL. This is now expanded with all ne essary Ad-ho fun tionalities.

This ar hite ture an be obtained on Figures 2.10 and 2.11. [8

ATM-Tranceiver
Wireless Access

ATM-Tranceiver

Wireless ATMLAN

ATM-Tranceiver

Wireless Access
Wireless Access

Figure 2.11:

ATM-Tranceiver

ATM-Tranceiver

Wireless Access

Wireless Access

Ad-ho WATM LAN a distributed ATM swit h

2. Asyn hronous Transfer Mode (ATM)

20

2.9

Wireless LAN standardization


ommer ially available produ ts and re ent prototyping of wireless LAN systems have
been developed primarily to provide wireless extension to existing wired LANs, espe-

ially

Ethernet.

But the in reasing importan e of personal mobility and high data rates in

onjun tion with standardization of ATM will reate demands for wireless ATM solutions
in the near future, where WATM shall be seen not only as a wireless extension of the wired
network, but as integral part of future mobile systems. In this hapter an overview of the
most important European and non-European a tivities an WLANs fo using on WATM is
given.

2.9.1 IEEE Standards


Already in 1997 the IEEE approved the new standard 802.11. This standard is intended to
play a very important role for the expandation of wireless LANs. 802.11 spe ies a single
MAC sub layer and 3 physi al (PHY) layers.

Two PHYs are for radio operating in the

2400-2483.5 MHz frequen y band at data rates of 1 and 2 Mbps, and one is for infrared. In
future, 802.11 shall be extended with a PHY for High Speed WLANs in the 5 GHz band
at data rates of 20 Mbps.

2.9.2 ETSI Standards


ETSI Te hni al Committee RES 10, whi h has re ently been in luded into
wireless Broadband Radio A ess Networks, is urrently developing a family of stanfor high speed radio te hnology alled HIPERLAN (High Performan e LAN). The

In Europe, the
BRAN
dards

family onsists of four network types. It spe ies a PHY with a data rate of app. 20 Mbps
in the 5 GHz ISM (

Industry-S ien e-Medi ine)

band. The other family members address

multimedia a ess to lo al and remote wired ATM networks as well as very high speed
wireless networks. A Spe i feature of the HIPERLAN standard is the apability of

ho networking without any ba kbone system (see Figure 2.12).

In early 1997, ETSI laun hed a se ond standardization ampaign on

ad-

wireless Broadband

Radio A ess Networks (BRAN). BRAN aims to dene the standards for servi e independent broadband radio a ess networks and systems having a peak data rate of at least
25 Mbit/s at UNI. BRAN networks shall be apable of e iently arrying existing servi es
like voi e and ISDN as well as providing the transport me hanism for future servi es.

2.9.3 ATM Forum


Although the

ATM Forum is not an o ial standardization body, still it plays a signi ant

role in the wireless ATM standardization, be ause it represents the industry's point of
view. The

ATM Forum

is sin e June 1996 engaged with the WLAN standardization and

is mainly o upied with mobility support matters in wired ATM networks.

2.10

Wireless ATM resear h

2.10.1 European WATM proje ts


European WATM resear h is on entrated in the ACTS program, funded by the

Community.

European

ACTS is divided in several subproje ts, ea h for spe i usage s enarios.

2.10. Wireless ATM resear h

21

Mobile Wireless Networks

Statical Wireless Networks

HIPERLAN
Type 1

HILERLAN
Type 2

HIPERLAN
Type 3

HIPERLAN
Type 4

Wireless LAN
MAC

Wireless ATM
MAC

Wireless LL
(Lokal Loop)
MAC

Wireless ICN
(Interconnect)
MAC

Wireless LAN

Wireless ATM

Wireless LL

Wireless ICN

PHY (5 GHz)

PHY (5 GHz)

PHY (5 GHz)

PHY (17 GHz)

20+ Mbit/sec

20+ Mbit/sec

20+ Mbit/sec

150+ Mbit/sec

ATM Oriented HIPERLANs

Figure 2.12:

The four HIPERLAN types [5

2.10.1.1 MEDIAN
The ACTS MEDIAN proje t implements a high speed

wireless Customer Premises LAN

(WCPN/WLAN) pilot system for multimedia appli ations. It is a very high speed, very
high frequen y system, with a data rate of 155 Mbit/s at a frequen y of 60 GHz.

2.10.1.2 Magi WAND (Wireless ATM Network Demonstrator)


The ACTS Magi WAND proje t has a similar s ope, although at a lower frequen y
band(5 GHz) by using lower data rates (20 Mbit/s).

Main obje tive is to spe ify a wire-

less ustomer premises a ess system for ATM networks that extends the servi es and
benets of ATM networks to the mobile user.

2.10.1.3 SAMBA (System for Advan ed Mobile Communi ations)


The ACTS SAMBA Proje t promotes the development of a broadband ellular radio extension of B-ISDN. The proje t fo uses a trial platform providing transparent ATM onne tions up to 34 Mbit/s in a ellular radio environment at 40 GHz. The user appli ations
in mind are a wireless TV amera and a medi al appli ation.

2.10.1.4 ATMmobil
The most ambitious WATM proje t apart from the ACTS a tivities is ATMmobil. It is a
four-year resear h program funded by the German Government. Its main obje tive is to
develop new on epts for broadband mobile multimedia ommuni ation, built on an ATMbased infrastru ture allowing for data rates up to 155 Mbit/s. It onsists of four sub-proje ts
whi h over spe i elds of the entire proje t:

Broadband Cellular ATM A ess:

Aim is to investigate a ellular ATM system enabling

wireless broadband multimedia a ess, supporting ATM transmission of up to 20 Mbit/s.

2. Asyn hronous Transfer Mode (ATM)

22

Wireless ATMLAN:

Obje tive is to develop a unied framework ar hite ture for both

entralized and ad-ho WATMLAN systems (see Chapter 2.8).

Integrated Broadband Mobile System(IBMS):

This subproje t investigates integration

of low and high data rate servi es for indoor and outdoor mobile ommuni ation
systems based an ATM.

ATM Radio-in-the-Lo al-Loop:

This subproje t overs the last mile, providing wireless

broadband a ess to subs ribers of the publi network.

2.10.2 WATM resear h in the United States


Most a tivities on WATM in the United States are on entrated in the resear h laboratories
of

AT&T and Lu ent Te hnologies.

2.10.2.1 SWAN (Seamless Wireless ATM Network)


The SWAN Proje ts from the

Bell Labs investigates the ombination of wireless a ess and

multimedia network omputing in an indoor setting. The system enables a data rate of
625 kbit/s in the ISM band at 2.4 GHz.

2.10.2.2 BAHAMA
A se ond proje t at the

Bell Labs is alled Broadband Adaptive Homing ATM Ar hite ture

(BAHAMA). Main obje tive is to innovate and advan e the system on ept of an ad-ho
wireless ATM LAN (see Chapter 2.8 supporting multimedia servi es for indoor appli ations.
Mobiles are linked to Portable Base Stations (PBS) at 25 Mbit/s with radio or infrared
opti s. S ope of this thesis is to evaluate the performan e of BAHAMAs MAC proto ol.
The on lusions will be onfronted with urrent results on existing

ATMmobil proto ols.

2.10.2.3 MII (Mobile Information Infrastru ture)


Bell Labs, performed together with Sun Mi rosystems, is
alled Mobile Information Infrastru ture(MII). It is partially funded by the United Stated
The Latest WATM resear h at

government and the National Institute of Standards and Te hnology. The early work on
the proje t was based on the network model developed in the BAHAMA proje t.

MII

employs an ar hite ture of wireless ATM LANs inter onne ted through ATM swit hes and
supports both the network model of BAHAMA PBSs, enabling end-to-end ATM links, as
well as IEEE 802.11 ompliant wireless links onne ting with standard (Ethernet) wireless
LANs. More on BAHAMA and MII an be found in the Chapter 3. [8

CHAPTER 3

The BAHAMA network

and

he BAHAMA network is a WATMLAN proje t generated at the AT&T Bell Laboratories. It onsists of two types of network elements:

Mobile Endpoints (MEs).

Portable Base Stations

(PBSs)

Ea h PBS is responsible for managing the tra within its

own (mi ro) ell (see Figure 3.1). The PBS-to-PBS ba kbone links an either be wired or
wireless and oer a high bandwidth, sin e they are of high speed (Gbps).

The PBSs an be pla ed in an arbitrary topology inside the network to support the mobile users oering high exibility due to their

Ad-ho features

(i.e. self-re onguring and

self-organizing apabilities)(see Chapter 2.8). MEs assumed to be multimedia laptop omputers, whi h will be able to ommuni ate with ea h other at 2-20 Mbps using the PBS
ba kbone LAN or dire tly (peer-to-peer), if they are lose by. The later will not be dis ussed any further, sin e it is optional.

WT

WT

PBS

WT

Figure 3.1:

WT

The BAHAMA ell [14

There are many problems to be solved in su h a wireless ATMLAN. In this hapter BAHAMAs ar hite tural aspe ts will be presented rst.
BAHAMA

mobility management

aspe ts.

Mobility management is a keyword, be ause

wireless users are free to roam inside the network.

Homing Algorithm,

The hapter will lose with some

Additionally, a novel algorithm, the

is presented supporting this on ept, while guaranteeing orre t ell

sequen e delivery (see Se tion 3.4).

3.1

System Ar hite ture

A key question regarding wireless ATM is whether to transport ATM ells over the wireless interfa e or to lo ate the ATM Adaption layer (AAL) in the mobile hosts or the Base
Station (see Chapter 2.3).

If AAL is lo ated in the mobile hosts, and ATM ells are transported over the wireless
interfa e, a ustom wireless interfa e will be needed to transport ATM ells over wireless
hannels. Be ause ATM is end to end, mobile hosts an benet from su h properties as
QoS-provisioning and multimedia servi e support. The disadvantage is that most urrent

3. The BAHAMA network

24

Wired Network
Ad Hoc Wireless LAN

Gateway
PBS
Portable Base
Station

Portable Base
Station

Gb/s

Portable Base
Station

Gb/s

2-20 Mb/s

WT

WT

WT

Figure 3.2:

WT

WT

WT

WT

WT

The BAHAMA System ar hite ture [7

hosts (mobile and xed) la k ATM apability and will not readily benet from the end-toend ATM approa h (see Figure 3.3).

Cross-connect
WATM
network

Applications
TCP
IP
AAL
WATM
Custom wireless

ATM
switch

WATM-ATM
conversion
WATM
ATM
Custom
wireless PHY

Figure 3.3:

WATM
network

ATM
switch

WATM-ATM
conversion

ATM ATM
PHY PHY

ATM
PHY

WATM
Custom
wireless

Applications
TCP
IP
AAL
WATM
Custom wireless

The BAHAMA proto ol sta k [6

If AAL is lo ated at the Base Stations, the requirement that end hosts must have ATM
apability an be relaxed and a standard wireless interfa e su h as 802.11 an be adopted
for mobile hosts. Both approa hes are targeted within the MII (Mobile Information Infrastru ture) wireless LAN (see Figure 3.4).

The MII approa h with end-to-end ATM is alled

Broadband Adaptive Homing ATM Ar hi-

te ture (BAHAMA). The MII approa h with a non-ATM wireless segment is alled wireless

virtual LAN (WVLAN). It onsists of a network of PBSs, as does the BAHAMA LAN.
However, a WVLAN PBS has an ATM swit h and IEEE 802.11 interfa e ards, instead of
BAHAMA air interfa e ards. Both networking options ommuni ate with standard ATM
networks using a Gateway ATM swit h that provides proto ol onversion and interfa ing
fun tions.

3.2

Proto ol Sta ks

With its end-to-end ATM approa h, ATM networking has two potential working modes.
The rst mode, ommonly referred to as

native mode ATM has appli ation programs that

dire tly use an AAL (see Figure 3.5 (a)). The se ond mode, whi h provides ATM as a transport medium for the existing

Transmission Control proto ol/Internet proto ol

(TCP/IP),

3.3. The BAHAMA VP/VC on ept

Cross-connect

802.11
wireless
LAN

Applications
TCP
IP
802.11 MAC
802.11 PHY

25

ATM
switch

802.11
wireless
LAN

ATM
switch

MAC-LANE
bridge

MAC-LANE
bridge

LANE
802.11 AAL5
MAC
ATM
802.11 PHY PHY

LANE
AAL5
ATM
PHY

ATM ATM
PHY PHY

Applications
TCP
IP

802.11
MAC

802.11 MAC

802.11 PHY

802.11 PHY

AAL - ATM adaptation layer


AAL5 - AAL type 5
LANE - LAN emulation
PHY - Physical Layer

Figure 3.4:

Wireless Virtual LAN proto ol sta k [6

is known as TCP/IP over ATM(see Figure 3.5 (b)).


The tele ommuni ation market ex epts that most video and voi e appli ations will use native mode ATM. The BAHAMA sta k in orporates both native mode ATM and TCP/IP
over ATM alternatives. In this task, ustom wireless has a ustom physi al layer, multiple
a ess layer and a data link layer (see Figure 3.3).
For the wireless virtual LAN (WVLAN) the standard wireless interfa e overs the physi al
layer, data link layer and multiple a ess layer. Mobile hosts ommuni ate through hybrid
parts onsisting of 802.11 links and ATM onne tions (see Figure 3.4).[6

Applications
AAL

AAL - ATM adaptation layer


IP - Internet protocol
TCP - Transmission control protocol
WATM - Wireless ATM

Applications
TCP

WATM

IP

Custom wireless

AAL
WATM
Custom wireless

(a)

(b)

Figure 3.5:

3.3

proto ol sta k [6

The BAHAMA VP/VC on ept

In wired ATM networks the whole

Virtual Points/Virtual Conne tions (VP/VC)

on ept

is based on xed endpoints (see Chapter 2.2). But in wireless networks, mobility is a very
important issue. So, in the BAHAMA proje t a new VP/VC on ept that supplies slow
mobility (i.e. walking speed) was introdu ed. [7[13
In order to keep the PBSs simple and low- ost while maintaining high data rates (Gbit/s),
the need for VP/VC translation was eliminated.

This was done by assigning an unique

3. The BAHAMA network

26

VPI 9a - VCI 3
VPI 9b - VCI 3
1
PBS

VPI

9a

VPI

6
PBS

3
PBS

9a

7
PBS
5
PBS

VPI 9a - VCI 1
VPI 9b - VCI 2

4
PBS

2
PBS

8
PBS

VPI 9b

VPI 9b

VPI 9a - VCI 1
VPI 9b - VCI 2
VCI 9a - VCI 3
VCI 9b - VCI 3

9
PBS

Figure 3.6:

The VP/VC on ept [7

VPI to every PBS. As a onsequen e, a VPI orrespond to a parti ular destination PBS,
rather than to a virtual path of base stations and links. Virtual trees allow multiplexing
ells from dierent sour es to the same destination onto the same virtual tree. An example
of this an be obtained on Figure 3.6. VPI 9a and VPI 9b are assigned to PBS 9. In other
words, all ells with VPI 9a or 9b are routed to the destination PBS 9. To distinguish ells
from dierent onne tions, ea h PBS ontrols the VCIs. For example, (VPI 9a; VCI 1)
and (VPI 9b; VCI 2) are X-to-Y onne tions.

3.4
A

The Homing Algorithm

homing algorithm for routing ATM ells in wireless ATM networks is presented here, in

whi h users may move perhaps even during the middle of an a tive session. The example
that follows on Figure 3.7 illustrates this algorithm. Boxes 1-8 represent the PBSs, and
ir les A and B denote two (mobile) users that are ommuni ating with ea h other. A' and
B' denote the lo ations of A and B at a later time. The term lo al PBS is used to refer to
the PBS asso iated with a mobile's urrent position. A and B have wireless onne tions
to the network (at PBSs 2 and 7 respe tively), and the high-speed links between PBSs an
be either wired or wireless.

As lokal PBS

3
PBS
5
PBS

Source Home
for A-to-B
Connection

1
PBS

2
PBS

4
PBS

Bs lokal PBS

8
PBS

7 Destination Home
for A-to-B
PBS Connection

6
PBS

Figure 3.7:

The Homing Algorithm [7

To maintain in-sequen e ATM transmissions as users move during a onne tion, the on ept of a

Sour e Home Station and a Destination Home Station is introdu ed.

These refer

to parti ular PBSs asso iated with a onne tion and play the main role in maintaining ell
sequen e.

ATM ells from user A that are destined to user B are rst routed from A to the Home PBS
for A. The ells are then routed along a pre-determined virtual path from the Sour e Home
PBS to the destination Home PBS where they are buered and then delivered in-sequen e

3.4. The Homing Algorithm

27

to B's Lo al PBS.
Initially, a virtual path from 2 to 7 (passing through 4 and 6) transports ATM ells from
A to B. That is, PBS 2 is the Lo al PBS for A and also the sour e Home PBS for the
A-to-B onne tion. Similar, PBS 7 is the Lo al PBS for B and also the Destination Home
for the A-to-B onne tion. When A moves to lo ation A' (with a wireless onne tion to its
new lo al PBS 3), the ATM ells are rst routed along a pre-determined path from PBS
3 ba k to home PBS 2, and then along the virtual path from 2 to 7. If B has also moved,
for example to B', then B's Home PBS 7 will forward the ATM ells to B's Lo al PBS 8
(whi h is its urrent position) again using a pre-determined path.
Sin e it is ine ient always to re-route ells to/from Home Stations, the lo ations of the
Home PBSs are slowly updated, as users move through the network. For example, after A
moves to A', PBS 3 an be the new Sour e Home. Mobility aspe ts are not subje t of this
thesis and will not be dis ussed any further.[7

28

3. The BAHAMA network

CHAPTER 4

The Medium A ess Control Proto ol DQRUMA

n a wireless network that supports multimedia tra , the Medium A ess Control
(MAC) proto ol needs to share the limited bandwidth e iently and fair. The MAC

Distributed-Queueing Re-

proto ol used in the BAHAMA network is alled DQRUMA (

quest update Multiple A ess).

We onsider a ell with a PBS and N WTs (see Figure 3.1).

Fixed-length pa kets (ATM ells) arrive at the mobiles a ording to diverse random pro esses.

These are buered at the mobiles until they are transmitted uplink to the Base

Station. Pa ket transmission follows a hannel a ess proto ol embedded at the PBS. The
PBS broad asts downlink pa kets that are destined for mobiles within its ell.

