close

Вход

Забыли?

вход по аккаунту

?

A direct microwave M-QAM adaptive transmitter for fixed wireless ATM networks

код для вставкиСкачать
INFORMATION TO USERS
This manuscript h a s been reproduced from the microfilm master. UMI films the
text directly from the original or copy submitted.
Thus, som e thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality o f th is reproduction is dependent upon the quality o f the copy
submitted.
Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleedthrough, substandard margins, and improper alignment
can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete manuscript and
there are missing pages, these will be noted.
Also, if unauthorized copyright
materia! had to b e removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning
the original, beginning at the upper left-hand comer and continuing from left to
right in equal sections with small overlaps. Each original is also photographed in
one exposure and is included in reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6” x 9” black and white photographic
prints are available for any photographs or illustrations appearing in this copy for
an additional charge. Contact UMI directly to order.
Bell & Howell Information and Learning
300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA
800-521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
NOTE TO USERS
This reproduction is the b est copy available
UM I
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A DIRECT MICROWAVE M-QAM
ADAPTIVE TRANSMITTER FOR FIXED
WIRELESS ATM NETWORKS
A Thesis
Submitted to the College of Graduate Studies and Research
In Partial Fulfilment of the Requirements
for the Degree of
Doctor of Philosophy
in the Department of Electrical Engineering
The University of Saskatchewan
Saskatoon, Saskatchewan
By
A bbas M o h a m m a d i
Fah 1998
@Copyright Abbas Mohammadi, 1998. All rights reserved.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1*1
National Library
of Canada
Bibliotheque nationale
du Canada
Acquisitions and
Bibliographic S erv ices
Acquisitions et
services bibliographiques
395 Wellington Street
Ottawa ON K1A0N4
Canada
395, rue Wellington
Ottawa ON K1A0N4
Canada
Your Me Votro reference
OurfUe Notre reference
The author has granted a non­
exclusive licence allowing the
National Library o f Canada to
reproduce, loan, distribute or sell
copies of this thesis in microform,
paper or electronic formats.
L’auteur a accorde une licence non
exclusive permettant a la
Bibliotheque nationale du Canada de
reproduire, preter, distribuer ou
vendre des copies de cette these sous
la forme de microfiche/film, de
reproduction sur papier ou sur format
eiectronique.
The author retains ownership of the
copyright in this thesis. Neither the
thesis nor substantial extracts from it
may be printed or otherwise
reproduced without the author’s
permission.
L’auteur conserve la propriete du
droit d’auteur qui protege cette these.
Ni la these ni des extraits substantiels
de celle-ci ne doivent etre imprimes
ou autrement reproduits sans son
autorisation.
0-612-37899-3
Canada
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UNIVERSITY OF SASKATCHEWAN
College o f Graduate Studies and Research
SUMMARY OF DISSERTATION
Submitted in partial fulfillment
o f the requirements for the
DEGREE OF DOCTOR OF PHILOSOPHY
by
Abbas Mohammadi
Department o f Electrical Engineering
University o f Saskatchewan
Spring 1999
Exam ining Committee:
Dr. A. Hi rose
Dean's Designate, Chair
College o f Graduate Studies and Research
Professor T.S. Sidhu
Chair of Advisory Committee, Department of
Electrical Engineering
Professor S. Kumar
Supervisor, Department o f Electrical Engineering
Professor H.C. Wood
Department o f Electrical Engineering
Professor D.E. Dodds
Department of Electrical Engineering
Professor C. Williamson
Department of Computer Science
External Examiner:
Professor D.R. Conn, Chair
Department o f Electrical and Computer Engineering
McMaster University
1280 Main Street West
Hamilton, Ontario, Canada
L8S 4L7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
PER M ISSIO N TO USE
In presenting this thesis in partial fulfilment of the requirements for a Doctor
of Philosophy degree from the University of Saskatchewan, the author agrees
th at the libraries of this University may make it freely available for inspec­
tion. The author further agrees th at permission for copying of this thesis in
any manner, in whole or in part for scholarly purpose may be granted by
the professor who supervised this thesis work or, in his absence, by the Head
of the Department or the Dean of the College in which this thesis work was
done. It is understood th at any copying or publication or use of this thesis
or parts thereof for financial gain shall not be allowed without the a u th o rs
w ritten permission. It is also understood th a t due recognition shall be given
to the author and to the University of Saskatchewan in any scholarly use
which may be made of any material in this thesis.
Requests for permission to copy or to make other use of material in this
thesis in whole or part should be addressed to:
Head of the Departm ent of Electrical Engineering,
University of Saskatchewan,
Saskatoon, Saskatchewan,
Canada, S7N 5A9.
i
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A B ST R A C T
Wireless ATM plays a key role in the realization of broadband wireless net­
works. The transmission of various classes of traffic and the provision of
bandwidth on demand over a wireless channel poses a number of new tech­
nical challenges. This thesis addresses the design of a low cost adaptive
transm itter for fixed wireless ATM/B-ISDN systems with emphasis on opti­
mum use of wireless network resources.
A new architecture for a direct microwave wireless ATM transm itter is
proposed. The transm itter capacity adaptation is implemented by using an
admission control metric and an M-QAM m odulator. The two main compo­
nents of the transm itter are: an M-QAM control unit and a direct microwave
QAM m odulator unit. The M-QAM control unit is used to select an opti­
mum modulation level for the QAM modulator. The modulation level is
adjusted based on the bandwidth demand, QoS requirements, and outage
conditions of the wireless ATM link. The direct microwave QAM m odulator
unit transforms the broadband traffic to a modulated microwave signal that
is suitable for transmission over a wireless network.
The required bandwidth of the broadband traffic is estimated using an
effective bandwidth metric. An analytical relation, called the capacity reduc­
tion factor, is derived to represent the performance degradation due to the
wireless channel and channel fading in a B-ISDN network. Using the effective
bandwidth m etric and the capacity reduction factor, a QoS metric for the
wireless broadband network is introduced. This metric is termed as, modi­
fied effective bandwidth. This metric is used to adapt the M-QAM m odulator.
ii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A nother significant contribution of this research work is a new architec­
ture for the direct QAM m odulator.
This is based on use of PIN diode
reflection attenuators. The PIN diodes operate in forward bias condition
thereby overcoming the speed lim itation problem due to charge storage. Us­
ing residue theory, analytical results to model the large signal forward bias
operation of PIN diodes are presented. This theory also examines the tran ­
sition tim e of a PIN diode with bias changes from a reverse bias to a forward
bias.
The direct microwave QAM m odulator implementation using MIC and
silicon MMIC technologies is examined. While a realization using MIC is
simple and straightforward, a silicon MMIC realization offers a very cost ef­
fective solution.
A system study was conducted to exam ine the operation of the adaptive
direct microwave M-QAM m odulator in the wireless channel with ATM traf­
fic. The operation has been examined for different wireless channels and for
various classes of traffic. The call acceptance and outage performance are
com pared w ith those for a fixed QAM m odulator. The results show th a t the
proposed system can be used for im plem entation of cost effective adaptive
transm itters for broadband wireless applications.
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A C K N O W LED GEM ENTS
I would like to express my sincere gratitude to Professor Surinder K um ar,
my research supervisor, whose vision and keen interest are the inspiration of
this research work. I am grateful to him for the opportunity to benefit from
his knowledge, encouragement, and advice. I am grateful to all my research
advisory committee members; Professor Hugh Wood, Head of the D epart­
ment of Electrical Engineering, for his invaluable help during the course of
this research and providing me a number of opportunities in the departm ent,
Professor David Dodds for his continuous encouragement and support, and
Professor Carey W illiamson who always helped me in this thesis.
I thank the director, Mr. Cliff Klein, and staff of Telecommunications
Research Labs (TRLabs) who provided me an excellent milieu to work on
this thesis. I thank the professors, staff, and students of the departm ent
of the Electrical Engineering for making this time enjoyable. I also th an k
my friend, Dr. David Klymyshyn for his support during the course of this
research.
I am grateful to Telecommunications Research Labs for providing finan­
cial support as a Research Scholarship. The scholarship provided by Iran
ministry of culture and higher education is gratefully acknowledged. T he
Natural Sciences and Engineering Research Council(NSERC) is also grate­
fully acknowledged for its financial assistance.
A special thanks w ith appreciation is extend to my family for their love
and support.
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
D E D IC A T IO N
To the Loving M em ory
of m y father, Mr. Saeed Mohammadi
fo r his unlimited love.
V
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF C O N TEN TS
P E R M IS S IO N T O U S E
i
ABSTRACT
ii
ACKNOW LEDGEM ENTS
iv
D E D IC A T IO N
v
T A BL E OF C O N T E N T S
vi
LIST O F T A B L E S
xii
LIST OF F IG U R E S
xiii
LIST OF A B B R E V IA T IO N S
xxiii
1 IN T R O D U C T IO N
1.1
2
1
Broadband Wireless C om m unications.........................................
2
1.1.1
ATM N e tw o rk s ....................................................................
2
1.1.2
Radio A T M ...........................................................................
3
1.1.3
B andw idth Estim ation in Radio ATM Network
4
1.1.4
A daptive M o d u latio n ...........................................................
5
1.1.5
M odulator Im plem entation.................................................
6
1.2
Literature R e v ie w ............................................................................
6
1.3
Research O bjectives............................................................................. 11
1.4
Thesis O rg an isatio n .........................................................................
....
11
F U N D A M E N T A L C O N C E P T S F O R W IR E L E SS A T M T R A N S ­
M IT T E R
14
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.1
..............................................
14
2.1.1
ATM Basic C o n c e p ts...........................................................
14
2.1.2
ATM S e rv ic e s........................................................................
16
2.1.3
B andw idth Demand and QoS R e q u ire m e n ts .....................17
2.2
ATM M u ltip le x e r...............................................................................
17
2.3
Bandwidth E s tim a tio n .....................................................................
20
2.4
End-to-End QoS Guarantees in Radio ATM N etw orks................ 24
2.4.1
Cell Loss Ratio Due to Congestion
2.4.2
Cell Loss Ratio in a Radio Channel
.................................
25
..................................25
2.5
Adaptive Radio ATM T ran sm itte r..................................................... 28
2.6
M odulation Techniques for Fixed Radio A T M ................................28
2.7
3
Fundam entals of ATM Networking
2.6.1
Q uadrature Amplitude M o d u la tio n .................................
2.6.2
Spectral Efficiency of QAM
2.6.3
Power Efficiency
29
.................................................. 30
......................................................................32
QAM M odulator Im p le m e n ta tio n .....................................................33
2.7.1
Classic Im plem entation............................................................ 33
2.7.2
D irect Microwave Im p lem en tatio n ........................................ 34
2.8
Radio Microwave Link D e s ig n ........................................................... 34
2.9
Summary
...........................................................................................
36
B A N D W ID T H E S T IM A T IO N O F B R O A D B A N D T R A F ­
38
F IC
3.1
Characterisation of Broadband T raffic.............................................. 39
vii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.1.1
3.1.2
3.2
Video Traffic
........................................................................
39
Ethernet T r a f f i c ......................................................................40
E stim ation of the Required Bandwidth of B roadband Traffic . 40
3.2.1
Simulation M o d e l......................................................................41
3.2.2 Bandwidth E stim ation for Empirical Video Traffic . . .
42
3.2.3 Bandwidth E stim ation for Ethernet T ra ffic ......................... 42
3.3
Bandwidth Estim ation Using Statistical M o d e l s ........................... 44
3.4
Effective Bandwidth using the Fractional Brownian Motion
A p p r o a c h ................................................................................................ 46
4
3.4.1
Comparison of Results for VBR Video T r a f f i c ...................48
3.4.2
Comparison of Results for Ethernet T r a ffic ......................... 50
3.4.3
Experim ental Estim ation of the Effective Bandwidth . 54
3.5
Modified Effective B andw idth for Radio A T M .............................. 57
3.6
S u m m a r y ...........................................................................................
59
A D A P T IV E M -Q A M F O R R A D IO A T M IN M U L T IP A T H
F A D IN G C H A N N E L S
4.1
4.2
60
Requirements for an A daptive T ra n s m itte r.................................... 61
4.1.1
Capacity Variation Due to Q o S .............................................61
4.1.2
Capacity Reduction Due to ChannelF a d i n g ....................... 63
Adaptive M-QAM M o d u l a t o r ...........................................................65
4.2.1
M-QAM in AWGN C h a n n e l................................................... 66
4.2.2
M-QAM in M ultipath Fading C h a n n e l ................................66
viii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2.3
4.3
4.4
5
M-QAM for ATM Cell Transmission in Multipath Fading 70
Adaptive M-QAM M odulator for Radio ATM Applications . . 73
4.3.1
Modified Effective Bandwidth for M -Q A M ........................ 73
4.3.2
Adaptive M-QAM A rc h ite c tu re ........................................... 74
S u m m a r y ...............................................................................................78
D IR E C T M IC R O W A V E M -Q A M M O D U L A T O R H A R D ­
79
W ARE
6
5.1
Modulator A rchitecture........................................................................79
5.2
Theoretical Model for Direct QAM M o d u la to r..............................81
5.2.1
A n a ly s is ..................................................................................... 81
5.2.2
Numerical S im u latio n .............................................................. 94
5.2.3
Direct Microwave QAM M o d u la to r.................................... 96
5.3
Distortion in the PIN Diode M o d u lato r........................................ 108
5.4
S u m m a r y ............................................................................................ I l l
H A R D W A R E IM P L E M E N T A T IO N OF D IR E C T M IC R O W A V E
112
QAM M O D ULATO R
6.1
Direct Microwave M-QAM Implementation using MIC Tech­
nique
112
6.1.1
Subsystem A rchitecture
..................................................... 112
6.1.2
Q uadrature Coupler
6.1.3
Ring C o u p l e r ......................................................................... 116
6.1.4
SPDT S w it c h ......................................................................... 118
............................................................113
ix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.2
6.1.5
Variable A ttenuator with Phase Shift Compensation . . 120
6.1.6
Power Combiner
6.1.7
QAM M o d u la to r ..................................................................... 126
.....................................................................126
Direct Microwave M-QAM Implementation Using Si-MMIC
T e c h n iq u e .............................................................................................. 128
6.3
7
6.2.1
PIN Diode in Si-MMIC T echnology.................................... 130
6.2.2
Subsystem Design on Si-MMIC Technology....................... 130
Summary
............................................................................................. 138
P E R F O R M A N C E OF F IX E D W IR E L E SS A T M T R A N S ­
M IT T E R
7.1
141
Adaptive Direct Microwave M-QAM Transm itter for Radio
A T M ........................................................................................................141
7.1.1
7.2
7.3
Advantages of using Effective Bandwidth Metric . . . .
141
Adaptive M-QAM P e rf o rm a n c e ..................................................... 144
7.2.1
LAN T ra ffic .............................................................................. 146
7.2.2
VBR M PEG Video T r a f f ic ....................................................149
7.2.3
Multiple Sources
7.2.4
Discussion of Results of the E x p erim e n ts.......................... 160
.................................................................... 156
Adaptive Direct Microwave M-QAM m o d u la to r ........................ 161
7.3.1
Adaptive Direct Microwave M-QAM Implementation
using MIC te c h n iq u e ............................................................. 161
7.3.2
Direct Microwave M-QAM Modulator Implementation
using Si-MMIC T e c h n iq u e ................................................... 162
x
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7.4
8
Sum m ary
............................................................................................. 166
C O N C L U S IO N S A N D F U T U R E R E S E A R C H
169
8.1
C o n c lu s io n s ..........................................................................................169
8.2
Contributions of this Research Work
8.3
Future W o rk ..........................................................................................175
..............................................174
REFERENCES
176
A P P E N D IC E S
176
A C A P A C IT Y R E D U C T IO N F A C T O R IN M -Q A M IN R A Y L E IG H
CHANNEL
191
B P H A S E S H IF T C O M P E N S A T IO N IN A R E F L E C T IO N A T ­
TENUATOR
193
C B IA S C IR C U IT F O R Si-M M IC D IR E C T M ICR O W A VE Q A M
M ODULATOR
C.l
195
Lowpass F ilte r ......................................................................................195
C.1.1
LowpassFilter Implementation on Si-MMIC Substrate 195
xi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
L ist o f Tables
2.1 QoS Requirements
............................................................................
18
2.2 Terrestrial Link D e s ig n ......................................................................... 37
3.1 The Video traffic s o u rc e s ......................................................................48
7.1 Average and standard deviation for modulation level, out­
age, effective bandw idth, and modified effective bandwidth for
Ricean C h a n n e l................................................................................... 148
7.2 Average and standard deviation for modulation level, out­
age, effective bandw idth, and modified effective bandwidth for
Rayleigh C h a n n e l................................................................................149
7.3 Traffic characteristics of various video
t r a f f i c ............................ 152
7.4 Average and standard deviation for modulation level, out­
age, effective bandw idth, and modified effective bandwidth for
Ricean Channel for video tr a f f ic ...................................................... 155
7.5 Average and standard deviation for modulation level, out­
age, effective bandw idth, and modified effective bandwidth for
Rayleigh Channel for video tr a f f ic ...................................................155
7.6
Statistical Param eters for multiple MPEG-1 Video Traffic in
a Ricean C h a n n e l................................................................................ 160
xii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
L ist o f Figures
1.1
The typical model for radio ATM com m unications...................
3
1.2
The reference model for fixed radio ATM communication services
4
2.1
ATM cell format for network node interface (N N I)...................
15
2.2
Connections for ATM n e tw o rk in g ................................................
16
2.3
An ATM m u ltip le x e r.......................................................................
19
2.4
A Finite Buffer with VP as an o u tp u t..........................................
21
2.5
An ATM multiplexer with VP as an output
22
2.6
a) A Nonblocking Switch b)Equivalent circuit for capacity es­
.............................
tim ation
23
2.7
The reference model for fixed point to point radio ATM link . 24
2.8
a) CLR versus BER for random and bursty channelsb)A
comparison between cell loss estimation results for the bursty
channel with the previously published results [25]........................... 27
2.9
An adaptive m odulator in a time varying c h a n n e l....................... 29
2.10 Constellations for QAM s i g n a l s ..................................................
29
2.11 Raised cosine pulse power s p e c tru m ................................................ 32
2.12 BER versus CNR for a QAM m o d u la to r .......................................33
2.13 Block diagram of a heterodyne QAM m o d u la to r..........................34
2.14 Block diagram of a Direct QAM m odulator....................................35
xiii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.1
a) F irst sixteen seconds of Jurassic Park videotrace b) Total
trace(27 m in u te s ) ...................................................................................39
3.2
The E thernet traffic t r a c e .................................................................. 41
3.3
Equivalent circuit for capacity e stim a tio n ....................................... 41
3.4
Sim ulated results for the required bandwidth of a VBR video­
trace Jurassic Park movie with buffer size as a param eter . . .
3.5
43
Simulated results for the required bandwidth of a VBR video­
trace Silence o f the Lambs movie with buffer size as a param eter 43
3.6
Simulated results for the required bandwidth of a VBR video­
trace Star Wars movie with buffer size as a param eter
3.7
. . . .
44
Sim ulated results for the required bandw idth of aggregate VBR
videotrace movie with buffer size as a param eter (phase shift
160 m sec for each source)
3.8
...................................................................45
Required bandwidth of Bellcore E thernet traffic with buffer
size as a p a r a m e te r ................................................................................45
3.9
Effective bandwidth of E thernet traffic source with U =.7 and
C L R = 10"6 .............................................................................................49
3.10 Effective bandw idth of E thernet traffic source with buffer size
20 KBytes and C L R = 10- 6 ............................................................... 49
3.11 Effective bandw idth of a VBR videotrace Jurassic Park movie
using Eq.(3.4) (Buffer size as a p a r a m e te r ) .................................
51
3.12 Effective bandw idth of a VBR videotrace Silence of the Lambs
using Eq.(3.4) (Buffer size as a p a r a m e te r ) .................................
51
3.13 Effective bandwidth of a V B R videotrace Star Wars using
Eq.(3.4) (Buffer size as a param eter)
...............................................52
xiv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.14 A comparison between the effective bandwidth and the peak
rate for different v id e o tra c e s............................................................... 52
3.15 Effective bandwidth of the aggregate VBR video traffic using
Eq.(3.4) (Buffer size as a param eter)...................................................53
3.16 Effective bandwidth of the Bellcore Ethernet d ata traces using
E q .( 3 .4 ) ...............................................................................................
53
3.17 a) Simulation model of an E thernet source in ATM sim ulator
b) Simulation model for multiple Ethernet sources...........................54
3.18 The experimental results for the effective bandwidth of a selfsimilar source with 80 percent utilisation with three H values.
55
3.19 The experimental results for the effective bandwidth of a self­
similar source with different utilisation for H = .8 ..........................
56
3.20 The effective bandwidth of 10 self-similar Ethernet source with
U = .8 ............................................................................................................56
3.21 Tolerable CLR due to buffer overflow as function of BER for
Q oS = 10-5 as a function of burst s i z e .......................................
58
3.22 Effective bandwidth for random error and bursty channel for
utilisation 50 percent and H = .7 .......................................................
4.1
58
Effective Bandwidth as a function of p c l r and D max for Rmean —
2 Mbps, a = 104, and H = .7................................................................ 63
4.2
Shannon capacity reduction factor due to channel fading for
Ricean (K = 5) and Rayleigh c h a n n e ls ............................................65
4.3
T he constellation size based on the number of identical video­
traces
.......................................................................................................67
xv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.4
The bit error rate of M-QAM m odulator(Exact and upper
b o u n d ) ......................................................................................................68
4.5
C apacity decrease due to fading in MQAM (K=5 in Rice chan­
nel)
4.6
......................................................................................................... 69
The constellation size plotted as a function of the number of
connections of Jurassic P a rk m ovie for a C L R = 10-6 and
Rayleigh fading c h a n n e l ......................................................................70
4.7
The constellation size vs. number of connections of Jurassic
P a rk m ovie for CNR=30 dB with a C L R = 10-6 ratio for a
Rayleigh fading c h a n n e l ..................................................................... 71
4.8
C apacity reduction factor in a random error channel due to
fading (K =5 for Ricean C h a n n e l) ..................................................... 72
4.9
C apacity reduction factor in a burst error channel due to fading 72
4.10 The modified effective bandwidth for different channels . . . .
74
4.11 A model for a fixed wireless ATM s y s te m .......................................75
4.12 MQAM m odulator for radio ATM using oneV P ............................76
4.13 The M-QAM control unit
.................................................................77
5.1
Direct Microwave QAM
....................................................................80
5.2
The P IN diode s tr u c tu r e ....................................................................85
5.3
The integration c o n t o u r .................................................................... 88
5.4
A multi-level d ata sequence
..............................................................90
5.5
The P IN diode forward bias resistance with I = .l m A ................... 95
5.6
The P IN diode forward bias resistance with 1=1mA
.................. 96
xvi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.7
Comparison between the model and M/A-Com measurement
for MA4P4000 at 100 MHz and 100 mA RF c u r r e n t..................... 97
5.8
Comparison between the model and measurement for SMP
1304 from Alpha at 100 m A ............................................................... 97
5.9
Basic structure used in the study of the PIN diode m odulator
98
5.10 Driver c u r r e n t ......................................................................................... 98
5.11 PIN diode forward biased re sista n c e .................................................. 99
5.12 PIN diode voltage in the basic s t r u c t u r e ........................................ 99
5.13 An architecture for direct microwave Q A M ................................... 100
5.14 a) A reflection multilevel PIN diode PAM modulator b)M odulator
response for the drive currents of .1,.3,.5 and 1 m A.......................102
5.15 Reflection M odulator without prefiltering and predistortion a)
Input current b)PIN diode resistance c) Modulator output
. . 103
5.16 Reflection Modulator using prefiltering(o: = .5) and w ithout
predistortion a) Input current b)PIN diode resistance c) Mod­
ulator o u t p u t ....................................................................................... 104
5.17 A model to study the direct QAM m o d u l a t o r .............................105
5.18 a)Bias in I path b)Bias in Q p a th c)PIN diode forward bias
resistance in I path d) PIN diode forward bias resistance in Q
p a t h ........................................................................................................106
5.19 a)PIN diode voltage in I path b)PIN diode voltage in Q path
c)PIN m odulator output in I path d)PIN modulator output in
Q p a t h .................................................................................................... 107
5.20 a) Tim e domain response of 64QAM b)Power spectrum of
64QAM m o d u la to r ..............................................................................107
xvii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.21 Harmonic distortion of PIN diode at f=2.5 GHz a) Harmonic
distortion in PIN diode at I 0 = .1 mA b)Harmonic distortion
in basic m odulator at I 0 = .1 mA c)Harmonic distortion for
the reflection modulator at I 0 = .1 mA d)Harmonic distor­
tion for the PIN diode vs. current e)Harmonic distortion for
the basic m odulator vs. current f) Harmonic distortion for the
reflection m odulator vs. c u r r e n t .......................................................109
5.22 Inter-m odulation distortion of PIN diode (frequencies are f=2.25
and 2.75 GHz) a) IMD in PIN diode at I 0 = .1 mA b)IMD in
basic m odulator at I 0 = .1 mA c)IMD in PIN diode at I 0 = .3
mA d)IMD in basic modulator vs. current e)IMD in PIN diode
vs. current f)IMD in reflection m odulator vs. c u r r e n t ................110
6.1 Direct QAM Modulator
.................................................................... 113
6.2 A Lange C o u p l e r ................................................................................. 114
6.3 Lange Coupler a) S14 and S12 as a function of frequency b)
S l l and S13 as a function of frequency c)Phase Sl4 and S ll
as a function of frequency
................................................................ 115
6.4 A R at-race or ring hybrid a) S14 and S12 as a function of
frequency b) S ll and S13 as a function of frequency c)Phase
S14 and S l l as a function of f r e q u e n c y ......................................... 117
6.5 A series SPD T switch implemented using Alpha PIN 1304
and a forward bias current 20 mA a)Insertion loss b)Isolation
c)R eturn l o s s ....................................................................................... 119
6.6 A reflection attenuator using Alpha PIN 1304 (without pre­
distortion circuit), a)Attenuation, b)Return Loss, c)Phase shift 122
xviii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.7
a)PIN diode resistance versus forward bias current, b) A ttenuation
of reflection atten u ato r versus PIN diode bias current, c)The
required pre-distortion characteristics d) Attenuation versus con­
trol v o l t a g e ........................................................................................... 124
6 .8
A reflection phase shifter using varactor SMV120412 from Al­
pha a)Phase shift b)Insertion Loss c)Return loss
6.9
...................... 125
Phase shift of SMV120412 as a function of control voltage . . 126
6.10 A Wilkinson power combiner a)Insertion Loss b)Return loss
c) Phase shift
........................................................................................127
6.11 I-Q path attenuation for M-QAM M odulation a)4QAM b)16QAM
c )6 4 Q A M
128
6.12 Constellation diagram for the direct M-QAM m odulator a)
4QAM. b) 16QAM. c) 64Q A M .......................................................... 129
6.13 The side view of a shunt PIN diode in S i-M M IC ........................ 130
6.14 PIN diode resistance as a function of forward bias current in
Si-M M IC..................................................................................................131
6.15 A Lange coupler in Si-MMIC implementation a) S u and
[dB] b) S n and S 1 3 [dB]
....................................................................131
6.16 A Rat-race coupler in Si-MMIC implementation a) S 3 4 and
S 2 4 [dB] b) S 4 4 and S 1 4 [dB] c) S 2 4 and 5 3 4 [d e g re e ]................... 132
6.17 An SPDT switch in Si-MMIC implementation a)Switch topol­
ogy b)Pin diode equivalent circuit c) Insertion loss d) Isolation
e) Return loss of s w itc h ....................................................................... 134
6.18 Attenuation of reflection attenuator as a function of bias current 136
xix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.19 A W ilkinson combiner in Si-MMIC implementation a) Inser­
tion loss, <Si2 and S 1 3 [dB] b) R eturn loss, S u [dB] c) Phase
shift, S 1 2 and S\z [degree]
.................................................................136
6.20 Layout of the polarity m odulator section of direct microwave
QAM m odulator
................................................................................ 139
6.21 Layout of the attenuator section of direct microwave QAM
m o d u la to r............................................................................................. 140
7.1 The adaptive direct QAM m odulator in a fixed wireless ATM
s y s te m .................................................................................................... 142
7.2 An adaptive direct QAM m odulator for wireless A T M ................ 143
7.3 a) Experim ental estimation versus effective bandwidth with
buffer size as a param eter for S ta rW a rs movie, b) Effective
bandw idth and peak rates of different video traces c)Experim ental
estim ation versus effective bandwidth with buffer size as a pa­
ram eter for aggregate video. d)The peak bandwidth, effective
bandw idth, and modified effective bandwidth for aggregate
video traffic. The cell loss ratio is 10- 6 in a Ricean channel . . 145
7.4
a) Sim ulation results for M-QAM in a Ricean channel for LAN
traffic based on the number of connections b) Constellation size
distribution in a Ricean channel for LAN traffic c) Simulation
results for M-QAM in a Rayleigh channel for LAN traffic based
on the num ber of connections d) Constellation size distribution
in a Rayleigh channel for LAN t r a f f i c ............................................. 150
7.5
ATM call acceptance improvement compared with fixed QAM
m odulator in Ricean and Rayleigh channels for com puter d ata
traffic
.....................................................................................................151
xx
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7.6
The outage improvement compared with fixed QAM in Rayleigh
and Ricean channels for computer d a ta t r a f f i c ............................ 151
7.7
a) Simulation results for M-QAM in a Ricean channel for VBR
video traffic based on the number of connections b) Constel­
lation size distribution in a Ricean channel for VBR video
traffic c) Simulation results for M-QAM in a Rayleigh channel
for VBR video traffic based on the number of connections d)
Constellation size distribution in a Rayleigh channel for VBR
video t r a f f i c .......................................................................................... 153
7.8
ATM call acceptance improvement comparison of the adap­
tive m odulator and the fixed QAM m odulator in Ricean and
Rayleigh channel for video t r a f f i c ................................................... 154
7.9
The outage improvement of the adaptive m odulator compared
with fixed QAM in Rayleigh and Rice channels for video traffic 154
7.10 a) Simulation results for M-QAM in a Ricean channel for mul­
tiple LAN traffic sources b)Constellation size distribution in a
Ricean channel for multiple LAN traffic sources c) Simulation
results for M-QAM in a Rayleigh channel for multiple LAN
traffic sources d) Constellation size distribution in a Rayleigh
channel for multiple LAN traffic sources
...................................... 157
7.11 a) Simulation results for M-QAM in a Ricean channel for mul­
tiple video traffic sources b)Constellaticn size distribution in a
Ricean channel for multiple video traffic sources c) Simulation
results for M-QAM in a Rayleigh channel for multiple video
traffic sources d) Constellation size distribution in a Rayleigh
channel for multiple video traffic s o u r c e s ...................................... 159
xxi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7.12 Constellation diagram for direct QAM m odulator a) 4-QAM.
b) 16-QAM. c) 64^QAM. d) 2 5 6 -Q A M .......................................... 163
7.13 The power spectrum of the Si-MMIC direct microwave and an
ideal 64QAM m odulator without using f i l t e r ................................ 164
7.14 Baseband Power Spectrum of M-QAM m odulator implemented
using Si-MMIC without using baseband filter for different lev­
els of m o d u la tio n ................................................................................. 165
7.15 Input current and pin diode forward bias resistance for I and
Q paths in 64QAM m odulator using a filter with roll-off factor
or = . 5 ..................................................................................................... 167
7.16 Baseband power Spectrum for 64-QAM using direct microwave
Si-MMIC implementation for 155(OC-3) Mbps data with dif­
ferent roll-off f a c t o r ..............................................................................168
B .l
Reflection attenuator and equivalent circuit for the PIN diode
term ination.............................................................................................. 194
B.2
PIN diode equivalent circuit with an open circuited stub in
s e rie s ........................................................................................................ 194
C .l
a)A lowpass filter as a bias circuit for the PIN diode b)Bias
circuit frequency r e s p o n s e .................................................................196
C.2
A MIM capacitor and a spiral inductor on Silicon substrate . . 197
C.3
Lowpass filter bias circuit implemented using Si-MMIC tech­
nology ..................................................................................................... 197
C.4
The atten u ato r layout along with the bias c i r c u i t ..................... 198
C.5
The SPDT layout with the bias c i r c u i t .........................................198
xxii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF A BBR EV IA TIO N S
AJBR
Available B it R ate
ATM
Asynchronous Transfer Mode
AWGN
Additive W hite Gaussian Noise
BER
Bit Error R ate
B-ISDN
Broadband Integrated Services Digital Network
BPSK
Binary Phase Shift Keying
CDV
Cell Delay Variation
CDVT
Cell Delay Variation Tolerance
CLR
Cell Loss R atio
CNR
Carrier to Noise Ratio
CPWG
Co-planar Waveguide
CTD
Cell Transfer Delay
DC
Direct Current
FDMA
Frequency Division Multiple Access
GOP
Group of Pictures
HDTV
High Definition TV
IF
Interm ediate Frequency
LAN
Local Area Network
LMDS
Local M ultipoint Distribution Service
LOS
Line of Sight
LRD
Long-Range Dependency
MBS
Maximum B urst Size
MIC
Microwave Integrated Circuit
MMIC
Monolithic Microwave Integrated Circuit
MoD
M ultimedia on Demand
MPEG
Motion P icture Expert Group
MPSK
M-ary Phase Shift Keying
xxiii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
M-QAM
M-ary Quadrature Amplitude Modulation
PAM
Pulse Amplitude Modulation
PC R
Peak Cell Rate
QAM
Quadrature Amplitude Modulation
QoS
Quality of Service
QPSK
Quadrature Phase Shift Keying
RF
Radio Frequency
SCR
Sustainable Cell Rate
SONET
Synchronous Optical NETwork
SPDT
Single Pole Double Through
SNR
Signal to Noise Ratio
TDMA
Time Division Multiple Access
UBR
Unspecified Bit Rate
VC
V irtual Channel
VoD
Video on Demand
VP
V irtual Path
WAN
W ide Area Network
xxiv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.
