Reverse Link Channel Access Techniques CDMA Packet Networks · Reverse Link Channel Access...
Transcript of Reverse Link Channel Access Techniques CDMA Packet Networks · Reverse Link Channel Access...
Reverse Link Channel Access Techniques for CDMA Packet Networks
Subramaniam (Alagan) Anpalagan
h thesis submitted in conformity with the requirements
for the Degree of Doctor of Philosophy,
Department of Electricai and Cornputer Engineering,
at the University of Toronto
@ Copyright by S. -4npaIagan 3002 '
N a t i o ~ i Library 1*1 ofCanada Bibliothèque nationale du Canada
Acquisitions and Acquisitions et Bibliographie Services services bibliographiques 395 Wellingîon Street 395, rue Wellington Onawa ON KlAON4 ûttawaON K l A W canada CaMda
The author has granted a non- L'auteur a accordé une licence non exclusive licence aüowing the exclusive permettant ii la National Lbrary of Canada to Bibliothèque nationale du Canada de reproduce, loan, distniute or seil reproduire, prêter, disûibuer ou copies of this thesis in microform, vendre des copies de cette thèse sous paper or electronic formats. la forme de microfiche/iih, de
reproduction sur papier ou sur format électronique.
The author retains ownership of the L'auteur conserve la propriété du copyxight in this thesis. Neither the droit d'auteur qui protège cette thèse. thesis nor substantial extracts fiom it Ni la thèse ni des extraits substantiels may be printed or otherwise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits sans son permission. autorisation.
Reverse Link Channel Access Techniques for CDMA Packet Networks
Subramaniam (Alagan) Anpalagan
Degree of Doctor of Philosophy
Department of Electrical and Computer Engineering
University of Toronto
2001
Abstract
In this thesis, we propose, analyze and m e s s three novel radio channel access techniques CO
effectiveiy manage the interference in the reverse links of a cellular CDM.4 network with the
emphasis on the packet data applications. The first scherne is based on the combined use of
ratet powc and ce11 control. whereby base station receivers attempt to operate with equal
receive power. In the second approach, the temporal variations of the channels are esploiteti
to schedule packet-transmissions in different modes of operations. The third technique uses
the overlapping sectors for adaptive sectoring of a ceIl based on the non-uniform angular trafic
around the base stations. Al1 of these approaches are systematically developed based on the
fact that both the interference-reduction and the interference-balancing irnprove the CDSI.4
systern performance. The techniques in effect implernent interference control at the s'stern
level respectively (a) esploiting the source characteristics of the mobiles for network-wide
interference-balancing, (b) tirne-separat ing the sigfiaIs based on the relative charinel conditions
and dehying dominant interferers and (c) jointly seIectirig the best possible set of antenna
based on the the a n p l a r trafic at the antenna and making the user assignrnent that mininiizes
the total transmit potver. .Algorit hms for base station selection. transmission rate adjustment
and mobile power adaptation €or the abom schemes are aIso discussed. Sirnirlatioti crsdts and
numerical esamples that compare orirs with previoiisly reported techniques rlenionstratt~ t h
effectiveness of the rien- tec1iniqiics.
Acknowledgement s
1 would like to chank my supervisor Prof. Elvino S. Sousa for his invaluable advice.
guidance and support. He has always been accessible and very helpful,
Special thanks to my P h.D. Supervising Committee members: Prof. D. Hatzinakos and
Dr. 1. Katzela, Departmental Esamination Committee members: Prof. 1. Blake. Prof. B.
Francis and Prof. T. J. Lim, Esternal Examiner: Prof. P. Jeszensky and School of Gradiiate
Studies Exam Chair: Prof. kt. Thomson.
i am thankful to my friends in the Wireless Research Group for many thought-provoking
and stimulating discussions. I would also like to thank Rath, Benjamin and Farhad for proof-
reading part of my thesis. Special thanks to Halim and Wilson for your very interesting
thoughtful discussions. Diane B. Silva's and Sarah Cherian's administrative help are also
gratefully appreciated.
During the course of this research work, I !vas financially supported by the Governrnent
of Ontario, University of Toronto, Communications and Information Technology Ontario and
Norte1 Nettvorks and their assistance are v e l much acknowledged.
It is time to think of the past ... my high school teachers who tirelessly dedicated their
time and effort for 11s and who planted the seeds on us. especially Mr. Sivalingam and Mr.
Sivasubramaniam and many others. It is with great sorrow that ive miss you. Vaanar hIascer.
b u are remembered by your ileciication to the students.
1 am deeply grateful to my parents for their never ending cire and teaching me the values
of eclucation. 1 am rery much inclebted co my wife? .L[ekalao for her constant encouragement.
endless support and understanding. Our Arun has been wonderful bundle of j o -
To those who believe in higher education ... especially to my parents. This thesis is dedicated
to rny jather who has been a true champion of many achievements, who inspired and guided
us and, without ,whom I ,wouldn't be here defending my thesis.
Contents
Abstract
Acknowledgements
List of Figures
List of Tables
List of Abbreviations and Acronyms
List of Mathematical Terms
i
ii
viii
xiii
xiv
xvii
1 Introduction 1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 CDbIXSystems. . 3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Interference Control 4
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 4Iotivation ancl Objective 6
- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Thesis Organization r
2 Receive Power Control 9
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 [ntroduceion 9
2 2 Receive Pomer Controi (RPC) ProbIem and SoIution . . . . . . . . . . . . . . I l
. . . . . . . . . . . . . . . . . . . . . . . . . . . 22.1 CelIular Systrrn Ilo<lel 12
2-22 Sottrcc4metl RPC (SBRPC) . . . . . . . . . . . . . . . . . . . . . . . . 15
. . . . . . . . . . . . . 2.2.3 Xetwork-balanced RPC (XBRPC)
. . . . . 2.2.4 Capacity Evaluation in a Dual-Class System with Perfect RPC
. . . . 2.2.5 -4 Case Study: Fully-Loaded System and Interference-Balancing
. . . . . . . . . 2.3 Interference Power in Integrated Services FVireless Networks
2.3.1 Interference Power Variation -1nalysis . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Numerical Results
2.4 Minimizing the Interference Power Variation among Base Stations via Dynamic
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Receive Power Control
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Chapter Summary
3 Combined Rate/Power/Cell Control
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Introduction
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 System Model
3.3 Power and Cell Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 PC-only (Alg-0)
. . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 PC+CC (-Ug-I St AIg-11)
3.3.3 .A Potential Limitation in PC+CC Scheme . . . . . . . . . . . . . . . .
3.4 Two Algorithms for Combined Rate, Power and Cell Control . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 3.4.1 R/P/C Control (-4lg-III SI AIg-IV)
3-42 Optimal Solution in Mobile Transmit Bit Energy using Alg-III . . . . .
3.4.3 Comparison of Algorithms . . . . . . . . . . . . . . . . . . . . . . . . .
3.5 Simulation and Restdts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.1 Convergence of Algorithrtw . . . . . . . . . . . . . . . . . . . . . . . . .
3 -52 Average Bit Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.3 Minimum Transmit Bit Energy . . . . . . . . . . . . . . . . . . . . . .
3.5.4 Setwxk-wide Interference-Balancing . . . . . . . . . . . . . . . . . . .
3.5.5 Coverage Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Adaptive Sector Control 65
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Introduction 66
4.2 Fixed Overlapping Sectored Antenna Architecture (FOS-LA) . . . . . . . . . . 68
4.3 Minimum Mobile Transmit Power Solution . . . . . . . . . . . . . . . . . . . . 70
4.3.1 System Mode1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3.2 Interference Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
. . . . . . . . . . . . . . . . . . . . . 4.3.3 Minimum Mobile Transmit Power 74
-. 4.4 Spatial Interference Control Techniques and Algorithrns . . . . . . . . . . . . . ( 3
-- 4 -41 Radial Control and Cell-Breathing (CB) . . . . . . . . . . . . . . . . . r r
4.4.2 Azimuthal ControI and Cell-Slicing (CS) . . . . . . . . . . . . . . . . . 78
4.4.3 Hybrid Control and CeIl-Breathing plus Cell-Slicing (CB+CS) . . . . . 19
4.44 Flexibility in the Feasible BS.4.Wector . . . . . . . . . . . . . . . . . . 50
4 Outage Probability AnaIysis in a Hot-sector . . . . . . . . . . . . . . . . . . . 50
4.6 Simulation and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.6.1 Simulation Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.6.3 Performance of CB and CS Schernes with Sector Congestion Level . . . 89
4.6.3 Cornpanson of Received Eb/ l , Statistics with Four BSAA Schernes . . 90
4.6.4 Total Mobile Transmit Power . . . . . . . . . . . . . . . . . . . . . . . 92
4-63 Antenna Selection and Coverage Area . . . . . . . . . . . . . . . . . . . 94
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 Chnpter S u m r n a ~ 98
5 Distributed Inter-ce11 Interference Control 100
5 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.2 Systern Mode1 and ProbIem Description . . . . . . . . . . . . . . . . . . . . . . 103
. 5 2 . L Tirne-slottetl CDh.I.4 S-tern bfoclel . . . . . . . . . . . . . . . . . . . . 103
5.2.- inter-ce11 Intcrference 'rhimiiation Prol)lt!rri . . . . . . . . . . . . . . . IO4
. . . . . . . . . . . . . . . . . . . . 5.3 Tag Control and interference Management 106
. . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Tag Control and Scheduling 107
. . . . . . . . . . . . . . . . . . . . . . . . . . . 5.32 Resource Management 110
5.4 A Case Study: Achievable Data Rate in Traditional and Proposed Schemes . . 112
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.1 Need for Tagging 112
. . . . . . . . . . . . . . . . . . . . . . . . . . 5-42 Performance Cornparison 113
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Simulation and Results 115
. . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.1 Simulation Description 115
. . . . . . . . . . . . . . . . . 5.52 Location Independent Packet Scheduling 118
. . . . . . . . . . . . . . . . . . . . 5.5.3 Individual and System Throughput 120
. . . . . . . . . . . . . . . . . . . . . . . 5.5.4 Total Mobile Transmit Power 123
- - - . . . . . . . . . . . . . . . . . . . . . . . a . a Average Transmit Bit Energy 133
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6 Chapter Surnmary 125
6 Conclusions 127
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Thesis Surnrnal 127
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Thesis Contribution 130
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Future Research Directions 131
cii
List of Figures
1.1 Wireless and wireline access networks. . . . . . . . . . . . . . . . . . . . . . . 2
2.1 Reverse link interference mode1 in a cellular CDSIA nnetwork. . . . . . . . . . . 13
3.2 The bit error probability in a single-cell system. The ma-imurn number of
admissible voice users is 57 (minimum of the s-marked entries) with the SBRPC. 20
2.3 The bit error probability for wice users in ceIl L and cell 2 with and without
the NBRPC. !Vvl = IOt !VdL = 4 and :Va = 4. Cell 2 is made congested by
increasing the number of voice users to the mz~uirnum. . . . . . . . . . . . . . . 21
2.4 The bit enor probability for voice users in cell L and ce11 3 with and without
the NBRPC. Xvl = 54, iVdl = 4 and !Vd2 = 4. Yurnber of voice users in ce11 2 is
varied. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.5 (a) Total mobile transmit power dispensed in the system. (b) Ratio of receive
power level of voice users in ce11 2 to that of ce11 1 and ( c ) Ratio of inter-ce11
interference power to the total interrerence power. Total mobile transmit power
is mininiiitti when the reference receive power level ratio is 1. . . . . . . . . . . 25
2.6 Variation of interference power due to the integratioti of voice and data users. 29
2.7 Variation of interference power due to the integration of voice and data users
in sectors with tlifferent azimuthal angle. . . . . . . . . . . . . . . . . . . . . . 30
2.8 L$riatiori of interference power due to the integration of t-oice ancl data users
in cells ~ritti (Liffcrent radius. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1
2.9 The signal variation at the base station when only a single class of users is
present in a full-loaded single-ce11 s ~ t e m . Cal1 arriva1 process is Poisson-
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . distributed
The signal variation a t the base station when two classes of users are present
in a full-loaded single-ce11 system . Cali arrivals of different classes of users are
. . . . . . . . . . . . . . . . . . . . . . . . . . . . independent Poisson processes
The cornputation of the average mobile transmit power generated in a square
ce11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
-4n example of a user switching €rom one base station to another . . . . . . . .
Convergence of PC. PC+CC and RC+PC+CC algorithms . . . . . . . . . . . .
Total number of mobile switching that would occur during the power/rate ad-
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . justrnent
Achieved bit rates by al1 the users are added up and then averaged . . . . . . .
Average transmit bit energy required by different control algorithms . . . . . .
Total interference power at al1 the base stations with different interference con-
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . trol techniques
Normalized (by meanj standard deviation of total interference power at al1 the
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . base stations
Ce11 coverage with PC.only . Users are assigned to the closest (in distance) base
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . station
Ce11 coverage with PC+CC (Min Ts MP/M a.. Ru PP basecl): Csers are assigned
to a base station that requires minimum mobile transmit power . . . . . . . . .
3.10 Ce11 coverage tvith PC+CC-t-RC (Uin T s Eb/hias Rs PP basecl): Csers are
assignecl to a base station that recpires minimum transmit bit energy . . . . . 62
A mobile user i can potentially be covered by L antenna in the FOS-\-\. Antenna
"1" (denoted by solid line) receives signais frorn users mho rnay possibly be in
cornrniinication with L different (CO-located) antenna. . . . . . . . . . . . . . .
Spatial interference control techniques. Low data rate (LDR) and high data
rate (HDR) users CO-exist in the system. In (c) and (d). dashed and solid lines
denote the sectors of two different groups of non-overlapping sectors. Dotted
lines denote the shrinking of the cell/sector size. . . . . . . . . . . . . . . . . .
X regular cellular network model. Interference on the reverse Links is shown.
The center ce11 employs 3 sectors in (a) and 6 overlapping sectors in (b). The
. . . . . . . center ce11 experiences congestion in a sector as shown in Fig. 4.4.
-4 FOSA-4 [6,2] is irnplemented in the ceII, i.e., 6 sectors with 2-clegree of over-
Iapping: (a) a hot-sector has developed and (b) a hypotheticd base station
antenna assignment for mobile users. It can be seen that two groups of non-
overlapping sectors in the FOSAA [6!2] systeni. . . . . . . . . . . . . . . . . .
Outage probability for users in the hot-sector. . . . . . . . . . . . . . . . . . .
Hot-sector is location-wise randornized. . . . . . . . . . . . . . . . . . . . . . .
Complementary CDF of Eb/Io with sector congestion control parameter (p ) . A
(randornized) single hot-sector is included, . . . . . . . . . . . . . . . . . . . .
Complementary CDF of received Eb/Io for ail niobile iisers in iiniforrn angular
. . traffic in the center cell, ,û = O. No hot-sector is included in the simulation.
Complementary CDF of received Eb/to for al1 mobile users in non-~iriiforrn an-
guIar trafFic in the center ceII, $ = 2. A (randornized) single hoc-sector is
included. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4-10 Total mobile transmit powver in uniform angular traffic for a set of mobile user
Iocations. 3 = 0.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4-11 Total niotde transmit potver in nori-itniform ringtilar Traffic for a set OF tnobiIe
user locations. 3 = 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.12 The network coverage (for the CB-t-CS) with uniform angular traffic. = 0 .
. . . . . . . . . No hot-sector is included See Fig 4.24 for the center ce11 coverage 94
4.13 The network coverage (for the CB+CS) ivith non-uniforrn angular trafic: f i = 2 .
. . . . . . A single hot-sector is inciuded See Fig 4-15 for the center ceIl coverage 95
4.14 The center cell coverage with uniform angular traffic, P = O . 30 hot-sector is
included . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.1; The center ce11 coverage with non-uniform angular traffic? ,/3 = 2 . A single
hot-spot sector is included . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.1 Frame/slot structure of a CDbIh system . . . . . . . . . . . . . . . . . . . . . . 103
5 Minimizing the inter-ce11 interference. . . . . . . . . . . . . . . . . . . . . . . . 104
5.3 Fast scale signal variation in radio environment . Shadow fading is assumed to
be invariant for the duration of 120 dots . . . . . . . . . . . . . . . . . . . . . . 106
5.4 Tagging and scheduling . hi = totaI number of (hard and soft) tags issued by
each base station . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.5 An example of tagging with twvo base stations and two mobile users . . . . . . . 112
5.6 The average number of scheduled mobile users in each slot (from the simulation) . 117
. . . . . 3.7 Scheduled mobile users in distance and shadowing channel conditions 118
. . . . . 5.8 Scheduled mobile users in distance and shadowing channe1 conditions 119
5.9 Average (over 120 dots) achieved throughput . hl = 1 and M = 2 . . . . . . . . 121
5.10 Average (over 120 slots) achieveri throughput, hi = 5 and M = 10 . . . . . . . . 121
5.1 1 Histogram of the number of mobile users that achieved the average throughput
over the duration of the simulation ivhen M = I arid hi = 2 . . . . . . . . . . . 1'12
3.12 Histogram of the nimber of mobile users that achieved the average throughput
over the duration of the simulation when hl = 3 and hi = 10 . . . . . . . . . . 122
3.13 The mobile transmit power of active mobiIe tisers over the diiration of 1'20 dots
. . . . . . . . is siimmecl iip and t t i ~ t i aiveraged per slot w h i 5t = 1 and .LI = 2 123
3-14 The mobile transmit power of active mobile users owr the duration of 120 dots
is summed up and then averaged per slot when 51 = 3 and 34 = 10. . . . . . . 124
5.15 Average transmit bit energy dispensed in the systern over the simulation period. 125
List of Tables
2.2 The powver factors for different rate/service requirements when spread spectrum
bandwidth is 10 MHz. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 The bit error rate (xlO-j) for video users in ce11 1 and 2 with and without
implementing the NBRPC. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Performance measure rising fked and dynamic base station assignment. . . . . 36
3.1 Minimum ratio of total receive powver Q,/Q, at two base stations in order for
user i to switch from base station p to base station q. . . . . . . . . . . . . . . 49
-1.1 The number of antenna used in the simulation. See Section 4.2 for description
of L and S. . . . . . . . . . . . . . . . . . . + . . . . . . . . . - - - - . . . . . 87
-1.2 Total mobile transmit power for different BS.4.4 schemes followingl0 iterations.
Averaged over 12 hot-spot areas each mith 20 sets of mobile user locations. 3 = 2. 93
5.1 Transmission decision basetl on hard/soft tagging. . . . . . . . . . . . . . . . . 109
5 .2 Average tbroughput achieved over the simulation period. . . . . . . . . . . - * 120
3.3 Average mobile transmit power dispensed over the siniulation periocI. . . . . . 134
.5=l Average bit energy dissipatecl over the simulation periotl. . . . . . . . . . . . . 12.5
List of Abbreviations and Acronyms
1G/2G/3G
AiMPS
AoA
ATM
AWGN
BER
BS
BS A
BSAA
BSAA+PA
CAC
CB
CB+CS
CC
CDF
CDMA
CS
dB
DBSA
DCCH
FBSA
FDD
FDMA
FOSAA
GSM
lst/3nd/3rd generation
advanced mobile phone systems
angle-of-arriva1
asynchronous transmission mode
additive white Gaussian noise
bit error rate
base station
base station assignrnent
base station antenna assignment
base station antenna assignrnent pIus power
cal1 admission control
cell-breat hing
cell-breathing plus ceI1-sIicing
ce11 control
cumulative distribution function
code division multiple access
cell-slicing
decibel
dynamic base station antenna
dedicated control channel
f i e d base station antenna
frecpency-division cfuples
freclucncy division mu1 tiple acres
f i~ed overlapping sectoreci antrnna archicccture
global system for triobile corrlrriiiriications
HDR
IMT
ITU
LAN
LDR
MAI
MTP
MTSO
MU
NBRPA
PA
PC
PC+CC
PCH
PCS
PSTN
Qo s QPSK
RC
RP
RPA
RIPIC
R P L
RF
RX
SBRPA
high-date-rate
international mobile celecornmirnications
international telecommunication union
local area network
low-date-rate
multiple access interference
mobile transmit power
mobile telephone switching office
mobile user
network-balanced receive power allocation
power ailocation
power control
power plus ceIl control
physical channel
personal communication services
public svitched telephone network
quality of service
quadrature phase shift keying
rate control
receive power
receive power allocation
rate/power/cell
receive powr IeveL
radio frequency
receiver
source-basecl receiw poner alotxtiori
SDMA
Sm
SNR
SPC
SS
TCP/IP
TDD
TDM
TDMA
TF
TPC
TX
UMTS
WAP
WCDMA
space division mu1 tiple access
signal-to-interference ratio
signal-to-noise ratio
slow power control
spread spectrum
transmission control/internet protocol
tirne-division duplex
time-division multiplexing
time division multiple access
transmit power
transmit power control
transmitter
universal mobile telecommunication systems
wireless application protocol
wideband CDMA
List of Mat hematical Terms
system bandwidth
processing gain
SIR
SIR for '\.IL i
bit energy
interference power spectral density
Ebllo
required Eb/Io
Eb/I, for class c users
Eb/l, for ciass c users controlled by BS b
number of classes of users supportcd by the system
numbcr of BSs (or cell-sites) in the system
number of !dUs (or transmitters) in the system
nurnber of MUS in ce11 b
number of class c Mus in ce11 b
mean arriva1 rate of dass c users (Poisson process)
voice
data
length of a square ce11
bit rate
bit race of class c iisers
bit rate of 11L* i if it is to comniunicate to BS j
a service parameter For user z rneasured in Hz
transmit bit energ? of SiL i
cota1 crarisrtiit bit energ'.
svii
poiver factor for class c users
powr factor for class c users in ce11 b
radius of a circular ce11
angular beam midth of a sector
preferred RPL for class c users in ce11 b
preferred RPL a t ce11 b for a reference class users
home BS of MU i
home BS antenna of 4ILi i
distance between MC' i and BS j
reverse link gain between MC- i and BS j
fonvard link gain between MC i and BS j
fast-fading averaged out reverse link gain between M U i and BS j
slow-fading component between MU i and BS j
fast-fading component between MU i and BS j
user density in ce11 j
noise spectral density a t BS 1
total interference power at BS j
total interference power at BS b from class c users in ce11 j
total receive potver at BS j
average of the total rcceive potver at BSs in the nettvork
area averaged receive power at BS b from class c users in ce l l j
area averagecl receive p o t w at BS b from usecs in ceIl j
interference niütris of the systern
approsimated interference rnatris of the system
preferrecl RPL vertor of r~f~rt.ric:e cIass users in tlifferent cells
transmit power wctor of 1iC-s
svi i i
transmit power by MU i
transmit power by MU 2 if it is to comrnunicate with BS j
received power at BS j €rom MC i
received pilot power a t M U i frorn BS j
degree of overlapping
throughput achieved by user i
number of tags issued
average number of Mus scheduled when L 1 . I tags are issued
number of overlapping antenna that M U i can communicate to in FOSXX system
number of antenna a t a BS
number of sectors in a celI in a conventional non-overlapping antenna systern
angle between a reference line and the ceriter O € a sector
congestion control parameter in sectors
iteration index of an algorithrti
total number of iterations
frame length in tirne
slot length in time
total number of dots in a frarrie
number of MUS sctiedulecl in it dot
percentage of MUS scheduled in the network
a constant
probability of bit error
probability of outrige
pat h-loss exponent
dominant eigenvalue of the incr~rfcrencc macris
stantlarcl deviation
sis
Chapter 1
Introduction
The demand for wireless communication services continues to grow. Currently. voice and Iotv
bit-rate data services are provided in wireless networks. With the proliferation of the Internet
in our day-teday life, high bit-rate wireless chta applications are emerging. As a result?
wireless service providers are facing the challenge of deploying high-capacity wireless networks
rvithin the limited radio spectrum 11-51,
In tvireless networks, mobile terminals and base stations are connected via radio links
(channels). The communication from base stations to mobiles is called fonvard link com-
munication and from mobiles to base stations is called reverse link communication. These
communication channels are unreliable as they var- ivith time due to user movement and the
changing radio environment. The received signals in a radio environment can generally be
affected by two types of fading: shadorving and rnulti-path fading. In shadowing, the (slow)
signal variation is caused by contours or the terrain between trsers and the base stations. The
multi-path (fast) fading occurs when the signal arrives via different paths due to reflections of
the transmitted signal from the sirrrounding structures. Therefore, it is necessary to clevelop
appropriate channel access techniques to effectii-eIy hanclie the radio channels.
In cellular networks. the coverage area is divicted into several cells. each of which is servecl
t)y a base station as shown in Fig- 1.1. The base stations are trorinected t~picallx via wireline to
the mobile telephone switching center rvhich is the gateway to the wireline networks. The 1st
generation (1G) ntreless cellular systcms (e.g., -1k1PS) arc analog and provide voice services.
They are based on frequency d i ~ i s i o ~ multiple access (FD41.4). The second generation (X)
systems were developed to increase the voice capacity with improved cai i quality. They are
tligita1 and provide mainly voice services based on time division multipIe access (TDbIA)
(e.g., GSM) and code division multiple access (CDM-4) te-g., IS-95). The 1G and 2G systems
operate in circuit-sivitched mode where channels are reserwd and resources allocated for the
duration of the connection.
Figure 1.1: Wireless and wireline access networks.
The 3rd generation (3G) system are designed to provide higher data rates with better
quality. The packet-switched 3G networks can eficiently support integrated services. Le..
çarious services such as voice. emails. web-browsing and fax. al1 in a single estem. [n packet-
switched mode, channels and resources can be allocaced or de-allocated on a need basis rluring
a life of connecti~n. 3G systems are based on CDM.4 radio ziccess technology and two different
schemes have been adopted by the internationa1 telecommunication union. They are wicleband
CDhLk (WCDMA [6]) and time-division CD-IIA (TD-CDLI.4 [TI). The former d l be used
in freqiiency division duplesing (FDD) mode anci the latter in tirne division duplesing mode
(TDD). The FDD mode recpires separates frequency bands for €onvard and reverse links. The
T D D mode uses the same frequency band for both forw-ard and reverse links. but aiternates
the transniission direction in time.
CDMA Systems
CDMA is a multiple access scheme where ail the active mobile terminals transmit together ont0
the same frequency band. In CDMX systems. each terminal is assigned a distinct spreading
code. AI1 of these codes have noise-like characteristics with very small cross-correlation [dl.
Thc modulated signals are spread over a wider frequency band of several MHz (1.25 MHz in
IS-95 and 1.35/5/10/20 MHz in LVCDM-4. The signals can be separated at the receiver. for
example, by using a correlator that accepts only the signal energy from the desired user. The
undesired signals at the receivers contribute to the noise. in CDMA systems, al1 the available
channels can be used in evecy cell. Therefore, the main source of interference there is multiple
access interference (ILL41) and it comes from the users within the ce11 (intra-cell) and outside
of the ce11 (inter-cell). Hence. strict transmit power control is needed to control interference
in CDM.4 systems.
Some other features of CDM-4 technology are: sofi limit on capacity (i.e.. gracefiil per-
formance degradation), soft handoff (i.e., reduced link degradation during handoff). universal
frequency reuse (Le., no need for elaborate frequency planning). .Usa, the CDMA system
can combat multi-path fading using a Rake receiver since its bandwidth W is t-vpically much
higher than the coherence bandwidth of the channel in the outdoors. Therefore. it resoIves
multi-path components with time separation of at least l / l V [a] and then combines them to
niake an even stronger signal a t the receiver.