Time slot k-1

Time slot k

Time slot k+1

Packet Xmt Channel

Uplink

Request Access(Random) PGBK Request


(Contention-free)
Packet Xmt Channel

Downlink

Xmt_Perm for the Next Slot


ACK of Request Access

Figure 4.1:

DQRUMA proto ol stru ture [14

Frequen y-Division Duplex)

Two proto ol implementations are possible. The FDD (

mode

uses two dierent frequen ies, one for the uplink (mobile-to-base) and one for the downlink
(base-to-mobile) transmission.
downlink bandwidth, a TDD (

In order to a hieve dynami sharing of the uplink and

Time-Division Duplex)

slot allo ation s heme an be used.

Here the uplink and downlink hannels are time multiplexed on a single frequen y hannel.
This thesis on erns only FDD related proto ol matters.

4.1

DQRUMA des ription

Figure 4.1 shows the slot stru ture and Figure 4.2 shows a ow hart of the DQRUMA proto ol. DQRUMA onsiders a time-slotted system in whi h a
and a

Pa ket-Transmission (Xmt) Channel are

Request A ess (RA) Channel

formed on a slot-by-slot basis. If desired,

larger or smaller numbers of RA hannels an be allo ated than indi ated in Figure 4.1.

As the base station re eives transmission requests from the mobiles by listening to the RA

Request Table has has


A ess
Identi ation (A ess ID) and an asso iated eld ontaining Transmit-Request (Xmt-Req)

hannel, a

Request Table at the PBS is updated appropriately.

The

entry for every of the N mobiles in the system. Ea h entry ontains a mobile's short

4. The Medium A ess Control Proto ol DQRUMA

30

Listen to
Xmt_Perm
Channel

Empty Buffer

No

New Packet
Arrival?

Xmt_Perm?

Yes

Yes

Request Access
via
RA Channel

No

ACK?

Transmit Packet
with PGBK Req
via Packet Xmt
Channel
Yes
Buffer
Empty?
Yes

Figure 4.2:

No

DQRUMA ow hart at mobile [14

No

4.1. DQRUMA des ription

31

information. (e.g. whether or not the mobile has any pa kets to transmit).

When a pa ket arrives to a mobile with an empty buer, the mobile sends an one byte long
Xmt-Req to the Base Station via the uplink RA hannel, perhaps in ontention with other
mobiles a ording to some random a ess proto ol (e.g. Slotted ALOHA). The Xmt-Req
from a mobile in ludes the mobile's A ess ID. When a base station su essfully re eives
a Xmt-Req from a mobile, it sets the orresponding entry in the request table to indi ate
that this mobile has one or more pa kets to transmit (this is the Request Aspe t of the
DQRUMA proto ol). The PBS also a knowledges the re eption of the Xmt-Req by broad asting the mobiles A ess ID over the downlink

A knowledgement hannel (ACK hannel).

On e a mobile re eives positive a knowledgement (by listening to the downlink ACK hannel) that its Xmt-Req was re eived by the PBS, it listens to the downlink

Permission Channel

Transmit-

(Xmt-Perm) every time slot until it hears its own A ess ID, whi h

signies permission from the base station to transmit a pa ket during the next time slot.
(this is the Distributed-Queueing Aspe t of the DQRUMA proto ol). Pa kets are queued
at the mobiles and the PBS serves them a ording to a desired pa ket transmission poli y
(e.g. Round-Robin). The transmit permission is given ea h time slot to one of the mobiles
that has a non-empty Xmt-Req eld in the request table.

The Round-Robin algorithm an be des ribed as follows (see Figure 4.3):

 Slot timer a tive


 Xmt-Req eld of mobile 2 not empty
 Xmt-Perm for next UL slot to mobile 2
 Delete entry
 Pointer to next eld
Xmt_Req
Field #2
not empty

#1

#N

#2

#3

#4

Figure 4.3:

Round Robin s heduling

Ea h time a mobile transmits a pa ket using the uplink Xmt hannel, it also in ludes a
Piggyba king (PGBK) Request Bit to indi ate whether it has more pa kets to send. This
possibility oers a ontention-free Xmt-Req for the mobile transmitting a pa ket.

The

implemented DQRUMA-proto ol in ludes one byte for piggyba king purposes that arrys

4. The Medium A ess Control Proto ol DQRUMA

32

only one bit information.

The PBS he ks the PGBK Request Bit and updates the appropriate entry in the Request
Table. On e a mobile su essfully sends a Xmt-Req over the RA hannel and as long as its
buer remains not empty, the Xmt-Reqs for additional pa kets arriving at the mobile are
sent to the PBS by piggyba king the requests in a ontention-free fashion. In other words,
only pa kets arriving to a mobile's empty buer will send a Xmt-Req over the RA Channel.
The timing diagram on Figure 4.2 presents an overview of the DQRUMA-proto ol a tions
both at mobile and PBS.

Sin e there is just a single uplink frequen y, and sin e Xmt-Perm are sent downlink on a
Slot-by-Slot basis, it is su ient to use a single bit for the PGBK Request Bit. Also, the
Request Table just needs to indi ate whether or not ea h mobile has any more pa kets to
transmit. More information about the number of pa kets a mobile is waiting to transmit
is usefull:

 if there are multiple Pa ket Xmt Channels (e.g. on multiple frequen ies) per time slot,
 if the desired pa ket transmission poli y (implemented at the Base Station) depends
on the a tual number of pa kets queued at ea h mobile,

 if the round-trip propagation delay is large.

If the length of the (uplink) PGBK Request is less than the length of the (downlink) Transmit Permission, more guard time on the Uplink hannel than on the downlink hannel an
be in luded.[14

4.2

Proto ol states

DQRUMA's fun tionality for es the mobiles to a t and rea t in a determined matter. The
N buered mobiles in the system are ea h in one of the three states:
1. Empty
2. Request
3. Wait-to-Transmit
The possible states and the transitions between them are depi ted in Figure 4.5. Mobiles

Empty State.

When a pa ket arrives to the buer

of a mobile in the empty state, the mobile enters the

A mobile in the Re-

with empty buers are said to be in the

Request State.

quest State sends its Xmt-Req via the RA hannel and stays in the Request State until
its Xmt-Req is su essfully re eived by the PBS. When a mobile in the Request State
su essfully sends a Xmt-Req, it swit hes to the

Wait-to-Transmit State.

A mobile in the

Wait-to-Transmit state listens to the Xmt-Perm Channel until it hears its A ess ID, then
transmits a pa ket in the Xmt Channel in the next time slot, and also sends a ontentionfree Xmt-Req (if ne essary) using the PGBK Request Bit.
If the Mobile transmits a non-zero PGBK-bit indi ating that the mobile's buer is not
empty, the mobile stays in the Wait-to-Transmit State.
PGBK-bit it returns to the

Empty State.

If the mobile transmits a zero

A hannel error at the RA hannel ould disturb the transmission of a Xmt-Req.

The

mobile whi h send the Xmt_Req will not re eive an ACK within a predetermined time

4.2. Proto ol states

Protocol Actions
at Mobile

33

WT

PBS

Set ACK-timer

RA

(Collision)

Timer active
Set Back-off Timer

RA

Update Request Table


Send Acknowledge

ACK

Stand-By

Xmt_Perm with
Round-Robin

Xmt_Perm

Transmit Packet
with PGBK-bit

Protokol Actions
at Base Station

Send Downlink Packet


Pa

ck
et+
mt PG

BK

tX

e
ack

bit

set

Xmt_Perm
Pa

ck
et+
PG
t
m

tX

e
ack

Figure 4.4:

BK

no

ts

et

DQRUMA timing diagram

4. The Medium A ess Control Proto ol DQRUMA

34

Failure

EMPTY
STATE

PIGGYBACKING

PACKET
TRANSMIT

REQUEST
STATE

REQUEST
ACCESS

WAIT-TO-TRANSMIT
STATE

Success

Figure 4.5:

DQRUMA hannel model

margin and a timer will send the mobile ba k to the

Request State.

DQRUMA does not

provide a me hanism of a knowledging a PGBK bit. In ase of an errored transmission a


PGBK bit ould get lost and the mobile whi h send it would remain onstantly in the

to-Transmit

Wait-

State, without any possibility to exit. In su h a ase there are two possible

ways to exit the deadlo k:

 After a PGBK bit is send an additional timer ould be set at the mobile. If the timer
runs out after a desirable time, e.g. the sum of time slots equal to the number of
mobiles in the system (this is the time limit the s heduler would have needed to serve
the mobile), the mobile an move to the

Request State.

 The DL proto ol stru ture an be modied, in order to omprise a new PGBK_ACK


hannel, whi h would a knowledge the transmitted PGBK bits. The PGBK_ACK
hannel would then in lude 1 byte. The byte is required to address the ACK, sin e it
is not possible for the re eiving mobiles to know when to expe t it.

4.3

Slot and Burst Stru ture

Table 4.1 presents the ontribution of the single hannels, as presented in the previous
se tions, to the DQRUMA-proto ol:
In Figure 4.6 the proto ol's burst stru ture is indi ated. An uplink ontains two bursts.
The downlink information on the other side is transmitted in a single burst.
The rst UL burst happens in the RA hannel. This hannel an be a essed by every mobile in the lo al ell. The se ond burst in ludes the PGBK and the Pa ket-Xmt hannels.
This hannel is used by one mobile at the time, whi h got the Xmt-Perm by the PBS to
transmit its ATM ell.
In order to guarantee the proto ol's orre tness, propagation and hardware spe i param-

Guard
ele tromagneti waves (EM-waves).

eters need to be taken are of. The propagation delay is onsidered in the parameter:

Time.

This has to do with propagation speed of the

Sin e we propose a ell of one hundred meter radius, we hoose a guard interval of twi e
the time for an EM-wave to over this distan e. The parameter

Rise-and-Fall Time on-

tains hardware related inuen es. Both proto ol parameters extend the total slot length
as shown in Table 4.2.

4.4. DQRUMA Enhan ements

35

Uplink Channels

Number of bits

Request A ess(RA)

Piggyba king (PGBK)

Pa ket Xmt

424

Downlink Channels
A knowledge (ACK)

Transmission Permission (Xmt-Perm)


Polling

Pa ket Xmt

424

 In ludes 8 bit in the urrent implementation


 The polling hannel is only in the polling DQRUMA-enhan ement

present

(see se . 4.4.2)

Table 4.1:

Slot stru ture

RA

PGBK

Guard Intervals

Packet Xmt Channel

Uplink

Rise/Fall Times

Xmt_
ACK Perm

Figure 4.6:

Packet Xmt Channel

Downlink

DQRUMA slot bursts

The parameters ontribution for a single burst has a total physi al overhead of 1.96

4.4

s.

DQRUMA Enhan ements

The DQRUMA Proto ol presented in the previous se tion an be used together with any
random a ess proto ol for the RA Channel and any s heduling poli y a ording to dierent
tra and servi e requirements.

4.4.1 Collision Resolution


It is now possible that all entries in the Request Table have empty Xmt-Req elds. As a
result, the uplink Xmt Channel will be idle in the next time slot, sin e the PBS does not
know whi h mobile, if any, have pa kets to transmit.

However, to avoid wasting trans-

mission time, the (otherwise) idle uplink Xmt Channel an be onverted into multiple
RA Channels in the next time slot as an be obtained in Figure 4.7. This also serves to
relieve ontention on the RA Channel in ase many mobiles are lo ked in the Request State.

In the 'normal' mode of operation, when at least one mobile has a non-empty Xmt-Req
eld in the Request Table, the PBS uses the downlink Xmt-Perm Channel to announ e
whi h mobile an transmit a pa ket in the next uplink Pa ket Transmission Channel. If

4. The Medium A ess Control Proto ol DQRUMA

36

Parameter Name

Duration in

Guard Time

0.66

Rise Time

0.65

Fall Time

0.65

Table 4.2:

s

Guard Intervals

all the Xmt-Req elds in the Request Table are empty, then the PBS uses the downlink
Xmt-Perm Channel to announ e (by sending a multi- hannel RA) that the next uplink
Xmt Channel will be onverted into multiple RA Channels, allowing more mobiles in the
Request State to send their Xmt-Reqs a ording to some multi- hannel random a ess
pro edure. To eventually broad ast multiple A knowledgements, the next downlink Xmt
Channel is similarly onverted into multiple ACK Channels, delaying by one slot the downlink pa kets to the mobiles (see Figure 4.7).

PGBK

RA0

Packet Xmt Channel

ACK 0

RA 0

Xmt_Perm
Packet Xmt Channel

RA 1 RA 2

RAM-1

Uplink

Downlink

Uplink

Xmt_Perm

ACK 0 ACK 1 ACK 2

Figure 4.7:

ACK M-1

Downlink

multiple RA hannels

Even if some entries in the Request Table are non empty, it is still possible to dynami ally
onvert an uplink (downlink) Xmt Channel to multiple RA (ACK) Channels whenever the
PBS senses there are many mobiles in the Request State whi h annot broad ast their
RAs.

This an be a hieved by observing the number of ollisions on the RA Channel.

This adaptive ontrol of the RA bandwidth adds one additional time slot delay for uplink
and downlink pa ket transmissions, but the benet is that the PBS gets a more a urate
pi ture of whi h mobiles are waiting to transmit pa kets. This information may help the
PBS determine the order in whi h mobiles should transmit their pa kets. A multi- hannel
RA an also be released by observing the number of UL time slots with empty RA and
PGBK-bits. [14

4.4.2 RA Channel Polling


In this se tion a DQRUMA-proto ol enhan ement is presented. The idea is to extinguish
the ollisions on the RA hannel by simply polling every mobile with an empty eld in the

Request Table to send its transmission request (Xmt-Req), if its buer is not empty, within
the next uplink slot over the RA hannel (see Figure 4.8).

4.4. DQRUMA Enhan ements

37

In order to expand the proto ol to support the aforementioned fun tionality, some modi ations in the proto ol stru ture need to be made. All modi ations are on entrated on
the PBS stru ture. More pre ise, the s heduler must be provided with a se ond pointer,
whi h he ks the request table for empty elds in a Round-Robin manner. After nding an
empty eld the proper mobile ID is transmitted over the downlink

Polling Channel (PA)

(see Figure 4.8).

The total slot length still remains the same. Sin e in the uplink two

Guard and two Rise-

and-Fall intervals are in luded, the downlink slot length remains smaller even after an one
PGBK

PGBK

byte long downlink hannel is added.

Uplink Packet

RA

ACK

RA_
poll

Xmt_
Perm

Uplink Packet

RA

Downlink Packet

ACK

RA: Request Access


PGBK: Piggybacking bit
RA_poll: Request Access Polling

RA_
poll

Xmt_
Perm

Uplink

Downlink Packet

Downlink

ACK: Acknowledge
Xmt_Perm: Transmission Permission

Figure 4.8:

RA Polling

4.4.3 Ideal DQRUMA


The

DQRUMA ideal system is a modi ation of the original DQRUMA, in order to elim-

inate the inuen e of the slotted ALOHA request a ess hannel on the proto ol, so that
the performan e of the Round-Robin algorithm an be evaluated. This is a hieved by simply writing the RAs dire tly to the RT, as soon as they are about to leave the terminals,
without being transmitted over the RA hannel. This is explained in Figure 4.9.

PBS

RA

RA

RA

RA

Uplink

UL slot

Figure 4.9:

The prin iple of ideal DQRUMA

4.4.4 Nominal Rate on DQRUMA


4.4.4.1 Introdu tion of the Nominal Rate (NR)
NR is a new te hnique of omputing an alternative data rate of a onne tion.

4. The Medium A ess Control Proto ol DQRUMA

38

The NR is dened as the minimal data rate to serve a tra sour e, so that it an guarantee ertain QoS requirements. It an be al ulated from the following tra and QoS
parameters (see Chapter 2.4):

 Peak Cell Rate: P CR


 Maximum Burst Size: MBS
 maximum Cell Transfer Delay: maxCT D
 Sustainable Cell Rate: SCR
The following system parameter is also needed:

 Buer size: nB

If the buers are smaller than the

nB

 MBS

Maximum Burst Size, then the NR is:

N

nB
 P CR
 MBS
MBS

(4.1)

This equation fullls the ondition of serving as many ells as possible with the PCR, so
that the rest of the burst an be put into the buers without any loss of ells.
If the buer is bigger than the

Maximum Burst Size,

then the ells of the burst an be

expanded, so that the last one an still be transmitted within the

Delay.

maximum Cell Transfer

maxCTD

T
arrival
time

transmission
time

Figure 4.10:

Burst expansion [21

arrival-axis in Figure 4.10 there is an in oming burst of six ells with a peak ell
1
of P CR =
T . On the axis beneath (transmission) the outgoing ell sequen e an be

On the
rate

obtained. As it an be seen, the last ell is transmitted within the time period maxCTD.

B an
burst size MBS :

The burst duration


maximum

be al ulated from the peak ell rate

B

= MBS
P CR

P CR

1
T

and the

(4.2)

Hen e, the nominal burst duration an be omputed:

N
From Figure 4.10 an be obtained:

= MBS
N

(4.3)

4.4. DQRUMA Enhan ements

39

N
So, it is for the nominal rate

N

N

= B + maxCT D

(4.4)

(see Equation 4.2, 4.3 and 4.4):

MBS
 P CR
 P CR  maxCT
D + MBS

(4.5)

Thus, it is for the nominal rate:

8
MBS nB  P CR
>
>
MBS
<
MBS
N  max > PCRmaxCTD
+MBS  P CR
>
>
:

(4.6)

SCR

SCR stands here for Sustainable Cell Rate ( f. Se tion 2.4). [17[21

4.4.4.2 The Virtual Frame


Several modi ations to the DQRUMAs s heduler need to be done, in order to integrate
the

Nominal Rate (NR) algorithm in the present system.

NR is invented for a frame stru -

ture, where the s heduler plans the hannel allo ation assigning the slots to the terminals
at the beginning of a frame.

"Virtual" Frame
VBR 1
VBR 2
VBR 3
CBR 1
VBR 4
VBR 1
VBR 2
CBR 2
VBR 3
VBR 4
VBR 1
CBR 3
VBR 2
VBR 3
VBR 4
CBR 4

CBR traffic

NR VBR : NR CBR
3
:
1

Figure 4.11:

Virtual Frame for a full RT

Sin e DQRUMA uses a slot-by-slot allo ation s heme, a new appli ation form of the NR
algorithm must be invented. For this, we introdu e the

Virtual Frame (VF) on ept.

The

4. The Medium A ess Control Proto ol DQRUMA

40

VF is a slot allo ation map for the future slot assignment and it oers the advantage of
hanging dynami ally.