IN T R O D U C T IO N
The beginning of a new millennium has coincided with the emergence of
a new generation of wireless systems collectively called broadband wireless
communication services. Following the emergence of the first generation of
wireless communication services, which covers the car phone and analog cel­
lular era, a second generation, characterised by the development of digital
cellular technology and cordless telephony came into being. A new genera­
tion is now under development with the aim of supporting broadband com­
munication services over a wireless channel. The broadband wireless system
must use an efficient networking technique to transm it the different classes
of traffic, such as com puter data, voice, and video, that have different traffic
characteristics.
The introduction of the Asynchronous Transfer Mode (ATM) network­
ing technique has been an im portant contributing factor in the realization of
wireless broadband communications [1]. The ATM is an efficient networking
technique th a t is designed to transm it different classes of traffic over the same
network and can be used as a platform for broadband wireless communica­
tions. Moreover, the ATM technique could also provide the capability for
delivering bandwidth-on-dem and to portable terminals. Further, the ATM
network has the feature of statistical multiplexing which maximises the use of
the available wireless network capacity when traffic originates from multiple
sources.
In a broadband wireless network, an optimum transm itter has to take
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
into account both radio communication models and broadband traffic char­
acteristics. This has opened a new challenge for the telecommunications
community. T here is presently a considerable amount of research on wireless
broadband networking using the ATM technique. Following some early re­
search before 1995, this subject is now entering the m ainstream in terms of
research, technology, and standardisation.
1.1
1.1 .1
B ro a d b a n d W ireless C om m u n ication s
A T M N etw ork s
The ATM technique, called a fast packet switching method, has been pro­
posed to carry different classes of traffic in the same medium by using short,
fixed length packets. The basic idea of the ATM is to segment the digital
d ata of different sources into a sequence of blocks, called cells, th a t are trans­
ported and routed through a telecommunications network. The length of the
cell is 53 bytes, with the first 5 bytes consisting of header information and
the remaining 48 bytes consisting of the information payload. B oth real time
traffic, such as voice and video, which may tolerate some loss but not delay,
and non real tim e traffic such as computer data, which may tolerate some
delay but not loss, should be efficiently transm itted using the same medium.
An ATM traffic source may demand a bandwidth with a high peak for
a short duration. These sources are generally referred to as bursty traffic
sources. A m ain feature of the ATM is the statistical multiplexing. Using
the statistical m ultiplexing concept, different classes of traffic are carried in
the same channel with a bandwidth smaller than the aggregate peak rate
of the sources. T he ATM network has been designed to use a high quality
transmission system , in which transmission bit errors are rare and randomly
distributed. This is a reasonable assumption for optical fiber based systems.
ATM networks using optical fibers have been realized successfully.
2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Multimedia Terminal
Multimedia Terminal
Multimedia Terminal
ATM
Network
v.
High Speed Wireless ATM Link
Optical fiber
Terrestrial
Cellular
Radio ATM Switch
F ig u re 1.1
Radio ATM Switch
The typical model for radio ATM communica­
tions
1 .1 .2
R a d io A T M
Radio ATM is an evolving technology that has not been tested in the
m arket yet. Although in the past two years several companies have pro­
posed different methods for implementing radio ATM technology, an overall
architecture has yet to be designed. A typical architecture for a radio ATM
network is shown in Fig.1.1. The model considers the ATM network using
ATM switches, optical fiber, and a wireless channel as transmission systems
[2], [3]. High speed wireless ATM links (e.g., STM-1 or OC-3) are installed
between the nodes th a t are not connected by fiber. These radio links may
be used for ease of deployment or to gain system portability, and to realize
a network backbone. As shown in Fig.1.1, lower speed ATM traffic is de­
livered to the m ultimedia terminals by radio ATM cellular networks. A key
element in this design is the radio ATM transm itter. Two different versions
of this unit have to be used in the network architecture shown in Fig.1.1 as
its operation has to be optimised to maximise the utilisation of the wire­
less network resources. The ATM radio transm itter for the fixed point to
point link will be different from the cellular radio type transm itter for the
m ultimedia transmission to the mobile terminals.
A reference model for the fixed radio ATM link is shown in Fig.1.2. As
3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Video
Dest.
Fixed
Fixed
Radio ATM
Radio ATM
Video
Dest.
Video
Source
F ig u re 1.2
The reference model for fixed radio ATM com­
munication services
shown in this figure, a number of sources with different traffic characteristics
compete to send their traffic through a high speed point to point radio link.
Each source uses a dedicated path to connect to the high speed radio ATM
site. It is assumed th at the connections from traffic sources to the radio
ATM site have enough bandwidth.
The paths can be wired or wireless.
Moreover, the traffic sources may have different traffic characteristics with
different quality of service (QoS) requirements. As a typical application for
this system, one may think of the individual sources as a number of video
suppliers trying to deliver their movies to a lot of houses in a geographically
diverse area (e.g., an island) using a high speed radio ATM link. Such models
can be used to study the recently introduced local multipoint distribution
services (LMDS). An important issue in this model is the estimation of the
radio bandw idth for an ATM network. This is discussed in the next section.
1.1 .3
B a n d w id th E stim a tio n in R a d io A T M N etw ork
ATM bursty sources are characterised by th eir peak rate, mean bit rate,
burstiness, as well as their required QoS [4]. An essential task in the ATM
network design is to develop a realistic traffic model to estimate the required
4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Iliiiiiilo
High Speed Radio Link
Video
Source
bandwidth of ATM connections.
This requirement for bandw idth has to
be matched with the capacity of the radio system which depends on the
available spectrum and wireless channel characteristics. A metric based upon
a realistic traffic model is obviously needed. Such a metric should be based
on:
• The QoS requirements of various connections,
• The traffic model,
• The wireless channel characteristics.
These requirements are studied in this thesis and a metric, called a Modified
Effective Bandwidth [5], is derived in Chapters 3 and 4.
1 .1 .4
A d a p tiv e M o d u la tio n
As stated above, in a radio ATM network, various types of traffic with
different quality of service (QoS) requirements are carried over a wireless
channel. The variable bandwidth demand for diverse traffic can be efficiently
transm itted over radio only if the radio ATM transm itter operates as an
adaptive unit th a t is controlled by the ATM traffic characteristics. Due to
the demand for the spectrum , novel adaptive techniques th a t can dynamically
allocate resources are highly desirable. The adaptation process may introduce
some additional delay which must be kept below an acceptance level.
To ensure an overall QoS, radio channel performance must also be taken
into account. Because of spectrum congestion and bandwidth lim itations, the
adaptive transm itter has to use comparatively high level modulations th at
are spectrally efficient. This makes the design of such an adaptive transm itter
a challenging task [3]. The requirements for the adaptive transm itter may
be summarised as:
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
• An ad aptation m ethod based on ATM traffic characteristics and wire­
less channel performance,
• A highly bandwidth-efficient modulation m ethod,
• A real time adaptation process.
These requirements are studied in this thesis and an adaptive multilevel
quadrature am plitude m odulator (M-QAM) is proposed and analysed in
Chapters 5, 6, and 7.
1.1.5
M o d u la to r Im p lem en ta tio n
As mentioned above, the radio ATM tran sm itter has to work at high bit
rates using a highly bandwidth-efficient m odulation technique. A low-cost
realization for the m odulator is highly desirable [6]. A low-cost and high
performance necessitate an architecture th a t is suitable for an integrated
circuit im plem entation [7], [8]. It may be concluded th a t the m odulator
must be implemented w ith the following features:
• Operation a t high speed data rates,
• Low cost hardw are realization,
• Im plem entation as an adaptive m odulator.
A new high perform ance direct microwave multi-level quadrature amplitude
m odulation (M-QAM) m odulator that is suitable for integrated circuit de­
sign is proposed [6], [9-12]. The m odulator im plem entation is discussed in
Chapters 5, 6, and 7.
1.2
L ite r a tu r e R e v ie w
During the last few years, several research groups in universities and in­
dustry have initiated research on radio ATM. While an overall architecture
6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
has yet to be designed, various architectures for radio ATM have been pro­
posed in the open literature [13-19]. While some architectures have been
introduced to handle term inal mobility, others are focused on a point to
point radio ATM link [20-24]. An experimental point to point high speed
link (2.488 Gbps) using a 1 GHz bandwidth in a 19 GHz band has been
realized for LuckeyNet network [20]. The link used a QPSK m odulator with
sub-harmonic pumping.
The performance of a point to point ATM link has been analysed by Gans
et al. [21]. The main conclusion of this work is th at it is im portant to make
the radio link error rate performance similar to the SONET network. The
use of an error code with an ATM interleaver was recommended in this work
as well. The transmission of high speed ATM cells in SDH frames has also
been studied in [22]. It is suggested th at a high level QAM m odulator is
the most suitable candidate for future high speed communications systems.
The measurement of cell loss ratio of ATM traffic over radio in a 20 GHz
band has been reported in [25]. The experimental results of this study are
in agreement with the results in [2].
In an ATM network, a transmission link is divided into a num ber of dif­
ferent virtual paths (VP), each with a unique identifier. Each of these VPs
may be then subdivided into a number of separate virtual channels (VC)
with a separate identifier. A VC is a communication channel th a t provides
for transport of ATM cells. A VP is a bundle of VCs. Two switching con­
cepts namely VP and VC switching, have been used based upon these logical
connections in ATM architecture [26-29]. Both switching concepts effectively
provide an efficient and flexible capacity allocation mechanism. However, the
VP switching needs a simpler ATM switch but is less flexible [27], [28].
7
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
In a broadband integrated network, a capacity controlled radio may be
designed using the virtual path switching concept. T he concept of VP switch­
ing to design a radio ATM was introduced in the early 1990’s in [24]. A
specific ATM switch design for radio ATM communications was proposed in
[23]. Based upon previously published theoretical results [24], the use of an
individual QAM m odulator th a t corresponds to each virtual path was rec­
ommended in [23].
The effective bandw idth as a metric of QoS is a metric th at has been
commonly considered to estim ate the capacities of virtual paths. The effec­
tive bandwidth (in bits per second) of a time varying traffic source (e.g., a
VP) represents the minimum required bandwidth th a t it needs to meet QoS
guarantees. The idea of an effective bandwidth for single and aggregate traf­
fic sources was originally proposed using a stochastic theory for d ata sources
[30], [31]. O ther m ethods such as large deviation approxim ations [32] and
uniform arrival and service models [33] have also been used to compute the
effective bandw idth of single and aggregate traffic sources.
An accurate estim ation of effective bandwidth requires a good statistical
model. A commonly used assumption has been to consider the source as
having exponentially distributed burst and idle periods and to approximate
the real source with an n-state Markovian source [34-37]. The autocorre­
lation functions of these conventional traffic models drop off exponentially.
However, certain recent measurements of broadband traffic, e.g., local area
network (LAN), wide area network (WAN) and VBR compressed video, re­
veal long-range dependency (LRD). An im portant class of stochastic models
that can account for LRD is self-similar models [38-40]. The autocorrelation
function in a LRD model drops off slowly (typically as a power function).
These interesting observations have changed views about traffic characteris-
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
tics. It is known th a t the effective bandw idth based upon Markovian models
underestim ates the cell blocking probabilities when a source exhibits only
m oderate long-range dependence [40]. Consequently, the effective bandwidth
of a source or aggregate of sources m ust be computed based upon a self­
similar traffic model. This is a challenging problem th at has been addressed
in networking theory [41-43]. A function to estimate the effective bandwidth
based upon the self-similarity has been recently reported [44], [45].
As previously mentioned, the adaptive realization of radio ATM trans­
m itter is very desirable [46]. On one hand, an adaptive tran sm itter may
dynamically increase the capacity of the link when there are demands for
more bandw idth; on the other hand, the adaptive transm itter may improve
the radio link availability and decrease the outage when the bandw idth de­
mands are not at peak. A capacity controlled multilevel QAM (M-QAM)
m odulator can be used to realize an adaptive radio ATM transm itter. The
idea of an adaptive QAM m odulator was originally introduced for fading
channels [47], [48], where an M-QAM modulator was used to improve the
system gain and throughput [49]. T he M-QAM is also considered to absorb
traffic variations in a system in which a spectrum band is assigned to each
virtual p a th [23].
Direct microwave m odulator implementation has also been studied ex­
tensively in the literature. In the early 1990’s, the idea of a direct microwave
m odulator was presented [50]. The conventional method of generating the
m odulation at IF, followed by an up-conversion to the transm it frequency,
is not efficient in power, size, and cost [51]. An extensive reported study
shows the advantages of a direct microwave modulator over the conventional
m ethod [52], [53]. In [54], a direct microwave QAM m odulator was imple­
mented by directly mixing an R F signal with a baseband signal. This method
9
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
of direct m odulation has also been used for QPSK and vector modulators [55],
[56]. An alternative m ethod of implementing a direct microwave m odulator
is to use a variable attenuator. A direct conversion m odulator for 7t/ 4 QPSK
using a variable atten u a to r in the 1.9 GHz band was reported in [57]. This
method has also been used to realize a vector m odulator [58], [59]. Direct
microwave m odulation techniques can be used for various modulation tech­
niques [58], [60], [61], and [62].
To implement a direct microwave m odulator with a variable attenuator, a
PIN diode is a suitable electronic device. A PIN diode can be used as a suit­
able electronically tuned resistor in the attenuator due to its power-handling
advantage [63]. Recent progress in MMIC technology for silicon microwave
devices [64], [65], as well as the growing demand for high speed communica­
tions, makes high speed modulator implementation using PIN diodes a viable
option [58].
As stated in Section 1.1.2, the objective of this thesis is to design an
cost-effective wireless ATM transm itter for fixed applications with an effi­
cient bandwidth utilization. The above literature review shows th at a good
architecture for a radio ATM transm itter has not yet been reported. There is
a great demand for a direct microwave m odulator due to the rapidly increas­
ing demand for high bit rate wireless ATM systems. An accurate adaptation
m etric to adapt the m odulator according to incoming traffic has not yet been
reported. A real tim e adaptation metric will improve the throughput of the
link and increase radio system availability. In addition, a direct microwave
high speed QAM m odulator implementation using a variable attenuator has
n ot been reported yet. This implementation will be very useful if a variable
attenuator can be implemented using PIN diodes. However, a suitable model
for analysing a PIN diode operation in a high speed application has not been
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
addressed in literature [66]. Each of these issues will be addressed in this
thesis.
1.3
R esea rch O b jectives
The objectives of the research work reported in this thesis are as follows:
• To design an optimum architecture for a fixed wireless ATM transm itter
• To derive a metric to estimate the required bandwidth of the ATM
traffic sources in the wireless network
• To design an adaptive transm itter th at optimises the utilisation of the
wireless resources
• To design a high performance QAM modulator for high speed data
rates
• To develop a hardware implementation for the proposed modulator,
leading to a cost effective radio ATM
• To model and characterise a PIN diode to use in the high bit rate
m odulator.
1.4
T h e sis O rgan isation
In addition to the introductory chapter, this thesis contains seven chap­
ters.
In C hapter 2, design concepts for a radio ATM transm itter are developed.
The focus is to introduce an architecture to realize an optimum fixed radio
ATM transm itter. Following a model extraction for an effective bandwidth
estimation, the physical layer issues in ATM networking, adaptive transm it­
ter design, and direct microwave QAM m odulator are described.
11
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In Chapter 3, an accurate metric to estim ate the required bandw idth of
broadband traffic in wireless channels is introduced. The metric is estab­
lished by examining the required bandw idth for empirical video and d ata
traffic traces, and developing a statistical model to estimate the effective
bandwidth in a wireless channel. The statistical results are compared with
those obtained from computer simulations.
Following extraction of a capacity reduction factor for m ultipath fading,
a model to realize an adaptive M-QAM transm itter, which optimises the
wireless resource utilisation, is described in C hapter 4. The transm itter per­
formance is analysed for different ATM traffic loads in Ricean and Rayleigh
channels.
In Chapter 5, a new direct microwave QAM architecture is introduced.
An analytical m ethod is developed to investigate the operation of direct mi­
crowave QAM m odulators using PIN diodes.
This requires an analytical
model to characterise the PIN diode for high speed operation.
Microwave integrated circuit (MIC) and silicon monolithic microwave in­
tegrated circuit (Si-MMIC) techniques for implementation of the direct mi­
crowave M-QAM m odulator are investigated in Chapter 6. The microstrip
and co-planar waveguide (CPWG) line circuits are used to realize the mod­
ulator. While realization using the MIC is simple and straightforward, the
Si-MMIC realization offers a low cost and high performance implementation
method.
A system study based on analytical, sim ulation and measurement results
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
is reported in C hapter 7. Finally, conclusions are presented in C hapter 8.
Some of the results of this research work have been patented [7], published
[2], [5], [6], [8], [10], [121], will publish [12], and subm itted to [3], [9], [11]
various journals and conferences.
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.
FU N D A M EN TA L CONCEPTS FOR
W IRELESS ATM T R A N SM IT T E R
T he design of a broadband wireless network using ATM techniques re­
quires th a t networking issues and radio transmission issues be considered
simultaneously. A practical design m ust take into account the ATM net­
working characteristics and wireless transmission parameters [67]. In this
chapter a brief discussion of some fundamental concepts in ATM networking
and high speed wireless communications are presented. The chapter is organ­
ised as follows. Following a review of ATM networking, a model is developed
to estim ate the required bandwidth of a high speed ATM link. The impact
of the cell loss ratio due to the wireless channel on the end-to-end QoS is
studied in the next section. This is followed by a discussion of an adaptive
m odulator for a fixed wireless link.
2 .1
2 .1 .1
F u n d am en tals o f A T M N etw ork in g
A T M B asic C o n cep ts
T he basic idea of the ATM technique is to segment a digitally coded
inform ation stream (such as data, voice, and video) into a sequence of el­
em entary blocks, called cells, th at are transported and routed through the
telecommunication network. The cell is the basic element in ATM network­
ing. It is a short and fixed length packet of information. Cells pertaining to
independent connections and with different generation rates can be carried
on the same link and are accepted as they are generated. This uses statisti­
cal m ultiplexing instead of fixed-frame tim e division multiplexing. T he ATM
technique is based on packet processing thus enabling the implementation of
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Information Field
Header
/
*■“
^
s
-* — _
✓
^
y
VPI
PT
VCI
-
^
CLP
HEC
______________________ Header___________________ ^
F ig u re 2.1
ATM cell format for network node interface
(NNI)
basic network functions using a high bit rate [68].
The cell, which is 53 bytes long, is made of two m ain fields: the header
used for switching, and the information field, which transports the user in­
formation. The content and structure of the header are different at the user
network interface (UNI) and at the network node interface (NNI) [3]. The
cell header is 5 bytes long and contains two identifiers to route two separate
logical entities: the virtual channel (VC) and the virtual path (VP). An NNI
cell header is shown in Fig.2.1. The header is divided into virtual path iden­
tifier (VPI), virtual channel identifier (VCI), payload type, cell loss priority
(CLP) and header error correction (HEC). HEC is employed to protect the
header from transmission errors. It can correct one bit error in the header
and detect a large class of multiple bit errors. In the sender, the eight bit
cyclic redundancy check (CRC) across the first four octets of the cell header
are calculated and the results are inserted in the HEC field which is the 5th
byte of the header. It is specified by the generator polynomial [68]
g(x) = x s + x 2 -t-rr-F l.
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.1)
VP
High Speed Link
VC
F ig u re 2.2
Connections for ATM networking
The physical layer is divided into a num ber of different virtual paths,
each with a unique VPI. Each of these V Ps may be then subdivided into a
number of separate virtual channels, each with a separate VCI. The physical
configuration is illustrated in Fig.2.2.
2.1.2
A T M S erv ices
The services to be provided by ATM networking may be categorised into
interactive services and distribution services.
Interactive services include
video and m ultim edia conferencing, transmission of high-resolution images,
document browsing as well as other applications [69]. Distribution services
involve distribution from centralised service providers to network users e.g.,
files, stored video, movies, images and so on. These services have various
QoS requirements. Different services in an ATM network are classified in the
following classes:
• CBR: Constant bit rate connections require a fixed bandwidth channel
with tight delay variation constraints (e.g., voice),
• VBR-rt: Real tim e variable bit rate (VBR-rt) services that require a
small delay and fixed tim ing relationship (e.g., audio/video for real
tim e conference),
• VBR-nrt: Non-real time variable b it rate connections that are used for
bursty applications which do not require real-time performance and in
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
which the delay does not have a heavy impact on th e service (e.g., data
transfer),
• ABR: Available bit rate services th at do not support bounded delay
for delivery but support variable bit rate traffic. T he network provides
a best effort service which is not suitable for real tim e applications,
• UBR: T he unspecified bit rate connections that are not specified by any
param eters allowing the user to send any amount of the d ata over the
network. However, using this service, the network gives no guarantee
on cell loss rate, delay, and delay variations.
Typically a large amount of the traffic in an ATM network is VBR services.
2.1.3
B a n d w id th D e m a n d an d Q oS R eq u irem en ts
Different ATM services require various bandwidth and QoS performance.
These requirements are specified by traffic and QoS related param eters. A
traffic param eter, e.g., peak cell rate (PCR), is specified for the source, and a
QoS param eter, e.g., cell loss ratio (CLR), describes the expected QoS. Ac­
cording to the ATM Forum [69], various QoS and traffic requirements include
cell loss ratio (CLR), cell transfer delay (CTD), cell delay variation (CDV),
cell delay variation tolerance (CDVT), peak cell rate (PC R ), sustainable cell
rate (SCR), maximum burst size (MBS), and minimum cell rate, among oth­
ers. An ATM call is adm itted under a traffic contract consisting of a subset of
these param eters. The QoS related param eters (CLR, C TD , CDV) and the
peak bandw idth requirements for voice, interactive video, media playback
(e.g., HDTV), and file transfer are shown in Table 2.1 [70].
2.2
A T M M u ltip le x e r
An ATM multiplexer is shown in Fig.2.3. In ATM terminology, a source
is a term inal. This could be a telephone handset, a video player, or a multi17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
T ab le 2.1
QoS Requirements
V oice
V ideo
Media P layback
F ile
CLR
1 0 - 3 _ 10“ 6
10"9
1 0 - 10
IO-12 _ 10-9
CDV(m sec)
0 -5
5
CTDmax (777.5 6 c)
10-150
10
1000
PCR (Mbps')
8 -6 4 kbps
10
44(SCR=15)
media computer. When a source wants to transm it information, it requires
the establishment of a virtual channel (VC). Once a source has established a
VC, it generates a stream of cells, each cell consisting of 53 bytes. A typical
cell stream generated by a VC consists of silent periods, during which no
cells are generated, and activity periods, during which cells are generated
at a variable rate. The group of cells generated during an activity period is
called a burst.
An ATM multiplexer consists of a data buffer and a high speed link; the
buffer receives the cells generated by the established VC and transm its these
cells, one after another, onto the high speed link [37]. A VC’s allowable cell
delay and cell loss are specified by its quality of service (QoS) requirements.
To guarantee th a t all established VCs meet their QoS requirements, the mul­
tiplexer may have to deny certain VC establishment requests.
When a VC demands capacity to be transm itted using an ATM mul­
tiplexer, if there is enough capacity in the high speed link, its request is
accepted; otherwise the call will be rejected. An admission policy is needed
to control the call acceptance process. A simple admission control is the peak
rate admission policy [69]. The peak rate admission constrains the aggregate
peak rate to be less than the transmission capacity of the high speed link. It
enforces these constraints by rejecting a VC establishment request when the
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
burst
burst
VC
Finite
Buffer
VC
cell
High Speed Link
VC
F ig u re 2.3
An ATM multiplexer
VC’s peak rate added to the sum of the peak rates for the established VCs
exceeds th e transmission capacity. As an example, suppose a source has an
average bandw idth of 20 Mbps and a peak bandwidth 55 Mbps. Peak band­
width allocation requires that 55 Mbps be reserved at the high speed link for
this specific source, independent of whether or not the source transm its 55
Mbps continuously.
Let C denote th e transmission capacity of a high speed link; K denote
the num ber of services, and b\ , ..., 6* denote the peak rates for the K services.
The V C profile is( n l5 ...,nk), where n k is the current number of established
V C s of service-k. Since V C s arrive and depart,the V C profile changes with
time. Peak rate admission admits a new service-k V C if and only if
&* + 5 > n , < C .
i=i
(2.2)
Consequently, at all times the VC profile (n-i, ..., t) satisfies
J 2 b kn k < C .
k=l
(2-3)
On the other hand, statistical multiplexing perm its the aggregate peak
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
rate to exceed th e transmission capacity. This technique can utilise the link
more efficiently, allowing the link to transm it at its maximum rate even when
some of the established V C s are silent. To implement the statistical mul­
tiplexing, we m ust determine w hether a given collection of established V C s
meets the QoS requirements.
An effective bandwidth admission is an admission policy which is easy to
implement when the ATM m ultiplexer operates in the statistical multiplexing
mode. This policy is characterised by an effective bandwidth vector (b\ ,.... bek )
and it adm its a new service-k VC if and only if
+
(2.4)
1=1
where (n i,..., n*) is the current VC profile. However, determining suitable
values for b\ , ..., b% is a challenging problem in ATM networking theory.
2.3
B a n d w id th E stim a tio n
In an ATM network, a virtual path is an information transport th at makes
a logical direct link between two nodes and accommodates a number of vir­
tual channels simultaneously. A predefined route is usually defined for each
virtual path in th e physical layer. Each virtual path has a bandwidth, in
other words, “capacity” , which defines the upper limit for the total virtual
channel bandw idth carried by it. V irtual paths are also multiplexed on phys­
ical transm ission links using cell m ultiplexing [71]. While m apping various
services to the V Ps is an active research subject [72],[73], one proposal sug­
gests having a separate VP for each sendee class or even for each different
set of QoS requirements within th e same service class. Although QoS control
is easier, the to ta l number of V Ps could become large.
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
burst
burst
VC
Finite
VC
Virtual Path
Buffer
cell
VC
F ig u re 2.4
A Finite Buffer with VP as an output
Let us again consider the multiplexer model shown in Fig.2.3. Each high
speed link in Fig.2.3 can be considered a virtual path. This is illustrated in
Fig.2.4. As may be seen, different VCs have to compete to use the available
bandwidth of a VP.
The above model may be used to examine the capacity of total VPs in
a physical layer. Fig.
2.5 shows this model with M VPs feeding into a
communication system. Considering the peak rate admission for this model,
one could compute the capacity for V P X as follows:
Y
5: Ci)
(2-5)
k= l
where (n^i, n2>i , ..., ^artli) is the current VC profile for V P \. Similarly, the
VC profile for VP{ will be
E M t l < Ci.
fc=1
(2.6)
Total capacity of the physical connection is obtained using superposition of
the individual VPs. Thus, the total admission policy of the link can be
written as follows:
Ki
Km
Y . fyfc.lTCfc.l + ... + Y , bk,Mnk,M < C,
1
fc = l
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.7)
VP
Finite
VC Vi
Buffer
VC K i.l
VC 1.2
V C ,,
Finite
VP.
Buffer
VC
Physical Layer
*1*
VCUM
VC2M
Finite
VP,
Buffer
F ig u re 2.5
An ATM multiplexer with VP as am output
where C is the total capacity of the high speed link. This equation can also
be written as
M
Kj
Ki
E E &*,•«*,- < c .
( 2 .8 )
i=i k-i
For the statistical m ultiplexing mode, the above equation is modified as fol­
lows:
M
K ij
E k=
E bh ^ , i
i= l
where ( n ^ , n.2 ,i,
(2.9)
< c,
1
nKiyi) is the VC profile for VPi, and (6f
6 ,,,..., beKi i) are
the efFective bandw idths of VCs in V P It is assumed th a t each output port
is connected to a VP.
In a nonblocking switch, as shown in Fig.2.6.a, a buffer has been used in
each output port and the point of connection occurs at the output ports [73].
A switch fabric is nonblocking if it can deliver all packets to the requested
output ports. A simplified model for bandwidth estim ations of a VP is also
shown in Fig.2.6.b. In a network node where the switch routes VPs, this
model can be used as well. Thus, the efFective bandw idth of the ith VP can
be shown to be:
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Resource
Manager
Buffer
Connections
K >
HO-
Non-Blocking
Fabric
HO—
Connection^
Buffer
“
o -
b
F ig u re 2.6
a) A Nonblocking Switch b)Equivalent circuit for
capacity estim ation
BW f = J2 iljn v .
k=1
(2.10)
Then, the capacity of the total link will have the following relation with the
efFective bandwidth of the individual VPs:
M
Y , BW i <C-
(2.11)
I— 1
As can be seen, the efFective bandwidth of the VPs are related to available ca­
pacity of the link.Moreover, this relation showsth at the required
in a high speed link
bandwidth
isbound by the linear superposition of the V Ps’ efFec­
tive bandwidth. These interesting results will be used to design an adaptive
transmission system in the following chapters.
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Video
Source
High Speed Radio Link
Video
Dest.
□
Fixed
Fixed
Radio ATM
Radio ATM
V ideo
Dest.
Video
Source
F ig u re 2.7
T he reference model for fixed point to point ra­
dio ATM link
2 .4
E n d -to -E n d Q oS G u aran tees in R a d io A T M N e t­
w orks
End-to-end performance with multiple traffic types and multiple QoS
requirements needs the derivation of an overall bound th a t always guarantees
the operation of a wireless ATM network with desired QoS [69]. Estimation
of this bound resulting in an optimum network utilisation is a very difficult
task. While extraction of an end-to-end delay bound is in itself an interesting
networking problem th a t has attracted much attention [74], [75]. the focus
of this research is on the cell loss performance bound. The cell loss ratio is
defined as being the ratio of discarded cells to transm itted cells.
A reference model is considered to estimate a bound for cell loss ratio
in a fixed radio ATM link. This model is shown in Fig.2.7. As may be
seen, two different mechanisms may cause cell loss ratio in the wireless ATM
communications. W hile congestion results in cell loss ratio at nodes, the radio
link type of the physical layer introduces bit errors resulting in an increase
of the cell loss ratio. T he bit errors in an optical fiber link are random and
occur rarely. However, for the radio channel the conditions change with time.
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Such a channel could suffer from severe degradation for some fraction of time
[74].
2 .4.1
C ell Loss R a tio D u e to C on gestion
In an ATM network, congestion could occur at various switches an d /o r
buffers when buffers overflow and cells are dropped resulting in a decrease in
the throughput. In a nonblocking switch, if there are more requests to use a
VPi than it can handle, the cells in the ith buffer will start to drop.
2.4.2
C ell Loss R a tio in a R a d io C hannel
The desired channel for the transmission of ATM traffic is one with a
low bit error rate. The study of the problems associated with transmission
of ATM traffic over a wireless channel is im portant because the original
ATM network was designed to use a high performance physical layer such
as optical fiber. In this section, the im pact of a wireless channel on ATM
traffic is examined. In the following analysis, only the CLR resulting from
the corruption of the header in the ATM cells is considered.
Q uality o f Service for th e R an d om Error C hannel
W hen errors occur randomly, the error distribution is binomial.
The
header error correction (HEC) has been designed to correct a single bit error
in the 40 bit header [76]. Using binomial distribution, the CLR in this case
is given by:
40 /4 0 \
C L R r = £ ^ J p n(l - p ) 4Q~n,
(2.12)
C L R r ~ 780 x p 2,
(2.13)
where p is the bit error rate and C L R r is the cell loss ratio in the random
error channel. Equation (2.13) relates C L R r to the bit error rate in the
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
random channel and shows th at CLR is a second order function of the bit
error rate. In Fig.2.8.a the relation between bit error rate and CLR for a
random channel is presented.
Q u a lity o f Service in th e B u rst Error C hannel
B urst errors increase the probability of cell discard [5j. A suitable prob­
ability model in this case is the Newman-A contagious model [77]. The two
param eters m x and m 2 in this probability model depend on the average burst
length (3, the header length Z, and the probability of bit error p. These pa­
ram eters are given by m L = ^ and m 2 = /?. The probability of k bit errors
P [ X = k] is given by:
P ( X = k) =
] T (!le -0 )X
71=0 P
n'
K-
(2.14)
Considering th a t the cell loss ratio occurs due to uncorrected header error,
the cell loss ratio for the bursty channel is given by:
C L R b = 1 - P ( X = 1) - P ( X = 0),
(2.15)
An approxim ation to Eq.(2.15) can be given by:
C L R b = 1 - e~40p^ [ l + - ^ ( 1 + 0)e~0].
(2.16)
In Fig.2.8.a the relation between bit error rate and CLR for the bursty chan­
nel is plotted with burst width as the param eter. For a given BER a higher
burst w idth actually results in slightly lower CLR. This may beattributed
to the fact th a t when the errors are lumped together they affect fewer cells.
These analytical results are compared with the previously published mea­
surem ents in Fig.2.8.b [25]. As may be seen, the agreement is quite good for
a burst w idth of two bits.
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
10 '
e
2m
eO
j
e
U
2
5
10'
10‘
10 '
10'
10
10 '
10'
10'
Bit Error R ato
10‘
Analytical
M easu rem en t
10 *
•3 10
cc
‘
10‘
10‘
10*
Bit Error R ate
(b)
F ig u r e 2.8
a) CLR versus BER for random and bursty chan­
nels b) A comparison between cell loss estim ation
results for the bursty channel with the previously
published results [25].
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.5
A d a p tiv e R adio A T M T ran sm itter
An im portant characteristic of an ATM network is the bandw idth on de­
mand feature. In this network, the total number of connections varies with
time, and thus, the bandwidth demand varies with time. As stated before
the ATM sources require different QoS. For instance, while the voice traffic is
mostly delay sensitive and relatively loss insensitive, the data traffic is quite
sensitive to loss although some loss can be tolerated. Further, even in a sin­
gle class of traffic, different applications require different QoS. For example
with video traffic, TV broadcasts and video on demand (VoD) services have
more relaxed delay requirements th an video conferences and m ultim edia on
demand (MoD) where a strict delay bound is required to support interac­
tivity.
Also, while TV broadcast may require high picture quality, video
conference and VoD/MoD services may trade off, at times, picture quality
for high bandw idth efficiency [78]. All these issues make an adaptive system
th at can provide bandwidth on dem and highly desirable.