The WCDM.4 systems offer several improvernents over IS-93 systems. WCDMA systems
will implement coherent detection in the reverse link. fast potver control in both reverse and
hrward links and sophisticated interference control niechariisms using adaptive antenna [9-
131. Higher data rate services can be providecl titie to the increased bandwidth in FVCDhI-4
systerns. Siulti-rate transmission is made possible in \VCDbiA systems by using either multi
codes [l-1-17] or ri single code with variable spreatlirig gain control [15.19]. CVe consicier the
Iatter in oiir work.
1.2 Interference Cont rol
The CDM--\ system is interference-limited. That is, any reduction in the interference power can
directly translate into capacity increase 1201. A substantial amount of research [9.13.15,21-271
has been done to increase the system capacity by controlling the multiple access interference
for the reverse links. In all, the ultimate goal is to reduce the mobile transmit power as much
as possible while rnaintaining the desired signai-to-interference ratio (SIR) to al1 the admitted
users.
Transmit power control is a scheme by which the transmit power of the terminals are
controlled according to some criterion, usuaiIy the required quality of service. The near-far
problem [28] necessitates transmit power control in the reverse links. This type of power
control is esecuted 800 and 1500 commands per second in IS-95 and WCDM-A respectively.
Since each user is a potential source of interference to other users. any reduction in the transmit
power of a user also yields a reduction in the mdtiple access interference, mhich helps al1 the
other users, increasing the system capacity.
Receive power control [29-321 is also an important interference control mechanism. In the
reverse links, the desired receive power level for each user has to be set so that the transmit
power control can maintain it. In 2nd generation systerns where a single class of service
(Le., voice) mris predominantly provided, power level setting was simpIe in that al1 the users'
signals were set to be received at the same receive powr IeveIs. However. in 3rd generation
and future wireless sustems, due to the integration of tarious services. receive power levels will
also depend or1 the service requirernents- There. the receive power Ievel setting and adjusting
have to be done coritinuously depending on the network trafic loads as well as the service
requirements.
Another tcchniqiie to suppress the multiple access interference in CD.\[=\ systerns is to
isolate the signiils spatially [20.33.341. This is irnplemented rising ceIl sec-toring techniques
with tlirectio~iiil iuitcririas and hence avoicling somc of the interference cf iat wouId otherwise
prevail. Most of the IS-95 systerns rvere implemented with 3 sectors. The K'CDBL:\ systems
support the deploymetit of adaptivc antenna arrays [9] that can effectively direct riiills to the
interfering mobiles and hence controi the interference.
Temporal interference contro1 can aIso be implemented where transmit potver is controlled
depending on, for example, the instantaneous channel or traffic conditions. In packet-based
networks, the system tias the fiexibility to scheduIe only those packets that wilr muse the
minimum interference tu others. This involves additionai overhead due to the interference
management, however, it will increase the systern capacity.
Since CDh'IA systerns are interferencelimiteci, any means that reduce the transmit porver
(without deteriorating the service quality) wili increase the systern capacity. By rnaintaining a
radio Iink with minimum power when there is no activity at the transmitter helps in this regard.
This phenornenon is generally characterized by the activity factor which basically indicates
the percentage of time the tratismitter is active during a connection. In voice cornmimication,
this factor is typically 0.3 - 0.4. In our work, rive assume full transmission (i.e.. activity factor
of 1) ror al1 the sources. However~ with lower activity level. one can expect to increase the
systern capacity than with the full transmission case.
The single user receiver is the basic receiver design for direct spread CDMA system. It
is simple in complexi- but its performance suffers in the presence of large 41-41, Multiuser
receivers use knowledge of al1 the user specific pseudo random sequences to esploit the structure
of the multiple access interference. In multiuser cletectioii, the irsers are joiritly detected for
their mutual benefit iristead of being estimatecl as in single user detection. Therefore. for
those users whose spreacIing sequences are knotvn at the receiver (e-g. intra-cell iisrirs at the
base station receiver). rritrlti ttser tletection can be appIied - that c m in prinçiplc tilirriinate the
intra-ce11 interference, As a resuIt. the total interference is recluced a t the receiwr giving rise
to increasecl capacity. In orir work ive consiclet the singIe user receiver o t i 1 ~ CD1LI q-stems
consiriered for 3G and I ~ t y u n r i iricorporate some type of miilciuser detection crapiikilitic~ iri the
receiver structiirc.
1.3 Motivation and Objective
The telephone networks that were deployed many years ago were circuit-switched and opti-
mized mainly for voice communications. However, with the Internet, data traffic has surpassed
voice lately, forcing network vendors and operators to upgrade their networks or deploy sep-
arate networks. Then, the integration of multi-media services into a unified network was
envisioned as a cost-effective solution. This evolution of services, traffic and network is not
just limited to wireline networks, but rather, it has a parallel analogy in wireless networks as
Hence, we need to devise techniques that can handle integrated services efficiently in
CDMA networks in the next generation wireless systems.
The capacity analysis of the CDb1.A system has been done extensively, mostly for voice
users [9,13,22-251. The capacity trade-off studies between different types of users (e.g.. data
and voice) have also been investigated recently in [15,26.35-381. However. application-specific
characteristics (e.g.? delay tolerance) have not been fully exploited from the point of view of
minimizing the interference to irnprove the system performance.
The main objective of this thesis is to develop new channel access techniques to improve the
system performance through spatial and temporal interference control for the reverse links of
a CDMA packet network. The mobile transmit power is the rnost important radio resource. it
is essential that mobile terminals use the lowest possible amount of transmit power to achieve
a certain quality of service. This estends the mobile terminals' batte- life and reduces the
interference seen by other iisers - which increases the system capacity. The adaptivc concrol of
the transmit power is a major method of interference management in wireless comrniinication
systems. In our work. ive develop schemes and algorithms that inteltigently ailocate ancl
control the power-resource in wireless networks.
The 3rd generation wireless systems provide packet data sert-ices. In a packet data system.
information is separatecl into discrete packets that can be inclependently transmitted ancf re-
assembled iit the receii-ing end. As long as the iriformation is not tirne sensitive- the systetn ciin
transmit packets whenever it wants to. The various wireless data applications (e-g., file transfer
and web browsing) that are packet-oriented, delay-insensitive and throughput-demanding are
emerging with the wireless Internet. Poice calls require constant bit rates and are del--
sensitive whereas many data connections require variable bit rates and, are error-sensitive
and delay-tolerant. In the latter applications. ive have the flexibility of delaying the packets
if it helps achieve better system performance. tVe exploit the source characteristics of data
applications to adaptively adjust the transmission rates to reduce the total mobile transmit
power. tVe also consider a packet-scheduling mechanism that minimizes the interference by
exploiting the channel conditions and delaying the dominant inter-cell interferers. Spatial
signai isolation is considered using Kued overIapping sectors to distribute the interference
among sectored antenna and thus irnproving the system performance.
Thesis Organization
in Chapter 2, we consider the receive p o w r control problem in the reverse links of a CDhI.4
network that supports users with rnulti rate/service requirements. There. the receive power
levels are determined for a user relative to those of others. The interference power generated
by users with various service requirements is analyzed assuming that the signais from mobile
terminals are always received at preferred power levels. Then, we investigate the interference-
balancingl as a means to improve the sy t em performance.
In Chapter 3. we focus our research on eficiently supporting del-tolerant tvireless data
applications (e-g. emails) in cellular networks where higher spatial and temporal interference
variations at base station antenna are espettecl. -4 nover combined rate, power and ceIl
(R/P/C) control scheme is proposeti aritl studied. Two algorithms. one tiirectly minimizing
'By interference-balancing, we mean thar the totd interference power at each antenna is approsimately
equal. If ive assume that total interference priu-er and total receive potver are neariy equai. then each base
station operates tk-ich qua1 receive power.
the average transmit bit energy and the other indirectly using measurecl pilot power. arc
discussed. Simulation results are also presentcd.
In Chapter 4, the problem of base station antenna assignment with minimum average rno-
biIe transmit power is studied for non-orthogonal reverse links of the CDi\ll=\ systems with
h e d overlapping sectored antenna architecture (FOS-CL4). It is first noted that the traditional
sectored cellular system has limitations in switching users between in-ce11 sectors and also
out-of-ce11 sectors in moderatel-loaded ne tworks. It is then shown t hat by employing over-
lapping sectors in FOS-LA, we can expIoit the Aexibility of assigning a user to one of possibly
many potential antenna to effectively support the non-uniform angular traffic. The process
of dynamic ce11 sectoring to control the interfereuce is differentiated two-fold as cell-breathing
and cell-slicing and the latter can be viewed as being the angular counterpart of the former
radial scheme. The hybrid scheme, cell-breathing plus cell-slicing, is shown to yield the o p
timal solution in terms of minimum average mobiIe transmit power in a FOSX.4 system. A
simulation study is also done to demonstrate the effectiveness and the flcxibility of the FOSX.4
in handling non-uniform angular loads.
In Chapter 5 , a novel channel access scheme based on the relative real-time channel condi-
tions and the dominant interferer avoidance. is proposed and studied in a time-slotted CD.Vl.4
systern. The scheme is implemented via distribution of tags that indicate the relative channel
gains with the base stations. It is shomn that the system as a whole benefits when impie-
rnenting the proposed scheduling scheme since those allowed to transmit get increased data
rates because of the reduced interference. Tag management is also discussed for practicril
impIementation. Soft-tagging offers flesibiiity in allowing mobile terminais that cause mod-
erate inter-ce11 interference to transmit uricler certain conditions sirch as longer starved Lime.
Simulation results are also presented.
Finally. Chapter 6 suinmarizes the n-ork ancl cliscusses some of the implemencation issues
that have to be addressetl before irnplcrn~nting tlie proposed schemes.
Chapter 2
Receive Power Control
In this chapter, we analyze the effect of the interference power variation among the base
stations on the reverse link performance of a CDSi.4 cellular system that supports integrated
services (e-g., voice and video).
2.1 Introduction
The interference management in CDM.4 systems incliides (fast) transmit power control (TPC)
and (slow) receive power control (RPC). In transmit power control: mobile transmit power is
regulated according to some criterion, for example. to rnaintain a required signal to interference
ratio (SIR). In receive potver control. the receive power levels (RPLs) are determinetl for
every user to satisfy the required servicc quality. Once the receive power control niechanism
sets the preferred receivc power levels. transmit power control attempts to maintain it. The
performance of a CDhI.4 system is interference-limited. hence. tightiy coupled with both
transmit power control and receive potvcr control. \\è assume perfect transmit power control
and study the receive powr control probleni in ttiis chapter.
CVe first determine the receive power level for ii mobile user ac the home hase station
relative to those of other itsers. LVe take a snapsfior of the network and cleterniine the receive
power levels by solving a split-optimization problern; one simply solving a systern of equiitions
and the other solving an eigenvalue problern. The former sets the receive power levels for the
users relative to those of in-ce11 users. which ive call source-based receïue power control and is
dependent on source characteristics such as bit rate and bit error rate. The latter sets the
receive power levels relative to those of out-of-ceII users: which we call network-balanced receive
power control and is dependent on ~etwork characteristics such as traffic conditions. We also
analyze the interference power variation with time at a base station due to the integration of
different classes of services assuming that the signals are always received at preferred power
levels. It is shown that the supporting of high/low data rate or high/low quality of service
recpirenients contributes to the increased signal variation at the base stations. Finallt-. it is
shown that the higher the interference porver variation among the base stations is. the higher
the total mobile transmit power will be. Hence, the need for interference-balancing among the
base stations is ernphasized to rninimize the total mobile transmit power.
The rest of the chapter is organized as follows: In the riext section, the receive power
levels are determined for different classes of users in a cellular systern with Lied assignment
of users to base stations. The effectiveness of the source-based and network-balanced receive
power control are illustrated using numerical examples in a single-ce11 and rnulti-ce11 system
respectivel. The performance irnprovernent via interference-balancing is also ernphasized
there. In Section 2.3: the interference power variation due to the integration of services
within the single systern is analyzed. This is follorved bu the cliscussion on the performance
improvernent using the dynamic base station assignmerit in Section 2.4. Finallq. Section 2.5
sunimarizes the chapter.
Receive Power Control (RPC) Problem and Solu-
t ion
In a system mhere a single class of service (i.e., voice) was mainly provided, receive power
control was simple in that al1 the users' signals were set to be received at the same power
levels. However: in the future systems with the integration of multiple services into a single
system, these receive porver Ievels will be different as will be discussed in this section.
The preferred receive power levels are set differently for different users based on sorne cri-
teria. For example, high data rate users are received a t a higher level than lower data rate
users. The base station (BS) sets the preferred receive porver levels, in CDMA systems with
homogeneous traffic based on the therma1 noise considerations (in single-ce11 systems [?O]) or
netivork Ioad conditions (in cellular systems [SI). Hoivever. in CDSI.4 systerns with heteroge-
neous traffic, the service requirements of multiple classes of users have also to be taken into
account in setting these power levels.
The previous work on receive power control for multi-class users [18,?7,39,4OI focused
mainiy on single-cell systems and claimed that, by incorporating the inter-ce11 interference
coefficient into their madels, celIular systerns can be similarly studied. However, with the
integration of services, different users contribute different1y to the multiple access interference:
therefore, the receive power concro~ problern has to be studied in detail for rnulti-ceIl systems.
[n such a sustem. power management schemes not onIy niitiga~e the near-far probiem [281
and take inter-ce11 interference into account but also balance the natuta1 differences between
various classes of services.
LVe motinte our investigation from the foIlort-ing facts:
L. in a miiIti-celI system wich a singIe class of trsers. cells witti higher toad need to operate
at higher reçeiw power levels than those with lowr ioarls [SI and
2. in a siriglc-ceII systtm with rnulti-class of irsers. ttsers with tiigher r;itr/sen~icc rcquire-
ments need to be albcated higher power than those with lower requirements so that each
user receives the requirecl service by properly balancing the intra-cell interference 1181.
The concept of SIR-balancing that yields a fair distribution of the interfere~ice power in the
sense that al1 the users experience equal SIR Ievel has been studied estensively for a single
class of users. In [dl], the problem of SIR-balanced transmit power control rvith the single class
of users in spread spectrum systems (vas first presented in rnatris notation, then transrorrned
into an eigenvalue problem, and solved anaiytically using matris algebra. CVe show that
in a cellular CDMA system with multi-ratejservice users, if proper receive powr contrd is
implemented within and between different cIasses of users in each ceIl with the systcrn-wide
class-wise SIR-balancing, the receive power control is also an eigenvalue problem.
2.2.1 Cellular System Mode1
We study the reverse links of a cellular CDMX systern. The mobile users are classified ac-
cording to their data rate (r) and Eh/& (y) requirements where Eb and 1, are bit energy and
noise power respectively. For esample: class c users require Eb/Io of */, and have a bit rate of
r,. For simplicity, we assume t hat the mobiIes always transmit at t heir required source rates,
i.e.. the transmit bit rates are assumed to be constant during the connections. Hoivever. the
received Eb/ I , varies depending on the Ioad conditions (and hence interference) in the ceIis.
Let C, B and :V be the nrimber of cIasses, base stations and mobile users (SIUS) in the systern
respectively. The base stations and classes are numbered I ... B and 1 ... C respecti\~ely.
Hence? N = rFL -VI. where .VI is the number of usen in the cell j and -VI = XE=, ?ik,. ivhere
-Lk, is the number of clcas k users in cell j. The communication quaiit- is determincd by the
received Eb/Io which is given for CDM-4 systems by TH. where r is the signai-CO-noise ratio
(SIR) and H is the pcocessing gain. TherrnaI noise is not included in the analysis since we
nssrimc a large capacity CDSIA syscern. in which multiple iiccess interferenc~ is the tloniinant
source of interference. Total spreatling bandwiclttl is II-.
Each base station employs perfect (fast) transmit power control to its users such that the
receive power Ievels a t its antenna are equal for al1 the users of the same ciass. So soft handooff
is implemented. We also assume that users are not moving and transrnitters arc always active-
Al1 the users of the same class are treated equally witliin the system in terms of received Eb/I ,
(with ciass-wise system-wide SIR-balancing).
Figure 2.1: Reverse link interference mode1 in a cellular CDMA network.
Fig. 2.1 shows a cellular network with B base stations. Let Pkj be the receive power level
at home base station j for class k users. The quantity QkjYb is the total received powver h m
al1 the class k users in ce11 j? at base station 6 if the users adjust their power so that they are
received with unity powver at their respective home base stations. Here, we do a many (power
sources in a ceIl)-to-one mapping, Le.: the receive power from a11 the tisers of class k in ceil
j , onto ce11 b is derioted by a scalar, Qk1,b- Sote that Qkb.b = :Vkb,VEL. b. Therefore. the total
receive power at base station 6, Qb = CI xkQkj,bPkj- The total interference power for a class
c user communicating mith a base station 6: equals to (Qb - Pcb)-
The receive power control problern can be stated as folloms: Given that the base station b
supportsL !VCb users with delivered2 Eb/ ï , of -fcb and bit rate of rc,Vcl 6 . fintl the receive power
vector Pb = [Pib ..- PCbj,Vb, such that the total3 receive power. El Et Pb? is minimized.
CVe work with the receive power in this chapter, hoivever, the mobile transmit power is
important in practice. Mobile transmit power not only contributes to the interference but
equally importantly determines the battery life in the reverse link communication. Hence, the
reduction of mobile transmit power is an absolute requirement. CVe first show that the receive
power control mechanism minirnizes the surn of the receive power at the base stations. Then,
Ive argue that this will translate into minimizing the average mobile transmit power in the
system.
The Eb/I, for class c users (with the transmission rate of r,) controlled by base station b
can be written as:
Practically, Qb is a random quantity that depends on the traffic: source activity levels and
time-varying propagation conditions: hence, y,b is a random variable. Hoivever. we do a s n a p
shot analysis here, Le.. the netaork is assurned to be static a t an instance and then derive the
receive power levels for different users. In the foilowing subsections, Ive first relate the receive
power levels of al1 the classes to that of a reference cIass in each ce11 and then detemine the
relationship between receive power levels of the reference classes in different cells.
'Yote that we are considering the fixed asignment of users to the base stations here. Therefore. the number
of users is known and fkcd at least for the duration of this power-Ievel setting consideration.
%ote that cal1 admission control rvill determine Ncb1Qc7 b and hence is determined accordingl-
3.\ûte that if ICI = Jibr Le.. each mer forms a cIass in cell 6, then xk P k b is the total receive potver from
al1 the users in ce11 6 .
2.2.2 Source-based RPC (SBRPC)
Eq- (2.1) can be re-written as:
Let us consider a base station (b') and its receive power levels (Pcb., Vcj a t an instance. Eq.
(2.2) has to be satisfied for al1 c a t b'. Since the left-side of (2.2) is the same Qc a t base station
b', let it be a constant K. The above argument is valid for any base station 6: therefore.
That ist in ce11 b' with C nurnber of classes of users,
lf rate/service requirements are known for a class of users. we can cornpute the receive power
level relative to that of a refererice class using (2.3). Let this reference class be denoteci as c'.
Therefore, For any class c in any ce11 6 ,
where
Xote that Iceb = l.Vb. Therefore. in any ce11 b. receive power vector for different classes is
given by
In (2.6). ttie receive powr levels are set basecl on source reqriirernencs such as bit rate and
bit error rate; hence. rfiis rriechanism is calIec1 soiircc>-hasetl receive power control ISBRPC).
1-5
It is applicable regardless of the inter-ce11 interference (or eqriivalently the nimber of users of
anj* class in other cells) since the inter-ce11 interference affects al1 the intra-ce11 iisers equaIIy
at the antenna. However, we still need to knotv the receive power level of the refercrice class
(Pc-b) a t each base station in order to determine the actual receive power.
Since the receive potver levels (and hence the mobile transmit potver") are increased by
fcb relative to the reference class? we cal1 fd the power factor mhich can be approxirnated
(assurning that the contribution by any user to the total interference power is minimal) for
moderate rate/service requirements asl
Y&c fcb = -
7c.b Tc-
1 Ea/fo(r) 1 data rates ( r ) = 9600 x 1
Table 2.1: The power factors for different ratelservice requirernents when spread spectrum
bandtvidth is 10 MHz.
Table 2.1 shows power factors for different rate/service requirements that are relative to
the reference class users (i.e.. voice services with Ï,. = 5 dB and rc- = 9.6 Kbps). it can be
seen that these receive power IeveIs can be as high as 250 tirnes that of the voice services,
for reliable data transmission rate of 1.92 SIbps. Hence, the receive power levels can v a n
significantly chie to the tiorninance of fetv high ser?-ice/ratc iisers.
'In the Appendiz A. ive (iiscias the rei;itionship berween the receire power and ttic average transmit power.
It can be seen from (2.4) that the receive power controI involves optimizirig t w variables
fcb and Pc-b in order to determine Pcb:Vc. 6. 'Ve have seen how to compute fcb- Xext, ive
consider optirnizing PcSb in a cellular network.
2.2.3 Network-balanced RPC (NBRPC)
Since the receive power levels of al1 the classes cân be specified in terms of the reference class
in each ce11 (see (2.4)). we mork with only the receive power levels of the reference clasç (Le..
Pc.b,Qb) in the sequel. CVe ce-mite (2.1) for class cg users in ceIl b as:
I V -(c.b = - pc- b
B : 'db. TC. El=, Qkj,bPkj - Pc.6
Since Pkj = fkjPc-,: (2.8) can be writtcn as:
The solution in terms of Pc-b '~ to (2.9) requires solvinga set of B algebraic equations. Nith the
system-wide class-wise SIR-baiancing, Le.. = -yc (and hence fcb = f,).Vc. 6. (1.9) becornes
where 1 is the interference niatrix tvhose elerrients are defined as Iij = 4,,, and Pc- is the receive
power vector for the reference c l a s i~scrs at B base stations, Le.. P,- = [Pc-[ ... PceB]-
The existence of unique positive solutions to (2.11) in terms of ceceive power vector. Pr-.
is guarrinteed by the Perron-Frobenius tIico- of stochastic
the receivc puwr level of the referenrtt ('Ixss tic al1 the base
from the suliition to ('2.1 1) and t t i ~ r iwiw power Lcvels of
17
matrices [QI. \Ve now know
statioris. Le.. [Pc-[ ...
al1 the rli~sses in ail the cclls.
Le., VbtPb(= If lb - - - fCb]TPc.b), from (2.6) . Since the receive power is determined by the
network conditions with SIR-balancing in the system? \ire cal1 this procedure network-balanced
receive power control (NBRPC).
If ive assume that for a large number of users uniformly distributed over the cell, then wve
can simplify Qkj,b as, Qkj,b = .$kj,bNkj7b'ktjT b; j # b, where ?,bkj,b is the total receive powver of
class k users from cell j on ceIl b averaged (by the number of users) over the entire ce11 area.
Therefore, 1 can be simplified as 1'. It shouId be notecf that lJlkb,b = 1,Vb.
In the receive potver control, the intra-ceII interference between different classes of users
and the inter-ce11 interference between different cells are controlled respectively by the SBRPC
and the NBRPC to provide the required services to al1 the admitted users. It can be seen that
the receive power control mechanism firstly determines the receive power to different cells and
then to different classes. Once the receive power level is determined for a user, the channel
loss determines the mobile transmit power.
In [43,441, the transmit poiver problem nas solvecl for the transmit power and it tva
s h o w that the solution to the eigenvdue problem will be minima1 in the total mobile transmit
power in the system. In our ivork, we considerccl the receive power and by similarity with the
transmit power problem, ive shoived that the siim of the receive power will be minimum. In
the Appendis At we relate the receive power and average total transmit power to show that
minirriizing the receive power also rninimizcs the average total mobile transmit power.
2.2.4 Capacity Evaluation in a Dual-Class System with Perfect
W C
In this section, we analyze the performance of the SBRPC and the NBRPC in terms of admis-
sible number of mobiles using a niimerical exampie. The general system mode1 is described
in Section 2.2.1. Mobile users are assumed to be randomly uniformly distributed within each
cell. The signal fades according to 1-" law! where 1 is the distance From the transmitter to the
receiver and a is taken to be 4. Lsers are assumed ta be in communication with a base station
that is the closest in the distance sense (ie., no shadowing). The system capacity is defined
as the total number of users that can be supported with the required quality of service.
For illustrative purposes, let us consider two classes of users, voice (class v) and low-
resolution video (class d). The communication qua1ity requirements specified in terms of bit
error rate, are P,,, = 10-3 (with 7, = 9.5) and = IO-' (with = 15.18). The speech bit
rate is 9.6 Kbps and coded low resohtion video has 64 Kbps [451. Let the reference ciass be
v. Therefore, f, = 1 and fd = L1.53 from (2.5). The spreading bandwidth. W = 10 MHz.
For the computation of the bit error rate, Gaussian approximation is used for the multiple
access interference and the bit error probability is calculated using P. = ferfc( Jz). where erfc(.) is the standard complernentary error function. Though the above bit error
probability function is for coherent binary FSK modulation scheme. we do not assume any
specific modulation schemes in out analysis. The above function for the bit error probability
wfi corisidered t'or the purpose of numerical iIlustration.
SBRPC in a Single-Ceil Svstem
\\ firsc consider a single-ce11 sptern trith and without iniplernenting the SBRPC. The number
of vitlm lisers is fised a t 4 and the maximum number of voice tisers that satisfies the bit error
riitr r~*cpirement is tleterrnirictl. Ttir receiw potver ItvcIs are set according to (2.6) wheri
ir~ipltwit!riting the SBRPC ancl cqirall!- set for al1 the uscrs ivlitm SBRPC is not implementrtl.
Figure 2.2: The bit error probability in a single-ce11 system. The maximum number of admis-
sible voice users is 37 (minimum of the x-marked entries) with the SBRPC.
Fig. 3.2 shows the bit error rate tvith and without the SBRPC for both voice and video
users. Without the receive power control, voice users receive better service while video users
receive poor service. In fact, tve cannot admit even a single mice user with 4 satisfied video
users without implementing the SBRPC. With the SBRPC. voicc iiscrs are affected since we
are basically allocating the power levels in proportion to their service requirement. The bit
error rate curve is shifted upwards (worsening) for voice users and downwards (improving) for
video uscrs. As a resultt the maximum number of voice users that can be supported increases
to approsimately .Si with 4 satisfied video users. it can be verified that by the cal1 admission
control which is described in the Appendis BI the above susterri will allow 57 voice users to
co-esist mith 4 vicleo users.
We have seen that proper SBRPC is neecfed to increase the capacity baIancing the mutual
intra-crll interf~rencc- In the following. we analyze a two-ce11 sustent ;ssiirning that the SBRPC
is implcrntlntetl in each cell. That is. ne set Pdb = 11-53 x in bath cells (6 = 1.2) and
Eh/[ , of vicieo tisers is related to that of voice users via.
"Id = fd w- = 18.15, in both cells.
- / d u + wf"f,)-L - I d )
NBRPC in a Two-Ce11 Svstem
We sirnulate a congested ce11 scenario by increasing the load of a ce11 as rnuch as possible. For
simplicity, an equal number of video users (4) is admitted in both cells and the number of
voice users in ce11 1 is 10. Hence, ceIl 1 is made lightly-loaded and ceIl 2 is the congested ceI1.