The VF stru ture takes pla e beside DQRUMA's

Request Table

(RT). The terminals still ompete for the RA hannel to update the proper elds in the
RT. At the beginning of a time slot the s heduler he ks the VF to nd out whi h terminal/servi e should be transmitted and then looks at the
this terminal/servi e has the proper eld set.

Request Table

to see, whether

The following example demonstrates this

method. It on erns 10 dierent data onne tions from 10 dierent terminals:

 four VBR terminals (VBR 1-4)


 four CBR terminals (CBR 1-4)
 two ABR terminals (ABR 1-2)
At the beginning of the onne tions, their nominal rates are being omputed a ording
to se . 4.4.4.1. For this example we assume a nominal rate ratio of VBR to CBR tra
equal 3/1. Nominal rate is not adjusted to ABR servi es. ABR ells are buered and not
transmitted as long as

RT always full:

real time (rt) onne tions have their elds in the RT set.

If the Request Table has always the rt- onne tion elds set, then the

s heduler will serve the terminals a ording to the omputed NR ratio, namely 3/1, as

shown in Figure 4.11. Conne tions within a servi e lass are not served onsequently,
unless one or more onne tions have not updated their Xmt-Perm elds. It is more
like an internal multiplexing taking pla e, in order to a hieve fairness for all servi e
lass onne tions (see Figure 4.11).

One servi e lass missing:

For the ase that a servi e lass is missing (e.g. VBR), the

existing one (e.g. CBR) will be ontinuously served until there are no rt- onne tions
to be served or a VBR terminal updates its proper eld in the RT (see Figure 4.12).

"Virtual" Frame
VBR 1
VBR 2
VBR 3
CBR 1
CBR 2
CBR 3
CBR 4
CBR 1
CBR 2
VBR 4
VBR 1
VBR 3
CBR 3
ABR 1
ABR 2
ABR 1

Figure 4.12:

No VBR Xmt_Perm fields set

RA from VBR Terminals 1,3 and 4

No rt connections

Virtual Frame for one (or none) rt onne tion

4.5. Other proto ols

41

4.4.5 TDD implementation


Furthermore it is possible to a hieve dynami ontrol of the uplink and downlink hannels
in a TDD (time-division duplex) implementation, where the uplink (mobile-to-base) and
the downlink (base-to-mobile) are time-division multiplexed on the same frequen y hannel. Thus, the PBS an respond to urrent onditions and hange the amount of bandwidth
assigned to the uplink and downlink hannels.

This dynami sharing of uplink and downlink bandwidths is managed by treating the PBS
like a 'super mobile'. The PBS simply keeps an entry for itself in the Request Table, and
announ es over the Xmt-Perm Channel its own ID (whi h every mobile re ongnises) when
it is going to transmit a 'downlink pa ket' during the next time slot. In other words, in a
TDD implementation, the ID announ ed over the Xmt-Perm Channel indi ates whether the
following time slot will be for uplink or downlink ommuni ation. The pa ket transmission
poli y at the PBS will most likely need to serve the PBS more often than the mobiles (e.g.
using a weighted Round-Robin poli y), sin e the PBS needs to transmit all the 'downlink'
tra . The TDD proto ol implementation will not be dis ussed in the following.

4.5

Other proto ols

This se tion presents the

slotted ALOHA

proto ol whi h is used for transmission of the

mobile's Xmt_Reqs. Further, a shord des ription of three existing MAC proto ols is given.
Their omparison with the DQRUMA proto ol an be found in Chapter 6.

4.5.1 Slotted ALOHA


The multiple a ess proto ol ALOHA was developed in 1970 at the University of Hawaii
and it was designed for ground based radio broad asting systems. The basi idea is that a
user transmits data whenever he has data pa kets to send. If the transmission failed the
transmitter waits for an arbitrary time until it retransmits the ollided data pa ket. The
Slotted ALOHA-Proto ol denes slots in whi h data pa kets an be sent with the advantage that a sending station is not interfered by a station

starting with

transmission. The

maximum throughput for the pure ALOHA proto ol is 1/2e and for the slotted ALOHA
proto ol the maximum throughput an be al ulated to 1/e.
The simplest way to implement a slotted ALOHA proto ol would be to provide always one
RA hannel per slot. The base station evaluates the terminal transmissions and re ognizes
if there was a ollision. Via a feedba k the terminals are informed whether the RA slot
was su esfully a essed. Ea h ollided terminal draws a random number (maxRBO) that
spe ies how many slots to wait until a essing the RA hannel again. The range of this
number has to be set to a meaningful value.
until the next a ess takes pla e.

If the range is too big, mu h time passes

The result would be that a terminal does not get the

apa ity needed for the transmission of LLC-PDUs, sin e too mu h time passes, until the
base station re eives the dynami parameters.

On the other hand the probability for a

new ollision in reases if the range for the random number is too small and as result the
system ould be ome unstable.

In Systems where a random a ess is used relative often it might be a bottlene k to oer
only one RA slot per period be ause the probability for a ollision is relative high then and
the ollision resolution takes too mu h time. This fa t leads to a dierent implementation

4. The Medium A ess Control Proto ol DQRUMA

42

for the slotted ALOHA proto ol for ollision resolution.

4.5.2 The Dynami Slot Assignment Proto ol DSA++


The

Dynami Slot Assignment

(DSA++) proto ol was developed at COMNETS at the

University of Te hnology in Aa hen for transmission in mobile broadband systems.

Its

basi idea is to assign the apa ity dynami ally to the users depending on their a tual
requests. Dynami parameters depi t as exa tly as possible the ommuni ation request of
a mobile terminal. They are transmitted on the uplink to the base station. With this information the base station is able to make the slot allo ation for the users within its radio ell.

Time
Frame

Frame

DownInfoPDUs

Frame

UpInfoPDUs
TTA

DSA++

RA RA TTA

PeriodCtrlPDU

Figure 4.13:

DSA++ proto ol proto ol Stru ture [18

The DSA++-Proto ol organizes the ommuni ation into frames or so alled signalling periods. Ea h frame starts with a broad ast PDU whi h signals the reservation of slots within
one frame. The broad ast PDU is alled the

Period ConTRoL (PCTRL) PDU.

A signalling period onsists beside the PCTRL-PDU of downlink and uplink transmission
phases. Between the hange of the send dire tion a

Trans eiver Turn Arround

(TTA) time

is in luded. This is done to let the send signal de rease before the re eption starts. The
RA slots belong to the RA hannel of the DSA++ proto ol. This hannel is a essed by
terminals that have urgent dynami parameters or a knowledgements to transmit.

One

proto ol whi h ome to onsideration for the RA hannel is slotted ALOHA. The feedba k
for the ollision resolution of the RA hannel is broad asted within the PCTRL-PDU and
the mobiles are informed whether their transmitted pa kets were ollided or not.

After a terminal be omes an UL slot assigned, it an use the PGBK option oered in
ea h data slot. Thus, the terminal an transmit its request for the next data frame without needing to a ess the RA hannels avoiding possible ollisions whi h would delay the
transmission.

Figure 4.13 shows a DSA++ proto ol omposing frames of a onstant length.


kind of implementation is to use a variable length for the signalling period.

Another

If there is

nothing to send, only empty frames are sent. Then only PCTRL-PDUs a ompanied with
the TTA time and one or more RA slot as oered in default, allo ate the hannel apa ity.
Before ommuni ation takes pla e, the signalling period has to be omposed.

After the

4.5. Other proto ols

43

s heduling is nished, the apa ity assignments for the just beginning period are sent
within the PCTRL-PDU to the terminals. [18

4.5.3 FDD DSA++ MAC-Proto ol


This proto ol is invented for the SAMBA proje t.
in Figure 4.14.

Phase

Its blo k stru ture an be obtained

The FDD DSA++ proto ol operates in two phases.

plans the next signalling period.

In the

Transmission Phase

The

reservation

the ATM ells are

transmitted. The terminal requests for the next signalling period are transmitted within a

signalling burst in the reservation Phase.

The DL signalling is broad asted to the mobiles

within a DL signalling burst.

Downlink
Signaling Burst

Downlink Signaling Period

Announcements

Downlink

....

....
Feedback
Reservations

Uplink
N-1

Slot N

N+1

N+2

Offset

Random
Access
N+3

Uplink Signalink Period

Figure 4.14:

Signalling of the FDD DSA++ Proto ol [16

A DL signalling burst ontains the following parameters:

 A reservation request for every UL slot


 An Announ ement Message for every DL slot
 A Feedba k Message for every Random A ess CH of the last UL signalling period
 A eld for system signalling, e.g. Information Channels
Announ ement messages oer the possibility to the mobiles of swit hing to the power-save
mode, without desyn hronization or loosing addressed ells. Binary Feedba k messages are
used for ollision resolution of the requests transmitted in the Random A ess Mode. The
Random A ess CH are used for request transmission, but they are also usefull for terminal
registration. The Slotted ALOHA proto ol is performed on these hannels. [16

4.5.4 Energy-Burst MAC-Proto ol


Energy-Burst

is a MAC proto ol invented by

Philips.

Its blo k stru ture is depi ted in

Figure 4.15. The data frame is divided in time slots that ontain an ATM ell and signalling information, whi h an be used for

Piggyba k transmission.

Thus, the Base Station

an re eive apa ity requests for the next frame from the terminals that use these slots.
There is also a possibility that a terminal does not own a slot. Su h terminals use a spe ial
proto ol feature: The

Energy-Polling mode.

4. The Medium A ess Control Proto ol DQRUMA

44

At the end of ea h MAC frame there are many small hannels (1-2
is adjusted to a

single

s).

Ea h of them

WT. A terminal, whi h wants to request Xmt permission sends a

polling CH. The Base Station adjusts in the next


MAC-frame a super slot to the mobile. In this super slot the mobile an transmit its rst
short energy signal in the a ordinate

ATM ell together with its a tual tra parameters. The positioning of the energy-bursts
at the end of a frame redu es the number

tran eiver turn around time (TTR) to a minimum

of two. [17

time

...

(n-1) MAC-frame

(n) MAC-frame

downlink

TTA
0
1

(n+1) MAC-frame
TTA
11
00
00
11
00
11
00
11
00
11

uplink

1
0
0
1
0
1
0
1

Uplink-Data-Sequence
superslot

PCTRL-PDU
Period-Control
PDU

ATMcell

ATMcell

normal
slot

normal
slot

ATMcell

ATMcell

energybursts

superslot-signalling field

Figure 4.15: MAC-frame with Energy-burst Signalling(Philips) [17

...

CHAPTER 5

BONeS Simulator

ONeS Designer is a software pa ket for design and analysis of system ar hite tures,
networks and proto ols.

of a system.

BONeS Designer

models the proto ol/messaging layers

It an be used to dene the system ar hite ture with an abstra t model

that treats the system as a olle tion of shared resour es (CPUs, buses, memory, database
servers, et .), queues and delays.
The system ar hite tural model des ribes the ow of transa tions through various pro essing elements.

By running a simulation on the ar hite tural model, performan e metri s

su h as response time, throughput, utilization, and error probability an be evaluated.

5.1

Building Simulation Models

A typi al

BONeS Designer proje t involves reating data stru tures, onstru ting system

models, running simulations and evaluating the results.

Data stru ture formats, blo k

diagrams and simulation plots are all displayed graphi ally.

5.1.1 Constru ting Data Stru tures


The

Data Stru ture Editor

(DSE) an be used to dene and hierar hi ally organize data

stru tures. They ontain basi information about an entity, su h as size, pro esses, resour e
requirements, and status.

The data stru tures also ontain performan e measurement

information. This is useful for generating statisti s on performan e, su h as throughput,


delay and utilization.

5.1.2 Building System Models


The

Blo k Diagram Editor

(BDE) is used to graphi ally onstru t system models.

a model is built, all onne tions are automati ally type and onsisten y he ked.

As

Envi-

ronmental models, modules, and fun tional building blo ks are stored in Designer's Core
Library. The Core Library an be further extended by reating new blo ks out of existing ones, by reating blo ks with C or C++, using a Designer-generated template, or by
onstru ting nite state models using the optional Finite State Ma hine editor.

5.1.3 Running Simulations


The

BONeS Designer

Simulation Manager

monitor simulation studies.

(SM) is used to set up, start, ontrol and

First exe ution parameters (tra rates, buer sizes, bus

bandwidth, CPU and disk speed, et .) must be set and probes whi h olle t and pro ess
data during the simulation must be pla ed.

5.1.4 Evaluating Simulation Results


The

Post Pro essor

(PP) is used to organize and display data generated by simulations.

Multiple simulations of the same model are grouped together in libraries so that statisti al

5. BONeS Simulator

46

measures (su h as averages and onden e intervals) and performan e measures (su h as
throughput and delay) an be omputed and displayed as fun tions of independent variables
(su h as the system load).

5.1.5 Intera tive Simulation Manager


The

Intera tive Simulation Manager (ISIM) allows to exer ise intera tive ontrol and anal-

ysis during exe ution of a simulation. System parameters an also be altered and repeat
the pro ess as often as desired to optimize a design. Some ISIM Features are:
1. Animation of simulation progress by highlighting the urrent exe ution point (blo k
and port).
2. A variety of step, next, break, and ontinue ommands for intera tive ontrol of
simulation exe ution.
3. Breakpoints or ags to automati ally suspend exe ution of the simulation when a
spe ied ondition is met.
4. Graphi al display of instantaneous (not pre-re orded) values of data in any part of
the system.

BONeS Designer

BONeS Designer

Optional Products
Finite State Machine
Editor
Interactive Simulation
Manager

Data Structure Editor


Block Diagram Editor
DESIGNER
Symbol Editor

MAC Modules

Core
Library

Simulation Manager

Network Modules

Post Processor

Projekt Editor

Figure 5.1:

BONeS-Designer Stru ture [11

5.1.6 Proje t Editor


The

Proje t Editor

(PE) provides a means of organizing large proje ts and managing

them more ee tively.

Proje t Editor allows better visualizing of proje ts by displaying

hierar hi al views and using olor to demonstrate dependen y and other relationships.
Figure 5.5 gives an overview of a simulated System.

From left to right all basi system

templates are mapped as well as their relative hierar hi al order (see Se tion 5.4.1).[1

5.2

Simulation ar hite ture with BONeS

In order to build a layered System using the fun tionality of

BONeS Designer

all layers

should have an uniform stru ture, so that a single layer an be easily taken from the system
to mount a new one. The ommuni ation between the single layer blo k elements o urs
over stri t dened interfa es. A layer blo k ontains three main omponents:
The

Control Unit

5.3. Interfa e primitives

47

Sub-Blo ks
The Interfa es
The

Higher Layer Interface

Control Interface

Functionality
Unit
Control
Unit

Sub-Block A

Control Interface

Higher Layer Interface

Sub
Blocks

Control
Unit

Lower Layer Interface

Sub-Block B
Lower Layer Interface

Figure 5.2:
The

Control Unit

Blo k stru ture [11

primarily initializes the

Sub Blo ks.

These an ommuni ate over the

interfa es with other layers. This happens dire tly or over the

sub-blo ks an ontrol the signalling sequen e in the System.

Control Unit,

so that the

A layer ommuni ates with the environment in three ways (see Figure 5.2). These are:

 Higher Layer Interfa e


 Lower Layer Interfa e
 Control Unit
The fun tionality of a module an be spread over many single blo ks. A typi al example
of su h an ar hite ture is presented in Figure 5.2 (Sub-Blo k B).

5.3

Interfa e primitives

It is very important for a modular ar hite ture, that internal and external ommuni ation
is well dened. The external ommuni ation happens by using a ommuni ation set of, so
alled,

Primitives, whi h onsist of a ertain stru ture.


Source
Address

Destinat. Primitive
Protocol Data Unit
Address Type

PCI

Source
Address

SDU/User Data

Figure 5.3:

Destinat. Primitive Interface Control


Address Type
Information

General
Type

Special
Type

Control
Data

Data- & Control Primitives [11

The primitives an be en apsulated into data pa kets whi h are of two types (see Figure
5.3).

Both types have the same internal stru ture.

The  Sour e and the  Destination-

5. BONeS Simulator

48

Address an be the logi al address of a layer or sub blo k. The

ISO/OSI Referen e Model

denes four primitive types:

 Request
 Indi ation
 Response
 Conrmation
The main dieren e between a  Control-Primitive and a  Data-Primitive is fo used on
the nal Field of the Pa ket (see Figure 5.3).
tains the

Interfa e Control Information(ICI),

For a  Control-Primitive this eld onwhi h further ontains other data types

(see Figure 5.4):

 General Type
This type mostly spe ies layer and terminal features, su h layer initializing,

ne tion Admission Control (CAC) and intra-layer


Spe ial Type

Con-

ommuni ation.

It spe ies layer signalling pro edures and it is appli ation spe i .

 Control Data

It ontains ontrol information for stati al or dynami parameter ex hange.


A Data Primitive (DP) ontains a

Proto ol Data Unit (PDU) instead of the ICI. The PDU

also uses a pa ket stru ture, whi h in ludes:

 PCI, Proto ol Control Information


 SDU or User Data
The latter en apsulates ATM ells or signalling information.[2[3[11[19

Name: Interface Control Information [PhilipsArchitecture]


Date: Tuesday, 4/14/98 11:33:47 pm [1]
Name

Type

Subrange

General Primitive Control Type General Primitive Control Type ...


Special Primitive Control Type ROOTOBJECT
...
Data
ROOTOBJECT
...

Figure 5.4:
5.4

The Data Stru ture

Default Value
Init
...
...

Interfa e Control Information [11

Implementation

5.4.1 PED and System ar hite ture


In this se tion the Wireless Terminal (WT) stru ture and an entire system in

Designer

BONeS

simulation environment will be introdu ed. For attaining an overview over the

hierar hi al simulation system the

Proje t Editor (PED)

is used. The simulation ar-

hite ture an be obtained in Figure 5.5.


At the higher level a system template an be found whi h ontains for example a omplete
simulation system of 10 WTs and a PBS (see Figure 5.6).
ments are the

Other important system ele-

hannel and the Init Terminal Addressing element, where the global memory

5.4. Implementation

49

Simulation Architecture:4 @ 1/29/98, 6:19 pm

[ 3-Oct-1998 11:10:24 ]

Init Global memories

Channel Template

Process Channel Delay Template

System Template
Channel Adapter Template

Terminal Template

Layer Management Template

Data Processing Template

Layer Template

Control Unit Template

Figure 5.5:

Simulationsar hitektur [11

initializing takes pla e. The simulation stru ture follows exa tly the ar hite ture planed
in the PED, as it is shown in Figure 5.5 and 5.6.