In addition, a radio channel is essentially a time varying channel. The
channel variation results in a different bit error rate (BER) and cell loss
ratio (CLR) variations. Thus, the radio channel must be m onitored so as
to m aintain th e QoS. A simple m ethod to adapt the system to the channel
variation is to use a channel estim ator in conjunction with a feedback channel
where feedback channel only carries the carrier [79]. This is shown in Fig.2.9.
2.6
M o d u la tio n T ech n iq u es for F ix ed R ad io A T M
Due to th e spectrum lim itation in a radio channel, a high bandw idth
efficiency m odulation method is required. As discussed above, an adaptive
realization of the m odulator is also highly desirable. M-level m odulation
techniques such as M-QAM, and M-PSK are possible candidate m odulation
methods for fixed radio ATM because of their high bandwidth efficiency and
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Adaptive
l Time Varying
Transmitter
Output
Demodulator
! Radio Channel
Channel
Estimator
Feedback
Channel
F ig u re 2.9
An adaptive modulator in a tim e varying chan­
nel
IQ
•
•
•
•
16-QAM
F ig u r e
64-QAM
2 .1 0
Constellations for QAM signals
their adaptability features [79]. When the radio bandwidth efficiency and
power requirements are examined, the M-QAM offers a better trade-off. The
QAM m odulation is discussed in the next few subsections.
2.6.1
Q u a d ra tu re A m p litu d e M o d u la tio n
Quadrature am plitude m odulation is a two-dimensional linear modulation
technique and the QAM m odulated signal may be expressed as:
sm(t) = A mcu(t)cos2'Kfct —A m3u(t)sin27rfct;
m = 1,2,.... L,
(2.17)
where u (t) is a signal pulse, f c is carrier frequency, A mc and A ms are the
inphase and quadrature signal amplitudes, and L = y/M . Constellations of
this m odulation in the I-Q planes are shown in Fig.2.10.
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.6.2
S p ectra l E fficien cy o f Q A M
The lowpass equivalent o f the QAM signal is given by:
v ( t ) - ^ 2 ( a n + j b n)u(t - nT)]
an, bn = ± 1 , ± 3 ,..., ± L — 1 ,
(2.18)
n
where T is the symbol rate, and u(t) is pulse shape. The bandpass signal
sm(t) can be related to the lowpass signal v(t) through
sm (t) = Re[v(t)ej2^fct],
(2.19)
The autocorrelation function of sm(t) is given by:
4>ss(t ) = Re [(f) (t ) ej27r^cT],
(2.20)
where (j>vv(r) is the autocorrelation function of the equivalent low pass signal
v(t). Then, the power spectral density of $ ( / ) is given by:
•M /) =
- h ) + $ » ( - / - /=)]
(2 -2 1 )
where $ vv( f) is the power spectral density of v(t). It can be shown [80] that
the power spectral density of v(t) is given by:
® .v (/) =
(2 .2 2 )
where U( f ) is the Fourier transform of u(t) ajid $ a ( f ) is the power spectral
density of the information sequence. The QAM signal may be considered to
consist of two PAM signals conveyed by the cosine and sine carriers. Thus,
the spectral efficiency of the QAM will be twice th at its of PAM components
in the inphase and the quadrature channels. Substitution for $ u ( f ) in the
above equation gives the power spectral density the PAM signal carries by
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
cosine carrier as [80]:
« „ ( / ) = ^ r m f ) \ 2 + $ i I f \ U ( ~ ) \ 26 U - ^ ) ,
m=—oo
J
(2.23)
where g. and cr2 are the mean and variance of the information sequence. As
may be seen, the mean value of the PAM symbols is zero (/i = 0), and, thus,
Eq.(2.23) can be simplified as being
* » ( / ) = “ \U (f)\2.
(2.24)
Thus, the spectral efficiency of the PAM signal is controlled by the pulse
shape u(t). The raised cosine pulse is a commonly used pulse shape for the
digital communications in bandlimited channels. The signal u(t) for this
pulse shape is given by:
{)
sin{irt/T) c o s jr a t/T )
nrt/T 1 - (2a t/T Y -
(
}
where a is called as roll off factor . The Fourier transform u(t) is given by:
T i f O < |/| <
( 1 —0 )
2
r
f [1 - s i n ( f - i f / a ]
if
< [/I <
The normalised spectrum of the raised cosine pulse shape is shown in
Fig.2.11.
Thus, the bandw idth efficiency of QAM m odulator can be adjusted by
param eter a. As may be seen from Fig.2.11, the channel bandwidth, \¥ , is
approxim ately equal to ( l+ o :) /T and, since 1 / T = C/log2(M) symbols/sec,
we obtain the result:
" = W
T - (6ps/H z)
W = 717+7 oc
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
( 2
' 2 6 )
alpha=.25
alpha=.5
alpha=.75
alpha=l
4
20.9
CL
1 0 .8
CO
0 0.7
o 0.1
CL
0.4
0.2
0.6
0.8
fT
F ig u r e
2 .1 1
Raised cosine pulse power spectrum
where rj is M-QAM spectral efficiency, and C is the M-QAM transmission
capacity.
2 .6 .3
P o w e r E fficien cy
Although spectral efficiency may be the prim ary criterion, the power ef­
ficiency of the selected method is also an im portant factor in system design.
Power efficiency is a measure of the received power needed to achieve a spec­
ified bit error rate (BER). For an additive white Gaussian noise (AVVGN)
channel, the probability of symbol error in the M-QAM system is estimated
as follows [80]:
MQAM =
2 (1
- 7 M )eTfc
1- 1(1“ 7
2 (M — 1 )
S )erfc
(fe rrT )^
where k is even (k is bits per symbol and M = 2k) , 7 &is the average signalto-noise ratio (SNR) per bit, and M is the m odulation level. BER results
for coherent dem odulation with 2, 4,
6
and
8
bits per symbol and assuming
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
'-2
■*
■o
■+
~
10
4QAM
F
16QAM
64QAM U
256QAM \:
W 10'
10
'
10
'
20
10
F ig u re
2 .1 2
25
CNR[dB]
30
35
40
BER versus CNR for a QAM m odulator
perfect clock and carrier recovery in a Gaussian channel [81] are shown in
Fig.2.12.
2 .7
Q A M M o d u la to r Im p lem en ta tio n
The QAM m odulator implementation has a considerable impact on the
radio ATM performance and cost. Different design approaches used in the
past could be categorised into the two categories of heterodyne and direct
m odulation method.
Each approach has its specific advantages and dis­
advantages, and a selection is based upon system requirements and design
objectives.
2 .7 .1
C lassic Im p le m e n ta tio n
The QAM m odulator has been traditionally realized using the heterodyne
m ethod shown in Fig.2.13. The QAM mapper in the figure divides a binary
sequence with rate
into two binary symbol streams, each with a rate ft, / 2 .
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Pulse
Shaping
Data
QAM
MAP
0 -f
(E )- "
W
HP A
90 °
Pulse
©
Shaping
F ig u re 2.13 Block diagram of a heterodyne QAM modulator
A 2-to-y /M level baseband converter is used to convert the streams into
y /M -level baseband PAM signals in the inphase or the quadrature paths.
These symbols are shaped by the raised cosine filters. The filtered I and Q
baseband signals are used to modulate the inphase and quadrature outputs
of an IF oscillator. The modulated signal at the IF is up-converted to the
desired transm it frequency in one or more steps. As may be seen, mixers are
commonly used in this realization. A power amplifier is used to boost the
modulated signal at the transm it frequency to the required power.
2 .7 .2
D ir e c t M icrow ave Im p le m e n ta tio n
The classic heterodyne design described above suffers from disadvantages
such as design complexities, high costs, RF filter requirements, and d ata rate
lim itation if IF frequency is low. The direct implementation methods could
effectively overcome these limitations. The basic idea of direct microwave
implementation is to use the baseband signal to modulate a carrier at the
desired transm itted frequency. A basic direct microwave modulator imple­
m entation is illustrated in Fig.2.14.
2.8
R a d io M icrow ave L ink D esig n
It is well known th at m ultipath fading has a significant adverse effect
on the performance of a wireless communications system. In a broadband
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
►
Pulse
Shaping
Data
QAM
HP,
MAP
Pulse
Shaping
9(P
F ig u re 2.14 Block diagram of a Direct QAM m odulator
microwave system, the fading generally consists of a combination of flat and
frequency selective components. The flat component is a time-varying, fre­
quency independent attenuation of the channel response and, hence, of the
signal. This attenuation can be thought of as the slow variation of the chan­
nel loss over a broad range of frequencies. The frequency selective component
of the fading could be considered to be the frequency selective variation of
the power response. It shows itself either as a monotonic gain change (or
slope) across the radio channel bandwidth, or as a dip (or notch) within the
bandwidth. M ultipath fading arises from the fact th at the signal propagates
along several paths, each of different electrical length. At the receiver, the
relatively delayed signal components interfere with each other, and this leads
to the frequency selective effects described above. In a line of sight (LOS)
microwave link, the received signal generally consists of a strong direct wave
and some reflected waves. A Ricean distribution model for fading can be
used in this case [82].
The radio link tolerance to m ultipath fading is usually specified through
the radio link outage probability. This m etric is closely related to the radio
link fade margin and system gain; the system gain being the difference be­
tween transm it output power and receiver threshold sensitivity for a given
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
bit error rate. T he system gain is given by the following relation [83]:
F M — Gs — Lp — h p —Lp + G t + Gr,
(2.27)
where Gs is the system gain, F M is the fade margin, Lp is the free space
path loss(dB), L p is the feeder loss, Lp is the branching loss, and Gp and
G r are gains of transm itter and receiver antennae. The outage probability
is related to the fade margin through the following relation [83]:
10log(l - R ) = 30log(d) + l0 lo g (6 A f) - 7 0 - F M ,
where
(1
(2.28)
—R) is the outage probability, d is the p ath length in k m , / is the
carrier frequency in G H z , and A is the terrain roughness factor.
A typical example of a short haul radio link for an 18 GHz system with
64Q A M m odulation is given in [5]. The path length is 10 km. The same
link is used for the performance study in Chapter 7. The height of transm it
and receive antennae is selected as 7 meters so as to achieve an LOS path of
14.3 km. Both antennae are parabolic dishes with diameters of 2 feet and a
nominal gain of 38.9 dB. The HPA was assumed to be a commercial MMIC
with power outp u t of 23 dBm. The results of this link design are summarised
in Table 2.2.
2.9
Sum m ary
In this chapter different issues to realize a transm itter for fixed wireless
ATM applications were studied. The study examined both ATM networking
and radio transmission issues. A relation was derived to relate the effective
bandwidth of th e VPs to the available capacity of the link. It was shown
th a t CLR in wireless ATM networks results from buffer congestion and radio
channel BER. A relation between BER and CLR was derived for a radio
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
T a b le 2.2
Terrestrial Link Design
T ra n sm itte r
T r a n sm itte r Power
23 dBm
T r a n sm itte r Antenna Gain
3 8 .9 dB
Antenna H eigh t
7 m
Space Link
L oss
1 4 2 .1 9 dB
T o ta l L oss
6 4 .3 9
M u ltip a th L e v e l
-3 0 dBc
R e c e iv e r
R e c e iv e r Antenna Gain
3 8 .9 dB
N o ise F igu re
3 dB
R e c e iv e SNRj
4 6 .8 1 dB
T h e o r e t ic a l SNR&(10-8 )
21 dB
Im p lem en tation L oss
5 dB
Fade Margin
2 0 .8 1 dB
O u ta g e (fo r 1 4 .3 km)
.Q30/°(AvailabLlity99.970/ 0)
channel. The concept of an adaptive m odulator was presented to provide
bandwidth on demand over a wireless channel. The QAM m odulator was
selected as a suitable modulation technique to realize an adaptive m odulator
in a wireless channel. Different implementation techniques to realize an MQAM m odulator were investigated and a direct microwave implementation
was selected as a better choice for fixed wireless ATM applications. The
following chapter discusses the concept of bandwidth estimation in a wireless
broadband network.
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.
B A N D W ID T H ESTIM A TIO N OF
B R O A D B A N D T R A FFIC
The bandwidth estim ation of the broadband traffic sources is an impor­
tan t issue in realizing a wireless broadband network- The concern is how a
unique traffic param eter should be defined to represent the required capacity
in a wireless broadband network. This issue is im portant in wireless ATM
design as the capacity is closely related to the cost and complexity of the
radio ATM.
A straightforward design is to develop a radio to provide enough band­
width for the peak rate of existing traffic sources. Although this solution is
simple, it is neither an optim um nor a practical design. It is not optim um
because at a given tim e using all available resources, it only provides service
to some of the users. It is not practical because a very' broadband radio sys­
tem needs a very broad spectrum band and a rather complex system design.
A better alternative method is a demand based design. In this method,
the bandwidth is assigned to the different traffic sources based upon their
traffic characteristics. This m ethod, however, requires th a t the traffic sources
describe their traffic and QoS param eters before a link can be established.
In this chapter an accurate m etric to characterise the required bandwidth
in a wireless ATM network is introduced. The m etric is established by ex­
amining the required bandw idth for empirical video and d a ta traffic traces,
and developing a statistical model to estim ate the effective bandwidth in a
broadband network. The results of the statistical model are compared with
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
120
£: 100
•Jurassic Park MPEG-1
80
60
2
40
lU
Time [sec]
h 100
Jurassic'Park MPEG-1
co 40
200
F ig u re 3.1
400
600
800
1000
Timetsec]
1200
1400
1600
a) First sixteen seconds of Jurassic Park video­
trace b) Total trace(27 minutes)
simulation in Section 3.4.
3 .1
C h a ra cterisa tio n o f B ro a d b a n d Traffic
It is well known th a t a broadband traffic source, e.g., video or data, may
be described using statistical models.
T he investigation of the empirical
traffic traces can be the first step to build a suitable statistical model. This
is the subject for the next subsection.
3 .1 .1
V id e o Traffic
As an example of a video traffic source, the M PEG - 1 encoded trace of the
Jurassic Park movie is examined. In the MPEG-1 encoding for this trace,
the individual packets correspond to the d a ta in video frames produced at a
constant frequency of 25 frames per second [84]. The trace record consists
of the number of bits per frame in a sequence of 40,000 consecutive frames.
Fig.3.1 shows the bits per frame of the trace. A strong periodic component
can be seen in the trace which arises from the M PEG - 1 encoding format. In
39
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
this format groups of
12
frames are encoded into a fixed pattern (known as
the Group of Pictures (GOP)) of I-frames. P-frames and. B-frames according
to varying amounts of motion compensation between frames. For this trace,
I-frames carry the most d ata and occur every
12
frames.
As may be seen, the number of bits per frame is a time varying function.
Thus, the trace requires a variable channel capacity for transmission. On the
other hand, if a peak rate allocation system is used, the bandwidth has to be
reserved according to the maximum bit rate. For this trace, the maximum
frame size is 119632 bits, and for the frame rate 25 frames per second, a link
with 1.1 Mbps capacity will be required if peak rate is used as the allocation
criterion.
3.1.2
E th ern et Traffic
The d a ta trace of the Ethernet traffic source is also examined. These data
were collected on different Bellcore LAN networks in August’89 by Bellcore
researchers [85]. The first 500 seconds of this traffic is presented in Fig.3.2.
The trace contains the time stam ps and length of packets in bytes. As may
be seen, the required capacity is once again a time varying function with the
average lower than peak rate of 10 Mbps of Ethernet traffic.
3.2
E stim a tio n o f th e R eq u ired B a n d w id th o f B ro a d ­
b a n d Traffic
The above discussion of the video and E thernet d ata traces shows th at
the required bandwidth for the broadband network is time variable. It is also
clear th a t th e peak rate bandwidth allocation is not a bandwidth efficient
mechanism. An efficient bandw idth allocation policy may be designed if one
relates the required capacity to the desired quality of service. In this section,
a series of simulations is conducted to estim ate the required bandwidth of
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.5
-•-Ethernet Traffic
21.5
0.5
200
1 0 0
F ig u re 3.2
300
Time [sec]
400
500
600
The Ethernet traffic trace
Connections
Buffer
—
O
■7
F ig u re 3.3
Equivalent circuit for capacity estimation
the real traffic traces based on their desired QoS.
3.2.1
S im u la tio n M o d el
As mentioned earlier, the objective is to separate the ATM switch oper­
ation from the cell transmission system. A model for a non-blocking ATM
switch was discussed in Chapter 2 . As explained in Section 2.3, the switching
operation is done by an ATM fabric that is independent from the capacity
estimation section. The operation of the ATM switch fabric has been ex­
tensively investigated in the literature [69]. A model for cell transmission is
presented in Fig.3.3.
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3 .2.2
B a n d w id th E stim a tio n for E m p irical V id e o T raf­
fic
A basic m ultiplexer was simulated to estim ate the required bandwidth of
the video traces. It is assumed that all the d ata of the frame arrive back
to back at th e incoming link rate (Table 3.1), and th at they have been con­
verted into ATM cells with 48 bytes of frame data. No header was added
for sake of simplicity. Cells th at do not fit into the buffer at the time of
arrival are discarded. The only performance metric of interest is the cell
loss ratio (CLR) a t the multiplexer’s output buffer. The delay constraints
were not considered. The simulator is used to estim ate the required band­
width of an M PEG-1 encoded Jurassic Park movie. The total trace contains
40000 frames (approxim ately 27 minutes of video). The data trace consti­
tuted different fram e sizes in bits [84]. The results of the simulations are
presented in Fig.3.4. As may be seen, by increasing the buffer size, the re­
quired bandw idth decreases for a constant cell loss ratio. The simulation was
also repeated for the M PEG - 1 encoded Silence of the Lambs and Star Wars
movies with 40000 fram e traces. The results are shown in Fig.3.5 and Fig.3.6.
The next sim ulation was conducted to estim ate the required bandwidth of
the aggregate of th e three MPEG-1 encoded movies, namely Jurassic Park,
Silence of the Lambs, and Star IFars.The result is shown in Fig.3.7. A phase
shift 160 m se c was selected between the videotraces to make sure th at the I
frames are not all in synchronization with each other. As may be seen, the
required bandw idth of the aggregate of the movies is less than the required
bandwidth of th e superposition of the individual traces. This dem onstrates
the statistical m ultiplexing gain for the aggregate traffic.
3.2.3
B a n d w id th E stim a tio n for E th ern et Traffic
A sim ilar sim ulation was also conducted to estim ate the required band­
width of th e E th ern et d ata trace. The trace of Ethernet LANs measured by
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
*
+
*
Buffer Size=50KBit
Buffer 3ize=100KBit
Buffer Size=200KBit
,-z
,-3
Jurassic;Park;
g 10'
10 '
200
F ig u re 3 .4
300
400
500
EfFective Bandwidth [Kbps]
600
700
Sim ulated results for the required bandwidth
of a VBR videotrace Jurassic Park movie with
buffer size as a parameter
+
*
*
Buffer Size=50KBit
Buffer Size=100KBit
Buffer Size=200KBit
The Silence of the :Laxnfas;=;i
10
g 10
10
200
F ig u re 3.5
300
400
500
Effective Bandwidth[Kbps)
600
700
Sim ulated results for the required bandwidth of
a V BR videotrace Silence of the Lambs movie
w ith buffer size as a param eter
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
10
10
+
*
»
-1
'
'
Buffer Size=50KBit
Buffer Size=100KBit
Buffer Size=200KBit
sS ta p W a rs
2 10'
10
,-s
10
'200
F ig u re 3.6
300
400
500
Effective BandwidthfKbpsl
600
700
Simulated results for the required bandwidth of
a VBR videotrace Star Wars movie with buffer
size as a parameter
Bellcore in August 1989 was used for the simulation. The required bandwidth
as a function of the QoS for this trace is shown in Fig.3.8 with buffer size
as a param eter. The simulated results presented in this section show that
an optimum network requires a traffic bandwidth estim ation. An analytical
model which can be used for this estimation is developed in the next section.
3.3
B a n d w id th E stim a tio n U sin g S ta tis tic a l M od els
The ATM network is a connection oriented network meaning that a traf­
fic source requires to establish a VC before transmission. This is achieved
by first introducing the traffic parameters and the QoS parameters to the
network from the traffic sources. These parameters can be used to develop a
statistical model to estim ate the required bandwidth of the broadband traffic.
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
:
I . . : .' L . . .
Memory Size=200 KBit
Memory Size=500 KBit
:::::::
:
= = : ; ! : = = = ; ! = i l i i = i i = ! i : i i = H = ! ! I H i i i i = i i = ! ! n i : 5 ^ ! i ; i ! : i = E : : i ! i ! i i i ! i i i i : : ! ! == : ! = i i l l : : = n n n
srinn!niin=ii=nnni!H!ininnnnn!iiniini!inn;iii!nT4SSini;nnininHnnniiin
500
1500
1000
Effective BandwidchfKbps]
F ig u re 3.7
Simulated results for the required bandw idth of
aggregate VBR videotrace movie with buffer size
as a param eter (phase shift 160 m sec for each
source)
• 50KBit
+ 500KBit
» 1 MBit
fjBelkoresEtoen^b
-3
f 5-
1.4
F ig u re 3.8
1.6
1.7
1.8
Effective BandwidthlMbps]
Required bandwidth of Bellcore E thernet traffic
with buffer size as a param eter
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The required bandwidth of the broadband traffic is generally characterised
by an effective bandwidth. The effective bandwidth of a time varying traffic
source represents the minimum required bandwidth that it needs to meet QoS
guarantees . Assigning an effective bandw idth function to each traffic source,
which depends not only on its mean bandwidth but also on its burstiness,
gives a suitable tool to estim ate the required bandwidth of th a t source [5],[8 6 ].
3.4
E ffectiv e B a n d w id th u sin g th e F raction al B r o w n ­
ian M o tio n A p p ro a ch
In a series of papers [38],[39],[40] researchers from Bellcore have reported
the measurement results of real traffic sources. Their results show th a t the
different bursty traffic sources have a self-similar or long-range dependent
behaviour. For a long range dependent input process or self-similar process
the tail of the queue has the following large buffer asymptotic behaviour [40]:
-P[Q > II] ~ e-7n\
where II is the buffer size,
7
(3.1)
is a constant and e = 2 —2 H. is between zero
and one. H lies between .5 and 1 , and is the H urst param eter or self-similar
param eter. As compared to this, the large buffer asymptotic behaviour of the
queue tail in a conventional Markovian model with no long range dependent
behaviour is asym ptotically exponential [87]. This may be approxim ated by:
P[Q > n ] ~ e-T,n
where
77
(3.2)
is a positive constant. As m ay be seen, for H = .5 the two asym p­
totic relations (3.1) and (3.2) are identical. The estimation of the effective
bandwidth using exponentially distributed idle and burst models underes­
tim ates the cellblocking probabilities for real traffic sources[40],[86]. This
error is considerable even for m oderate long range dependence, e.g., H = .7.
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The estimation of the effective bandwidth of a self-similar traffic source is a
challenging problem.
It is known th at if traffic arrival could be considered as being a Gaussian
process, a Weibull distribution may be used to estim ate the buffer overflow
[44]. Based upon the fractional Brownian motion approach, a theory has been
recently proposed to deal with effective bandwidth of self-similar sources [45].
According to this theory the following equation holds for self-similar traffic:
= n
(3.3)
where the constant f 2 depends on iJ, the mean burst size, and the quality of
service but is independent of the utilisation and buffer size. An explicit re­
lation for the constant
is not straightforward. However, an approximation
of the upper bound may be used to estimate the effective bandwidth as [45]:
B W eff = R n ean“^~
(K (H )VC 5 ^ F ^ ) V * , 7V(2«)n -(i-ff)/H f l i/(M )i
(3.4)
where
K{ H) = H h {1 - H )1~h ,
(3.5)
and P c l r is the cell loss ratio, and a is the variance coefficient defined to
be the variance over the mean (bit-sec), H is the Hurst param eter of the
stream (a dimensionless measure of long range dependence between .5 and
1
), and n is the buffer size (in bits). This effective bandwidth equation is
used to estimate the required bandwidth of the Ethernet traffic. The effective
bandw idth as a function of buffer size and Hurst param eter for a utilisation
factor of “u = .7” and a CLR of 10- 6 is shown in Fig.3.9 (The utilisation
factor is defined as u = Rmean/Rpeak)- As may be seen, increasing the H
param eter generally increases the required bandwidth for a given quality of
service and buffer size [8 8 ]. The effective bandwidth is also examined as a
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
function of the H param eter and the utilisation for a CLR of IQ-6 . As shown
in Fig.3.10 increasing utilisation also increases the effective bandwidth for a
given quality of service and an H parameter.
T ab le 3.1
The Video traffic sources
Mean [Mbps]
Var.
C o f f .[ b - s ]
Hurst
J u r a s s ic Park
.326950
4 . 928e4
.8448
S ile n c e o f th e Lambs
.182788
6 .4 8 4 e 4
.8959
S ta r Wars
.232830
7 .2 5 0 e 4
.8458
A ggregate stream
.742568
6 . 039e4
.8959
3 .4 .1
C o m p a riso n o f R esu lts for V B R V id eo Traffic
The traffic param eters of a VBR entertainment video source (MPEG
encoded “Jurassic Park” movie) was estimated using this method and the
results are shown in Table 3.1. The Group of Pictures (GoP) format used
for the encoding is IBBPBBPBBPBB [84] and the frame rate is 25 frames
per second. The trace contains 40000 frames (approximately 27 minutes
of video).
For this movie, the Hurst parameter was estimated as being
H = .845. The effective bandw idth of the videotrace was estimated using
Eq.(3.4) and the results are shown in Fig.3.11. As may be seen, increas­
ing the buffer size decreases the required bandwidth for a given quality of
service. A comparison of Fig.3.4 and Fig.3.11 clearly establishes that the
effective bandwidth estim ation method offers a close upper bound for the
required bandwidth of this videotrace. This comparison is also repeated for
the Silence of the Lambs and Star Wars movies in Figures 3.12 and 3.13. As
may be seen, the effective bandw idth estimation using Eq.(3.4) again offers
a suitable upper bound estim ation for the required bandwidth. On the other
hand, Fig.3.14 compares the effective bandwidth with the peak rate allo­
cation. As may be seen, using the effective bandwidth can efficiently save
bandwidth.
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
S ett-Sim ilar E thernet S o u rce
M em ory SIze[K Bytes]
F ig u re 3.9
Effective bandwidth of E thernet traffic source
with U =.7 and C L R — 10"
6
Self-Similar Ethernet Source
Utilization Factor
0.S
Hurst Parameter
F ig u re 3 .1 0 Effective bandwidth of Ethernet traffic source
with buffer size 20 KBytes and C L R = 10~ 6
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In the next step, video traffic consisting of an aggregate of three M PEG
videotraces (Jurassic Park, Silence of the Lambs, and Star Wars ) was ex­
am ined. Each trace contains 40000 frames (approximately 27 minutes of
video). Their individual and aggregate traffic characteristics are shown in
Table 3.1. The required bandw idth based on ATM switch sim ulation for ag­
gregate traffic using 200 and 500 K bits shared buffer is shown in Fig.3.7. The
aggregate traffic is also examined using Eq.(3.4). The Hurst param eter of the
aggregate traffic is assumed to be the highest H param eter of the individual
videotraces [8 8 ]. The effective bandw idth of the aggregate of the videotraces
as calculated by Eq.(3.4), is shown in Fig. 3.15. As may be seen, the effective
bandw idth again offers a close upper bound on the required bandw idth of
real aggregate video traffic sources.
3 .4 .2
C o m p arison o f R e s u lts for E th ern et Traffic
Eq.(3.4) is also examined to estim ate the required bandw idth of Ethernet
d a ta traces collected by the Bellcore researchers. The mean rate for this
d a ta trace was estim ated to be m = 1.3620M6ps, the variance coefficient
a = 264410 bit-sec, and the H urst param eter H =
.8
[85]. The results are
shown in Fig.3.16. The required bandw idth of this d ata trace was already es­
tim ated using an ATM m ultiplexer simulation and the results were presented
in Fig.3.8. As may be seen from Figures 3.8 and 3.16, Eq.(3.4) offers an accu­
rate estim ate of the required bandw idth of these video traces. However, this
estim ate is not very close for th e d a ta traces. Indeed, the fractional Brow­
nian approach offers a b etter approxim ation when used in a highly utilised
network using large buffer size [45],[8 8 ]. The utilisation factor of this d a ta
trace is about 14 percent.
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
trnTTT7TTnTTTTTTT!TT
x
•
+
O
1 0 '1
B=50 KBit
B=100 KBit
B=200 KBit
B=500KBit
B=1 MBit
«
J
10
” 10 -
1 0 -5
10"6
200
300
400
500
Effective Bandwidth [Kbps]
600
700
F ig u r e 3.11 Effective bandwidth of a V BR videotrace Juras­
sic Park movie using Eq.(3.4) (Buffer size as a
parameter)
10 °
................................. --- x
*
+
10"1
B=50 KBit
x B=100 KBit
* B=200 KBit
+ B=500KBit
10"2
•2
£03
I
-3
•:;;t”•*j*::::::: I*:::
10
10“ * i i i i i i i i l i i i l i l i i H i l i i i n H i i i i i M i i i i i l i i i i h i i i i i H f x i
1 0 '5
10
^200
300
400
500
Effective Bandwidth [Kbps]
600
700
F ig u r e 3.12 Effective bandwidth of a V BR videotrace Silence
of the Lambs using Eq.(3.4) (Buffer size as a pa­
rameter)
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10'
*
*
+
o
10
10
B=50 KBit
B=100 KBit
B=200 KBit
B=500KBit
B=1 MBit
'
2 10
10'
10
'
200
300
400
600
700
500
Effective B andw idth [Kbps]
800
900
F ig u re 3.13 Effective bandwidth of a VBR videotrace Star
Wars using Eq.(3.4) (Buffer size as a parameter)
1200
1100
1000
c.
900
Q-.-
■e— >
S 800
m 700
600
500
•------ •
o-------o
x------- x
«------- •
o-------o
»------- ■<
Rpeakl (Jurassic Park)
Rpeak2(Silence)
Rpeak3(StarWars)
BWeff(Jurassic Park)
BWeff2(Silence)
BWeff3(StarWars)
400
10
10
'
10
Buffer Size [KBits]
F ig u re 3.14 A comparison between the effective bandwidth
and the peak rate for different videotraces
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
:::::::::::: :: I
rTTTT;
*
+
o
600
800
B=200 KBit
B=500 KBit
+ B=1MBit
o B=5MBit
1000
1200
1400
Effective Bandwidth [Kbps]
1600
1800
F ig u re 3.15 Effective bandwidth of the aggregate V B R video
traffic using Eq.(3.4) (Buffer size as a parame­
ter).
•
°
°
Buffer Size=500 KBit
Buffer Size=1 MBit
ifj i i t l l I I : I f : =
:::::: i:1:
: : : : : : : : : : : : : : : : : : : : H:ii:!i=!i=;M :n : ;^
;
2
2.5
3
3.5
Effective Bandwidth [Mbps]
F ig u re 3.16 Effective bandwidth of the Bellcore E thernet
d ata traces using Eq.(3.4)
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Buffer(stze=B)
Source
Source 1
Source 2
End '
. N'ode;
Source
Switch
(size=B)
Sink
Source 1
n
BufferTsize=B)
Source 2
Switch
aJ(size=B)
Source 3
Source 3
Source N
Sink
V\'od.
Source N
F ig u re 3.17 a) Simulation model of an Ethernet source in
ATM sim ulator b) Simulation model for m ultiple
E thernet sources.
3 .4 .3
E x p e r im e n ta l E stim a tio n o f th e E ffectiv e B a n d ­
w id th
An experimental estim ation of the effective bandwidth can be done using
com puter simulation. The ATM-TN simulator was used to obtain a relation
between the desired quality of service and the effective bandwidth [41],[89].
The simulation model is shown in Fig.3.17.
N e tw o rk M o d e l of a S in g le E th e r n e t S ource in A T M -T N
As shown in Fig.3.17.a, this network consists of a source, two end-nodes,
and a per-port switch. T he traffic source is a self-similar E thernet source.
T he source is connected to a buffer of size B. The objective is to study the
quality of service based on th e cell loss ratio. As the source is a LAN source
which is insensitive to delay, the delay variation is not considered here.
N e tw o rk M o d e l o f M u ltip le E th e r n e t S o u rces in A T M -T N
The estim ation of the effective bandwidth of multiple E thernet sources
has been done using the topology shown in Fig.3.17.b. The network consists
of 1 0 E thernet sources. The sources are characterised by utilisation factors as
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ethernet seff-sfrnilar Source(U=flO%)
'
8 10
8
10 '
6.2
8.4
6.S
6.8
9
9 .2
9.4
9 .6
9.8
Effective Bandwidth[Mbps]
F ig u re 3.18 The experimental results for the effective band­
width of a self-similar source with 80 percent
utilisation with three H values.
well as Hurst parameters. This network has 11 end nodes. All of the sources
are connected to a shared buffer of size B. The capacity of the connected
links of the multiplexed sources is 10Mbps. The capacity of the output link
was a variable in the experiment.
E x p e r im e n t R e s u lts
The results of the experiments for a single Ethernet source are presented
in Figures 3.18, 3.19, and 3.20. In Fig.3.18 the quality of service for a self­
sim ilar Ethernet source with utilisation of 80 percent is shown for different
values of the H urst parameter.
It may be seen th at increasing the self­
sim ilarity ( higher H) increases the required bandw idth for the same quality of
service. This behaviour was also predicted by the analytical model. Fig.3.19
shows the effective bandwidth for different utilisation with the same Hurst
param eter H = .8 .
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0
1
6.5
5.5
7
7.5
8
8.5
9.5
Effective BandwidthfMbps]
F ig u re 3.19 The experimental results for the effective band­
width of a self-similar source with different util­
isation for H = . 8 .
1 0 self-sim ilar Ethernet sources(U=80%)
10 '
ir
o 10 _3
H=.7
10
“*
10
““80
82
84
86
88
90
92
94
96
98
100
Effective Bandwidth[Mbit]
F ig u re 3.20 The effective bandwidth of 10 self-similar Ether­
net source with U = . 8 .