The mauinium number of admissible voice users in cell 2 is cletermined while achieving the bit
error rate requirements for al1 the users in the system.
without NBRP with NBRPC
1 1 t I
50 52 54 56 58 60 62 64 66 68 70 Number of Voice Usen in Cell 2 (Nwp)
Figure 2.3: The tiit error probability for voice irsers in cd1 1 ancl ce11 2 with and mithout the
SBRPC. lL*uI = IO. iVdI = 4 and !Vd2 = 4- Ce11 2 is made congested bu inçreasing the nuniber
of voice uscrs co the rnx~irnum,
We asstirne for sirriplicity. ark,.b = i~' .Vk. j. 6. and j # b. -4 two-ce11 susteni with sqiiare relis
is considered in the investigation. It can be shown using a simple integratian that th = 0.09
for the mode1 described in subsection 2.2-1 with square cells. Hence, r 1
If ive assume system-mide cIass-!vise SIR-balancing, receive power levels can be obtained by
solving the eigenvalue problem given by (2.1L) with I = 1'. The bit error rate for voice users
in both cells are shown in Fig. 2.3 versus the number of voice users in the congested cell. Et
c m be seen that the XBRPC helps increase the number of voice users in the congested ce11 to
63 (minimum of the O-marked entries) from 59 without the NBWC (which is x-marked).
Table 2.2 shows the received bit error rate for video users in each cell with different number
of voice users. With the NBRPC? al1 video users receive the required service {of Pc.d = IO-')
while 63 voice users are active. Tt can also be noted that without implementing the SBRPC.
the cells with 1ow loads (in our case, cell 1) get better performance. For example. we notice
in Fig. 2.3 that voice users in cet1 1 receive much better service than those in ce11 2 in the
absence of the NBRPC.
ive can also note that the SBRPC ensures that video users receive their services in pro-
portion to what voice users receive in terms of bit error rate in each ce11 since they are related
via (2.22). In effect. the SBRPC is Iocaily implernented in each cell witithout any knomledge of
other cells and the XBRPC is done at the network level mith the knowleclge of al1 the cells-
Homevert as will be donc in Chapter 3 it is desirable! from the point of t iew of minimizing
the total mobile transmit power? to combine the SB RPC and XBRPC to interference-balance
the system. The teceive power IeveI ratio (PU2/P, , ) of voice users arc 4-78. 3-38 and 5.8'7 for
Nu, = -50, 60 ancl 70 respcctiveiy- Xote that the receive power Ievels olvitfco users also change
sirnilar[y as reIatecl by (2.6). It is general1y true cliat the smaller the receiuc power Ieveist the
smaller the niobile trarismit porver. Therefore. it can be seen that the congestccl cell capacity
is increased iic tlie espense of the mobile transmit poiver duc to the tliffcrt~ncc in the receive
power levels.
1 Cell l(w/o XBRPC) 11 0.0003 1 0.0004 1 0.0005 1 0.0007 1 1 Cell l(withNBRPC) 11 0.23 1 O.j0 1 0.78 1 1.38 1 1 Cell 2(w/o NBRPC) 1) 0.37 1 1.05 1 2.51 1 5.19 1 1 Ce11 2(with NBRPC) 1) 0.23 1 0.50 1 0.78 1 1.38 1
Table 2.2: The bit error rate ( x IO-') for video users in ce11 I and 2 with and without irnple-
menting the NBRPC.
2.2.5 A Case S t udy : Fully-Loaded S ystem and Int erference-Balancing
We now consider a congested system by Ioading both cells to their lirnits. Note that ive were
abIe to admit 57 voice users with 4 video users in a single-ceIl systern. However. with the
sarne nurnber of video users. the number of admissible voice users has to be reduceri due to
inter-ce11 interference in rz muIti-ce11 system. We cornputed this number to be 5.1 voice users for
the two-ce11 systern iising the cal1 admission control described in the Appendis B. Therefore,
we have NUL = 54, Ndl = 4 and Nd = 4. Hence, ce11 1 is near-congested and me want to
cornpute the rnavirnurn possible nurnber of voice users in ce11 2. This nurnber is found to be
54 as shown in Fig. 2.4. This is true since this configuration has symrnetric interference levels.
generated by an equal nimber of video users and voice users in each cell. Whether ive have
synmetric or non-symrnetric user distribution level. performance depencts funclameneally on
the total mutual interference between cells. Fig. 2.; is obtained for the user configuration
jiist discussed above, Tt. can be seen frorn the figure that the total rnobik transmit potver is
minimuni when the cell loads (or in tact mutual inter-ce11 interference) are balancecl. i.e.. an
equal number of users esists in boch ceiIs - which irnpIies in our nioriel an eqiral aniount of
interference at the base stations. At this point. the reference receive p o w r levels iirp equal in
cadi ceil (PL.?/ Pcl = 1)- \Ve n i I l stucly interfer~ncr-balancing in cletail lacer iri Stv-tiori 2.4.
Figure 2.4: The bit error probability for voice users in ceH 1 and cell 2 with and without the
NBRPC. Nui = 34, Ndl = 4 and -Va = 4. Xumber of voice users in cell 2 is varied.
We sornetimes used the term load-balancing in a Ioose sense. Load-balancing is a special
case of interference-bdancing. In a systern with homogeneous users, this is equivalent CO
balancing the nurnber of users in the cells; hence called load-balancing. In a system tvith
heterogeneous users: ive interference-balance the system - this gives an additional dimension
in the cal1 admission control since we have flesibility with the nurnber of users of various classes
whose source and sen-ice requirements can be esploited or tradd-off to improve the system
performance, In a practicaI scenario: cells are not evenly loaded and we want to muimize the
capacity while minirnizing the mobile transmit power. in the nest section, ive analyze how
the interference at the base stations varies due to the support of niulti-rate/service tisers and
also due to the srnaller cells in future wireless networks.
Figure 3.3: (a) Total mobile transmit power tlispensed in the system, (b) Ratio of receive
power level of voice users in ceII 3 to that of cell I and (c) Ratio of inter-ce11 interference
power to the total interference power. Total mobile transmit ponrer is minimum mhen the
reference receive power level ratio is 1.
2.3 Interference Power in Integrated Services Wireless
Networks
in the last section, we analytically derived the receive power levels for every user in the system.
There, the mutual interference between users of the same ce11 and different cells was balanced
to provide the required services to a11 the admitted users. In this section, assumirig that the
signals are always received a t the preferred IeveIs. ive analyze the impact that the integration
of multi-ratelservice users within a single system has on the interference power at the base
stations. In the reverse links of a CDM.4 system, the interference is generally generated by a
large number of sources. This beneficially influences the distribution of the interference porver
in that the variance of the power is reduced. Hoivever. in the evolving cellular systems with
the integration of various services (e-g., voice, data and video), we increase the variation in
the interference power as notecl in chis section.
With the increasing demand for wireless services, cells shrink in size radially (in the form of
cell-splitting) and azimuthally (in the form of cell-sectoring) to effectively combat the multiple
access interference and thus to support more users- It was noted in [461 that srnaller cells
increase the traffic non-uniformi~y. In our work, ive show that the supporting of users with
rnulti-ratelservice requirements causes the interference power at the base stations to va- even
with perfect receive power controi and no ce11 shrinking. This suggests that in future wireless
systerns, higher variation in the interference power can be expected causing the delivered
service to vary
2.3.1 Interference Power Variation Anaiysis
.4 simple mode1 is assunieci in the folIowing investigatiori. CVe consider a systern with a singie
circi~lar cell of unit radius. It is assumed that the class k c-alIs are generatecl randornly accorcling
CO t h Poisson clistribiition within the ceil with the rrieari rate of r r r k . Vk. The validity of t Iir
Poisson mode1 in the wireless systems is out of the scope of this work. However, it simplifies
our investigation while shedding somc light into the expccted behavior in real systems.
Since we have a single-ce11 system, we drop the ce11 index j from the number of k class
users (Nk j ) and mi t e it as Nk. The probability that the number of class k users generated in
an area of A equals N is given by,
with both the mean and the variance equal to mkA. Let us compute a statistical parameter
STDI which refers to the non-uniformity of the interference power at the base stations. This
parameter is defined as the standard deviation of the interference power divided by the mean
of the interference powver at the base stations.
The signals from different classes of iisers are received at different levels as disciissed in
Section 2.2. The random quantity Q: the total instantaneous receive (or interference) potver
at the base station, with no inter-cell interference (Le., single cell) and neglecting thermal
noisel is Q(. ) = Ck NkPk, where Pk is the receive potver level of class k. With perfect receive
power control, Pk = JkP, where P is the receive power level of the reference class and fk is
the power factor of class k.
Therefore, the mean of Q, E[Q] is given b_v,
And:
The cal1 arrival processes between different classes are assurned to be independent. Therefore.
with the Poisson arrival process.
ru, rn, . if i # j . E [.V,:V, =
r + m . c it hprwise.
Hence,
Using (2.13) it can be shown that,
Therefore, the variance of Q is given by7
Hence,
Sext we analyze the effect on cell-shrinking on the interference power variation.
With Ce11 Sectoring
Cell-sectoring has been user1 in 1s-95 type systems with 3 sectors/cell and in future wireless
systems, it can be implemented wittr an adaptive antenna array. It can be shown that. in a
sector of angular width of 9 (within a ceIl of unit radius),
€rom (2.15), one can see that the srnaller the angular width of a sector, the higher the variance
of the interference power.
With Ce11 Splitting
Cells are split a s the user population increases. .4ssuming for simplicity that they are made
srnalier circular cells tvith base stations at the ceriter. it can be shown that
where R is the ce11 radius of the split-ceIIs. Frnni (2.19). une can see that the smaller the ceIl
sizes. the higher the variance of the interferenre potver.
2.3.2 Numerical Results
Let us consider twvo types of users, voice (v) and data [ d ) , for simplicit. The receive power
levels are P, and Pd for voice and data services respectively. Hence,
Q(.) = lV"P, + !VdPd.
Therefore, for a particular ce11 radius and sectoring, from (2.17).
where K is a constant. The power factors are normalized relative to the voice calls (with bit
error rate = and rate = 9.6 Kbps); therefore, f, = 1.
R = 1 . e = m
0.12 2 3 4 5 6 7 8 9 10
Power Factor. t-d
Figure '7.6: i~ariation of interference power due to the integration of voice and data users.
While keeping the number of average voice caIIs fked. the pomr factor (Le., rehtive
service/rate reqiiirernents) is varieci from 1 (hornogeneoi~s type) to 10 (heterogeneous type).
Three (lifferetit i1rriva1 rates (nid = 2.5.10) for data caIIs are consitlered witli m, fisetl at 50.
L\ïthoiit l o s of getieratit- let K = I. Fig. '2.6 shows the (riorriializetl) standard deviacion
29
Figure 2.7: Variation of interference power due to the integration of micc and data users in
sectors with different azirnuthal angle.
of interference potver with power factors. When fd is changed from 1 to 10. the STD, is
increased by about 50% from about 0.14 to 0.22,
While keeping the number of average voice and data calls tked at 50 and .5 respectivel_v?
the power factor wu varied for different ce11 sizes. Figs. 2.7 and 2.8 show the measure of
the interference power variation when the cells are sectorized and split respectively. in the
ceIl-sectoring case. the STDI increases by about TT% from 0.18 CO 0.32. when fd is increased
froni 1 ta 10 for 19 = ïî/6 (i.e., when the ce11 is divided into 12 sectors). In the çelI-splitting
case. the STD, iricrcases by about 67% from 0.54 to 0.92. when Id is increils~rl from 1 to 10
for R = 0.26 (Le.. the cet1 is shrunk radialty one-fourth of the origina1 radiirs). Therefore. the
efïect on the interference power variation due to the integration of services in srnzrller celIs is
significant.
Figure 2.8: Variation of interference power due to the integration of voice and data users in
cells with different radius.
g 0.8 - 2 al t
0.7 - - - a m,, = 50
u g 0.5 -
y, = 9.5 rw = 9.6 kbps
d U>
5 0.4
2 0.3
0.2
0.1
STD, = 0.7013 1 l
_ _ - - - - _ _ - - - - - - - - _ - - - _ - - - - _ - - - - _ _ - - - - _ - - -
-
Figure 2.9: The signal variation at the base station wheri otily a singk c las OF irsers is present
in a fiilly-loaclecl single-ce11 -stem, Cal1 arriva1 process is Poisson-clistribiitett.
1 2 3 4 s 6 7 a 9 t O Power Factor. 1-d
Signal Variation with Time
We simulated the arrival of a single class and two classes of users separately and then computed
the STDI. Figs. 2.9 and 2.10 respectively show how the interference signals vary at the base
station with time in a single cell-system if the arrival process of users from each class is
Poisson-distributed. Fig. 2.9 is obtained for a single class of users with the parameters shown
in the figure and the observed STDr is 0.1013. Fig. 2.10 is for two classes of users with
the pararneters shown in the figure and the observed STD, is 0.2559. -4s evident from the
figures, when the users of various classes CO-exist and receive the required services. the power
variation at the base stations increases. in both cases, systern is fullyloaded and we used
the cal1 admission control formula (derived in the Appendiu B) in deterrnining the average
nirrnber of users from each cliss. It can be seen chat the analytical results obtained using
(2.20) are also comparable with the simulation resdts. Here. we have to take the ratio of the
two STD's to account for the constant. The value of this ratio is 2.67 and 2.47 respectively
from the analytical and simulation results.
We have seen that factors such as diverse user requirements, ~raffic non-uniformity and
smaller cells contribute to higher variance in the interference power at the base station. Other
factors can be imperfect transmit poufer controlf bursty nature of the sources and severe
channel conditions. in real systems, al1 of the above contribute CO the larger variation in
the interference power. In this section, ive considered a single cell and investigated how the
interference power varies at the base station. In a system with multiple cells. the interference
power will also vary with tirne at each antenna and also bctween different antenna. The
power control algorithrns opprate based on the previoiis trieasurernents aricl if signal varies
unespectedIy the control niecfiariism rnay not be able to follom sucti variation. Hence. the
delivered service varies açcordiiigly as w e k
Since the interference powr at the base scacions are espected to Vary sigriificantly in a
systeni that supports rriulti-rarv/service users. the interfcrcric*t~balancing is stiicli~vl in [lit. nest
Figure 2.10: The signa1 variation at the base station when two classes of users are present
in a fully-loaded single-ce11 systern. Cali amvals of different classes of users are inriependent
Poisson processes.
section as a means to rninirniïe the mobile transmit power.
2.4 Minimizing the Int erference Power Variation among
Base Stations via Dynamic Receive Power Control
We have seen in subsection 2.2.3 that the receive power levels are set based not only on the
rate/serc.ice requirements (Le., SBRPC) but alsa on the levels of interference €rom neighboring
cells (Le., NBRPC). There, ive assumed fixed base station assignment of mobile users to base
stations (snap-shot analysis). IE uiibsection 2-25: we have seen that operating with different
interference power at different base stations adverseiy affects the performance of a CDM.1
cellular system. tri this section: we consider dynamically changing the base station assignrnent
to minimize the interference power; hence, the ceceive power Ievels are adjusted (i.e.. dynamic
NBRPC) based on the traffic conditions in the network.
This mechanism is also caIIed in the literature as "cell-breathing where cell coverage area
shrinks/expands as a result of mobiles srvitching base stations. The celi-breathing can be
implemented using pilot power cantrol where the (fomard link) pilot poiver is set inverseiy
proportional to the total receive power at the base station receiver (as discussed in subsection
3.3.2). In this controt mechanism, there is a possibility that some mobiles may not be under
the coverage of any base station at al1 as a resuIt of weak pilot power received at the mobile.
That is, there wii1 be a fiole in the coverage - wtiich is an undesirable from the service point
of tiew.
Since our purpose is to generate tarying leveIs of interference power at differenc base
stations and analyze the systeni perforniance, ive do this with homogeneous users but with
different loads in ceils (hence chop the class index in the notations). \Ve assume a single
dass of users with svystem-wide SIR-balancing. In each cell. users maintain poiver coritrol
sa thst receive powr Ievefs of al1 clie iiscirs at their respective hase stations arp e<lriiil. Let
j(= 1: 3,3,4) be square cells in a regular celliilar system. This load-irnbalanced cellular system
has = N, 35 = '2Y.Y3 = 3.V: :V4 = 4rV, whcrc Y is assumed to be a large number. Sote
that the power Factor is equal for al1 users in the system. We take $,,* = .$ = 0.09.Vj, b. and
( j # 6). Therefore,
1' =
With the above interference matrk ((I), the receive power vector can be found by solving the
eigenvalue problem (with system-wide SIR-balancing). Hence, P = [1.00 1.43 2.51 10.18]~.
From the above result, we can see that cells with the higher loads need to operate at higher
power levels relative to the lower load cells so that they attain a greater Ievel o l inter-ce11
interference suppression. So far, we considered fised base station assignment (FBSX) for each
user. In the sequel, wc considcr dynamic base station assignment (DBSX).
If ive allow users to svitch base stations. some of those in the congested cells can be
arranged to be connected to base stations in the lighter cells. We tried several combinations'
of loads in the cells and report a combination that gives a minimum mobile transmit power
out of those combinations. Correspondingl- the load levels in each ce11 are: 22N. 3.0K
and 2.4N respectively for ce11 1. 2, 3 and 4. The respective receive power vector at the base
stations are: P = [1.00 1.22 3.65 1.221~.
Table 2.3 summarizes some key meastrres for both fixed and dynarnic base station assign-
ment. There. STD[load]: stantlartl deviatioti of ttic Ioad levels. STD[RPLI: standard deviation
of the receive powr levels, E[.LiTP]: rnean of the total mobile transmit power. E[Q]: mean of
the receive poxer at the base stations. STD[Q]: standard deviation of the normalizetl (by the
mean) total receive power at the base stations. \ \ e used (2.26) in evaluating the total mobile
"n Chapter 3. the pon-er arljiistment and base station sclection algorithms are disctissed that distribiitively
achieve the mininium niobile transmit potver for eacii tisclr.
Table 3.3: Performance measirre rising fi'red and dynamic base station assignment.
transmit power. From the table, the total mobile transmit power dispensed in the systeni
in the dynamic base station assignment is 1.9 units whereas fised base station assignment
required 5.2 units. We can also make the following observations from the results shown in the
table. -4s the load is balanced (i.e., variance of the ce11 loads is reduced), the variance of the
receive power levels is reduced and most irnportantly so is the total mobile transmit power (see
first 3 columns). As well, the mean ancl variance of the receive (or interference) power at the
base stations also clecrease - which inriicates that the base stations operate with alrnost eqiial
total receive (or interference) power. Therefore? Ive conclude that the interference-balanced
system (or a system with equal receive power) improves the CDMA system performance.
2.5 Chapter Summary
We considered the receive power contro1 problem for users nrith mtilti-ratelsenrice requirements
in a cellular CDMA netwvork and showved that the problern can be split into two optimization
problerns: source-basetl receive power control and network-balanced receive power control. \Ve
have seen how both of these methods balancc the interference among the tlifferent classes of
users ancl also aniong different cells in orcler to eficiently accornrnodate integrated services in
a single systerri.
The need for interference-balrincing t m eitiphasized in orcler to minirnim the total mobile
transmit powr and hence EO increae the mpiicity. The cell-breathing cechniqiie \vas shonn
to tlynarnicall- atljiist the receit-e poae: Icl-PIS hasecl on the trafic conditions in the netwrk.
The understanding of the receive power control and cell-breathing rnechanisrns helps us devise
techniques later in our study in effectiveiy handling the real trafic in the nest generation
cellular systems. In Chapter 3: ive disciiss one may of balancing the interference power among
base stations in order to decrease the total mobile transmit power in the network. In Chapter 4.
we use overlapping sectors to distribute the interference power among sectored antenna in order
to reduce the total mobile transmit power. In Chapter 3, we propose a mechanism to rninimize
the inter-ceIl interference exploiting the channel conditions in the scheduling of packets,
Appendix A: Relationship between Receive Power Leveis and Total
Mobile Transmit Power
We derive a simple and useful Formula to calculate the average total transmit porver dispensed
by the mobiles in a CDM-4 network. This also relates the receive power to the transmit power.
Figure ?.II: The computation of the average mobile transmit power generated in a square
ceil.
Let us assume a large number of iisers (iV,) in each (square) ceIl j and everl cell is eqiiai
in size ancl shape as shown in Fig- 2.1 1. The base stations are locatetf ac integer points (1 .111)
rr-herr I = ... - -ID. -2D. 0.2D. 4D ... m t l rn = ... - 4D. -20.1). 2D. -ID... tvhere ZD is t h
length of the square cell. The user density in each ceIl j is givcn by p, = 3. CVe assume a
single-class of users with -stem-wide SIR-baIancing; therefore: the receive power Ievels for the
users communicating with a base station are equal. The distance-dependent path-loss modei
is assumed for radio channels with parameter a.
For a user i in communication with base station j whose receive power Level is P,, the
mobile transmit power is given by
where di = J(xi - 1)2 + (yi - m)*, (x,, yi)? and position coordinates of the user. K is a
constant with dimension of length raised to -a and without loss of generality, let K be 1.
Therefore, the total mobile transmit power dispensed by users in ce11 area j is,
Hence, the total mobile transmit power from d l the mobiles in the sustem.
From (2.21) and (2.22), (2.33) becomes,
Then, the mean mobile transmit potver is given as,
Since a large nuniber of users is assumed. ive can w i t e (2.25):
+D +D Let ~ ( o , D) = & J-D J-o ( ~ ~ - t z ~ ) " ~ ~ d U d z . Hence, the (norrnalizcd) average total mobile
transmit power (MTP) dispensed in the systern is given as,
[t should be noted that for srnaller values of d the attenuation factor cannot be a. In îact,
it is srnaller for smaller distances. Therefore, we have to define different values of a for
different values of d and then do the derivation. For simpiicity, we assurned uniform value of
a throughout the ce11 of interest. Even with spatially varying a, E[MTP] cc C j iVjPj-
For example, if the path-loss exponent is -4 and the length of the square-cell is -4. then
~ ( 4 , 2 ) = 2.49. The average total mobile transmit power dispensed in the system in propor-
tional to the nurnber of users in the ce11 and their receive power levels at the respective home
base stations. CVe use ( 2 2 6 ) for the approximate estimate of the total mobile transmit poiver
in the system.
It should be noted that the average mobile transmit power is a linear combinatiori6 of the
receive potver levels in Our simplified rnodel. If ive minimize the receive power levels as done in
the receive power control in Section 2.2, ive also minimize the average mobile transmit poiver.
Appendix B: Derivation of Equivalent Bandwidth Based Cal1 Admis-
sion Control from the Solution to Receive Power Control Problem
We derive the equivalent-bancltvicith baseci cal1 admission control arising from the formulation
and soIution of the receive power control problem.
We assurned in Section 2.2 that the transrriission rate for class c users, (r ,) . is fisecl for the
cal1 duration. In the general case, it can be clepentlent on the traffic conditions in the ce11 iincl
therefore, ive assume it to be rcb in ce11 6. Therefore, from (2.10),
"ote that tve nssurned E ~ e d base station assignriierit and hence 1% is a constant parrinieter Iiere.
PI; * 1 *j,bPc-i -i '~b .bPc-b = (1 + )P,yb. tlb j i tb (yc-arc-a)
3 Pc-b = CJfb Qj.bPc'j , Qb. 1 f Cv/(-ic-brc*b) - \kb,b
Since, Pcmb > 0,Qb: the denominator has to be positive. Therefore,
Let us assume, CV/(rc.byc-b) >> 1. Hence,
Using the approximated powr factors. fk, = 7:C:::l it can be seen that
Therefore, in every cell, cal1 admission control is defined as:
where wk = Nkrk-ik. The above inequality ciefines the region for the instantaneous cal1 admis-
sion.
The above caI1 admission control is an upper-bound since it is deriveci taking only the
intra-ceIl interference into account. Le.. by having a constraint on the denorninator in (2.27)
which basicaliy depends only on the intra-ce11 interference. The tighter bound can be obtainecf
taking the inter-ce11 interference into consicleration. Typical . for ce11 6. it is in the form of
c:=, wk < bbII' where O < ()à < 1. and (ib ciepends on the inter-ceIl interference e s p e r i ~ n c ~ d
at base station 6.
Chapter 3
Combined Rate/Power/Cell Control
In the last chapter. tve showed that a reduction in the variation of the interference power
among base stations will improve the performance in ~~~~~~4 systems. The hot-spots (Le..
areas that generate relatively higher traffic than the others) can cause the base stations to
operate with higher interference power. We also notcd that users with diverse rate and service
requirernents contribute further to such variation among base stations. Dynarnic base station
assignment was considered as one way of reducing the interference power variation. Korvever.
as will be seen later, it has limitations in stvitching the mobile users (.LIuS) in moderateiy-
loaded networks. Hence: we investigate other means to interference-balance the system to gain
performance advantage.
In this chapter. nre propose and study a schenie that atternpts to reduce the interference
poiver variation among the base stations. We consider the reverse links in a CD&[.\ network
ivith tiigher degree of traffic fiiictuations with space and time. In our scheme. transmission
races of those mobile users in the congested (rion-congestecl) cells are decreased (increaseci).
Hence. the proposed scheme is appropriate for del- insensitive applications. \Ve consider
the niinirnization of the average transmit bit energy subject to rnaintaining individuai target
Eb/lo for each user. Two algorithms. one clirectly rriiriiniizing the transmit bit energy and tir
or lier inclirectIy using meiirittretl pilot power. are giveti. Bot h idgorithnis select the optimal
base station if forward and reverse link gains are equal; however. the latter is decentralizeti
and uses only Iocal measurements and is arrienable for practical inipler~ientation.
3.1 Introduction
The effective use of power control (PC) allows universal spectrum reuse in cellular CDM-1
systerns. The use of the same frequency in al1 the cells helps implernent easy base station
selection, We refer to the dynarnic base station selection in this chapter as ce11 control (CC)
where cell coverage area shrinks or espands naturally as a result of users changing the base
stations (BSs). It has been shown in [47] that joint PC and CC can improve the overall
capacity especially in a network with uneven traffic distribution. The joint PC and CC is
collectively known a s cell-breathing where ce11 cmerage area shrinks/espands as a result of
mobiles switching base stations. In rate control (RC): data transrtiission rates are acljusted
with tirne. RC has been proposed previously; for esample. the transmission rates of data
users can be increased during inactive period of other users (e.g.. voice users) [48]. In our
work, we study the combined use of PC. CC and RC in a CDM-4 network with the goal of
rninimizing the mobile transmit power (MTP) required to transit a one bit. The data rate is
âdjusted depending on the congestion in the network. The optimal home base station and the
corresponding transmit power are found for each user using iterative algorithms.
In the early work [28,43,49] on PC. the assignment of risers to base stations was assumed
to be known and fixed. -4 centralized PC algorithm that requires global information on al1 the
link gains and the transmit power in the netrvork mas proposecl to balance the SIR in ail the
links. The main tlrawback of this algorithm is that it requires continiioiis access to all radio
paths in c he system; therefore, its use in practical cellular systertis is limited. A distributed
PC algorithni chat converges to the optimal one proposed in [49I is giwn in [431 where each
user tvorild roritrol its own transmit power based on only the local information about the link
gains.