5.4.2 Terminal stru ture


In the BAHAMA network all wireless terminals own the same fun tionality and so they

WT-Address ontains the mobiles unique ID of


NumGen Minimum and NumGen Maximum are used on the RA han-

have the same stru ture. The parameter


one Byte length.

nel for determining the Random Ba k O for the slotted ALOHA proto ol (see Figure 5.7).

All terminals in lude the omponents listed next, in order to be onform with the ISO/OSILayer model (see Figure 5.7):

 ATM Sour e Layer ! higher layer


 MAC Layer
 Channel State Adapter & Channel Termination ! physi al layer
Every layer has the stru ture shown in the Figure 5.8. For ommuni ation between neighbour layer primitives are ex hanged through two dierent interfa es: The
and the

Data Primitive Interfa e.

Control Primitive

The rst one is used for the inter-layer information ex-

hange and the se ond one for the transport of the a tual data.

The

Management Layer

an a ess all layers and initializes all modules, so it is of great

importan e for the system. At this point every module re eives over the

Interfa e an unique identi ation

number.

Control Primitive

5. BONeS Simulator

50

DQRUMA System mit Base Station FDD with 10 Mobiles

Init Terminal
Addressing

Init

Execute
In Order

[ 15-Nov-1998 11:44:44 ]

2
1

Init

Init
Error Free
Channel w/Distance&Collisions
DL
Error Free
Channel w/Distance&Collisions
UL

Init
DQRUMA
Terminal

DQRUMA
Base Station

DQRUMA
Terminal

Init

P Slot Time UL

Init
DQRUMA
Terminal

DQRUMA
Terminal

P ACK&Xmt_Perm Length
P Time for Ack
P PCI Lenght

Compute
Statistics

P Request Lenght
Init

Init
DQRUMA
Terminal

P # DQRUMA Terminals
DQRUMA
Terminal

P Slot Time
P Xmt Delay
P Mean Data Rate

M Channel State Matrix


M Goto Set Position

Init
DQRUMA
Terminal

M Flow Counter Vector DL


M Receiver State Vector DL

P SCAN Time

Init

P Maximum Queue Size-MAC


DQRUMA
Terminal

M Channel Packet Exchange DL


M Channel Packet Exchange UL
M Flow Counter Vector UL

P NumGen Minimum
P NumGen Maximum
P Bit/sec

DQRUMA
Terminal

DQRUMA
Terminal

M Receiver State Vector UL

P Receive Range

Init

Init

P # WT
P RA Delay
P # Simulation Terminals
E RA to Xmt alarm

Figure 5.6:

System

5.4.3 The MAC Layer


Medium A ess Control)

The MAC (

Layer ontrols the medium a ess for the ommon

transmission hannel. This thesis is fo used on medium a ess matters of the BAHAMA
network.

The MAC layer for the PBS is a bit dierent from the mobile's MAC layer, sin e it is extended in order to in lude the fun tionality for providing the hannel rights to the mobiles.
It ontains three important modules (see Figure 5.9): The

Handler and the MAC Pro essing Unit.


The

Control unit

Control Unit,

the

Data Port

is at rst responsible for the forwarding of the initialization pro ess to

the sub-modules. Further, it supports the routing of in oming and outgoing primitives.

The two

Data Port Handler take on the task of ltering information from in oming MACData Primitives, so

PDUs, as well as en apsulating the out oming MAC-PDUs in form of


that they an be forwarded to the next layer.
The

MAC Pro essing Unit has also a sub-blo k stru ture and ontains the following mod-

ules, that manage all MAC-layer spe i fun tions (see Figure 5.10).

 PDU Pro essing Unit:


The

PDU Pro essing Unit en apsulates or de omposes PDUs.

The header of in oming

PDUs is evaluated and a ording to the data won, the PDUs are used for intra layer
purposes or they are forwarded to the

Data Port Handler

to be routed to another

layer.

 PCI Evaluation Unit:


The

PCI Evaluation Unit extra ts Base Station data from the in oming MAC header
MAC Units for further analysis.

and then forwards them to the blo k

 MAC Unit at Mobile:

5.4. Implementation

51

DQRUMA Terminal OneDLBurst

[ 15-Nov-1998 12:25:39 ]

ATM Poisson
Source Layer
Test
Init

Layer Management
Template

Control

PX

MAC Layer
w/DQRUMA
at Mobile
OneDLBurst

PY
PZ
P SCAN Time
M Goto Set Position
M Receiver State Vector
M Channel State Matrix

LD

Control

M Flow Counter Vector

M Channel Packet Exchange DL


Channel Adapter
w/Position
& Collisions
& Energy Detection
2 TR

M Channel Packet Exchange UL


M WT-Address
P Maximum Queue Size-MAC

Data
Channel

P NumGen Minimum
P NumGen Maximum
P Time for Ack
P RA Delay

WT->Channel
Termination
DL/UL

P WT-Number
P Request Lenght
P Slot Time
P PCI Lenght
P # Simulation Terminals
P Mean Data Rate
P Destination Number

Figure 5.7:

Terminal

This is one of the most important modules in the system, where most of the MAC
proto ol work is a omplished. A ording to the MAC information oming in from
the Base Station, this module transmits the buered ATM ells at the appropriate
time slots.
For modeling the Base Station some modules need to be hanged as it an be seen in
Figure 5.11. These are the following:

 MAC Unit at Base Station:


The

MAC Unit at the Base Station

represents the kernel of the system, sin e it

determines whi h mobile at whi h point will use the shared hannel. The permissions
are given in a Round Robin a ess s heme.

 MAC generate at Base Station:

After determining the mobile's ID whi h will take permission to send an ATM ell over
the next Uplink Slot, the ID is en apsulated in an Downlink MAC header potentially
together with an A knowledge of a su essful Request A ess and a Downlink ATM
ell.

5. BONeS Simulator

52

Layer Template

[ 3-Oct-1998 12:02:10 ]
HIGHER LAYER INTERFACE
Higher Control

Control
Unit Template

Data Processing
Template

Seperate
Ports

M Block Address

One_Way

Control

Higher Data

Lower Control

Lower Data

LOWER LAYER INTERFACE

Figure 5.8:
MAC Layer w/DQRUMA at Mobile OneDLBurst

Layer-Template

[ 15-Nov-1998 13:05:08 ]
Higher Data

Higher Control

M MAC Layer Address


P Maximum Queue Size-MAC
Port

P NumGen Minimum
P NumGen Maximum

Data
MAC Processing
w/DQRUMA at
Mobile OneDLBurst

P Request Lenght
P RA Delay

Data Port
Handler
Port

P Slot Time

PDU

Data
PHY

PDU

CSMA MAC
Control
Unit

One_Way

P WT Adress
P Destination Number
M Source Address

Lower Data

Lower Control

Figure 5.9:

One_Way

Control

Data Port
Handler
PDU

P PCI Lenght

PDU

P Time for Ack

MAC-Layer

5.4. Implementation

53

MAC Processing w/DQRUMA at Mobile OneDLBurst


Control Data

Edit Control
Data in DQRUMA
uplink

SDU In/Out

Data

CP

[ 15-Nov-1998 13:31:18 ]

MAC Generate
at Mobile
OneDLBurst

SDU-CP
Buffer

ON/OFF
Switch

In
CP

CP for
Request
Access

Out

MAC PDU Processing


Unit w/DQRUMA
at Mobile
OneDLBurst

PHY CP
MAC PCI Evaluation
at Mobile
OneDLBurst

CP Out

Figure 5.10:

P PCI Lenght

P Maximum Queue Size

P WT Adress

P Initial Queue State

P RA Delay

P Destination Number

P Slot Time

P Request Lenght

M Source Address

P NumGen Minimum

P Time for Ack

P NumGen Maximum

MAC-

MAC Processing w/DQRUMA at Base Station OneDLBurst

PDU In/Out

MAC Pro essing Unit at mobile

[ 15-Nov-1998 13:34:53 ]

Control Data

SDU

SDU-CP Buffer
at Base Station
OneDLBurst

MAC PCI Evaluation


at Base Station
OneDLBurst

Medium Access
Strategy
OneDLBurst

MAC Generate
at Base Station
OneDLBurst

CP for Ack&Xmt_Perm
at BS OneDLBurst

MAC PDU Processing


Unit w/DQRUMA
at Base Station
OneDLBurst

Out

Data

In
CP

PHY CP
CP Out

Edit Control
Data in DQRUMA
OneDLBurst

CP
PDU

P Maximum Queue Size

M Source Address

M SDU Memory

E Slot Timer

P Initial Queue State

P PCI Lenght

M Ack arrival Memory

M INDEX

P Destination Number

P #WT

M Request # in Table

M Request Table

P Slot Time

Figure 5.11:

MAC-

MAC Pro essing Unit at Base Station

54

5. BONeS Simulator

CHAPTER 6

Simulation s enarios
6.1

DQRUMA slot length

The length of the DQRUMA slot strongly depends on the transmission rate of the physi al
(PHY) hannel (CH). In the following, the slot expansion between two data rates is being
illustrated. Some of the systems were simulated for PHY data rates of 2 and 10 Mb/s, respe tively. Sin e at higher PHY data rates the slot be omes smaller, the hannel apa ity
will be in reased. This raises the proto ols throughput drasti ally.
As a result, the length of the single slot hannels does not remain the same. This is depi ted in table 6.1 and 6.2. A burst always in ludes a

guard interval

and

Rise/Fall time.

As shown in table 4.2 in hapter 4 the total guard time for a single burst is a onstant
with a value of 1.96

s.

Additionally, the PGBK hannel is modelled with 8 bit in ontrast

to the original DQRUMA, whi h proposes 1 bit.

ttrans ) of n bits for a data rate of Rdata , it is:


n
(6.1)
ttrans =
Rdata

In order to al ulate the transmission time (

The variable

n ontains the length values of the DQRUMA hannels from Table 4.1.
2 Mb/s

Uplink

10 Mb/s

Burst Duration [ s

Burst Duration [ s

5.96

2.76

PGBK + Pa ket_Xmt

217.96

45.16

Total slot length

223.92

47.92

RA

Table 6.1:

Uplink Bursts

2 Mb/s
Downlink

10 Mb/s

Burst Duration [ s

Burst Duration [ s

ACK + Xmt_Perm + Pa ket_Xmt

221.96

45.96

Total slot length

221.96

45.96

Table 6.2:

Downlink Burst

As it an be obtained from Tables 6.1 and 6.2 the slot for the DL is smaller than for the
UL. This results from the proto ols burst stru ture (see hapter 4). Sin e the DQRUMA
prin iple is based on a time slotted stru ture, where both UL and DL are divided into slots
of the same size, the UL slot is hosen to be the DQRUMA slot.

burst overhead.
guard time and the rise/fall time of a burst. As the data rate rises the
transmitted faster, while the time for the burst overhead remains onstant. This

Another important feature of higher data rates is the growing impa t of the
This omprises the
bits are

6. Simulation s enarios

56

ee t an be seen in the Table 6.1. Although the data rate be omes ve times higher, a
slot is transmitted only 4.67 times faster (see Equation 6.2).

Slot size 2Mbps


Slot size 10Mbps
6.2

:92s
= 223
47:92s ' 4:67

(6.2)

DQRUMA Simulations at low data rates (2 Mb/s)

At the beginning, the DQRUMA proto ol was dimensioned for a data rate of 2 Mb/s. Two
systems with 5 and 10 wireless terminals were simulated (see Figure 6.1). Poisson sour es
have been used. Ea h terminal has a buer with a apa ity of 100 ATM ells. No
and

Rise/Fall

time was onsidered here.

With a

load adjustment fa tor

Guard

8 iterations be-

tween 20% and 80% of the hannels apa ity were simulated. The proto ol parameters an
be found summarized in Table 6.3.
The

Random Ba k-o

parameter for the slotted ALOHA proto ol of the RA hannel is a

random number between 1 and 10 or 20.

This is the number of slots that the terminal

must wait after a ollision, until it transmits its RA again. The impa t of this proto ol
parameter is very strong and an ause long delays, if it turns to be too small or too large
(see Se tion 4.5.1).

The aspe t of this parameter will be fo used in another simulation

later in this hapter (see Se tion 6.2.1 and 6.3.1.1).

The same simulations with the same parameters were also done for the

ideal system. This


request

is a system where the request a ess is automati ally known to the PBS so that the

table is at on e updated without the need of the RA hannel.

between the ideal and the normal

The dieren es dete ted then

systems illustrate the impa t of the slotted ALOHA

at the RA hannel to the DQRUMA proto ol.


DQRUMA System mit Base Station FDD with 5 Mobiles

Init Terminal
Addressing

Init

Execute
In Order

[ 25-Nov-1998 17:30:28 ]

2
1

Init

Init
Error Free
Channel w/Distance&Collisions
DL

DQRUMA
Terminal
4
Init

Init

Error Free
Channel w/Distance&Collisions
UL

Init

DQRUMA
Terminal
2

Init

DQRUMA
Terminal
3

DQRUMA
Terminal
1

DQRUMA
Terminal
5

Compute
Statistics

M Channel Packet Exchange UL

P Slot Time UL

P # DQRUMA Terminals

M Flow Counter Vector UL

P ACK&Xmt_Perm Length

P Slot Time

M Receiver State Vector UL

P Time for Ack

P Xmt Delay

M Channel State Matrix

P PCI Lenght

P Mean Data Rate

M Goto Set Position

P Request Lenght

P SCAN Time

M Flow Counter Vector DL

P RA Delay

P Maximum Queue Size-MAC

M Receiver State Vector DL

P Bit/sec

M Channel Packet Exchange DL

P Receive Range

P NumGen Maximum

P # WT

E RA to Xmt alarm

Figure 6.1:

DQRUMA
Base Station

P NumGen Minimum

Simulation example with 5 terminals

For these simulations basi performan e metri s were omputed and displayed, su h as
mean delays, omplementary distribution fun tions to hara terize the proto ol, as well as
number of ollisions, in order to measure the impa t of the RA hannel allo ation to the
total proto ol performan e.

6.2. DQRUMA Simulations at low data rates (2

Mb/s)

Parameter

Value

Terminals

5/10

Sour e Type

Poisson
2 Mb/s

Data Rate
Buer

100 ATM ells


10/20 slots

Max. Random BO
Load

20%-80%
0.22

Slot Time

Table 6.3:

57

ms

5/10 Terminal original DQRUMA Simulation

The mean ell delay on the UL an be seen in Figure 6.2(a) as a fun tion of the total system
load. The load is equally shared over the terminals in ea h of the systems simulated (5
and 10 WTs). The mean delay displays the mean time from the point one ell is generated
at a terminal sour e until the time it rea hes the Base Station.

First, the big impa t of the RBO parameter on the delay is proved. The dieren e between
the delays for a RBO value 10 and 20 is app. 0.5

s.

It seems, that a RBO of 20 is too

large for both systems.

ATM cell delay Uplink

[ 16Sep1998 11:38:40 ]

max ATM cell delay Uplink

[ 16-Sep-1998 11:38:40 ]

max ATM cell delay Uplink

5.0

0.050

4.5

0.045

4.0

0.040

3.5
3.0
2.5
2.0
*

1.5
*

1.0
*

0.035
0.030
0.025
0.020
*

0.015
0.010

Maximum Delay [s]

Mean Delay [ms]

ATM cell delay Uplink

0.5

0.005

0.0

0.000

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.2

0.3

0.4

0.5

Load
5 WT 110 (solid)
5 WT 120 (solid)
* 5 WT ideal (solid)

Figure 6.2:

0.6

0.7

0.8

Load
10 WT 110 (dashed)
10 WT 120 (dashed)
10 WT ideal (dashed)

(a)

5 WT 1-10 (solid)
5 WT 1-20 (solid)

10 WT 1-10 (dashed)
10 WT 1-20 (dashed)

* 5 WT ideal(solid)

10 WT ideal (dashed)

(b)

Mean and maximum delays for a 2 Mb/s system

For small RBOs the ollisions in the RA hannel will grow rapidly, be ause after a ollision has happened the RA from the terminals whi h suered the ollision will need to
be retransmitted within a small window (number of onse utive Request A ess CHannels
(RA_CH)), eventually in ontention with other terminals whi h send their RAs with a
high ollision possibility.

6. Simulation s enarios

58

Choosing large RBO values an also have bad results for the mean delays of the ATM ells.
This an be explained onsidering the delay some ells an experien e if they are transmitted at the nal RA_CH of the Ba k O window. If a ollision happens then, that means
that a new RBO interval must be determined, whi h again might be too long, so that the
total transmission time is extended. This parameter aspe t (large RBO) an be obtained
from Figure 6.2(b). The proto ols worst ase for the ell delay is depi ted here over the
total load. This diagram shows long delays that an be aused by onse utive ollisions. It
an be seen, that a RBO parameter value of 20 auses a maximum delay growth in the diagrams of app. 10

s.

The RBO will be optimized in further simulations later in this hapter.

An important feature shown in the diagrams in Figure 6.2 is the growth of delays as the
number of system terminals in reases. For a greater number of terminals these move more
often in the random a ess state, sin e they produ e a lower load and their buers are
more often empty. So, there is a greater number of pa kets that arrive in empty buers.
Su h pa kets must request a

Xmt_Perm

over the RA_CH. Pa kets that arrive next an

be transmitted with the PGBK option after the terminal has transmitted its RA su essfully. More RAs also means more RA_CH ollisions, that in rease the delay of the ell
transmissions.