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Fig.3.20, illustrates the effective bandw idth of multiple sources. In this
experiment, utilisation of all sources were considered to be 80 percent. The H
param eter for different sources is similar. The buffer size of shared memory
is 100 cells and the sim ulation time is fixed at 10 seconds. Again the effect of
H param eter variation was studied in homogeneous sources. Increasing the
H param eter of sources increases the required bandwidth, provided th a t the
quality of service is kept constant. The ATM-TN simulator was also used to
estimate the effective bandw idth of the self-similar traffic sources [8 8 ].
3.5
M o d ified E ffe ctiv e B a n d w id th for R ad io A T M
The cell loss resulting from buffer overflow is independent of the cell
loss resulting from channel bit errors. The effect of cell loss due to a non­
ideal physical layer such as the radio channel is an increase in the required
bandwidth. Thus, a modified effective bandw idth function has to be defined.
The total CLR (C L R T) is related to the CLR due to buffer overflow (C L R a)
and the CLR due to the non-ideal physical channel (C L R c). Thus, the total
C L R t can be estim ated as follows:
C LR t —CLR0 +
(1
— C L R 0)C L R c.
(3.6)
Using Eq.(3.7), the CLR due to buffer overflow can be expressed as
CLR
CLRt ~ CLR‘
CLR° ~
1 -C L R ,
'
((3’7)
Fig.3.21 shows th a t the desired C L R t of the network is equivalent to the
C L R 0 for the random error channel. However, for a burst error channel, the
tolerable C L R a decreases as a function of the BER and the burst length.
Using this relation, the estim ation of the effective bandwidth for a non-ideal
channel is straightforw ard.
The required modification is to use Eq.(3.6) in
Eq.(3.4), where C L R t is the desired CLR for a guaranteed quality of service.
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
*tz
8 7
O
©
3 6
m
*
Burst Channel(Beta=2,7.12.17)
Random C hannel
o
® 5
"3O
CC
o 4
jp
XI
Beta=2'
S 3
-S
o
I—
2
1
10
’
-10
10
.-a
’
•9
10 ’
10'
.7
10
Bit Error Rate
’
F ig u re 3.21 Tolerable CLR due to buffer overflow as function
of BER for Q oS = 10- 5 as a function of burst
size
S elf-S im ilar E th e m e r S o u rce (U=50% & H =.7 p = .0 0 0 0 l)
10* '
-x—R an d o m
C h an n e l
—o — B u rst C hannel(B eta=2)
ua>
u>
>
5
C
O
o
c3
a
52.
5.4
5 .6
5.8
Effective Bandwidth [Mbps]
6.2
6.4-
F ig u re 3.22 Effective bandw idth for random error and bursty
channel for utilisation 50 percent and H=.7.
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The modified effective bandwidth for an E thernet source with U = .5, H =
.7, B E R = 10-5, and burst length of two bits is shown in Fig.3.22. It should
be pointed out th a t the burst error effect can be effeciently resolved using an
interleaving technique. However, the extra delay due to interleaving should
be considered when an end-to-end delay bound is studied.
3.6
S u m m a ry
In this chapter it was established th a t an optim um ATM network requires
a mechanism for estim ation of the traffic bandw idth. An analytical model as
well as com puter sim ulation results for effective bandwidth estimation were
presented for various types of traffic. The effective bandwidth procedure was
modified to accomm odate the bursty behaviour of the radio channel. In the
next chapter another im portant issue, namely, the adaptive m odulator for
the radio ATM network is addressed.
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.
A D A P T IV E M -QAM FOR R A D IO ATM
IN M ULTIPATH FADING C H A N N ELS
In a broadband wireless network, an optimum tran sm itter has to take into
account both radio channel and broadband traffic characteristics. Newer ser­
vices are being continuously introduced and the comm unications network is
becoming more dynamic and complex. These services require various bandwidths and QoS. As discussed in the last chapter, the QoS requirements can
be described by an effective bandwidth metric and different traffic calls have
different effective bandw idth requirements.
On the other hand, in a radio transmission system , m ultipath fading
results in the reception of a multitude of reflected and delayed signal compo­
nents [90],[91]. Although the performance degradation due to the multipath
fading is more severe for the mobile network, it affects the performance of
fixed radio links as well. It is well known th a t m ultipath fading in mobile
communications has a Rayleigh distribution, and its distribution in the line
of sight link is Ricean [82],[92]. In a fixed channel, the simplest method to
overcome the fading effect is to make the transm itter power large enough, or
the bit rate low enough so as to get a satisfactory error probability during a
specific fraction of the time. Although this is simple, these methods result
in low efficiency [90].
For an optim um broadband radio system, a more sophisticated design is
clearly required. Overall, the system should provide an adaptive capacity
to maximise the available spectrum usage while m aintaining the radio link
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
performance. Further, this system must be implemented in a cost effective
manner.
In this chapter, AWGN channel, and Ricean and Rayleigh fading channels
are examined. The results outlined in this chapter are applicable to severe
fading scenarios, e.g., a short haul link in a crowded city (LMDS applications)
or a land mobile radio. It is shown th at a dynamic bandw idth demand system
can be realized by using an adaptive transm itter. Finally, a multilevel QAM
m odulator is introduced to implement the adaptive transm itter.
4 .1
R e q u ir e m e n ts for an A d a p tiv e T ra n sm itte r
From the previous discussion it is clear th at a broadband wireless ATM
system th a t is designed to carry multimedia traffic has to adapt to the QoS
requirements as well as to the radio channel variations [69],[93]. The following
subsections address the capacity variation due to QoS and radio channel
performance.
4 .1 .1
C a p a c ity V ariation D u e to Q oS
The number of users in an ATM network is tim e varying and these users
also have various QoS requirements which may be specified by the cell loss
ratio and delay constraints. The effective bandw idth as a metric of QoS was
derived in the last chapter. Eq.(3.4) is rewritten below:
BW e/ / = Rmean~>~
(K(H)^/-2log(j>cLR))l/H<Tm2H)U - ^ - H)/HR i l Z P .
(4-1)
K {H ) = H h {1 - i? )1_jEf,
(4.2)
where
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and cr is the variance coefficient.
The above equation has to be further
modified to take into account the delay lim itations of ATM sources. If the
maximum tolerable delay is Dmax, the relation between maximum delay and
buffer size may be written as:
B W eff
(
^ Dmax-
(4-3)
Using (4.3), Eq.(4.1) can be modified as:
B W ef f = Rmean~\(K (H )sj-2log{jpCLR))llH^
!r>(BW ,„Dmax)-'-1- ' r>li'R)llg « ) .
(4.4)
This equation can be rearranged to collect the effective bandwidth terms at
the left hand side as:
B W « //(1 —Rmeon/BWef f ) =
( K ( H ) J-2 lo gtp cL R ))
■
(4.5)
As may be seen, this is a nonlinear equation th a t relates the effective band­
width to the required cell loss ratio,
Pc
lr
-,
and maximum delay, Dmax. This
equation has to be solved numerically to obtain the effective bandwidth value
for a given cell loss ratio and maximum delay.
As an example, for an ATM traffic source with the Rmean = 2 Mbps,
a = 104 bit-sec, and H = .7, the effective bandw idth as a function of p c l r
and Dmax was computed using Eq.(4.5) and results are presented in Fig.4.1.
As majr be seen, the required effective bandwidth is a function of cell loss
requirement and delay constraint. This suggests that i f optimum transmission
of multimedia traffic is desirable, a flexible transmitter design is essential.
62
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a.6
20
-5
log(CLR)
F ig u re 4.1
10
-1 0
0
M axim um Delay[msec]
Effective Bandwidth as a function of
Pc l r
and
DmnT for Rmean = 2 Vlbps, o’ —
—10 , and B — .7.
4.1.2
C a p a city R e d u c tio n D u e to C h a n n el F ading
A model has been introduced for an ATM transmission system in Chapter
2. As already discussed in Section (2.11), the effective bandwidth of a high
speed link, C , has the following relation with the effective bandwidth of
individual VPs.
E B W sffi < C,
(4.6)
1=1
where N is the number of VPs and B W ejji is the effective bandwidth of the
ith VP.
The available capacity in a radio ATM link is a function of channel per­
formance as well. Toestim ate the capacity reduction
due to fading, the
Shannon capacity is examined first. The well known Shannon limit for an
AWGN channel is given by [80],[94]:
Cw = Wlog2(l + 7 )>
63
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(4-7)
where W is channel bandwidth and
7
is signal to noise ratio. For a fading
channel, the Shannon limit can be modified as [95]
C f = [ Wlog2{l +
(4.8)
where p( 7 ) is the probability density function of the received CNR. Thus,
the Shannon lim it for a fading channel decreases by a factor:
rc = £f_ = I y lo90- +7)p(7)<*7
Cw
log2( 1 + 7 )
(4.9)
where <fc is called the capacity reduction factor .
For LOS radios the fading distribution is Ricean. For a Rice channel, the
probability density function of the received signal to noise ratio is given by
[81]:
P(7) = i ± r ^ i p ( - A ' - ^ (1 + K ) ) h ( 2 j l ( K 2 + K )),
(4.10)
where T is th e average power of 7 and K is the Rice param eter (K —>0 for
a weak direct wave and K —> 0 0 for a strong direct wave [96]). I0 is the zero
order modified Bessel function.
For Rayleigh fading, the probability density function of the received
C N R , p( 7 ), is given by:
(4.11)
where T is th e average power of 7 . The capacity reduction factor due to Rice
and Rayleigh fading channel is computed and the results are presented in
Fig.4.2. An approxim ate analytical model for a Rayleigh channel has been
used in [70],[95]. T he capacity reduction factor using this approximation is
also included to show a comparison with exact results.
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CQ
0.8
Ricean
Rayleigh
Approximate for Rayleigh[70],[95]
0.75
F ig u re 4 .2
15
20
Average CNR[dB]
30
Shannon capacity reduction factor due to chan­
nel fading for Ricean { K = 5) and Rayleigh
channels
For a fading channel, the available capacity in an ATM link using Eq.(4.6)
and Eq.(4.9) can be modified as:
J ^ B W e f f i < e C w.
(4 .1 2 )
1=1
A dynamic bandwidth demand could be accommodated by a variable capacity
transmitter as long as Eq.(4-12) is satisfied.
4.2
A d a p tiv e M -Q A M M o d u la to r
According to Eq.(4.12), a variable capacity tran sm itter is required to real­
ize an optimum radio ATM transm itter. An M-QAM m odulator can provide
high spectral efficiency and is suitable for im plem entation as an adaptive
transm itter for radio ATM. The following subsections describe the perfor­
mance of M-QAM as an adaptive m odulator in the wireless channel.
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4 .2 .1
M -Q A M in A W G N C h an n el
According to Eq.(2.26), the spectral efficiency of the M-QAM m odulator
using a raised cosine filter with roll-off factor a is given by:
C
IoqiM
where M is the m odulation level, C is the channel capacity, and W is the
channel bandwidth. Thus, the number of possible VP connections in an
AWGN channel for an M-QAM modulator is given by:
( 4 . 14 )
This shows th a t the modulation level can be selected using the following
criterion:
lo g iM
>
(4 .15)
As a typical example for calculation of the constellation size, M-QAM is
studied in an AWGN channel using a raised cosine filter with a roll-off factor
a = .5 with a 40 MHz channel bandwidth for a number of similar video­
traces, namely, Jurassic Park with B W eJ-f = 824.23 Kbps, and Silence of
the Lambs with B W e/ f = 626.96 Kbps. A buffer size of 200 KBits, and cell
loss ratio
10
" 6, are used. The results of this calculation are shown in Fig.4.3.
4 .2 .2
M -Q A M in M u ltip a th F ad in g C h an n el
In the previous section, a capacity reduction factor was derived for the
Shannon limit. This factor shows th a t the capacity is reduced in the presence
of fading. Using a bit error rate upper bound for the M-QAM modulation,
the capacity reduction factor can also be computed for M-QAM modulator.
A bit error rate upper bound for the M-QAM m odulator has been derived in
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o
55
oa
o
eo
U
Jurrasic Park (B=200Kbits)
Silence of the Lambs(B=200Kbits)
50
100
150
200
250
300
Number of MPEG-1 Video Connections
F ig u re 4.3
The constellation size based on the number of
identical videotraces
[74]. The bit error rate for an AWGN channel for the M-QAM modulation
can be bounded by:
37
2 (M -
Pb — . 2 exp{-
1
)
(4.16)
This upper bound has been compared with the exact results in Fig.4.4 and
as may be seen from this figure, the bound is quite tight. From Eq.(4.16),
the modulation level is obtained as follows:
M ~
1
—
37
2log(5pb) '
(4.17)
By substitution for M in Eq.(4.13) the capacity reduction factor in the mul­
tipath channel for M-QAM modulator is given by:
_ fy lo g ijl ~ 37[2foff(5p6)] 1)p(j)d''f
Sm q a m ~
log2{l _ 3 7 [2 ^ ( 5 p 6) ] - 1)
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(4.18)
4QAM E xact
4QAM U p p er b o u n d
16QAM E x act
I6QAM U p p er b o u n d
64QAM Exact
16QAM U p p er b o u n d
256QAM E x act
2S6QAM U p p er b o u n d
20
25
CNR[dBI
F ig u r e 4 .4
The bit error rate of M-QAM m odulator(Exact
and upper bound)
A closed form relation for Rayleigh fading is derived in Appendix A and is
given by:
-e x p (-^ g E fe l)(g +
Zm
q a m
~
+
L n (l —3 7 [2 /n ( 5 p 6 )]-1)
(4.19)
where E=.577215 is the Euler constant. The computed results for the ca­
pacity reduction factor for the Rayleigh and Ricean channels are presented
in Fig.4.5 for different values of BER. Using this factor for bandw idth allo­
cation, Eq.(4.15) can be modified as:
N
v-'
mv
^ €m q a m W I°92M
X . B W . r / i < --------
(4.20)
This leads to the following criterion for the constellation size selection:
log2M >
(1 + a) S£i BWejfj
MQAM
68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(4.21)
0.98
Ricean Channel
0.96.
— Rayleigh-Channel
0 0.88
-*
-o
-+
-«
-fl
-£>
-A
-*
§.0.86
0.841
0.82
0.8
20
F ig u r e 4.5
22
24
28
30
32
Average CNRtdBl
CNR[dB!
BER=1 e -2 Numerical
BER=1e-4 Numerical
BER=1 e -6 Numerical
BER=1 e -8 Numerical
BER=1 e -2 Analytical
BER=1e-4 Analytical
BER=1 e -6 Analytical
BER=1 e -8 Analytical
34
36
40
Capacity decrease due to fading in MQAM (K=5
in Rice channel)
The required M-QAM constellation size in AWGN and Rayleigh fading
channels using J u ra ss ic P ark movie videotraces for a cell loss ratio 10- 6
with a raised cosine filter with roll-off factor a = .5 with 40 MHz channel
bandw idth is presented in Fig.4.6.
As expected, the possible number of
connections for a specific level of M in a fading channel is less than th at for
an AWGN channel. It may also be seen that accepted connections increase
when the average of the received CNR is increased. As shown in Fig.4.5,
the capacity reduction due to multipath fading is a function of the required
BER. The required constellation size for similar videotrace connections, for
different values of B ER with a CNR=30 dB and Rayleigh fading is presented
in Fig.4.7.
69
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
AWGN
Fading <CN R>=20[dB]
Fading <CNR>=40[dB]
Ol
0 6
O
I 5
g4
©
U
>1
1 3
O'
50
100
150
200
250
Number of Connections
F ig u re 4.6
The constellation size plotted as a function of the
number of connections of Jurassic P a rk movie
for a C L R = 10~ 6 and Rayleigh fading channel
4 .2 .3
M -Q A M for A T M C ell T ran sm ission in M u ltip a th F ad ing
Using Eq.(2.13), an approximate relation between B E R and CLR for a
random channel is given by:
Pi - ^
where prCLR *s
<4.22)
ce^ i°ss ratio due to random error channel.
Using (4.22) in (4.18), the capacity reduction factor for a random error
channel is obtained as:
^m q am ~
fy log2p- - 3 7 [% (p £ LjR/31.2)]
log2(1 _ 37 [ W 0 W 3 1 .2 ) ] - 1 )
(4.23)
where £mq AM is the capacity reduction factor for the M-QAM in a random
channel. A sim ilar relation can be derived for a burst error channel. Using
Eq.(2.16), the bit error rate and cell loss ratio relation in a burst channel is
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8
AWGN
Fading (B E R = 1 e-2 )
Fading (B E R = 1 e-6 )
tWe
C
•J5
CB
'3
-u .
a4
o
U
1 3
O'
50
F ig u r e 4 .7
100
150
Number of Connections
200
250
The constellation size vs. num ber of connections
of Jurassic P ark movie for CN R=30 dB with a
C L R = 10~ 6 ratio for a Rayleigh fading channel
given by:
Pb
where
PPCLR
40 :
(4.24)
is the burst length and Pqcr is the cell loss ratio due to burst channel.
Substitution of Eq.(4.24) in Eq.(4.23) gives the capacity reduction factor for
the burst channel as:
cb
9m
q a m
f y 1092(1 - Z ' y j i o g W P c L R / ^ ) ]
lM 7)<*7
h g 2 { 1 _ 3 7 [ ^ ( / 5 p ^ L K / 4 0 ] - 1)
(4.25)
where Zm q a m *s ^ e capacity reduction factor for th e M-QAM in a burst
channel.
T he capacity reduction factor for a given cell loss ratio in the
random and burst error channels are presented in Fig.4.8 and 4.9 for Ricean
and Rayleigh channels.
71
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
K icean
£0.95&
o
CLR=1e-3
CLR=1e-6
CLR=1e-9
CLR=1e-12
*3 0.8
30
A verage CNR[dB]
F ig u re 4 .8
Capacity reduction factor in a random error
channel due to fading (K=5 for Ricean Chan­
nel)
Kicean
§ 0.9
Rayleigh
«
-
•3
0.85-,
es
a.
;
B u rst Cbahnel(P=2)
25
30
CLR=1 e -3
CLR=1e-6
CLR=1 e -9
CLR=1 e - 1 2
35
40
A verage CNR[dB]
F ig u re 4 .9
Capacity reduction factor in a burst error chan­
nel due to fading
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.3
A d a p tiv e M -Q A M M o d u la to r for R a d io A T M A p ­
p lic a tio n s
As shown in previous sections, the assigned bandw idth is a function of
the cell loss ratio. As explained in Section 2.4, the cell loss ratio may result
from cell blocking or radio channel impairment. In the following section, a
multimedia radio m odulator is designed according to these characteristics.
4 .3 .1
M o d ifie d E ffe c tiv e B a n d w id th for M -Q A M
A general relation between the transmission capacity and the number of
the perm itted connections may be w ritten as:
Y , B W effi(pcLR) ^ C,
(4.26)
€ m q a m (P c l r ) *■=1
where Pclr 1S
ceU loss ratio due to the channel and Pqlr Is the cell loss
ratio due to cell blocking. A new metric is needed for use in bandwidth
allocation in radio ATM. This new metric, termed as the modified effective
bandwidth, has been originally defined as:
Y
t= l
BWmeffi(jPcLRi PcLR)
£ C,
(4-27)
where
B W meffi{ p c LR-. Pc l r ) = T~c
7~c
\
$ M Q A M \P C L R J
i=i
(4.28)
is the modified effective bandw idth of each traffic connection. As an example,
the modified effective bandw idth for an Ethernet source with Rmean — 5
Mbps, H = .7, and a = 1 Mb-s is illustrated in Fig.4.10. The buffer size
is 10 Kbits and channel b it error rate is 10-6. As may be seen, the highest
bandw idth is required for a Rayleigh channel. The required bandwidth in
a Ricean channel is a function of Rice parameter, K , and lies between the
73
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ethem er Source (Rmean=5 Mbps & H=.7 & p= 1e-6 & sigma=1 M bit-s & B=10 Kbits)
10 '
■«
■o
-S>
■a
10
AWGN (Random error)
AWGN (Beta=2)
Ricean (Beta=2.K=5,CNR=25 dB. and QAM Modulation)
Rayleigh Channel(Beta=2,CNR=25 dB and QAM Modulation)
'
5.2
5.4
5.6
6.4
5.8
6
6.2
Effective Bandwidth[Mbps]
6.6
6.8
F ig u re 4.10 The modified effective bandwidth for different
channels
level selection for the M-QAM can be obtained using Eq. (4.14):
log2M >
(1
+ °°
_
(4
, 9)
where B W mef /i is the modified effective bandwidth of VP{.
4 .3 .2
A d a p tiv e M -Q A M A rch itectu re
Based on the above theoretical results, an architecture for a radio ATM
system using an M-QAM m odulator is proposed in Fig.4.11. The function of
the adaptive M-QAM control unit is to select an optimum modulation level.
The adaptation is based on an ATM call admission task and wireless channel
condition. The demodulator can be informed about the level of modulation
by different techniques [97],[98]. The control unit for an adaptive M-QAM
control unit is presented in Fig.4.11. The number of virtual paths is N and re­
connections are demanding to use V P i, i = 1,..., AT at time tj. The operation
74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Wireless
Channel
- MQAMr'T
Demodulator
Feedback
Channel
F ig u re 4.11 A model for a fixed wireless ATM system
of the control unit is described as follows:
• Different connections, to be transm itted using V P t- i = l,...,vV, in­
troduce their traffic and QoS param eters (e.g., tolerable cell loss ratio
Pclr
and maximum delay Dmax) to the connection gate control block.
• The effective bandwidth of V P s are evaluated by using the traffic pa­
rameters of the new connection and information from the buffer size
manager.
• A buffer size is selected for each VP according to minimum tolerable
delay. Eq.(4.5) is solved to find the effective bandwidth. The buffer size
is selected according to the delay constraints of each VP using Eq.(4.3).
• This is followed by a modified effective bandwidth estimation of the
individual VPs. To do this, the capacity reduction factor, £m q a m , is
computed according to Eq.(4.19).
• The total value of the modified effective bandw idth is computed accord­
ing to the new values of the modified effective bandw idth of individual
V P s , respectively.
• The m odulator capacity is checked with the instantaneous bandwidth
demand, and the modulation level is controlled according to the instant
bandwidth demand.
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Capacity Request
Yes
VCs
No
M ay MQAMleveT
i^be mcreased7T:;
Wireless
S ^ t ii n a tq & |
F ig u re 4.12 MQAM m odulator for radio ATM using one VP
• The outage estim ator examines th e outage percentage according to the
new m odulation level. The outage estim ation is to be computed using
a link design program which is initialised according to the specific radio
link.
• The k ( M = 2k) is increased so long as the desired system outage is
achieved.
• A positive acknowledgement from the outage estimator, implies th at
the gate control will perm it the new connections to use the V P t and
radio link.
As may be seen, the M-QAM control u n it is designed to operate in harmony
with a non-blocking ATM switch. The performance of this unit has been
studied in C hapter 7.
76
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Channel;
Connections(VCs)
Do not accept the call
Accept the Call
Delay Parameters
-BtiffefSize.
VP
Effective Bandwidth
H^axtHp . R«al=5:Rnl
VP. Effectiv e Bandwidth
VP
X r an j•Cl­
i “
^m ean j ffj
Effective Bandwidth
H=ma.x(Hj) . R ^ =X Rm
^•RmeinjCTj
H=max (Hj) . F'mean “ ^ ^mean
R rtxan
rtvtan
'i
IIS
&23&
m m
S^siftSW
a m
m
TI
I
I
W inefE
m m w vz
No
j (M)
-‘'A1 ' ■:
t
6 « .,
No
MQAM Control
F ig u re 4.13 The M-QAM control unit
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4 .4
Sum m ary
To optimise the ATM radio transm itter performance, an adaptive MQAM modulator has been proposed and analysed in different wireless chan­
nels. An M-QAM control unit was introduced to select an optimum modu­
lation level. An implementation method for such a unit was presented. The
hardware architecture for a direct microwave M-QAM modulator is described
in the next chapter.
78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.
D IR E C T MICROWAVE M -Q AM
M O D U LA TO R H A R D W A R E
An adaptive transm itter for radio ATM applications using M-QAM was
introduced in the last chapter. The proposed m odulator could be used for a
system with high bit rate multimedia traffic. The cost efficiencies of the m od­
ulator is another im portant design consideration. The conventional m ethod
of QAM m odulation is not efficient in power, size, and cost. In addition to
the IF sections, such a method also requires a transm itter power amplifier
which may include a linearizer circuit. This could have an impact on the
cost of the radio. Direct microwave modulation at transmission frequency
offers an inexpensive, efficient solution which leads to a compact circuit th a t
is suitable for MMIC implementation.
In this chapter a novel structure is proposed for direct microwave QAM
m odulator im plem entation without a power amplifier. The proposed design
uses high speed attenuators implemented with PIN diodes. A new analytical
model to analyse a forward biased PIN diode is presented in Section 5.2.
This is followed by a detailed examination of the direct M-QAM m odulator
performance.
5.1
M o d u la to r A rc h itectu re
A new architecture of the direct microwave QAM m odulator is presented
in Fig.5.1. As shown, the power oscillator output is divided equally between
the inphase and the quadrature paths using a quadrature hybrid. A 180
degree hybrid in each path is used to modulate the oscillator signal with
79
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Data
Data
Modulation
Control
Splitter
_ _I_,
Threshold
Detector
Power Osc.
Ring
SPDT
Coupler
Switch
Variable
Attenuator
Variable
Phase
Shifter
Lange
Power
Combiner
Coupler
Ring
SPDT
Coupler
Switch
Variable
Attenuator
Threshold
Phase
Shifter
Predistortion
& Prefiltering
Detector
F ig u r e 5.1
Variable
Direct Microwave QAM
the sign of the baseband signals. The positive sign in the baseband signal
selects the inphase o utput of the 180 degree hybrid and the negative sign in
the baseband signal selects the out-of-phase o u tput of the 180 degree hybrid.
The rectified baseband signal in the inphase path is used to adjust the level of
an attenuator in th e inphase path. Similarly, the rectified baseband signal in
the quadrature p ath is used to adjust the attenuator level in the quadrature
path.
Pulse-shaping is realized using a raised cosine filter in each path.
Using this m odulator, the filtering can be done on the baseband signal before
applying the baseband signals to the variable attenuators. Moreover, due to
the nonlinear relation between the control voltage and the attenuation for the
PIN diode attenuators, baseband pre-distortion becomes essential. The pre­
distortion and pre-filtering can be implemented by using the same memory
circuit [10]. An inphase power combiner is used to combine the inphase and
quadrature signals.
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.2
T h e o retic a l M o d e l for D irect Q A M M o d u la to r
The main block in the proposed direct modulator is a high speed variable
attenuator. A common m ethod to realize a variable attenuator is to use PIN
diodes. The PIN diode has traditionally been used for high power microwave
switching applications as well as a variable attenuator. However, to use a
PIN diode circuit as a high speed modulator certain design issues have to be
resolved. The main problem is the hysteresis effect when a PIN diode is used
in the ON-OFF mode [6 6 ]. In this mode of operation, when the diode condi­
tion changes from forward bias to reverse bias, a minimum tim e is required
to clear out the stored charge accumulated during the forward bias. This
time is a function of the carrier life time as well as the ratio of the forward
to reverse bias current [99].
This speed lim itation can be overcome by operating the PIN diode in
the forward bias mode only. Hence, an accurate PIN diode model is re­
quired to characterise high speed forward biased operation at microwave and
millimeter-wave frequencies. W hile the PIN diode has been extensively stud­
ied, an accurate model th a t describes the PIN diode operation in response to
high speed d a ta is not available. In this section, the PIN diode operation in
response to high speed multilevel d ata is studied. The residue theorem and
complex inversion formula are used to obtain the electron density function.
A device model is derived from this density function. The method results in
a closed form formula for th e PIN diode resistor under forward bias.
5.2.1
A n a ly sis
The fundamental equations for PN junction analysis are the current den­
sity and the continuity equations [100]. The current density equations consist
81
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
of a drift component and a diffusion component and are given by:
Jp =
Jn =
q (jip p E -
D p V p ).
(5 .1 )
q(pnn E + DnV n ),
(5 .2 )
where Jp and Jn are hole and electron current density, pp and pn are the hole
and electron mobility factors, and Dp and Dn are diffusion constants of holes
and electrons respectively. The well known continuity equations are:
r-7 T
V .Jp —
_
rdp
q.
,
Ctt
Tp
.dn
-
P ~ Pn
_
n —nv
V .J n —
x
9 p )i
,r
(°-3 )
.
Z
,
(°-^)
where p and n are hole and electron minority carrier densities, q is electron
charge, pn and nv are equilibrium hole and electron densities, rpand r n are
hole andelectron carrier life times, and gp and gn are hole and electron
generation factors. The other im portant equation is Gauss’ law:
V .(e E )
=
p,
(5 -5 )
P = q(P — n + N d - N a).
(5 .6 )
where p is the net charge density, e is perm ittivity and iVd and N a are donor
and acceptor densities. T he generation term is usually very small and it can
be ignored. Moreover, the assum ption of heavily injected holes and electrons
is usually valid(n »
n n and p »
pn).
Using these assumptions, the
continuity equations may be simplified to:
- V
4 =
« ( |
+ f ) .
V .J„ = q & + - ) ,
Ot
Tn
82
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(5.8)
These equations have to be solved to obtain the electron and hole densities.
Assuming th at the electric field E has only transverse components, i.e, ^
0
=
, and the electron and hole densities are a function of x and f, the electron
and hole densities equations may be written as follows:
dn
_ d2n
n
~di ~ n dx* ~ Tn
dp
dt ~
(5‘9)
d2p
p
p~daJ ~ Tp
(3‘10)
Assuming a sym m etrical structure, electron and hole carrier densities may
be considered to be the same. It is also possible to approxim ate rp = r n = r
and define an am bipolar diffusion constant D =
2
D nD p/( D p + Dn). Using
these assumptions and taking the Laplace transform of (5.9) result in:
d2N ( x ,s )
,1 + sr
dx2
~ ^ rD ^
where N {x, s )is the Laplace transform of n(x, £). Eq.(5.11) has the following
solution:
iV'(x, s ) = kie V W * + k2e ' / 1^ ~ x.
(5-12)
Using this solution, and considering a symmetrical PIN diode as shown in
Fig.5.2, we can assume N ( x ,s ) = P (x , s). For a symmetrical PIN diode
the carrier density is minimum at the center of the intrinsic region (x=
0
in
Fig.5.2). The R F carrier density is mostly concentrated in the boundary of
n and p region [1 0 1 ], so th at:
s) + N ( x , s )
] |„ 0
=
s)
| „
0
= 0.
(5.13)
This condition results in k x = k2 in (5.12) and this equation may be written
as:
/ 1 -f- ST"
iV(x, s ) = 2kxcosh(J ^ ---x).
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(5-14)
Using (5.14), ki may be solved in term s of the charge density at x = -y , so
th at:
ti =
,
2
( 5 . 15 )
c o sh (y /^ f)
Using ki in (5.14), the charge density in the Laplace domain is obtained as:
, 3)
l l + ST
c o s h ( f i ± j f f ) COS
V
tD
Next, the solution for iV(x, s ) in response to the bias current is examined.
At x = y , the hole current is zero, so th at the following condition is valid:
D dP£ ' S)
3) =
f tn E .N ( ~ , s).
(5.17)
The ambipolar mobility factor is defined as /z = 2fjmnP/([in + fj.p), so that:
»)•
<5-18)
W hen the PIN diode is forward biased with high speed data, three current
components have to be considered. The first component is a DC bias which
makes diodes operate under the forward bias condition. The second compo­
nent arises from the incoming d a ta signal that is used to control the level of
the attenuator. The last current component is the RF current component at
the frequency
uj0
. Under higher power operation, the R F signal harmonics
have to be considered as well. Thus, the total current in the PIN diode may
be expressed as:
^ + TD(s) + I x (s) = A q(fiN E +
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(5.19)
p
I
-W/2
F ig u r e 5.2
N
W/2
The PIN diode structure
Using Eq.5.18, Eq.5.19 may be written as:
£ + I D(s) + I R(s) =
.
(5.20)
Substituting from (5.16) in (5.20), results in:
t
n
7 +
T° ( s ) +
« s)
(— s) / i~!~5r
= iA q D -
/tt
^
Q
<5-21)
From this equation, iV^-y, s ) may be written as:
W
JV(— ,s) =
2 A q D ^ s in h (^ f)
»
From Eq.(5.16) and (5.22), the electron density function can be given by:
s) = . . „ / r r l V,
4 - r . . „ (-r + ^ W + /*(*))•
2At D ^ s m h ( ^ f ) *
(5.23)
The charge density, N"(x, s), may be separated into three components as
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
N ( x , s) = Ni(x, s) + N 2(x,
s
)
-f- N 3(x,
s),
(5.24)
/
---- ( 5
f ) »
25)
where,
Ni(x, S) =
cosh(yf^jfx)
Ky.-T.D
2 X 9 r>v^ « n h ( v^
N 2{X, s ) =
cosh (J^ ^ x)
----- / D(s),
r - V
2 A q D ^ ± fsin h (^ f)
(5.26)
N z{x, s) =
co sh (J ^ ^ -x )
W rD
------J*(s).
2 A q D ^ s in h {^ f)
(5.27)
T he three time domain components of densities are obtained by taking
the inverse Laplace transform,i.e.,
n x(x, t) = L~l (Ni(x, s)),
(5.28)
n 2(x,
)),
(5.29)
n 3(x,t) = L~l (N3(x,s)).
(5.30)
t)
= L~l (N2(x,
s
In order to obtain the time domain components, the residue theorem and
complex inversion formula can be used.