The work in [43.49] does not address the problem of congestion in the network. Lsers
in the congestcd ceils rvould be transmitting higher power than those in the lightly-loaded
cells. This increases the interference and decreases the capacity of the interference-limiter1
CD&[_- systems. Basically, load-balancingl among cells (or interference-baiancing among base
stations) !vas not employed t h e . Then, in [47,50], the minimization of the total mobile
transmit power by dynarnicdly changing the base stations for each user depending on the
congestion in the cells was considered; Le., increasing (decreasing) the coverage of the lightly
(highly) loaded cells. By integrating both PC and CC, we can handIe congestion gracefully in
the face of traffic fluctuations.
Vie have seen in Chapter 2 that the cells with higher loads have to operate a t higher receive
power levels. By ernploying cell breathing techniques. some of the users in the congested
cells can be made to connect to cells with lotver loads: hence operating with relatively lower
receive power levels that results in lotver total rnobiIe transmit power. Hotvever. the total
interference power at different base stations still varies considerably since we cannot switch
as many users as ive wish to balance the loading in the cells. Since the interference-balanced
system performs better, we propose to batancg the total interference potver (or receive power
in a large bandwidth spread spectrum channel) at the base stations via controlling the data
rate.
In the proposed scheme, ive âdaptively adjust each user's data rate depending on the
congestion at the home base station as wel1 as on the average congestion in the netivork. The
basic idea in the proposed scheine is that i f a transmitter esperiences relatively higher(loiver)
congestion at the base station receiver, then it reduces(increases) the transmission rate. In
effect. we combine the adaptive ceU control [U] and the aclriptive rate control [481 in the
'-4s noted in suhsection 2.2.5, it is appropriate to use the term -1oad-balancing" and "interference-
balancing" in the si-stem with homogeneous and heterogeneous users respective- '[t is impossible CO do the coniplete interference balancing; howxer- ive rcduce the rriean and vaniince uf
the ir~cerferencc puiver antong base stacions in the netivork.
process of congestion control. To this enci. ive propose two iterative rate/porver/cell (RIPIC)
control algorithms that minimize the transmit bit energy for a given Eb/Io for each user.
The resc of the chapter is organized as iollotvs: The system mode1 is described in Section 3.2.
The cell-breathing technique and implementation algorithrns are discussecl in Section 3.3 €01-
Iowed by the rate control combined with the cell-breathing in Section 3.4. Simulation resuIts
are presented in Section 3.5. Finally, Section 3.6 summarizes the tvork in this chapter.
System Mode1
A set of M transmitters that shares the same CDMA channel is considered. The network
consists O € B base stations. CVe focus on the reverse link (from mobile to base station). The
Iink gain betteen transrnitter i E (1, ...? N ) and receiver j E {lt ... B) is denoted by G,, and
the mobile transmit pomer of i is given by &. For an omni-directional antenna with unity
gain in all directions. the signal power received at receiver j irom transmitter i is Pi] = G,A. CVe assurne the presence of the external noise at each base station receiver. The spectral
density of the esternal noise a t base station j is denoted by $/2- Hence. the external noise
power is, ~ 1 . l ; ~ where I.V is the spread bandwidth of the channel. Let b(i) be the home base
station of i, i.e.. transrnitter i cornrnunicates tvith receiver b ( i ) . The desired signai at receiver
b ( i ) is equal ta ~ ~ ~ ( ~ ~ j > , , while the interfering signal porer from other transrnitters plus noise
pomer is,
Ler die total receive poiver at receiver j be Qj. That is. QI = ~t==, e h , & i rb ll- The average
of the total receive power at al1 the base stations is given b . 0. That is,
[t is assumeci that the base station controller compittes Q and transrnics it via control ctizinnels
to ail the niol>iIe users in tlitb tiettrork. Cl,> ma>- replace the total interference p o w r 1,. II? the
easy-to-measure total receive power Q,: since the clifference will have negligible effect on the
SIR in a large bandwidth spread spectrurn channel. Therefore, ive use Q instead of I in the
rest of the chapter.
For sirnplicity, ive assume a single-class of users in the analysis. However. the proposed
technique can be applied in rnulti-rate rnulti-service systems as wvell. We consider applications
that can tolerate variable del- of packet arrivais. Such applications are now prevalent on the
Internet (e.g., push applications such as stock and weather updates). In digital communication.
bit error rate is determined by the received bit energy-to-interference density ratio (Eb/Io).
For CDM-4 systems, Eb/lo is given by r H . where is the SIR and H is the processing gain.
Let r:eq and -/yq be the required data rate and Eb/Io of user i. If the instantaneous data
transmission rate is ri then Hi = n. The user signals are spread with variable gain depending
on the data rate. Therefore. SIR for user i is given b~
The Eb/l, is given by,
Let us define a service parameter Q, as. ai -,,ri. Hence. iPi = C i T i . In order to get 9.
the usual SIR is multiplied by CI/ ' to get a measure in Hz. This allows us to work with either
y or r or both. In our work. r is adjusted while keeping the -/ iïxed to the required E6/l0.
Since the data rates are adjusted. the proper performance meiisure of interest is the bit e n c r e
that is needed to get the reqiiired Eb/l,. Let the transmit bit energ-- for user i be deCined as.
8; r 8. Ancl the average transmit bit energv for the systeni is 8 r $ XE, ei. Our goal ri
is to minimize the transmit bit energl; for each user and then in the ?stem. Therefore. the
problem of cornbined rate/power/cell cotitrol cari be formulatecf as folloivs:
Find: b ( i ) . Oit i i m i e : et. Sirbject to: 7i = mtZreq. Bi > O? ~ i .
The difference betweeri oiir schenie arid previoiis relatetl w r k [-LT.30] is that ne rior o d y
adjirst ( D i (to be precise. r,) in each iteration of the algoritlini but it is set bascd or1 the
congestion at the honie base station as well as in the network. Hence, a t each iteration (n) of
the algorithm, we have
As will be seen later, the rates (ri) are set by (3.13). Hanly [47] and Yates [50] have solved the
similar problem and given an iterative algorithm to find the optimal mobile transmit power
and the base stations assurning that Qi(n) = cPi,Vi. In our scheme. in every iteration. Qi(n)
can be adjusted. We show that if there exists a feasible solution (in terms of b ( i ) , F, and r,),
the algorithms will find the solution. FVe discuss power and ce11 control in the next section.
3.3 Power and Ce11 Control
Here, only power control is employed- It is assumecl that the assignment of users to base
stations is known and fixed. Users connect to the strongest3 (in radio distance sense) base
station at the connection setup time. Therefore. the base station assignment is given by,
b ( i ) = arg mas{Gi,) - (3.5) 1
An iterative decentralizecl PC algorithm proposed in the literature [49. 511 gives the mobiIe
transmit power of user i a t (n + l ) lh iteration as:
req req where,:Di(n) = yi r, . it has been shown ttiat if ttit? solution exists, iteration given by (3.6)
converges to an optinial solution from an? arbitra- porver vector. in the sense that the total
niobile transmit potver of alL the tisers is rriiriiniizd with the E ~ e d assignmcnt of users to
3 K tve assume distance-dependenc path-loss rriodel for links (no shadowing), then this tvoiiItI be closest in
physical ciistance.
base stations. The main problern with PC-only scherne is that Ive accornrnodate iisers in
congested cells at the expense of higher mobile transmit power. Also there is a high variatiori
of the interference power arnong base stations. By allowing the users to switch base stations
and hence balancing the Ioad arnong celIs, ive can reduce the total mobile transmit power as
discussed next.
Here, both power control and ce11 control are ernployed. \rVe allow users to change their
base stations dynamically over the connection duration. CVe have seen previously that when
implementing PC+CC and the load is balanced among cells, the total mobile transmit power
is minirnized. Since the lightly Ioaded cells accept users from heavily loaded cells and they
are operating with relatively lower receive power levels. mobile transmit power of these users
are rediiced. This results in smalIer receive power (anci interference power) in the hot-spot
ce11 than what would have been, had these users linked to the base station in the hot-spot
cell. Our motivation to combine rate control with power/cell control is also inspired by the
observation that balancing the interference power arnong base stations can reduce the total
mobile transmit power. Vie discuss two algorithms that can implernent the PC+CC.
Alg-1: PC+CC (Min Tx MP based)
In this algorithm [di], each mobile transmitter (Ts) cornputes the reqiiired transmit mobile
power (MP) to a11 the base stations (or ri subset of thern in a Iirnited case) that achieve the
requirecl Eb/I,; then, selects the base station that recluires minimum Tx MP.
That isl
req req Again, Q i ( n ) = ri . The mobile transmit power, P, is given by (3.6) with b ( i ) çelecteci
by (3.7). It has been proven f471 that titis aIgorithm converges and firids a home base station
with minimal transmit power for each user if the solution exists.
Alg-II: PC+CC (Max Rx PP based)
In this algorithm, each base station transmits its pilot power (PP) inversely proportional
ta its total rneasiired powerJ at its receiver. Hence, the pilot power of base station j is,
' For a user i7 the receive (Rx) pilot power ist $ ( n + l ) =
where G:, is the forivard link gain between transmitter (base station) i and receiver (user) j .
Each user senses the receive pilot power and picks up the strongest that determines the home
base station. That is,
b ( i ) = arg mau{Aij). J
The mobile transmit power, pi is given by (3.6) with b(i) selected by (3.9).
3.3.3 A Potential Limitation in PC+CC Scheme
We show thiit poiver and celI control scheme (also knorvn as cd-breathing) has limitation in
its ability to switch users between base stations in rnoderately-loaderl networks. We illustrate
the above facc using an esample.
Let us corisider a CDhIA network with tivo circular cclls (of unity radius). Reciprocity in
fornarcl ancl reverse links is assumed. We use the pilot-posver basecl base station assignment
algorithm ( Alg-II) iri selecting the home base station.
Let ris consider a particuhr user ir d distance frorri base station p as shown in Fig. 3.1.
In orcler Eor niobile Iiser i to switch base station from p to (1. che folloiring coriditiori. whicti
.'Sote that wc :issunie Q, > O.Vj, sincc ive incIurIe the tion-zttro thermal rioisc at each Liase station.
Figure 3.1: An esample of a user switching from one base station to another.
cornes from (3.8) and (3.9), has to be satisfied:
where Qj is the total receive power at base station j and Gij is the (reverse or fonvard) link
gain between user i and base station j. With the distance-dependent path loss mode1 for
channels, (3.10) becomes.
where cr is the propagation constant.
Table 3.1: 4Iinirniim ratio of total receive power Qp/Q, at two base stations in order for user
i CO switch from base station p to base station (1.
Table 3.1 shows for different d's. the rriininitrrn ratio of total receive powr at base stations,
p aritl ri. that is rt~c~iiirerl if a mobile user i were to switch froni p to q- It is eviclent froni the
table that if rf = 0-5 and CI = 4: then Qp has to be at l e s t as midi as SOI), in order for the
smitching to occitr. En a high capacity cellular system: cells tvill be moderatel-loadecl and
this condition car1 tiardly oçcur. As a result, it is very unGkely that ceIl-breat tiing will heIp
those users close to a base station if they need to switch.
We can integrate rate control with power and ce11 control to make cell-switching possible
and to minimize the interference variation among base stations as seen next.
3.4 Two Algorithms for Combined Rate, Power and
Cell Control
In this section, we propose two algorithms that allow different transniission rates and minirnize
the transmit bit eriergy for each user. If the data rates are fixed then this is equivalent to
minimizing the transmit power as done in [U].
3.4.1 R/P/C Control (AlpIII & Alg-IV)
üsers adjust (if necessaru) their rates, power and home base stations. The data rate of user i
is adjusted asl
a and K is a positive constant. The value of c will be dependent on the where 'T,(n) = Q,l,,l
variation of the interference power at the base stations. such that the higher the variation
is, the larger the rate atijttstment should be. For esample. it can bc a function of standard
cieviation of the total interference power at the base station. The formula (3.12) helps adjust
the rate in relation to the congestion IeveI. We do not study siich a frinction iri tletail and for
sirnplicity. Ive wiirtie z = 0. therefore.
Hence: (bi(n) = -/yqri(n). The rate setting given in (3.13) aIloivs some uses to get more and
some less than tvhat tvaç origifially required depending on the congestion with time. it can be
seen that in the proposed scheme transmission rates of those users in congested (Light) cells
are decreased (increased). Since our goal is to interference-balance the system, we set the
rates as given by (3.13). If we have a minimum data rate constraint: then Ive can incorporate
it into our algorithm. The mobile transmit power of user i has to be set as given by (3.14) in
order to get the required Eb/I, with the given data rate in the radio environment.
By substituting (3.13) into (3.14)? Ive get
req req - ~ ~ ( n + 1) = 'Yi Ti Q(n)
bVGib(i) -
It is noted that the data transmission rate depends on both Qb(,) and Q and the mobile
transmit power depends only on Q. We have seen why we need to set the rate as given hy
(3.13) and the mobile transit power by (3.15). Yext we discuss two algorithms that implement
the combined rate/power/cell control.
Alg-III: PC+CC+RC(Min Tx Eb based)
This algorithm is similar to Mg-1. User i cornputes the required transmit bit energy co base
station j :e i j :Vj : with data rate and niobiLe transmit power given by (3.13) and (3.15) respec-
tively. Since we minimize the bit energ?. let
P, (r1 + 1) Q,,(n + L) - j = (1. -.-. B ) .
r,{n + 1)
From (3.13) and (3.25):
where r,, is the date rate if user 1 were to connect to bise station j and.
It can be seen that base station selection proceciure is the same as in Mg-[. Hotwver. the
selccted base stations rvili not necessari- bc identicai' because of the difFerence in total receive
power at base stations.
Alg-TV: PC+CC+RC(Max Rx PP based)
This algorithm is similar to AIg-II. Base station selection is the same as in -4lg-II. -4 user
C' i receives the pilot power iIij(n + 1) = P;(ni and then selecrs the home base station as
b ( i ) = argmavj {A,). The data rate and mobile transmit power are given by (3.13) and (3.15)
respec tively.
3.4.2 Optimal Solution in Mobile Transmit Bit Energy using Alg-
III
Hanly [LI71 and Yates [XI] proposed an algorithm5 each to choose home base station b ( i ) and
corresponding transmit porver fi subject to 7, = f q , that rninimizes 1 pi. There. users begin
witharbitrary positive power levels P(O) ivhere ~ = [ h ... hlT. Then. they adapt their
power Ievels inductivety as follows: Given that the users are transrnitting with power P(n) a t
step R, they cornpute the values at the next step. ~ ( n + 1) by first computing P,J(n c 1I.b
via,
i d then selecting the home base station as b ( r ) = arg rnin,(~~~} and the nert transmit paver
"'Qh'li'n'- Here. u, corresponds CO a parameter that depends on the source data ab+l) = ,&(,,
rate and requiretl quzility of service for user i. The? considerecl a fked rate sciienie ir-hereby
r, = r:eq,Vi.
"See Section K in [di/ for more detriiis and Sections 11 arid [II in [SOI. Xote that ae have modifieci the
riocations and sirnplified the prwntation.
Hanly - Hanly considered spread spectrum channel and hence, lui - riri/(ll; and Yates consid-
H Q ~ ~ Y - w).ote~ eretl the non-spread spectrum case and hence. luY"tes = r,- It can be seen that wi -
since yiri = CVr, (see Section 3.2)- Basicalllv, both solved the same problem from a different
point of view. The solutions if they exist (in terms of home base station selection and mobile
transmit power) given by Hanly [47] and Yates [30] are not necessarily the same though both
minimize the total mobile transmit power.
Let us now analyze the proposed algorithm, Mg-III. We are minimizing O,, where Elij =
proposed frorn (3.17). Therefore, in the proposed scheme, w i L W ,
= yi /W. Comparing mith
Hanly's approach where C is rninimized with Wi = *,iri/l.V, in our scheme 1 ei is minimized
with wi = yi/W. Note that ei = i.e., data rate and transmit power are adjusted to rl
minimize Oi. Therefore, by following Hanly's argument. we can conclude that the power
adaptation algorithm given by (3.15) with the data rate set by (3.13) will yield the optimal
solution in the minimum mobile transmit bit energy for each user. Et should be noted that the
proposed algorithms will find the optimal solution only if such solution exists. The conditions
for the existence of solutions is analyzed using matris algebra theory in the Literature (see
also [-Li, :O]).
3.4.3 Cornparison of Algorit hms
We compare the above-discussed algorithnis here. Alg-l(i1i) requires the knowledge of the
total receive power at each base station and link gains from each mobile user to each base
station. Hoivever. in Alg-lI(iV) users can easily select the home base station using the pilot
poivcr. Then. each mobile user needs to know only the reverse link gain and the total receive
powcr at the tiotne base stations to determine the mobile trarisniit porver. That is. in Ag-[(III).
a i i s t ~ 1 ~ieetls to know Q,.G,,.Vj whereris in Alg-[I(I\')- i c oril! requires Q,.Gi,. j = b ( ~ ) in
ortler to tleterr~iine the home base stacion. it can be provetl chal both Alg-[(III) ancl Ag-II(I\-)
st*lec-t the sanie home base station for each user if foriv;irtl and reverse link pairis are cqii;iI
(see t he Appentlis C). Hence, Alg-II(IV) implements Mg-[(III) wi th local knowledge about
the Iink gain aritl total receive porver at the home base station. To deterniine the data rate
and mobile transmit power in the proposed algorithms, a user needs to know Q and the total
receive porver at the home base station. tt is assumed that Q can be included within a control
channel. Therefore, we conclude that Mg-IV is the right choice for practica1 impiementation
since it uses only local measurements to set the data rate and mobile transmit power.
The difFerences between the previous cell-breathing schernes and the proposed schemes are
1. The data rate, ri is adjusted in Mg-III(Xlg-IV) while not in AIg-I(.4lg-[I),
2. The transmit power, fi is dependent on QD(,) in Alg-I(Alg-ll) but on Q in Mg-IIl(A1g-
IV),
3. The minimimion is done over pi in .Mg-[(;\lg-II) and over Oi in .Mg-III(;\lg-N).
3.5 Simulation and Results
The performance of the proposed aigorithm is evduated by simulating the same system as
in [47I. -4 network with 36 base stations is taken for study. The base stations are located at
integcr points (1. m). 1 , rn = 1, .... 6, as shown in Figs. 3.8-3.10. For siniplicity ive assume that
ai1 users reqiiire the sarne Eb/I, (ab) anci transmission rate (r,): r, = 8 Kbps and -j, = 6.77
dB. With It' = 1.25 M-iz: this corresponds to 4, = 38 KHz. The link gain is niodeleci as
Cl, = L/<$, where d , is the distance between user r ancl base station receiwr j . First. ive
have taken 400 trsers and distribuced them uniformly over the network. Then. we create a
hot-spot aroiintl the base station at (4.4) by adding 20 more i~sers rrnifornrly. This barely
makes the Cised assignrnent infeasibk rvhiIe making the dynarnic iissignment feasible.
\\k (10 a sriap-siiot. arialysis. ive.. rit a tirtie insttirit OF t h systeri.1. Tri practice. iisers are
triobile and hot-spots c~tiringe with tinie: hotwcer. r w art1 (~oiisidering th^ network at an instance.
CVe run the algorithms Mg-O to Mg-IV for a maximum of 15 iterations. Lsers are allowed to
connect to one of 36 base stations. We consider al1 the users in the network when evdhating
the performance. We assume equaI reverse and fonvard link gains. Therefore. the Ag-[(III)
and Alg-[[(IV) will give the same performance since they are in fact the same except for their
way of implementation.
3.5.1 Convergence of Algorithms
Figure 3.2: Convergence of PCI PC+CC and RC+PC+CC algorithms.
Let us first analyze how the aIgorithms perform in terms of convergence. \Ve define a user
to have unacceptable Eb/l , if it $vas not within 0.5 dB of its target Eb/I,. In Fig. 3.2. we plot
the fraction of mobile iisers with unacceptable Eb/I, versus the nuniber of iterations. Though
we report onlx thc rcsiilts obtained in one instance. this gives an inclictition of the convergence.
A1g-O has not corivcrgecl even after 1; iterations. because the trafic wnfigiiration is not
feasible when iniplcrtitwtirig the fised ~ ~ I S P statioii usignrnent. Other algorithms converge.
Alg-III(IV) and Alg-I(I1) converge after 5 and 10 iterations respectively. The faster conver-
gence with R/P/C control can be explained as follorvs: The cell control in which mobile users
are switched till the optimal solution is achieved, is directly affected by the rate control. That
is, users decrease/increase the rates and avoid being handed off to other cells, which other-
wise would occur with no rate control. This is evident when we look a t the total number of
- AIg-&PC )- - + -. + AIg-l:PC+CC(Min Tx MP) 0
- o- Alg-ll:PC+CC(Max Rx PP) +0 o Alg-lll:PC+CC+RC(Min Tx Eb) - ec
- - - Alg-IV:PC+CC+RC(Max Rx PP,
0 A A - - - - 0 5 10
Iterations (K)
Figure 3.3: Total number of mobile switching that would occur during the power/rate adjust-
ment.
handoffs that occur during the 15 iterations in Fig. 3.3. It should be noted that we show the
cumulative total in Fig. 3.3. That is, when we report a number For an iteration, we add all
the handoff that occurred in each of the previous iterations (including the current iteration).
From the figure. it can be seen that the R/P/C control has Fewer hancloff than in PC-iCC
control. Since the PC-only algorithm does not allow mobik switching. therr is no handoff.
8000 A * * * A A * A * A A * A v - - T v v w F - v v v v -
I l
7980 - f PC+CC(Min Tx MP) I I
7960 - I PC+CC+RC(Min Tx Eb) 1 - * - PC+CC+RC(MW RX PP 1
7440 - !
Figure 3.4: Achieved bit rates by al1 the users arc aclded iip and then awragecl.
3.5.2 Average Bit Rate
Fig. 3.4 shows the average bit rate achieved by the users for different schemes with the nurnber
of iterations. -4fter 15 iterations, the bit rates from al1 the users are sumrned up to get the
average bit rate. Alg-O, Alg-1 and Mg41 do not adjust the transmission rates: as a result,
their achieved rate remains a t S Kbps. In Mg-III and Alg-W. some users (that comniiinicate
with the base station at (-1.4) transmit at lower rate (= 7.7 Kbps) whilc others (that are not
in the hot spot ce!!) transmit at higher rate (h: 8.1 Kbps). As a result. the average bit rate is
about 7.8 Kbps.
We noted in Section 2.3 that due to cell-shrinking and multi-meclia users. the receiveci
signals va- significant Iy. Wit h the rapid traffic fluctuations spatiallu (Le.. hot-spots rhange
wïth time). ive espect each user to get the required amrage bit rate ovcr the duracion of the
connection.
3.5.3 Minimum Transmit Bit Energy
In AIg-O. Alg-I and Mg-III data rates remain unchanged. However. in our scherne. the data
rates are adjusted. Therefore, we cannot compare the mobile transmit power in different
schernes. Instead, the ratio of the mobile transmit power to the bit rate (transmit bit enerp)
has to be compared. Fig. 3.5 shows the average bit energy for al1 the 420 users with the
nurnber of iterations. Compared to the scherne that uses only PC and CC. our algorithnis
reduce the average bit energy by about 25% (frorn 6.4 to 4.8 units) when RC is ernployed.
The adjustment (decreaselincrease) in the data rates helps more to reduce the mobile transmit
power, -4s a result, the transmit bit energy is reduced.
Atg-I:PC+CC(Min Tx MP) AIg-II:PC+CC(Max Rx PP) Aig-III:PC+CC+RC(Min Tx Eb)
21 O 5 10
lteralions (K)
Figure 3.5: Average transmit bit energy reqriired by different controi algorithms.
The interference power at al1 the base stations are added up to compute the mean interference
power in the systern. It is the smallest when using the aIgorithms Mg-III(IV) as shown in
Fig. 3.6. In our scheme, ive atternpt to balance the to td interference power at each base
Figure 3.6: Total interference power at al1 the base stations with different interference control
techniques.
station by adaptively changing the data rates. Fig. 3.7 shows the (norrnalizecl by the mean)
standard deviation of the interference power at base stations - which can be an indication
of the interference-balanced system. The smaller standard deviation indicates that the base
stations operate mith almost the same anioitnt of receive (or interference) pomer. It is ciear
frorn the figure that the .-&-III and Alg-11- rediice the high variation in the total interference
power arnong the base stations and herice acIiieve the interference-balancing.
I O 5 10 15
Irerations (K)
Figure 3.7: NormaIized (by mean) standard tieviation of total interference power at al1 the
base stations.
3.5.5 Coverage Area
In Figs. 3.5-3.10, we group the mobile users that cornmunicate with a base station to form the
coverage area for the particular ceII. The coverage of a base station is loosely defined as the
largest physical area nrhere the mobile users that the base station serves reside. Fig. 3.8 shows
the coverage area when no dynamic base station assignment was implemented. It can be seen
that the center ceIl has a larger coverage compared to the other two cases. Fig. 3.9 shows
the cell coverage using PC+CC algorithrn, Some users in the congested ceil are now being
connected to surrounding base stations. There is not much difference in terms of coverage
with PC+CC and PC+CC with RC as shown in Fig. 3.LO.
I Basxi M PC oniy
Figure 3.5: Ce11 coverage with PC-on]> 'sers are assigned to the closest (in distance) base
station.
1 PC+CC(Based on Min Tx MP and Max Rx PP)
Figure 3.9: Ce11 coverage with PC-tCC (Min Ts MP/hIas Rs PP based): Csers are üsçigned
to a base station that recjuires minimum mobile trarismit power.
Figure 3.10: Cell coverage with PC+CC+RC (Min Ts Eb/4Tas Rs PP baseci): Users are
assignecl to a base station that requires rriiriiniiim transmit bit erierC;!=
3.6 Chapter Summary
I
We proposed ri scheme that combines power, rate and ce11 control to be inipleniented as a
means of congestion control. The new scheme helps balance the interference among base
stations while the traditional cell-breathing, which was shown to have limitations in switching
mobile users between base stations in moderately-loaded networks, fails to do so. Since some
data applications can tolerate delayed packet arrivals, ive exploit them in balancing the system.
,%BR (available bit rate) applications are not guaranteed any rate; however. they receive any
services only when the resources are available. Therefore, =\BR services can be best supported
using the proposed scheme. In the proposed scheme, transmission rates of those usew in the
congested (light) cells were decreased (increased). Two algorithrns: one directly minimizing
the bit energy and the other indirectly using measurecl pilot power. were given. Each jointly
optirnized the base station selection and the mobile transmit bit energy Insteüci of optimizing
the mobile transmit p o w r and data rate separately, ive set the data rate in such a way to
interference-balance the system and then rninimized the mobile transmit bit energy.
Simulation resiilts showed that our algorithrns converge faster than the schertie without rate
control. and minimize the average transmit bit energy. It was also noted that a 25% reduction
in transmit bit energy can be achieved on average using our schemes over the scheme nrith no
rate control. These were achieved via interference-balancing among base stations (or in a way
making the antennas operate with equal receive power) by adjusting the data transmission
rates. With the traffic fluctuations spatialIy with time. ive expect each user to get the required
average bit rate over the duration of the connection.
Appendix C: Proof that Alg-I(II1) and Alg-II(IV) Select the Same
Celi-Sites if Forward and Reverse Link Gains are Equal.
We pmw chat cht% pilot-poner based base station selection algorittim. :\IO-il(I\') chooses the
same but* station selecteci by the ocher algorithm. .Mg-[(III) if foriurrl and reverse Link gairis
are equal for a user.
In Alg-1, user i selects the base station at (n i L ) ' ~ iteration as, from (3.7):
Since
~ , , ( n + 1) = and Qj>O,Vj, WGij
b ( i ) = arg min{ QiQ, (n) ) = a rgmg{ cvcij }- 3 \ t ' ~ ~ ~ 1 @iQjtn)
Since is independent of j , we have
G, arg mau{-) QAn)
as done in Alg-II (see (3.9)). Similarly: it can be shown that Alg-III and Alg-I\' select the
same base stations. This concludes the proof. i
In practice. mobiles rnake several short-term measurernents of the signals to average out
the fast-Fading cornponents. Theii the base station selection is based on the long-terrn niea-
surernents which depend on the loss due to the distance and the shadowing. Since these are
reciprocal on the forwml and reverse links. user can use either the fonvard link or reverse
Iink gain measurernents in switching cells. Therefore. by replacing G with 6'. we showeci that
Mg-[(III) and Ag-II(ni) woiild select the same base station for each user.
Chapter 4
Adapt ive Sector Control
The distribution of interference power with the angle of arriva1 (Ao.4) a t the base station
antenna is non-uniforrn in wireless networks typically due to the existence of hot-spots. A
hot-spot can be defined as an area where higtier traffic is generated relative CO other areas,
Hot-spots can be a result of a larger number of catls originating from an area or due to a
srnalier number of higher data rate users. The integration of diverse services (e.g., voice and
data) into a single system further contribiites to such variation. The traditional sectored
cellular system cannot handIe the non-uniform interference power effectivery- In this chapter,
wve consider an antenna architecture with fixeci ovedapping sectors (called FOS.4-A) that offers
flexibility in the assignrnent of mobile users CO antenila. In the FOS-4.4. o d y a subset of the
overlapping sectors (or anterina) will be selected atLaptiveIy depending on the traffic conditions
around the base stations. The basic idea in ernploying the F0S.A-1 is that if we cari distribute
the interference power and reduce the interference power variation among the active antenna,
then ive can redrice the total rnobiie transmit power clispensed in the sustem.
4.1 Introduction
In trâditional cellular systems, signals from different antenna are directed towards different
sectors avoiding the CO-channel interference 1201. However. the signals from different antenna
often overlap in space - which in one way helps implement soft handoR in CDhIA networks.
This type of overlapping cari be referred to as soft-overlapping where the interference due to
such overlapping is kept under a threshold. The soft-overlapping can occur from the antenna
belonging to the same base station (BS) or to the different base stations. In the fixed overiap-
ping antenna architecture (FOS-AX), we allow (controlled) overlapping of sectors where more
than one co-located antenna can cover any space in a cell. This type of overlapping is referred
to as hard-overlapping where the interference due to the overlapping is above a threshoId, In
our work. me consider the FOS=\-A and hence the hard overlapping. This (hard) overlapping
poses some problems: in an orthogonal CD4I-4 system (i.e.. forward links). it becomes the
code-assignrnent problem [52] where no two codes can be used simultaneously in a space and
in non-orthogonal CDMA system (Le., reverse links). it becomes a signal-to-interference ratio
(SIR) control problem where mobile transmit power (MTP) and base station antenna assigri-
ment (BSA-A) have to be managed to control the interference. Our work Focuses on the latter
problem specifically in the hot-spot scenario which simulates the non-uniform interference
power around the base stations.
The hot-spots introduce large blocking probability in practice. Many techniques have
been propoçed in FDMX and TDMA systems to alleviate the hot-spot problem including cell-
spiitting, cell-sectoring, cell-overlaying, channel borrowirig and channel sharing (see [53I for a
comprehensive survey on these techniques). The above techniques are mainiy related to re-
asissigning the channel arountl the hot-spot area [54,55]. However in CDM-4 systerns. a11 of the
cells/sectors can irse the same channel with strict transmit power control. Siost of the previous
work on non-tiniform trafic either in TDSI-A or €DAI-\ [S. 56-37] or in CD4i.4 [.LG.aS, 591
s-stcbnis clcalt trith the irnbalance of loatl Ir\-els artiong rells. not sectors. in [GO-611. non-uniforrn
distribution of mobile users within a ce11 ivas consiclered and adaptive sectoring ( without using
overlapping sectors) was considered to reduce the congestion in the hot-spots.
As notecl in Chapter a fem high data rate users may cause enormous interference to
a large number of low date rate users. We can confine the higher interference power that
cornes from the high data rate users using antennas with narrower bearns. -4ntenna arrays
with adaptive beam forrning [621 have been proposed to suppress interference by steering the
nulls to the potential interferers. Though they can handIe the angular load variations very
effcctivelv. their cost and complexity are of great concerns for irnplernentation- One alternative
to the adaptive antenna array system is to use the switched rnulti-beam antenna system in
which several narrow beams are used to cover a cell. A beam receiving the desired signa1
with the highest signal strength is selected. In [34], a switched multi-beam antenna system
is considered where the beam selection is done on a frame-by-frame basis. In our work, we
propose to employ the fised overlapping beams (sectors) because of the low complexity and
the possible evoliition from the current fixed sectored cellular system. The overlapping sectors
can be employed either in one or more cells depending on the traffic conditions.
We differentiate the process of ce11 sectoring twofold in the FOSXA as ceil-breathing and
cell-slicing, which can be viewed to occur based on the radial and angular loads respectively.
The "cell-breathitig': is a general term to describe the shrinking/e.upandirig coverage of a cell
(or sector) that cari occur in either radial or atigular directions or both. However. in the
literature it has been used to refer to only in radial direction and ive too continue to do so.
-4s show in subsection 3.3.2 the cell-breathing can be iniplemented using pilot power control
where the (forwartl link) pilot power is set inversely proportional to the total receive poiver a t
the base station receiver,
[ r i a systeni with high interference power variation with the -40-4. WP show that cell-
breathing aloricb is not effective: hence. ive propose to ernploy fiset1 overlapping sectors to
coritrol the intctrfcrerice azirnuthally - n-hich WC cal1 cell-slicing. The omrlapping sectors arc
iisetl as a rxltBa~is to implement the angiilar counterpart of the radia1 cd1-brcathing scherne.
Therefore, the hybrid scheme (cell-breathing + cell-slicing) provides advantage in both an-
gular and radial directions. LVe show that the base station antenna asçignnierit plus power
adaptation (BSX-A+P-4) algorithm proposed in [4Ïj can be used to select the required set of
antenna from a pool of fixecl antenna in the FOS=\,-\. The BSAXtP.4 algorithm implements
the hybrid scheme and is shown to minimize the total mobile transmit power in the system.
The rest of this chapter is organized as follows: The FOSAA is described in the next
section. Then. in Section 4.3, the FOSA-4 is analyzed for the reverse links and it is shown
that there exists an optimal solution in minimum mobile transmit poiwr and that can be
achieved using BS.-\A+PA algorithm. In Section 4.4, the spatial interference filtering and the
use of overlapping sectors to handle non-uniforrn angular loads are discussed. In Sections 4.5
and 4.6, performance of the FOSAX system are evaluated analyticaly and via simulation
respectively in the hot-spot scenario. Finally, Section 4.7 concludes the chapter.
4.2 Fixed Overlapping Sectored Antenna Architecture
(FOSAA)
In the traditional sectored cellular system such as IS-95. a cell is divided into 3 or 6 sectors
each being covered exc1usivety by non-overlapping antenna. There. highly directional fixed
antenna are used to suppress the CO-channel interference. In such a susteni. there is a lirnited
fiesibility in assigning users to different sectors of the same cell because of the limiteci antenna
gain outside the main coverage area. -4s a result. this traditional sectored ceIlular architecture
hcks the capability of supporting the non-uniforrn angular Loads.
In the F0SA.A. the radiation patterns of antenna are arranged in sirch ii ilr- that a user
an'where in the cell cari cornmunicate to more than one antenna with eqtixlly strorig radio
pathl. This is achieved by having a set of co-locateci fiseci anci cIirectionaI antenna whose
'Since antcnna are CO-locaterl the slow fading and tiist~uic~clepenclenc loss are eqiiiil h r r i a niohile user to
a set of ( L ) co-locntetl nntennn in s systeni ttiat impieriients the FOSAA. The hotnc h i ~ c ~ . station nntenna is
I
! !
coverage overlap spatial- Since each user h a a freedorn to select one of many antenna with
equai radio path for communication, 1% have more than one choice in the assignment of users
to antenna rvhich can be done based on the angular Ioad levels.
Figure 4.1: A mobile user i can potentially be covered by L antenna in che FOS-4-4. Antenna
"1" (denoted by solid line) receivcs signals from users rvho may possibty be in communication
with L different (CO-located) antenna.
Let the total nurnber of antenna at a base station be J . Each antenna generates a radiation
pattern that can azimuthally cuver the angle of 0. We assume that these antenna are uniformlj-
distributed2. Let the number of overlapping antenna covering any space in the system be L -
which is also rcferred to as degree of ovcrrlapping. Hence, it can be shown that.
selectcd baseci an the above channel conditions ( a h averaging out the fast rnrilti-patli fading). Hence. we
conclttde thac al1 L antenna are equdly ai.nilable for a niobile user co seIeet for the operation.
'That is. both the azimuth angle of each swtor arid angriiar separation betwecn the centcr of each mtrmria
where S = $, and S is also the number of sectors in a cell as in the traditional (non-FOSX.4)
system, tvith no overlapping of sectors. Fig 4.1 shows L(> 1) antenna cornring a user
i and antenna "1" receiving signals from users who are in communication with possibly L
different antenna in the FOSA.4. Note that O < L < .J and L = in our rnodel. Hence. we
denote the FOSAA as [J,L] in the rest of the chapter. Every antenna in the system transmits
pilot power whose signal strength is inversely proportiond to the total measured power at
the respective receiver. This helps mobile users (Siüs) to sivitch base stations anytime if
necessary depending on the interference experîenccd at the antenna. Several antenna beams
can be formed and directed in such a way to create overlapping coverage. Therefore, the
F0S.U can be a software implementation in practice.
It can be argued that instead of using J overlapping sectors with angular width of 0 and
L clegree of overlapping in a cell. one can use J non-overlapping sectors with angular width of
6/ L and get the similar performance. For esarnple. sis overIapping sectors with angular width
of 1'20" vs siu non-overlapping sectors with angular midth of 60". =\part from the benefit of
tising the FOSXA to handle non-uniform anguIar traffic. the foilowing benefits support the use
of overlapping sectors: (a) the use of overlapping sectors (combined Mth dynamic intelligent
code management) increases the fonvard link orthogonal code ce-use efficiency as shown in [52].
(b) there will be a practical limit on the mininiuni size of any sector and hence the use of
overlapping sectors with relatively larger angular width may be desirable and (c) the number
of handoffs3 due to mobility can be reduced usirig the overlapping sectored antenna.
Minimum Mobile Transmit Power Solution
In this section, ive analpe a CDSI.-\/FOSA.I system to show that there esists a minimum
mobile transmit power solution.
"1s will be noter! later. since a user can switch anytinie EO an ancenna that requires triininitirn transrriit
powr, softer handoffs r d 1 occur even mithoiit [nohility-
4.3.1 System Mode1
bVë consider the reverse links whose quality are characterized by &/Io. -1 user i requires Eb/Io
of li and data rate of ri. Transmitter activity factor of iinity is assumeci for al1 the users. Each
user is equipped with only one omni-directional antenna. Hoivever: a base station uses severai
antenna that overlap in coverage to serve a cell. Uniform antenna gain is assumed for al1 the
antenna. Xeither soft handoff nor mobility is considered. Total spreading bandwidth available
is W. The spectral density of the external noise at antenna receiver j is given by q, /2, so that
the externa1 noise power at the receiver is r1jl.V-
-4 set of iV transmitters shares the same CDM.4 channel in the reverse link. The network
consists of B base stations and the FOSAA [J?L] is implemented at eachJ base station. That is,
.I antenna are mounted at each base station with L degree of overlapping. Therefore, the tocal
number of antenna used in the system is J x B. Hence theoretically, a user can cornmunicate
with J x B antenna. thoiigh practically with a small siibset of them. The iink gain between
transmitter i E (1, ..., N ) and receiver j E (1. .... J x B ) is denoted by Gij and the mobile
transmit power is R.
4.3.2 Inter ference Analysis
Let a ( i ) be the home base station antenna of user i. Le.. trarismitter i communicates Nith the
receiver at a(i ) . Let U ( a ( i ) ) and Ü ( a ( i ) ) be a set of users who are inside and outside of the
sector covereci by a ( i ) respectively, From U ( a ( i ) ) . let H ( a ( i ) ) be a set of users (not including
user i) whose home base station antenna is a ( i ) and the rest. H ( a ( i ) ) . is comrnunicating with
other antenna CO-located with a(i ) . Hence. H ( a ( i ) ) = { U ( a ( i ) ) - H ( a ( i ) ) - i ) . Antenna u(1)
wiII be receiving signals from i and interference from H(a( i ) ) . H ( a ( i ) ) arici Ü ( a ( i ) ) . Typically,
if iisers are nssignetl based only on mobile-to-base station distance. H ( j ) - H ( j ) and Ü(j) i d
'PracticaII. the FOSAA can be implemented in sclccted cells depencling on the trafic conditions as we do
in the sirriu1;ition stiidy later.
be in-sector: out-of-sector and out-of-ce11 users respectively for base station j -
The desiretl signal for user i a t receiver a ( i ) is equal to ~ , ~ ( , ~ i > , . ivhile the intcrfering signal
power from other users is given by,
where the first: second and third terms come from intra-sector, inter-sector and inter-ce11
interference respectively- It can be seen that the 2nd tenn in (4.2) is generally negligible in
traditional non-FOS-1.1 system (Le., only soft-overlapping). However in the FOSA-1, this term
can be substantial and p lay a major role in the assignment of users to the antenna based on
the angular loads.
Eq. (4.2) can be written in a general form as,
Therefore, the received Eb/I, for user i can be written as,
LY G i a ( , ) R = - - , i = 1, ..-: N and a ( i ) E (1. .... J x B). r i Ia(i)
The problem is CO fincl a( i ) and f i subject to satisfying yi,Vi7 such that Z, pi is rninirnized.
The difference betiveen this problem and the one solved in [47] is that the clioice for a(i) has
increasecf L-fold with L different choices at each base station for each mobile user i. The
choice cen be made btised on the interferencc potver at the respective antenna. This is jointly
done by finriin:: the assignment vector AI and mobile transmit power veccor P. where
and
P = [P[ ... P.V]?
Since (4.4) sacisfies for al1 -V users in the system.
The above system of equations can be written in a matris form as.
(I - H}P = b,
Ynrnitn(n) where 1 is the N x !V identity rnatrix. b is the N x 1 vector defined by b, = c__, and H is
the 1V x !V (weighted normalized) link gain rnatrix (with m and n respectively denote column
and row indices) defined as,
The indices m, and n correspond to mobile and base station respectively. It can be seen that
H is non-negative. irreducible and stochastic. Hence, Perron-Frobenius [421 theop guarantees
the existence of a dominant positive eigenvalue A. If X < 1: the above nratri'r equation (4.5)
has a positive solution in transmit ponters tvhile sat is l ing the rate and service requirernents.
If the traffic configuration m e force h < 1 and hence no solution will esist that satisfies al1
the users' service reqiiirernen~s. *
There are nurnerous papers that ana ly~e the interference rnatris H and the solutions P
(sec [47.50] and refererices therein) for non-FOSA.4 system. it iras reported in [4T] that.
larger valire of X inclicates network cotigestion and as a resiilt higher interference power at
the receiver. Also notetl in [al] that srnaller value of X tviI1 yield faster convergence of potver
adaptation algorithrns'. We irill ciisciiss latcr in subsection 4.4.4 hon- ,\ is retlticeti in cliffcretrt
schernts.
4.3.3 Minimum Mobile Transmit Power
By introducilig additional CO-located antenna with overlapping coverage in the FOSA.4 sys-
tems, ive do not disturb the channel conditions but give additional Aesibility to users to sivitch
antenna since each user can communicate to more than one antenna with equally strong ra-
dio path. In the FOS.\A, each mobile user has the freedorn of selecting an antenna €rom an
increased antenna space. Cornpared to the non-FOSA-A, an L-fold increase in the number
of sectors (or antenna) can theoretically be espected if every ce11 is to use the overlapping
sectors. Hence, with the available antenna size of J x B, the size of the possible base station
antenna assignrnent vector is (J x B)*. Let a be one of the assignments and also ~ ( a ) and
H(a) be the corresponding mobile transmit power vector and normalized interference rnatrk
respectively. Hence, the mobile transmit power vector is said to be feasible untler assignrnent
a, if
In the above we assume that each user uses just enough power to achieve the desired quality
of service.
Hanly [47] and Yates (501 have solved the combined problern of selecting the base station
antenna and finding the mobile transmit power in general form and proposed an iterative
Base Station Antenna Assignment plus Power Adaptation algorithm which we calr BS.4--\+P.A
algorithm in this chapter. The BS--L4+P=\ algorithm that we discuss is the same as in [-4Ï] but
implemented differently using pilot power. In this iterative algorithm. base station antenna is
seiected as given by (-4.10) and the mobile transmit power is given by (4.9).
In FOSA.4 system, sectors are relatively shifted to overlap. \\te use the BS.LA+PA al-
gonchrn in finding the home antenna and adapcing the mobile transmit potver to minirnize
the total mobile transmit power. In the nest section. ire discuss the algorithnis for differerit
ceIl-sectoring cases with the c~iialitütive illiistration of these interfer~iice control cechntqu~s.
4.4 Spatial Interference Control Techniques and Algo-
rit hms
CDMX technology allows for al1 the available channels to be used in every cell/sector. There-
fore, CO-channel interference from within and neighboring cells/sectors contribute to the total
interference. Spatial interference filtering ['10,33,34] is the leading technique for SIR improve-
ment in CDMA networks and it can be achieved using ce11 sectoring techniques with either
fised or adaptive antenna.
r - LDR uxn ic) CS 0 : HDR uxrs l JFB+CS
Figure 4.2: Spatial interference control techniques. Low data rate (LDR) and high data rate
(HDR) users CO-exist in the systern. In (c) and (d): tlashed and solid lines denote the sectors
of two different groups of non-overlapping sectors. Dotted lines denote the shrinking of the
cell/sector size.
ive can replace the total interference Il fin (4.3)) by the eas-to-rneasure total receive
p o w r Q, shere QI = zk~k,&r since the difference ivill have tiegligible effect on SIR in a
large baticlwitlth spread spectrurn channel. Therefore. we use Q instead of I in the rest of
the chapter. Al1 of the following base station antenna assignment algorithrns are implernented
iceratively. Let n be the iteration index.
lit. tliscuss four types of spatial interference filtering tediniqiies iising an esample stioivn
iri Fig. 4.2 cniploying: (a) traciitionzil 3-sector anttBnria with fiscd BS-4-1 (b ) triiditiorial
scctor(r-011)-hreathing. (cl 6 sectors with o~erliippinp r.ovtbragtB pliis rio cdl-breitthing and ( (1)
76
6 sectors with overiapping coverage plus cell-breathing. The above schemes are respectively
referred to as fixcrl, ccll-breathing (CB), cell-slicing (CS) and celi-breathing plus cell-slicing
(CB+CS) in this chapter. In the first twot no FOS.4.4 is assumecl and in the Iast two, the
€OSA-4 is implemented. We include the fised and the CS schemes just for the cornparison
purposes thoirgh in practice the CB and the CB+CS are used. Our niain goal here is to
show the versatility and Hexibility of the CB+CS scheme in the FOS-\=\ in the face of spatial
interference powr variation especially with the Ao.1 and compare it with the CB scheme.
Let us consider a case where sector B is overloaded (Le., ho t-spo t sector) and sectors A and
C are not overloaded, as shown in Fig 4.2. In such a case: fked BSA.4 scheme does nothing
but block some users in sector B to avoid SIR degradation to already admitted users. In h e d
BSA.4: users connect to the strongest (in the radio distance sense) base station antenna and
therefore in general,
a ( i ) = arg ma.{GiL): Yi, 1 = 1: ..., S x B. 1
and the mobile transmit power is given in the literature [49,51] as,
The main problem with the fked BSA\ scheme is that the system accommodates mobile
users in congested cells/sectors at the espense of higher mobile transmit power. By allowing
the mobiIe users to switch base stations, we can reduce the total mobile transmit power as done
nest in the clvnarnic BS.4.L We nom allow mobile users to change their base station antenna
dynamicalIy baseci on radial, angular and both radial/angular load levels at the antenna in
CB, CS and CB+CS respectively. As mentioned earlier. we use the BSA.4+P.A algorithm
whicti is implenimted using the pilot powr control (See Section 3.3 for more details on how
to irnplenient pilot power based BS-\=\+P.\ algorïthm). At the nt" itcration of the algorithm.
bise station anterina 1 transmits its pilot porver inversely proportiotial tr i its total nieasuretf
powver, Ql(n), at its receiver. For user i, the receiwv pilot powver is:
where G:[ is the forwxrd link gain between transmitter (antenna) 1 and receiver (mobile user) i.
The pilot power strength has been normalized to unity for al1 the base stations. Mobiles sense
the receive pilot power and pick up the strongest that determines the base station antenna.
That is,
a ( i ) = arg ma..{il ), Vi. 1
(4.10)
The mobile transmit power. ~ i ( n ) is given by (4.9) with the home base station selectcd wing
(4.10). It should be noted that a ( i ) is selected based not oniy on link gain but also with the
congestion level consideration at the base station antenna receiver.
In the CB, we can assigti a mobile user to one of (possibly) many antenna of the adjacent
cells. In the CS, with overlapping sectors? we can assign to one of (at least) L antenna of
the same cell. In the hybrid CB+CS scheme. a11 possible antenna (in-cell ancl out-of-cell) are
considered in the assignment for evey mobile user. In al1 of the dynamic BS.4-4 schemes. ive
use the same algorithm; however. the difference lies in the pool of base station antenna that
is available as candidates for a mobile user-
4.4.1 Radial Cont roi and CelLBreat hing (CB)
CVe assume that no F0SA.I is implemented here. In the CB scheme. theoretically as well as
prac~ically. any niobile user can connect to any antenna in the systeni. provicled that there
is a strong radio path between them. Lsing the cell-breathing technique [-!TI. sonie of the
users in sectar B can be arranged to be servcd by the adjacent light cells/sectors as shown in
Fig. -l.2(b). Pilot power of sector B can be reduced t hat shrinks the sector size (or loari) and.
hence forcing the mers at the sector edge to link to other base stations. Here. the BS-4-4 and
the correspontIirig mobile trar!sniit potver are given h - (4.10) n-ith i = 1. .... S x B aritl (-1.9)
respect ivel-
This technique radial13 limits the users: i.eo users above a certain radio distance from the
base station can be controlled in terms of connectivity. Here, interference power tvas controlled
by assigning a mobile user to a lightly loaded (out-of-cell) sector/cell.
4.4.2 Azimuthal Control and CeU-Slicing (CS)
We have seen in Chapter 2 that one high rate/service user can cause a lot of interference to lom
rate/service users. If a high data rate/service user happens to be near the base station then
the cell-breathing technique would not heIp. However, with overlapping sectors. we have a
choice of assigning this user to one of ttvo adjacent (in-cell) scctors as s h o w in the Fig. - 1 3 ~ ) .
This assignrnent can be done according to the angular load conditions. Hence. this technique
azîmuthally limits the interference. rnaking it more suitable for non-uniforni interference with
the -40.4.
We implement the F0S .U here: hoivever. no mobile user can connect to an out-of-cell
antenna (Le.. no CB). In practice. there will be no system that implements only the CS
schenie. However, for cornparison purposes, ive consider the CS scherne. Here. a mobile user
i can connect to one of L (in-cell) antenna whose coverage potentially includes it. The BS.4.4
and the corresponding mobile transmit powver are given by (4.10) with 1 = l..... L x S and
(4.9) respectively.
It can easily be proven ttiat if a mobile user is ucder the sector-coverage of L co-located
antenna. it d l altvays select an antenna whose total interference potver is minimum. provided
that forward link gains between al1 L antenna and the mobile user are eqiial when irsing the
pilot power basecl BS.iX+PA algorithm. Therefore. due to the overlapping coverage. i t is
possible for the mobile users to bc handed off among co-locatcct antenna based on angiilar
loarl levels.
One can view as cclls bthg sl ic~d by the antenna in the çoverage using this sctierric
ticricr. tve cal1 this techniqiir cdl-sliczng, In the FOSAA. the BS.U+P.-! algoritfini s ~ i r r t s
the required set of antenna frorn a pool of k e d antenna: as a result, the cell-slicing natiirally
occurs. However, due to the nature of the radiation pattern, we can achieve the ccll-slicing
discretely only. By using a large number of Lied sectors or adaptive antenna array, we may be
able to get smaller "slices". This sector/antenna arrangement provides us with the Hexibility
to handle the hot-spots.
4.4.3 Hybrid Control and Cell-Breathing plus Cell-Slicing (CB+CS)
The cell-breathing technique is appropriate for non-uniform radial load and the cell-sliciiig
is appropriate for non-uniform angular Load. Hence, ive propose a hybrid scheme ernploying
both CB and CS (as shown in Fig 4.2(d)) to be used in real traffic conditions.
,\gain, we implement the FOSA.4 here. In practice, a communication channel is initistlly
established between a base station antenna and a mobile user through the pilot powr control.
If mobile users are assigned based on the received pilot power throughout the system. Le..
there will be no restriction on the base station antenna available for the assignment: in which
case, we d l in fact be implementing both the cell-breathing and the cell-slicing. Hence. the
antenna search space is increased L-fold in the CB+CS scheme cornpared to the CB schemc
because of the additional antenna used in the overlapping coverage.
The algorithrns that help select the BS.4.4 vector A, and the niobile transmit power vector
P are given by (4.10) tvith 1 = 1. .... L x S x B and (4.9) respectively. In a way! the FOS-1.A
provides an avenue for Hanly's BS.-\A+PX algorithm to breathe radially (continuoiisly) and
azirnuthally (discretely) further tlecreasing the total mobile transmit power.
If the loads in every sector/celI are the same, then there is nothing to be gained froni
the cell-breathing or employing ovcrlnpping coverage. Homever, in the face of Kuctuations in
traffic levels or interference with the -40-ly these schemes can perforni well. in gctieral. t h e
cnn be many (fkecl) overlapping seccors. not necessari[! wich eqiial overlapping coveragt*. and
there can be riarroiver sectors as wll.
4.4.4 Flexibility in the Feasible BSAA Vector
Let us analyze the matrix H that characterizes the solution. In the fked base station assign-
ment case, for user i, home base station a ( i ) is not changed once selected a t the cal1 setup
phase. Therefore, the matrïx component $ ' Y i . j remains unchanged. If a feasible solution
does not exist with a particular BSA-4 vector. G cannot be maneuvered to make it feasible
with another BS-4-4 vector. Hoivever, in the dynamic base station assignment case. the above
component can be maneuvered giving rise to many possibilities in t e m s of assignments. -4s a
result, the in-feasibility condition that existed in k e d assiprnent case may become feasible
with another base station vector in the dj-namic case,
In FOSXX, ive give additional flexibility in the choice of BS-4-4 vector. Let us assume that
the a cell-breathing scheme has converged to a BSAX vector with a particular matrix. Now
with the introduction of L - 1 additional CO-located antenna at a base station, each user i has
at least L - 1 additional possibilities witti equaI radio Iink to choose a ( i ) if necessary. The
choice of a( i ) is done depending on the interference power with Ao.4 a t the antenna receiwrs.
Depending on the selected a( i ) . ith row will have different values in the new matrix. In al1
the rows, the denominator remains the same since the link gain G,?V j rernains unchanged.
The numerator changes with the choice of n(i). Some users who were under the coverage of
previous antenna beam may not necessarily be inside the coverage area of the selected antenna
beam. Also some who were not under the coverage of the selected antenna previously are now
come under the coverage. Therefore. ive can see that users are given additiorial antenna choices
which will manipulate the matrix H to lead to minimal mobile transmit power solution.
4.5 Outage Probability Analysis in a Hot-sector
CVe now analyze the outage probability for users in a hot-sector using a simplified niociel. This
is tlorie to cornpare two cases. orle with F0S.U (iisitig the CS scherrie) and che other wichout
FOSS.4 (using the fisetl schenie).
In future wireless systems. the high data rate users are e'ipected to cause significant in-
terference to the low data rate uscrs and hence, the hot-spot (caiised by a user with killer
applications) can develop anywhere anytime. In our work. Ive consider the mobile users with
equal data rates but create a hot-spot with a relatively larger nurnber of mobile users. One
could use a m k of high and low data rate users to simulate the sarne effect, i.e., non-uniform
received power a t the base station antenna with the .\o.\.
The angiilar width of the antenna beam ( 8 ) can be made smaller to contain the signals of
interest to a smaller area thus lirniting the interference to others. Hence, hot-spots of different
sizes (that is measured in relation to 8 ) or intensity can be effectively handled. The srnaIler the
beani width the more the capacity gain, if a large nurnber of srnaller sized hot-spots develops
in a cell. However, this increases the soft handoffs due to user rnobility. We do not study
the optimal size of the angular width of the sectors to handlc the different number/size of
hot-spots, rather, 8 is fixed at 120".
.\ regular cellular CDhfiA network s h o w in Fig. 4.3 with circular cells of unit radius
is considered in the following analysis. Since the FOSAA system is expected to yield bet-
ter performance in non-uniform angular traffic. we consider a case where uniform and non-
uniform rnobiIe user distributions are assumed in radial and angular directions respectivel.
The F0SA.I rnay be implernented in the congested cells only; as such, Ive focus our inves-
tigation on a particular ceIl with non-uniform angular traffic. A hot-sector is created inside
the center ce11 as shown in Fig. -l*-l(a). Aritenna are nanied frorii I to 6 as shown in Fig. 4-4.
[t can be seen that antenna fit 1.2.3 and antenna # 4.5.6 form two non-overlapping antenna
groiips respectively. The center ce11 in Fig. 4.3 is covered witti 3 non-overlapping sectors in
the non-FOSJLX and 6 overlapping sectors in the FOS.-\.\ systeni. That is. tve implement a
[62[ FOS.\A (i-e., 6 anteriria with 2-degree of overlappirig) in the center ce11 only. Perfect
swtoring is assumed for al1 the antenna. The radiatiori pattern of each antenna covers the
eritire sector tvith unit gain everywhem. The other ct4ls arc ricither sectorecl nor overlapping.
$0 cell-breathing is irnplementetf for simplicity in the iiriiiIysis.
Figure 4.3: -1 regular cellular network model. Interference on the reverse Iinks is shown. The
center ce11 employs 3 sectors in (a) and 6 overlapping sectors in (b). The center ceIl experiences
congestion in a sector as shown in Fig. 4.4.
x : mobiles * : connected to rintcnna with dashed lint: O : connected to antenna with solid linc
Figure -1.4: A FOS=\-4 [6,1] is implemented in the cell. i-e.. 6 sectors with '2-degree of overlap-
ping: (a) a Iiot-sector has developeci and (b) a bypottietiçal h i ~ c station antenna assignrnent
for mobile ~iscrs. It can be seen that t ~ o groups of non-overlapping sectors in the FOS-\-\
[ 6 2 ] systeni.
Fig. 4.-4(b) shows a hypothetical base station antenna assigriment in the FOSA-4 for users
in the congested center cell. In this exaniple. the users in the hot-spot area are connected
to antenna # 4 and # 6. Xntenna # 1 and # 5 are not selected for communication. As
will be seen later from the simulation results, not al1 siu antenna will necessarily be selected
a t an instance. This is due to the relatively higher intra-sector interference to some of the
antenna. In the non-FOSX,-\. case. al1 three antenna in the center ce11 are used since there
is no flexibility in the choice of antenna for users. In the following analysis, we assume that
onIy four antenna are selected in such a way to distribute the interference power among the
antenna when employing the FOSAA.
We assurne that an equal number (N) of users is present each in every other cell. The
average inter-ce11 interference from other cells is proportional to the nurnber of other ce11
interferers. However, the inter-ce11 interference varies according to lognormal distribution.
The mobile users are assigned to antenna based only on the physical location (i.e.. neither
shadowing nor cell-breathing).
Hot-sector is created as follows: First. 31V users are uniformly distributed in the center
ce11 and then ON, (p 2 O) mobile users are added uniformly in the hot-sector. Mobiles are
synmetricalIy distributed about u = O" as shown in Fig. 4 4 a ) within the hot-sector. u is
rneasured anti-clockwise as s h o m there. Here. 13 can be regarded as a congestion control
parameter for the hot-sector. Hence. The probability density function of user distribution in
the center cel1 can be given as,
[-7 ot hern-ise.
It cari be seen that = O corresponds to a iiniform mobile tfistribiition in the center ceIl iintl
fi = 1 corresponds to a scenario where 50% of mobile iisers are in the hot-sector and the rest
in clic* i i t tier tn-O sectors of the center ceIl.
\\P iinalyze the receivecl Eb/Io for iisers in the hoc-sector aïeit only \Ve do not consider
the users in ottier sectors or cells in the ânalysis: however, al1 tlie iisers are taken into accoirnt
in the simulation later. Homog~neous users (Le., data rate. r, = r ? required Eb/I,, y, = y,
processing gain, Hi = H,tli) are assurned in the systern. The Eb/fo for user i in the hot-sector
can be written as,
where fintro and Iint,, are the intra-sector and inter-ce11 interference respectively In our inter-
ce11 interference model, linter = TN? for any antenna at the center ce11 base station. We
assume r to be a lognormal random variable with O dB mean a d B standard deviation. We
also assume perfect transmit power control with equal receive power levels for intra-cell users:
hence, = N,, where LV, is the number of users whose signais are received at antenna, S .
The randomness in the received signal power from the intra-scctors has been removed by the
perfect transmit power control. The outage occurs when the esperienced Ea/ lo is Iess than
the required Eb/Io. Therefore, the outage probability is given by. Pmt = P[yi < y/. We use
the statistics about r in finding the outage probabilit.
Non-FOSAA System (with Fixed Scheme)
In this caset the center ce11 employs 3 non-overlapping sectors. Fised BSX-4 scheme is imple-
mented (i.e., no cell-breathing). Here, the hot-sector antenna (fit 1) supports (1 + $)N users
on the average. Hence, rintra = (1 + 3 ) 9 atid Tint,, = riV. Therefore. (-1.12) can be tvritten as.
Therefore, the outage probability in the non-FOSA,l systeni.
CG p..-~osrr = / 1 (ln 2)' out
Xnon-FOSAA &UT
whicli can be sirnplified as,
where erfc() is the cornplernenta~ error function and is defined as erfc(x) = 5 e-y2d9.
FOSAA System (with Cell-Slicing Scheme)
In this case, Ive implement overlapping sectors (Le., FOSJLA[6,2]) and only 4 antenna are
selected when using the BSAX algorithm. Note that there is no inter-sector interference in
this case in the hot-spot area. Becailse, the antenna are selected splitting the interference
between two non-overlapping antenna. CVe consider the users in the hot-sector area only.
Antenna # 4 (or # 6) supports (1 + $ ) N users on the average. Hence, Following the derivation
for the non-FOSAA case, we can show that,
H 5 mhere XFOS.~;I = - - (1 + 2). N7
The erfc() is a non-increasing function in its argument. It can be seen that Y F O S ~ I . ~ >
\ ( n m - ~ ~ ~ - . ~ ~ , V , d > O; as a result, the FOS.4-4 systern will have lower outage probability for
ilsers in the hot-spot area.
Performance Cornparison: Fixed vs Cell-Slicing
\Ve asurne the followîng parameters in cornparing both schenies: N = 11. H = 1-28,: = 6.8s
clB. cr = -1 clB. We pIot the oucage probabilities in Fig 4.3 iaing the derivet1 forrniilae. (4.14)
x 1 c l (4.15).
o. 1 1 O 0.2 0.4 0.6 0.8 1 1.2 1.4 l.6 1.8 2
Sesor Congaanon Pinmater. 0
Figure 4.5: Outage probability for users in the hot-sector.
From the figure, we can see that as the congestion lever in the sector increases, the difference
in the outage performance betwwn the FOS.&% and the non-FOS.4-4 schemes becomes more
apparent with the former outperforming the latter. It can be seen that when = 3 (Le.. 50%
of the users in the ce11 are concentrated in a sector), about 25% improvement in the outage
probability can be expected using the FOSXA.
Using the flesibility in the antenna selection in the FOS-4-4. ive are distributing the in-
terference between two antenna that covers the hot-sector in the FOS.4.4. This Nil1 adversely
affect the users in the neighboring sectors/ceIls. Hoivever. if ive can retfuce the interference
powr mriation among active aantenna in the system. then WC cari expect to reduce the tom1
interference power in the system as shorvn in the nest section via simulation.
4.6 Simulation and Results
In this section, ive validate the analytical observations via computer simulation. The per-
formance measures such as the total mobile transmit power. Eb/I, statistics and minimum
number of antenna required are considered when implernenting different BSA-4 schemes.
4.6.1 Simulation Description
-4 network with 36 base stations is considered in our study. The base stations are located at
integer points (1, m) where 1, m = 1, ..., 6, as s h o w in Figs. 4.12-4.13. Let the ce11 around
(4. 4) be denoteti as the center cell. This ce11 is covered with 3 non-overlapping sectors in the
non-FOS.4.4 and 6 overlapping sectors in the FOS.\.& system. Fixed and CB schemes are
irnplemented in the non-FOSX-4 and the CS and the CB+CS schemes are in the FOSA.4. -4
hot-sector is created inside the center ce11 as shown in Fig. 4.4(,2). Table 4.1 shows the number
of antenna useci in the simulation.
Table 4.1: The number of antenna used in the simulation. See Section 4.2 for description of
L and S.
Schemes:
First. we have taken 400 mobiIe users and tlistribilted them randomly iiniformly over the
network. i-e.. owr [0.3.6.5]~[0.3.6.5]. Since only the center ceil is CO\-cred with sector antenna.
ive fiirther itilil 22 (= L? x %J) niobile iisers r;indornIy uniforrril- iiroiinil the b i ~ e station
there. Ttlerrforf~. thr total number of mobile iisers in the center cell. before creating the hot-
S=3,L=I1'i in hot-spot ce11
S=L.L=1 in other cells
sector, is approsimately 33 and they are evenly distributeci among sectors. Finally? 11 x @
mobile users are added randomly uniformly in the hot-sector. Thc link gain is modeled as
G, = i/4>,. ivhere dij is the distance beriveen mobile user i and base station receiver j . We
assume -fi = 6-85 dB, Hi = 128, Qi. Though Ive assume homogeneous mobile users in the
following, the results and analysis are equally applicable to the heterogeneous case.
CVe use the previously discussed iterative BSA-4 algorithrns (see Section 4.4) in determining
the home aritenna for each mobile user. For a certain set of mobile locationst the Eb/Io's for
al1 the mobile users in the network are compiited and recorded following 10 iterations of the
BSAA algorithms. This process is repeated 20 times to account for the randomness in the
locations of the mobile users. Hence, a total of 10 x 20 x (122 + 1 13) data points are collected
to plot the complementary cumulative distribution function (CDF) of the received Eb/ l0 in
the network.
Figure 4.6: Hot-sector is location-wise randomizetl.
Thts hot-spot cari deverop anyhe rc within a c d . in the simulation. the location of the
hot-spots are ranclorriized within a cell as follows: First. me gcnerate high intensity of trafic
in an arca covered by a sector of m-idth 120" and obtain the rcstilts iü- described earlier. Then.
a new hot-spot of angular width 120' is selected by moving the ceritrr of the hot-spot (v = 0"
asic in Fig. -1.6) by 130" counter-clockwise. 'I riis proceclirre is rcpeatrd 12 tirnes covering thr
entire r d l . it shot~lcl be notecl that the orientation of the iltittknnii rerriains fisecl duririg th -
abovc procedure.
4.6.2 Performance of CB and CS Schemes with Sector Congestion
Level
We first compare two schemes namely cell-breathing and the çell-slicing in delivering the
required link quality as the congestion in the hot-sector varies. By changing the value of
p, different amount of traffic is generated in a single hot-sector. -4s mentioned earlier. the
location of the hot-sector is randomized with the fixed antenna orientation.
Figure 4.p Complementary CDF of Eb/l , with sector congestion control parameter (9). A
(randomized} single hot-sector is included.
Fig, 4.7 shoivs the cornplernentary CDF of the receivetl Eb/f , for different values of 3. Lt
tan be seen that as increases, Le.. ivhen congestion Icvel iricreases in the hot-spot area. the
difference in performance betweeri che CB and the CS sciicrnes b~cotiies more apprirent. For
exxnpte. whtm 3 = 1. the CB anci the CS schenies tIciimr Eb/i , of 6.33 dB. 90% arid 99% of
Figure 4.8: Complementary CDF of received
Eb/Io for al1 mobile users in uniform anguIar
traffic in the center cell, f l = O. Yo hot-sector
is included in the simulation.
Figure 4.9: Complementary CDF of received
Eb/Io for al1 mobile users in non-uniform an-
gular traffic in the center cell. 3 = 2. A (ran-
domized) single hot-sector is included.
the time respectively while when ,B = 2, they deliver 65% and 96% of the time respectiveIy
as depicted in Fig. 4.7. This shows that the cell-slicing scheme can effectively support higher
degree of non-uniform angular load variation thnn the cell-breathing.
4.6.3 Cornparison of Received Eb/Io Statistics with Four BSAA
Schemes
We compare the Eb/I, performance of the BS.U schemes discussed in Section 4.4. -4s men-
tioned earlier, in ail the schemes, after 10 iterations of the algorithms, the received Eb/lo
of ail the mobile users were recorcicd Two cases. j3 = O and 3 = 2 were considered: mcti
respectivel~ simulating the i~riiforrn and nori-uniform angular traffic arouritl the base statiori
at (4.4). Figs. 4.8 and 4.9 st~otv the cornplertientary CDF ril the receivecl for al1 riirihilt~
9 1
users in the network. For clarity: a portion of the graph is shown iri a different scale.
Uniform Azimuthal Load (6 = 0)
In tbis case, there are 422 mobile users in the network; with 3 times mobile lisers in the
center ceIl as in any other cell. Sectoring is employed in the center cel1. With the above
feasible configuration of mobile users, the network can deliver the required quaiity (of 6.85
dB) to a11 the mobile users 99% of the time for al1 the cases incIuding the fked assignment
as shown in Fig. 4.8. It can be seen from the figure that al1 but the fked assignment yields
similar performance. As espected, there is no noticeable gain in employing the ceil-breathing
or the celi-slicing, because of the fairly uniform mobile user population and hence uniform
interference power with the Ao.4 at the base stations.
Non-Uniform Azimut ha1 Load ( P = 2)
The traffic intensity in the hot-sector is increased from 11 to 33 correspondhg to 3 = '7.
Consequently, the received quality drops significantly, especially when employing the fixed
assignrnent and cell-breathing schemes as s h o w in Fig. 4.9. The fixed assignment. CB. CS
and CB+CS schemes provide the required quality (of 6.85 dB), 33%, 63%, 94% and 98% of the
time respectively. With its inherent inability to handle the hot-spots: the fimd assignrnent
scherne suffers the worst while the hybrid schenie (CB+CS) perforrns the best. Both the CS
and the CB+CS perforrri comparatively since a e do not exploit the CB vep- niuch in the
simulation; because, our focus is on non-uniform angular loacis ancl not on radial loacls. CVe
c m see from Fig. 4.9 that if the systern is requiretl to provide a guarantee on the delivereri
Eb/lo 99% of the tirne. the CS and the CB schemes will deliver approximateiy 6.8 d B and
6 dB respectively- The gain in the Eb/lo for the CS scheme can be translated into capacity
increase in the CD&[=\ systerns.
1Ye have seen analytically in the prerious sertion that as the congestion in the sector
increases, the cell-slicing sr-tienie oiitperforrns the fisctl scherne- Sirnrdntion resir1ts shown in
Fig. 4.9 aIso draw the same conclusions6. For -/ = 6.53 dB. ive notice about 60% improrenient
in the outage probability between the CS and fised schemes in the simulation (see Fig. 4.9)
whereas 25% irnprovement was noted in the analysis (see Fig. 4.5) tvhen a = 4 ciB. The
difference may be due to several factors inclucling the assumption of lognormal distribution for
the inter-ce11 interference power and the value of a. However, both analytical and simulation
results point to the same conclusion that the cell-slicing (using FOSAA} can offer performance
improvement in the face of the non-uniform angular loads.
4.6.4 Total Mobile Transmit Power
Figs. 4.10-4.11 show the required total mobile transmit power with the number of iterations
for a11 four schemes for a set of mobile user locations in a hot-sector (when u = 0). Again,
twvo cases, @ = O and 8 = 2 are considered. -4s clear from the figures, the CB+CS requires
the smallest amount of power while the fised assignment requires the largest amount since the
former is the minimal solution. In the uniform angular trafic case ( P = O), as expected. no
significant difference in mobile transmit powr is noted. With the hot-sector scenario (J = 2)?
the fked assignment diverges (with no feasible solution) while others converge.
The cell-slicing algorithms (namely CS and CB+CS) select a set of antenna such a way to
avoid those that are experiencing h e a l interference (from usew in the hot-spot area) as can
be seen in Fig. 4.15. By Iinking to the lightly-loatied antenna of the same-ce11 (cell-slicing) and
not linking to the antenna of the other-ce11 (cell-breaching), mobile users operate ancf receive
the required service with lower mobile triinsrnit power.
The total mobile transmit power reqiiirenients for al1 the users are s h o w in Trible 4.2
for different schemes with ) = 2. Thesc results are reported follotving 10 iterations of the
algorithms and averaged over 12 hot-spot locations each with 20 mns witti clifferent mobile
user locations. Tt is clear from the table that the CB+CS scheme requires the Ieast arnount of -
"t ~hourd bc nnteci that P,,,t = P[Î, < y] = 1 - Pr;, > :l.
93
Figure 4.10: Total mobile transniit power in Figure 4.11: Total rnobile transniit power in
irniforrn angular traffic for a set of mobile user non-uniform angular Traffic for a set of rno-
locations? B = 0. bile user locations, $ = 2.
1 1 BSA-4 schemes 1
Table 4.2: Total mobile transmit p o w r for ciifferenr BS=\.\ schernes followingL0 iterations.
.\veragecl owr 12 hot-spot areas each with ZO sets of mobile user locations. 3 = 2.
Total MTP 23.35 37-43 18-72 15.04
power since it is optimal in minimum transmit power aiid the CB scherne reqiiires about 30%
mare p o w r than in the CBSCS scheme ahen 9 = 2. We also have to look at the de1ivered
quality of service mith this total transrnitted power. The Fig. 4.9 confirms that the CB+CS
scheme provides better quality than the CB scheme.
Since we have fairly uniform radial loads in this simulationr both the CS and the CB+CS
schemes perform comparably well. With radial load fluctuationsr we can expect to see the
CB+CS outperforming the CS scheme.
4.6.5 Antenna Selection and Coverage Area
We now see how mobile users are assigned to base stations by looking at the coverage area
for each base station. We group the mobile users that communicate with an antenna to form
a coverage area. Figs. 4.12-4.13 show the coverage when irnplernenting the CB+CS scheme.
Since our focus is on the center cell, this ce11 is shown separately in Figs. 4.14-4.15.
Figure 4.12: The network coverage (for the CB+CS) with uriiforrri angular trafic. Y = O. So
hot-sector is in(-liidtd. See Fig. 4-14 for the center cell coverage.
Figure 4.13: The network coverage (for the CB+CS) with non-uniform angular traffic. 3 = 2.
-4 single tiot-sector is included. See Fig. 4.15 for the center ceil coverage.
Figiire 4.14: Tlie venter ceIl coverage with iiniform iingiilnr trafic. 3 = 0- So hot-sector is
incf udcd
Figure 4.15: The center ccll coverage with non-uniform angular traffific. 3 = 2. -4 singIe
hot-spot sector is included.
Tt shouId be borne in mind that in Figs. 4.l4-4.15? the adjacent solid-line antenna and the
dotted-line-antenna overlap though it is not explicitly clear in the graph.
Uniform Azimuthal T r a c (,û = 0)
Fig. 4.22 shows a snap shot of the mobile user distribution in the network ancl antenna coverage
of each base station antenna when implementing the CB+CS scheme for 3 = O. [t is clear
that there is no clifference in the coverage area in Fig. -4.14 between iniplenienting CB aricl not
implementing it, in the case with tiniforrn mobile user distribution. The celI-slicing aigorithm
uses only four of the siu ancenna in scrving the mobile users as s h o w iri the bottoni two
graphs of Fig. 4.14.
Non-Uniform Azimuthal TraFfic ( / = 2)
Fig. 4-13 shows a snap shot of the mobile user distribution in the network and the antenna
coverage of each base station lzntenna when implementing the CB+CS scheme for 3 = 2.
Hot-spot ce11 is shown in Fig. 4.15 from which we can make the following observations:
1. The hot-spot ce11 shrinks radially in coverage when implementing the cell-breathing (see
topright graph).
2. The cell-breathing scherne is not sought when the cell-slicing can handle the scenario
(see bottom-right graph).
3. With the overlapping coverage. the antenna that tvould potentially receive most of the
interference power (e.g., in our case. that physically covers rnost of the hot-spot area) is
not selected when implernenting the cell-slicing scherne (see bottom two graphs).
Ping-pong Effect and Softer HandofT
By introducing additional antenna with overlapping coverage and selecting dynamically only a
set of antenna that minimize the total mobile transmit potver, wve increase the processing (Le..
switching the mobile users betmen antenria) a t the base stations. If the received signal varies
rapidly with the .\o.\. then the set of selected antenna tvill also change. It should be noted
that even when an antenna is not usccl for communication. it will still be trarisrnitting pilot
power based on the received power in otrr CDILI.\/FOSAA model. This will allotv mobile uscrs
to switch to this antenna if necessa- Therefore, when the receivetl signals txry rapidly. tliis
wiI1 catise ping-pong effect on the antenna choices for mobile iisers. Since the users are hancled
off betireen sectors of the sarne base station (Le., softer handoff). there is no signaling overhead
involvecl between base stations. Deprrirlirig on the degree of traffic fluctuations. the riuriit)t~r
of hatitloffs will be large or sniall. Thib haridoff may inctir deiay and rnxF not be acceptablt* CO
sornp &Iay sensitive applicatiotis sticti as voice. Thereiorc. ttir proposed architecture will h t ?
more suitable for clel- tolerant applications otily.
4.7 Chapter Summary
The problem of base station antenna assignment with minimum mobile transmit power was
studied for CDMA networks with fixed overlapping sector antenna architecture where more
than one co-located antenna was used to cover any space in the network. i t was first noted
that the non-FOSXA has limitations in switching users between in-ce1 sectors. It was then
shoivn that by employing overlapping sectors in the FOSAA, we can exploit the flexibility
of assigning a user to one of possibly ma. potential antenna to effectively support the non-
uniform anguIar traffic. It was also proven that the problem of selecting a set of antenna from
a pool of overlapping antenna and assigning the risers to them in the F0SA.i with minimum
mobile transmit power is a specia1 case of a gerieral problem for which the solution has been
reported in the literature. The process of dynamic cell sectaring was differentiated twofold as
cell-breathing and ceIl-slicing and the latter can be viewed as being the angi~lar cotlnterpart
of the former radial scherne. The hybrid scheme, CB+CS: was sho~vn to yield the optimal
solution in minimum total mobile transmit power in the FOS-LI system.
The simulation results clearIy demonstrated the effectiveness and the HexibiIity of the
FOSAA in handling non-uniform angular loads. The FOS-4-4 outperforms the non-FOS-4.4
in the delivered signal quality in the hot-spot scenario. As the congestion lever increases. the
difference in Eb/ l , performance between the CB and the CS schernes becornes rriore apparent
with the latter oritperforming the former. The hybrid scheme. CB+CS. outperforms a11 the
other exploiting breathing and slicing techniques, in minimizing the total mobile transmit
power in the system. Though the CS and the CB+CS schemes performd coniparativeIy
w e l in our simulation: the latter hybrid scheme is espectcd to do better in the presence o l
non-uniform radia1 loads and in shadowing radio erivironment. Ttiese performance gains w r e
realized at the espense of increifitul nimber c i l antenna. though only a srtbsec will be seler.teti
any time depending on the traffic conditions. Also the niimber of softer handoffs ririthout user
mobi1ity that would occur in this architecture depends on the degrce of traffic fluctuations at
the base stations with the angle of arrival.
Chapter 5
Distributed Inter-ce11 Inter ference
Control
In CDbI.4 systems, the main source of iriterference is the multiple access interference which
cornes frorn interferers within the ceIl (intra-ceII) and outside the cell (inter-cell). It is difficult
to coordinate the transmissions from users in different cells to minimize the inter-ce11 interfer-
ence. .A centrahed base station controller can pcrform the coordination with the knowledge
of the channel gains betweeu al1 the mobile users to al1 the base stations. This may be practi-
caIIy possible in the point-to-multipoint communication (as in the fortvard links); hotvever. it
is difficdt in the reverse links due to the multipoint-to-point communication. In this chapter,
we propose and study a distributed mechanisrn to minimize both the intra-ceII and inter-ce11
interference in the reverse links. It is a heuristic method that esploits the real-tirne channel
conditions anci that delays/avoids the dominant interferers.
5.1 Introduction
I t is espectecl that the nest generation rrjreless necnorks ivill be mninl!
hear-!- and a sizablr portion of the traffic d l be tfcIay-rolcrant because of the it-ireless [riternet
10 1
applications. In such networks, mobile users contencl tvith the time-varying radio channels
for the successful transmissions of the packets. Therefore, proper interference management
techniques are needed to make mavirnal use of the limited wireless resources. This can be done
taking the channel conditions and the application requirements into account as discussed in
this chapter.
A scheduling scheme for high data rate transmission \vas proposed for the forward links
in [Ci-$ There: each base station schedules only one user to receive packets in each slot. Intra-
ce11 interference was not a main issue there since only a single user was active in each s1ot in
each cell. If more than one user closer to the ce11 boundary' do happen to be scheduled in the
same slot, each causes enormous inter-celI interference to each other. In such scenario: inter-
ce11 coordination is required to manage the inter-ce11 interference: othenvise. the performance
will be degraded significantly. The above problem also esists in the reverse link case. The
coordination of the transmission in the reverse links is more difficult than in the fomard links.
In our work. we consider minimizing the inter-ceII interference without inter-ce11 coordination
in the reverse links where one or more users will be scheduled to transmit packets in each slot.
In the proposed technique, the intra-ceI1 interference is also implicitly reduced: hoivever: our
discussion in the sequel rnainly focuses on inter-ce11 interference reduction.
The basic iclea behind the proposed scheme is iis follows: if a mobile tias good (in the relative
reverse link channel gain sense) instantaneous chmnel gains with more than one base station,
it i d 1 del- its transmissions in order to reduce the inter-ceIl interference. This mechanism
limits those mobile users that will cause higher inter-ce11 interference. In our scheme. rnobiIe
users that not only have stronger channel gains to their respective home base stations but also
catise relatively lower inter-ce11 interference are sçhecluled [or transmissions. Though a mobile
user with a stroriger link gain to its home base station is somecimes not allowed to transmit
because it causrs higher interference to ~ h e iisers iti üt lerrst one otlier cell. it is sliown that
l . l ~ ~ s ~ for il1usrr;irive purposes. WC assume tlistaiicc-depentlenc patti ioss mode1 here.
the system benefits, because, thase a[Ioived to transmit get increasecl data rates because of
the reduced inter-ce11 interference. The un-scheduled mobile users that had stronger !ink gain
but caused reasonably higher inter-cell interference wait for their turns. The proposed scheme
is implementeci by the distribution of tags by recekers among transmitters. The number of
issuetl tags is a system parameter. tt depends on the expectecl trafic. the number of cells
and the propagation conditions in the network. Different types of tags can be issued giving
additional Hexibility in the interference management. The proposed scheme is suitable for
variable bit rate: delay insensitive applications such as wireless Internet data.
The idea of delaying the transmissions during the adverse channel conditions has been
investigated for data applications [48,65] in CDMJI systems. There. only the interference to
the home base station was considered in delaying the transniissions. In another work [49I in
narrowband systems, the users with weaker links were removed from the active user set in order
to help others with better channels. The procedure there is very computing-intensive because
the removal is done one user at a time and after each removal the SIR-balancing is done to see
il any more users have to be removed from the system. In our scheme. ive consider interference
to the home base station and other base stations as well. and implement it without any global
coordination, However, in the proposed scheme. each base station nieasures the signal strength
of theoretically al1 (but practically those in the neighborhood) users and transmits the tagging
infortnation in the control channel. Then. each mobile listens to many (or at least to those
with strong pilot power) control channels to rnake a decision on the packet transmission.
The rest of this chapter is organizecl as follows: -4 CDSIA systerri with time-slotted struc-
ture is tlescribecl in Section 5.2. Then. we discuss how the tags arc mariaged and transmission
decisions are macle in Section 5.3. Rate setting and power adaptation algorithms are also dis-
cusserl there, Cn Section 5.4, it is shown that the proposed schetliilirig scheme will outperform
the traclitional coilnterpart in achieving the minimum transmit bit cnergy-. Performance resiilts
from clic siriitilation are discusscci in Section -5.3 lollowed b - ttir siinimary in Section 5.6.
5.2 System Model and Problem Description
Interference avoiclance and control can be done by means o l ciynamic resource allocation at
the do t level. In this chapter, we consicler minirnizing the inter-cell interference in each slot
via intelligent schecluhg of packets.
5.2.1 Tirne-slotted CDMA System Model
CVe consider the reverse links oEa c e h l a r CDU.4 system that supports transmissions of packets
in slots (cg., CVCDM.4 or TD-CDMA). In such a system. in the same frequency band and tirne
tirne TDMA- feature
data + ovethead user 1
data + overhead
data t overhead
Figure 5.1: Frame/slot structure of a CDM-4 systenr.
do t 1: :VI mobile users are active. each using a mobile-specific spreaciing cocle. The slot/franie
structure of this CDM-4 system is illustratecl in Fig. 5-1 where IC: .st TI ancl T tlenote the spread
spectrurn banrlwidth, the number of dots per frame, the duration of a frarne and the duration
of a slot respectively- In this time-slotted sFsteni, transniission in each dot is controIlec1 and
transmitters are assuniecl to be capable of cransmittiiig diffcrcnt rates i n different slots.
5.2.2 Inter-ce11 Interference Minimization Problem
Our goal is to achieve mauirnum throughput with mininiiini transmit power by controlling
the individual user's transmission. A user is scheduled depenciing on how much interference
(both intra-ce11 and inter-cell) it generates- First, a base station seIects a set of iisers with
strong channel gains with it for transmissions (effectively those cause minimal intra-ce11 in-
terference). Then, those users that cause relatively higher inter-ce11 interference are blocked
from transmissions.
The link gain is modeIed üs Gq = LijSij/d;? where dijt Si,, Lij are the distance. slow-
fading and fast-fading components between mobile user i and receiver j respectively. a is
the propagation constant. For mobile user i, the home base station is chosen to be O if
G:, > G:,,Vj # O. rhere G8 = 5. That is, the fast-varying component of the received signal dpi
power is averaged out in selecting the home base station.
Figiire 5.2: Siinimizing the inter-ceII iriterference.
Fig. 5.2 shows a mobile iiscr 1 wmrnunicating with base station j . It also causes iriterference
to other mobile iisers coniniiiiiirxting rvith other base statioris. in the schetliiling of packets
105
based on the real-time channet gains. user i will be typicallv schectuled when G,, is I q e .
Howei.er, icleally, ive nant to schediile it packcts h m user .i when & is large, i-e.. wlien
the channel gain i l t h the home b u e station is scronger and the chnnnei gains to al1 other base
stations are weaker. This d l invaIve a large number of measurements and message passings.
Therefore, ive resort to a heuristic approach in which the dominant inter-ce11 interferers are
delayeci from transrnitting via tag contrai. Tagging helps detect the dominant interfererç
rvithout resorting to centralized control.
We assume a Iarge-capacity system in which medium access interference is the dominant
source of interference; hence, thermal noise is neglected. Let b(i) ,r i , fi be the honie base
station, transmission rate and mobile transmit power of mobile user i respectively. Hence.
the desired signal at base station receiver b ( i ) is equal to ~ i b ( i ] i > , . while the interfering signal
power from other transrnitters is.
Therefore, Eb/Io [or mobile user t , is given by,
CI' ~,&>i: "fi = -- Vi.
rz [b ( i )
Since data rates are adjustedo the proper performance mesure of interest is the transmit
bit energy which is defineci as the ratio of the dispensed transmit power to the achieved
throughput. Achieved throughpiit is cornpiited aiter accounting for the bit error. Let t h e
achieved throughput for user i in dot 1 be Q i ( l ) which depends on the r, and -fi in slot 1. Ttien
the transmit bit eriergy of user i is defineri S.
"") if user i is scheduled in dot 1 ; m: oit[) =
where j>,(l) is the transmit porver of user i iri dot 1- The awnge transmit bit energ.- in t h
systeni is giwn b - 8 $EL, Er 8,(1). where !V is the total nimber of mobile users tliiit
were admitted in the system. It shoiild be noted that out of !V iisers atlrnitterl into the systeni.
only !VI will be alloaed to transmit in dot f . We w n t to minimize the average transmit bit
energy in the system by way of intelligent packet scheduling. This is achieved by exploiting
the fast-varying temporal variations in the channel conditions in the scheduling of packets.
Fig. 5.3 shows the (simulated) signal strength with time in Rayleigh-faded radio environment
for 120 slots.
1 O i r
Figure 5.3: Fast scale signal variation in radio environment. Shadow fading is assumed to be
invariant for the duration of 120 dots.
5.3 Tag Cont rol and Interference Management
The sucress of the proposed packet schecltiling scheme depends on the tag tlistrihiition and
control algorithms.
5.3.1 Tag Control and Scheduling
The tag control mechanism determines the aiimber of tags. the types of tags aricl to whom to
issue the tags in each dot. Optimal tag control along with rate ancl power control is coniples in
the tirne-varying radio environment. becaiisc it is difficult to predict who will be transmitting
in each slot due to the random nature of the channel conditions. Therefore. we resort to a
heuristic approach whereby every base station issues a predetermined number of tags (M)
a t the beginning of each dot. The nurnber of tags issued will depend on the number of base
stations, the masimuni number of mobile iisers that can be supported in the system and the
radio environment in the network. This nimber can be approximately calculated for different
modes of operations (more details will be available later in the Simulation section).
In the proposed scheme, each base station measures the reverse link gains from each mobile
user to itself. Then! it ranks the gains in decreasing order and issues tags to the I I iisers that
have the strongest link gains with it. Therefore. in the proposcd scherne. JI = 1 corresponds to
the spread-spectrirm TDM mode of operation and M > 1 corresponds to scheduling multiple
sirnuItaneous transmissions as in traditional CDbIX systems. it is assumed t hat mobile users
can listen to multiple control channels from base stations in each slot for tag information. Tag
control (discussed in Section 5.3) deterniines the nurnber of mobile users that are allowed to
transmit in each slot. The number of schecliiled mobile users in slot 1 is NI(s .tl) since the
dominant inter-cell interferers arc blocketl from transmitting.
Different types of tags can be issricd to give Besibility in the decision making process at
the mobile users. For esample. a s_vstem ni* decide to issue twvo types of tags. hurd and soft.
Hard tags can be issued for the top qiiarter of the mobile iisers that have the strongest reverse
link gains and soft tags for the nest quarter. The rest of the mobiles clo not get any tags. This
type of tag distribution forms three partitions in the radio channel space. if shadowing and
muiti-path fading are noc presenc. t h will form three groups of mobile mers in rings arouncl
the base stations. One c m gencralizr ttiis iciea for a larger niirtihcr of tag types. Sest. IW
discuss how hard and soft tags can be used by the mobile users in the decision niaking process.
If a mobile user receives a hard tag, it mcans thiit it has a strong link gain rvith the base
station that issued the tag. If that base station is riot the mobile user's home base station
(and assuming no macro diversity), then if this rnobile user transmits, it i d 1 cause high inter-
ce11 interference. On the other hand, if a mobile user receives a hard tag from its home base
station: then it signifies the strong channel condition for transmission. Also, if a mobile user
receives a soft tag from its home base station, then it cloes not have a strong channel condition
but if it needs to transmit for some other reasons such as already-longer starved time For data
transmission, it can do so. If it receives a soft tag from other base stations, it signifies the fact
that it would not cause significant inter-ce11 interference if allowed to transmit. Thecefore, in
order for a mobile user (i) to transmit. two conditions have to be satisfieti. They are:
r it shouId receive a tag (either hard or soft) from its home base station. b ( i ) . and
a it should not receive a hard tag from any other base stations, j # b( i ) ,V j .
The above conditions respectively ensure that only those mobile users with strong link gains
are scheduled for transmission and that this transniission would not cause higher inter-ce11
interference to other mobiIe users. Table 3.1 shows the decision making process at the mobile
user. It shoultl be noted that our scheduling schenie is irriplemented in a faster scale? perhaps
every slot, i.e.: 667p s. Soft handoff mechanism helps mobiles rnaintain links during the
rriobility and deep fade scenario. The soft hantloff decisions are typically macle not on dot-bu-
slot channel condition basis rather in many slot intervals. There can be a relationship bettwen
the proposed scbcme and the soft handoff mechanisn:. It al1 depends on the clecision ride that
il systern implements. In one soft handoff mechanism. if one base station (from the active base
station set) directs to decrease the transmit power. chat is to say in our case7 that base station
has issued a hard tag (due CO strong çtiannel concliciotl). in the soft hancloff t~iocle. the mobile
irill mduce its porver and in out case. if the received tiatcl tag is From the home base statioii
only and none froni others. then niohilcv arcb allomed to transmit. The variorrs decision riiles
109
Hard
Nil 'Yil/Hard/So€t
#
1
2
3
Table 5.1: Transmission decision based on hard/soft tagging.
and their relationships to the soft handoff mechanism can be further investigated.
b(i)
Hard
Hard
Soft
A n Exarn~ie: Decision Mak ine Process
We illustrate how the mobiles decide to transmit upon receiving the tags using an example.
Fig. 5.4 shows two base stations (A and B) and five mobile users (1 to 5). The number of
tiard and soft tags issued is 2 each (hl = 4). Link gains are shown adjacent to the links.
For each mobile user, its home base station is shown inside 0. The active (Le., scheduted
for transmission) and inactive links are s h o w with solid and dashed lines respectively. The
letters besides the mobiles. h and s? respectively dcnote the fact that the mobiIes are hard-
taggecl and soft-tagged by the corresponding base stations. fié use the Table 5.1 in making
the tlecisions. The mobiles 2 and 1 will I c scheduled because they satisfy conditions # 2 and
.-j respectively. The mobiles 1. 3 and j will not be scheduled because of the conditions # 4- 6
and 3 respectively.
Soft-tagging offers fiexibility in allowing mobile users that cause moderate inter-ceII inter-
fereiice to transmit uncter certain conditions such as longer starved time. Another use of soft
taggirig is in congestion control. For esrirnpIe, if a sytcrri ttsperiences service <legatfatiori due
to congestion, then al1 the soft tagging can be t r ea td as tiard tagging and the transniission
j # b(i).V3
Yil
Soft
clecision
Yes
Yes
Hard L
N o
Figure 5.4: Tagging and scheduling. hl = total niiniber of (hard and soft) tags issued by each
base station.
procedure !vil1 become as follows: only if a mobile user receives a tag €rom its home base
station and none from other base stations. will it be allowed to transmit.
CVe assume that there exists a dedicated control channel (CCH) for each physical (PCH)
channel so that each slot can be managed. Depenciing on how many types of tags are issued,
a different number of bits has to be used in the CCH. We expect that the number of differ-
ent types of tags to be small (between 1 and 4) and hence the tagging inforrnatiori can Lie
transmitted in the CCH using 1 or 2 bits.
5.3.2 Resource Management
\'e have discussed how the decision is made at the mobile users to transmit or retreat wheri
they receive tags. Sest, tve discuss how to determirie the transmission rate and power for each
aiobile ttser.
Transmission Rates
tt is clifficult to preciict exactly which mobile users are transmittirig in each slot due to the
nature of the proposed distributed schecluiing in the ranclom radio Channel conditions. In
general, the data rate has to depend on the required average rate and minimum/mxximum
rates for each mobile user. If Ive know who transmits in each slot and the necessary link gains.
then transmission rates for each mobile user may be computed by a centralized controller. We
do not consider the rate-determination problem in detail in chis ~vork, but instead use a simple
method. The maximum possible data rate is shared among al1 the transmitting mobile users,
Since the minimum spreading gain is 1. the rnaximum data rate is set as r,, = W. The data
rate for scheduled mobile user ( 2 ) is set as follows in slot 1:
whem K(M) is the average number of mobile users that would transmit tvhen .CI tags are
issued by each base station. More details on how to determine K ( M ) are available later in
the Simulation section. The above rate-setting fairlu shares the allowable transmission rates
among scheduled mobile users in the systern.
Power Adaptation
The corresponding mobile transmit porver for the chta rate set h - (5.4) is given as:
where 1,(1) and G,,(l) are totaI interference power at base station j and reverse link gain
betwen base station j and niobile i respectit-el? in slot 1. The niobile transmit power in slot
( 1 t L) depencls on the required Eb/Io. transmission rate. link gairis and the total interference
power at the tiorrit. hase station in dot 1. Ué assirrnc that the signais do not var>- drastically
in relatioti to the slot size so chat the rate setting arirl poner aclapcaticin algorithms car1 irork
effectively.
5.4 A Case Study: Achievable Data Rate in Traditional
and Proposed Schemes
CVe analyze the proposed (tagging) and traditional (non-tagging) schenies using a simple mode1
where twvo base stations (A and B) ancl two mobile users (1 and 2) and are present in the
system as shown i n Fig. 5.5. ;LI = 2 (1 hard and 1 soft ttag).
0 : B S I.2:MU ():homeBS :active link - - - - - - - : inxrive link
Figure 5.5: An example of tagging mit h twvo base stations and two mobile users.
5.4.1 Need for Tagging
The neecl for interference control (via tagging) is disctissetl nest. For sirnplicit_v. let us assume
that the home bifie station of mobile users 1 and 2 are A and B respectively Le., e l .+ > G I B
and GZB > C2.-l. 30te that Ci, represents the Iink gain between mobile user i and base station
j. The data rates are denoted by r i and rz respective1:. for mobile tisers I and 2. In a sIot.
With t h same required Eb/Io and equal amount of tlispensetl power. i.e.. y; = 1: P, = P. i =
l,?.
Gl.4 r i x - G2 B and r2 cc -. GA Gl B
That is: if and Gis are both large. then rl tvill be large and r? will be small. The
same argument goes also with respect to GZB and G2-4- Therefore. in general? if a mobile
user has strong channel conditions with its home base station as iwll as with another. then
it has irndesired situation as far as the system is concerned. That is, its signaIs rnay be
received with good quality a t its base station, however, it causes increased interference to
users comniunicating with other base station. CVe del- those irsers that cause larger inter-cell
interference. It is expected that due to the fast-varying nature of the channels. the deIayed
users will be able to get their turn sooner or later to transmit packets.
5.4.2 Performance Cornparison
It is obvious that the mobile users with stronger channels have to transmit lower power than
those with weaker channels to receive the same quality of service. it can be argued that in
the proposed scheme, mobile users will be transmitting more power than othenvise required
(in the traditional scheme) since we are blocking some mobile users with strong channels and
allowing some other mobiIe users with less-than-good channels to transmit. In fact. we biock
only those mobile users that cause higher inter-ce11 interference and as a result, the interference
is reducecl at the base stations tvhen other mobile users' packets are ciecoded a t the receiver.
Since thc base station operates with redticetl interference it reqiiires its scheduled mobile osers
to transmit lowr power. Therefore. it can bc seen that the tliffercnce in the increased mobile
transmit power dire to the scheduling of iisers with less strong channels is made up by thc
decrease in the reqilired mobile transmit power by the base station. If the difference is large.
then the s-teni benefits. Let ils now compare both schenies again iisitig the esample stiorvn
in Fig. 5.3 to show chat the proposed schetne hris ttie at1r;iritagt in temis of acIiiembIe data
rate-
Ttadit iond Scheme
In the traditional scheme, both mobile users are allowed to transmit and hence?
LC' C.> kV G and sirnilarly r2 = Hence, the total instantaneoiis slot data rate is T(&$ + 2).
Proposed Scheme
Let us assume that each base station issues 1 f m d tag and 1 soft tag. The relative strength
between link gains GLa4 and G2.4 (GLB and G2B) determines which mobile user gets the hard
or soft tag from base station A (B). There are four possible combinations with EWO set of link
gains. It can be shown that one of them degenerates to the case in (3.6) and the other two
will not change the transmissions compared to the traditional scherne.
Cl4 - > 1 and Gl5 - > 1. G2il G25
Therefore, we have either the same situation as in the traditional scheme or the situation that
resuIts from the case (0.6) and we will focus on the latter case.
By consulting Table 3.1, mobile user 1 decides not to transmit because it received 2 hard
tags and mobile user 2 will transmit since it received 2 soft tags. Let us nosv compute the
achievable data rate in this situation. Theoretically. there \vil1 be no interference since niobile
user I is inactive and there is no thermal noise at the receiver. Let us look at a worse case
in which che data Stream is transmitted using two non-orttiogotial codes (causing rnutual
interference): eacti carrying equaI amount of bits. Then. it can he showri that the total data
rate is f- Tlierefore, we can see that the proposed schenie « d l do better tfian traditional
scheme in ternis of achieved data rate if
Tliat is. if conditions in (5.6) and (5.7) premil. then the proposeci scheme mil1 outperforni the
traditional scheme. It is expected that in the time-varying radio environment, this would be
possible. We have iised a simple numerical example for illustration. In the next section, we
implement our scherne in a larger realistic environment and analyze the performance.
Simulation and Results
CPé evaliiate the performance of the proposed scheme via computer simiilation. We compare
it with a scheme where :CI tags are issued as in the proposed scheme. but al1 itl are allowed
to transmit. This scheme is referred to as conventional scherne in this section.
5.5.1 Simulation Description
Cellular System
-4 network with 9 base stations is set up for the study as shown in Figs. 3.7-5.8. We have
taken 100 mobile users and distributed them randomly uniforrnly over the network, i.e.. over
[0,3]s[0.3]. In our simulation. we assume the folIowing parameters: Cl;. S . Tf and T are 1.23
MHz. 15. 10 ms and 667 ps respectively. We assume that the channel conditions rernain
invariant within a slot. We anal-me the system performance with tlifferent values of issued
tags: 31 = 1.2.5.10. Only hard-tagging is impiemented in the sittidiitiori.
Mobile users and base stations use single ornni-directional antenna For transmission and
reception respectively. The radiation pattern of each ariterina çovers the cntire ce11 with tinity
gain ev~rytvhere. Since our scherne exploits temporal channel variations. ive nin our algorithms
for about S secoricls (or eqiiit-aleritly 1'300 dots) with shatlow aritl Riiyleigh faciing. The siorr-
fading is assiinied to be log-norrriiilly distribtited with nirari O tlB arid stanclard deviation of
S dB. The fast fatlirig is zissunicd to be R q k i g h discri1)iitrd. The chiinnel gains bettvreri
mobile iiscrs antl biise stations arc iusurnetl to be inclepenclently icleritically clistribritetl antl
the path-loss cxponcnt, a is takcn to bc 4. Perfect transmit potver control is assumed ancl no
soft hancloff is consitlered.
Once a random location for a mobile user is determined its location is unchanged for
120 sIots (eqtiivalent of SO nis) in order to include the temporal fast-varying channel fading.
We assume that the mobile users are rnoving slowly and hence the change tlue to tlistance-
clepencIent path loss and shadowing loss remain iinchanged for the duration of a t l e s t 120
slots. After 120 slots, a new location is determined according to the iiniform distribution. A
total of 100 such locations are generated for each mobile user.
CVe denote a block as having 1'20 dots and hence the shadowing is realized from block to
block. 'Ve take measurements in each slot (667 ps), then for each block (80 ms) and finally
for the dtiration of the simulation (S sec).
Rate Setting
We use the previously discussed algorithms in determining the transmission rates from (5.4)
and transmit power from (5.5). for every scheduled mobile user in each slot. Therefore. we
need to know more about K ( M ) . We performed an e.xperiment in the simulated network to
compute K(M) for different ancf the mean statistics are depicted in Fig. 5.6. There. the
average number of mobile users scheduled to transmit by each base station in each slot is
shown for different radio environment verstis the number of issued tags. M. For example.
when 10 tags are issiied by each base station. then on average about 7 or S iiserç tvill be
schedulecl from a ceil in the tirhan environment. In the case of the conventional scheme. we
take K(M) = :II.
Throughput Computation
\Te ttse ttie following methocl to conipute the throughptit. Let C(L) cienote thix size of ttie
pack~t transmitted from user i i t i slot L. For siniplicity. tve assunie that one dot fits a p2icket
Figure 5.6: The average nurnber of scheduled mobile iisers in ench dot (frorn the ~irii~liition).
of size, C i ( l ) = ri(l)T. The crror detection is donc on a dot hy slot basis. WC definc the
achieved throughput of mobile user i in slot L as,
where f(.) is the Eb/~o-tieperitient bit error rate function. It depends on the modulation
scheme being used. The r,(l) is the transmission rate and is the experienced Eb/Io for user
i in dot 1. The pr3bability of correct reception of a bit. when the received Eb/Io is 1. is given
by 1 - f (x). With the additive white Gaussian multiple access interference asstiniption and
using the differential BPSK modulationl ive can take f (x) = O.Se-'. Eq. (.3.5) lielps us to
take Eb/Io into account in cortiputing the achiemble throughpiit. The achievtd tliroughput
for each user is computetl in each dot and aggregated for ~ h e block and then for ctie systern.
The rnethod of cornpirtirig thti throughput is not crucial to our scherne as long as the same
methocl is irsed in conipiiring the sctienies-
Sest. ive conipire t h e performance of the proposecl scherrie it'itti conventioriiil 'içhenie. In
particular. ive invcstigi~te the fol1oiving: tdio gets scheciuled in each dot- ivh;lt tiiruiighpiit
Figure 5.i: Scheduled mobiIe users in distance and shadowinp channel coriditions.
individual user and the system achieves during the siniulation time of S sec and the dispensed
total mobile transmit energrr in the system.
5.5.2 Location Independent Packet Scheduling
We want to see hotv the coverage of mobile users is infiuenced by the proposccl scheduling
scheme. It can be argued that. when using such a channel-basecf schedding scheme. some
mobile users will be disadmntaged due to their phyçical locations with respect to the base sta-
tion locations. For esample, when there is no fading (Le.. only with distance-tlepentlent path
10s): mobile users phpicallv close to the base stations will have the advmtage of transmitting
al1 the time. In such a case. our tagging scheme limits those rnobik users at the celt boundary
from ever trarismi tting. Honever, in real propagation conditions. there i d be fading (both
slowv-sçale ancl fast-scale): as a result. the distance-clependence effect wilI be alIeviatecl and
those mobile users tvith goocl radio distance to the honie base station and relatire-ely worse to
the 0 t h base stations id1 gr3t ci, transniit.
Figs. 5.7-5.8 stiow mobik iisrrs (denoted nith *) that are scMdeci Tor trëinsrriissiori at
Figure 5.8: Scheduled mobile users in distance and shadoming channel conclitioris.
2
1 S
1
as-
an instance of the dot transmission For different values of issttec1 tags when shadow fading is
present, We note from the figures that tlie fading decreases the dependence of the coverage on
the distance betiveen the mobile users and the base stations. Mobiles closer to the home base
stations are sometimes not scheduled, Le., the distance-dependence effect is elirriinated. The
fast-fading aIso helps achiew the fair distribution of the allocated rates among mobile users
as will be noted in the nest section.
Let 6 be the average percentage of mobile users scheduled to transmit (out of 100 mobile
users in the network) in a dot when shadow and Rayleigh fading are prcsent in the system,
From Fig. S.6, the valries o f € are approsimatdy 8.7%. 17.8%. 41% and 67% respectiveIy for ,Ci
= 1: 2, 3 ancl 10. That is. for esample? tire can assume thac approsimately 67 mobile iisers will
be transmitting in each dot when each base station issues 10 tags per dot in a 9-ce11 network
with 100 admitted iisers iising our scheme.
L . 1 . - A A .. .
'm. - . - ,
- . 1
S .
. ..- J A - 5
I . - I W . i
8 .: ' s
O 5 1 1 5 2 2.5 3
Table S.?: Average throughput achieved over the simulation period.
Conventional
Proposed
5.5.3 Individual and System Throughput
We compute the achieved throughput (after accounting for the bit error) for conventional ancl
proposed schemes. Figs. 5.9-5-10 show the achieved throughput in each block (averaged over
120 slots). As the number of issued tags increases, the throughput decreases. Because more
mobile users are transmitting and increasing the interferencc and hence the bit error rate.
the throughput is decreasetl. It can also be seen that. as the number of issued tags increascs
from 1 to 10. the differerice in the achieved throughput becorties more apparent betwecri both
schemes. This is due to the fact that when 1Ci is srnall, we do not fuily take adtantage of
the tagging in controlling the transniissions and hence the interference. Table 5.2 shows the
average throughput achieved over the cluration of the simulation for SI = 1. 2. 5 and 10.
Equally important to systern throughput is also to analye the individual throughput.
In Figs. 5.11-5.12, we plot the histograrn of mobile users that achieved different throughput
over the durütion of the simulation. As the number of tags increases, the spread in the
achieved throughpirt witieris in the proposed scheme comparecl to the conventional scheme
though the former yields better throughput performance. For esample. the normalizeil (bu
the rnean) standard cleviation of the intlividual throughput is 0.0; and 0.10 respectir.ely for
the conventiona1 and proposecl schemes. when bI = 10. The d e r the spread the Iess fair the
system towards users in delivering the throughput. Me espect this spread to shrink ~viieri the
simulation is rirn for a longer periotl.
1.18 x 106
1-18 x 106
1 . 5 x O
133 x 10'
3.53 x 10"
3.7; x 10'
1.3'2 x 10"
1.66 x 10"
Figure 5.9: Average (mer 220 dots) achieved throughput. ?iI = 1 and 1.1 = 2.
' 6 -
Figure 3.10: Arerage (over 1'20 rilors) a(-tiieved throughpiit. LI = 5 and II = 10.
. rtd
tu- -
Figure 5.11: Histogram of the number of mobile ilsers chat achieved the average throughput
over the duration of the simulation when 41 = 1 and .LI = 2.
Figiirc 5-12: Histogram of the number of m o f d ~ tisers chat achiereci the average ttiroughpuc
over the ciuration of the sirnidation mhen N = 5 iind L i = 10.
Figure 5.13: The mobile transmit p o w r of active mobile iisers over the duration of 120 sIots
is sumrnetl up and then averaged per dot when SI = I and .LI = 2.
5.5.4 Total Mobile Transmit Power
Throughput alone is not a good performance measiire in comparing two schemes in radio
networks. The mobile transmit power is very importarit in such systems since it determines the
size and life of the batteries (or the ce11 phones). Figs. 5.13-5.14 show the dispensed (averaged
over the mobile users) power in each block, Table Z.3 shows the total mobile transmit power
for the duration of the simulation. in our scheme. transmission rates are adjusted and so are
ttie mobile transmit power. -4s we see ncsc. the rnost important measiire is the ratio of the
mobile transmit power to ttie throughpiit which is the artioiint of energy recliiired CO transmit
one bit of information.
5.5.5 Average Transmit Bit Energy
Tht* miin goal OF the proposecl schemc is to achievtl tiigher throughpitt with lower pow~r by
coiitrolling the inter-ce11 interference via inceliigent packet scheclitling. Since the chta rittes
are ~Iytiiimically acljustetl alninst in t ~ ~ y dot. the appropriate mcBwilre o l performancr is 1
Figure 3.14: The mobile transmit powr of active mobile users over the duration of 120 dots
is surnmed iip and then averagecl per slot when 41 = 5 and 1,I = 10.
Table Z.3: Average mobile transmit poirrer dispensed over the simulation period.
the dispenseci power used to transmit a bit (which is the bit energy). We compute the bit
energy for each mobile user in each slot over the simulation periotl ancl report the average bit
enerv. Table 5.4 shows the average bit energy dissipatecf in the sy t em in order to achieve
the average throughput given in TabIe 5.2 clispensing the average transmit power given in
Table -3.3. Fig. 5-15 shows it graphicaily with different M.
-4s c l eu from the figure. when SI = 1- no ciifference is rintetl. \\?th the increase in the
issuecl tiigs frorn 1 to 10. the proposetl scherne outperforrns the coriveritional sctienie by about
30% in t lie r r;insrnit bit mer,-.. \VP htwefit from ~ h e tagging-tiasi~cI sctit~tlriling when niore tags
iirc distrit)iitd aniong the niobile iis~rs and allowing therri to ~t iakrb ckc.isions. The users c l e l -
their transrtiissions if it helps others achieve better throiigtiput witti less transit power.
Nurnber ot tags issueo. M
Figure 5-15: Average transmit bit energy ctispensed in the system over the simulation periotl.
I i
5.6 Chapter Summary
Table 5.4: Average bit energy dissipated owr the simulation period.
(
0.641 x IO-.' / 4.26 x IO-.'
We proposed a tagging-baed packet scheduling scherne chat cari be thought of as a dytiarnic
distributecl iriter-ce11 interference controI mectianism tvittioiit an? coordination aniong base
stations. The number of tags thiis issued clepcnds on tlitb ttspectcd traffic. the niiriiber of
0.469 x 10-'
0-414 x
Conventional 1 0.347 x 10-6 10-"-599 x 10-'' Proposed 3.02 x IO-'' 0.349 x 10-6
cells and the propagation conditions in the niitwork. Diffcrerit types of tags cztn Lie isstiecl
giving nclditional Hesibility in the interference nirinagenient. in otrr schenic, mobile mers chat
not only havc stronger channel gains to their rcspcctivc home base stations but also cniise
relatively lower inter-ce11 interference are schecliilecl for traiismissions. The proposecl sctieme
is suitable for \'BR (variable bit rate) del- insensitive applications siich as ivireless Internet
data.
SirniiIation results in shadow and rnultipath fading environment showed that our scheme
outperforms the conventional scherne by 30% in transmit bit energy when ench base station
issues 10 tags in a 9-ceIl network with 100 admitted users. The scheme exemplifies a coopera-
tive approach whereby fast-varying temporal variations in the channel conditions are espioited
to scheclule the packets. .As the number of issued tags increues. the difference in performance
between the conventional and the proposed schemes becortics more apparent with the latter
outperforming the former.
Chapter 6
Conclusions
in this clissertation, ive focused our work on reverse links in the nest generation cellular CDhI.4
networks to improve the system performance by means of effective temporal and spatial inter-
ference control ancl management. CVe showed thnt. both the rnerin of the mobile transmit power
and the mriarice among the base station rcceive powers have to be rediiced to improve the sys-
tem performance. Various wireless data applications that are packet-oriented. delay-tolerant
and throughput-demanding have emerged recently with the wireless Internet revolution. These
developments required us to research appropriate system architecture. irnplementation tech-
niques and control rnechanisms in future CDMA networks taking into account the different
service reqiiirements of the applications.
6.1 Thesis Summary
In Chapter 2. me anal-ed the effect on the system capacity ctue to the integration of different
wireless communication senices. This stucij- helped us dirtw otir investigation and devise
techiiiqiies to improve the qsteni performance. CVe first consitlercd the receive power control
prot~lem at the base stations for niulti-media mers in a celliilar CD_\[=\ network ancl showet1
t h ttiv prot)letri can bc split irito two optiniizacion prob1c:riis. Orie is basecl on source char-
acteristics arid the other on thc netiwrk trafic conditioiis: tlic fornier is irnplemented locally
within thc cclls and the latter glohally in the nctn-ork. \\é showcl how the interference amorig
clifierent types of iisers and as well as among different cells van be balancecl in order to ef-
Ficiently accomtnodate different rate/service users. The neetf for interference-balancing was
ernphasizecl in order to minimize the total mobile transmit potver and hence to increase the
svstem capacity-
In Chapter 3. ive proposed a schenie chat combines power. rate and ce11 control to be
implemented as a means of congestion control. There. tratisi~iission rates of those users in
the congestetl (non-congested) cells are decreased (increased) proviciing the required average
throughput among users: hence. the proposed schenie is siiitable for applications that can
tolerate delayeci packet arrivais. Two distributeci iterative algorithtns, one directly minimizirig
the bit energy and the other indirectly iising measured pilot potver. were given. Simulation
results showet1 that Our algorithms converge Fister thari cell-brcathing scheme and minimize
the average transmit bit energy. These are achieveti via iritcrfererice-balancing among base
stations via aclaptively adjusting the data transmission rates.
In Chapter 4. we investigated overlapping sectors as a means to distribute the interference
power aniong sectored antenna and hence to improve the systeni performance. The problem
of base station antenna assignrnent with minimum mobile transmit power was studieci for
CDhi-4 networks with fised overlapping sectored antenna architectiire (FOSAX). It ms first
shown that the non-FOSA.4 has limitations in s~vitching users between in-cell sectors and also
out-of-cell sectors in moderatel-loaded networks. It ivas theri s h o w that by employing over-
lapping sectors in €OSA-4, nre can esploit the Aesibifity nf asigriing a user to one of possibly
niany potential anterina to effectivelu support the rion-uniforcri angtilar traffic. In FOS.4-4.
the process of dynamic cell sectoring is tiifferentiated two-foltl as cell-hreathing (CB) and
cell-slicing (CS) and the latter can be vieiveci as being the arigi~tar counterpart of the forrrier
riulial scheiiit~. The fiybricl schenie. CB-tCS. is shunri to \- i i+l t lit. optirrial solritioti iri rriini-
rriiim total niohilc rraristriit p o w r in a FOS.k-4 s>-stcwi. \Ye &O stiowed that the pilot-powr
baser1 antennu assignnient and power aclaptation algorithni ç;in natrirally iniplemerit hot11 cell-
siicing ancl ceII-brcathing, hencc controlling spatial interfercrice effectiwly. Siniulation resdts
tlernonstratecl the flexibility and effectiveness of the F0S.U in non-iiniforrn angular traffic.
The FOSA-4 outperforms the non-FOS.\=\ in ternis of received signal quality in a hot-spot
scenario. -4s the congestion level increases. the difference in SIR performance between the
cell-breathing and cell-slicing schemes becomes more apparent with the latter outperforrning
the Former. Though CS and CB+CS schemes performed coniparatively well in the sirnrilacion.
the latter hybrid scheme is expected to do better in the presence of non-uniform radial Ioads
and in shadowing radio environment. These performance gains are realized at the expense of
an increased nurnber of antenna, though only a subset will be active at any time depending
on the trafic conditions.
In Chapter 5. wc studied a new tagging-based channel access scheme to schedule packets in
such a way to reduce the inter-ce11 interference. This schenie exploits the temporal variations
of the channel conditions and delays the dominant inter-cell interferers. In this scheme. mobile
iisers that not only have stronger channel gains CO their respective home base stations but also
cause relatively lomer inter-ce11 interference are scheduled. Simulation results in a reaIistic
radio propagation environment showeci that our scheme oucperfornis the conventional schenie
in minimizing the transmit bit energy* The schenie exemplifies a cooperative approach ürnong
niobile users whereby fast-varying teniporal variations in the channel coiiclitions are exploitecl
to schcdule the pnckets. As the nurnber of issiietl tags increases. the ciifference in performance
bctweeri the conventional and the proposecl sclicnic becoiries more apparent wirti the latter
outperformitig the Former. This scheme is aIso applicable for del--tolerant applications siich
as wireless Iriternct data. Different types of tags can be issi~ecl to give additional flesibility in
interference niimagement and congestion cotitrol.
6.2 Thesis Contribution
In this thesis clissertation. WC
1. showed that the receive power controf problem for users with mufti-servicc rcqiiirements
c m be soIvecl by the split-optimization approach and revealed ttiat in order to improve
the susteni performance, the variance of the interference power among the base stations
has to be redimcl ancl the interference-balancing via source and network control can
achieve it,
2. proposed ancl analy~ed a novel combined rate/power/cell (R/P/C) control scheme that
gracefully itnplements congestion control to minirnize the transmit bit cncrgy for delay
tolerant wireiess data applications,
3. investigated the iise of a novel fixed overlapping sectored antenna arctiitectiire as a rneans
to flesibly anci effectively handle the non-iiniform angiilar traffic at the hase stations in
multi-service wireless networks by distribiiting the interference among sectored antenna
and thus improving the system performance especially in a hot-spot environment,
4. proposed and studied a novel tagging-based schediiling scherne that JistributiveIy con-
trois the channel access to mauirnize the system throughput for variable bit rate appli-
cations by esploiting the temporal variations in the channe1 conditions and delaying the
dominant interferers.
5. cleveloped alporithms for home base station assignrnent. rate acIjustnient. packet schedul-
ing and power aclaptation to be used in the proposed schernes-
The materials in Chapters 2. 3: 4: and 5 have been piiblisheci in [30]. [66]. [G1.651 and [691
respectiwi.
Future Research Directions
We clisctiss several possible extensions of our work and also oiitlirie a niiniber of issues that
have to be addressccl in real implementation.
L. In the R/P/C control scheme,
0 [n our worki we performecl a snap-shot analysis: however, che clynamics of the
R/P/C control with time have to be further stuclied. Performance anaiysis in the
prcsence of shadowing can also be clone to understand its cffect on the achieved
individual throughput. These will help alleviate the disadvantage that the users in
congested area may esperience for a short period of time when irnplementing the
proposed scheme.
0 Communication of the mean interference potver? Q: to al1 tlie iisers is criicial in bal-
ancing the interference among base stations and determining the rates and adapting
the potver. This information can be iricluded iri the control chaiinel czssociated with
each reverse link.
O We assurned that the total interference power at the base stations varies slowiy
so that the algorithms can handle the user assignment to base stations. Fucther
stuciy is neederl to uriclerstarid the dynarnics of the niobile-switching and activat-
ing/deactivating an antenna during the process of congestion control (Le-. when
spatial ancl temporal interference v a - mith time).
O Potentially inherent anterina diversity gain in the reverse links can bc esploited to
improve the SIR. The gain that can be achieved iising adjacent sectors in FOS--\,-\
can be irivestigated. The correlatiori betweeri the signai receivctl at ttvc differerit
co-locaterl anterina will determine the amount of irriprovcnient.
a By incorporatilig the TDM cornponent. iw cari control the interfereiic~ flcsibly and
cffectivcly sincc WC have the freedom to choosc thc appropriate sector to opcrnte
with and also tinie slot to transmit. Hoirever. these woilld involve signaling and
processing overheaci in rnanaging both the space and time dornain with inter-ce11 co-
ordination. One can investigate the CDL~iA/FOS=\=\/TDhi with proper sector/slot
assignrncnt niechanisrns.
3. In the tagging-based scheduling scherne,
a CVe need to tletcrmine the number of tags to be issueci ( A I ) in the network where
d l depentls on network load, network size and operating radio environment. In our
approach, ive set it to a fised nurnber for al1 the base stations. However. it cari
also be set dynarnically by the base stations depending on the above-rnentioned
conditions.
a Better algorithnis for determining the transmission rates can be cieveloped. CVe
used a simple rate-setting algorithm that evenly divides the rates among potential
mobile iisers. However, algorithms that dynamically set variable rates to mobile
users can be devised.
a This triggirig-based scherne detects those niobile u s e s that rnay potcntially cause
higher inter-ce11 interference ivithout incirrring any cornplex signalirig for inter-cell
coorclination. The use of this scheduhg schcme in HDR-type [64[ systems with
the focris on forivard links can be investigüted.
Both the R/P/C congestion control and tagging-based interference control sclienits are promis-
ing to be implenientetl for wireless data applications that can tolerate delayed packet amvals.
The use of overlapping sectored antenna kas also been investigated as a rneans to effectively
hancile the non-iiriiforrti iingiilar trnffic n-hich is ver>- prevaient in r d netwrks. Ive espect
that al1 of the ahow rrt-tmicliies can be usetf tngettier ta irriprow the systeni perfurniance.
Bibliography
[l] E. Lee and D. 4lesserschmit t, Digital contmvrrication. KIuwer -4caciernic Pu blishers. 2nd
Editioti, 1998.
[2] V. Garg and .J. Wilkes: Wireless and persona1 co~nmunications systems. PrenticeHall,
1996.
[3] T. S. Rappaort. LV~reless communications: principles and practice. Preritice-Hall. 1996.
[;ll Ai .J. Viterbi. CDMA: principles of spreud qectnrm communl'cutzons. Addison-CVesle_v,
1995.
[51 L. E. Slillcr riricl .J. S. Lee. CDMA systems engineering handbook Artech House. 1998.
[6] T. Ojanpera and R. Prasad, Wideband CDkfA for third generation mobile communica-
tions. Artech House, 1998.
[Tl C. Mhailescu and et al-: "Performance etaluacion of a dynarnic resowce allocation algo-
rithrn for CSlTS-TDD systerns." VTC-Spriny: 2000.
181 E. Sousa. "Lecture notes for ECE 1543s.- ECE Department. U71.rrer.sit~j of Toronto, 1998.
[91 J . Liherti ancl T. Rappaport, &AnaIytical resuIts For capacity iniprovernents in CDM-4."
IEEE Trans. Veh. Technol.? vol. 43. pp. 650-690- August 1994
[IO] 1.- Li anci et xi.. .'PerFortnance etaluation of a ceIlirlar base station rniiitiheatri antenna.-
IEEE Truns. bëh. Techr~ol.. ro-ol. -16. pp- 1-9. F e b r i i a ~ 1997.
[ I l ] R, Sclinialenberger and J . Blanz. .-[riHuencc of the distribution of rriotile iism on a C/I
balancecl rnulti-beani aritenna digital cellular radio system." PIMRC. pp. 898-902. 1997.
[12] Y. Hara and et al.. "Or1 the capiicity of ceIlular systems nith niiiltibearii iiritenrias."
PlhlRC. 1998.
[13] -4. F. Saguib and et al.. "Capacity improvement with base-statiori antenria arrays in
cellular CD&[-1." IEEE Tmns. Veh. Technol., vol. 43. pp. 691-695. Aiigiist 1994.
[14] C.-L. I and R. Gitlin, "MuIti-code CDS[-4 wireless personal commiinication rietworks."
lCC, pp. 1060-1064, 1995.
[IS] D. Ayyagari and -1. Ephremides. -'Cellular multicocle CDMA capacity for integratetl
(voice anci data) services," IEEE .J. Select. Areas Commun.. vol. 17. pp. 928-938. 1Iay
1999.
(161 H. Schottcn ancl et al., ".4ctaptive mrilti-rate multi-code CD>[-4 systerns." VTC. pp. 782-
783, 1998.
Cl71 M. hIadkour and S. Gupta, "SIulti-rate muIti-code using fwt for mobile antl personal
corn ni uni cation^,^ PIMRC. 1998.
[18] C.-L. 1 and K. Sabnani. --Variable spreading gain CDMA with adaptive control for true
packet switching wireless networks." [CC. pp. 725-730, 1995.
[Tg] C.-L. 1 antl K. Sabnani. .-\-ariable spreading gain CD4i.A with aclaptive control for inte-
gratetl traffic in wireless rietwork." VTC. 1995.
1201 K. Gilhousen and et al., .'On the capacity of a cellular CD1L.A sifstenit" IEEE Tmns.
Veti. Technol-. vol. 40. pp. 303- 312. 11- 1991.
[-II C.-C. Lee and R. Sceele. ,*Effeçr of soft and sofcer Iiandoffs or1 CD1I.A systerri c:api\city.-
IEEE k n s . Veh. Techriol.. vol- -17. pp- S3&541. Augiist 1995.
133
[-21 34. Jansen and R. P m a d , .-Capacity. ttiroiigfipiic ancl clclay iinalysis of a celliilar DS-
CDM-4 systeni with irnperfect poner coritrol and iniperfect scctorizatiori." IEEE Trans.
Veh. Technol.. vol. 44, pp. 61-74: Fehriiary 1995.
1231 A. Viterbi and A. Viterbi. "Erlang capacity of a power controlled CDlL.4 system." IEEE
J. Select. Areas Commun.. vol. 11. pp. SW900. -4ugust 1393.
[24] R. Prasad and et al., zC~p~ ic i t y analysis of a ccllular direct seqiience cotlc division n id -
tiple acccss systern with irnperfect powr control." IEICE Trans. Commun.. vol. EÏ6-BI
pp, 894-905, August 1993.
[-a] E. Kudoh. "On the capacity of DS/CDSI.4 cellular mobile radios untler imperfet trans-
mitter power control." IEICE Trans. Co~nrnun.. vol. EÏ6-Bt pp. 386-593. August 1993.
[26] .LI. Yagatsiika and et al.. "Data trafic control and capacity evaluations for voice/rlata
integrated trarismission in DS-CD.\[=-\." IEICE Trans. Commun.. vol. ESl-B, pp. 1353-
1363, July 1998.
[27] E. Cianca and et al., ;'.An approach to masimize the capacity of a multimedia CDSI.4
wireIess system," VT. pp. 909-913. 1998.
[28] R. Xettleton and H. Alüvi, "Power control for a spread spectrum ceIIular mobile radio
system," VTC. pp. '242-246. 1983.
[29] J. Zoii and C'- Bhargam. "Optirnizetl power allocation for misecl rate trafic in a DS-
CDS[-4 cellular systeni," lEEE Electromc Letters, pp. 1902-1903. October 1993.
[30] -4. S. Anpalagan and E. S. Sousa. "On the receive power allocatiorl in a celIitIar multimectia
CDht-4 systern with fisecl base station iwignenient." In Proc. IEEE VTC-$rang.. 2000.
[32] J . Su and t.c al.. "Optitnimion of powr allocation in a multicel1 DS/CDSiA system ivith
heterogencous traffic." [CC. 1999.
[32] D. Kanade and S. Gao. ..Optirna1 powr idlocation for the reverse litik iti a multimedia
D S - C D U systein." ICC. 1999.
[33] L. C. Godara. "Applciations of antennzi arrays to mobile communications: Part i: per-
formance, improvernent, feasibility and system cons ide ration^.^ Proceerlings of the IEEE.
vol. 85, pp. 1031-1060. Jiily 1997.
[34] T. Matsumoto and et al., .'Beani-selection performance analysis of a switctied mdtibeani
antenna system in mobile communicaitons e~ivironments~" IEEE Trr1n.s. Veh. Technol..
vol. 46, pp. 10-19, February 1997.
[35] -4. Sampath and d. Holztman. "-4ccess control of data in intergrated voiceldata CDSf.4
systems: benefits and tradeoffs?" IEEE ./. Select. .-Ireas Commun.. vol. 1.5, pp. 151 1-1526.
October 1997.
[36] A. Brand and H. Aghvami, "Perforrnitnce of a joint CDbIA/PR4IA protocol for mixed
voice/data transmission for third generation mobile communication.'^ IEEE J. Select.
Areas Commun., vo1. 14, pp. 1695-1707. December 1996.
[37] K. Das and S. Morgera, "Interference and SIR in integrated voiceldata wireless DS-
CDM-4 networks - a simulation study:" IEEE ./. Select. Areas Commun.. vol. 15, pp. 1527-
1538. October 1997.
(381 S. Oh and K. Wiseman. "Integration of voice and data traffic in CDhIA networks iising
dynamic spreaciing gain control." [CC. 1998.
[39] A. Dabak and et al.. "Intcgrated voicc and video network iising variable rate CD.\[=\."
[CC. 1996.
[401 J . Kim ancl et al.: "A p o w r control striiccure for niultimedia traffic." VTC. 1999.
[411 R. Xettleton. "Trafiic theory and interfercrice miinagement for a spreacl spectrum celIiiIar
mobile radio system..' [CC. pp. 24.5.1 -5. 1980.
[42] F. R. Gantmacher. The theoq oJ n1ntrice.s. \oI. -1. Chelsea Press, Sew lkrk. 199(1.
[43] J . Zander: "Performance of optinium transnsittcr power control iri cellular radio systenis.'
IEEE Trans. Veh. Technol., vol. 41. pp. 5142. February 1992.
(441 S. -4. Grandhi anci et al.: "CentraIized potver control in cellular radio systerns." IEEE
Trans. Veh. Technol.. vol. 42, pp. 466-468: Sovcrnber 1993.
1451 J. Zou and et al.: "Efficient methods for high-rate data transmission in mobile and per-
sonal communications," Mernoria Tecnicn ME-YICON, 1994.
[46] S. Sato and et al., "-4 performance analysis of non-uniform trafic in microcell sustcrns."
[CC, pp. 1960-1964. 1993.
[Ji] S. Hanlrv, "An algorithm for cornbinecl ceIl-site selection and pomer control to triasiitiize
cellular spread spectrum capacicy," lEEE J . Select. Areas Commun.: vol. 13. pp. 1333-
1340, September 1995.
[48] S. Rarnakrishnan and J . Holtman, "A scheme for throughput mairimization in a dual-class
CDMA system," IEEE J. Select. rlreas Commun., vol. 16. pp. 830-844, August 1998.
[da] J. Zander, "Distributed cochannel interference control in cellular radio systems." lEEE
Trans. Veh. Technol.. vol. 41. pp. 305-311. Aiigust 1992.
[JO] R. Yates anci C.-Y. Huang, "tntegrateci poirier control ancl base station management.-
IEEE Trans. Veh. Techn01.~ vol. 14, pp. 635-64-4. August 199.5.
[al] S. A. Craridhi aiid et al.. .'Distributecl potver cotitrol in cellular radio systenis." IEEE
Trnns. Commun.. vol. 42, pp. 226-243. Feh./.\hrch/Apd 1994.
[a31 1. htzela and 11. Saghstiirieli. "Cliiiritiel wigtitrierit schenies for c r l l ih r riiohile telecom-
munication systems: a cornprelicnsivc siirvey." IEEE Personal Cornrrricn. Mug.. pp. 10-31.
Jiine 1996.
[ S 4 T. kum and W. Wong, "Hot-spot traffic relief in cellular systerns," lEEE J. Select. Areas
Commun.. vol- 11. pp. 934-939, Aiigust 1993.
[35] A. S. -4npalagan ancl E. S. Sousa, "Channel borrowing schemes in sectored cellular systems
mith worst case SIR analysis," In Proc, IEEE ICUP. 1998.
[S6] Y. Argyropoulos and et al.: -Dynamic ctiannel allocation in interference-limitecl celhdar
systems with uneven traffic distribiition." IEEE Trans. Veh. Technol.. vol. -18. pp. 224-
232' January 1999.
(571 W. Park and et al., .'Performance nnrilysis of traffic load shedding for mobile communi-
cation systems," ICUPC, pp. 306-3 10. 1997.
[a81 K. Takeo and et al.. "-4 base station selection technique for up/downIink in CDS[-4
systems," VTC, 1999.
[59j .J. Qiu and J . Mark, .'A clynamic load stiaririg algorithm through power control in celhlar
CDBI-A," PIMRC. 1998.
[60i 31. ivIahmoucli and E. Sousa. .*Aclaptivc sector size control in a CD.LI.4 system using
Butler matrix," VTC, 1999.
[Gl] S. Sato aricf Y. Amezawa, "-4 stiidu of tlunartiic zone control for CDSI.4 mobile radio
communications," ICUPC. pp. 306- 3 10. L99T.
[621 J . \,\ïriters. "Srnart antenna for wireless systenis.'- [EEE Personal Commun. Mq.. pp. 23-
35. Februan 1998.
[631 P. Laricaster, The theoq O/ rnatricr*.~. -4cacler1tic press, ?ien- l\ork. 1969-
1 :39
[64[ P. Bencicr and et al.. TDSL4/HDR: A hanclwidt ti rfiïrictit tiigti-speetl wirelws cI;ita
service for nomczdic iisers." IEEE Corrrrn irn. !b¶w~. . pp. 70-77. .l uly -1000.
[66] A. S. Aripnlagan and E. S. Sousa. ---4 cornhinctl rate/powr/cell coritrol scherne for clelay
insensitive applications in CD!Li s~stcnis." In PTOC. IEEE Glohecom. 2000.
[GT] A. S. Atipakagan and E. S. S011sa. . i ; lclapti~~ ce11 cectoring rising fiset1 overlapping sectors
in CDhI.4 net\vorks." h l Proc. IEEE [CC, '2001.
[65] A. S. Anpalagan and E. S. Sousa. --Pcrformatice analysis o l a cellular CDM-4 network
with filceci overlapping sectors in non-tiriifortn azirrithal traffic-" In Proc. IEEE [CC. 2001.
[69] -4. S. Anprilagan and E. S. Soirsa. "-4 tagging-baed mediuni access schenie for wireIess
data applications in LDU.\/TDhI rietworks." In Proc IEEE PIMRC. 2001.