Request to Xmt Delay vs Load

[ 16Dec1998 10:59:23 ]

Max Request to Xmt Delay vs Load

Request to Xmt Delay vs Load

[ 16-Dec-1998 10:59:23 ]

Max Request to Xmt Delay vs Load

3.0

0.050
0.045
0.040

Maximum Delay [s]

Mean Delay [ms]

2.5

2.0

1.5

1.0

0.035
0.030
0.025
0.020
0.015
0.010

0.5
*

0.2

0.3

0.4

0.5

0.6

0.7

0.005

0.0

0.000

0.8

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Load
5 WT 110 (solid)
5 WT 120 (solid)
* 5 WT ideal (solid)

10 WT 110 (dashed)
10 WT 120 (dashed)
10 WT ideal (dashed)

(a)

Figure 6.3:

Load
5 WT 1-10 (solid)
5 WT 1-20 (solid)
* 5 WT ideal (solid)

10 WT 1-10 (dashed)
10 WT 1-20 (dashed)
10 WT ideal (dashed)

(b)

Mean and maximum delay between a rst pa ket and

Xmt_Perm

As expe ted the delays grow in both diagrams with a higher load. This happens be ause
at a higher load more pa kets arrive at a mobile's buers and queue. Con luding it an
be said, that the higher the load gets the qui ker the buering delay grows. At a lower

 20%) the diagrams in 6.2(a) represent the s heduling delay from the Round Robin

load (

algorithm, sin e it an be supposed that the buers are almost empty and the ollision
rate is low.
The urves for the

ideal

systems an also be found on the bottom of the Figures 6.2(a)

6.2. DQRUMA Simulations at low data rates (2

Mb/s)

59

and 6.2(b). Sin e no ollisions o ur here all the mean delay is aused by the s heduling
algorithm.

Figures 6.3(a) and 6.3(b) exhibit the mean and maximum delay from the time a ell arrives to an empty buer until the time the

Xmt_Perm for this ell is transmitted for two

systems (5 and 10 WT). The maxBO parameter takes the values 10 and 20 for both systems. Additionaly, the ideal 5 and 10 WT systems are simulated, in order to distinguish
between the

slotted ALOHA inuen es on the RA_CH and the s heduling delays from the

RR s heduler at the base station. The urve's form is detailed in the following.

% Collisions in RA Channel vs Load

[ 21-Aug-1998 20:48:34 ]

% Collisions in RA Channel vs Load


0.35

Collision Rate [%]

0.30

0.25

0.20

0.15

0.10

0.05

0.00

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Load
5 WT 1-10 (solid)
5 WT 1-20 (solid)

Figure 6.4:

10 WT 1-10 (dashed)
10 WT 1-20 (dashed)

Collision rate on the RA Channel

For a low system load the diagrams in Figure 6.3(a) only ontain the s heduling delay.
With growing load, this delay will be rising until it rea hes a maximum, whi h is found to
be at a load of 60% for 5 WTs and 70% for 10 WTs. After this point the mean RA-toXmt_Perm delays fall again. This be omes lear, if we onsider the fun tionality of the
PGBK bit. The PGBK bit is deployed for requesting Xmt_Perm for ells that arrive in
non-empty buers. This is more likely to happen for higher loads, e.g.

60-70%. Then the

number of terminals in the system that request Xmt_Perm via PGBK bit rises, so that
the the RA_CH is used less frequently.

In the Figures 6.3(a), 6.3(b) the

ideal system serves as referen e showing the impa t of the

ollisions happening in the RA_CH. In 6.3(a) both ideal diagrams start from the same
point and begin to drift apart for higher load values. This is be ause the

request table

is

strongly o upied for higher values, so that the Xmt_Reqs in the table must wait longer
until the s heduling algorithm serves them.

Figure 6.3(b) presents the maximum RA-to-Xmt_Perm times. The diagrams on the top of
the gure an be explained in a ordan e to the urves for the mean ell delays in 6.2(a).
The most interesting results are those for the ideal systems with 5 and 10 WT. The
maximum time the s heduler needs to serve a terminal is here shown. Sin e the systems

6. Simulation s enarios

60

Request Table

ontain 5 and 10 WTs, the

in ludes 5 and 10 elds. Hen e, the maximum

times happen as the table is full, so that the s heduler must wait 5/10 slots until is serves
the last of the elds. Up to a load of 80% the maximum delay for 5 and 10 WTs is almost
identi al.

Thus, the s heduling strategy has only minor impa t on the dieren e of the

maximum delays for the systems using the RA_CH.


Figure 6.4 presents the ollision rate on the RA hannel as fun tion of the system load.
These urves have a similar shape as those in Figure 6.3(b) for the same reason as depi ted
there. For a low system load the RA_CH utilization is low. Then it rises with a growing
load until it rea hes a maximum. Beyond this point more and more ells transmit a positive set PGBK bit be ause the system buer o upan y rises and the RA_CH utilization
fa tor falls.

PGBK/Request vs Load

Mean Buffer Occupancy

[ 27-Aug-1998 12:06:05 ]

[ 19-Aug-1998 14:58:03 ]

Mean Buffer Occupancy

PGBK to all requests


2.0

0.8

1.8

Mean Buffer Occupancy

0.7

Request Rate

0.6

0.5

0.4

0.3

0.2

0.1

1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0

0.0

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Load

10 WT 1-10
10 WT 1-20
10 WT ideal

0.2

0.3

0.4

Figure 6.5:

0.6

0.7

0.8

Load

10 WT 1-10 UL
10 WT 1-20 UL
10 WT ideal

(a)

0.5

(b)

PGBK-to-RA ratio and Mean buer o upan y

Figure 6.5(a) depi ts the PGBK-to-RA ratio for a 10 WT system. The maxRBO parameter takes the values 10 and 20 and the ideal 10 WT system is added as a referen e. As the
total system load in reases the PGBK-to-RA ratio rises. This be omes lear if the mean
buer o upan y is onsidered (see Figure 6.5(b)). With growing system load the buers
are getting stued and more ells starting ontaining positive PGBK bit.

On the other

hand the RA Channel utilization drops be ause the terminals transmit most of their ells
over the PGBK CH without the possibility of suering a ollision.
The DL ell delay is dis ussed in Figure 6.6(a).

All ells destined for the mobiles are

transmitted from the PBS via downlink. Sin e a FDD system is proposed, there are 2 Mb/s
for the DL transmission.
following.

The DL slot assignment me hanism model is des ribed in the

There is a Round Robin s heduling instan e that he ks all PBS DL buers,

whether or not there are DL pa kets to be send. These are s heduled then from the RR
s heduler. Generally it an be said, that this is an

M/D/1 model:

6.2. DQRUMA Simulations at low data rates (2

ATM cell delay Downlink

Mb/s)

61

Complementaty Distribution Function

[ 19-Aug-1998 14:58:03 ]

Downlink Delay
1.6

0.1

1.2
*

1.0
0.8
*

CDF

Downlink Delay [ms]

1.4

0.6
*

0.4
*

0.2

0.3

0.01
10WT 1-5
10WT ideal

0.2
0.001

0.0

0.4

0.5

0.6

0.7

10WT 1-20

0.8
10WT 1-10

Load
5 WT 1-10
5 WT 1-20

10 WT 1-10
10 WT 1-20
10 WT ideal

* 5 WT ideal

0.0001
0.005

0.01

0.015

(a)

Figure 6.6:

0.02
Delay [sec]

0.025

0.03

0.035

(b)

Downlink ell delay and Complementary Distribution Fun tion(CDF)

 Poisson Sour es ! Markov arrival pro ess


 Constant pa ket length (ATM ells) ! Deterministi server pro ess
 A single server
M/D/1 model. Using the approximative equation for the
expe ted waiting time for the DL ells (DL ), it is
There is no exa t solution for an

DL =
In equation 6.3

 is

1
2    (1

the servi ing rate and

 the

)

(6.3)

total system load. The simulation result

for the DL ell delay omes very lose to the analyti al model.

The dieren e an be

explained if it is onsidered that the analyti al model oers just an approximation and the
simulation results have a larger onden e interval for higher loads.

The UL Complementary Distribution Fun tion(CDF) of a 10 WT system for a system load


of 70% an be found in Figure 6.6(b). The maxRBO parameter is variated in 3 steps: 5, 10
and 20. The relative positioning in the diagram remains the same as were in Figure 6.2(a).
There are signi ant dieren es between the

ideal System and

the other ones. Generally,

it an be said that the three normal systems experien e similar delays for most of the
ells (see top left area in Figure 6.6(b)).

slotted ALOHA data


Es aloha , an be obtained from

The utilization of the RA_CH an be ompared with the theoreti al


(see Figure 6.7). The theoreti al slotted ALOHA e ien y,
the equation

Es aloha = Load  e Load

(6.4)

0.04

6. Simulation s enarios

62

Succ. Req vs All Req/#slots

[ 27-Aug-1998 12:06:05 ]

S-ALOHA throughput

[ 27-Aug-1998 12:06:05 ]

S-ALOHA throughput

1.0

1.0

0.9

0.9

0.8
0.7
0.6
0.5
0.4
0.3

*
*

0.2

Request / all requests

Request / Number of Slots

Succ. Req vs All Req/#slots

0.8
*

0.7
*

0.6
*

0.5

*
*
**

0.4
0.3
0.2

0.1

0.1

0.0

0.0

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.2

0.3

All requests / # slots


10 WT 1-10 (dash)

0.5

0.6

0.7

0.8

All requests / # slots

* S-ALOHA (solid)

* S-ALOHA throughput (solid)

(a)

Figure 6.7:

0.4

10 WT 1-10 throughput (dash)

(b)

S-ALOHA proto ol performan e in the RA_CH

Equation 6.4 produ es almost the same result as the simulation. The same is shown for
the

S-ALOHA throughput, Ds aloha .

This is equal

Ds aloha = e

# requests
# slots

(6.5)

6.2.1 Simulations with a variable number of terminals


The obje tive of this simulation is to measure the mean delay as a fun tion of the number
of terminals in the system.

The

random Ba k-o

(RBO) is also being varied.

The val-

ues for the RBO are 10 and 20. The number of terminals varies between 4 and 20. The
simulation is performed for a hannel data rate at 2 Mb/s. Again, no guard intervals are
onsidered in the slot. The load is set onstant to 3.5% of the a tual hannel apa ity per
a tive terminal. An

ideal system

was also simulated with the aforementioned parameters

(see Table 6.4).

Parameter
Terminals
Sour e Type
Data Rate
Buer
Max. Random BO

Load/Terminal
Slot Time

Table 6.4:

Value
4-20
Poisson
2 Mb/s
100 ATM ells
10/20 slots
3.5%
0.22

ms

Parameters for a system with variable number of terminals

6.2. DQRUMA Simulations at low data rates (2

UL Delay vs # 20 Terminals

Mb/s)

63

[ 17-Dec-1998 11:48:21 ]

UL Delay vs # 20 Terminals
0.016
0.014

Mean Delay [s]

0.012
0.010
0.008
0.006
0.004
0.002
*

0.000

5.

10.

15.

20.

# Simulation Terminals
varWT RBO 1-10 (solid)
var WT RBO 1-20 (dash)

Figure 6.8:

* varWT ideal (dot-dash)

Variation of the terminal number

In Figure 6.8 the results are shown for a s enario where the number of terminals is being
varied. The maxRBO parameter is also variable. Its value alters between 10 and 20. For
an

ideal

system the variation of the number of terminals in the present s enario has no

impa t on the mean ell delays.

This means that the s heduling is independent of the

terminal number. It an also be obtained, that the maxRBO value of 10 results in better
delays for systems up to 12 WTs. For systems with more than 12 WTs a greater maxRBO
value should be hosen. The maxRBO parameter will be optimized in a further simulation
for spe i s enarios in luding 10, 20 and 50 wireless terminals.

6.2.2 System with dierent load per terminal


Parameter
Terminals
Sour e Type
Data Rate

Value
5
asymmetri Poisson
2 Mb/s
100 ATM ells

Buer
Max. Random BO

10/20 slots

Total Load

30%-70% of the CH ap.

Load/Terminal

WT1: 1/2-WT2: 1/4-...- WT5: 1/32

Guard Time
Rise/Fall Time
Slot Time

Table 6.5:

0.22

ms

5 Terminal Simulation with asymmetri onne tions

6. Simulation s enarios

64

5 WT Weighted Load/Terminal

[ 16-Dec-1998 17:07:59 ]

5 WT Weighted Load/Terminal
4.0
3.5

Mean Delay [ms]

3.0
2.5
2.0
1.5

*
*

0.3

0.4

0.5

0.6

1.0
0.5
0.0

0.7

Load

ideal Terminal 1 - solid


ideal Terminal 5 - solid

*
*

Terminal 1 (20) - dotted


Terminal 5 (20) - dotted

Terminal 1 (10) - dashed


Terminal 5 (10) -dashed

Figure 6.9:

Weighted load simulation

This s enario omprises 5 terminals and a PBS (see Figure 6.1).

In this simulation the

terminal sour es (Poisson) produ e asymmetri load, in ontrast to the previous ones. In
this ontext, asymmetri means that the terminals do not generate the same load.

The

load distribution is being hosen, so that the rst terminal generates 1/2 of the total load,
the se ond one 1/4 , the third one 1/8, et . A load adjustment fa tor varies the total load
between the boundaries 30% and 70% of the hannels apa ity.

Round Robin)

Sin e the s heduling algorithm at the base station (

does not support a

weighted option, e.g. to serve the terminals with a higher load more often rather these
with a lower load, we expe t to see a proto ols insu ien y for a weighted s enario with
dierent servi e lasses. All simulation parameters an be retained from the Table 6.5.

Figure 6.9 presents the impa t of dierent loads on systems with


ing poli y.

Ea h system is here represented with two urves.

Round Robin

s hedul-

One for a terminal whi h

generates 1/2 (Terminal 1) and a terminal whi h produ es 1/32 (Terminal 5) of the total
system load. The mean ell delays for all systems with a low terminal load fa tor do not
hange as the load grows. That is be ause the ell generation rate is very low, so that the
RR s heduler manages to serve these terminals in time before ells starting sta k in the
terminals buers. This hanges for a high terminal load fa tor.

Su h terminals an maintain onne tions whi h represent real time servi es. Su h servi es
might need to be served more often than normal onne tions. Sin e RR serves every terminal with one slot for a omplete table y le, the terminals with high rate onne tions
will not have the opportunity for their ells to be served properly. These are get sta ked,
so that the buering time dominates the mean ell delays.

6.3. Simulations at high data rates (10

Mb/s)

65

Simulations at high data rates (10 Mb/s)

6.3

The simulation series presented next is performed for a hannel rate at 10 Mb/s for ea h
dire tion, DL and UL. Radio propagation and hardware spe i parameters are mentioned
and expand the proto ols slot length.

This omes very lose to real proto ol working

onditions, so that we an extra t useful results. The simulation runs also in lude more
terminals with RBO parameter being optimized. In addition, two proto ol extensions are
introdu ed.

The

DQRUMA polling

proto ol (see Se tion 4.4.2), eliminates the ollisions

on the RA hannel by polling the terminals to send their requests, if needed, in a

Robin

manner. The

DQRUMA with ollision resolution

Round

supports the ollision resolution

s heme that is presented in the Se tion 4.4.1.

6.3.1 Original DQRUMA simulations


In this se tion simple DQRUMA s enarios are presented for a number of 10, 20 and 50
terminals. The slot stru ture and length an be obtained from the Tables 6.1 and 6.2 in
se tion 6.1. An example of a simulation environment with 20 terminals an be found in
Figure 6.10.

DQRUMA System mit Base Station FDD with 20 Mobiles OneDLBurst

Init Terminal
Addressing

Init

Execute
In Order

Error Free
Channel w/Distance&Collisions
DL

2
1

Init

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

Error Free
Channel w/Distance&Collisions
UL

Compute
Statistics
OneDLBurst

[ 26-Nov-1998 11:49:41 ]

DQRUMA Base
Station
OneDLBurst
P Destination 0
P Destination 1
P Destination 2
P Destination 3
P Destination 4
P Destination 5
P Destination 6
P Destination 7
P Destination 8
P Destination 9
P Destination 10

M Channel State Matrix

Init

DQRUMA Terminal
OneDLBurst

Init

DQRUMA Terminal
OneDLBurst

M Goto Set Position

P Destination 14

M Receiver State Vector DL


M Channel Packet Exchange UL
M Flow Counter Vector UL
M Receiver State Vector UL

P Destination 12
P Destination 13

M Flow Counter Vector DL


M Channel Packet Exchange DL

P Destination 11

P Destination 15
P PCI Lenght

P NumGen Minimum

P # Simulation Terminals

P Bit/sec

P Request Lenght

P NumGen Maximum

P Mean Data Rate

P Receive Range

P # DQRUMA Terminals

P Time for Ack

P SCAN Time

P # WT

P Slot Time

P PCI Lenght Downlink

P Maximum Queue Size-MAC

P RA Delay

P Destination 16
P Destination 17
P Destination 18
P Destination 19
P Destination 20

Figure 6.10:

20 WT simulation

After an optimization of the RBO parameter in Se tion 6.3.1.1, the s enarios have been
simulated for the best values found. The optimized RBO values dier for s enarios with
dierent numbers of terminals.

Guard and Rise/fall times were added to the original slot,

in order to be ome reality- lose results. These values are explained in Se tion 4.3. A simulation parameter overview is given in Table 6.6. Larger buers were pla ed on terminals
and Base station, whi h an now take up to 256 ATM ells.

6.3.1.1 Impa t of the Ba k O parameter


This simulation will estimate the impa t of the RBO parameter on the DQRUMA proto ol
for three systems ontaining 10, 20 and 50 WTs. The system load is said to be 70% of the
CH apa ity. This spe i load value is hosen, be ause it was found, that most ollisions
happen at this point.

The RBO parameter was altered within dierent boundaries for

6. Simulation s enarios

66

Parameter

Value

Terminals

10, 20 and 50

Sour e Type

symmetri Poisson
10 Mb/s

Data Rate
Buer

256 ATM ells

Max. Random BO

10WT: 7 - 20WT: 15 - 50WT: 45

Total Load

20%-88% of the CH ap.

Guard Time
Rise/Fall Time

0.66

Total Slot Time

Table 6.6:

s

s
s

ea h 0.65
47.92

Simple DQRUMA simulation parameters for 10, 20 and 50 WTs

ea h system (see Table 6.7). The rest of the simulation parameters used remain unmodied (see Table 6.6).

Terminals

RBO Values

10

5-25

20

5-25

50

30-55

Table 6.7:

RBO Parameter Variation

The results from this simulation an be obtained in Figure 6.11.

Choosing too small

maxRBO values an drive the ell delays higher. The same happens if the parameter is
set to a high value. The RBO parameter value that auses a minimum delay was hosen
as an optimum value for further simulations.

6.3.2 DQRUMA Polling simulations


The simulations des ribed in this Se tion investigate the behaviour of the

Polling

proto ol (see Se tion 4.4.2).