D C R e sp o n se
The DC current is used to ensure th a t the diode operates in the forward
bias condition only. The Laplace transform of electron density resulting from
the DC current N i ( x , s) is given by:
cosh(\p^fx)
I
N 1(x,s) = ---------W td J =------ o
2 A q D ^ $ ^ sin h (^ f) *
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
£5>31j
This equation may be written as:
N l { x ’ s) = 2 X j D F' ( x ’ s)'
(5'32)
cosh(J^f-x)
Fi(x, s ) = — ; - - V tD>- - — .
s \ / W sin /i( \ / w v )
(5.33)
where,
This function has a pole at the origin (s = 0), a second order pole at s = —■£,
and infinite poles where the sinhyperbolic function is equal to zero,i.e, at
11
-f* S T
W
sin h (V ~ 7 d ~
= °’
(5-34)
or,
/1
2
—
i—st
VTd~ T = j2k*
k = _ 2 , _1, lj 2’-
(5‘35)
k = 1.2,...
(5.36)
These poles are given by:
1
„.2kir
sk = - ~ D ( ^ - f
As shown in Fig.5.3, to obtain the residues, an integration contour with
radius Z m may be considered where,
_.
,
X. . X
I ttvk , 9 .
Zm = {rn + - ) ( - + D{—— ) )
J
r
W
.
m = integer
.
« —,
(o.37)
This choice for Z m insures th at the contour does not pass through any of the
poles. Now, the residues for different poles can be obtained.
For s = 0, the residue R 0 is given by:
S, =
(5.38)
smh( u w )
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
m
m
F ig u re 5.3
The residue for poles s = sk, k =
The integration contour
1
,
are given by:
R k = lims->Sk(s - sk)Fi(x, s)est.
(5.39)
Now
,.
.2/C7T
1+ S T
hm^ y - p D ~ = 3 W '
(o-40)
so that the residue R k can be expressed as:
4 Dcos(2£fx)e~[i+D(2& ' ^ t
Rk = W c o s ( k l r ) ( - i - D ( ^ - Y ) '
•=
<5- « )
The only pole left in the equation is the second order pole at s = —
Its
residue is given by:
R’ = Hms_ _ x - ^ ( { s + —)2Fi(x, s)est).
r as
t
<§;
1
(5.42)
is a good practical assumption [6 6 ], which can be used in the above
equation so as to give the residue:
o'
2 D ,-
d r,
1
s C0
R = w h m ^ T s [{s + 7 }
5
/l( V w F x )
s
5£i
‘e
]
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
( 5 -4 3 )
After taking the limit, a simple result for R ' is obtained as:
>
=
2D r =t
w ~e ' •
<5-44)
Using the integration contour shown in Fig.5.3,
1
r
M
—— <b estFi(x, s)ds = R 0 + EL + ^
Z/K J J Cm.
fc=
R k.
(5.45)
j
Using the residue theorem to estimate the above integral, the electron density
function corresponding to the DC current is obtained as:
( )
Ul{X' j
Io ( y^
COsh{^ D )
2AqD
sinh( ^ )
2Dr
W 6
'
™ 4 D c o s ( ^ x ) e ~ £ + D^ W t
& Wcos{h-K)[-± - D { ^ \
(5.46)
As may be seen from (5.46), as time t tends to infinity, the second and third
components approach zero and the electron density after a transition time
can be obtained from the firstterm.
This isconsistent with the well known
steady state relation [6 6 ], [1 0 2 ]. The second and third terms also show th a t
diode operation in the on-off mode is limited to slow speed operation only.
D a ta R esp o n se
The iV2 (x, s ) component in Eq.5.26 corresponds to the data signal and is
defined as:
s) =
s),
where,
(5.47)
____
coshiJ
F2( x , s ) =
W
<
• ~ -fp ( s ) .
5 .4 8
4 ^ § s in h { ^ \)
A PAM sequence with period T is considered as the data signal. Such a
signal is shown in Fig.5.4 and it can be expressed as:
iD(t) =
A k{u(t - kT) - u(t - (k + 1)T)),
*=o
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(5.49)
(t)
Ak
T
T
T
F ig u re 5.4
A multi-level data sequence
where A * is the am plitude of signal. The Laplace transform of this signal is
given by [103]:
OO
9
c '7"*
A
f D(s) = T — e~kTssinh( V ) .
*=o 5
2
(5.50)
Substituting (5.50) in (5.48), F2(x,s) may be w ritten as:
F2(x,’ s ') =
coshQ ^x)
~ 2Afc _stT . t , sT,
----- V ] — o~e skTsi n h ( V—O ).'
, - - —- , /■ r^-/ l J - r .t W\
(5.51)'
/lX r v T
'
Hence, this component of density function is given by:
m x , s) = _L_(
£ ^ t e-*Ts,mh[iL )y (5.52)
Using the final value theorem in Eq.(5.52), h‘77zt_i.0 0 n 2 (x, t) = l i m b o s N 2(x, s),
the electron density function because of the d ata signal is given by:
1
,
^
cosh(J^^x)
“ 2.4fc
sin h {s/S J } £
_ , fcT
.
sT
(5.53)
It may be easily seen th a t the electron density goes to zero when t —>■oo.
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
This shows th at the forward bias operation highly decreases the lim itations
on the PIN diode operation speed. These interesting results can be used to
design high speed multilevel digital modulators [7], [10], [11] and [12].
R F R esp on se
The electron density for the third component of the current, i.e., R F
current can also be computed in a sim ilar way. For simplicity an RF signal
with only a sinusoidal component is considered, so th at:
*rf(£) = B c o s iu t ),
(5.54)
The Laplace transform of this signal is:
Bs
= s2 + u ;2 ’
^
where B
component
(5‘55)
is the amplitude of the sinusoidal and
of currentresults in the
uj
is its frequency.
This
third component of electrondensity,
Ar3 (x, s), and it can be w ritten as:
cosh(J^^-x)
■ ^ ( l , s ) = --------------.
W
Bs
rD
( 5 .5 6 )
2 A q D ^ 5 i n h ( f i j § li - )
* 2
+
“ 2
or:
N 3 (x, s )
=
0 A ~d F 3 (s,
(5.57)
s),
where:
F3 (x, s )
s cosK J^x)
= ---------------- v - 1 - D
------------.
) ( s 2 + “ 2)
This function has a second order pole a t s =
(5.58)
two poles at s = jui and
s = —juj and an infinite number of poles because of s i n k function zeros. For
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the poles at s = j u and s = —jcj, the residues are:
L i = lims^ j u [(s - ju>)F3(x, s ) e s£],
(5.59)
L 2 = lima+ . j u [ ( s + j u ) F 3(x, s)est],
(5.60)
or:
cosh ( J 1+J n Tx )ejuJt
Li =
----------- j=- -
— .— .
,
(5.61)
,
(5.62)
f)
cosh(J
L2 =
Lx)e~]UJt
— / - .
7—
As before, the residues corresponding to the poles resulting from the sink
function zeros can be obtained as:
Lk = Zims-+Sfc[(s - sk)F3(x, s ) e s£].
(5.63)
This equation may be w ritten as:
‘
i D {=± - D ( ^ ) c o s ( ^ x)
Wcos(M)[(=± - D ( l = ) 2)2. + U2]
The other pole of F 3 (:r, s) is at s = —-p, and this pole is a second order pole:
i ! = lim
-I- - ) 2 F 3 (x , s ) e s£].
’■ a s
r
(5.65)
Using the approximation:
.
I + s t W
1
s ,n h { y ~ F D ~
1 + s t W
T> ~ V~ 7 d ~ T ’
(5,66)
The residue is obtained as:
—t
r'
fd
e"
W t (±)2 +
u 2'
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
( ‘ }
Now, the inverse Laplace transform of Fz(x, s) can be obtained as:
~T~~z f
—7
As a result, the
TJ J C m
713
estFz{x, s)ds = L l + L 2 + L' + Y I L k.
(5.68)
(x, f) term can be w ritten as follows:
1 + ju lT W
2
D
W
t
e-r
W cos{kTr)[{
+ cu2
As may be seen, this density component includes a harmonic term and two
transient terms. T he factor u in the denominator for the transient terms
shows th a t at the higher frequency operation these two transient terms dis­
appear rapidly.
P I N D io d e V o lta g e
The total electron density, n ( x , t ) , consists of three components and is
given by:
n ( x ,t ) = n i(x, t ) + ri2 (x, t) + nz(x, t).
(5.69)
The conductivity equation gives:
<r(x, t) = q(finn(x, t ) + fipp{x, t)).
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(5.70)
Assuming p ( x ,t ) = n ( x ,t) , the conductivity can be written as:
a(x, t) =
2
qpn{x,1).
(5 .7 1 )
Using O hm ’s law, the resistance of a cylindrical conductor of electrical con­
ductivity <x, length W along the current path, and cross section A is :
Using this relation, the PIN diode voltage may be w ritten as:
(5.73)
5 .2.2
N u m e r ic a l S im u lation
Forward B ia se d R esistan ce
As a typical example, the model was applied to study the response of
a cylindrical PIN diode with radius r = .78 mm, p = .061 m 2/ V , D =
15.6 x
1 0
_ 4 C7tz2 / s , r = 5pse c,W = 28pm. As a first step, the PIN diode
is considered to operate in its linear region without any bias variation (only
a constant DC bias), and it is assumed that no RF signal is applied to the
diode. T he DC current values are assumed as .1 mA and 1 mA. The results
obtained from th e above model were compared with the previously reported
results [66],[104]. As may be seen in Figures 5.5 and 5.6, the final value
of the PIN diode resistance approaches the well known results [6 6 ], [104].
However, the resistance value during the transition is much higher than its
final value. It is an im portant factor th a t limits the PIN diode speed as
a data m odulator for on-off operation modes . For instance, for the diode
under study, th e maximum speed of d ata m odulation in on-off mode is less
than 100 kbps. T he numerical results are also compared with th e measure-
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.2
F ig u r e 5.5
0 .4
0.6
0.8
1.2
1 .6
1.8
The PIN diode forward bias resistance with I= .l
raA
m ent result of the commercially available PIN diodes, MA4P4000 from M/A
Com and SMP-1304 from Alpha. As may be seen from Fig.5.7, there is a
good agreem ent between M /A Com measurement [105] and the results of the
above model. T he comparison between this model and the results in [106] for
A lpha’s diodes is shown in Fig.5.8. Again, a good agreement was obtained.
P I N D io d e a s a High. S p e e d P A M M o d u la to r
The m odel developed above was used to study a basic multilevel PAM
m odulator. This circuit is shown in Fig.5.9. An RF signal with peak value of
1
volt and frequency of 2.5 GHz is applied to the PIN diode (Vin — c o s ( 2 tt
2.5 x 10 9 t)).
x
T he data rate was selected as 4 M sym/sec and data level
symbols correspond to the drive currents of .1, .3, .5, and
1
mA. To avoid the
charge storage problem the diode was always operated in the forward bias
mode. In Fig.5.10, the drive current for the PIN diode is shown. To avoid
the transition tim e region, the diode operation was examined after .04 msec.
T he PIN diode resistance because of data drive is shown in Fig.5.11. The
PIN diode voltage is obtained using Eq.(5.73) and is shown in Fig. 5.12. As
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o
This paper
Relj [661.[1Q4[
8 10
0.4
0.6
0.8
1.2
1.8
1.6
X 1 0 '5
F ig u r e 5.6
The PIN diode forward bias resistance with 1=1
mA
may be seen, the net DC level of current controls the resistance of the PIN
diode in the forward bias condition, and the level variation does not impose
any lim itation on m odulator performance.
5 .2 .3
D ir e c t M icrow ave Q A M M o d u la to r
The above theoretical results were applied to analyze a direct QAM mod­
ulator structure introduced in Fig.5.1. For practical implementation, the
variable m odulator section and sign modulator section can be replaced by
a reflection type PIN diode attenuator and an SPDT switch, respectively.
Such a direct microwave modulator is shown in Fig.5.13.
H igh S p e e d R eflectio n PA M M odulator U sin g P I N D iod e
The key block in the direct microwave m odulator strcture is a reflection
multilevel PAM m odulator. The circuit block consists of two PIN diodes
driven by high speed data and a hybrid as shown in Fig.5.14. A multilevel
baseband PAM signal is used as the drive signal for the PIN diode while an
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
S eries resistance vs. forward current for HIPAX PIN diode from M/A COM
10
x
10
c
This Method
M/A Com M easurem ent
'
10'
m
10
10 '
Forward Current[Amp]
F ig u re 5.7
Comparison between the model and M/A-Com
measurement for MA4P4000 at
100
MHz and
100 mA RF current
S e rie s resista n ce vs. forward cu rren t for 1 3 0 4 PIN d io d e from Alpha
This M ethod
Alpha M easu rem en t
<r i<y
* 101
10 '
10
'
Forw ard CurrentfAm p]
F ig u r e
5 .8
Comparison between the model and measure­
ment for SMP 1304 from Alpha at 100 mA
97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
—
ooooo^nr
Data
+ Vj (t) -
z s=50
—i—
ift)
r u
\
F ig u re 5.9
+
Zl =50
V ft)
in
Basic structure used in the study of the PIN
diode modulator
10
'
<E
10
4.05
4.1
4.15
4.2
4.25
4.3
tim e[sec]
4.35
4.4
x 10‘5
F ig u re 5.10 Driver current
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
S 10
coe
a.
4.05
4.1
4.15
4.2
time[sec]
4.25
4.3
4.35
4.4
x 10,-5
F ig u re 5 .1 1 PIN diode forward biased resistance as a func­
tion of tim e
0.3
0.2
0.1
-0.3
F ig u re
4.05
5 .1 2
4.1
4.15
4.2
time[sac]
4.25
4.3
4.35
4.4
x 10'
PIN diode voltage in the basic structure as a
function of time
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1
;0 omb"inef|
Gdupleri
m m m
F ig u re 5.13 An architecture for direct microwave QAM using
the variable attenuator in I and Q paths
100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
R F signal pum ps the RF power in the modulator. If the PIN diode resistance
is denoted by R d , the output signal of the attenuator is given by [1 0 2 ]:
S 2 1 = 2SZ1p r ( t ) S'C41»
‘i
(5.74)
where
(5.75)
and Sfi is the coupled voltage transmission coefficient of the coupler,
is the direct voltage transmission coefficient of the coupler, and Z Q is the
characteristic impedance. By using an ideal quadrature coupler, Eq.(5.75)
describes the reflection modulator response. Again, in this experiment, the
d a ta rate was selected to be 4 Msym/sec and pump frequency was selected
to be 2.5 GHz with a peak amplitude 1 volt. A four level data signal cor­
responding to the drive currents of .1, .3, .5, and 1 mA was applied to the
diode. The response is shown in Fig.5.14.
In the next experiment an eight level PAM signal was applied to the
m odulator. The drive current and PIN diode resistance are shown in Figures
5.15.a and 5.15.b. The modulator output for a lower carrier frequency is
also shown in Figure 5.15.C. As may be seen, a high speed m odulator could
be implemented a t 40 Msym/sec. Higher speed is also feasible. T he use
of predistortion for further improvement of the m odulator performance is
described in the following chapters.
This experim ent was repeated by using prefiltered baseband data. The
filter was a raised cosine filter with roll-off factor a = .5. The drive current,
th e PIN diode resistance and the m odulator output are shown in Fig.5.16. As
may be seen, a high speed modulator was realized. However, predistortion
must be implemented to linearise the output.
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
High Speed Driver
9 0.6
5541
4.05
4.15
4.2
time[sec]
425
x 10-
(b)
F ig u r e 5 .1 4 a)A reflection multilevel PIN diode PAM mod­
ulator b) M odulator response for the drive cur­
rents of .1,.3,.5 and 1 mA.
102
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
<
.1 . 0.8
c
©
(a)
L.
3 0.6
o
g. 0.4
c
0.2
40
_L
40.05
_L
40.1
40.15
40.2
40.25
_L
_L
40.3
40.35
40.4
40.45
40.5
time[microsec]
(b)
*0o
1 10'
40.05
40.1
40.15
402
40.25
40.3
40.35
40.4
40.45
40.5
time[microsec]
S'
0.5
(C)
2 -0.5 L
40
40.05
40.1
40.15
402
40.25
40.3
40.35
40.4
40.45
time[microsec]
F ig u re 5.15 Reflection M odulator without prefiltering and
predistortion a) Input current b)PIN diode re­
sistance c) M odulator output
103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
40.5
PIN dlode Reslstance[Ohm]
40
10
40.05
40.1
4 0 .1 5
402
40.25
40.3
40.35
40.4
40.45
40.5
40.3
40.35
40.4
40.45
40.5
40.3
40.35
40.4
40.45
40.5
time[microsec]
'
10
,o
40
40.05
40.1
40.15
40.2
4025
Modulator output[v]
time[microsec]
1
0
•1
40
40.05
40.1
40.15
40.2
4025
time[microsec]
F ig u re 5.16 Reflection M odulator using prefiltering(a
=
.5) and without predistortion a) Input current
b)P IN diode resistance c) Modulator output
104
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
s e t)
QAM Signal
F ig u re 5.17 A model to study the direct QAM modulator
D ir e c t M icrow ave QAJM M o d u la to r
An architecture to realize a direct microwave QAM m odulator was pre­
sented in Fig.5.13. The analog signal processing section prior to the reflection
m odulator realizes a BPSK m odulator in each path where the symbol rate
is 1 /T sym/sec. Then, the input signal to the PIN diode m odulator can be
w ritten as
OO
Sr(t) = ^ 2 aku(t — kT)cos(2'Kfct),
(5.76)
k = —oo
where ak =
1
when the symbol has positive polarity, and ak =
symbol has negative polarity; and u(t) =
1
— 1
when the
in interval —T f 2 < t < T / 2 and
zero elsewhere. A sim ilar expression can be written for the Q path. T here­
fore, the direct microwave m odulator can be modelled using Fig.5.17. The
model is used to realize a 64QAM direct microwave m odulator. The input
d a ta rate was selected to be equal to OC-3 (155.52 Mbps). While an eight
level PAM signal with symbol rate of 25.92 Msymbol/sec was used in each
path, a four level rectified PAM signal was applied to each attenuator. Fig­
ures 5.18.a and 5.18.b show the bias currents for the I and Q attenuators. As
may be seen, these signals have been filtered using a raised cosine filter with
roll-off factor a = .5. T he PIN diode forward bias resistances in response to
these biases are shown in Figures 5.18.C. and 5.18.d. The PIN diode voltage
responses to bias voltages are shown in Figures 5.19.a and 5.19.b, and the
outputs of the m odulator in I and Q paths are presented in Figures 5.19.C and
5.19.d. The time and frequency domain responses of the 64QAM m odulator
are shown in Fig.5.20.a and Fig.5.20.b.
According to Fig.5.20.b, the band-
105
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.4
<
E
0.8
.
e
i
0.6
0.8
0.4
3 0.6
CQ
0
0.4
02
1
Q
scu
02
40
42
44
46
time{microsec]
48
40
42
44
46
time[microsec]
b
48
40
42
44
46
timefmicrosec]
48
a
PIN diode Reslstance[Ohm]
10“
10
8 10
.0
10
40
42
44
46
time[microsec]
C
48
d
F ig u r e 5.18 a)Bias in I path b)Bias in Q path c)PIN diode
forward bias resistance in I path d) PIN diode
forward bias resistance in Q path
106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-0.4
42
44
46
42
tim e[m icrosecl
44
46
tim e[m icrosec]
P -0.5
42
44
46
42
44
46
tim e[m icrosec]
tim e[m icrosec]
F ig u re 5.19 a)PIN diode voltage in I path b)PIN diode volt­
age in Q path c)PIN modulator output in I path
d)PIN modulator output in Q path
539148484853535353235323232348484853485353
timefsecl
a-o
-1 0
I
- 20
0
-3 0
1
-4 0
o
-50
C/3
Z
Z1
2.2
Z3
2.4
2.5
2.6
2.7
2.8
Frequency[GHz]
F ig u re 5.20 a) Tim e domain response of 64QAM b)Power
spectrum of 64QAM modulator
107
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
width of the proposed direct microwave 64QAM m odulator for transmission
of OC-3 signal is about 40 MHz.
5.3
D is to r tio n in th e P I N D io d e M o d u la to r
The PIN diode electron density in the steady state case was obtained as:
Io
V rD cosh (-^ =)
2AqD
sin h { - ^ = )
where Bi is the amplitude of the sinusoidal signal. The PIN diode voltage is
obtained using Eq.(5.73), where i(t) is as follows:
M
i(t) = I 0 + Y2 BiCosuit.
i=i
(5-77)
The Harmonic distortion and power spectrum for the PIN diode, basic atten­
uator, and reflection attenuator using an input with peak value of
1
volt and
frequency of 2.5GHz with drive current .1 are compared and shown in Fig­
ures 5.21. As may be seen, the level of harmonic distortion is always better
than 50 d B c. Also, by increasing the drive current, the harmonic distortion
decreases. T he measurement results for nonlinear PIN diode operation has
already been reported [107]. A good agreement has been found between these
results and measurement results. The second and third inter-modulation dis­
tortions for PIN diode, basic attenuator, and reflection attenuators are also
shown in Figures 5.22. As may be seen with the higher drive current, the
level of inter-m odulation distortion decreases. Moreover, th e level of inter-
108
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Power[dBJ
t ii
ICO
-JOO
-a*o
• xq
-*o
2
3
4
S
I
7
I
•
10
Frequency [GHz]
Frequ«icy[GHz]
Powcr[dB]
r
t
Frequency [GHz]
O
\oo
i
0.*
os
p_4
i\£
a
Forward DC ^currentfmA]
Forward DC^Zuirent[mA]
F ig u re 5 .2 1 Harmonic distortion of PIN diode at f=2.5 GHz
a) Harmonic distortion in PIN diode at I Q =
.1
mA b)Harmonic distortion in basic modu­
lator a t I0 =
.1
mA c)Harmonic distortion
for the reflection m odulator at I c
.1 mA
d)Harmonic distortion for the PIN diode vs. cur­
rent e)Harmonic distortion for the basic modu­
lator vs. current f) Harm onic distortion for the
reflection m odulator vs. current
109
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
0
-60
-100
ca
Sv*
<U
o£
ca
-a
u.
CJ
£
a
•SO
a*
>250
0
>
2
3
4
S
«
7
f
t
9
I
-330
O
Frequ'encyfGHz]
b
Frequ^icy[GHz]
c
o
c
to
5
c
o
CS
■oo
E
ou.
c
Forward DC current[mA]
O
l-r
O
ee
G
Forward DC current[mA]
Forward DC current[mA]
F ig u r e 5.22 Inter-modulation distortion of PIN diode (fre­
quencies are f=2.25 and 2.75 GHz) a) IMD in
PIN diode at I0 =
.1
mA b)IMD in basic mod­
ulator a t I 0 = .1 mA c)IMD in PIN diode at
I0 = .3 mA d)IMD in basic m odulator vs. cur­
rent e)IMD in PIN diode vs. current f)IMD in
reflection m odulator vs. current
110
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
modulation distortion is always better than 50 dBc -
5 .4
S um m ary
In this chapter, an accurate model to characterise a PIN diode as a vari­
able resistor driven by high speed digital data and RF pump signals was
presented. The model extracts the electron density function resulting from
these signal drives. This is used to obtain a closed form expression for the
PIN diode resistance. The residue theorem and complex inversion formula
are used to solve the PN junction equations. A high speed reflection pulse
amplitude m odulator which is an essential block for implementing a more
complex digital modulator is also examined using this model. Moreover, a
new architecture for an M-QAM m odulator using PIN diode attenuators was
presented. A detailed discussion of the modulator hardware implementation
is presented in the next chapter.
I ll
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.
H A R D W A R E IM PLEM EN TA TIO N OF
D IR E C T MICROWAVE QAM
M O DULATO R
An architecture for a direct microwave QAM m odulator was proposed
and analysed in the last chapter. In this chapter a hardware implementation
of such a m odulator is addressed. The main objective is to design a direct mi­
crowave QAM modulator. The m odulator implementation using microwave
integrated circuits (MIC) and silicon monolithic microwave integrated cir­
cuits (Si-MMIC) techniques is described. The MIC realization is simple and
straightforward. The carrier frequency used to implement the MIC version
was selected to be 2.5 GHz. This implementation uses very low cost PIN
diodes, m aking this version an attractive solution for this band of frequency.
Although silicon monolithic microwave integrated circuit (Si-MMIC) is still
an evolving technology particularly for the higher microwave and millimeter
wave bands, it offers a very low-cost realization making it a very desirable
option [108]. T he direct m odulator described in the last chapter is very suit­
able for implementation using this technology. The Si-MMIC version was
implemented a t a center frequency of 18 GHz. This implementation can be
scaled to m illim eter wave frequencies as well.
6.1
D ir e c t M icrow ave M -Q A M Im p le m e n ta tio n u sin g
M IC T ech n iqu e
6 .1 .1
S u b s y s te m A rch itectu re
The architecture of a high power direct QAM m odulator is shown in
Fig.6.1. A Lange coupler is used to create the required R F signals for the I
112
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Lange
Coupler
SPOT
Coupler
-90
•180
Lange
Power
Coupler
Combiner
Lange
Coupler
SPOT
-90
-180
F ig u re 6.1
Coupler
Direct QAM M odulator
and Q paths [4]. In each path, a rat-race coupler is used to get the inphase
and out of phase RF signal. T he subsequent circuit blocks in each path con­
sist of a single pole double throw switch (SPDT) switched by the polarity of
incoming d a ta , a reflection attenuator implemented by using forward biased
PIN diode term inations and controlled by the absolute values of a baseband
pulse am plitude modulation (PAM) signal, and a variable reflection phase
shifter to ad ju st the phase error. To realize the modulated signal filtering
and to com pensate for nonlinearities, a pre-filtering and pre-distortion block
may be used to shape the baseband PAM signal in each path [16],[7]. Fi­
nally an inphase power combiner is used to combine signals from two paths
into one high power QAM m odulated and filtered output at microwave fre­
quency. A lthough the center frequency is selected to be 2.5 GHz in a 500
MHz bandw idth for MIC implementations, a sim ilar design may be repeated
for different bandw idth and center frequencies. The various parts of this
m odulator are described in the following sections.
6 .1 .2
Q u a d ra tu re C ou p ler
As may be seen in Fig.6.1., the power oscillator signal must be divided
into inphase and quadrature components. A Lange coupler, shown in Fig.6.2,
113
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
F ig u re
6 .2
A Lange Coupler
is a suitable circuit to realize this block [63]. An ideal Lange coupler is
a quadrature hybrid th at equally divides input signals between direct and
coupled ports with quadrature phase shift. Moreover, using a Lange coupler,
a very broadband quadrature hybrid can be realized. In the Lange coupler,
bonding wires are used to interconnect the microstrip lines. The length of
coupler should be A/4 in the operating frequency and the width of the fingers
and their gap are estim ated according to the coupling factor. A microstrip
Lange coupler was implemented using a 25 mil thick substrate with er = 10.
The w idth of fingers, W , their gaps, S , and length of coupler at 2.5 GHz
are W = 1.827 mil
(1
mil=.0254 mm), S = 1.970 mil, and L = 481.352 mil,
respectively. The coupler arms are 50 f2 lines with W 1 = 23.57 mil. The
circuit was designed using Libra-EEsof [109] . The layout and performance
of the Lange coupler is shown in Fig.6.3. As may be seen from Fig.6.3.a,
over the 500 MHz band, the through and coupled ports have almost the same
am plitudes. Moreover, Fig.6.3.c shows th a t the two outputs have a 90° phase
shift. Thus, the Lange coupler can operate efficiently as a quadrature hybrid.
As may be seen from Fig.6.3.b the return loss and isolation performance of
this coupler is very good over the 500 MHz band.
114
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-* S14[dB]
"O S12[dB]
5-*
CO ^
2.1
2^
2.3
2.4
2.5
a
2.6
Z7
2.8
Z9
-10
-*
-0
« -20
0
)
S11 [dB
S13[dB;
o8-30$:
CO-40
-50
2.1
2.2
_i_
_i_
2.3
2.4
_ i_
2.5
2.6
2.7
2.8
2.9
b
CM
CO
-*
-a
100
V
*Q ■" 0
08
« .-1 0 0
r-a-r:
irr.* ....* ..
2
2.1
2.2
2.3
2.4
2.5
O
<S14 degree
<S12 degree'
0
t---*---f2.6
2.7
2.8
"O " 0 " O
-* *2.9
FrequencyfGHz]
c
F ig u re 6.3
Lange Coupler a) S14 and S12 as a function of
frequency b) S l l and S13 as a function of fre­
quency c)Phase S14 and S l l as a function of
frequency
115
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
T)
6 .1 .3
R in g C o u p le r
As seen in Fig.6 . 1 , an 180° coupler is required to provide the inphase
and out of phase components to the R F signal in each path. A suitable
circuit to realize this block in MIC technology is the rat race (ring) hybrid.
Such a coupler is shown in Fig.6.4. When a signal is applied to port
1
, the
outputs at ports 2 and 3 are inphase, and when a signal is applied to p o rt
4, the outputs at po rts
2
and 3 will have a 180 degree phase shift. T he
characteristic im pedance of the coupler arms is 50 ft, and the characteristic
impedance of the ring is 70.7 ft. A standard design process was followed to
realize this coupler [63]. The coupler was implemented on microstrip w ith a
25 mil thick substrate with er = 10. The circuit was designed and optimised
using Libra EEsof [109]. In Fig 6.4, the physical dimensions of the coupler are
W l = 23.57 mil corresponding to 50 ft lines, W = 10.14 mil corresponding to
70.7 ft lines, and R = 455 mil. As may be seen from Fig.6.4.a, the coupling
and through outputs are over 3 ± .2 dB range over a 500 MHz bandwidth.
A good return loss a n d isolation (better than 20 dB) were obtained over the
band. According to Fig.6.4.c, the phase shift between coupling and through
ports is close to 180°.
116
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
©
CM
CO
"2©
-*
-0
=9-
00
Tj-
S34[dB'
S24[dB;
CO■
CO
-6
2.1
2.2
2.3
2.4
2.1
2.2
2.3
2.4
T
T
2.5
2.6
2.7
2.8
2.9
2.5
b
2.6
2.7
2.8
2.9
a
-1 0
°B -3 0
-5 0
10
-
T
T
ff
T
<S24 'd e g re e
<S34 'd eg ree'
100 '
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
Frequency[GHz]
F ig u re 6.4
A Rat-race or ring hybrid a) S14 and S12 as a
function of frequency b) S l l and S13 as a func­
tion of frequency c)Phase S14 and S l l as a func­
tion of frequency
117
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3
6 .1 .4
S P D T S w itch
An SPDT switch must be implemented to provide the polarity modula­
tion for the RF signal. The SPDT is controlled by the sign of the incoming
data. Using an SPDT, 0° and 180° outputs of the ring coupler are selected
based on the positive and negative polarities of the incoming data. A series
configuration, shown in Fig.6.5, was used to realize this switch as the series
configuration has the advantage of wider bandw idth compared with parallel
topology. The switch uses a PIN diode in each arm. The diode bias was
controlled by the d a ta polarity.
To realize a high performance QAM m odulator, an SPDT switch with
high isolation is required. While traditional implementation of series SPDT
switches on microstrip suffer from low isolation, a coplanar waveguide (CPWG)
implantation could solve this problem[110]. Although CPWG implementa­
tion highly improves the isolation of the switch, it increases the insertion loss
slightly. Fig.6.5 shows the insertion loss, isolation and return loss of an ideal
line, as well as those of a microstrip , and a CPW G series SPDT switch. The
PIN diodes are SMP1304 from Alpha industries [106], biased using a 20 mA
current and unbiased with a zero current. The measured scattering parame­
ters of the diodes provided by the m anufacturer are used in the simulation.
For the microstrip line implementation, the best performance is obtained
with the following dimensions in Fig.6.5: W i = W 2 = Wz = 23.57 mil, and
Ax = 18.65 mil, A 2 = A 3 = 100 mil, and L \ = 1000 mil, L 2 =
£ 3
= 682.51 mil,
D i = 344.70 mil, D 2 — Dz = 0. As may be seen from Fig.6.5, the optimized
microstrip switch characteristics are close to an ideal line switch performance.
The microstrip implementation has an insertion loss of about .5 dB and an
isolation of about 16 dB. A return loss of about 20 dB is also feasible with
this implementation. The optimized dimensions for a coplanar waveguide
implementation are: W \ = W2 = Wz = 23.57 mil, and Ax = 66.7 mil,
118
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
D-
©
t
o ------D,
W,
©
r= r 0
~Q
“55'
<
o
o
c -2
t:(D
U1
CD
■ *
Basic
Q---------s Microstrip
x--------- « CPWG
- -4
2.2
2.3
2 .4
2.5
2.6
2.7
a
-1 0
m
33
1 -2 0
<> •
•
•"
J5
ocn
-3 0
2.2
-1
-
2.3
2.4
b
2.5
2.6
2.2
Basic
Microstrip
CPWG
-*■
o
*
Basic
Microstrip
CPWG
2.7
o
20 .
-4 0 L-
-*
e -------- 0
*-------- •<
2.3
2.4
2.5
2.6
2.7
Frequency[GHz]
c
F ig u re 6.5
A series SPDT switch implemented using Al­
p h a PIN 1304 and a forward bias current 20 mA
a)Insertion loss b)Isolation c)Return loss
119
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
■A-2 =Az = 100 rail , and Li = 698.5 mil, L 2 =
mil,
£>2
£3
= 332.1 mil, D\ = 128.2
= Dz = 100 mil with a constant gap space of 1 mil. As may be
seen from Fig.6.5, an insertion loss of about 2.2 dB is significantly more than
the microstrip version. The isolation of around 25 dB is much b etter than
the microstrip version. The insertion loss can be lowered by using a higher
current than 20 mA for the forward bias arm . The return loss is acceptable
within the desired bandwidth. B etter performance could be achieved using
higher cost diodes.
6.1 .5
V a ria b le A tte n u a to r w ith P h a se Shift C o m p e n ­
sa tio n
R eflection A tte n u a to r
A voltage-controlled attenuator must be used to attenuate the signal to
the levels required for QAM implementation.
The control signal for the
attenuator in each path is the absolute value of the multi-level PAM signal.
A reflection atten u ato r using two PIN diode term inations at the output of a
Lange coupler was used. In a reflection attenuator, the return loss depends
on the hybrid performance. This allows an independent design for the desired
insertion loss characteristics over the operating bandwidth. This attenuator
is shown in Fig. 6 .6 . The PIN diodes are SMP1304 from Alpha industries. The
measured scattering parameters from the manufacturer were used to design
the circuit. The attenuator response for different PIN diode bias currents
is shown in Fig. 6 .6 .a. The attenuator return loss is shown in Fig.
. .b. If
6 6
PIN diode impedance is denoted by Z d , the output signal of the attenuator
is given by [1 1 1 ]:
S2 1 =
2
S ^ p r S :rC
(6 .1 )
where
(6 .2 )
120
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and S 2 1 is the coupled voltage transmission coefficient of the coupler,
is the direct voltage transmission coefficient of the coupler, and Z Q is the
characteristic impedance. Thus, the attenuation (in dB) is given by:
A tt = -lO Z osdSjil )•
(6.3)
As seen in Section 6.1.2, a Lange coupler has very good quadrature hybrid
characteristics. Eq.(6.3) may be simplified as:
A tt = -lQ log{\pr\2).
(6.4)
In the forward bias operation, the PIN diode may be approximated by a vari­
able resistor with impedance, Zd = Rd [102]. Using Eq.(6.2), the attenuation
for a PIN diode attenuator is given by:
A tt = - 1 0 iOS( | | i 5 A | 2).
■LLd “t " £
(6 .6 )
q
The PIN diode forward bias resistance as a function of bias current for an
Alpha SMP1304 PIN diode is shown in Fig.6.7.a. In this measurement Rd
represents the real value of Zd. Because of package parasitic elements, the
results of Chapter 5 cannot be directly used to estim ate Rd- A curve fitting
method was used to model the diode forward bias resistance:
Rd =
.0009
.787
39.3 „
-------- =5 - + — + 1.26.
Ld
1d
1d
—73
(6 .6 )
In Fig.6.7.a, curve fitting results are compared with the measured results.
From Eq.( 6 .6 ) and Eq.(6.5), it is clear th at the attenuation (in dB) has
a nonlinear
relation w ith the bias current.
On the other hand, a linear
relation between the control voltage, Vc , and the attenuation of the reflection
121
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Filtered
Baseband signal
s
Output
©
Input
©
I
9
H
m
,
r
*
v
2
* ............
*
^
c
2
i ------------- B
t
<D
I
............... r
15
=j
c
t
[ ] --------------- 5 ----------------E 3----------------U
2 0
" U --------------- a -
^
-------------------------------------- i---------------- . . . -------------£ _
2.4
2 .2
*-*
“
------------------------------i
2 .6
1=1 mA
I=5 mA
1=10 mA
I=20 mA
I=50 mA
2 .8
m
■o
W
W-2 0
1=1 mA
I=5 mA
1=10 mA
I=20 mA
I=50 mA
o
c
3
ffl
a:
2.2
2.4
2 .6
2 .8
b
100
o) 50'
1=1
mA
I=5 mA
1=10 mA
I=20 mA
I=50 mA
-5 0
2.6
Frequency[GHz]
c
2.4
2 .2
F ig u r e
6 .6
2 .8
A reflection attenuator using Alpha PIN 1304
(without pre-distortion circuit), a)A ttenuation,
b) Return Loss, c) Phase shift
122
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
m odulator is required. T his requirement is critical when a filtered baseband
signal is used as the atten u ato r control signal [112],[113]. The nonlinear
relation between the forward bias current (in mA) and attenuation (in dB)
is shown in Fig.6.7.b. A pre-distortion circuit has to be used to achieve a
linear relation between the control voltage and attenuation. For the 25 dB
dynamic range of the atten u ato r a seventh order polynomial provides the
required predistortion. T he required predistortion characteristics are shown
in Fig.6.7.c. After the linearization the attenuation variation as a function
of control voltage is shown in Fig.6.7.d.
R eflection P h a se S h ifter
Due to parasitic elements such as a junction capacitor and a package
inductor, the phase shift of the reflection atten u ato r varies with the bias
current. The phase shift for a reflection attenuator using Id as a param eter
is shown in Fig. 6 .6 .c. This phase shift has to be compensated. In Fig.6 .1 ,
a reflection phase shifter was included to compensate for the phase error
associated with different levels of attenuation. The phase shifter was realized
using a Lange coupler and varactor diode term inations.
The phase shifter uses a Lange coupler as a hybrid and two varactor
diodes ( SMV 120412 from Alpha industries) [106]. T he phase shift, insertion
loss, and return loss of th e phase shifter are shown in Fig. 6 . 8 as function of
frequency. As may be seen from Fig. 6 .8 , the attenuation level of the phase
shifter is bounded between —.5 ±
.1
dB. The return loss is also within an
acceptable range.
The phase shift as a function of control voltage is shown in Fig.6.9. As
may be seen, the phase shifter has an almost linear phase relation (with
accuracy ± 2 degree) w ith control voltage. This makes it quite suitable to
compensate the phase error associated with different levels of attenuation.
123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
M easurem ent R esults
C urve fitting
10
'
10
•15
10
< -2 0
•25
**
10
10
>F' o r w a r d b ia10s c u rre n t[m A 10
]
10
10
'
-3 0 ,
10
15
20
F o rw a rd b ia s c u rre n tim A !
25
30
25
D ata
Polynom ial fit
20
15
•15
10
< -2 0
•25
0 .4
0.6
C o n tr o l V o lta g e fV o U l
F ig u re 6.7
-3 0 ,
0.8
0.2
0.4
0.6
C o n tro l V oltage(V oltJ
a)PIN diode resistance versus forward bias cur­
rent, b)A ttenuation of reflection attenuator ver­
sus PIN diode bias current, c)The required pre­
distortion characteristics d) Attenuation versus
control voltage
124
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.8
30
Phase[degree]
ES— ---->
*— ----#•
50
2.2
2.4
r
T
2
2.2
Return Loss[dB]
b
2.6
P— --- E>
B--- --- EI
x--- --- K
H---- ----f----ir
v=o
V=4
V=8
V=10
V=12
3
30
2.4
2 .6
Frequency[GHz]
2 .2
v=o
V=4
V=8
V=10
V=12
T
2.4
40
0— --- >
*— ----*■
----K
4---- ---- !•ft --- if
2 .8
Attenuation[dB]
0
2.6
H-------- tif--- — if
v= c
V=4
V=8
V=10
V=12
2 .8
c
F ig u re
6 .8
A
reflection
phase
tor SMV120412
from
shifter
using
varac­
Alpha a)Phase shift
b) Insertion Loss c) R eturn loss
125
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
■o
-*
M easurement
Linear Approximation
a 15
10
12
Control Voltage [Volt]
F ig u re 6.9
Phase shift of SMV120412 as a function of con­
trol voltage
6 .1 .6
Pow er C om biner
To combine the inphase and quadrature path signals, an inphase power
combiner is needed. A Wilkinson power combiner was used for this purpose.
The layout of such a combiner is shown in Fig. 6 .1 .a. The performance of this
power combiner is presented in Fig.6.10 as well.
6 .1 .7
Q A M M o d u lator
An MIC version of a direct microwave QAM m odulator a t a center fre­
quency of 2.5 GHz and bandwidth 500 MHz was realized using the above
circuits. The I and Q paths attenuation is shown in Fig.6 . 1 1 . As may be
seen from the results in Fig.6.1I, a high performance and broadband M-QAM
m odulator was realized using these circuits. The number of levels is selected
by the control voltage, Vc. The constellation diagram of the direct microwave
QAM m odulator over a 500 MHz bandw idth is presented in Fig.6.12 for the
126
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-2
m
~g.
m
CO
-*
-o
S12[dB]
S13[dB]
_ 3 ,
08
CM
CO
-4
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
3
-1 0
S11[dB] _
Y“
co
-4 0
-5 0
2.1
2.2
2.3
2.4
2.5
2.7
2.9
T
■ *
00
00
2 .8
T
T
<S12 & <S13
2.6
b
■o
<S12[degree'
<S13[degree'
,
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
Frequency[GHz]
c
F ig u r e 6.10 A Wilkinson power combiner a)Insertion Loss
b)Return loss c)Phase shift
127
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.9
3
I-
~T~
6
-B-
8
10
-1-Q Path-Att.-irv4QAM
12
14
16
18
2^5
6
CQ
2, 8
c*
_o 1 0
12
m
3
C -14
©
2.35
2.4
2.45
2.5
2.55
2.65
2.7
2.6
FrequencyfGHz]
- - -----1---------- 1—
I----------1■ ■ “ r" m
i
"i
--- »y . . --- - r ••.
---...........
W
—’------- t?----
2.75
---- *----- -----:--------- *---------- ------- --- »---------- ---------- ¥----------
5 16
18
2.25
J-Q Path AK. in 16QAM
__I---------- 1_.
2.3
2.35
1.
2.4
_
1
2.45
)-Q Path XtL in 64QtAM
m
■n 8
c 10
o
12
(0
3
2.3
. _I
2.5
1
1
1
1
2.55
2.6
2.65
2.7
! --------!----
’1---------~ 0"
2.75
1
------- —
— *----- -----:--------- *---------- :------ — *--------------- ----- $----------
C 14
©
tS 16
<
-18
225
. . A.—^ ---- --------- A---------- 1.......
?
1
T
i
2.3
2.35
2.4
2.45
A............'. ...
__l
2.5
2.55
0"
-------^
—
f----------r- . ---- r I-----^—
2.6
2.65
2.7
2.75
Frequency[GHz]
F ig u r e
6 .1 1
I-Q p a th attenuation for M-QAM M odulation
a)4QAM b)16QAM c)64QAM
4QAM, 16QAM, and 64QAM cases. An ideal dem odulator was assumed for
obtaining these constellation diagram . As may be seen from these diagrams
the vector error m agnitude is quite low and the proposed m ethod results in
a high perform ance m odulator.
6.2
D ir e c t M icrow ave M -Q A M Im p le m e n ta tio n U sin g
S i-M M IC T ech n iq u e
As stated before, a cost effective realization is an im portant design ob­
jective for the direct microwave QAM m odulator. This objective can be
realized by using a monolithic microwave integrated circuit (MMIC) imple­
mentation- T he selection of particular MMIC technology is also very im128
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
..—
—
........
.............................
...... ;......
\
3
............- %
.........
7
..........
-% ......... 7 — i r ........
■<?-............
............... d o ........
...............
2
t
...................................f •
- is - —
jfi ............: .
................
...............
-3
-4
>
jp .............[ .
>
-2
-3
.......................
%b-
b
I
%
•y
i
-
3
4
3
<f
% 1 8
t
8
*
I s
&
9
*
\ *
6»
*
*
:
0*
8»
:
4
oa
:
6
^
!
%
%
*>
\
:
4
s
s
*D
:
«0
:
<to
:
*
I *
*
%
*
6
%
%
i
%
o*
8
%
%
:
^
**
%
%
!
^
-2
-8
*
! \
-6
j *
-4
|
#
|
i
------------- i_________j—
- 8 - 6
- 4 - 2
:
8
: ^
»
C
2
i
6
8
F ig u re 6.12 Constellation diagram for the direct M-QAM
m odulator a) 4QAM modulator,
b) 16QAM
modulator, c) 64QAM modulator
129
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Transmission Line
Silicon Substrate
Ground Plane
F ig u re 6.13 The side view of a shunt PIN diode in Si-MMIC
portant.
The GaAs MMIC technology has been commonly used due its
excellent microwave circuit characteristics. On the other hand, Si-MMIC is
an evolving technology th a t promises low cost implementation of microwave
and millimeter circuits [114], [115]. The direct microwave QAM m odulator
has an architecture th a t is quite suitable for this technology as it is easy to
design a shunt PIN diode in Si-MMIC technology. This section presents an
Si-MMIC realization of the m odulator at 18 GHz.
6.2.1
P I N D io d e in S i-M M IC T ech n ology
The side view of a grounded PIN diode realization using Si-MMIC is
shown in Fig.6.13. As an example, the substrate thickness was selected as
100 iim with an I region width of 96 frm. For a 50 Q microstrip line on a
h = 1 0 0 iim thick silicon substrate, the first hybrid cut off mode, / ^ e , , is
200 GHz [116]. Using Eq.(5.73), the PIN diode resistance is presented in
Fig.6.14. As discussed in the last chapter, the results of this equation are
accurate and can be efficiently used for Si-MMIC circuit designs.
6.2.2
S u b sy ste m D e s ig n on S i-M M IC T ech n o lo g y
Lange C oupler
Two im portant param eters concerning the circuit design using Si-MMIC
are perm ittivity and loss tangent tan5. The measured values for silicon are
er = 11.68(±0.7percent), and tan5 = 1.3 x 10-3 (±30perceni). W ith this
130
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
r *v : "rrrr:..... '*•:* ; : ..........■
•!I!'!!:iE!r!!!rlH!nYH4i!:E::r!^
====== : : n : i : ; ^ u i H : : : : : I : : ; ; : ; : ; Z:<Z±Z:?SjJ3:: J::; -31
::: IZ: :: I : u; '::L ;." i
“HiililElliHEiHHI-HiilHnHiEiEHiHHIillinElii
^ 1 0
(0
:
;_
:i
li :i
;
_ ;_ _ l
;
:
|313 i
; : : :: rr
10
;
;
i : i : :: I : Z: l\11ZT:
r ; : : ;:i
10
Forward CurrentfAmp]
F ig u re 6 .1 4 PIN diode resistance as a function of forward
bias current in Si-MMIC
1------------r
“i------------r
• --------- «
Q---------O
-1
S14[dB]
S12[dB]
,-2
cn
* -3
u* a s , a - a - e «
t
I
18.5
19
_J______________I
16
16.5
17
17.5
18
19.5
20
a
-10
-i------------r
*------ *
Q------ o
-20
co
co
1
oS -30,
« t—*—«—* «—»—*-- «—*U' U~ 71' O -ft U'; 0 0 ...p...Q-.T®1
-o
-6
S11fdB]
S13[dB]
-- 0 -- B" -a—9—ii
co
-40
-50
16
16.5
_l_______ L.
19
19.5
_I_______ I_______ I_______ L_
17
17.5
18
18.5
Frequency[GHz]
20
F ig u re 6.15 A Lange coupler in Si-MMIC implementation a)
S 1 2 and
5
x4 [dB] b) 5 n and
5
x3 [dB]
131
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CM
S34[dB]
S24[dB]
O
CO-2
» g- g-
;
"0
oS—4
w
-6
16
16.5
17
17.5
13
18.5
19
19.5
20
-1 0
S44
dB -o
W -40
-50
16.5
18.5
17 5
T
T
19.5
b
T
-*
■o
<S24 degree
<S34 degree
m_10 Olr~s—e16
16.5
17
17.5
18
18.5
Frequency[GHz]
c
19
19.5
20
F ig u re 6 .1 6 A Rat-race coupler in Si-MMIC implementation
a) S3 4 and S 2 4 [dB] b) 5 4 4 and Su [dB] c) S2 4
and
534
[degree]
substrate the m easured attenuation of 50 Q microstrip line is 0.6 dB/cm . A
m icrostrip Lange coupler was designed and optimised on a silicon substrate
with a 100 /zm thickness. The w idth of fingers, their gaps, and coupler length
are W =7.88 fj.m, S=7.12 /zm, and L=1504 /jm, respectively. The results are
presented in Fig.6.15. As can be seen, a high performance coupler over 4
GHz bandw idth is realized using this technology.
R in g C o u p le r
A ring coupler was also designed and optimised for a silicon substrate.
The design m ethod is similar to the previous rat-race coupler and the dimen­
sions are W l= 7 8 .6 /im for the 50 Q lines, W =30.86 iim for 70.7 f2 line, and
radius of the ring R=1514 /zm. The results are presented in Fig.6.16. As
may be seen, the ring coupler operates as a high performance 180° coupler
132
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
over a wide bandwidth. If the miniaturisation of the circuit is desired, a
lumped-element equivalent circuit or a reduced-size technique can be used
instead of distributed implementation [58],[117],[118]. However, using these
techniques decreases the bandwidth of the coupler.
S P D T S w itch on Si-M M IC
As discussed, a shunt PIN diode can be easily implemented in Si-MMIC
technology. The implementation of SPDT switch using shunt diodes is shown
in Fig.6.17 [63]. A current 1=30 mA (corresponding to .4
forward bias
resistance) was selected to bias the diode. A simple equivalent circuit, shown
in Fig.6.17, is suitable as a model for Si-MMIC PIN diode. Due to high
relative dielectric constant for silicon, the fringing capacitance (in air) around
the I region is relatively small, and the capacitance calculated using the
parallel plate capacitance formula below provides a useful estimate of the
junction capacitance C j.
where A is the junction area. For a square junction, A = 78.(rfim 2, resulting
in a C j = 6.7f F . The insertion loss, isolation, as well as the return loss of
shunt SPD T switch are shown in Fig.6.17. As may be seen, high performance
SPDT switch over a very wide bandwidth is achieved.
R eflection A tte n u a to r U sin g Si-M M IC T echnology
By using Eq.(6.5) and the estimated Rd from Fig.6.14, a reflection atten­
uator can be realized in Si-MMIC technology. As may be seen from Fig.6.18,
a given attenuation value can be realized using two bias current values th at
correspond to PIN diode resistance being smaller or bigger than 50 Q. To
avoid ambiguity, the bias current was limited from .3 mA to
6
mA. As may
be seen in Fig.6.18, this current interval provides a 24 dB dynamic range for
the attenuator.
133
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ZO'—
16
-’
-
16.5
17
17.5
_
.
18
-
18.5
-
19
19.5
20
Frequency[GHz]
e
F ig u re 6.17 An SPD T switch in Si-MMIC implementation
a)Switch topology b)P in diode equivalent circuit
c) Insertion loss d) Isolation e) R eturn loss of
switch
134
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
As for the MIC realization, a predistortion is required to linearise the
control voltage versus attenuation.
straightforward.
limited to the .3
The predistortion realization is quite
As may be seen from Fig.6.14, when the bias current is
mA to
6
mA interval, the PIN diode forward bias resistance
is always less than 50 Q. Then, Eq.( 6 . 2 ) may be written as:
IptI = Zz ~Jr '
(6'8)
or
= Z(
i
-J ^ | )
1
+
|p t |
'
Defining control voltage for an ideal diode as the interval .05 < Vc < .9 V,
one may develop a predistortion relation assuming \pj-\ = Vc. The PIN diode
resistance could be approxim ated as [1 0 2 ],
W2
R* = w
r d
'61°)
A simple relation between Vc and Id is obtained as
r
1+^c
rd = o2 —
/ i t Z a( l—
- tTTvc)
(6-n J
This predistortion was used for the implementation of the m odulator. Its
performance is reported in the next chapter.
W ilk in son P ow er C om b in er in S i-M M IC
To provide an inphase power combining, a single section Wilkinson power
combiner was designed and optimised w ith a center frequency of 18 GHz and
a bandwidth of 4 GHz. Similar to the MIC version, the combiner consists of
a 50 Q line input and output m icrostrip lines (78.6 fim wide), two uncoupled
70.7
jim lines(30.85 [jlin wide), and a 100 Q. isolation resistance. T hin film
135
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-5
S -10
-15
-2 0
-25 L
10
'
10
“
10
10
10
10
‘
Forward bias current[mA]
F ig u re 6.18 Attenuation of reflection attenuator as a func­
tion of bias current
----------------1--------------- r
.......
“ 1■
--------- 1
1
!
1
*---------*
g -------- 0
Sl2[dB]
S13[dB]
1
16
16.5
17
175
18
18.5
19
19.5
20
-10
S 11[dB]
-40
-50
16.5
T
b
T
■*
100
16
19.5
18.5
17.5
o
16.5
17
17.5
18
18.5
19
<S12 degree
<S13 degree
19.5
20
Frequency[GHz]
C
F ig u re 6.19 A Wilkinson combiner in Si-MMIC implementa­
tion a) Insertion loss, S 12 and
£13
[dB] b) Return
loss, S n [dB] c) Phase shift, S 1 2 and S 1 3 [degree]
136
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
resistors have a resistance given by [115]:
R
= ~Ps
w
( 6 .1 2 )
where I is the length of the resistor, w is the w idth of the resistor, and ps is
the sheet resistance of resistive material in f2/square. For Si-MMIC, ps = 342
fi/square results in a width of 192 p m for a length of 56 pm . As may be
seen from Fig.6.19, the insertion and return loss in two arms are equal.
M -Q A M M odulator U sin g Si-M M IC
In addition to its lower cost, a Si-MMIC M-QAM modulator offers a
number of performance advantages as well [1 2 ]. The parasitic and package
elements have low values and the junction capacitor has a very small value as
well. These characteristics result in low phase shift variation in the multilevel
attenuator. Moreover, the phase variation for a single frequency operation
can be compensated by using open circuit stubs. This method is discussed
in Appendix B. Thus, the phase compensation circuit is not needed provided
th a t the operation is limited to a narrow band operation.
A direct microwave QAM modulator was implemented using Si-MMIC
technology [1 2 ]. As before, the modulation was generated using polarity
m odulation and attenuation control in the I and Q paths. The layouts of
the polarity modulator section and the attenuator section are presented in
Figures 6.20 and 6.21. The bias circuit design is described in Appendix C.
T he performance of the modulator is presented in the next chapter.
As stated before, a distributed im plem entation of the couplers and power
combiner have been used to realize the m odulator. This technique results
in a higher bandwidth. A lumped-element equivalent circuits may be used
137
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to m iniaturise th e couplers and power combiner. However, the design of the
lumped-element circuits must be somewhat empirical, and it needs precise
inductor models based on a specific foundxy process. Moreover, the design
becomes difficult at frequencies above 20 GHz [118].
6.3
Sum m ary
In this chapter microwave integrated circuit (MIC) and silicon monolithic
microwave integrated circuit (Si-MMIC) im plem entations of the direct mi­
crowave QAM m odulator were presented. The MIC implementation uses
microstrip and co-planar waveguides as well as discrete silicon PIN diodes.
The constellation study over the 500 MHz bandwidth shows an error vector
magnitude b e tte r than 10 percent for the MIC m odulator. The Si-MMIC
implementation uses microstrip line. PIN diodes were realized in silicon sub­
strate. W hile realization using the MIC is simple and straightforward, the
Si-MMIC realization offers a low cost and high perform ance modulator.
138
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
F ig u re 6.20 Layout of the polarity modulator section of di­
rect microwave QAM modulator
139
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
F ig u re 6.21 Layout of the attenuator section of direct mi­
crowave QAM m odulator
140
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7.
PE R FO R M A N C E OF FIX ED
W IRELESS ATM TR A N SM IT T E R
The architecture and hardware for an adaptive direct microwave M-QAM
m odulator was described in th e last two chapters. In this chapter the opera­
tion of the radio ATM transm itter for different wireless channels and under
various traffic loads is examined. The Ricean and Rayleigh wireless channels
are considered. The traffic sources are VBR computer data and video traffic.
7.1
A d a p tiv e D ir e c t M icrow ave M -Q A M T ran sm itter
for R ad io A T M
A model for the adaptive direct microwave QAM modulator is presented
in Fig.7.1. As may be seen, this transm itter consists of two units. The MQAM control unit is used to provide an accurate estimate of the required
bandwidth in a wireless channel. The direct microwave QAM modulator
unit transm its the m ultim edia traffic through a wireless network. A control
signal from the M-QAM control unit to the m odulator unit assures th at
an optimum response to the bandwidth demand can be accommodated in
a wireless channel. A detailed architecture for the transm itter is shown in
Fig.7.2. Different sections of this modulator were described in the previous
chapters. In this chapter the functionality as well as performance of different
sections are examined.
7.1.1
A d v a n ta g es o f u sin g E ffectiv e B a n d w id th M etric
The effective bandw idth estimation is done in the M-QAM control unit.
Using the effective bandw idth metric, a statistical model to estimate the re-
141
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
srnmm
.^Switch
SGoiiirbll*
©nects^s
■Microwave
i>
w m im m
Wireless
MQAMrV
Channel
Demodulator
*•:?••■;•t'/'r.
Feedback
Channel
F ig u re 7.1
;.CtianneLf?^
iE s tim a td r ^
The adaptive direct QAM m odulator in a fixed
wireless ATM system
quired bandwidth was used. This model is a real tim e model, th a t provides a
close upper bound estimation for various traffic. The model was discussed in
Chapters 3 and 4. The accuracy of this model is evident from Fig.7.3.a where
the actual required bandwidth of the S ta rW a rs movie is upper bounded with
its effective bandwidth using buffer size as a param eter. The advantage of
using the effective bandwidth instead of the peak rate bandw idth for different
videotraces is quite clear from Fig.7.3.b for cell loss ratio 10-6 . As can be
seen from this figure, a bandwidth saving between 25 to 40 percent for video
transmission can be easily achieved. The saving percentage is a function of
the buffer size. Figure 7.3.c compares the actual required bandwidth of the
aggregate of three MPEG-1 coded movies (using Table 3.1) with its effective
bandwidth. As can be seen from Fig.7.3.b and 7.3.c, using the effective band­
width also shows a statistical multiplexing gain for aggregate video traffic.
For instance, while the effective bandwidth of the videotraces are 760, 590,
and 680 Kbps for a buffer size 350 Kbits, respectively, their aggregate traffic
requires only 1500 Kbps bandwidth using
1
Mbits buffer size. This shows a
statistical multiplexing gain of about 1.35. The peak ra te and effective band­
width are compared for aggregate video traffic in Fig.7.3.d. As may be seen
the peak rate requires about 3 Mbps capacity, while the effective bandwidth
reduces it to around
1 .5
Mbps. These results show th a t a high bandwidth
saving can be achieved using the effective bandw idth m etric. It should be
pointed out th a t this metric provides an accurate as well as a simple solution
142
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ConnectionsfVCs)
Accept the Call
i
I M
r
Do not accept the call
Delay Parameters
Connection Gate Control
Buffer Size
Managaer
i
Vp
Effective BiptfwkXh
H *idu.(H ^) . R —i-
ER
R<m—»I R_
SR„,°. ’
.
o.
R
_H=m-
BW
a c ttl
Vp
VP( Effective BaadwtdSi
*1 R
BW£ffi_
BW
E MQAM
E le ctiv e B utfw xfth
H«q»(^) . R^^ b Z R^
« >1
BW
BW
; MQAM
, MQAM
Estimation
?
MQAM
CNR
(1 + a ) Z B W m d f i
No
W
log ,(M/2)= log . (M)
log , (2M)= log , (M)
Yes,
No
(1+ CC.)
log
W racffl
w
Yes
Outage
O.K.?
MQAM Control
Threshold
©
Ring
Predistortion
Sc Pre filtering
Variable
SPDT
Attenuator
Switch
Variable
Phase
Shifter
Power
Lange
Combiner
Coupler
Variable
Ring
Coupler
Switch
A tte n u a to r
Shifter
Predistortion
Sc Prefiltering
Threshold
F ig u re 7.2
variable
An adaptive direct QAM m odulator for wireless
ATM
143
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to a very complicated bandwidth estimation problem.
As pointed out before, besides the different bandwidth requirements for
various traffic sources, an extra factor arises due to the wireless network. As
discussed in C hapter 4, a wireless channel capacity decreases due to fading.
A capacity reduction factor due to fading was introduced to take this into
account. A new metric, termed the modified effective bandwidth, was in­
troduced to include the effect of channel performance on capacity allocation
policy. Estim ation of this metric is the main task of the M-QAM control
unit. Fig.7.3.d compares the modified effective bandwidth with peak rate
and effective bandwidth for aggregate video traffic for cell loss ratio 10-6 in
Ricean channel. As may be seen, using the modified effective bandwidth, a
higher bandwidth has to be reserved. This results a guaranteed QoS in the
wireless channel.
7.2
A d a p tiv e M -Q A M P erform an ce
In the previous section it was shown th at using effective bandwidth in­
stead of peak rate allocation highly saves the bandwidth. In this section, a
simulation study is conducted to evaluate the performance of the adaptive
M-QAM m odulator. The M-QAM control unit uses modified effective band­
width to adapt the M-QAM modulator. The objective of the study is to
evaluate the advantages of using an adaptive structure over a conventional
fixed QAM m odulator. The adaptation was realized using an M-QAM con­
trol unit. The operation of the M-QAM control unit is summarised as follows.
According to Fig.7.2, the ATM connections introduce their traffic and
QoS param eters to the gate control port of the M-QAM control unit. Ac­
cording to their V P addresses, the connections are directed to different V P s.
Then, the traffic characteristics of these VPs are recalculated to include the
144
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1200
•----- > B=50KBitExp.
o — O B=100 KBit Exp.
---«
B=200 KBit Exp.
■-------
8=50 Kbit Ana.
o— o
B=100 KBit Ana.
1100
-
I
j
800-
I
1
600500-
200
300
400
500
600
700
800
900
400
Effective BandwidthfKbps!
(a)
(b)
2600
B=1 MBit Ana.
2400
KBit Exp.
I
j
2000 -
I
1600
1400
1200
400
600
600
1000
1200
1400
1600
1800
2000
Peak rate
Effective Bandwidth
Modified Effective Bandwidth
1000
Effective BendwidthlKbpsI
(C)
(d )
F ig u re 7.3 a) Experimental estimation versus effective
bandw idth with buffer size as a param eter
for S ta r W a r s movie.
b) Effective band­
width and peak rates of different video traces
c)Experim ental estimation versus effective band­
width with buffer size as a param eter for ag­
gregate video. d)The peak bandw idth, effective
bandwidth, and modified effective bandw idth for
aggregate video traffic. The cell loss ratio is 10~6
in a Ricean channel
145
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
characteristics of the new connections. The buffer sizes are selected by the
buffer size manager according to delay constraints. This is followed by an
estimation of the effective bandw idth and capacity reduction factor. Using
these param eters, the modified effective bandw idth metric is calculated. A
new m odulation level is considered by the M-QAM control unit if the new
level of m odulation satisfies the outage conditions. This is used to adjust the
level of the direct microwave M-QAM.
7.2 .1
L A N Traffic
In the first experiment, a large number of ATM connections, with LAN
traffic, are applied sequentially to an adaptive transm itter. The system uses
three VPs. A connection should be directed to a XT according to its V P ad­
dress. The connection selects a particular VP randomly in this experiment.
Although the delay is not a critical QoS param eter for data transmission,
a maximum delay bound is used during sim ulation to perform a buffer as­
signment policy. Each call m ust introduce its traffic as well as its QoS pa­
rameters. The required traffic param eters in this experiment are mean rate,
variance coefficient and H urst param eter. The QoS parameters are cell loss
ratio and maximum tolerable delay. These param eters are generated using
the MATLAB random num ber generator. The range for different param eters
are limited to: 1 < Rmean < 10 Mbps, 105 < a < 106 bit-sec , .5 < H < 1,
HU6 <
Pc l r
< 10-3 , and .001 < D max < .01 sec.
A VP initially carries traffic having a mean rate between 1 kbps and 15
Mbps, using a buffer size between 100 kbits and 1 M bits, and having vari­
ance coefficient, a, between 104 and 5 x 103 bit-sec, respectively. The Hurst
param eters of VPs is also random between .5 and 1. When a new LAN con­
nection is applied to system, the initial values for VPs are regenerated. The
channel is considered with burst errors. The burst width of the channel is a
uniform random variable between 2 and 10.
146
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The traffic is to be transm itted through a LOS radio link. The radio
link discussed in Section 2.8 was used in the simulation. The link design
param eters, presented in Table 2.2, were used in the outage block. As a
result of the traffic varieties of the ATM calls, th e network requires a variable
bandwidth. The M-QAM control unit estim ates this demand by using the
modified effective bandwidth metric. This dem and is satisfied by M-QAM
level variation. The feasibility of using a new modulation level is examined
by the outage block. According to bandwidth demand, the constellation size
is a variable between M = 2 and M = 256. This results in a constellation
size k of 2 to 8. The objective of simulation is to extract the distribution of
the effective bandwidth, B W e/ f , the modified effective bandwidth, B W mej f ,
the constellation size k , and the outage. The experiments are conducted for
Ricean and Rayleigh channels.
O p era tio n in R icean F ading
A large num ber of ATM calls are sent to the M-QAM control unit. The
constellation size, k , outage, B W ej j , and B W mef j variations are shown in
Fig.7.4.a. As expected, the constellation size, outage, effective bandwidth,
and modified effective bandw idth change with the bandwidth demand. The
probability of using any of the constellation sizes is presented in Fig.7.4.b.
As may be seen, the traffic can be often transm itted using a constellation
size k = 2 and k = 3. A 256-QAM(fc = 8) would be used if a conventional
design had to be selected. T he average and standard deviation of different
param eters are shown in Table 7.1. These values only have statistical mean­
ings. Fig.7.5 shows the acceptance improvement compared over th a t for the
different fixed QAM m odulators. As may be seen, in a Ricean channel a
good improvement compared with the fixed QAM can be achieved. The
outage perform ance of an M-QAM is also compared with the different fixed
147
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
QAM m odulators in Fig.7.6. As may be seen, while a slight outage deficiency
is tolerated compared with a low level fixed QAM, the adaptive modulator
offers considerable performance advantage over the fixed high level QAM
m odulators.
T a b le 7.1
Average and standard deviation for modulation
level, outage, effective bandwidth, and modified
________________effective bandwidth for Ricean Channel__________
M odulation l e v e l
Outage
BWe[Mbps]
BWem[Mbps\
A verage
9 .9 6 3 2
.00125
5 6 .1 6 6
6 0 .8 5 4 5
std
1 8 .2 7 7 3
.00255
2 2 .0 2 8
2 6 .0 0 2 2
O p e r a tio n in R a y le ig h F a d in g
Although a Ricean channel model is valid for a LOS link and a fixed
wireless ATM link must be studied under this category, a performance study
under Rayleigh channel conditions has some use, e.g., for comparison with
the Ricean channel or expansion of the model to land mobile communica­
tions. Fig.7.4.c shows the constellation size, outage, effective bandwidth and
modified effective bandwidth variations for Rayleigh channel. The constel­
lation size distribution for the Rayleigh channel is shown in Fig.7.4.d. The
average and standard deviation of these parameters are also shown in Table.7.2. As can be seen, compared with the Ricean channel, while the effective
bandw idth is almost the same (traffic characteristics are independent from
the channel), the constellation size and modified effective bandwidth have a
higher average and variance. This behaviour is predictable due to the nature
of the Rayleigh channel.