DQRUMA

The parameters used here are exa tly the same as

those used for the normal-DQRUMA (see Se tion 6.3.1), even if the slot stru ture does
not remains the same. That is be ause the additional length on the DL for the required

Polling CH, does not ex eed the dieren e between UL and DL be ause of the se ond burst

on the UL (see Se tion 6.1):

polling CH length
< Slot length UL
CH data rate

Slot length UL

(6.6)

The polling system diagrams together with the ones for the normal DQRUMA an be
found on Figure 6.12. As also mentioned in Figure 6.2(a), the delays go higher if the number of terminals in the system in reases. This is valid for both DQRUMA implementations,
polling and normal DQRUMA.
In ase of the polling systems, the mean delay remains onstant even for high load val-

 80%) in ontrast to the normal

ues (

DQRUMA delays, whi h rise onstantly until

the system overow at app. 90% of the system load.

Figure 6.12 shows that a random

a ess s heme is better than polling for small system load.

After a spe i load value,

the terminal polling oers signi ant better results sin e no ollision resolution is needed.

6.3. Simulations at high data rates (10

10Mbps varbackoff 10-20-50 WT

Mb/s)

67

[ 16-Dec-1998 19:19:23 ]

variation of the maxRBO value


4.0
3.5

Mean Delay [ms]

3.0
2.5
2.0
1.5
1.0
0.5
0.0

0.

10.

20.

30.

40.

50.

60.

maxRBO
50wt var backoff(solid)
20 WT var backoff(dashed)
10 WT var backoff (dotted)

Figure 6.11:

Variation of the RBO parameter

The point from whi h polling starts to be better an be obtained in 6.12. It is the point
where the diagrams for the

polling and the normal DQRUMA systems ome together.

This

onjun tion point is not at the same pla e for systems with dierent number of terminals,
but moves towards the top right orner of the diagram as the number of terminals grow.
Figure 6.13 shows the omplementary distribution fun tion for the 10 and 20 WT systems
for both simple and polling DQRUMA. The load adjustment fa tor for all systems was
hosen to be 0.8. It an be seen that the polling DQRUMA oers mu h better results for
both systems.

6.3.2.1 Polling simulation with weighted load


This is a simulation onsisting of 20 WTs and a PBS with asymmetri terminal load. The
total load varies between 20% and 80% of the CH apa ity. The load ontribution of the
rst ve terminals is the same as des ribed in the simulation in Se tion 6.2.2 (see Table 6.8).
Other simulation parameters do not alter referable to Table 6.6.

Terminal

Contribution to System Load [%

1/2

1/4

3
4

1/8
1/16

1/32

6-20

1/64

Table 6.8:

DQRUMA polling with dierent load per terminal

6. Simulation s enarios

68

ATM cell delay Polling/DQRUMA

[ 22-Nov-1998 20:51:31 ]

ATM cell delay Polling/DQRUMA


3.5

Mean Delay [ms]

3.0
2.5
2.0
1.5
1.0

*
*

0.5
0.0

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Load
DQRUMA 50WT (solid)
DQRUMA 20WT (solid)
DQRUMA 10WT (solid)

Polling 50WT (dash)


Polling 20WT (dash)
Polling 10WT (dash)
* Downlink (dotted)

Figure 6.12:

Mean ell delays for simple DQRUMA and DQRUMA polling

6.3. Simulations at high data rates (10

Mb/s)

69

Complementary Distribution Function - Load 0.8


1

CDF

0.1

0.01

20WT DQRUMA

10WT DQRUMA
0.001

10WT DL

10WT Polling

20WT Polling
0.0001
0.002

Figure 6.13:

0.004

0.006
Delay [sec]

0.008

0.01

0.012

CDF for simple DQRUMA and DQRUMA polling

Figure 6.14 depi ts the simulation results for the DQRUMA polling system for a weighted
sour es s enario. The left diagram presents the mean delays for four terminals. As it was
for the normal DQRUMA weighted s enario the terminals with the higher ontribution
to the total system load experien e longer delays for a system load over 55%.

The situation hanges for lower load adjustment fa tor values. The delays remain onstant
and are getting longer the smallest the ontribution to the system load is. This an be
explained with the RR polling s heme at the PBS (see se tion 4.4.2). The polling me hanism polls the mobiles, whi h have empty elds in the

Request Table.

The mobiles, whi h

have greater ontribution to the system load will have more often their RT elds full than
the other ones. This shows the utilization of the terminal buers in Figure 6.15. So, if a
request from a terminal with a low load ontribution fa tor arrives, then it is more likely
to wait until the s heduler serves it. The dieren e in the delays between WT1 and WT6
for the range of 0.2-0.55 of the system load is 2 Slots.

In the right diagram the delays for the

rst ells in an empty buer an be obtained.

All

terminals but the one whi h produ es 1/64 of the system load have similar RA-to-Xmt
delays. For WT 6 the delay time for a ell that arrives in an empty buer is 12.5 Slots.
For a higher load the delays for rst ells grow higher, be ause of the larger number of
requests in the RT.

6.3.3 Simulations with Collision Resolution


In the following simulations the ollision resolution algorithm will be investigated as proposed in hapter 4. All system spe i parameters remain the same as depi ted in Table
6.6.

From these simulations (10, 20 and 50 WTs) we expe t an important boost of the

6. Simulation s enarios

70

20 WT 10Mbps poll WL

RA-to-Xmt delays weighted polling

[ 21-Nov-1998 21:54:42 ]

[ 21-Nov-1998 21:54:42 ]

weighted load RA-to-Xmt delay

1.6

1.6

1.4

1.4

1.2

1.2

Mean delay [ms]

Mean delay [ms]

weighted load mean delay

1.0
0.8
0.6

1.0
0.8
0.6

0.4

0.4

0.2

0.2
0.0

0.0

0.2

0.3

0.4

0.5

0.6

0.7

0.2

0.3

0.4

0.5

0.6

0.7

delay1(dash)
delay1(dash)

Load

Load

delay3(solid)

delay3(solid)
delay5(dotted)

delay5(dotted)

delay8(dot-dash)

delay8(dot-dash)

(a)

Figure 6.14:

(b)

Mean delays for a 20 WT system with dierent load per terminal

proto ols performan e, e.g. signi ant fall of the mean delays.

This is more likely to

happen as the number of terminals grows, be ause of the on luding raising of the ollisions on the RA hannel. In this ontext, a very important issue is that of determining the
slot at whi h the ollision resolution will happen. For this, two methods were implemented.

The rst method, also named

MAC ollision resolution

(MAC_CR) uses MAC layer in-

The se ond one is named PHY ollision


resolution (PHY_CR) and needs PHY layer information in order to make the ollision resolution happen. Additionally, the s heduler initiates for both CR algorithms the Multi_CH
mode in ase of an empty Request Table. Determining the time point at wit h a CR should
formation to determine the relevant time slot.

happen is very important. The sele tion of the wrong slot will delay the proto ol a tion
wasting valuable time.

MAC ollision resolution:

This rst type of the CR me hanism ex lusively uses informa-

tion available at the MAC layer. The RA and the PGBK CH are onstantly he ked
to nd whether they ontain positive Xmt requests. If the proto ol dete ts a spe i
number of slots, parameter

Nma , that do not en lose a RA or a PGBK bit set, then

the CR is turned on.

PHY ollision resolution:

Another possibility of making a CR is to gain PHY layer in-

formation about the number of onse utive ollisions happened in the RA hannel
(parameter

Nphy ).

This method supposes an extra intra-layer ommuni ation (MAC-

PHY). Eventually it is more attra tive to use the MAC_CR, sin e all additional
fun tionality for the CR an be pla ed in the same layer.
Figure 6.16 shows the advantages a ollision resolution algorithm an oer.

The gure

ontains the mean delays for the three systems with CR (10, 20 and 50 WTs) as well as the
mean delays of the original DQRUMA simulations des ribed in Se tion 6.3.1 as a referen e.
These advantages, expressed in the redu tion of the mean delays, are getting bigger as the

6.3. Simulations at high data rates (10

Mb/s)

20 WT weighted polling mean buffer occ

71

[ 21-Nov-1998 21:54:42 ]

20 WT weighted polling mean buffer occ


9.

Mean Buffer Occupancy

8.
7.
6.
5.
4.
3.
2.
1.
0.

0.2

0.3

0.4

0.5

0.6

0.7

WT 1(dash)

Load

WT 3(solid)
WT 5(dotted)
WT 8(dot-dash)

Figure 6.15:

Mean Buer O upan y

number of system terminals rises. As the generated load per terminal falls, the terminals
will use the RA_CH more often to send their Xmt permission. The in reasing utilization
of the RA_CH will rise the ollision propability and the overall proto ol performan e will
drop signi antly.

System

Nma

(MAC_CR)

Nphy

(PHY_CR)

maxRBO

10 WT

20 WT

50 WT

40

Table 6.9:

Parameters for ollision resolution

Using a ollision resolution s heme as proposed in 4.4.1, we oer additional RA CH to


the terminals in order to make their requests. The gain from the usage of this algorithm
rea hes its maximum for a total system load of app. 50% for all systems. For a 50 WT
system the mean delays for the range between 0.3 and 0.5 of the system load are even
better than the delays of a 20 WT system without ollision resolution.

For all systems there is a load of app. 70%, where the extra ollision resolution slots have
a bad inuen e on the system's delays, making them rise. This is the point where the loss
of bandwidth be ause of the CR frames ompensates the positive impa t from the extra
RA slots. From this point on, CR should not be performed.

The hoise of the parameter that makes a ollision resolution happen,

Nma

and

Nphy

for

MAC_CR and PHY_CR respe tively, has proven to be a di ult matter, sin e the systems rea t very sensitive to parameter modi ations. Even onse utive arithmeti al values

6. Simulation s enarios

72

20 WT 10 Mbps Cres cell delay best

[ 16-Dec-1998 18:11:06 ]

Collision Resolution
5.0
4.5
4.0

Mean Delay [ms]

3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.2

0.3

0.4

0.5

0.6

0.7

0.8

Load
10 WT (solid)
10 WT MAC CR(dash)
10WT 3 col

20 WT (solid)
20 WT MAC CR(dash)
20 WT PHY CR (dotted)

Figure 6.16:

50 WT(solid)
50 WT MAC CR(dash)
50 WT PHY CR(dotted)

Collision resolution

0.9

6.3. Simulations at high data rates (10

Mb/s)

73

an produ e ompletely dierent results regarding the mean delays. Figure 6.16 presents
the best results found using CR in the dierent systems. Table 6.9 lists the optimal values
found for every of the simulated systems.

Sin e a CR method is oered, the maxRBO

parameter that was optimized by the simulation in Se tion 6.3.1.1 an be set to a smaller
value.

Nr. of Collres

[ 16-Dec-1998 18:11:06 ]

Number of Multi_RA slots


0.30

Multi_RA slots [%]

0.25

0.20

0.15

0.10

0.05

0.00

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Load
10 WT MAC
10 WT PHY(dash)

Figure 6.17:

20 WT MAC CR (solid)
20 WT PHY CR (dash)

50 WT MAC CR (solid)
50 WT PHY CR (dash)

Number of Multi_RA slots to the total number of slots

Figure 6.17 exhibits the number of Multi_RA slots for CR relative to the total number of
slots for all systems with CR simulated (e.g. 10, 20 and 50 WTs). For a low system load
the number of CR slots is very high ex eeding 25% for the 20 and the 50 WT system. This
is be ause the

request table (RT) utilization is low and the possibility of an empty RT high.

Additionaly, the RA hannel utilisation is low (see Figure 6.18), so that a few onse utive
ollisions are likely to happen ( ause for PHY_CR) and many onse utive slots are empty
of Xmt_Reqs or PGBK ( ause for MAC_CR). As the total system load grows the number
of Multi_RA slots drops signi antly until it rea hes a minimum. This minimum strongly

maxRBO, Nma

depends on the variable system parameters (

and

Nphy )

and the number

of terminals in the system.

Higher loads ( > 60%) result in an in reasing of the Multi_RA slot number.

This is a

side-ee t of the CR as proposed in the present systems. When a CR happens the terminals will use the CR algorithm based on

slotted ALOHA in order to pass their requests

to the PBS. The terminals whi h will not re eive a positive Multi_CH ACK on the next
DL slot, as a result of a ollision or errored transmission, will hange ba k to the original
DQRUMA working mode and perform the random Ba k-O pi king a random number
(out of 1-

maxRBO).

The terminal must then wait for this number of slots until it tries

6. Simulation s enarios

74

to transmit its Xmt_Req to the base station. As the number of terminals in the system
grows, the Multi_RA CH utilisation drops and more terminals will need to use the RBO
mode after a CR slot. This in reases the possibility of new ollisions on the RA hannels
after a CR.

There is one more important matter related to the RBO after a CR. If a new CR hap-

maxRBO

pens after a number of slots smaller the

parameter, then potentially su esfull

Xmt_Reqs will not be able to be transmitted be ause the normal proto ol working s heme
will be interrupted. Small

Nma

or great

Nphy

values ombined with the maxRBO param-

eter an have major inuen e on the system perfoman e ausing often a CR to happen.
This must be taking are of when hoosing the system parameters.

ODLB collisions

[ 16-Dec-1998 18:11:06 ]

collision rate
0.7
0.6

% Collisions

0.5
0.4
0.3
0.2
0.1
0.0
0.2

0.3

0.4

0.5

0.6

10 WT 7 (solid)
10 WT MAC CR 5 5 (solid)

Load

10 WT PHY CR 5 3 (solid)

0.7

0.8

0.9

20 WT 15(dash)
20 WT MAC CR 7 7(dash)
20 WT PHY CR 12 3(dash)

50 WT (dotted)
50 WT MAC CR 25 2(dotted)
50 WT PHY CR 40 2 (dotted)

Figure 6.18:

Collision rate for original DQRUMA and DQRUMA CR

Figure 6.18 depi ts the RA utilisation for the three aforementioned systems (10, 20, and
50 WTs). The simulation results for the normal DQRUMA proto ol are here ompared
with those from the systems with CR. Generally, it an be obtained that the RA CH
utilisation for the CR systems is lower than in the original DQRUMA systems. This is
be ause the total amount of the RA hannels grows signi antly using CR (ea h Multi_RA
slot ontains 17 single RA CH for a CH data rate of 10 Mbps/s).

For some parameter

ongurations or for a high number of system terminals the RA utilisation might drop
suddently as explained in Figure 6.17. This happens for example in the 50 WT CR systems.
The previous simulations have shown that the DQRUMA proto ol with CR is unable to
serve 50 terminals for a total system load greater than 60%, as the ollisions beyond this
point dominate in the RA hannels. The systems with 10 and 20 WT are stable for higher
system load values up to 70%.

6.4. Proto ol omparison

6.4

75

Proto ol omparison

6.4.1 DQRUMA vs. TDD DSA++


In this simulation DQRUMA will be ompared with the DSA++ proto ol a ording to an
existing DSA++ s enario.

This omprises 10 WTs with

Poisson Sour es

that maintain

symmetri onne tions with the Base Station. Ea h onne tion transfers a load of 1.1 Mb/s.
This is the load that enters the LLC layer and onsists of ATM ells. On the other hand,
the PHY hannel provides a gross data rate of 50 Mb/s (see Figure 6.19).

Generated Load
11,1 Mbps

Generated Load
11,1 Mbps

WT

PBS
LLC

LLC

MAC

MAC

PHY

PHY
25 Mbps
25 Mbps

Figure 6.19:

Simulation ar hite ture DSA ++

The DSA++ terminals own an LLC and a PHY layer, as it an be obtained from Figure
6.19, in ontrast to the DQRUMA terminals modelled in this thesis. This must be taken
are of when determining the simulation parameters.
As mentioned above, the 20 onne tions produ e a total of load 22.2 Mb/s. This means that

22:2 Mb=s ' 57864:6 ATM ells=s


48byte  8

(6.7)

are fed to the MAC layer, whi h result that a single ell must be transmitted within a
period of

1
57864:6 ATM ells=s ' 17:282 s
The hannel rate is hosen to 50 Mb/s in order to omply with a load of 100%.

(6.8)
This is

be ause a ell at the LLC has a length of 48 byte, while a PHY burst is 108 byte long. For
a load of 22.22 Mb/s at the LLC it is at the PHY

Mb
22:22 Mb=s  108byte
48byte ' 50 =s

(6.9)

6. Simulation s enarios

76

LLC-Header

MAC-Header

ATM cell

DSA ++ Slots
.......

RA

PGBK

.......

Figure 6.20:

ATM cell

DQRUMA Slot

Comparing the slot stru tures

Sin e DSA++ is a TDD proto ol, another simpli ation needs to be done in order to
ompare it with the FDD DQRUMA. For this, we assume ea h UL and DL data rate to
be 25 Mb/s. This also means, that the slot size will be doubled a ording to Equation 6.6
and 6.8 to be now 34.57

s.

This slot size refers to the normal DSA++ slot and annot be used for a whole DQRUMA
slot, sin e the latter also in ludes a request a ess hannel. DSA++ transmits su h information at the end of a frame. So, additional time per slot for the RA hannel must be
added to the al ulated slot size (see Figure 6.20).
First, we substra t from the 34.57
(1.96

s):

s (Slot size DSA)

PGBK &Pa ket Xmt = Slot size DSA

the time for the burst overhead

(Guard time + R=F

PGBK &Pa ket Xmt ) 432 DQRUMA


RA , is given next:

Within this time (

time) = 32:61 s

(6.10)

bit are transmitted. For the 8 RA

bit the transmission time,

RA = PGBK &Pa ket Xmt 

RA length
P GBK + P a ket Xmt

= 0:604 s

(6.11)

Guard interval and Rise/Fall time must be added to the value found for the RA hannel,
be ause it is transmitted in a single burst:

RAburst

= RA + Guard time + Rise time + F all time = 2:564 s

(6.12)

Now adding the values found by the equations 6.10 and 6.12 provides the total DQRUMA

dqrumaslot ):

slot length (

dqrumaslot

= PGBK &Pa ket Xmt + RAburst = 37:134 s

(6.13)

This is the slot length set in the DQRUMA system.


The

Poisson Sour es

are dimensioned in the following. As already mentioned, one sour e

generates a maximum of LLC load equal 1.1 Mb/s. Sin e one ATM ell in ludes at the LLC
layer 384 bit, the sour e produ es:

6.4. Proto ol omparison

77

1:1 Mb=s = 2890:625 ells=s


384bit

So, the

(6.14)

mean Poisson inter-pulse time an be approximated to:

1
2890:625 ells=s ' 3:46  10

4 s= ell

(6.15)

Another important issue for approximizing the DSA++ simulation is the absen e of a
PHY layer in the DQRUMA system. For this, we simply substra t the extra bandwidth

NOTE: DSA PHY slot length = 108 bit, DSA MAC slot length
RMACdqruma , is:

needed for the PHY layer (

= 62 bit). Hen e, the MAC data rate,

RMACdqruma

62byte = 14:352 Mb=s


= 25 Mb=s  108byte

(6.16)

Table 6.10 summarises the most important DQRUMA simulation parameters for the omparison with the DSA ++ proto ol.