Moreover, by comparing Table 7.1 and 7.2, one can see that the average
outage in th e Rayleigh channel has been increased. The call acceptance per148
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
formance is shown in Fig.7.5. As can be seen, an M-QAM generally increases
the traffic acceptance in different channels; however, a higher performance
can be seen in Rayleigh fading. This is due to higher channel capacity varia­
tion in Rayleigh fading. The outage performance for the Rayleigh channel is
also compared with the different fixed QAM m odulators in Fig.7.6. The su­
perior outage performance using the adaptive M-QAM m odulator is evident.
T a b le 7.2
Average and standard deviation for m odulation
level, outage, effective bandwidth, and modified
effective bandwidth for Rayleigh Channel_____
M o d u la tio n l e v e l
Outage
BWe[Mbps]
BWemiMbps]
Average
1 3 .5 9 4 4
.0 0 1 8
5 7 .1 3 0 7
6 8 .6 7 1 5
std
2 8 .8 9 9 3
.0 0 4 0
2 2 .5 5 5 8
2 8 .6 4 8 7
7 .2.2
V B R M P E G V id eo Traffic
In the next experiment, a number of MPEG-1 encoded videos are used as
the traffic sources. These video sources and their traffic characteristics are
shown in Table 7.3. [84], [119]. The number of video sources simultaneously
directed to a VP was a random number between 0 and 30, and the number
of copies from a single video is between 0 and 5. As in the previous example,
the QoS param eters
P
c l r
and Dmax, were obtained using a random number
generator with range 1 <D max < 10 msec, and 10-9 < p c l r < 10~6. Again,
three VPs, each initially carrying traffic with a mean bit rate between 100
kbits and 15 M b its, a buffer size between 100 kbits and 1 M bits and a
variance coefficient between 104 and 5 x 105 bit-sec were considered. The
Hurst param eter of V Ps was also random between .75 and .95. The channel is
considered to be a bursty channel with the burst width uniformly distributed
between 2 and 10. The radio link characteristics were similar to the link
studied in Section 2.8. The outage block examines the link with param eters
listed in Table 2.2 to estimate the outage probability.
According to the
149
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.5
Constellation Distribution for Ricean Channel
0.4
500
1000
1500
Ja 0.3
02
0.1
5*400
§200
O
$s „o
o.
500
1000
4 _
6
Constellation Size
1500
(a)
(b)
j
Outage%
3 -------------
-2
Aa
5 ------------- —
- ...........: -----
2 ......-.........
!
0
(c)
F ig u r e 7.4
BayieighBistriboaticn------
3
UuU razk
BWem
BWe[Mbps]
i
2
i
4
6
Constellation Size
i
(d)
a) Simulation results for M-QAM in a Ricean
channel for LAN traffic based on th e number of
connections b) Constellation size distribution in
a Ricean channel for LAN traffic c) Simulation
results for M-QAM in a Rayleigh channel for
LAN traffic based on th e number of connections
d) Constellation size distribution in a Rayleigh
channel for LAN traffic
150
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
100,
Ricean
Rayleigh
90
80
S 70
> 60
5 40
30
20
Constellation. Size
F ig u re 7.5
ATM call acceptance improvement compared
with fixed QAM modulator in Ricean and
Rayleigh channels for computer data traffic
20
x 10'
§ 10
CL
Ricean
Rayleigh
-5
Constellation Size
F ig u re 7.6
The outage improvement compared with fixed
QAM in Rayleigh and Ricean channels for com­
puter d a ta traffic
151
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
bandwidth demand, the constellation size can be varied between M = 2 and
M = 256.
V a r.
C o e f f .[ b - s ]
Hurst
.326950
4 .9 2 7 7 e 4
.8448
S ile n c e o f th e Lambs
.182788
6 .4 8 4 3 e 4
.8959
S ta r Wars
.232830
7 . 25e4
.8458
T erm inator I I
.272625
2 .7 3 3 4 e 4
.89
Mr.Bean
.441175
1 .0 8 5 2 e 5
S o c c er
.678225
9 .0 1 7 3 e 4
.91
i
J u r a s s ic Park
00
tn
Mean[Mbps]
O p eration in R icea n Fading
A large number of ATM video connections were sent sequentially to the
M-QAM control unit. The number of videotraces in a call was limited to 30.
T he constellation size, outage, B W eff, and B W mef f variations are shown in
Fig.7.7.a. As may be seen, as bandwidth dem and changes, the constellation
size, outage, effective bandwidth, and modified effective bandwidth change.
However, the traffic variation for video are less than that of the LAN type
traffic. The probability to use any of the constellation sizes is presented in
Fig.7.7.b. The mean and variance of different param eters are shown in Table
7.4. Fig.7.8 shows the acceptance improvement compared with the fixed
QAM modulators. The outage performance of an M-QAM is also compared
w ith the different fixed QAM modulators in Fig.7.9. As may be seen, while
a slight outage probability has to be tolerated compared with th at of the low
level fixed QAM, the adaptive m odulator offers a much better performance
compared with th a t for the fixed high level QAM modulators.
152
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.9
0.8
• Constellation. DnthbutiaD<m-Rxcecxr Chmimef ••
0.7
2000
0.6
>»
= 0.5
j»e
=£0.4
2000
CL
0.3
0.2
2000
0.1
0
2
3
4
5
6
Constellation Size
2000
(b)
0.7
Bl Uli
200
400
600
1000
1200
Constellation Distribution in Rayleigh Distribuatico
II mill! IHI HIS
II111 M ill
III 11 111 IIIIIK
1400 1600
0.6
1800 2000
05
= 0.4
200
400
600
800
1000
1200
1400 1600
1800 2000
0.3
200
400
600
800
1000
1200
1400 1600
1800 2000
200
400
600
800
1000
1200
1400
1600
1600
2000
(c)
F ig u re 7 .7
(d)
a) Simulation results for M-QAM in a Ricean
channel for VBR video traffic based on the num­
ber of connections b) Constellation size distribu­
tion in a Ricean channel for VBR video traffic
c) Simulation results for M-QAM in a Rayleigh
channel for VBR video traffic based on the num­
ber of connections d) Constellation size distribu­
tion in a Rayleigh channel for VBR video traffic
153
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7
100
Ricean
Rayleigh
80
c« 70
'v
>
O 60
ucu
S
C5 40
cu
-tJ
S 30
20
10
1.5
2.5
3.5
4.5
Constellation Size
F ig u re 7.8
ATM call acceptance improvement comparison
of th e adaptive modulator and the fixed QAM
m odulator in Ricean and Rayleigh channel for
video traffic
x 10'
<D
O)
■*
Ricean
■o Rayleigh
-1
F ig u re 7.9
1.5
2.5
3.5
Constellation Size
4.5
T he outage improvement of the adaptive modu­
lato r compared with fixed QAM in Rayleigh and
Rice channels for video traffic
154
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
T a b le 7.4
Average and standard deviation for m odulation
level, outage, effective bandwidth, and modified
effective bandwidth for Ricean Channel for video
traffic______________________________________
Outage
M odulation l e v e l
BWe[Mbps] BWemiMbps]
A verage
4 .8 2 8 0
.0 0 0 5 3 5
4 0 .2 6 2
4 3 .5 8 4
std
2 .6 8 4
.0 0 0 3 7 5
1 1 .5 0 6
1 4 .1 7 2
O p e r a tio n in R a y le ig h F a d in g
The M-QAM control unit operation was also examined for the Rayleigh
fading channel for the video traffic. Fig.7.7.c shows the constellation size,
outage, effective bandwidth and modified effective bandwidth variation of
dynamic bandwidth demand. The mean and variance of these param eters are
also shown in Table.7.5. Comparing Table 7.4 and 7.5, one can see th a t the
average outage in the Rayleigh channel has been increased. The constellation
size distribution for the Rayleigh channel is shown in Fig.7.7.d. As can be
seen from Fig.7.7.d, a higher m odulation level is required to transm it similar
traffic in a Rayleigh fading. The call acceptance performance is shown in
Fig.7.8. The outage performance for the Rayleigh channel is also compared
with the different fixed QAM m odulators in Fig.7.9. The outage performance
in the high level QAM is evident.
T a b le 7.5
Average and standard deviation for m odulation
level, outage, effective bandwidth, and modi­
fied effective bandw idth for Rayleigh Channel for
M odulation l e v e l
Outage
B\ie[Mbps]
BWem[Mbps]
A verage
5 .4 8 8
.0 0 0 6 2 7
4 0 .6 7 2 5
4 8 .2 5 9 3
std
2 .8 3 1
.0 0 0 3 9 6
1 1 .2 6 6 4
1 4 .4 7 1 5
155
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7.2.3
M u ltip le Sources
In the previous experiments, a single ATM call was directed to a VP
at a time. A general model is to apply the multiple inputs to each VP.
The performance of the M-QAM control unit in Ricean and Rayleigh fading
channels for Ethernet traffic and video traffic is examined in this section.
M u ltip le L A N Traffic Sources
In the first experiment, a number of E thernet sources were simultane­
ously directed to the different VPs. T he number of connections for each
VP was between 1 and 4.
The range for different parameters are lim­
ited to: 2 < Rmtan < 5 Mbps, 105 < cr < 106 bit-sec , .5 < H < 1,
10-6 <
pclr
< 10-3 , and .001 < Dmax < .01 sec. There are three VPs and
each VP initially carry traffic having a mean rate between 1 kbps and 10
Mbps. The other traffic parameters of V Ps are similar to the previous exper­
iments. The experiments were done for Ricean and Rayleigh fading channels.
Fig.7.10.a shows the constellation size, outage, effective bandwidth and mod­
ified effective bandwidth variations in a Ricean channel. As can be seen, the
effective bandwidth and modified effective bandwidth variations can also be
seen in multiple source experiment. Fig. 7.10.b shows the constellation distri­
bution in a Ricean channel. As may be seen, the M-QAM control unit selects
the constellation size k = 4 with higher probability. Fig.7.10.c shows the con­
stellation size, outage, effective bandwidth and modified effective bandwidth
variations in a Rayleigh channel. A comparison between Fig.7.10.a and 7.10.C
shows a higher variation in Rayleigh channel compared with Ricean chan­
nel. Fig.7.10.d shows th a t the most common constellation size in Rayleigh
channel should be k =5 compared to k = 4 for Ricean channel.
156
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.45
0.4
200
400
600
BOO
1000
1200
1400
1600
1800
0.35
2000
0.3
= 0.25
£ 0.2
0.15
0.1
2*400
0.05
§•2
00
5<s CD
0.
500
2000
1500
1000
(b)
(a)
0.45
0.4
200
400
600
BOO
1000
1200 1400 1600
1800
0.35
2000
0.3
no
= 0.25
0
500
1000
1500
2000
I
£.100
0.2
0.15
....—
5 400
0.1
2000
1--
0.05
§200
9
m0
»
500
1000
i --------
1500
2000
(c)
(d)
F ig u re 7.10 a) Simulation results for M-QAM in a Ricean
channel
for
multiple
LAN
traffic
sources
b)Constellation size distribution in a Ricean
channel for multiple LAN traffic sources c) Sim­
ulation results for M-QAM in a Rayleigh channel
for m ultiple LAN traffic sources d) Constellation
size distribution in a Rayleigh channel for mul­
tiple LAN traffic sources
157
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Multiple Video Traffic Sources
An experim ent was also conducted to study the multiple video traffic
sources in Ricean and Rayleigh channels. In th is experiment, each VP was
requested to tran sm it between 0 and 30 MPEG-1 videotraces, and the num­
ber of copies from a single video is between 0 and 5. Table 7.3 shows the
traffic characteristics of various video traffic sources. The initial values of
VPs are sim ilar to previous example. Fig.7.11.a shows the constellation size,
outage, effective bandwidth and modified effective bandwidth variations in
a Ricean channel. As may be seen, there is less variation compared with
LAN traffic. T his is due to lower mean rate of video traces compared to
LAN traffic. Fig. 7.1 l.b shows the constellation distribution in a Ricean
channel for m ultiple video sources. Fig.7.11.c shows the constellation size,
outage, effective bandwidth and modified effective bandwidth variations in
a Rayleigh channel. A comparison between Fig.7.11.a and Fig.7.11.c shows
th at a higher bandw idth is required in Rayleigh fading. This can also be seen
by comparison between Fig.7.11.b and 7.11.d where operation in a Rayleigh
channel in average requires a higher m odulation level to transmit the similar
traffic. O ther interesting results can be seen from outage results. While two
experiments used sim ilar m odulator and same input traffic, a Ricean chan­
nel resulted th e b e tter outage performance compared with Rayleigh channel.
The average and standard deviation of the different parameters for multi­
ple video sources are shown in Table 7.6. As can be seen, while the effective
bandwidth is sim ilar in Rayleigh and Ricean channels due to same input traf­
fic, the average of the modulation level in Rayleigh channel is higher than
th at in Ricean channel. This is due to higher modified effective bandwidth
requirements in a Rayleigh channel, and is a function of the param eter K in
a Ricean m odel. This experiment shows th at a Rayleigh channel requires 12
percent more bandw idth compared to a Ricean model with K = 6.
158
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.8
200
400
GOO
800
1000
1200
1400
1600 1800 2000
0.7
0.6
200
400
600
800
1000 1200
1400
1600
1800 2000
0.4
0.3
200
400
600
800
1000
1200
1400
1600
1800 2000
•5.400
*
|200 kx-L i.i
Ia o
200
0.1
11. > .t*L
400
600
t iiiilliiil.lL. J-l»i Jl. I .iuri,.. ,i;«
800
1000
1200
1400
1600
1600 2000
(a)
(b)
0.5
0.45
200
10 p —
400
I ■ ■ - r -11
600
600
i
"r
1000
1200
1400
1600 1800 2000
-Onstettsticn-DathbutunTn'Rayie^gh Channel
0.4
.
0.3
200
400
600
000
1000 1200
1400
1600
1600 2000
|0 025
£ 0.2
0.15
200
400
600
800
1000
1200
1400
1600
1800 2000
7400j---- 1---- 1---- 1---- 1---- «---- 1---- «---- 1---- 1--S
2
:
l... ........I.....:.... If...
§
. . j t . i 2! I
200
0.1
0.05
i. .. j .
400
600
800
1000
1200
1400
1600 1800 2000
(c)
(d)
F ig u r e 7.11 a) Simulation results for M-QAM in a Ricean
channel
for
multiple
video
traffic
sources
b) Constellation size distribution in a Ricean
channel for multiple video traffic sources c) Sim­
ulation results for M-QAM in a Rayleigh channel
for multiple video traffic sources d) Constellation
size distribution in a Rayleigh channel for mul­
tiple video traffic sources
159
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
T a b le 7.6
Statistical Param eters for multiple MPEG-1
___________
Video Traffic in a Ricean Channel____________
M odulation l e v e l
Outage
BWe[Mbps]
BWem [Mbps]
M ean(R icean)
1 6 .8 0 5 2
.002208
7 6 .6 0 7 0
8 2 .2 0 1 8
s td (R ic e a n )
2 1 .9 6 7 9
.003069
2 3 .4 2 4 0
2 7 .8 1 4 3
Mean. (R a y le ig h )
2 2 .0 8 4 6
.00295
7 8 .2 0 7 0
9 2 .8 8 4 9
s td (R a y le ig h )
2 4 .0 0 6 9
.00335
2 3 .6 6 8 8
2 9 .3 3 1 0
7 .2 .4
D isc u ssio n o f R e su lts o f th e E x p e rim en ts
The previous experiments were conducted to study the performance of
the M-QAM m odulator for a fixed wireless broadband network. These ex­
periments presented the following results:
• The M-QAM control unit highly saves the bandwidth while providing
a guaranteed QoS in wireless networks.
• The M-QAM control unit provides a flexible transmission system th at
highly improves the call acceptance and outage condition in wireless
networks.
The advantages of using the effective bandwidth allocation compared with
the peak rate allocation were discussed in Section 7.1. It was found th at
a bandw idth saving between 25 to 40 percent for a single video MPEG1 trace can be easily achieved. This saving is close to 50 percent for the
aggregate of three video traces due to statistical multiplexing gain. In a
LAN type traffic, depends on utilisation and buffer size a high bandwidth
saving is also achievable. For example, a bandwidth saving b e tter th an 60
percent was obtained for Bellcore A ugust’89 Ethernet traces in C hapter 3.
Moreover, using M-QAM control unit results in a high acceptance and outage
improvement compared with conventional QAM modulators. As an example,
in Fig.7.4.a one sees th a t the demand to use a more than 100 Mbps in a
160
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ricean channel is very lim ited. Thus, an adaptive m odulator frequently
operates in lower m odulation level (e.g., k=2, and 3) and it improves the
system outage. The outage improvement is due to using a lower m odulation
level compared with a 256-QAM conventional choice. Moreover, the system
uses the high m odulation level when there is a high bandwidth demand. As
seen from Fig.7.5 and 7.6 as well as 7.8 and 7.9 a high acceptance and outage
improvement can be achieved by using the M-QAM control unit.
These
results are also valid for m ultiple traffic sources.
7 .3
A d a p tiv e D ir e c t M icrow ave M -Q A M m od u la to r
The second unit of the adaptive transm itter is a direct microwave QAM
m odulator. As shown in Fig.7.1, the modulation level of direct microwave
M-QAM m odulator is varied by using a control signal from the M-QAM
control unit. W hen bandw idth demand is low, a low level modulation, e.g.
4QAM, is used. The use of low level M-QAM improves the system gain
and decreases the outage. A low level modulation should also be used when
a high QoS is required. O n the other hand, when the bandwidth demand
is high, a high level direct microwave QAM can be used provided th a t the
required QoS can be met. T he simulated constellation diagrams for different
levels of QAM using two adaptive m odulator implementations are presented
below to illustrate the high adaptive performance over a broad bandwidth.
7 .3 .1
A d a p tiv e D ir e c t M icrow ave M -Q A M Im p lem en ­
ta tio n u sin g M IC te ch n iq u e
This section outlines a performance study for a direct microwave QAM
m odulator using MIC. T he constellation diagrams of the direct microwave
M-QAM m odulator with M =4, 16, 64, and 256 and a center frequency 2.5
GHz are shown in Fig.7.12 over 500 MHz bandwidth. As can be seen, there
is an error vector m agnitude (EVM) in the constellation diagram. This error
161
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
is created due to broadband operation of the m odulator. As may be seen,
the error vector m agnitude is less than 6 percent over the full bandwidth
for 4QAM. This increased to 7 percent for 16QAM. For 64QAM this error
is about 8 percent. T he error vector magnitude for 256QAM is less than 10
percent. This increase is due to the attenuator response at higher attenuation
levels. As may be seen in Fig.6.11, the attenuation characteristic is not
completely flat. The error vector magnitude is in the acceptable level for
practical systems.
7 .3 .2
D ir e c t M icrow ave M -Q A M M o d u la to r Im p lem en ­
ta tio n u sin g S i-M M IC T ech n iqu e
As stated before, direct microwave QAM m odulator realization using SiMMIC technique is a highly cost effective implementation. Such a direct
microwave m odulator design was described in the last chapter. A perfor­
mance study was conducted for this m odulator using the model presented
in Fig.5.17. In the forward bias mode, the shunt PIN diodes could be con­
sidered as a resistances. The PIN diode I region width was 100 [im. T he
carrier life tim e was r = 5usee. A predistortion circuit was designed using
Eq.(6.12). The input d a ta were selected to be at OC-3 (155.52 Mbps) rate.
An eight level PAM signal with symbol rate 25.92 Msymbol/sec was used in
each path. This translated to a four level PAM signal to each attenuator.
The power spectrum of the direct microwave 64QAM modulator and a
theoretical 64QAM m odulator without using baseband filter are compared
in Fig.7.13. As may be seen the performance of the direct microwave QAM
m odulator is very close to an ideal QAM m odulator. In addition, the level
of the out of band spectrum and channel bandwidth are similar to an ideal
64QAM m odulator. The degradation in the power spectrum of the direct
microwave 64QAM m odulator is mainly due to the predistortion circuit.
162
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.5
■%....
%-1.5
r- . i_- - - - - r - ■■
1
1 I 9
*
j %
15
9
1* ^
%
%
j
i
9
9
|
9
*b 1
%
%
|
9
9
<9
1
<9
•
\ *
9
<9 \ <J»
0P I *>
tf>
0 \ %
«
t* : U«
% i <(•
%
%
9 i
t
! %
\
e
| %
J.- - - - -
-6
*
! .
-4
-2
(c)
1
-3
1
L.
%
2
____
6
1 ----1------ 1- -
1 8 9 9 9 9 9 9’
.....
*>*>*>%%%« 9 9 9 0 0 0 0 0*
9 9 0 0 0 0 0 0*
•t
« « i
0 O 0 f 0 « « IB
0 9 M 0 0 %• 0 41« 9D
«*>tfi m- * 9 0 9 0 0 %«««<» «D
0 . ft f f 0
i(>9 9 0 0 0 9 I %**%<*<*<*»
< 0 9 0 0 0 9 9 1 * * %%%<* **
-— * ' ? <o '< r i r r Tr-|-YW *"<r«.....
9 9 9 f e i t :!
•*"%~%~VWV-%r.....
...
----- 1----- j----i_
i------ t----- _i----- 1
op v
. . 1
4
4
3
— .4 , - v v - w i - i -
1- - - -
:
9
9
9
9 \
? \
-4
----- J----- f■“ T-- -- 1
«b j %
•B :
-8
1
.5
-.5
1%
1
-1.5
8
-20
-15
-10 -5
(d)
5
10
F ig u re 7.12 Constellation diagram for direct QAM modula­
tor a) 4-QAM. b) 16-QAM. c) 64-QAM. d) 256QAM
163
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
15
20
Theoretical 64QAM
Direct 64QAM
-5
-3 0
..n
-3 5
-4 0
100
150
Frequency [MHz]
F ig u re 7.13 The power spectrum of the Si-MMIC direct mi­
crowave and an ideal 64QAM m odulator without
using filter
In the next experiment, the bandwidth effeciency of a direct microwave MQAM m odulator inplemented using silicon MMIC was compared with th a t of
a theoretical M-QAM m odulator. The data rate is OC-3 and no pulse shaping
filter was used. As can be seen from Fig.7.14, a lower channel bandwidth can
transm it an OC-3 d a ta rate by using a higher m odulation level. A theoretical
channel bandw idth 38.88, 25.92, and 19.44 MHz is required to transm it an
OC-3 signal using 16, 64, and 256 QAM modulator. As can be seen from
Fig.7.14, the performance of direct microwave M-QAM m odulator is very
close to the theoretical M-QAM modulator.
A raised cosine filter w ith roll-off factor a = .5 was used in the next
experiment. Fig.7.15 shows the control voltage in I and Q paths as well as
the PIN diode resistances corresponding to these signals. As may be seen, the
PIN diode resistances vary between 0 and 50 Q. The power spectrum of the
164
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Theoretical 4 QAM
Theoretical 16QAM
Oirect 16QAM
2 . -10
8-15
2 -2 5
-3 0
-35
-35
200
150
250
100
150
F requency [MHz]
(a)
(b )
Theoretical 64QAM
Oirect 64 QAM
Theoretical 2S6QAM
Direct 256Q AM
-5
-20
-2 0
5 -2 5
-2 5
-3 0
-3 5
-35
-40
100
-4 0
150
Frequency (MHz]
(c)
100
(d)
F ig u re 7 .1 4 Baseband Power Spectrum of M-QAM modula­
tor implemented using Si-MMIC without using
baseband filter for different levels of modulation
165
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
150
direct microwave and an ideal 64QAM m odulator using a predistortion and a
prefilter with roll-off factor a = .1, a = .5, a = .7, and a = .95 are presented
in Fig.7.16 for OC-3 data rate (155.52 Mbps). As may be seen, the power
spectrum in this figure demonstrates th at prefiltering limits the out of band
spectrum. As expected, the best out of band performance requires a higher
channel bandwidth. As may be seen from Fig.7.16.a, although a filter using
roll-off factor a = .1 results in a narrower channel spectrum, it has a high out
of band power spectrum . The best out of band power spectrum performance
can be achieved w ith a close to one. However, as can be seen from Fig.7.16.d,
a higher channel bandwidth is required in this case. A suitable trade-off is
to use a roll-off factor a = .5. As may be seen from Fig.7.16, while the pass
band characteristics of the theoretical QAM and direct QAM are similar, the
level of out of band spectrum is higher in the direct QAM m odulator. This
is due to the nonideal predistortion circuit.
7 .4
S u m m a ry
To examine the operation of the adaptive direct microwave QAM trans­
m itter in a wireless channel a performance study was conducted. The oper­
ation in Ricean and Rayleigh fading channels were compared for computer
d a ta and video traffic. Is is shown th a t using modified effective bandw idth
m etric highly saves the bandwidth and improves the ATM call acceptance
and wireless channel outage. It was also found th a t the modified effective
bandwidth metric demands more bandwidth in a Rayleigh channel compared
with a Ricean channel for similar traffic load. Moreover, the performance of
the direct QAM m odulator implemented using MIC and Si-MMIC techniques
was studied in this chapter. A comparison between the direct m odulator and
theoretical QAM m odulator shows a very good agreement.
166
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a) 0.8
O)
CO
o 0.6
2 0.4
0.4
O 0.2
Q0.2
0.5
0.5
time[microsec]
time[microsec]
(b)
0.5
0.5
time[microsec]
time[microsec]
(C)
(d)
^60
F ig u re 7.15 Input current and pin diode forward bias resis­
tance for I and Q paths in 64QAM m odulator
using a filter with roll-off factor a = .5
167
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-10
Theoretical 64QAM(alpha=.1)
Oirect 64QAM(alpha=.1)
Theoretical 64QAM(alpha=-5)
Direct 64QAM(alpha=.S)
-10
E -2 0
-20
-40
-4 0
1 -5 0
-5 0
-60
-60
-7 0
100
-70
150
100
150
F requency [MHz]
(a)
(b)
Theoretical 64QAM(alpha=.7)
Oirect 64QAM(a(pha=.7)
Theoretical 64QAM(alpha=.95)
Direct 64QAM(aIpha=.95)
-10
-1 0
£ -2 0
S -5 0
S -5 0
-60
-7 0
100
-7 0
150
100
Frequency [MHz}
(C)
(d )
F ig u r e 7.16 Baseband power Spectrum for 64-QAM using
direct microwave Si-MMIC implementation for
155 (OC-3) Mbps d a ta with different roll-off fac­
tor
168
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
150
8.
CO NCLUSIO NS A N D F U T U R E
R ESE A R C H
As stated in C hapter 1, this research work has been done to fulfill the
following objectives:
• To design an architecture for a fixed wireless ATM transm itter with a
focus on broadband applications.
• To derive a metric to estim ate the required bandwidth of the ATM
traffic sources in the wireless network scenario.
• To design an adaptive transm itter that adapts to the bandwidth re­
quirements and the channel variation. An efficient use of the channel
bandw idth was the overall goal for such a transm itter.
• To develop a suitable architecture for the transm itter.
• To develop a hardware implementation for the proposed modulator.
• To model and characterise the performance of the proposed transm itter
for different types of ATM traffic and for different radio channels.
8.1
C o n clu sio n s
Based on the research work reported in the previous chapters, it can be
stated th a t all these objectives were realized. T he results are summarised as
follows:
It is shown th a t for design of an optimum radio ATM transm itter, both
169
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the broadband traffic and channel performance have to be m onitored. Taking
these requirements into account a new architecture for an adaptive transm it­
ter has been proposed.
A statistical model was developed to estimate the required bandwidth of
the broadband traffic. T he model uses a nonlinear relation to estim ate the re­
quired bandwidth as a function of cell loss ratio and maximum tolerable delay
using a self-similar traffic model. This is followed by a study using empiri­
cal traces to investigate the required bandwidth of com puter d a ta and video
traffic in an ATM network. Adequacy of the proposed effective bandwidth
metric was established by comparison with the computer sim ulation results
for different types of traffic. Moreover, the effective bandw idth allocation
was compared with th e peak rate allocation. It is found th a t a bandwidth
saving close to 50 percent (depends on buffer size) can be achieved for ag­
gregate MPEG-1 video traffic. In a LAN type traffic, depends on utilisation
and buffer size a high bandw idth saving is also achievable.
A metric, called Modified Effective Bandwidth, has been introduced for
the wireless network. It has been shown that the capacity requirement for
such a network is a function of QoS, e.g., cell loss ratio and maximum delay;
as well as channel characteristics, e.g., fading behaviour. A Capacity Reduc­
tion Factor th a t takes into account the channel conditions was introduced.
Using this factor a m athem atical equation for the modified effective band­
width has been proposed as a QoS metric for wireless ATM.
The concept of modified effective bandwidth was applied to Ricean and
Rayleigh channels and it was found that the modified effective bandwidth
metric demands more bandw idth in Rayleigh channels compared with Ricean
channels for sim ilar traffic load. This is due to the nature of the Rayleigh
170
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
channel. An experiment with multiple video sources showed th a t a similar
traffic load requires 12 percent more bandwidth in a Rayleigh channel com­
pared to the Ricean channel with Rice param eter of AT = 6.
An M-QAM modulator was proposed as an adaptive m odulator for ra­
dio ATM. It was shown th at the modified effective bandw idth can be used
to adapt the M-QAM m odulator to traffic characteristics as well as channel
conditions thus resulting in an optimum transm itter.
An M-QAM control unit has been designed. This unit assigns a buffer size
to each VP according to the delay constraints. By using the modified effec­
tive bandw idth metric, the required bandwidth for each VP was calculated
in this unit and a modulation level was specified in the M-QAM modula­
tor. It was shown th at such a unit monitors the instantaneous bandwidth
dem and continuously while maintaining the outage conditions. Simulation
studies showed that the call acceptance improvement is close to 20 percent
compared to the average fixed QAM m odulator. An outage improvement
b etter than .02 was obtained compared to the highest level fixed QAM Mod­
ulator in all experiments.
A novel method to realize a QAM m odulator has been introduced. The
proposed m ethod is suitable for high speed and high bandwidth operation.
A key issue in the design of the proposed m odulator was a model for
th e PIN diode as a data controlled attenuator. An accurate model to char­
acterise a PIN diode as a variable resistor controlled by high speed digital
d a ta and R F pump signals was developed. Using the residue theorem and
complex inversion formula, an analytical model was derived. This model was
171
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
employed for com putation of the switching time for PIN diodes. The the­
oretical results obtained using the proposed model were compared with the
measurement results of the commercial PIN diodes. An excellent agreement
was achieved. Expected performance of the QAM m odulator was computed
using circuit models and it was shown th at the proposed m ethod can be used
to realize an adaptive modulator.
An im plem entation for the MIC version of the m odulator was proposed
using silicon diodes, m icrostrip and coplanar lines. A predistortion filter to
provide a linear relation between the control voltage and reflection attenu­
ator was proposed. T he constellation study over the 500 MHz bandwidth
shows an error vector m agnitude better than 10 percent for the MIC version
of the m odulator.
A highly cost effective direct QAM m odulator can be realized using SiMMIC technology. Keeping this in view, subsystems were designed using
Si-MMIC technology. Using these designs, a highly cost effective high bit
rate adaptive M-QAM m odulator for radio ATM can be realized. A com­
parison between the proposed direct QAM m odulator and theoretical QAM
m odulator shows an excellent agreement. This new m ethod to implement
the QAM m odulator has great potential.
Finally, a system study was carried out to examine the operation of adap­
tive direct microwave QAM modulators in the wireless channel for transmis­
sion of ATM traffic. T he operations in Ricean and Rayleigh fading channels
have been examined for computer data and video traffic. T he call acceptance
and outage characteristics were compared with those of the fixed QAM mod­
ulator. T he results show th a t the proposed adaptive m ethod can efficiently
meet the expected high demand for a low cost transm itter for broadband
wireless applications. Some results of this research work have been patented
172
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[7], published [2], [5], [6 ], [8 ], [10], [121], will publish [12], and subm itted to
[3], [9], [11] various journals and conferences.
173
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.2
C o n trib u tio n s o f th is R esearch W ork
• An adaptive direct microwave M-QAM transm itter for wireless ATM
applications has been designed. The transm itter uses an M-QAM mod­
ulator capacity controlled by modified effective bandwidth. T he mod­
ulation control unit estimates the required bandwidth of traffic by con­
sidering traffic densities and channel performance.
• A new metric, called Modified Effective Bandwidth, has been introduced
to assign the actual bandwidth to an ATM connection in a wireless
channel and an analytical relation has been derived to calculate this
metric.
• A new factor, called Capacity Reduction Factor fo r M-QAM, has been
introduced in Rice and Rayleigh fading channels and an analytical re­
lation is derived to calculate this factor in various fading channels.
• A new closed form relation based on residue theory has been derived to
study the operation of PIN diodes. The relation is the only available
relation th a t accurately estimates the transition time of the PIN diode
when its bias is changing from reverse bias to forward bias.
• A new architecture for the direct microwave QAM modulator has been
introduced th at results in a high performance and low cost implemen­
tatio n of the QAM modulator.
• A microwave integrated circuit has been implemented for the direct
microwave QAM modulator that is scalable in frequency.
• A Si-MMIC has been implemented for the direct microwave QAM mod­
ulator which results in a simple, low cost, and high performance mod­
ulator for broadband applications.
174
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.3
F u tu re W ork
The fabrication of the hardware and field trials in fixed wireless ATM de­
ployment is the next step. T he direct microwave QAM m odulator introduced
in this research work can be realized using MIC and Si-MMIC technologies.
A power oscillator can be designed on the same MMIC chip as th e modu­
lator using an IMPATT diode oscillator th a t can be easily implemented in
Si-MMIC technology. Thus, the complete transm itter can be implemented
on a single MMIC chip. Meanwhile, the other architectures to realize a QAM
m odulator as well as Si-MMIC implementation issues should be studied. The
design reported in this thesis should be examined at different microwave and
millimeter-wave frequencies.
The proposed system can be developed to operate with various access
techniques, e.g., FDMA and TDMA. The results of this method can be ex­
tended to other high speed services such as the new millimeter wave band
local m ultipoint distribution services (LMDS).