Figure 6.21 presents the simulation results for the

omparison between DQRUMA and the DSA++ proto ol as proposed in se tion 4.5.2.

mean delay hara teristi s for both proto ols, as well as the omplementary distribution fun tion (CDF) for a load adjustment fa tor of 0.8.

The diagrams show the

DSA++
Xmt S heme
S heduling
Collision Resolution

FDD

Earliest Due Date

Round Robin

Splitting

RBO

50 Mb/s

ea h 25 Mb/s for UL and DL

PHY Channel
Number of Conne tions
Load per onne tion
Buer Size

20

20

1.1 Mb/s

1.1 Mb/s

256

256

17.282

Slot Time

DQRUMA

TDD

Table 6.10:

s

37.134

Simulation parameters

In the DL both systems own the similar mean delays.


a signi ant UL delay dieren e.

s

Nevertheless, the diagram shows

This dieren e an be understanded, if the following

aspe ts are onsidered.


In order to ompare a FDD (DQRUMA) with an TDD (DSA++) system, some important
limitations were needed. First, we shared the available PHY data rate equally to the UL
and the DL CH (ea h 25 Mb/s). One important feature of the DSA++ is that the boundary between UL and DL in the frame is not xed, but moving a ording to the urrent
tra onditions. This means that if there is no tra in the DL, all available slots in the
frame will arry UL ells. Ignoring this feature an ae t the fun tionality of the DSA++
proto ol in a great manner.
There are two more reasons for the high delays in the DQRUMA asso iated with the proto ols intern stru ture. The rst one has to do with DQRUMA's s heduler. Round Robin
is implemented here as proposed in the relevant papers. On the other hand, the implementation of DSA++ used for the omparison uses

Earliest Due Date (EDD) as a s heduling

strategy. EDD is found to be mu h better than the most s heduling methods available and

6. Simulation s enarios

78

10WT poisson Rapp mean delay

[ 20Nov1998 22:25:04 ]

DQRUMA / DSA++ cell delay


1.0

mean cell delay [ms]

0.9
0.8
0.7
0.6
0.5

DQRUMA UL

0.4
0.3

DSA++ UL

DSA++ DL

0.2
0.1
DQRUMA DL

0.0

0.3

0.4

0.5

0.6

0.7

0.9

0.8

Load

Figure 6.21:

DQRUMA vs. TDD-DSA++: Mean delays

oers better results than RR. The se ond reason is the absense of a ollision resolution
algorithm (The DQRUMA proto ol with a ollision resolution method was investigated in
Se tion 6.3.3). The DSA++ uses an algorithm alled

Splitting for ollision resolution.

For the aforementioned reasons the urrent implementation of the DQRUMA proto ol
oers poor results ompared with the TDD version of the DSA++ proto ols.

6.4.2 DQRUMA vs. FDD DSA++


This simulation is going to ompare the DQRUMA proto ol with the FDD version of the

RXmt ) was:

DSA++ proto ol. In the DSA++ s enario the transmission rate (

RXmt = 50000 ells=s

(6.17)

nmobiles ).

The same s enario was simulated for a number of 6, 10 and 14 mobiles (

Poisson

sour es were used. The total adjusted load was hosen to be 75% of the transmission rate,
e.g. 37500 ells/s. This load is provided by all the mobiles residing in the urrent s enario.
Now, from the load per terminal, the

mean time between ells (tpoisson) an

be omputed

a ording to equation 6.18:

 1
Xmt
tpoisson =
nmobiles
The slot length in the FDD DSA++ system (DSAFDD ) is
 0:75  R

DSAFDD
whi h onsists of a
whi h is:

= 21:33 s

(6.18)

(6.19)

guard interval of 1.33 s and the time used for the ell Xmt (DSA ell),

6.4. Proto ol omparison

79

Complementary Distribution Function - Load 0.8


1

CDF

0.1

0.01
DQRUMA UL

DSA ++ UL

DQRUMA DL
0.001

DSA++ DL

0.0001
0.001

Figure 6.22:

0.002

0.004

0.005

0.006

DQRUMA vs. TDD-DSA++: CDF for a Load of 80%

DSA ell
Within a period of

0.003
Delay [sec]

DSAFDD

= 500001 ells=s = 20 s

(6.20)

the DQRUMA PGBK&pa ket_Xmt burst must be transmit-

ted. As shown in Se tion 6.4.1, the length of the RA hannel must be added to

DSAFDD , in

order to get the DQRUMA slot. Using the same pro edure followed by equations 6.10-6.13

DQRUMA ) for this simulation

the total DQRUMA slot length (

an be omputed to:

DQRUMA ' 23:04 s

(6.21)

The simulation parameters an be found gathered on Table 6.11.


FDD DSA++
Xmt S heme

DQRUMA

FDD

FDD

gross data rate

50000 ells/s

50000 ells/s

Generated Load

37500 ells/s

37500 ells/s

Slot Time

Table 6.11:

21.33

s

23.04

s

Simulation parameters DQRUMA vs. FDD DSA++

Figure 6.23 presents the results for the omparison of the DQRUMA proto ol with the
FDD-DSA++ proto ol.

Ea h proto ol is represented by three urves for 6, 10 and 14

terminals.

The simulation result for the omparison between the DQRUMA and the FDD-DSA++
proto ol shows for both systems the same behaviour. As the number of terminals in the
system grows, the delays grow higher. Sin e the total system load remains the same for
every of the three systems simulated (6, 10 and 14 WTs) the load per terminal is smaller,

6. Simulation s enarios

80

so that the terminals use more often the random a ess CH to transmit their requests.
More ollisions happen then, whi h push the delays higher. This happens to both systems.
On the other hand, the DL for the DQRUMA system remains the same sin e the RR DL
s heduler at DQRUMAs PBS has to serve the same load that is generated there ( onne tion are symmetri , e.g. UL and DL have to transfer the same load).

DQRUMA 14WT
0.1

Cell Delay CDF

DQRUMA 10WT

0.01
DQRUMA 6WT

DSA 6WT

DSA 10WT

1e-3

DSA 14WT
1e-4

Figure 6.23:

20

40

60

80

100
t/Slots

120

DQRUMA vs. FDD-DSA++: CDF for UL and DL delays for a Load of 75%

As it an be seen in Figure 6.23 the DQRUMA proto ol oers shorter transmission times
for most of the ells as long as the number of terminals in the system is small (e.g. 6 WT).
The time axis is divided here in DSA++ slots. In that ase app. 70% of the ells for a
6 WT system an be transmitted faster by the DQRUMA system.

This is aused by a

signalling delay of the DSA++ proto ol that lasts app. 10 DSA++ slots. Within this time
the dynami parameter ex hange is performed so that the s heduler an assign the frame
slots to the terminals. Sin e DQRUMA uses a Slot-by-Slot assignment, no su h delays are
aused. For a 10 WT system DQRUMA still transmits app. 60% of its ells faster than
DSA++, while for a 14 WT system the situation is inverted.

Here, DSA++ transmits

60% of its ells faster.

It an be obtained that DQRUMA auses long delays, espe ially for greater numbers of
system terminals. This has to do with the hara teristi of the slotted ALOHA proto ol
used. In ase of onse utive ollisions the total delay for the ollided terminals will grow
high.

DQRUMA uses in this implementation no ollision resolution algorithm.

other hand FDD DSA++ uses an algorithm alled

On the

splitting for ollision resolution purposes.

6.4.3 Nominal Rate (NR) s heduling


The inuen e of tra on dierent onne tion lasses will be investigated in this se tion.
20 terminals take part in the present s enario. Every terminal has one onne tion to the

6.4. Proto ol omparison

81

PBS. The simulation environment an be seen in Figure 6.24. Tables 6.12 and 6.13 omprise the simulation parameters.

DQRUMA FDD 10WT ODLB 4VBR 4 CBR 2 ABR NR scheduling

Init Terminal
Addressing

Init

Execute
In Order

Error Free
Channel w/Distance&Collisions
DL

2
1

Init

Init

DQRUMA Base
Station OneDLBurst
multiple Sources
NR scheduling

Init
Init
DQRUMA rtVBR
Terminal
OneDLBurst

Error Free
Channel w/Distance&Collisions
UL

Compute
Statistics
OneDLBurst

[ 25-Nov-1998 14:21:38 ]

Init
DQRUMA CBR
Terminal
OneDLBurst

DQRUMA rtVBR
Terminal
OneDLBurst

DQRUMA CBR
Terminal
OneDLBurst

DQRUMA ABR
Terminal
OneDLBurst
Init

Init

Init
DQRUMA rtVBR
Terminal
OneDLBurst

M Channel State Matrix

Init
DQRUMA CBR
Terminal
OneDLBurst

DQRUMA rtVBR
Terminal
OneDLBurst

DQRUMA CBR
Terminal
OneDLBurst
Init

M Goto Set Position


M Flow Counter Vector DL
M Receiver State Vector DL
M Channel Packet Exchange DL

Init
Init

M Channel Packet Exchange UL


M Receiver State Vector UL

Init
DQRUMA CBR
Terminal
OneDLBurst

Init
DQRUMA rtVBR
Terminal
OneDLBurst

M Flow Counter Vector UL

Init
P Slot Time
Init

DQRUMA ABR
Terminal
OneDLBurst

DQRUMA CBR
Terminal
OneDLBurst

DQRUMA rtVBR
Terminal
OneDLBurst
Init
DQRUMA CBR
Terminal
OneDLBurst

P SCAN Time
P Maximum Queue Size-MAC

Init

Init

DQRUMA CBR
Terminal
OneDLBurst

Init

P Bit/sec
P Receive Range
P # WT

DQRUMA rtVBR
Terminal
OneDLBurst

DQRUMA rtVBR
Terminal
OneDLBurst

DQRUMA ABR
Terminal
OneDLBurst

P RA Delay
P Mean Inter-Pulse Time CBR
P PCI Lenght Downlink

P Number of pulses per burst

P NumGen Minimum

P Inter Pulse Time

P NumGen Maximum

P Mean Delay Between Bursts

P Time for Ack

P Mean number of pulses per burst

P PCI Lenght

P Inter-Pulse Time (during burst)

Init

P Request Lenght

DQRUMA ABR
Terminal
OneDLBurst

P # DQRUMA Terminals

Figure 6.24:

Multiple servi e lass s enario

rtVBR, eight CBR and four ABR onne tions, transmitting with the same
mean data rate. This means that the tra will also onsist of 40% rtVBR ells, 40% CBR
ells and 20% ABR ells. The load was so omputed, that the sum of the nominal rates is
There are eight

equal to the hannel transmission rate.


The ATM ells in the E-Burst proto ol (see Se tion 4.5.4) are transmitted into

frames,

MAC

ea h with a size of 50 ells. DQRUMA instead used a slot-by-slot allo ation so

that the NR method must be re-designed (see Se tion 4.4.4.1). As already done for previeous s enarios, the slot length for the DQRUMA must be re omputed in order to be ome
omparable results.
In the E-Burst proto ol one ATM ell is transmitted in a

Normal slot within normal

= 14

s.

This time in ludes R/F and guard times that must be extra ted, in order to get the time
of transmission of the 53 bytes of the ATM ell. Then the RA burst length must be added
to get the total slot length.

Following the same pro edure des ribed by the equations

6.10-6.13 the DQRUMA slot length,

dqruma , an be omputed to:

dqruma ' 16:183 s

(6.22)

Within this time the whole DQRUMA slot must be tted so that it an be transmitted. It
must be taken are of

guard

and

Rise/fall times.

Hen e, we an form an equation, whi h

RMAC ):

provides the ne essary transmission speed on the MAC layer (

dqruma  2  (guard + rise + fall ) +

DQRUMA slot length


RMAC

(6.23)

The equation 6.23 simply shows that the the MAC slot onsists of two bursts, in ludes a
total of

DQRUMA slot size = 440 bit

and should not ex eed

Solving the equation, we nd for the ne essary Xmt data rate:

dqruma .

6. Simulation s enarios

82

RMAC

 35:89 Mb=s

(6.24)

This is the hannel rate that must de adjusted for the DQRUMA simulation. The sour e
hara teristi s are listed in the Table 6.13.
E-Burst

DQRUMA

number of onne tions

20

20

buer spa e

500 ATM- ells

500 ATM- ells

PCTRL-PDU duration

s
14.7 s
2.2 s
14.7 s

number of slots per frame

50

Normal-Slot

14.0

Super-Slot
Energy-Burst

Table 6.12:

16.183

s

no frames

System and hannel parameter of the 20 onne tions simulation

parameter

rtVBR

CBR

ABR

sour e hara teristi

onstant burst length

deterministi

On-O-sour e

n/a

30 (average)

bursts per se ond


average rate
PCR
maxCTD, a ess delay

Table 6.13:

30

Mbit=s
10.00 Mbit=s
20 ms, 1.5 ms

Mbit=s
1.43 Mbit=s
5 ms, 0.0 ms

1.43

1.43

Mbit=s
20.00 Mbit=s
1.43

n/a

Sour e hara teristi s of the 20 onne tions simulation

6.4.3.1 Impa t on CBR and rt VBR onne tions


Figure 6.25 present the omplementary distribution fun tions of the ell delays for CBR
and rt VBR servi es. Ea h of the two diagrams ontains the CDF fun tions for RR, EBurst NR and DQRUMA NR s heduling. For CBR servi es the DQRUMA-NR is better
than the E-Burst-NR. This has rst to do with the signalling of the E-burst proto ol, so
that the rst CBR ells an be transmitted after app. 1 ms. A se ond reason is asso iated
with the approximation of the nominal rates for the rt servi es. After the omputation of
the NR [in %, values had to be round o for the integration into the

virtual frame.

In a

long MAC frame the exa t nominal rates an be used.


For CBR servi es it is shown that the demand for a

maxCTD of 5 ms

an be guaranteed

and still a signi ant margin to the 5 ms boundary is present. For the RR s heduling the
CBR delays own same hara teristi as the ABR delays for the same s heduling strategy.
This depends on the s heduling method, whi h serves the terminals a ording to their
mean data rates and not a ording to the a tual needs at a spe i time point. This also
applies for the VBR servi es.
E-Burst-NR manages to transmit the VBR ells within the VBR

maxCTD of

20 ms. On

the other hand, DQRUMA-NR ex eeds the time limit for about 0.5 ms. This an be explained, if the approximations needed for the appli ation of the NR to a slot-by-slot system
are onsidered.

6.4. Proto ol omparison

83

VBR traffic

CBR traffic

Round Robin

Round Robin

maxCTD = 20ms
0.1

maxCTD = 5ms

0.1

0.01

E-Burst NR

CDF

CDF

E-Burst NR

DQRUMA NR

0.001

0.0001
0.0

DQRUMA NR

0.01

0.001

0.01

0.02
0.03
Delay [sec]

0.04

0.05 0.0001

0.001

(a)

Figure 6.25:

0.002

0.003
0.004
Delay [sec]

0.005

0.006

(b)

Complementary Distribution Fun tions for VBR and CBR servi es

6.4.3.2 Impa t on ABR onne tions


The CDF for the delays of the ABR ells is depi ted in Figure 6.26.

The

Round Robin

algorithm serves all onne tions a ording to their mean data rate, without taking are
of the hara teristi s of the dierent tra sour es.

If a sour e generates a long burst,

this will not be a ounted for by the s heduler, so that the ells will be sta ked in the
buer ausing long delays. One more signi ant dieren e between the DQRUMA and the
Energy-Burst proto ol is the maximum system throughput.

E-Burst has a greater maximum hannel apa ity than DQRUMA. Additionaly DQRUMA
has a greater proto ol overhead. It is shown on previous simulations that DQRUMA rea hes
its limits and is overloaded for a generated load of app. 90% of the CH apa ity. Sin e the
present E-Burst simulation is dimensioned with a system load of 102%, it an be seen that
DQRUMA will be pushed to its limits and will not be able to serve the terminals properly.

The E-Burst-NR s heduling transmits almost 85% of the ABR ells faster than RR s heduling. The rest ones are served qui ker by the RR s heduler. This happens be ause RR serves
the terminals with the negotiated mean data rate at the beggining of the onne tion. In
ontrast to RR, the E-Burst-NR only serves an ABR terminal, after real time (rt) onne tions have pla ed its ells in the frame a ording to the NR, and the MAC-frame still is
not lled.

The DQRUMA-NR s heduler serves app. 95% of the ABR ells slower than the E-Burst
s heduler. This has to do with the implementation of the DQRUMA-NR method. A ording to Se tion 4.4.4.1 we hoose to transmit ABR ells, only if no rt servi es are present in
the request table of the BS s heduler. So, it is more likely to serve ABR ells in a frame
stru ture, be ause the slot planning happens at the beggining of the frame for the next 50
slots. If there are no more requests for rt servi es, ABR is served. For a slot-by-slot slot

6. Simulation s enarios

84

allo ation s heme (DQRUMA) the Xmt_Perm announ ements happen at every slot and
as rt servi es enter their requests onstantly, these are being preferred.
The simulation with NR s heduling serves the rt servi es a ording to their needs, so
that for CBR and rt VBR onne tions no buer overows o ur. On the other hand, as
des ribed above there are signi ant losses for the ABR ells (see Figure 6.27). This has
to do with the servi ing strategy the DQRUMA NR s heduler uses.

ABR traffic

DQRUMA NR

CDF

0.1

Round Robin

E-Burst NR

0.01

0.001

0.0001
0

0.1

0.15 0.2 0.25


Delay [sec]

0.3

0.35

0.4

Complementary Distribution Fun tions for ABR servi es

DQRUMA CLR
0.40
0.35
0.30
cell loss ratio

Figure 6.26:

0.05

0.25

DQRUMA NR

0.20
0.15
0.10

DQRUMA RR

0.05

E-Burst NR

0.00
0.2

0.4

0.6

0.8

1.0

Load

Figure 6.27:

Cell loss ratio for ABR servi es

CHAPTER 7

Con lusion
7.1

Summary
he main target of this diploma thesis was the implementation and performan e evaluation of the DQRUMA MAC proto ol of the BAHAMA Network and its omparison

with other existing wireless ATM MAC proto ols.

At rst, a short introdu tion of the main ATM features was presented, followed by the aspe ts of the wireless expansion and the problems related to the transmission over wireless
links. Two possible wireless ar hite tures were then depi ted. Chapter 3 is dedi ated to the
des ription of the BAHAMA network invented by AT&T's

Bell Labs.

The following hap-

ter gives a detailed presentation of BAHAMA's medium a ess proto ol alled DQRUMA,
and the enhan ements made within this thesis. Chapter 6 presents all simulations made,
also evaluating the results won.
The original proto ol stru ture has been modelled with

BONeS Designer

and the rst

simulations proved the proto ol's orre tness and do umented the impa t of several proto ol parameters, su h as

Random Ba k O, Slot time and Number of Mobiles.

Based on the original DQRUMA proto ol two enhan ements were presented, designed and
evaluated. The results were ompared with the original system. The rst enhan hement
was the elimination of ollisions for the Xmt requests of the terminals. This was a omplished by polling the terminals with empty

Request Table

elds to transmit their Xmt

Requests over the next RA hannel.


The se ond improvement modies the s heduling strategy by taking are of the sour e
hara teristi s. For this, the existing

Nominal Rate

method was modied, in order to be

used in a slot-by-slot assignment s heme. In this ontext, a new stru ture was proposed,
the so alled

Virtual Frame

(VF). The VF proposes a dynami stru ture, whi h de ides

over the slot assignment, based on existing

Request Table information.

Finally, the performan e of the aforementioned systems was analysed and ompared with
two existing

DSA++

implementations (FDD and TDD) and the

using dierent simulation s enarios.

Energy-Burst

proto ol

In order to ompare partially dierent systems, a

omparison s heme was proposed and simpli ations needed to be made.

7.2

Outlook

During this thesis new questions raised, whi h should be investigated, in order to get a
more pre ise view of the DQRUMA proto ol.
The system presented in this thesis proposes an error free hannel. So, a realisti simulation system should ontain a PHY layer of a hannel model (e.g. Gilbert-Elliot model).
Further, a LLC layer with a ARQ algorithm should be installed, thus providing the system

7. Con lusion

86

with an error orre tion me hanism to de rease the ell loss ratio on the wireless link.
The DQRUMA proto ol presented in this thesis is the FDD implementation for two frequen ies (UL and DL frequen e), but more frequen ies an be used. The impa t of severals
frequen ies ould be investigated in future proje ts. The DQRUMA proto ol an also be
designed for a TDD implementation.
super mobile owning an entry in the

This an be a hieved by treating the PBS like a

Request Table.

The s heduler will most likely need

to serve the PBS more often than the mobiles, e.g. by using a weighted Round Robin
pa ket transmission poli y. A TDD implementation would also eliminate the need of many
simpli ations made for the omparison of a FDD system (DQRUMA) with TDD systems
(TDD-DSA++, Energy-Burst).
The present ollision resolution algorithm highlighted many problems, when it was used in
systems with many terminals or for high system load values. The result evaluation showed,
that it ould be more e ient, if a modied algorithm would use onse utive Multi_CH
slots for ollision resolution.
Handover and inter- ell interferen e are items not yet analysed for the DQRUMA proto ol.
This has to be done sin e BAHAMA proposes a multi ell environment.

LIST OF FIGURES

2.1

ATM Time Division Multiplex

. . . . . . . . . . . . . . . . . . . . . . . . .

10

2.2

ATM Cell Stru ture at UNI and INI [25 . . . . . . . . . . . . . . . . . . . .

10

2.3

ATM Cross Conne ts and ATM Swit hes . . . . . . . . . . . . . . . . . . . .

11

2.4

ATM Referen e Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

2.5

Ar hite ture of a ellular mobile ATM network [26 . . . . . . . . . . . . . .

15

2.6

Merging a ellular ATM radio network into an ATM ore network [26

. . .

16

2.7

Proto ol Sta k of the ATM Air Interfa e . . . . . . . . . . . . . . . . . . . .

18

2.8

Centralized WATMLAN Ar hite ture based on a distributed ATM swit h

18

2.9

Centralized WATMLAN Ar hite ture . . . . . . . . . . . . . . . . . . . . . .

19

2.10 Distributed WATMLAN ar hite ture . . . . . . . . . . . . . . . . . . . . . .

19

2.11 Ad-ho WATM LAN a distributed ATM swit h . . . . . . . . . . . . . . . .

19

2.12 The four HIPERLAN types [5

. . . . . . . . . . . . . . . . . . . . . . . . .

21

3.1

The BAHAMA ell [14

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

3.2

The BAHAMA System ar hite ture [7 . . . . . . . . . . . . . . . . . . . . .

24

3.3

The BAHAMA proto ol sta k [6 . . . . . . . . . . . . . . . . . . . . . . . .

24

3.4

Wireless Virtual LAN proto ol sta k [6

. . . . . . . . . . . . . . . . . . . .

25

3.5

proto ol sta k [6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25

3.6

The VP/VC on ept [7

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

3.7

The Homing Algorithm [7 . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

4.1

DQRUMA proto ol stru ture [14 . . . . . . . . . . . . . . . . . . . . . . . .

29

4.2

DQRUMA ow hart at mobile [14

. . . . . . . . . . . . . . . . . . . . . . .

30

4.3

Round Robin s heduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31

4.4

DQRUMA timing diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33

4.5

DQRUMA hannel model

4.6

DQRUMA slot bursts

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

35

4.7

multiple RA hannels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

4.8

RA Polling

37

4.9

The prin iple of ideal DQRUMA

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.10 Burst expansion [21

. . . . . . . . . . . . . . . . . . . . . . . .

37

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38

4.11 Virtual Frame for a full RT

. . . . . . . . . . . . . . . . . . . . . . . . . . .

39

4.12 Virtual Frame for one (or none) rt onne tion . . . . . . . . . . . . . . . . .

40

4.13 DSA++ proto ol proto ol Stru ture [18 . . . . . . . . . . . . . . . . . . . .

42

4.14 Signalling of the FDD DSA++ Proto ol [16 . . . . . . . . . . . . . . . . . .


4.15

MAC-frame with Energy-burst

43
44

5.1

BONeS-Designer Stru ture [11

. . . . . . . . . . . . . . . . . . . . . . . . .

46

5.2

Blo k stru ture [11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47

5.3

Data- & Control Primitives [11 . . . . . . . . . . . . . . . . . . . . . . . . .

47

5.4

The Data Stru ture

. . . . . . . . . . . .

48

5.5

Simulationsar hitektur [11 . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

Signalling(Philips) [17

Interfa e Control Information [11

. . . . . . . . . . . .

5.6

System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

50

5.7

Terminal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51

5.8

Layer-Template . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

List of Figures

88

5.9

MAC-Layer

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

. . . . . . . . . . . . . . . . . .

53

. . . . . . . . . . . . . . . . . .

53

6.1

Simulation example with 5 terminals . . . . . . . . . . . . . . . . . . . . . .

56

6.2

Mean and maximum delays for a 2 Mb/s system

57

6.3

Mean and maximum delay between a rst pa ket and

. . . . . .

58

6.4

Collision rate on the RA Channel . . . . . . . . . . . . . . . . . . . . . . . .

59

5.10 MAC-

MAC Pro essing Unit at mobile . . .


MAC Pro essing Unit at Base Station

5.11 MAC-

. . . . . . . . . . . . . . . .

Xmt_Perm

6.5

PGBK-to-RA ratio and Mean buer o upan y . . . . . . . . . . . . . . . .

60

6.6

Downlink ell delay and Complementary Distribution Fun tion(CDF) . . . .

61

6.7

S-ALOHA proto ol performan e in the RA_CH . . . . . . . . . . . . . . . .

62

6.8

Variation of the terminal number . . . . . . . . . . . . . . . . . . . . . . . .

63

6.9

Weighted load simulation

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

64

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

65

6.10 20 WT simulation

6.11 Variation of the RBO parameter

. . . . . . . . . . . . . . . . . . . . . . . .

67

6.12 Mean ell delays for simple DQRUMA and DQRUMA polling . . . . . . . .

68

. . . . . . . . . . . . . .

69

6.14 Mean delays for a 20 WT system with dierent load per terminal . . . . . .

6.13 CDF for simple DQRUMA and DQRUMA polling

70

6.15 Mean Buer O upan y

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71

6.16 Collision resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

6.17 Number of Multi_RA slots to the total number of slots

73

. . . . . . . . . . .

6.18 Collision rate for original DQRUMA and DQRUMA CR . . . . . . . . . . .

74

6.19 Simulation ar hite ture DSA ++

. . . . . . . . . . . . . . . . . . . . . . . .

75

6.20 Comparing the slot stru tures . . . . . . . . . . . . . . . . . . . . . . . . . .

76

6.21 DQRUMA vs. TDD-DSA++: Mean delays

78

. . . . . . . . . . . . . . . . . .

6.22 DQRUMA vs. TDD-DSA++: CDF for a Load of 80%

. . . . . . . . . . . .

79

6.23 DQRUMA vs. FDD-DSA++: CDF for UL and DL delays for a Load of 75%

80

6.24 Multiple servi e lass s enario . . . . . . . . . . . . . . . . . . . . . . . . . .

81

6.25 Complementary Distribution Fun tions for VBR and CBR servi es

83

. . . . .

6.26 Complementary Distribution Fun tions for ABR servi es . . . . . . . . . . .

84

6.27 Cell loss ratio for ABR servi es

84

. . . . . . . . . . . . . . . . . . . . . . . . .

LIST OF TABLES

2.1

Chara teristi s of typi al ATM servi es [26

2.2

ATM Adaption Layer Types [25

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

13

2.3

ATM servi e ategories and their QoS parameters . . . . . . . . . . . . . . .

14

2.4

Te hni al data of the ATM air interfa e [5 . . . . . . . . . . . . . . . . . . .

16

4.1

Slot stru ture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

35

4.2

Guard Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

6.1

Uplink Bursts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

55

6.2

Downlink Burst . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

55

6.3

5/10 Terminal original DQRUMA Simulation

. . . . . . . . . . . . . . . . .

57

6.4

Parameters for a system with variable number of terminals . . . . . . . . . .

62

6.5

5 Terminal Simulation with asymmetri onne tions

63

6.6

Simple DQRUMA simulation parameters for 10, 20 and 50 WTs . . . . . . .

66

6.7

RBO Parameter Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66

. . . . . . . . . . . . .

6.8

DQRUMA polling with dierent load per terminal

. . . . . . . . . . . . . .

67

6.9

Parameters for ollision resolution . . . . . . . . . . . . . . . . . . . . . . . .

71

6.10 Simulation parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

77

6.11 Simulation parameters DQRUMA vs. FDD DSA++

. . . . . . . . . . . . .

79

6.12 System and hannel parameter of the 20 onne tions simulation . . . . . . .

82

6.13 Sour e hara teristi s of the 20 onne tions simulation

82

. . . . . . . . . . . .

90

List of Tables

LIST OF ABBREVIATIONS

AAL
ABR
ACK
ACTS

ARQ
ATM
ATML
B-ISDN
BER
PBS
CBR
CDF

CER
CH
CR
DCS 1800
DECT

DL
DSA
EDD
E-Burst
FDD
FDM
FEC
GFC
GSM

ISDN
LAN
LLC

ATM Adaptation Layer


Available Bit Rate
ACKnowlende
Advan ed Communi ations
Te hnologies and Servi es
Automati Repeat ReQuest
Asyn hronous Transfer Mode
ATM Layer
Breitband-ISDN
Bit Error Rate
Portable Base Station
Constant Bit Rate
C omplementary Distribution
Fun tion
Cell Error Ratio
CHannel
Collision Resolution
Digital Cellular System 1800
Digital European Cordless
Tele ommuni ations
DownLink
Dynami Slot Assignment
Earliest Due Date
Energy-Burst

Frequen y Devision Duplex


Frequen y Division Multiplexing
Forward Error Corre tion
Generi Flow Control
Global System for Mobile
Communi ation

Integrated Servi es Digital Network


Lo al Area Network
Logi al Link Control

MAC
MS
NNI
PCR

Medium A ess Control


Mobile Station
Network Node Interfa e
Peak Cell Rate
Period ConTRoL-PDU
Proto ol Data Unit
PiGgyBa King
PHYsi al Layer
Quality of Servi e
Request A ess
Resear h and te hnology
development in Advan ed
Communi ations te hnologies in
Europe
Radio A ess Layer
Segmentation And Reassembly
Servi e Primitive
Time Devision Duplex
Time Division Multiple A ess
Tran eiver Turn Around
Unspe ied Bit Rate
User Network Interfa e
UpLink
Variable Bit Rate
Virtual Channel
Virtual Channnel Identier
Virtual Path Identier
Wireless ATM
Wireless Terminal

PCTRL-PDU
PDU
PGBK
PHY
QoS
RA
RACE

RAL
SAR
SP
TDD
TDMA
TTA
UBR
UNI
UL
VBR
VC
VCI
VPI
WATM
WT
Xmt

Transmission

Xmt_Perm
Xmt_Req

Transmission Permission

Transmission Request

92

List of Abbreviations

BIBLIOGRAPHY

[1

Altagroup Homepage: Datasheet BONeS Designer, November 1998.

available at

http://www.altagroup. om/alta/produ ts/newdatasheets/designer.html.


[2

BONeS Designer Modeling Guide.

[3

BONeS Designer Users Guide.

[4

Computers and Digital Systems Le ture: E ien y of slotted ALOHA,

Altagroup In . Te hni al Referen e.

Altagroup In . Te hni al Referen e.


November

1998. available at
http://www.eee.bham.a .uk/james-roxbyp/ee3 2/L7/index.htm.
[5

[6

HIgh PErforman e Radio Lokal Area Network (HIPERLAN), Requirements and Ar hite tures. Draft tr, Sophia Antipolis FRANCE, February 1997.
ETSI RES 10.

E. Ayanoglu, K. Eng, Karol M.J., Z. Liu, P. Pan ha, M. Veeraraghavan, Woodworth


C.B.

Mobile Information Infrastru ture. Te hni al report, Bell Labs Te hni al Journal,

1996.
[7

K.Y. Eng, M.J. Karol, M. Veeraraghavan, E. Ayanoglu, C.B. Woodworth, P. Pan ha, R.A Valenzuela.

A wireless broadband ad-ho ATM lo al-area network.

Wireless

Networks 1, pp. 161  174, 1995.


[8

[9

F. Fitsilis.

Netze.

Gegenberstellung vers hiedener Kanalzugrisprotokolle fr drahtlose ATM

Student thesis, Chair for Communi ation Networks, July 1997.

D. Fuhrmann. Entwi klung von Management Funktionen zum Betrieb eines drahlosen
Multihop Adho ATM LAN. n, November 98.

[10 A. Hetti h.

Tutorial for using the Simulation Ar hite ture in BONeS.

Proje t: Wire-

less ATM LAN, Mar h 1997.

Untersu hung der dynamis hen Auswahl des Central Controllers in einem
auf W-ATM/UMTS basierenden Adho -LAN. Diplomarbeit, Chair for Communi ation

[11 K. Hong.

Networks, November 1998.

Simulative Leistungsbewertung des Energy-Burst Polling MAC-Protokolls


fr ad-ho ATM LANs unter Ber ksi htigung der Kanaleigens haften. Master's the-

[12 D. Karnesis.

sis, Diplomarbeit, Kommunikationsnetze, RWTH Aa hen, 1997.

Mobility-Management and Media-A ess Issues in the BAHAMA Wireless ATM LAN. In IEEE International Conferen e on
Universal Personal Communi ations, pp. 758762, 1995.

[13 M.J. Karol, Veeraraghavan M., K. Eng.

An e ient demand-assignment multiple a ess proto ol for wireless pa ket (ATM) networks. Wireless Networks 1, pp. 267  279, 1995.

[14 M.J. Karol, L. Zhao, K.Y. Eng.

[15 H. Kopka.

LATEX:Eine Einfhrung.

Addison-Wesley, Bonn, 1993.

Simulative und analytis he Untersu hung eines Kanalzugrisprotokolls


mit s hneller Kollisionsausung fr mobile ATM-Netze unter Ber ksi htigung von
verzgertem Feedba k. Diplomarbeit, Chair for Communi ation Networks, Mar h 1996.

[16 A. Kraemling.

Bibliography

94

Entwi klung einer Verbindungannahmesteuerung zum Betrieb eines drahtlosen Multi-Hop Ad-ho ATMLAN. Diplomarbeit, Chair for Communi ation Net-

[17 R. Kuntze.

works, June 1998.

Implementation and examination of methods for guaranteeing the QoS in


wireless ATM systems. Diplomarbeit, Chair for Communi ation Networks, July 1998.
M. Lott, A. Hetti h. Ar hite ture for an Extendable Platform for WATM System
Simulation and Spe i ation. Proje t: Wireless ATM LAN, January 1998.
M. Lott, A. Hetti h. Requirements for an extendable Platform for WATM System
Simulation and Spe i ation. Proje t: Wireless ATM LAN, January 1998.
I. Meiri k. Untersu hung von Verfahren zur dynamis hen Kanalvergabe auf BAsis von
periodis hen Zeits hlitzen im WATMLAN. Diplomarbeit, Chair for Communi ation

[18 P. Laing.

[19

[20

[21

Networks, Navember 1998.


[22 T.L. Saaty.

Elements of queueing theory.

[23 M. Veeraraghavan, M. Karol, K.Y. Eng.

wireless ATM LAN.

M Graw-Hill, 1961.

Mobility and onne tion management in a

Journal on Sele ted Areas in Communi ations, Vol. 15, No. 1,

1997.

Inplementation and Analysis of Conne tion-Managament Pro edures in a Wireless ATM LAN. In IEEE Conferen e on Universal Personal Communi ations, Vol. I, pp. 380  386, 1996.
B. Walke. Kommunikatiosnetze und Verkehrstheorie I&II. Vorlesung an der RWTH
Aa hen. Chair for Communi ation Networks, Aa hen, 1993.
C. Walke. Development and Performan e Evaluation of SDMA-Te hniques for Wireless ATM Networks under Consideration of the QoS. Diplomarbeit, Chair for Com-

[24 M.L. Veeraraghavan.

[25

[26

muni ation Networks, De ember 1998.

A demand-assignment a ess ontrol for


multi- ode DS-CDMA wireless pa ket (ATM) networks. In INFOCOM 1996, pp. 713

[27 L. Zhao, M.J. Karol, M. el Zarki, K.Y. Eng.

 721, 1996.