175
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
R EFER EN C ES
[1 ] T.S. R appaport, Wireless Communications, Principle and Practice,
Prentice Hall, P T R 1996.
[2] A. Mohammadi, D. Klymyshyn, and S. K um ar, “A 155.52 Mbps ATM
radio in 18 GHz band using TCM-QAM” , Proceedings of the 1996 Inter­
national Wireless Conference, Wireless 96, Calgary, Alberta, pp. 424430, July 1996.
[3] A. M ohammadi, and S. Kumar, “An adaptive MQAM modulator for
fixed wireless ATM networks” , Submitted to IE E E Journal on Selected
Areas in Communications.
[4] C.S. Chang, and J.A . Thomas, “Effective bandw idth in high speed digi­
tal networks” , IE E E Journal on Selected Areas in Communications, Vol.
13, No. 6 , pp. 1091-1099, August 1995.
[5] A. M ohammadi, S. Kumar, and D. Klymyshyn, “Characterisation of
effective bandw idth as a metric of quality of service for wired and wireless
ATM networks” , IE E E International Conference on Communications,
IC C ’97, M ontreal, pp. 1019-1024, June 1997.
[6 ] S. Kum ar, A. Mohammadi, and D. Klymyshyn, “A direct 64-QAM mod­
ulator suitable for MMIC implementation” , Microwave Journal, Vol. 40,
No. 4, pp. 116-122, April 1997.
[7] A. M oham madi, S. Kumar, and D. Klymyshyn, Direct QAM Modulator,
Canadian and U.S. Patents approved March 1998.
[8 ] S. Kum ar, A. Mohammadi, and D. Klymyshyn, “A direct microwave
frequency m odulation for a 155.52 Mbps 17.7 to 19.7 GHz radio for ATM
176
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
traffic” , Proceedings o f the 8th Asian Pacific Microwave Conference,
A P M C ’96, New Delhi, India, Vol. 4, pp. 1447-1550, December 1996.
[9] A. Mohammadi, and S. Kumar, “An adaptive M-QAM in fixed wireless
ATM networks”, Invited paper to 1999 IEEE International Conference
on Communications, IC C ’99.
[10] A. Mohammadi, S. K um ar, H.C. Wood, and D. Klymyshyn, “Charac­
terisation of the microwave PIN diode for high speed d a ta m odulation” ,
Proceedings of the International Semiconductor Device Research Sym­
posium, IS D R S ’97, Charlottesville, VA, December 1997.
[11] A. Mohammadi, S. Kum ar, H.C. Wood, and D. Klymyshyn, “Time do­
main characterisation of the microwave PIN diode driven by high speed
data”, Submitted and revised fo r IEEE Transactions on Microwave The­
ory and Techniques.
[12] A. Mohammadi, and S. Kumar, “A low-cost direct microwave QAM
modulator using silicon MMIC technology”, Accepted fo r 1999 IEEE
M T T -S International Topical Symposium on Technology for Wireless
Applications, Vancouver, February 1999.
[13] D. Raychaudhuri, L.J. French, R.J. Siracusa, S.K. Biswas, R. Yuan,
P. Narasimhan, and C.R. Jonston, “WATMnet: A prototype wireless
ATM system for m ultim edia personal communications” , IE EE Journal
on Selected Areas in Communications, Vol. 15, No. 1, pp. 83-95, January
1997.
[14] M. Shafi, A. Hashimoto, M. Umehira, S. Ogose, and T. Murase, “Wire­
less communications in the twenty-first century: A perspective” , Pro­
ceedings of the IEEE, Vol. 85, No. 10, pp. 1622-1638, October 1997.
[15] S.F. Bush, S. Jagannath, R. Sacchez, J. Evans, S. Shanmugan, and V.
Frost, “A control and management network for wireless ATM system” ,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1996 International Communication Conference, IC C ’96, Vol. 1 , pp. 459463, June 1996.
[16] Y. Shoji, K. Tsukamoto, S. Komaki, “Proposal of the radio high­
way networks using asynchronous time division multiple access” , IEIC E Transactions on Communications, Vol. E79-B, No. 3, pp. 308-315,
March 1996.
[17] J. Mikkonen, M. Niemi, and P. Nieminen, “A system scenario for wire­
less broadband access with multimedia support” , Proceedings of the
IC U P C ’97, San Diego, October 1997.
[18] N. Passas, S. Paskalis, and D. Vali, “Quality-of-service oriented medium
access control for wireless ATM networks” , IEEE Communications Mag­
azine, Vol. 35, No. 11, pp. 42-50, November 1997.
[19] N. Morinaga, M. Nakagawa, and R. Yokohama, “New concepts and
technologies for achieving highly reliable and high capacity multimedia
wireless communications systems”, IEEE Communications Magazine,
Vol. 35, No. 1, pp. 34-40, January 1997.
[20] M.J. Gans, T.S. Chu, P.W. Wolniansky, and M. Carloni, “A 2.5 Giga­
bit 23-mile radio link for LuckeyNet” , Proceedings o f G LO BECO M ’91,
Phoenix, pp. 1065-1068, November 1991.
[21] J.B. Cain, D.N. McGregor, “A recommended error control architecture
for ATM networks with wireless link,” IEEE Journal on Selected Areas
in Communications, Vol. 15, No. 1 , pp. 16-27, January 1997.
[22] A.H. Aghvami, O. Gemikonalu, and S. Kato, “Transmission of SDH
signals through future satellite channels using high level modulation
technique” , IE E E Journal on Selected Areas in Communications, Vol.
10, No.
6
, pp. 1030-1036, August 1992.
178
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[23] T. Okada, T. Takao, and T. Shirato, “Feasibility study of variable m ulti­
level QAM modem for wireless ATM network”, IE IC E Transactions on
Communications, Vol. E79-B, No. 3, pp. 316-326, March 1996.
[24] S. Komaki,
“Theoretical analysis of a capacity controlled digital mi­
crowave radio” , IE IC E Transactions on Communications, Vol. J73-B-II,
No. 10, pp. 498-503, October 1990.
[25] A. Sato, N. Omura, M. Ito, and T. Ito, “Broadband ATM wireless trans­
mission characteristics in 20 GHz band” , 1997 IEEE International Con­
ference on Communications, IC C ’97, Montreal, pp. 1549-1553, June
1997.
[26] K. Sato, S. O hta, and I. Tokizawa, “Broadband ATM network architec­
ture based on virtual p ath” , IE E E Transactions on Communications,
Vol. 38, No. 8 , pp.
1 2 1 2
- 1 2 2 2 , August 1990.
[27] S. Ohta, and K. Sato, “Dynam ic bandwidth control of the virtual path in
a Asynchronous Transfer Mode network” , IEEE Transactions on Com­
munications, Vol. 40, pp. 1239-1247, July 1992.
[28] F. Vakil, “A capacity allocation rule for ATM networks”, Proceedings
of the G LO BECO M ’93, Houston, Texas, pp. 406-415, November 1993.
[29] B.H. Ryu, H. Ohsaki, and H. Miyahara, “Design algorithm for virtual
path based ATM networks” , IE IC E Transactions on Communications,
Vol. E79-B, No. 2, pp. 97-107, February 1996.
[30] D. Anick, D. M itra, and M.M. Sondhi,
“Stochastic theory of a d a ta
handling system with m ultiple sources” , Bell System Technical Journal,
Vol. 61, pp. 1871-1894, O ctober 1982.
[31] R. Guerin, H. Ahmadi and M. Naghshineh,
“Effective capacity and
its application to bandwidth allocation in high-speed networks” , IE E E
179
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Journal on Selected Areas in Communications, Vol. 9, No. 7, pp. 968981, July 1991.
[32] E.P. Kelly, “Effective bandwidths a t multiple-class queues” , Queueing
Systems, Vol. 9, pp. 5-16, September 1991.
[33] R.J. Gibbens and P.J. Hunt, “Effective bandwidths for the m ulti-type
UAS channel” , Queueing Systems, Vol. 9, pp. 17-28, September 1991.
[34] A. Elwalid, and D. M itra, “Effective bandwidth of general Marko­
vian traffic sources and admission control of high speed networks” ,
IE E E /A C M Transactions on Networking, Vol. 1, No. 3, pp. 329-341,
June 1993.
[35] A. Elwalid, D. M itra, and R.H. W entworth, “A new approach to allo­
cating buffers and bandwidth to heterogeneous, regulated traffic in an
ATM node” , IE E E Journal on Selected Areas in Communications, Vol.
13, No. 6 , pp. 1115-1127, August 1995.
[36] F. Kelly, “Notes on effective bandw idth” , in , Stochastic Networks: The­
ory and Applications, Oxford University Press, pp. 141-168, 1996.
[37] K. W . Ross, Multiservice loss models fo r broadband telecommunication
networks, Springer-Verlag, London, 1995.
[38] W.E. Leland, M.S. Taqqu, W. W illinger, and D.V. Wilson, “On the self­
similar nature of E thernet traffic” , IE E E /A C M Transactions on N et­
working, Vol. 2, No. 1, pp. 1-15, February 1994.
[39] M.W. G arrett, and W . Willinger, “Analysis, modelling and generation of
self-similar VBR video traffic” , Proceedings of ACM Sigcom m ’94, Lon­
don, pp. 269-280, August 1994.
[40] J. Beran, R. Sherman, M.S. Taqqu, and W. Willinger,
“Long-range
dependence in variable bit rate video traffic,” IEEE Transactions on
Communications, Vol. 43, N o.2/3/4, Feb./M arch/A pril 1995.
180
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[41] Y. Chen, Z. Deng, and C. Williamson, “A model for self-similar E th ­
ernet LAN traffic: design, implementation, and performance implica­
tions” , Proceedings of the 1995 Sum m er Computer Simulation Confer­
ence (SC SC ’95), Ottawa, Ontario, pp. 831-837, July 1995.
[42] W. Willinger, M.S. Taqqu, R. Sherman, and D.V. Wilson, “Self­
similarity through high-variability: statistical analysis of E thernet LAN
traffic at the source level” , IE E E /A C M Transactions on Networking,
Vol. 5, No. 1, pp. 71-86, February 1997.
[43] B. Tsybakov, and N. Georganas, “On self-similar traffic in ATM queues:
definition, overflow probability bound, and cell delay distribution” ,
IE E E /A C M Transactions on Networking, Vol. 5, No. 3, pp. 397-409,
June 1997.
[44] I. Norros, “A storage model w ith self-similar input” , Queueing Systems,
Vol. 16, pp. 387-396, 1994.
[45] I. Norros, “On the use of fractional Brownian motion in the theory of
connectionless networks”, IE E E Journal on Selected Areas in Commu­
nications, Vol. 13, No. 6 , pp. 953-962, August 1995.
[46] M. Naghshineh, and A.S. Acampora, “QoS provisioning in micro-cellular
network supporting multiple classes of traffic” , Journal on Wireless N et­
works, Vol. 2, No. 2, pp. 195-203, 1996.
[47] W .T. Webb, and R. Steele, “Variable rate QAM for mobile radio” , IE E E
Transactions on Communications, Vol. 43, No. 7, pp. 2223-2230, July
1995.
[48] W . Webb, “Spectrum efficiency of multilevel m odulation schemes in
mobile radio communications” , IE E E Transactions on Communications,
Vol. 43, No. 8 , pp. 2344-2349, August 1995.
181
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[49] T. Ue, S. Sampei, and N. Morinaga, “Symbol rate controlled adap­
tive m odulation/T D M A /T D D for wireless personal communication sys­
tem s” , IE IC E Transactions on Communications, Vol. E78-B, No. 8 , pp.
1117-1124, August 1995.
[50] S. Kumax, “Directly m odulated VSAT transm itter,” Microwave Journal,
pp. 255-264, April 1990.
[51] A.A. Abidi, “Direct conversion radio transceivers for digital communi­
cations” , IE E E Journal o f Solid State Circuits, Vol. 30, No. 12, pp.
1399-1410, December 1995.
[52] L.E. Larson, R F and Microwave Circuit Design for Wireless Comm uni­
cations, Artech House, Boston, 1996.
[53] G.A.S. Machado, Low-Power HF Microelectronics a unified approach,
LEE Press, 1996.
[54] I. Telliez, A. Couturier, C. Rumelhard, C. Versnaeyen, P. Cham ­
pion, and D. Fayol, “A com pact monolithic microwave demodulator for
64QAM digital radio links” , IE E E Transactions on Microwave Theory
and Techniques, Vol. 39, No. 12, pp. 1947-1953, December 1991.
[55] R. Pyandiah, P. Jean, R. Leblance, and J. Meunier, “GaAs monolithic
direct linear (1-2.8) GHz Q PSK m odulator” , 19th European Microwave
Conference, London, pp. 597-602, September 1989.
[56] F.L.M. van den Bogaart, and R. Pyndiah, “A 10-14 GHz linear MMIC
vector m odulator with less th an
.1
dB and
.8
degree amplitude and phase
error” , Proceedings of IE E E Microwave Symposium, Dallas, Texas,
pp.465-468, June 1990.
[57] T. Sas aid, and S. Otaka, “A GaAs direct-conversion 7t / 4 shifted QPSK
m odulator IC with 0-28 dB variable attenuator for 1.9 GHz personal
handy phone systems” , IE E E GaAs IC Symposium, pp. 241-244, 1995.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[58] I.D. Robertson, M M IC Design, IEE Press, 1995.
[59] L.M. Delvin, and B .J. Minnis, “A versatile vector modulator design for
MMIC” , Proceedings of IE E E Microwave Symposium, Dallas, Texas, pp.
519-522, June 1990.
[60] D. Klymyshyn, S. Kum ar, and A. M ohammadi, “A 360 linear microwave
phase shifter with a F E T frequency/phase multiplier” , Microwave Jour­
nal, pp. 130-137, July 1997.
[61] D. Klymyshyn, S. K um ar, and A. Mohammadi, “GMSK M odulator
Formed of PLL to W hich Continuous Phase Modulated Signal is Ap­
plied” , Canadian and U.S. Patents Approved, April 1998.
[62] S. Lucyszyn, I.D. Robertson, and A.H. Aghvami, “24 GHz serrodyne fre­
quency translator using a 360 degree analog CPW MMIC phase shifter” ,
IE E E Microwave and Guided Wave Letters, Vol. 4, No. 3, pp. 71-73,
March 1994.
[63] D.M. Pozar, Microwave Engineering, Addison-Wesley, 1990.
[64] J.F.
Luy,
and P. Russer,
Silicon-Based Millimeterwave Devices,
Springer-Verlag, Berlin, 1994.
[65] J.F. Luy, K.M. Strohm, J. Buechler, and P. Russer, “Silicon monolithic
millimeter-wave integrated circuits” , IE E Proceedings-H, Vol. 139, No.
3, pp. 209-216, March 1992.
[6 6 ] W.A. Davis, Microwave Semiconductor Circuit Design , Van N ostrand
Reinhold, NY, 1984.
[67] K. Pahlavan, A. Zahedi, and P.Krishnam urthy, “Wideband local access:
wireless LAN and wireless ATM” , IE E E Communications Magazine,
Vol. 35, No. 11, pp. 34-40, November 1997.
183
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[6 8 ] J. Legras, “ATM (Asynchronous Transfer Mode): Overview, Synergy
w ith SDH, and deployment perspective” , in ,SO N ET-SD H A Sourcebook
o f Synchronous Networking,” C.A Siller, M. Shaft, IEEE Press, 1996.
[69] M. Schwartz, Broadband Integrated Networks, Prentice Hall, NJ, 1996.
[70] A. Kassler, and P. Schulthess, “Mobile broadband applications and the
Magic WAND system” , Proceedings of A C T S Workshop on Personal
Mobile Multimedia, Singapore, September 1997.
[71] B.C. Lindberg, Digital Broadband Networks and Services, McGraw-Hill,
NY, 1995.
[72] K. Liu, D.W. Petr, V.S. Frost, H. Zhu, C. Braun, and W. Edwards,
“Design and analysis of a bandwidth management framework for ATMbased broadband ISDN” , IEEE Communications Magazine, pp. 138145, Vol. 35, No. 5, May 1997.
[73] I. Habib, “Bandw idth allocation in ATM networks” , IEEE Com m uni­
cations Magazine, Vol. 35, No. 5, pp. 120-121, May 1997.
[74] S.J. Golestani, “Congestion-free communications in high speed packet
networks” , IE E E Transactions on Communications, Vol. 39, No. 12, pp.
1802-1812, December 1991.
[75] A.K. Parekh, and R.G. Gallager, “A generalised processor-sharing ap­
proach to flow control in integrated service networks: single node case” ,
IE E E /A C M Transactions on Networking, Vol.l, No. 3, June 1993.
[76] D.E. Dodds, and L. Du, “ATM Fram ing using CRC bytes”, 1994 IE E E
International Conference on Communications, New Orleans, pp. 410414, June 1994.
[77] J. Newman, “On a new class of contagious distribution applicable in
entomology and bacteriology” , Annals of Math. Statistics 10, pp. 35-57,
1939.
184
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[78] D.J. Reininger, D. Raychaudhuri, and J.Y. Hui, “Bandwidth renegotitiaion for V BR video over ATM networks,” IE EE Journal on Selected
Areas in Communications, Vol. 14, No. 6 , pp. 1076-1086, August 1996.
[79] A.J. Goldsmith, and S.G. Ghua, “Variable-rate variable power MQAM
for fading channels,” IE EE Transactions on Communications, Vol.45,
No. 10, pp. 1218-1230, October 1997.
[80] J. G. Proakis, Digital Communications, 3rd ed., McGraw Hill, NY, 1995.
[81] S. Sampei, Applications of Digital Wireless Technologies to Global Wire­
less Communications, Prentice Hall, 1997.
[82] J. Doble, Introduction to Radio Propagation fo r Fixed and Mobile Com­
munications, Artech House, MA, 1996.
[83] K. Feher, Digital Communications: Microwave Applications, PrenticeHall, Englewood Cliffs, NJ, 1981.
[84] O. Rose, “Statistical properties of MPEG video traffic and their impact
on traffic modelling in ATM systems” , University Wurzburg Institute
Computer Science, Technical Report 101, February 1995.
[85] W. Leland, M.S. Taqqu, W. Willinger, and D.V. Wilson, “On the
self-similar nature of Ethernet traffic (Extended Version)” , IE E E /A C M
Transactions on Networking, Vol. 2, No. 1 , pp. 1-15, February 1994.
[8 6 ] M. Nomura, T. Fujii, and N. Ohta, “Basic characteristics of variable
bit rate video coding in ATM environment” , IE EE Journal on Selected
Areas in Communications, Vol. 7, No. 5, pp. 752-760, July 1989.
[87] A. Papoulis, Probability, Random variable, and Stochastic Process, Mc­
Graw Hill, NY, 1991.
[8 8 ] A. Patel, and C. Williamson, “Statistical multiplexing of self-similar
traffic: Theoretical and simulation results” , Proceedings of the 1997
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
IA ST E D Conference on Applied Modeling and Simulation (A M S ’97),
Banff, Alberta, pp. 298-302, July 1997
[89] P. Gburyzynski, T. Ono-Tesfaye, and S. Ramaswamy, ‘"Modelling ATM
network in a parallel simulation environment: a case study” , Proceed­
ings of the 1995 Sum m er Computer Simulation Conference (SC SC ’95),
Ottawa, Ontario, pp. 869-874, July 1995.
[90] J.K. Cavers, “Variable-rate transmission for Rayleigh fading channels” ,
IE E E Transactions on Communications, Vol. 20, pp 15-22, February
1972.
[91] A.J. Goldsmith and P.P. Varaiya, “Capacity of fading channels with
channel side inform ation”, IEEE Transactions on Information Theory,
Vol. 42, pp. 1986-1992, November 1997.
[92] M.K. Simon, S.M. Hinedi, and W.C. Lindsey, Digital Communication
Techniques, Prentice Hall, 1994.
[93] M. Naghshineh, and M. Willebeek-LeMair, “End to-end QoS provision­
ing m ultimedia wireless/mobile networks using an adaptive framework”,
IE E E Communications Magazine, pp. 72-81, November 1997.
[94] R.G. Gallager, Inform ation Theory and Reliable Communications, W i­
ley, 1968.
[95] W .C.Y. Lee, “E stim ate of channel capacity in Rayleigh fading envi­
ronm ent” , IEEE Transactions on Vehicular Technologies, Vol. 39, pp.
187-189, August 1990.
[96] P.B. Papazian, G.A. Hufford, R.J. Achatz, and R. Hoffman, “Study of
the local m ultipoint distribution service radio channel” , IEEE Transac­
tions on Broadcasting, Vol. 43, No. 2, pp. 175-184, February 1997.
186
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[97] T. Okada and S. Aikawa, ‘T h e design of all digital multi-level m odulator
and its application to variable capacity transmission” , IE IC E Transac­
tions on Communications, Vol. J75-B-II, No. 6 , pp. 325-336, June 1992.
[98] T. Okada, T. Takao, and T. Shirato, “Feasibility study of variable multi­
level QAM modem for wireless ATM network,” IEIC E Transactions on
Communications, Vol. E79-B, No. 3, pp. 316-326, March 1996.
[99] J.C. McDade, and F. Schiavone, “Switching time performance of mi­
crowave PIN diodes” , Microwave Journal, pp.65-68, December 1974.
[100] H.A. W atson, Microwave Semiconductor Devices and Their Circuit A p­
plications, McGraw Hill, NY, 1969.
[101] W. Reiss, “Volterra representation of a forward-biased PIN diode” ,
IE E E Transactions on Electron Devices, Vol. ED-28, No. 12, pp. 14951500, December 1981.
[102] K. Chang, Handbook of Microwave and Optical Components , C hapter
4, “Semiconductor control devices, PIN diodes” , by J.F. W hite, John
Wiley, Vol. 2, 1990.
[103] M.R. Spiegel, Advanced Mathematics fo r Engineers and Scientists,
McGraw-Hill, 1980.
[104] D. Leenov, “The silicon PIN diode as microwave radar protector at
megawatt levels,” IE E E Transactions on Electron Devices, Vol. ED-11,
No. 2, pp. 53-61, February 1964.
[105] M /A-CO M , R F and Microwave Semiconductors, Burlington MA, 1994.
[106] Alpha Industries Inc., Semiconductors and M M IC , Woburn, MA, 1996.
[107] R.H. Caverly, and G. Hiller, “Distortion in PIN diode control circuits” ,
IE E E Transactions on Microwave Theory and Techniques, M TT-35, No.
5, pp. 492-501, May 1987.
187
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[108] P. Stabile, A. Rosen, “A silicon technology for millimeter-wave mono­
lithic circuits” , R CA Review, Vol. 45, pp. 587-605, December 1984.
[109] H P-EEsof High-Frequency Design Solutions, Release 6 .0 , July 1995.
[110] K.W . Kobayashi, L. Tran, A.K. Chi, and D.C. Striet, “A 50 MHz-30
GHz broadband co-planar waveguide SPDT PIN diode switch with 45
dB Isolation” , IE E E Microwave and Guided Wave Letters, pp. 56-58,
February 1995.
[111] S. Lucyszyn, and I.D. Robertson, “Analog reflection topology build­
ing block for adaptive microwave signal processing applications” , IEEE
Transaction on Microwave Theory and Techniques, Vol. 43, pp. 601-611,
March 1995.
[112] K. Kathiravan, S. Kumar, and G. Harron, “Combing prefiltering and
predistortion in a high power direct PSK m odulator”, Proceedings of
IE E E International Microwave Symposium, Dallas, Texas, pp. 877-880,
June 1990.
[113] S. Kum ar, “Power amplifier linearization using MMICs” , Microwave
Journal, pp. 96-104, April 1992.
[114] J.D. Cressler, “SiGe HBT technology: a new contender for Si-Based RF
and microwave circuit applications” , IE EE Transactions on Microwave
Theory and Techniques, Vol. 46, No. 5, pp. 572-589, May 1998.
[115] P. Russer, “Si and SiGe millimeter-wave integrated circuit” , IEEE
Transactions on Microwave Theory and Techniques, Vol. 46, No. 5, pp.
590-603, May 1998.
[116] R.K. Hoffman, Handbook of Microwave Integrated Circuits, Artech
House, 1987.
188
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[117] R.W. Vogel, “Analysis and design of lumped and lum ped-distributedelement directional coupler for MIC and MMIC applications,” IE E E
Transactions on Microwave Theory and Techniques, MTT-4.0, pp. 253262, February 1992.
[118] T. Hirota, A. Minakawa, and M. Muragchi, “Reduced-size branch line
and rat-race hybrid for unipolar MMIC’s”, IEEE Transactions on Mi­
crowave Theory and Techniques, M TT-38, pp. 270-275, March 1990.
[119] Z. Zhang, J. Kurose, J.D. Salehi, and D. Towsley, “Smoothing, statis­
tical multiplexing, and call admission control for stored video,” IEEE
Journal on Selected Areas in Communications, Vol. 15, No. 6 , pp. 11481166, August 1997.
[120] I.S. Gradshteyn, and I.M. Ryzhik, Table of Integrals, Series, and Prod­
ucts, Academic Press, New York, 1980.
[1 2 1 ] S. Kumar, D. Klymyshyn, and A. Mohammadi, “Broadband electron­
ically tunable ring resonator filter with negative resistance coupling,”
Electronics Letters, Vol. 32, No. 9, pp. 809-810, April 1996.
[122] W. Kang, I. Chang, and M. Kang, “Reflection type low-phase-shift
attenuator,” IEEE Transactions on Microwave Theory and Techniques,
Vol. 46, No. 7, pp. 1019-1021, July 1998
[123] C. Chi and G.M. Rebiz, “P lanar microwave and millimeter-wave
lumped elements and coupled-lines filters using micro-machining tech­
niques,” IEEE Transactions on Microwave Theory and Techniques,
M TT-43, pp. 730-738, April 1995.
[124] J. Rieh, L. Lu, P.B. Katehi, P. Bhattacharya, E.T. Croke, G.E. Ponchak, and S.A. Alterovitz, “X- and Ku-band amplifier based on Si/SiGe
HBT’s and micromachined lumped components” , IE E E Transactions
189
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
on Microwave Theory and Techniques, Vol. 46, No. 5, pp. 68-5-694, May
1998.
190
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A. C A P A C IT Y R E D U C T IO N FACTOR IN
M -Q A M IN RAYLEIGH CH A N N EL
The capacity reduction factor in a Rayleigh channel for M-QAM mod­
ulator is defined by Eq.(4.18) as follows:
c
_
Sy *0 0 2 ( 1 - 3-f[2ln{5pb)]~l )p { j) d j
Sm q am
iog2{1 _ 37 [2jn (5P6 )]-i)
'
(A’1}
where p ( j) is the probability density function of the received C N R in
Rayleigh channel and is obtained as follows:
P(7 ) = ^ e “ r j
(A.2)
where T isthe average power of 7 . The num erator of E q.(A .l) can be
written as:
I m cam = I (092(1 - j
)
P
(7)d7.
(A.3)
Replacing Eq.(A.2) in E q.(A .i) results in
Im qam =
^
I
M l -
*”•
<A -4 >
This integral can be solved as follows [120]:
, e
, 2ln(5pb)
2ln(5pb) ^
I m q am = —log2exp(------— — ) £ t-(— — — ),
where
00
xk
Ei{x) = E + l n ( - x ) + E r T 7 ’
fc=i M l '
191
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
,A^
(A.o)
and £ = .5 7 7 2 1 5 is the Euler constant. The Eq.(A.5) can be written as:
, _ _ e
,
I m q am = —log2exp{
2in(5p6) w * , , , -2 ln {o p b)
2ln{5pb) x _ , 2ln(opb)
—— ){E + ln{-----— -----)+ ( — — — ) + ( — — — )7 2 .2!+„.)
(A.7)
As can be seen, the higher order terms can be ignored if T is big enough.
Then, Eq.(A.7) can be approximated as follows:
7
e
7
2/n(5p&) ^t r, , ,_ ,-2Z n(5p6) N t ,2ln(5pb)„
I m qam — —log2exp{
— — )(E + ln{
—-----) + (— —— )).
(A.8 )
Using Eq.(A. 8 ), the capacity reduction factor in Rayleigh fading using
M-QAM m odulation can be obtained as:
r —nf
Zmqam
21n(5pb)\ / p
j / -2ln(5pb)
L n(i _ 37 [2Zn(5p6 )]-i)
2ln(5pb) x
•
192
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
B. P H A SE SH IFT CO M PEN SA TIO N IN A
R EFL EC T IO N ATTENUATO R
A reflection attenuator is shown in Fig.B.l. As may be seen, this atten­
uator uses a quadrature hybrid and PIN diode term inations. A general
equivalent circuit for a P IN diode was presented in C hapter
6
. Using
Eq.(6.2), the phase characteristics of the reflection attenuator may be
obtained as:
T
UJL>S
r
UL,S
uiR'jCj
1+w2R±Cj2 .
u>R*Cj
1+u>2R jC p
^(R j) = tan l (—-----——------ ^ ------) - tan---(—----R S — Z a + T, +, u ^ R f C p
R s -r Z a -f
x
^ ----- )
1
+ u i - R l. C j l
(B-l)
As may be seen in E q .(B .l), the phase characteristic varies with the
variation of Rj. This can be attributed to the effect of the parasitic
elements Cj, Ls. This effect can be compensated using an open stub
circuit [121]. Fig.B.2 shows such a phase compensation method using
an open stub in the PIN diode circuit [122]. The impedance of an open
stub is given by:
( B
' 2 )
Thus, the attenuator phase shift for attenuator with an open circuit stub
is modified as follows:
uiR?Cj
,, „
,
0 (R i )
=tan
_ i,
b jL s
'(—
*
i _ L . .2 d 2 / ^ , - 2
~
Zo
~
T
‘«»<*) ,
------ —
K
n s — Zi0 -r l+(Jj 2 x*Cp
,
- l,
uiR?Cj
L d -L s
*
"
Zo
I a n (»)
) - f c m ----------------------- ^----- Wl--------- )
n s -t- ZJO -r l+UJ2 R 2 C : j 2
(B.3)
The stub length, d, can b e obtained by solving this equation with <$(50)
set to
0
(0 ).
193
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Quadrature
Hybrid
ci
RFC
R,
-oooooo—
L-a / V " - 1
PIN diode
_L
F ig u re B . l
Rj
Reflection attenuator and equivalent circuit for
the P IN diode termination.
ci
0— ffljo o -A /V v -
tan 8
L-VNA -
,C
f t
1
Rj
F ig u re B .2
PIN diode equivalent circuit with an open cir­
cuited stub in series
194
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C. BIAS C IR C U IT FO R Si-M M IC D IR E C T
M ICROW AVE QAM M O DULATO R
The Si-MMIC implementation of the various subsystems for direct
microwave QAM m odulator was described in Chapter
6
. The SPD T
switches and variable attenuators in the direct microwave QAM modu­
lator circuit use PIN diodes. The PIN diode has to be suitably biased.
This bias circuit m ust apply the high speed d a ta to the SPDT switches
and variable attenuators. This circuit can be realized by a low pass filter.
The DC isolation between subsystems can be provided using coupling
capacitors.
C .l
L o w p a ss F ilter
The filter design depends on the d ata bandw idth and carrier frequency
of the m odulator. A second order B utterw orth lowpass filter with cutoff
frequency / c = 3.6 GHz can provide the desired frequency characteris­
tics. The required bandwidth of high speed d ata is less than 100 MHz
and carrier frequency is about 18 GHz. T he filter and its frequency char­
acteristic are shown in Fig.C .l.
C .1.1
L o w p a ss F ilter Im p le m e n ta tio n on S i-M M IC
S u b str a te
To implement th e lowpass filter on Si-MMIC substrate, a spiral induc­
tor may be used [123]. The dimensions of the inductor are shown in
Fig.C.2.a. This spiral inductor provides a 3.15 nH inductance in desired
bandwidth. This inductance value can be easily realized on Si-MMIC
technology [124]. A 1.26 pF capacitance was used in the lowpass filter.
195
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
R=50Q
W
----
L=3.15nH
C=1.26 pF < R = 5 0 Q
R etu rn L oss(ldeaf)
i
R eturn lo ss(S i-M M IC )
Insertion L o ss(td eaf)
[
Insertion Loss(Si-M M IC )
-5
—15
-2 0
...V
-2 5
-3 0 ,
F re q u e n c y [GHzI
F ig u r e C . l
a)A lowpass filter as a bias circuit for the PIN
diode b)Bias circuit frequency response
A metal-insulator-metal(M IM ) capacitor in Si-MMIC can use Sioo as di­
electric. Such a MIM capacitor is shown in Fig.C.2.b. A dielectric with
thickness d=2000 A 0 provides a capacitance .199 f F / f i m 2. Thus, a 1.26
pF capacitor requires metal electrodes with dimensions 80 x 80 /jm 2.
The lowpass filter, shown in Fig.C.3, is realized using MIM capacitor
and spiral inductor. The frequency response has been compared with
the ideal filter in Fig.C .l. A capacitor C = 2.95 pF may be used as a
coupling capacitor to provide the DC isolation. Using an MIM capacitor
on silicon substrate, a capacitor with dimensions
122
x
1 2 2
[im2 should
provide the desired capacitance. The SPDT switch and variable atten ­
uator circuits with their bias circuit are shown in Fig.C.4 and Fig.C.5.
196
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Metal Electrode
Dielectric Film
S=20 ^ m
W=20 n m
\\\\\\\\X\\\\\\\\\\\\\\\\\X\\\\\\\\\\\\\\\\\\\\\\\\\\N\\\\\NV.
v \ \ w w \ \ \\ \ w \ w \ \ \v \ v ^ \ w ^ \ \ v w w ^ \ \ \\ \ \ \ \\ \ \ \\
F ig u re C . 2
L=320 (i m
A MIM capacitor and a spiral inductor on Silicon
substrate
F ig u re C .3 Lowpass filter bias circuit implemented using SiMMIC technology
197
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
F ig u re C .4 T he atten u ato r layout along with the bias circuit
F ig u re C .5
T he SPD T layout with the bias circuit
198
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Документ
Категория
Без категории
Просмотров
0
Размер файла
8 866 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа