close

Вход

Забыли?

вход по аккаунту

?

THESEUS: A wavelength division multiplexed/microwave subcarrier multiplexed optical network, its ATM switch applications and device requirements

код для вставкиСкачать
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI
films the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter free, while others may be
from any type o f computer printer.
The quality o f this reproduction is dependent upon the quality of 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 material had to be 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.
UMI
A Bell & Howell Information Company
300 North Zed) Road, Ann Arbor MI 48106-1346 USA
313/761-4700 800/521-0600
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
THESEUS - A Wavelength Division Multiplexed / Microwave
Subcarrier Multiplexed Optical Network, its ATM Switch
Applications and Device Requirements
W eiX in
Submitted in partial fulfillment of the
requirements for the degree
of Doctor of Philosophy
in the Graduate School of Arts and Sciences
COLUMBIA UNIVERSITY
1997
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
UMI Number: 9728325
Copyright 1997 by
Xin, Wei
All rights reserved.
UMI Microform 9728325
Copyright 1997, by UMI Company. All rights reserved.
This microform edition is protected against unauthorized
copying under Title 17, United States Code.
UMI
300 North Zeeb Road
Ann Arbor, MI 48103
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
© 1997
Wei Xin
All Rights Reserved
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
ABSTRACT
THESEUS - A Wavelength Division Multiplexed / Microwave
Subcarrier Multiplexed Optical Network, its ATM Switch
Applications and Device Requirements
Wei Xin
A Terabit Hybrid Electro-optical Self-routing IJltrafast Switch (THESEUS) has
been proposed. It is a self-routing wavelength division multiplexed (WDM) / microwave
subcarrier multiplexed (SCM) asynchronous transfer mode (ATM) switch for the
multirate ATM networks. It has potential to be extended to a large ATM switch as 1000 x
1000 without internal blocking. Among the advantages of the hybrid implementation are
flexibility in service upgrade, relaxed tolerances on optical filtering, protocol
simplification and less processing overhead. For a small ATM switch, the subcarrier can
be used as output buffers to solve output contention. A mathematical analysis was
conducted to evaluate different buffer configurations. A testbed has been successfully
constructed. Multirate binary data streams have been switched through the testbed and
error free reception (< 10'9 bit error rate) has been achieved.
A simple, intuitive theoretical model has been developed to describe the
heterodyne optical beat interference. A new concept of interference time and interference
length has been introduced. An experimental confirmation has been conducted. The
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
experimental results match the model very well. It shows that a large portion of optical
bandwidth is wasted due to the beat interference. Based on the model, several
improvement approaches have been proposed.
The photo-generated carrier lifetime of silicon germanium has been measured
using time-resolved reflectivity measurement Via oxygen ion implantation, the carrier
lifetime has been reduced to as short as 1 ps, corresponding to 1 THz of photodetector
bandwidth. It has also been shown that copper dopants act as recombination centers in
the silicon germanium.
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
Table of Contents
C h ap ter 1
In tro d u ctio n
............................................................................... 1
1.1 Historical Perspective
......................................................................... 2
1.2 Overview of the Dissertation
References
C hapter 2
.............................................................. 4
................................................................................................. 5
A New ATM Switch: THESEUS
2.1 Introduction
7
......................................................................................
7
2.1.1 Broadband Integrated Services Digital Networks ...............................7
2.1.2 Asynchronous Transfer Mode ....................................................... 11
2.1.2.1 Transfer Modes ....................................................................... 11
2.1.2.2 ATM Ceil Header Fields
....................................................... 16
2.1.2.3 ATM Protocol Reference Model
..........................................
17
2.1.2.4 ATM Layer .............................................................................
18
2.1.2.5 ATM Adaptation Layer
2.2 ATM Switches
........................................................ 21
................................................................................ 25
2.2.1 Shared Medium Architectures
..................................................... 29
2.2.2 Shared Memory Architectures ....................................................... 30
2.2.3 Space Division Architectures
...................................................... 31
2.2.4 Some Examples of ATM Switch Architectures................................. 33
2.3 THESEUS - A New ATM Switch ...................................................... 39
2.3.1 Architectures of THESEUS
......................................................... 39
2.3.2 Testbed of THESEUS .................................................................. 44
2.3.3 Performance of Small ATM Switches
2.3.3.1 Introduction
........................................ 53
........................................................................
i
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
53
2.3.3.2 Solutions to Output Contention ............................................... 53
2.3.3.3 Comparison of Group and Random Assignments of Subcarriers •••• 54
2.4 Summary
References
............................................................................................ 58
....................................................................................................
60
C hapter 3 Heterodyne O ptical Beat Interference
L im itations on THESEUS
3.1 Introduction
.................................................... 64
.......................................................................................
3.2 Heterodyne Optical Beat Interference
64
................................................ 65
3.3 Experimental Verification .................................................................. 74
3.4 Improvement of THESEUS over Optical beat interference.......................78
3.5 Summary
References
C h a p te r 4
............................................................................................ 81
....................................................................................................
Device R equirem ents for THESEUS ..................................... 84
4.1 Introduction
.......................................................................................
4.2 Time-Resolved Reflectivity Measurement
4.3 Carrier Lifetime of Silicon Germanium
4.4 Summary
References
C h a p te r 5
82
.......................................... 85
............................................. 88
............................................................................................ 97
....................................................................................................
Conclusion
5.1 Summary
84
99
............................................................................ 101
...................................................................................... 101
5.2 Suggestions for Future Work ........................................................... 102
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
List of Figures
Fig. 2.1 (a) UNI cell format and (b) NNI cell format ...................................... 15
Fig. 2.2 ATM protocol reference model
Fig. 2.3 AAL structure
........................................................ 18
................................................................................
Fig. 2.4 A cross point and its states
23
............................................................ 32
Fig. 2.5 An 8 by 8 crossbar switch ................................................................ 33
Fig. 2.6 Starlite switch architecture
.................................................................35
Fig. 2.7 Moonshine switch architecture
Fig. 2.8 Knockout switch architecture
...........................................................36
.......................................................... 38
Fig. 2.9 Basic architecture of Terabit Hybrid Electro-optical
Self-routing Ultrafast Switch (THESEUS) ............................................. 39
Fig. 2.10 Block diagram of node n of THESEUS
.......................................... 41
Fig. 2.11 Wavelength tuning response to a step input current
Fig. 2.12 Block diagram of microwave subcarrier testbed
Fig. 2.13 ASK demodulator
............................42
.................................43
........................................................................... 46
Fig. 2.14 Two different microwave subcarriers over two
lasers with the same wavelength
...................................................... 47
Fig. 2.15 Received 6.3 Mbps data on 3.0 GHz subcarrier
with different modulation indices
..................................................... 48
Fig. 2.16 Received data on 3.0 GHz subcarrier with differentdata ra te s .................50
Fig. 2.17 Eye diagram of (a) 44.7 Mbps and (b) 139Mbps d a ta ......................... 51
Fig. 2.18 Optical spectrum
........................................................................... 52
Fig. 2.19 Average subcarrier utilization versus cell arrival rate ............................59
Fig. 2.20 Average cell delay versus cell arrival rate ......................................... 59
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
Fig. 3.1 Spectrum of optical beat interference from two optical carriers................. 70
Fig. 3.2 A laser beam as a train of wave components ........................................71
Fig. 3.3 Illustration of interference length resulting from combining
two laser beams
............................................................................. 73
Fig. 3.4 Block diagram of experimental setup
................................................ 75
Fig. 3.5 Spectra of microwave subcarriers with the presence of optical
beat interference, which is centered at (a) dc, and (b) 1.2 G H z ................. 76
Fig. 3.6 Signal to interference ratio versus optical frequency forsubcarrier
Fig. 3.7 Block diagram of modified version of THESEUS
........ 77
...................... 79
Fig. 4.1 Summary of energy gap values for Si/Si[.xGex strained-layer
superlattice on Si(001) and unstrained bulk alloy
...............................
88
Fig. 4.2 Schematic structure of SiGe .............................................................. 89
Fig. 4.3 Rocking curve of (004) reflection of SiGe
....................................... 90
Fig. 4.4 SIMS depth concentration profiles of Cu and Ge ................................ 92
Fig. 4.5 Set up for time-resolved reflectivity measurement................................ 93
Fig. 4.6 Time-resolved measurement for Sample 0 - 5 .................................... 95
Fig. 4.7 Carrier lifetime vs. total oxygen ion implantation dosage ....................... 97
List of Tables
Table 4.1 Oxygen ion implantation versus carrier lifetime ................................. 94
iv
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
Acknowledgement
I would like to express my sincere appreciation to professor Edward S. Yang,
my advisor, for his incredible patience, encouragement and advice. I am deeply grateful
to him for giving me another chance to pursue my Ph. D. at Columbia, after my former
advisor Prof. Kam Lau left, and a chance not only to learn so much about Science and
Engineering, but also to learn much about this New World, linguistically, culturally and
socially.
I am very grateful to Prof. Thomas E. Stem for introducing me to this optical
beat interference problem. Also I would like to thank Prof. Thomas C. Marshall for his
stimulating discussion about this problem when I had just started doing research on it
and felt confused and desperate.
Special thanks to Prof. Dave Auston, Binbin Hu, and Li Xu for helping me and
letting me using their lab to measure silicon germanium carrier lifetime. Special thanks
are extended to Hsing-Kuen Liou for providing me with the silicon germanium samples
when he was working at IBM.
I would like to acknowledge Dr. Michael Choy, Dr. Frank Tong for lending me
so many microwave components when they were at IBM. Without them this thesis
would never have been possible.
I would like to thank all my friends and colleagues at Columbia, especially Q. Y.
Ma, Ping Mei, Yue-Fei Yang, John Wei, Clifton Liu and Millie Ehrlich for their laughter
and help. I am very grateful to Millie for correcting errors in this thesis.
v
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
This thesis was partially supported by the Center for Telecommunications
Research (CTR), Columbia University.
vi
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
1
Chapter 1 Introduction
1.1
H istorical Perspective
Communication of information between human beings in early primitive
civilizations was accomplished mainly by transmission of mechanical acoustic waves
such as voice, drums and gongs. The transmission distance of this method was very
limited, even though some forms of this method still exist today, such as fire alarms,
police sirens, church bells, and Muslim calls for prayers, but for different reasons. To
increase the transmission distance, people started to use optical or visual means to
convey information. The ancient Greeks used fire beacons, mirrors and sunlight to
transmit information as early as 1500 BC [1,2]. The Chinese used smoke in the day and
fire beacons at night to report foreign invasions at the borders around 800 BC [2]. The
long distance transmission of information remained mainly optical until the discovery of
electromagnetism in the 18th and 19th centuries. In the 18th century, a variety of
electrostatic and electrochemical telegraph systems were seen in Europe [3].
During these two centuries, Charles Coulomb (1736-1806) demonstrated the law
of electrostatic force. Karl Friedrich Gauss (1777-1855) discovered the divergence
properties of electric fields. Andre-Marie Ampere (1775-1836) discovered the relation
between a steady current in a wire and the associated magnetic force. Michael Faraday
(1791-1867) found that a changing magnetic field induced a current in a wire, and vice
versa. James Clerk Maxwell (1831-1879) summarized and extended all the empirical
knowledge on electromagnetics into a single set of mathematical equations, the famous
Maxwell's equations. And Heinrich Hertz (1875-1894) confirmed Maxwell's theory [4],
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
2
During and after this period of the great discovery of electromagnetism, in 1837,
Wheatstone and Cooke of England invented the needle telegraph system. Mores and Vail
of the United States demonstrated an electromagnetic telegraph system over a 40-mile
line between Baltimore and Washington in 1844. In the 1890s, Marconi of Italy and
Popov of Russia invented the wireless radio telegraph systems. In 1899 Marconi sent
messages across the English Channel, and, in 1901, across the Atlantic Ocean between
Poldhu, England, and St. Johns, Newfoundland [3]. In 1876, Alexander Graham Bell
[5] patented a first telephone in the United States. Thus, an electromagnetic
communication era was bom.
The modem era of optical communication may be said to have originated with the
invention of lasers in 1958, and the first developments soon followed the realization of
the first lasers in 1961. In the 1960s, many people experimented with guided systems in
which the laser beam was confined to a transmission channel by lenses spaced at 10 m or
100 m intervals apart. K. C. Kao and G. A. Hockham [6] at the Standard
Telecommunications Laboratories in Harlow, England, came to realize that a much
simpler guidance system, which used a continuous glass fiber of a kind, might be used
for telecommunication. Their work may be said to have laid the foundation for the
subject of fiber optic communications [1].
Within the modem optical communication era, lightwave technologies can be
categorized by the wavelength used in the transmission of information, the way of
transmission, and the transmission distance [7].
The first-generation lightwave technology is centered at a wavelength of 0.85
(im, due to the availability of light source and ease of implementation. It uses AlGaAs
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
lasers or light emitting diodes (LEDs) as sources, multimode fiber as transmission
medium and silicon PINs or avalanche photodiodes (APDs) as detectors.
To reduce the fiber attenuation and dispersion loss, we come to the secondgeneration lightwave systems. This technology uses InGaAsP and germanium as sources
and detectors which operate at a wavelength of 1.3 (im. It uses both multimode and
single-mode fibers.
To further achieve maximum loss-limited transmission distance, a lightwave
system should operate at the wave length where fiber attenuation is smallest This occurs
in the region of 1.55 Jim wavelength and leads to the third-generation technology. It can
still use InGaAsP and Ge for sources and detectors. The fiber attenuation loss has been
reduced to 0.16 dB/km, together with the reduction of the dispersion loss by using
dispersion-shifted fibers and single mode lasers such as distributed feedback (DFB)
lasers.
The fourth-generation lightwave technology uses coherent detection which is
analog of a superheterodyne radio detection. Compared with the direct detection of the
three previous generations, this technique offers significant improvements in receiver
sensitivity and wavelength selectivity, thereby, increasing the transmission distance to a
much greater extent.
With the invention of erbium doped fiber amplifiers (EDFAs), the transmission
distance is increased tremendously without electronic repeaters. When Desurvire was
teaching at Columbia University, he argued that this would lead to the fifth-generation
lightwave technology [8].
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
4
As
we
are now in the period
of
transition
from
electromagnetic
telecommunications to fiber optic telecommunications, it should be emphasized that
developing hybrid devices and systems can take advantage of both technologies as in the
proposed new asynchronous transfer mode (ATM) switch: Terabit Hybrid Electro-optic
Self-routing Hltrafast Switch (THESEUS) [9].
1.2
Overview of the Dissertation
In this dissertation, A new ATM switching architecture, THESEUS, its physical
implementation, possible improvements, and device requirements are presented.
In Chapter 2, the evolution and specifications of a broadband integrated services
digital network (B-ISDN) and its ATM implementation is first reviewed, following the
approaches and categorization of R. O. Onvual [10] and M. de Prycker [11]. Then, the
architecture of THESEUS and its testbed implementation is presented.
Chapter 3 investigates the optical beat interference (OBI) and its limitations on
THESEUS and wavelength division multiplexing (WDM) / microwave subcarrier
multiplexing (SCM) networks. A simple, intuitive theoretical model describing the OBI
is developed. Based on this model, some possible improvements to overcome OBI are
discussed.
In Chapter 4, some device requirements for the THESEUS and WDM / SCM
networks are discussed. A new photodetector material, silicon germanium, is
investigated. This material offers possibly 1 THz bandwidth of detection.
Finally, Chapter 5 concludes the dissertation and suggests some future work.
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
5
R eferences
[1]
J. Gowar, Optical Communication Systems, London: Prentice-Hall International,
Inc., 1984.
[2]
J. E. Midwinter and Y. L. Guo, Optoelectronics and Lightwave Technology,
Chichester, England: John Wiley & Sons, Ltd., 1992.
[3]
W. R. Bennett and J. R. Davey, Data Transmission, New York: McGraw-Hill,
1965
[4]
A. D. Olver, Microwave and Optical Transmission, Chichester, England: John
Wiley & Sons, Ltd., 1992.
[5]
J. M. Wozencraft and I. M. Jacobs, Principles o f Communication Engineering,
New York: John Wiley & Sons, Inc., 1965.
[6]
K. C. Kao and G. A. Hockham, “Dielectric-fiber surface waveguides for optical
frequencies,” Proc. I.E.E., vol. 113, pp. 1151-1158, 1966.
[7]
S. E. Miller and I. P. Kaminow, Optical Fiber Telecommunications II, San Diego:
Academic Press, Inc., 1988.
[8]
E. Desurvire, Erbium-doped Fiber Amplifiers Principles and Applications, New
York: John Wiley & Sons, Inc., 1994.
[9]
W. Xin, Z. Zhang, and E. S. Yang, ‘THESEUS - A Terabit Hybrid ElectronicOptical Self-Routing Architecture for Large ATM Switches,” 1995 IEEE/LEOS
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
6
Summer Topical Meeting Digest on Technologies for a Gbbal Information
Infrastructure, August 7-11, 1995, Keystone, CO., pp. 24-25.
[10] R. O. Onvural, Asynchronous Transfer Mode Networks: Performance Issues,
Boston: Artech House, Inc., 1994.
[11] M. de Prycker, Asynchronous Transfer Mode: Solution for Broadband ISDN, 3rd
ed., London: Prentice Hall International (UK), Ltd., 1995.
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
7
Chapter 2 A New ATM Switch: THESEUS
2.1
Introduction
In this section and the next one, following the approaches and categorization of
Onvual [1] and de Prycker [2], the evolution, background and specification of broadband
integrated services digital networks (B-ISDNs) and ATM networks are reviewed.
2.1.1
B roadband Integrated Services Digital Networks
The concept of integrated services digital network (ISDN) has been evolving
since the International Telecommunicadons Union - Telecommunication Standardization
Sector (ITU-T), formerly known as the International Telegraph and Telephone
Consultative Committee of the International Telecommunications Union (CCITT),
adopted the first set of ISDN recommendations in 1984. The main feature of the ISDN
concept is the support of a wide range of voice and nonvoice applications in the same
network. ISDN extends the concepts of telephone networks by incorporating additional
functions and features of current circuit and packet-switching networks for data to
provide both existing and new services in an integrated manner.
Two interfaces are defined for ISDNs: Basic access and Primary rate access. The
basic access interface has a total bit rate of 144 Kbps, consisting of two 64-Kbps
channels and a 16-Kbps signaling channel. The primary rate access interfaces are defined
with total bit rates of 1.544 Mbps (i.e., T1 bandwidth) and 2.048 Mbps (i.e., El
bandwidth), including a 64-Kbps signaling channel. The T1 line is used extensively in
the United States, Canada, and Japan. The El tine as a ITU-T recommended standard, is
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
8
used in the rest of the world [3]. The remaining bandwidth at each type of interface is
divided into various combinations of 64-Kbps channels. It is also possible to use the
primary rate access interfaces without a signaling channel, with clear channel rates of
1.536 Mbps and 1.92 Mbps, respectively, for T1 and El links.
It is soon realized that higher bit rates are required for applications such as
interconnection of local area networks, video, image, and so forth, bringing the
standardization process to the introduction of broadband ISDN (B-ISDN) concepts. BISDN is conceived of as an all-purpose digital network. Activities currently under way
are leading towards the development of a worldwide networking technology based on a
common set of user interfaces and universal communications. Once deployed throughout
the world, B-ISDNs will facilitate worldwide information exchange between any two
subscribers without any of the limitations that can be imposed by the communication
medium or the media.
ITU-T (CC1TT) Recommendation 1.113 defines “broadband” as “a service or
system requiring transmission channels capable of supporting rates greater than the
primary rate”[4]. Currently, B-ISDN interfaces support up to 622 Mbps, with the
possibility of defining higher rates in the future.
The concepts of B-ISDN are summarized in ITU-T (CCITT) Recommendation
1.121 [5]:
B-ISDN supports switched, semipermanent, and permanent, point-to-point
and point-to-multipoint connections and provide on-demand, reserved, and
permanent services. Connections in B-ISDN support both circuit mode and
packet mode services of a mono and multimedia type and of a
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
connectionless or connection-oriented nature and in bidirectional and
unidirectional configurations.
A B-ISDN will contain intelligent capabilities for the purpose of providing
service characteristics, supporting powerful operation and maintenance
tools, network control, and management
Accordingly, B-ISDNs will support services with both constant and variable bit
rates and connection-oriented and connectionless transfers. At least conceptually, BISDNs not only support all types of communication applications that we know of, but
also provide the framework to support future applications that we do not fully understand
today.
ITU-T (CCl l l ) classified possible broadband applications into four categories
[6]:
1. Conversational services.
2. Retrieval services.
3. Messaging services.
4. Distribution services:
a. without user-individual presentation control;
b. with user-individual presentation control.
Conversational, retrieval, and messaging services are interactive services whereas
distribution services are classified into two subcategories according to the user interaction
for presentation control.
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
10
Considering these applications, B-ISDNs will support interactive and distributive
services, bursty and continuous traffic, connection oriented and connectionless services,
and point-to-point and complex communications, all in the same network. The types of
services that B-ISDNs are envisioned to offer can be characterized by one or more of the
following attributes:
•
High bandwidth;
•
Bandwidth on demand;
•
Varying quality of service parameters;
•
Guaranteed service levels;
•
Point-to-point, point-to-multipoint, or multipoint-to-multipoint connections;
•
Continuous or variable bit-rate service;
•
Connection oriented or connectionless services.
Accordingly, a B-ISDN should be capable of assigning usable capacity
dynamically on demand. B-ISDN switch fabrics should be capable of switching all types
of services. Furthermore, the network should take the bursty nature of some applications
into consideration in allocating the available bandwidth while guaranteeing the quality of
service for continuous bit-rate (CBR) services.
The introduction of highly reliable fiber systems into the access network provides
the necessary high bandwidth required for B-ISDN. However, there are a number of
issues that need to be resolved before B-ISDN networks can become a reality. As
technology advances rapidly to meet the need for high-speed communications, the
bottlenecks in communication networks are moving from the transmission media to the
communication processors. The throughput and end-to-end delay requirements of
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
11
applications are now limited by the processing power at network nodes, necessitating
fast network protocols. The suitability of current network protocols in B-ISDN has not
been fully addressed in the standardization committees. Congestion control is another
major area that needs to be satisfactorily addressed. B-ISDNs will support a very large
number of connections simultaneously in the network. Simple call admission as in
today's telephone systems, or hop-by-hop flow control used in current packet networks
can no longer be effective or used in B-ISDNs.
One issue that is resolved is the transfer mode: ATM is the transport mode of
choice for B-ISDN by ITU-T (CCITT). ATM is a packet-oriented switching and
muluplexing technique that uses short fixed-size cells to transfer information over a BISDN network. The short cell size of ATM at high transmission rates is expected to offer
full-bandwidth flexibility and high-bandwidth utilization, and provide the quality of
services required by applications with a wide range of performance requirements through
statistical multiplexing. The term asynchronous is used to reflect the fact that the cells of
an information unit may appear at irregular intervals over the network links. ATM is an
attempt to utilize the properties of both the packet and circuit-switched networks in an
integrated network.
2.1.2
Asynchronous T ransfer Mode
ATM is the transport mode of choice for B-ISDN. The transfer mode specifies
how information supplied by network users is eventually mapped onto the physical
network.
2.1.2.1
T ransfer Modes
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
12
ITU-T (CCITT) defines the transfer mode as a technique used for transmission,
multiplexing, and switching aspects of communication networks. Classifying the
communication networks according to the architecture and techniques used to transfer
data, the most commonly used types of networks can be categorized as follows:
•
Circuit-switched networks;
•
Message-switched networks;
•
Packet-switched networks:
•
Datagram packet switching;
•
Virtual-circuit packet switching.
Circuit switching is not flexible enough to support all B-ISDN applications,
where different types of applications have significantly varying bandwidth requirements.
Since each channel has a fixed bandwidth, it might be possible to define a basic channel
and allocate as many channels to different applications as they require. However, this
scheme introduces the problem of synchronization between the channels of a connection.
Furthermore, the selection of the basic channel bandwidth is a rather complicated issue.
If the basic rate chosen is too small, then, in addition to the synchronization problem
between channels of single connections, management of a large number of channels
complicates the implementation. If, on the other hand, a large bandwidth is assigned as
the basic rate, then a large amount of bandwidth would be wasted.
The disadvantage of the message switching transfer mode is that it is not well
suited for real time or delay-sensitive applications, such as voice, since the delay in the
network is rather unpredictable.
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
Packet switching is an attempt to combine the advantage of both circuit switching
and message switching. It is essentially the same as message switching, except that the
size of the information unit transmitted in the network is restricted to some maximum
value, which is on the order of few thousand bytes. Accordingly, user messages may be
segmented into packets before they are transmitted. Doing so allows the overlapping of
reception and transmission of packets of a message, thereby reducing the end-to-end
delay of messages, as the complete information unit has to be received at a node before it
can be processed and transmitted. The main disadvantage of packet switching is that
transmitting a message requires more overhead (i.e., packet header) per message than in
message switching, thereby reducing effective resource utilization in the network.
Essentially, each packet in the network is treated as an independent entity.
However, the packets need to be reassembled at the receiver to form the original
information unit before being passed to the user. Two approaches are used to handle
packet streams in the network.
In the datagram approach, each packet is treated independently and may follow
different paths to its destination. The main disadvantage of this approach is that packets
may arrive at the destination out of sequence, and sequencing packets to form
information units is a processing-intensive operation. Alternatively, end-to-end logical
connections or virtual-circuit connections can be established, similar to circuit switching,
before the transmission can start, and all packets of a message may follow the same path
in the network. This guarantees the sequential delivery of packets to the receiver but
requires a call set-up phase. ATM is the transfer mode of choice, and it is essentially a
connection-oriented packet switching where all packets are of fixed length. In general,
the chosen transfer mode of B-ISDN is envisioned to have the following properties:
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
14
1. Support all existing services as well as those with yet unknown characteristics
that will emerge in the future.
2. Utilize network resources as efficiently as possible.
3. Minimize the complexity of switching.
4. Minimize the processing time at the intermediate nodes to be able to support very
high transmission speeds.
5. Minimize the number of buffers required at the intermediate nodes to bound the
delay and the complexity of buffer management.
6. Guarantee performance requirements of existing and expected applications.
ATM is an attempt to meet all of these objectives in a unique manner. Compared with
other transfer modes, ATM has various features that extend the capabilities of current
packet-switching networks towards incorporating the most desired features of circuit
switching to support real-time traffic most efficiently.
In ATM, user information is transmitted between communicating entities using
fixed-size packets, referred to as ATM cells. An ATM cell is 53 bytes, consisting of a
48-byte information field and a 5-byte header. The ATM header includes minimal
functionality to reduce the intermediate node processing. Various applications with
different bandwidth requirements are easily supported, as bandwidth in ATM networks
is given on demand, as long as there are sufficient resources to support the application in
the network. ATM is a connection-oriented technique that provides an efficient means to
guarantee the quality-of-service requirements of applications. However, connectionless
services can also be supported relatively easily in an efficient manner. Due to fixed and
small cell sizes, buffer management and switching fabric designs are simplified.
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
15
Furthermore, buffer sizes at intermediate nodes are expected to be small, bounding the
cell delay.
It has long been argued how well ATM addresses the challenges of B-ISDNs
services, and at what expense. Independent of how valid the arguments made against it
are, ATM is the transfer mode of choice for B-ISDN, and, most likely it will be around
for several years to come.
Bit
I
1
...
Bit
4 5
GFC
VPI
VPI
VCI
1
4 5
VPI
1
VCI
VCI
Byte
VCI
VCI
VPI
PT
C
L
P
Byte
VCI
PT
HEC
HEC
Information field
Information field
53
53
(a)
(b)
Figure 2.1 (a) UNI cell format and (b) NNI cell format.
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission.
16
2.1.2.2
ATM Cell Header Fields
The ATM cell header consists of the following fields: generic flow control
(GFC), virtual path identifier (VPI), virtual channel identifier (VCI), payload type (PT),
cell loss priority (CLP), and header error control (HEC). The header format is different
at a B-ISDN user network interface (UNI) than it is in a B-ISDN network node interface
(NNI), as illustrated in Figure 2.1.
GFC is a four-bit field providing flow control at the UNI for the traffic originated
at user equipment and directed to the network, and does not control the traffic in the other
direction (i.e., network-to-user traffic flow). The GFC field has no use within the
network and is meant to be used by access mechanisms that implement different access
levels and priorities. Accordingly, this field is used as a part of the VPI at NNIs,
providing enhanced path-identification capabilities. Two modes of operation are defined
for the GFC field: uncontrolled access and controlled access. The uncontrolled access
has no impact on the traffic that users send to the network. In the case of controlled
access, the flow rate of cells generated by users is controlled at UNI.
ATM is a connection-oriented technique, and virtual circuits are required to be
established between the end nodes before transmission can start. As with any other
packet-switching network, routing of cells is performed at every node for each arriving
cell. VPI, an 8 or 12-bit field, together with VCI, a 16-bit field, contains the routing
information of a cell. The two levels of routing hierarchies, virtual paths and virtual
channels, are defined in ITU-T (CCITT) Recommendation I. 113 as follows:
•
VC is a concept used to describe unidirectional transport of ATM cells associated
by a common unique-identifier value, referred to as the VCI.
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
17
•
VP is a concept used to describe the unidirectional transport of cells belonging to
VCs that are associated by a common identifier value, referred to as the VPI.
A VP is a collection of a set of VCs between two nodes in a B-ISDN. A
predefined route is associated with each VP in the physical network. Furthermore, each
VP has its own bandwidth, limiting the number of VCs that can be multiplexed on a VP.
VPIs, in general, are used to route packets between two nodes that originate, remove, or
terminate the VPs, whereas VCIs are used at the end nodes to distinguish between
different connections. There are three bits in the ATM header to define the payload type.
The CLP field of the ATM cell header is a 1-bit field used for cell-loss priority. Due to
the statistical multiplexing of connections, it is unavoidable that cell loss will occur in a
B-ISDN. A cell with the CLP bit set may be discarded by the network during
congestion, whereas cells with the CLP bit not set have higher priority and will not be
discarded, if at all possible.
The HEC field is used mainly for two purposes: to discard cells with corrupted
headers and for cell delineation. The 8-bit field, when used for header error check,
provides single-bit error correction and a low-probability corrupted-cell delivery
capability. The field is also used to identify the cell delineation. The HEC value is equal
to the remainder of the division of the product x8 and the polynomial of order 31 with
coefficients being equal to the bits of the first four bytes of the header by the polynomial
x8+ x 2 + x + l .
2.1.2.3
ATM Protocol Reference Model
The B-ISDN protocol reference model (PRM) with ATM, as defined by ITU-U
(CCITT), is illustrated in Figure 2.2. This figure illustrates that the transfer mode is
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
18
ATM. The ATM adaptation layer (AAL) is service specific and at a high level consists of
two parts: continuous bit-rate (CBR) and variable bit-rate (VBR) services. For VBR
services, the layer is further divided into two sublayers: convergence, and segmentation
and reassembly (SAR). Higher layer functions are application specific and classified into
three categories: signaling, connectionless, and connection-oriented services.
Plane management functions
Signaling
and
control
CLNS
data
CONS
data
Video
Voice
Convergence
SAR
CBR
ATM
Access control
Physical layer
Figure 2.2 ATM protocol reference model.
2.1.2.4
ATM Layer
The ATM layer is common to all services and provides cell transfer capabilities.
That is, the ATM layer corresponds to the boundary between functions devoted to the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
19
header and those devoted to the information field. The characteristics of this layer are
independent of the physical medium used. The ATM layer provides cell multiplexing,
demultiplexing, and routing functions using the VPI and VCI fields of the cell header.
Furthermore, the ATM layer may supervise cell flow to ensure that connections stay
within the limits negotiated at the call establishment phase. It takes corrective actions to
make sure that the service qualities of connections that stay within the negotiated
parameters are not affected by connections that do not. The ATM layer is also
responsible for the cell sequence integrity for each source. No retransmission of lost or
errored cells is performed at this layer.
ATM is a connection-oriented technique requiring that end-to-end connections be
established before cells carrying user information can start flowing. Cell relaying
performed at intermediate nodes in the network forwards cells from one ATM entity to
another. Cells can be relayed from one VP to another or one VC to another, either in the
same or a different VP. The routing information used at a switching node has a local
meaning only. An explicit addressing with an end-to-end significance cannot be used is
mainly due to the short fixed-routing field size at the cell header. Switching from
incoming links to outgoing links is done by reading the routing field(s) of incoming
cells, and delivering the cells to the corresponding outgoing link port with the new
headers. The table entries at each node are written at the connection-establishment phase
for each connection. Two levels of connections are defined: virtual channel connections
(VCC) and virtual path connections (VPC).
A VCC is an end-to-end connection defined by the concatenation of the VC links,
which in turn are defined by the routing table entities of two nodes connected by a pointto-point physical link.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
20
The use of the VCC approach requires the use of large routing tables and imposes
a large amount of processing overhead for setting up connections. To reduce this burden,
VCs ate bundled into VPs and switched together by using the same VPI for each
connection in the bundle. Conceptually, a VP is the same as a VC, except that a VP
connection is defined over two or more physical links. VPs are semipermanent
connections, and routing tables for VP switching are preset by network management
functions. Hence, conceptually, a VP may be viewed as a link consisting of two or more
physical links.
In addition to routing, the ATM layer performs a cell multiplexing function,
which aggregates cells from different connections into a composite flow of cells and
marks the payload types at the PT field.
The ATM layer interfaces with the physical layer through a physical service
access point (PHY-SAP) using the request and indicate primitives. The ATM entity
passes one cell per request and accepts one cell per indicate primitive.
The ATM layer also interfaces with the ATM adaptation layer (AAL) through
ATM-service access points (ATM-SAP) similar to two PHY-SAP primitives. "ATMDATA.request" initiates the transfer of an ATM-service data unit (ATM-SDU) and its
associated SDU type to its peer entity over an existing connection. ATM-SDU is 48
bytes of user data including an AAL header to be transferred by the ATM layer between
peer-communicating upper layer entities. The loss priority and the SDU-type parameter
are used to assign the proper CLP and PT fields to the corresponding ATM-physical data
unit (ATM-PDU) generated at this layer. "ATM-DATA.Indication" indicates the arrival
of an ATM-SDU over the existing connection, along with a congestion indication (which
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
21
is set if the received ATM-SDU has passed through one or more network nodes
experiencing congestion) and the received SDU type.
ATM layer management functions at the user network interface are summarized in
two categories: fault management and traffic management.
2.1.2.5
ATM Adaptation Layer
The ATM layer deals with the functions of the cell header independent of the
information unit structure and the bit rate of the supported applications. This simplicity
and the flexibility of the ATM layer is achieved by leaving out various services required
in providing the quality-of-service requirements of the B-ISDN applications. In
particular, at the ATM layer, there is:
•
no information on the frequency of the service clock;
• no detection for misinserted cells;
• no detection for lost cells;
• no means to determine and handle delay variation;
• no awareness on the content of user information.
The main reason for not providing these functions at the ATM layer is that not all of these
services are required by every B-ISDN application. The role of the ATM adaptation layer
(AAL) is to provide, for each service class, the functionalities required in reaching their
desired quality of service.
AAL supports higher layer functions of the user and control planes and supports
connections between ATM and non-ATM interfaces. Information received by the AAL
from higher layers is segmented or collected to be inserted into ATM cells. Cells received
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
22
by the AAL from the ATM layer are reassembled to form the information or read out For
presentation purposes, the AAL functions can be classified into two categories:
continuous bit-stream oriented (CBR) services adaptation functions and bursty data
services adaptation functions. CBR services are those which require uninterrupted flow
of digital information, for example 64-Kbps voice services. Examples of AAL CBR
functions are:
•
Cell assembly / disassembly;
•
Variable delay compensation;
•
Mapping control signals into ATM cell stream;
•
Clock recovery;
•
Loss cell handling.
Although CBR services can be used to support current data services, doing so
does not take into account the bursty nature of services that are not continuous bit-stream
oriented. In particular, CBR services do not take advantage of idle periods between
information transmission. Accordingly, bursty data services are designed to provide
bandwidth savings by taking the bursty nature of such applications into consideration.
The functions performed by the AAL bursty data services include:
•
Segmentation of information units into cells;
•
Handling partially filled cells;
•
Reassembling cells to form information units;
•
Action on lost cells.
In general, AAL consists of the two sublayers illustrated in Figure 2.3: segmentation and
reassembly (SAR) and convergence sublayer (CS).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
23
Service specific
(may be null)
Convergence
sublayer
Common part
convergence sublayer
Segmentation and
reassembly sublayer
Segmentation and
reassembly sublayer
Figure 2.3 AAL structure.
The transmit side SAR layer receives CS-PDUs and segments them (or collects
them in case of CBR services) so that when the SAR header/trailer is added to PDU the
final payload fits into one ATM cell (i.e., 48 bytes). On the receiving end, the SAR layer
reconstructs the CS-PDUs from received cells and passes them to the CS. The CS
sublayer is subdivided into common part CS (CPCS) and service specific CS (SSCS).
The CPCS performs functions common to all AAL users, such as multiplexing and loss
cell detection. The service-specific requirements of different classes of users, for
example, timing recovery for real time applications, are implemented in the SSCS. For
services that do not require any specific function, the SSCS may be null.
The three sets of requirements of B-ISDN services used to classify AAL
functions are defined by the ITU-U (CCITT):
•
Time relation versus no time relation between source and destination;
•
Constant versus variable bit rate;
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
24
•
Connection oriented versus connectionless services.
This classification attempts to categorize all possible B-ISDN services into eight different
classes, of which four are explicidy defined:
•
Class A. This class corresponds to constant bit-rate connection-oriented services
with a timing relation between source and destination. The two typical services of
this class are 64-Kbps voice and constant bit-rate video.
•
Class B. This class corresponds to variable bit-rate connection-oriented services
with a timing relation between source and destination. Variable bit-rate encoded
video is a typical example of this service class.
•
Class C. This class corresponds to variable bit-rate connection-oriented services
with no timing relation between source and destination. A typical service of this
class is connection-oriented data transfer.
•
Class D. This class corresponds to variable bit-rate connectionless services with
no timing relation between source and destination. Connectionless data transfer
between two local area networks over a wide area network is a typical example of
this type of service.
Corresponding to these four classes, five types of AAL protocols are currently being
defined by ITU-T (CCITT), with other types of protocols possibly to be defined in the
future. The relationship of five AAL types to the four service classes are:
•
AAL type 1: (class A) connection-oriented, constant bit-rate service with timing
relationship between the sender and the receiver;
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
25
•
AAL type 2: (class B) connection-oriented, variable bit-rate service with timing
relationship between the sender and the receiver;
•
AAL type 3: (class C) connection-oriented, variable bit-rate service with no
timing relationship between the sender and the receiver;
•
AAL type 4: (class D) connectionless, variable bit-rate service with no timing
relationship between the sender and the receiver;
•
AAL type 5: (class D) connectionless, variable bit-rate service with no timing
relationship between the sender and the receiver.
2.2
ATM Switches
Conceptually, ATM networks are packet-switching networks, in that each ATM
cell in the network is transmitted independently; and connection oriented, as end-to-end
connections are established before the cell transfer can start. In simple terms, then, an
ATM switching node transports cells from the incoming links to the outgoing links using
the routing information at the cell header and information stored at each switching node
by the connection set-up procedure. In particular, a connection set-up task mainly
performs two functions at each switching node. (1) For each connection, it defines a
unique connection identifier to be used at the outgoing link. (2) It sets up routing tables at
each switching node to provide an association between the incoming and outgoing links
for each connection.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
VPI and VCI are the two connection identifiers used in ATM cells. In order to
uniquely identify each connection, VPIs are uniquely defined at each link and VCIs are
uniquely defined at each VP. The first step in establishing end-to-end connections is
determining a path from source to destination. At the end of this step, the sequence of
links to be used for the connection and their identifiers are known. In the case of pure
VC switching, a sequence of messages is exchanged between every neighbor node along
the path to set up a routing table entry associated with the connection. The entry basically
maps the incoming link identifier, VCI to outgoing link identifier, VCI. Consider three
neighbor nodes j - 1, j , and j + 1. The routing table entry at node j is set as follows.
Node j —1 sends a message to node j that includes information on the incoming and
outgoing links. Node j assigns a VCI to the connection and creates (incoming link
identifier, VCD part of the identifier. Node j then sends two messages: one to node
j —1 defining the VCI value used at the incoming link (which corresponds to an
outgoing link at node j - 1) and another one to node j +1 that includes information on
the incoming and outgoing links. Node j + 1 replies to node j , which includes the
required information to set up the (outgoing link identifier, VCD part of the table entry.
Once the routing tables at all nodes along the path are set, cells start to follow. At each
node, the VCI, together with the link it is coming from, is used for each arriving cell to
determine the outgoing link and the (new) VCI.
In the case of VP / VC switching, the procedure is similar to that of VC
switching, with the main difference being the number of routing tables set per
connection. VPs are semipermanent connections, in that routing tables for each VP are
preset by network management functions.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
27
In either type of switching, the connection identifier(s) at the cell header is read, a
table look-up is performed to determine the outgoing link and the connection identifier
value(s), the identifier(s) is updated to its new value(s), and the cell is switched from the
incoming link to the outgoing link.
An IV by IV switch can be viewed as a black box with IV input ports and IV output
ports that transports cells from any incoming link to any outgoing link. Incoming links
are connected to a switch fabric through input ports. After the cell header is processed to
determine its outgoing link, it is passed to the switch fabric, to be delivered to its
outgoing link. The interface between the switch fabric and the outgoing link is referred to
as the output port.
When more than one cell attempts to access an output port simultaneously, a
phenomenon called output conflict or contention occurs. When this occurs, only one of
the contending cells can be read out by the link. Others may either be stored in a buffer
temporarily until they can be read out or dropped. In output buffering, these cells are
stored between the switch fabric and the output port. In input buffering, storage is
provided between the incoming link and the input port, where cells that are blocked due
to output contention are kept until they can be delivered to the output ports. If the input
buffer is a FIFO, a blocked cell at the head of the queue blocks all other cells that may be
waiting in the queue. In particular, head of line (HOL) blocking occurs when some of the
cells that might be waiting in the queue, which could otherwise be switched to their
destination ports, are forced to wait for the HOL cell to be transmitted first. Finally,
internal blocking may occur if more than one cell contend for the same resource
simultaneously within the switch fabric. Similar to the case of output contention, all but
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
28
one of the contending cells may be temporarily stored at buffers, either within the switch
fabric (referred to as internal buffering) or at input buffers.
Different applications require different types of connections in ATM networks.
For example, a voice service may take place between two end points, whereas more than
two end nodes participate in teleconferencing. In general, four types of connections can
be defined:
1. Point-to-point connections, where the connection is established between two
entities. Most current services fall into this category.
2. Point-to-multipoint connections, where the cell stream generated by a source
node is distributed to two or more nodes. A typical example of this type of
connection is video distribution in which a video server serves multiple
destination nodes.
3. Multipoint-to-point connections, where information generated by more than one
node is transmitted to a single node. Data collection centers where data generated
at various sites are collected to a central location, such as in banking, require this
type of connection.
4. Multipoint-to-multipoint connections, where communications take place between
a group of users in which more than one user can simultaneously originate
traffic, to be distributed to all group members. Teleconferencing is a typical
application that requires this type of connection.
In terms of switching requirements, then, a cell is either switched point-to-point from an
incoming link to an outgoing link or point-to-multipoint from one link to a number of
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
29
outgoing links. The latter is referred to as multicasting or broadcasting if an incoming cell
is transported to every outgoing link. The other two types of connections, multipoint-topoint and multipoint-to-multipoint, can then be established as multiple instances of these
two types of connections.
With advances in transmission technology providing the bandwidth required to
support B-ISDN applications, the challenge is now to design switches that can deal with
several hundreds of millions of cells arriving to a node per second and to support
different types of connection requirements of applications.
Various ATM switches with different architectures have been proposed in the
literature. Following the classification of Tobagi [7], switch fabrics proposed for ATM
networks are classified into three categories:
•
Shared medium architecture.
•
Shared memory architecture.
•
Space division architecture.
2.2.1
Shared Medium A rchitectures
In a shared medium architecture, incoming cells are multiplexed into a common
medium, typically a bus or a ring. The medium speed, in general, is greater than or equal
to the sum of the transmission rates of the incoming links attached to it. Then, a small
FIFO that has a capacity to hold only a few cells is sufficient to store incoming cells until
they can access the medium. Output contention cannot occur in this architecture, as two
or more cells cannot arrive at an output port simultaneously. However, the arrival rate of
cells at a particular outgoing link may exceed the link bandwidth for a short period of
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
30
time, and output buffers are used to store cells that arrive at a faster rate than they can be
served.
Each output port is assigned a unique address. Once the outgoing link of an
arriving cell is determined, the output port address is added to each cell before it is
passed to the medium. This address is decoded by each output port interface to the
medium and filtered to determine whether to copy the cell to the output port or not. Cells
addressed to a particular output port are then copied to the output buffer to be read out by
the transmission link.
Shared medium architectures naturally support multicast / broadcast and perform
well when the medium speed is greater than or equal to the sum of the transmission rates
of incoming links attached to it. As the number of links attached to it and their speeds
increase, running shared medium at such very high rates may no longer be
technologically feasible, and the medium speed becomes the bottleneck. Accordingly,
shared medium architectures do not scale well and can support only a relatively small
number of ports. Alternatively, they can be the switching elements of large switches in
which each unit is connected to others according to some topology.
2.2.2
Shared Memory Architectures
A shared memory switch consists of a single dual-ported memory module shared
by all input and output ports. Incoming cells are multiplexed into a single stream and are
written to the shared memory. The memory is organized into logical queues, one for each
output port Cells at the output queues are also multiplexed into a single stream, read out,
demultiplexed, and transmitted on the output lines. In this architecture, the main
bottleneck is the memory access time to support both incoming and outgoing traffic.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
31
The memory can logically be organized into either full sharing or complete
partitioning. In the former, the entire memory is shared by all output ports and an
arriving cell is dropped only when the memory is full. An upper bound is imposed in the
latter for the number of cells waiting in the queue of each output port and a cell is
dropped if this limit is reached at a particular queue even though there is space available
in the memory. Full sharing provides a better cell-loss probability than complete
partitioning by utilizing the memory more efficiently (i.e., if there is a space, then the cell
is accepted). However, this scheme may not be fair at times when a burst of cells arrives
at a particular output port, reducing the space available and eventually causing service
degradation for other ports.
2.2.3
Space Division Architectures
There are two major drawbacks to shared medium and shared memory switch
architectures. First, multiplexing is required at the input side and demultiplexing at the
output side of the switch, restricting the scalability of the switch to support a large
number of ports. Second, buffer management and control functions are often centralized,
which increases the complexity of the switching node.
In space division switching, multiple cells from different input ports can be
transferred concurrently on multiple links. Each cell transfer requires the establishment of
a dedicated physical path through the switch from incoming to outgoing links. These
switches also allow the control to be distributed within the switch, thereby reducing its
design complexity.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
32
The basic building block in a space division switch is a cross point that can be
enabled or disabled by a control unit As shown in Figure 2.4, each cross point has two
input ports and two output ports and allows concurrent activation of two separate paths.
Input 2
Output 2
Input 1
Output 1
Bar state
Cross State
Figure 2.4 A cross point and its states.
Output contention in a cross point occurs when the two input ports
simultaneously request connection to the same output port If this occurs, one of the two
contending cells is granted access to the output port, whereas the other may either be
dropped or buffered temporarily until the port becomes available.
When buffers are used, they may be placed at the input ports or within the cross
point In either case, the buffer size is finite and the use of buffers does not totally solve
the problem of output contention. In particular, it is possible that the buffers may become
full, causing cells to be dropped due to the lack of space to store incoming cells.
Figure 2.5 illustrates an 8 by 8 crossbar switch, where each square corresponds
to a cross point. In general, N 2 cross points are used in an N by N crossbar switch.
Connection from any input port / to output port j is accomplished by engaging the
corresponding cross point (i.j) in the N by N matrix. As long as cells at each input port
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
33
are destined for different output ports, the crossbar switch allows N connections to be
simultaneously established, thereby achieving concurrent delivery of N cells.
Accordingly, it is an internally nonblocking switch. The main disadvantage of crossbar
switches is that the switch complexity grows with N 2, of which at most N are used at
any given time. Furthermore, there is a unique path between any input and output port,
and the loss of a cross point would prevent the connection between the two ports
involved.
1
2
3
4
5
6
7
Output
8 ports
Figure 2.5 An 8 by 8 crossbar switch.
2.2.4
Some Examples of ATM Switch A rchitectures
Various ATM switch schemes have been proposed for large N by N switches
with a small number of switch elements (smaller than N2). The main trade off is the low
throughput due to internal blocking, buffering, or sorting. One of the most famous ATM
switches is a Banyan network, originally introduced by Goke and Lipovski [8] in 1973.
The major property of a Banyan network is that there exists exactly one path from any
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
34
input to any output. Different subclasses of Banyan networks have been defined, of
which the Delta networks are the most famous. These Delta networks have the self­
routing property, i.e. independent of the input port at which, if the cell enters the Delta
network, it will always arrive at the correct output port
Turner [9,10] proposed a so-called St. Louis switch, which basically consists of
a copy network, a distribution network, a routing network and a number of broadcast
and group translators. All these networks are self-routing networks. The incoming cells
enter the copy network first, then the broadcast and group translators, distribution
network, finally the routing network and leave the switch. The copy network is
responsible for making the required copies of the incoming cells, such as in the case of
multicasting. The broadcast and group translators are responsible for the translation and
filling in of the routing header, that is, to translate the virtual path identifier or virtual
channel identifier to a real output address of the routing network. The distribution
network distributes the incoming traffic randomly over all its output ports, so that at its
output ports (also, the input ports of the routing network), the traffic is as much as
possible uniformly distributed over all links. Finally the routing network will ensure that
the cells will arrive at the correct destination, using the routing header of the cells.
A special class of ATM switches, which does not have internal contention of cells
passing through the switch, is called non-blocking switches. The most famous nonblocking switches are based on the Batcher-Banyan network topology. In 1984, Huang
and Knauer [11] proposed a Batcher-Banyan wideband digital switch called Starlite. It is
constructed out of self-routing, non-blocking interconnection networks. At the switch
input, unused cells are discarded by a concentrator network. A sort-to-copy and a copy
network are used to provide the multicast mechanism. A sort-to-destination and an
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
35
expander network route cells to their destinations. Figure 2.6 shows the Starlite switch
architecture. Later on, the Starlite approach was applied to an optical implementation, and
the problems with this optical switch were discussed in [12].
Input
v v v M / v v v v v v v v v v v
Concentrator
*
*
Sort to Copy
Copy Network
Sort (Batcher)
Trap
Expander (Banyan)
Output
V V V V V Y V V Y V y V V V V
Figure 2.6 Starlite switch architecture.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
36
Hui [13,14] proposed another switching architecture, also based on the sortrouting (Batcher-Banyan) concept, which was later called Moonshine. It uses another
principle than Starlite to avoid the output contention problem of the Batcher-Banyan
solution, occurring when multiple cells are destined to the same output port. It was
designed to handle variable length packets, so it is more general than a pure ATM switch.
To resolve the output contention, a 3-phase algorithm is proposed in combination
with input queuing at each input port, thereby causing HOL (head of line) blocking. The
effect of HOL blocking on the performance is rather negative, as is shown in [2]. The
maximum throughput is about 58% as for input queuing system. In addition, Moonshine
requires an internal speedup to perform the 3 phases during one packet time.
Sorting
Network
Banyan
Network
Figure 2.7 Moonshine switch architecture.
The problem of conflicting requests from a number of physically separated input
ports is resolved by 3 phases which are called the arbitration, acknowledgement and
sending phase. Figure 2.7 shows the architecture of the Moonshine switch. It consists of
a Batcher sorting network and a Banyan network. In Phase I (arbitration phase) of the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
37
algorithm, each input / sends a short request packet, which is just a source-destination
pair (/J.). The requests are sorted in nondecreasing order according to the destination
address y, by the Batcher sorting network. The request is granted only if j t is different
from the one above it in the sorted list, Thus it solves the output contention problem. In
Phase II (acknowledgement phase), the switch sends the sorted request packets back to
the input ports from the output ports of the Batcher sorting network via fixed
connections. The granted request sends the acknowledgement packet, which is a source
address i only, through the Batcher sorting network to sort and then the Banyan network
to reach to source input port i. Since each input port can send at most one request at
Phase I, there is no output contention at Phase II.
Input ports receiving
acknowledgements for their requests then transmit the full packet in the final Phase in
(sending phase) through the same Batcher-Banyan network, without conflict at the
output port. Input ports which fail to receive an acknowledgement retain the packet in a
buffer to retry in the next time slot, when the three phase cycle is repeated.
Another famous switch architecture call Knockout switch was proposed by Yeh,
Hluchyj and Acampora [15] for fixed-Iength packets (called Knockout I) and Eng,
Hluchyj and Yeh [16] for variable-length packets (called Knockout II) in 1987. The
architecture, illustrated in Figure 2.8, is based on a quasi-crossbar architecture. Each
input port drives a bus connected to all output concentrator / buffers. The Knockout
concentrator reduces the number of input ports to the output buffer from N to L
(L « N ). A shared buffer design, which works as a FIFO for the L buffers, precedes the
concentrator. As L is smaller than N, there is some cell loss in the concentrator.
The concentrator works based on the knockout tournament principle. Of the up to
N cells that enter the first level of the concentrator, only one winner takes the first spot.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
38
The remaining cells enter the second level of competition, and one winner takes the
second spot This procedure continues until L winners are determined. Each level of the
concentrator is made up of knockout elements that choose between two cells; the winner
goes to the next level of competition, the loser goes to the next stage to compete with
other losers.
Input
Interface
Modules
Input
Buses
Packet
Filters
N:L
Knockout
Concentrator
Bus
Interfaces
Shifter &
Shared Buffer
Output
Interface
Modules
N Output
Figure 2.8 Knockout switch architecture.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
39
2.3
2.3.1
THESEUS - A New ATM Switch
A rchitecture of THESEUS
Node 1
Node n
Node N
Node 1
star coupler
Node n
NodeN
Figure 2.9 Basic architecture of Terabit Hybrid Electro-optical
Self-routing Ultrafast Switch (THESEUS).
While the transmission medium is optical, currently available ATM switches are
exclusively electronic, and most are based on the shared memory concept. The
disadvantages of electronics include slow speed and complex circuitry, making large
ATM switch implementation prohibitively expensive. At present, the largest
commercially available switch is limited to 32 X 32. In order to make full use of the
optical bandwidth which extends up to 30 THz, many have proposed pure photonic
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
40
packet switching architecture using multiwavelength approaches or wavelength division
multiplexing (WDM). A major limitation of multiwavelength systems is the small number
of optical channels, typically less than 30 and possibly around 50 [17-19], making large
ATM switches impossible. Here a new hybrid electronics-optical architecture is
proposed. The simplification of processing overhead comes from the use of self-routing
fast tunable lasers. In addition, microwave subcarrier multiplexing (SCM) is
superimposed on the tuned optical wavelengths so that a large ATM switch is realizable.
SCM has an added advantage as a well-established dual-use (military / civilian)
technology with available commercial products.
As shown in Figure 2.9, the proposed switching architecture, Ierabit Hybrid
Electro-optical Self-routing LJltrafast Switch (THESEUS), combines and merges
electronics with optical technology [20], It uses tunable transmitters, fixed receivers and
a star coupler; thus it simplifies the protocols such that it eliminates the signaling process.
There are N optical carriers or nodes. Every carrier has M microwave subcarriers, which
makes the switch as large as NM x NM. For N equals 30 to 40 and M equals 20 to 60
[21, 22], The switch can be built as large as 1000 x 1000. Figure 2.10 shows the detail
of the node n. Each transmitting node is basically an M by 1 optical coupler. It combines
M optical input into one output; in turn, these M optical signals reach every receiving
node via an A by A star coupler. The M input ports of the node n consist of a buffer and
header processor, a selective microwave subcarrier transmitter and tunable optical
transmitter. Each incoming cell in the form of electrical data is first stored in a buffer.
The destination information of the cell header is then read to select the microwave
subcarrier by a subcarrier selector, and to tune the tunable laser via a tuning driver. The
cell's destination address can be split into two parts, the first part tuning the laser
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
41
wavelength and the second selecting the microwave subcarrier. The content of the header
is updated by the header processor according to the switching table. The optical carriers
are combined with the other ones at the node, and then sent to a star coupler to broadcast
to all the receiving nodes. For multicasting, the header processor makes necessary copies
with the new addresses and stores them in the buffer to be transmitted later. At the
receiving end, the signal is first demultiplexed optically, i.e. through a grating or a filter,
and detected by a photodetector. The RF signal from the photodetector is then split to M
microwave receivers through a power divider. The recovered baseband signal, which is
the original ATM cell with a new header is thus sent to the output link.
input n buffer &
header
p ro cesso r
subcarrier
selecto r
tuning
d river
microwave
subcarriers
tunable
laser
w a v ele n g th
sta b ilizer
feed b a ck
X n e 1 X1....A.N 1
T o star coupler-like
optical media
Transmitting Node n
m ic r o w a v e
h e te r o d y n e
receiver I
From
optical
m ed ia
narrow bandwidth
filter & receiver
baseband
signal
output nl
power
d ivid er
m ic r o w a v e
h e te r o d y n e
receiver M
output nM,
baseband
signal
Receiving Node n
Figure 2.10 Block diagram of node n of THESEUS.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
42
The main reasons for this architecture design are that the optical power budget
problem can be mitigated by using a smaller star coupler, and the number of costly
optical receivers needed is reduced to N, since the switch is geographically compact. For
the communication network application, an NM by NM star coupler is needed and every
user needs a pair of optical transmitter and receiver, as well as a pair of microwave
subcarrier transmitter and receiver, since a network is usually spread over a large area.
The proposed system does not have intermediate states; therefore, it has no
internal blocking. As for the destination contention, one may use input buffers to store
the contended cells and retransmit them until they succeed in transmission. The other
way is to use the traditional sort and trap networks which first detect cells with identical
addresses and then send them back to the input for re-entry in the next cycle. A more
detailed discussion about this will follow in the next chapter. Time division multiplexing
(TDM) is another option to avoid destination contention.
6.7 mA
c
o
3
O
9
a
c
0.22 nm (27GHz)
£
a>
c
a;
«
£
$
0 ns
10 ns
20 ns
30 ns
40 ns
50 ns
Time
Figure 2.11 Wavelength tuning response to a step input current. Trace 1 is the input
current. Traces 2 and 3 are the intensity outputs from the two sides of the Fabry-Perot
filter with positive and negative polarities, respectively, of wavelength-to-intensity
conversion. The average deconvolved rise time is 5.33 ns.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
subcarrier select control
I
I
2.2 GHzl-
2.2 G H zl
s
s
2.6 GHz I- t circulator amp
w
w To laser
comb
modulator •s i
1
3.0 GHz I
generator
(mixer)
t
t
c
c
- 3.4 GHz I
h
h
1
3.8 GHz I
data input
channel select filter bank
bandpass filter bank
(a) Subcarrier transmitter.
t
channel select conuol
From optical receiver
I
9 GHz L.O.
r l 4.3 GHzl-
preamp
- I 3.9 GHzlcomb
generator
- I 3.5 GHz
IF bandbass
filter 6.5 GHz]
2.8 GHz
H 5Z g5 3 "12.7 GHzlchannel select filter bank
data input
(b) Subcarrier receiver.
Figure 2.12 Block diagram of microwave subcarrier testbed.
ASK
demodulator
recovered
baseband
signal
''k
44
The other advantage of this architecture is that the speed of the switch is basically
limited by the speed of the processing of the header, the selecting of the microwave
subcarrier, and the tuning of the laser wavelength. The commercially available
microwave switches have a switching speed of 10 ns or less. N. K. Shankaranarayanan
et al.[23] have measured the tuning speed of three-section distributed Bragg reflector
(DBR) lasers. They claimed that the tuning rise time is about 5 ns. Figure 2.11 shows
their measurement result. With a 5 ns tuning current, the wavelength tuning speed is
about 10 to 15 ns, which is confirmed by Glance and Kobrinski [18,19]. For 1 Gbps
data rate, the 5 byte ATM header provides 40 ns to process and select the subcarrier and
tune the wavelength. Thus, this architecture is suitable for multirate ATM networks.
2.3.2
Testbed of THESEUS
A microwave subcarrier testbed has been constructed with amplitude-shift keying
(ASK) modulation for microwave subcarriers and direct intensity modulation for the
optical carriers, and with direct optical and heterodyne microwave detection. Figure 2.12
shows the block diagram of one microwave subcarrier transmitter and receiver. The
system uses a comb generator as a microwave source, and uses filters and switches to
select a microwave subcarrier instead of a simple voltage controlled oscillator because of
stability, tuning speed and ease of control. The comb generators generate microwave
signals from 100 MHz to 18 GHz with 100 MHz comb spacing. The switches, both
used in the transmitter and receiver, have a 100 ns switching speed. The filters used in
the channel select filter banks have 3 dB bandwidth of 50 MHz, which guarantees a
single microwave signal per channel since the comb spacing is 100 MHz. The center
frequencies of the filters are chosen as 2.2, 2.6, 3.0, 3.4, and 3.8 GHz. After the
subcarriers are amplified, the data steams are modulated via an amplitude-shift keying
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
45
(ASK) scheme through the mixers. The mixer is reverse used such that the subcarrier is
injected to the local oscillator (LO) port, baseband data is injected to the intermediate
frequency (IF) port to turn the diodes on and off, and the output of the modulated
subcarrier is taken from the radio frequency (RF) port The modulated subcarriers are
further filtered by the bandpass filter bank with 360 MHz 3 dB bandwidth to prevent the
adjacent channel crosstalk, even though it might cause some bandwidth limiting
distortion. Since the baseband data used in the testbed is lower than 150 bits per second
(bps), this kind of distortion is very limited. The subcarriers are further amplified to 7
dBm and directly injected into the Ortel lasers for intensity modulation with about 25%
modulation depth. The Ortel lasers are operating around 1310 nm and have modulation
bandwidth of 10 GHz and relative intensity noise (RIN) less than -140 dB/Hz. The
optical carriers are then sent to the star coupler to broadcast to all the receiving nodes.
At the receiving end, the optical carrier intended for local reception is first filtered
by a fixed narrow band optical filter, and is then detected by the Ortel 10 GHz receiver.
The RF signal from the optical receiver is about -45 dBm and is preamplified before it is
mixed with a local oscillator to upconvert to 6.5 GHz, as shown in Figure 2.12(b). The
local oscillator signal is obtained through a comb generator and a channel select filter
bank. The comb generator is the same as the one used in the transmitter. The filters are
centered at 4.3, 3.9, 3.5, 3.1, and 2.7 GHz, respectively, with 50 MHz 3 dB
bandwidth. By choosing a local oscillator, any one of the 5 subcarriers can be chosen for
reception. The bandpass filter is centered at 6.5 GHz with 300 MHz 3 dB bandwidth.
The chosen subcarrier is further downconverted with a local oscillator of 9 GHz and
filtered by a low pass filter of 2.85 GHz. The IF signal is then demodulated by an ASK
demodulator as shown in Figure 2.13.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
46
IF signal
power
divider
low pass
filter
recovered
baseband signal
delay line
Figure 2.13
ASK demodulator.
The ASK demodulator basically performs square function to recover the
baseband signal with a low pass filter. Suppose that the ASK modulated subcarrier has
the form of
l sc = A(l +m cos G)bst) cos 6)sct,
(2 . 1)
where A is the amplitude of the subcarrier, m the modulation index, 0)bs the angular
frequency of the modulating baseband signal, and 0)ie the angular frequency of the
subcarrier. By squaring it, it becomes
r->
I'sc
=
+
2
+ m~
—
- —
2 + m2
i
A *
.1
nr
■>
_
+m A~ coso)bst + — A ' cos2cobst
,
A" cos 2 corj
( 2 .2 )
+mA2[cos(2tyJC+cobs)t + 005 (2 ^
2
+
A 2[cos 2(coK
)r + cos 2(g)K- Q)bl )r].
Since cosc » G)bs, A low pass filter could be used to filter out second harmonic and high
frequency components. Equation (2.2) has only the dc and baseband terms as
2 + m2
-A2 +mA2 coso)bst.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.3)
47
Thus the baseband signal could be recovered. In the testbed as shown in Figure 2.13, the
OF signal from the downconverter is split into two paths, and then recombined at the
mixer to perform the square function. One path has a delay line, which is adjustable
manually, to ensure that the two input signals at the mixer have exactly the same phase.
The output of the mixer is then filtered by a low pass filter of 400 MHz to recover the
baseband signal. This low pass filter is very critical. If its bandwidth is too wide,
Equation (2.3) might include the second harmonic of the baseband signal such that
II
+ m A 2 +mA2 cos (Obst + ^ ~ A 2 cos2mfar.
2
4
(2.4)
transmitted
data
received
data
(a) 139 Mbps / 2.6 GHz subcarrier.
(b) 6.3 Mbps / 3.0 GHz subcarrier.
Figure 2.14 Two different microwave subcarriers over two lasers with the same
wavelength, (a) 139 Mbps / 2.6 GHz subcarrier, (b) 6.3 Mbps / 3.0 GHz subcarrier.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
48
Thus, the recovered baseband signal would have some second harmonic distortion
(SHD). Furthermore, the modulation index m plays an important role in the SHD, since
it is proportional to the square of the modulation index. If the modulation index can be
reduced low enough, this distortion can be almost eliminated.
(a) Modulation index m = 67.4%.
W J
(b) Modulation index m = 7.1%.
Figure 2.15 Received 6.3 Mbps data on 3.0 GHz subcarrier with
different modulation indices.
Figure. 2.14 shows the results of two channels with data rates of 139 Mbps on
2.6 GHz subcarrier and 6.3 Mbps on 3.0 GHz subcarrier. The ASK modulation indices
were 8.2% for 139 Mbps data and 67.4% for 6.3 Mbps data. Each subcarrier was then
modulated on a laser. The upper row is transmitted data and lower row is received data.
The received signals are slightly degraded with some noise and overshoot, especially for
the 6.3 Mbps signal. This might be mainly due to the SHD. Here a 400 MHz low pass
filter was used, whose bandwidth was large enough to include the second harmonic
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
49
signal. A 200 MHz low pass filter was also used instead, and no significant change was
observed. This was because that the bandwidth was still too wide for 6.3 Mbps data to
exclude the SHD. Even though it was narrow enough for 139 Mbps data, the received
data didn't have much SHD, since its modulation index was quite low.
The modulation index for the 6.3 Mbps data was changed from 67.4% to 7.1%,
and the received signal improved dramatically as shown in Figure 2.15. In Figure
2.15(a), the received data, of high modulation index of 67.4%, shows that the overshoot
is so large that it seems to split one bit into two or to double the data rate due to the
second harmonic distortion. Figure 2.15(b) shows a well recovered data with modulation
index of 7.1%. The bit error rate (BER) was measured for both channels. Even for
different modulation indices, both channels had less than 10-9 BER. Different data rate
signals of 1.5, 6.3, and 44.7 Mbps over the 3.0 GHz subcarrier were transmitted as
shown in Figure 2.16. The received 44.7 Mbps data has some degradation such as
overshoot in Figure 2.16(c) compared with the signals of 1.5 Mbps in Figure 2.16(a)
and 6.3 Mbps in Figure 2.16(b). The eye diagrams of 44.7 Mbps and 139 Mbps are
shown in Figure 2.17(a) and (b), respectively. The upper row is the transmitted data and
the lower row is the received data. The received 44.7 Mbps data has a quite symmetrical
open eye and little distortion compared to its original transmitted data. The 139 Mbps
data shows some distortions such as overshoot, rise and fall time degradation or time
jitter. It also shows a nonsymmetrical pattern, which indicates the nonlinearities of the
system. But nevertheless, the received data still has a quite open eye, which ensures an
error free (< 10"9 BER) data reception. Therefore, this switching architecture can be used
for multirate services in the packet switching networks or ATM networks, as long as the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
50
modulation index is kept low enough to reduce the second harmonic distortion effect, but
high enough for a reasonable reception.
(a) 1.5 Mbps.
(b) 6.3 Mbps.
i
(c) 44.7 Mbps.
Figure 2.16 Received data on 3.0 GHz subcarrier with different data rates.
However, there was no interference or crosstalk between these two channels,
since the signals were from two different lasers with the same wavelength but different
microwave subcarriers. Figure 2.18 shows the optical spectrum of the two lasers. Figure
2.18 (a) is from the Laser 1 with 3.0 GHz subcarrier and (b) is from Laser 2 with 2.6
GHz subcarrier. These are multimode lasers. For Laser 2 the highest mode is at 1.3012
(im. The nearest mode from Laser 1 is the seventh from left at 1.3010 (im. The mode
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
51
spacings are almost the same since these lasers are the same kind. Thus the smallest
wavelength difference is AA = 0.2 nm. The average wavelength at these two modes is
A = 1.2011 p.m. The corresponding frequency difference can be calculated from the
derivative of the equation, / = ^ , as
A/ = 4-AA = 35.44 GHz.
A
(2.5)
This frequency (which is beat interference,) is much larger than the 10 GHz frequency
range of the optical receiver. Since the photodetectors act like low pass filters, this beat
interference at the output of the optical receiver would not be received. This matter will
be discussed in more detail in the next chapter.
(a) 44.7 Mbps.
(b) 139 Mbps.
Figure 2.17 Eye diagrams of (a) 44.7 Mbps and (b) 139 Mbps data. The
upper row is transmitted data and the lower row is received data.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
52
100V:
l
le jm /d iu
i
l
AUR
t
i
*
•
l
l
;
;
;
;
t
•
•
•
•
a
•
a
•
•
•
•
a
a
a
|
:
i
i
:
:
■
a
a
•
■
•
•
•
•
;
50>. ;
:
:
:
:
•
a
i
i
•
a
i
•
l
»
•
•
•
a
a
a
•
•
|
•
•
•
*
•
♦
.
a
«
•
•
»
•
•
•
:
:
i
i
:
i
1
(a)
•
t
a
*
I
a
a
..................... i ..................... 1 . 1 . . . .
L ..L IX 1
:
1.2933
0
180:4
lO jftt/d iV
i
a
|
1-3033
1.3133
AUR
1
U
(b)
RES 0-In*
2iW div
Figure 2.18 Optical spectrum, (a) Laser 1 with 3.0 GHz subcarrier, (b) Laser 2
with 2.6 GHz subcarrier.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
53
2.3.3
2.3.3.1
Perform ance of Small ATM Switches
Introduction
The THESEUS architecture can be used for a small ATM switch or a local area
network (LAN). It has quite a different performance in terms of output contention.
Chiaroni et al. [24, 25] proposed a solution to output contention by using fiber
delay lines (FDLs) as optical buffers. But this technique complicates the protocols by
sorting the cells first before they enter the switch. Only then can the contended cells be
decided to enter the proper FDLs, corresponding to the proper positions in the queue of
the optical output buffer to meet the first in first out (FIFO) requirement. Therefore, this
technique increases the processing overhead, which in turn increases the time delay and
decreases the throughput.
The disadvantage of having complicated protocols can be overcome by using
THESEUS instead, whose microwave subcarriers are used as output buffers to resolve
the destination contention.
2.3.3.2
Solutions to O utput Contention
In the previous section, the implementation of THESEUS as a large ATM switch
was discussed, where the microwave subcarriers were used as part of the input and
output links. But for a small ATM switch or a LAN, the wavelength division optical
carriers are enough to link the input and output ports. That is, the number of the optical
carriers is equal to or larger than the number of the output ports. The microwave
subcarriers can be used as the output buffers to resolve the destination contention.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
54
The idea of contention solving is given as follows. If two incoming cells appear
at the two input ports 1 and 2 with the same destination address, the tunable lasers 1 and
2
will tune to the nominally same wavelength (assuming that the beat interference
frequency is much larger than the receiver bandwidth). Thus, the intended output port
will receive them at the same time, causing an output contention. But if the selection of
the microwave subcarriers 1 and 2 is made according to the source address, the intended
output port will read out one cell from subcarrier 1 first and store the other cell from
subcarrier 2 in the electronic buffer to read out later. Therefore the contention problem is
solved.
If the number of the microwave subcarriers is the same as the number of the
optical carriers, a perfect contentionless ATM switch is in hand. However, due to the
small modulation bandwidth of the lasers, the number of the microwave subcarriers is
limited to less than the number of the optical carriers. This is like the Knockout switch
situation, where the size of output buffer L is smaller than the number of input ports N.
Here we cannot have a knockout tournament to determine which subcarrier to choose,
since the simple protocol won’t know which cells are destined for the same output port.
This leads to two approaches. One is to divide N input ports into M groups, where M is
the number of microwave subcarriers. Any input port in group 1 will choose subcarrier
1, group 2 will choose subcarrier 2, and so forth. The other approach is to randomly
select any subcarrier from 1 to M for any input port regardless of any other input port's
behavior. The performance of these two approaches will be discussed in a given traffic
patter in the next subsection.
2.3.3.3
Com parison of Group and Random Assignments of Subcarriers
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
55
A slotted ALOHA access scheme [26, 27] is used to study the random
assignment of the microwave subcarriers versus group assignment to achieve higher
throughput, assuming simple immediate-first-transmission (IFT) protocol with no
capture and with a uniform traffic load. This analysis is based upon Markov chain
techniques and the use of combinatorial techniques. The analysis was originated by Yue
for the multichannel packet radio networks [26], and corrected by Zhang and Liu [27].
The system consists of N by N input and output ports, N wavelengths or optical
carriers, and M microwave subcarriers. The time is divided into slots of fixed length
equal to the transmission time of a single cell. All ports are synchronized, and all cell
transmissions are started only at the beginning of a time slot. Every input port has its
own buffer which can store at most one cell at any time. Once a cell is accommodated at a
buffer, it remains there until it is successfully transmitted. Each input port can transmit a
cell on a chosen subcarrier. It is assumed that the input port will know about its success
or collision immediately after the transmission. If two or more cells are simultaneously
transmitted on the same subcarrier, collision of the cells occurs. The input ports whose
transmission is unsuccessful retransmit their cells in future time slots. Therefore, each
input port is in either of two states: the thinking state, if it does not have a cell in its own
buffer to transmit; the backlogged state, if it has a cell awaiting or undergoing
transmission. A consecutive pair of thinking and backlogged periods is called a renewal
cycle. Each input port generates a cell destined for a particular output port with
probability 1/ N . To simplify the analysis, this probability is assumed to be one instead;
that is, all the input ports are sending cells to a particular output port. For the IFT
protocol, a thinking input port generates a new cell only at the beginning of a slot with
probability X, and when the new cell is generated, the port transmits the cell with
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
56
probability one. If the transmission is not successful, the port joins the backlogged state.
A backlogged input port cannot generate a new cell and retransmits the old cell according
to a geometrical distribution with mean 1/ pr slots.
When a collision occurs, the colliding cells are lost There is no recovery or
capture. Thus the capture restriction parameter u is set to be equal to one. The system
throughput 6 , which is defined as the average number of cells successfully transmitted
per slot, is given as in [27],
N N
N
miB(W^i)
®= *=0
X X
X
j=0* ; X#1=1 c,=1
/
;
f N — i\
\
4 \ n„ ' a) V ” ( i - p , r “ \
a„ J
<2-6>
where a = k - j + cs, cs is the number of successful subcarriers or the number of cells
successfully transmitted over the successful subcarriers, since only one cell is
successfully transmitted over one successful subcarrier, n is the number of cells that are
simultaneously transmitted out of N input ports. Jtj is the element of the (N + 1)dimensional row vector of the steady-state probability distribution n = [ k 0, tt, , •••, ;r„ },
and can be calculated by solving the set of linear equations, n = IIP, and 2 ^._0 Kj = 1•
Here the probability transition matrix P = (Pyt)is given by
N
P* = X
#1=0
m in (A f,/i) /
j
\
X
c ,= 0
k - ll- P .r i
\ n
N -f
a„
/ /V — / A
a j
(2-7)
wherethe conditional probability S(cJ/i) = Prob( cs cells are successfully transmitted I n
cells are simultaneously transmitted). This conditional probability can
be calculated
through recursive expressions as
|«, m) = 2L f ” Y 1- —]f —^1 S(cs - I u{ j ) \ n - j ,m - l ) ,
j=oUA
m 7Vmy
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.8)
57
where
0 if j = 0, or u +1 < j < N
1 if 1 < j < u
and S(cs \n,m) is the probability that there are cs cells successfully transmitted, given that
n cells are simultaneously transmitted over m subcarriers, and S(cf|n) = S(c,|n,M).
Therefore, with the initial conditions
S(0 |1,m) = 0 m > l and S(0 |1, 0 ) = 1,
S(l|l,m) = 1 m > 1 and S(l|l,0) = 0,
S(0|n, l) = 1n = 0 or n > u + l and 5(0|n, l) = 0 1 < n < u,
5(l|/i, l) = 0 n= 0 or n > u + 1 and S(l\n, l) = 1 1 < n < u,
5(% ,m ) = 0 k > n or k> m ,
5(-l|n, m) = 0 for any n, m,
The system throughput 6, or the average subcarrier utilization can be calculated as
(2.9)
and the average cell delay
(2. 10)
Assume a 16 x 16 ATM switch with 16 wavelengths and 4 microwave
subcarriers. It is also assumed that all 16 input ports are sending cells to a particular
output port through a particular wavelength, as assumed before. For the random choice
of subcarriers, N = 16, M = 4. For the group subcarrier assignment, the 16 input ports
are divided into 4 groups. Group 1 uses subcarrier 1, group 2 uses subcarrier 2, and so
on, therefore N = 4 and M = 1. The average subcarrier utilization and average cell delay
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
58
are calculated assuming pr =0.1 and u = 1 for no capture. Figure 2.19 shows the
average subcarrier utilization U versus traffic load X. The average cell delay £[D] in cell
length is shown in Figure 2.20. The solid lines represent the random subcarrier
assignment and the dashed lines represent the group assignment It is shown that the
group assignment of subcarriers has lower cell delay, and higher subcarrier utilization or
system throughput.
Since the cells arriving at the input ports are destined to all the output ports with a
uniform probability l/N, the actual arrival rate at any input port for a particular
wavelength is X/N. In the case mentioned above, the actual arrival rate is 0.0625A..
Therefore, the ATM is operating in the low cell arrival rate region of 0.1. The group
assignment of the subcarriers has 7% throughput and 19% cell delay improvements over
the random assignment
2.4
Summary
THESEUS is a self-routing wavelength-division / microwave subcarrier
multiplexed ATM switch for multirate ATM networks. It has the potential to be extended
to a 1000 X 1000 system without internal blocking. Among the advantages of the hybrid
implementation are flexibility in service upgrade, relaxed tolerances on optical filtering,
potentially low-cost components, protocol simplification, and less processing overhead.
For a smaller system, the SCM can be used to solve destination contention. The
calculation shows that the group assignment of the microwave subcarrier is better then
the random assignment in terms of throughput and cell delay.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
59
0.50-1
0.450.40-5 0 . 3 5 0.30-
random assignment: 16 input ports, 4 subcarriers
group assignment: 4 input ports, 1 subcarrier
0.25-
0 .2 0 0.2
0.6
0.4
0. 8
X
Figure 2.19 Average subcarrier utilization versus cell arrival rate.
1
■00
S
c
o
6
-
random assignment: 16 input ports, 4 subcarriers
group assignment: 4 input ports, 1 subcarrier
4 -
0. 2
0. 4
0.6
0. 8
X
Figure 2.20 Average cell delay versus cell arrival rate.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
60
R eferences
[1]
R. O. Onvural, Asynchronous Transfer Mode Networks: Performance Issues,
Boston: Artech House, Inc., 1994.
[2]
M. de Prycker, Asynchronous Transfer Mode: Solution fo r Broadband ISDN, 3rd
ed. London: Prentice Hall International (UK), Ltd., 1995.
[3]
M. Schwartz, Telecommunication Networks: Protocols, Modeling and Analysis,
Reading, Massachusetts: Addison-Wesley Publishing Company, 1987.
[4]
ITU-T (CCll'l) Recommendation 1.113, “Vocabulary of Terms for Broadband
Aspects of ISDN,” 1988.
[5] ITU-T (CCl lT) Recommendation 1.121, “Broadband Aspects of ISDN,” 1990.
[6 ] ITU-T (CCITT) Recommendation I. 211, “B-ISDN Service Aspects,” 1990.
[7]
F. A. Tobagi, “Fast packet switch architectures for broadband integrated services
digital networks,” IRE proc., vol. 78, pp. 133-167, 1990.
[8 ] L. R. Goke and G. J. Lipovski, “Banyan networks for partitioning multiprocessor
systems,” in Proc. 1st Annu. Int. Symp. Comput. Architectures, Dec. 1973, pp.
21-28.
[9]
J. S. Turner, “Design of a broadcast packet network,” in Proc. Infocom '86,
Miami, April 1986, pp. 667-675.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
61
[10] J. S. Turner, “Design of a broadcast packet switching network,” IEEE Trans.
Commun., vol. 36, pp. 734-743, June 1988.
[11] A. Huang and S. Knauer, “Starlite: A wideband digital switch,” in Proc.
GLOBECOM '84, Atlanta, GA, Dec. 1984, pp. 121-125.
[12] A. Huang, “The relationship between STARLITE, a wideband digital switch and
optics,” in Proc. ICC ’86, Toronto, Canada, June 1986, pp. 1725-1729.
[13] J. Hui, “A broadband packet switch for multi-rate services,” in Proc. ICC ’87,
Seattle, WA, June 1987, pp. 782-788.
[14] J. Y. Hui and E. Arthurs, “A broadband packet switch for integrated transport,”
IEEE J. Select. Areas Commun., vol. SAC-5, pp. 1264-1273, Oct. 1987.
[15] Y. S. Yeh, M. G. Hluchyj, and A. S. Acampora, “The Knockout switch: A
simple, modular architecture for high-performance packet switching,” IEEE J.
Select. Areas Commun., vol. SAC-5, pp. 1274-1283, Oct. 1987.
[16] K. Y. Eng, M. G. Hluchyj, and Y. S. Yeh, “A Knockout switch for variablelength packets,” IEEE J. Select. Areas Commun., vol. SAC-5, pp. 1426-1435,
Dec. 1987.
[17] C. E. Zah et al., “Monolithic integration of multiwavelength compressive-strained
multiquantum-well distributed-feedback laser array with star coupler and optical
amplifiers,” Electron. Lett., vol. 28, pp. 2361-2362, Dec. 1992.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[18] B. Glance, U. Koren, R. W. Wilson, D. Chen, and A. Jourdan, “Fast optical
packet switching based on WDM,” IEEE Photon. Technol. Lett., vol. 4, pp. 11861188, Oct. 1992.
[19] H. Kobrinski et al., “Fast wavelength-switching of laser transmitters and
amplifiers,” IEEE J. Select. Areas Commun., vol. 8 , pp. 1190-1202, Aug. 1990.
[20] W. Xin, Z. Zhang, and E. S. Yang, “THESEUS - A terabit hybrid electronicoptical self-routing architecture for large ATM switches,” 1995 IEEE/LEOS
Summer Topical Meeting Digest on Technobgies fo r a Gbbal Information
Infrastructures, August 7-11, 1995, Keystone, CO, pp. 24-25.
[21] R. Olshansky and V. A. Lanzisera, “60-channel fm video subcarrier multiplexed
optical communication system,” Electron. Lett., vol. 23, pp. 1196-1198, Oct.
1987.
[22] P. M. Hill and R. Olshansky, “A 20-channel optical communication system using
subcarrier multiplexing for the transmission of digital video signals,” J. Lightwave
Technol., vol. 8 , pp. 554-560, April 1990.
[23] N. K. Shankaranarayanan et al., ‘Dynamic wavelength tuning characteristics of a
three-section distributed Bragg reflector laser,” in Proc. OFC ’90, San Francisco,
CA, Jan. 1990, paper WM16.
[24] D. Chiaroni et al.,
“Rack-Mounted 2.5 Gbit/s ATM Photonic Switch
Demonstrator,” ECOC ’93, Post deadline paper.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
63
[25] F. Masetti, P. Gavignet-Morin, D. Chiaroni, G. Da Loura, “Fiber delay lines
optical buffer for ATM photonic switching,” in Proc. IEEE INFOCOM ’93, San
Francisco, CA, vol. 3, pp. 935-942, 1993.
[26] W. Yue, “The effect of capture on performance of multichannel slotted ALOHA
system,” IEEE Trans. Commun., vol. 39, pp. 818-822, June 1991.
[27] Z. Zhang and Y-J. Liu, “Comments on ‘the effect of capture on performance of
multichannel slotted ALOHA system’,” IEEE Trans. Commun., vol. 41, pp. 14331435, Oct.. 1993.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
64
Chapter 3
Heterodyne Optical Beat Interference Limitations
on THESEUS
3.1
Introduction
As mentioned in Chapter 2, different lasers operating at different wavelengths
will produce beat interference at the photodetectors, causing outage of the microwave
subcarriers which modulate the optical carriers. This subcarrier outage will severely
degrade the performance of THESEUS or WDM / SCM networks
In 1963, Magyar and Mandel [1] observed the beat interference fringes produced
by two independent ruby lasers, which were recorded by a high speed image tube with
40 ns exposure time. Desem [2-4] studied how the optical beat interference affected
microwave subcarriers. Shankaranarayanan et al. [5] investigated the statistics of the
microwave subcarrier outage due to the optical beat interference (OBI) in WDM / SCM
networks. Wood and Shankaranarayanan showed a significant bit error rate degradation
due to the OBI [6 ].
In section 3.2, the development of a theory to describe the heterodyne optical beat
interference and the calculation of the microwave subcarriers' signal-to-interference ratio
(SIR) are presented. Section 3.3 presents the experimental results, and the limitation on
the WDM / SCM network. In section 3.4, there will be a discussion of some aspects of
the improvements of THESEUS or WDM / SCM networks to overcome the optical beat
interference effect, based on the theoretical model [7, 8 ].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
65
3.2
Heterodyne Optical Beat Interference
In THESEUS or WDM / SCM networks, information is transmitted by
modulating a radio frequency subcarrier, which in turn is used to intensity modulate an
optical carrier. When the optical carriers from two or more lasers operating in a WDM /
SCM system are combined at a photodetector, optical beat interference occurs at a
frequency equal to the difference in the optical frequencies of the lasers. If this beat
signal occurs at a frequency near a subcarrier frequency, it will interfere with the
information being transmitted or cause subcarrier outage.
Assuming that lasers 1 and 2 are operating at wavelengths A, and A2. Each laser
is directly modulated by a microwave subcarrier f { and / 2, respectively. The subcarriers
have the power as
Px(t) = 1+ m, (r)cos27i/;r,
(3.1)
P2(t) = 1+ m2(r) cos ln f2t ,
(3.2)
where mx{t) and m2 (r) are the subcarrier modulation depths, and the functions of time
since the subcarriers are, in turn, modulated by the transmitted baseband information.
The electric fields of the two optical carriers have the form,
(3.3)
(3.4)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
phases of the optical carriers. Assume that the optical carriers have the same polarization.
At the photodetector, the electric field of the combined carriers has the form of
(3.5)
e{t) = ey{t) + e2{t).
The photodetector detects photons or the energy of the optical signal; thus the output
photocurrent of the photodetector is given by
< ( ')
=
* |« M
f
'/>(<)|£,(0|: + f>! (r)|£2(< f
(3.6)
=R
+JP, m
(()£,• (OB, ( t y w w u
where R is the responsivity of the photodetector. For the sake of simplicity, assume that
both lasers have the same optical output power of unity as,
|£,(<f = |£2(<f = £ ,M S « = E l W M = 1-
(3.7)
Equation (3.6) becomes
i(0 =
{p, (0 + P2 (0 + 2VF,(0A(0 cos[% (r) - % (r)]} -
(3.8)
Substitute Equation (3.8) with Equations (3.1) and (3.2),
2
+ m, ( 0 cos 2 jtfxt + m2(t) cos ln f 21
i{t) = R +2-^1 + m, (r)cos27rfjf + m, (t)cos2jtf2t + m, (t)m2(t) cos 2jtfxt cos 2nf21
•cos[% (0 - ^ ( 0 ]
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
67
Expand the square root into Taylor series, and keep only the zero and first order terms,
since only the subcarrier frequency range is of interest Rewrite the wavelength
difference in terms of frequency as,
V ,« - %(<) =
T T - « + * ,( » ) -
+ «>(<).
A —A
where Sf = c 2
1 is optical carrier difference in frequency, <5d>(r) = <!>, (t) - d>, (r) is
Aj A-,
initial phase difference of the carriers. This phase difference is the main reason that the
linewidth of the beat interference is the sum of the linewidths of the two optical carriers.
This matter will be discussed in more detail later in this section. The photocurrent is
given by
2
+ mx(t) cos 2 n f t + rr^lt) cos 2 xf 21
+2 cos[27r^r + <5£>(r)]
+ m,^ ~ jcos[2 ?r(<y + fi)t + <5d>(r)] + cos[27r(^f- f ) t +
i(t) = ^ + ^ W j cos[2jc( y + / , ) , + ,5*(,)J + cos[2jt(# - / 2 )r + <S*(i)]}
l*l ( - > ^ ') { c o i [ ^ + /, + / , > + tt>(0] + c o s [ 2 * ( # - /, - / , ) I + M>(0]}
m'(,)4miW { c o s ^ f f + / , - / , > + « > ( ; ) ] + cos{2*(# - / ,+ / ,) ! + « > ( ! ) ] }
(3.9)
Thus, the two subcarriers at /, and f 2 have the power (m2(t)} and (m 2 (t)),
respectively. The angle brackets denote time average. The beat interference noise is
centered at Sf with peak power 4; Sf + f and Sf - f with peak power (rn{ (r))/4;
Sf + f 2 and <§f- / 2 with peak power
(m^(r))/4;
<§f + / ! + / 2,
Sf~f\~fii
Sf + f ~ / 2, and Sf - f + f 2 with peak power (m,2(r)m2 (f))/l 6 . The last four beat
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
68
interference noise terms can be discarded, since the modulation depth is usually small,
and the peak power
smaller. There are 5 noise peaks in
consideration. Figure 3.1 shows a spectrum of beat interference from two lasers, each
modulated by a microwave subcarrier. The presence of the 5 noise peaks is clearly
shown, where each laser operates at a different wavelength and is modulated by a
microwave subcarrier with a modulation depth below 10%.
Assume that the lasers have Loientzian lineshape; therefore, the lineshape of the
individual beat interference peak is Lorentzian also, the linewidth of which is simply the
sum of the two lasers' linewidths. The Lorentzian lineshape is given by
A/
(3.10)
and
where / 0 is the center frequency and A/ is the linewidth. Thus, by integrating these 5
beat interference noises over the bandwidth 2 B (B is the baseband bandwidth), centered
at one of the subcarriers, say /,, the signal to interference ratio (SIR) for /, can be
calculated as
SIR = 10 log
(3.11)
where the signal power is given by
(3.12)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
69
and the beat interference noise power has the form of
4 2 i r [ ( / - # ) 3 + ( 4 f / 2 )! ]
, K (0)
4
M2 4 ( / - < ? - / . ) 2 + (A // 2 )2]
(«ifw )
A/
4
2 ^ [ ( / - ^ + / 1) 2 +(A //2)2]
(m22(r))
A/
4
2; t[ ( / - # - / 2)2 + (A //2 )2]
(m;(r))
A/
If one carries out the integration, the noise power is given by
( /,+ « - # )
-i -2(/l - B - S f j
t. .a. -n, 2--------------------tan
A/
A /
4/r
A /
A /
] 2(2/, - B —df)
(m,2 (0 ) ' -i 2(2/, + B ~ d f )
t a n ----------------------tan
4;r
A/
A/
(3.13)
2(/,
, 2 (Z - B - S f - f J
(< ^ w )
t a n -------------------------- tan ----------------------4n
M
A/
+
An
tan
-i 2(/i + B - S f + / , )
A /
. _,2(/,
-ta n
A /
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
70
Therefore, one can investigate how the beat interference affects the subcarrier by
changing the wavelength of the optical carriers, i.e., Sf. The calculation results with the
experimental data will be presented in the next section.
MKfi
5.3-559 S i:
ssf -3S dsn
SF«i FULL
arm
» as/
P IS 8U 3 W t
q as
s# «no
VF ,003
*
Figure 3.1 Spectrum of optical beat interference from two optical carriers.
Each carrier is modulated by a subcarrier with modulation depth at about 3%.
Dirac stated in his book that ‘each photon then interferes only with itself.
Interference between different photons never occurs’[9]. This has been said in a sense of
ensemble average over a long time scale, such as holography making where one laser
beam is split into two paths, and the difference between the two paths is shorter than the
coherent length, ensuring the same photon interference. In a short time scale, two
independent laser beams do interfere with each other, as observed by Magyar and
Mandel [1]. The optical beat interference linewidth has been calculated by Nazarthy et al.
[10]. This statistical optic method is basically to delay the photocurrent from the
photodetector with
t
seconds, then multiply with itself to calculate the autocorrelation
over time t. By taking the Fourier transform of the autocorrelation with respect of time r,
one finds the linewidth of the beat interference from the two laser beams. This method is
quite cumbersome and less intuitive, though mathematically very vigorous. Here is a
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
71
presentation of a simple and intuitive physical picture to calculate the Lorentzian shape
linewidth of the heterodyne OBI from two Lorentzian shape optical carriers as follows.
Lasers have stimulated emission and spontaneous emission with spontaneous
emission rate
(3.14)
where t sp is the spontaneous emission lifetime or carrier lifetime. This can be explained
as follows. Every t sp seconds, statistically, one photon will be emitted spontaneously
with a random initial phase. This photon will further stimulate other photons to be
emitted with the same initial phase. Thus, the electric field of the laser has the form of
<?(r) = £ 0 cos[2 # mr + <&(*)],
(3.15)
where £ 0 is the electric field amplitude, f m is the laser mth mode frequency, and 4>(r) is
the initial phase which changes in the time scale of t sp. One can picture the laser beam as
a train of wave components. Each component has a fixed phase relationship, and
between the components there is a phase disruption as shown in Figure 3.2. The length
of the component is called coherent length lc, and its corresponding coherent time has
the form
Figure 3.2 A laser beam as a train of wave components.
The component length is the coherent length.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
72
t.c
C
(3.16)
C
Intuitively, if one takes Fourier transform of Equation 3.15, one will get a
spectrum. The first term of the argument of the cosine function will give the Fourier
component f m, and the second term will give the linewidth, as stated by Yariv [11],
which seems like phase noise dominated line broadening. The Lorentzian shape
linewidth is given by Henry [12] as
where a is the linewidth enhancement factor, and / is the laser intensity. By definition,
the linewidth is related to the coherent time as
(3.18)
Now consider a situation such that two laser beams are combined at a
photodetector. Laser beam 1 and 2 have linewidth A/j and A/2, coherent time t cl and
t c2, and coherent length lcl and lc2, respectively. In Figure 3.3, the lines represent the
two laser beams, and the nodes on the lines denote the phase discontinuities. When these
two beams are combined together, one simply adds these nodes together. Between these
nodes, the two laser beams have fixed initial phase; thereby, the two laser beams
interfere with each other. The length between the nodes is defined as interference length,
and its corresponding time as interference time. If a photodetector has a response time
shorter that the interference time, it will be able to detect the optical beat interference, if
the beat interference falls within the bandwidth of the photodetector.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
73
■--------- --------- ■-----
------ ■------------------■--------- --------- ■----- -------------■
^2
beam 2
h
combined
Figure 3.3 Illustration of interference length resulting from combining two laser beams.
The lengths of the two laser beams are Kylcl or K2lc2, which are equal if the
lengths are infinitely long, where K{ and K2 are the number of nodes for the two laser
beams, respectively. It yields to
r -2 —K'l« y
A
c2
and the total number of nodes is
K = Kl +K1 = K l + - ^ L .
Li
Since all the discussion is based on statistics, the average interference length is
number of total nodes
lim
*i-*~ K
1
.
+ K\K\
lA
1
J_ + J _
L
U
or
1-JL _L
i‘'i ~ i*c,I
*ic2,
if one multiplies the speed of light c on both side of the equation, it becomes
J _ _ J L + _1_
*i Tel rc2
'
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
74
thereby, the optical beat inference linewidth is given by
Af. = — = Afl +Af2
which is the sum of the two laser beam linewidths. If these two beams are somehow
correlated, such as in the holography, the wave components are aligned, therefore the
interference linewidth is just the laser beam linewidth.
If a photodetector has a response time shorter than the interference time, one
would be able to observe the optical beat interference noise. In the experiment, which
will be presented in the next section, 6 and 40 GHz photodetectors are used,
corresponding to 170 and 25 ps response time. The optical beat interference linewidth is
measured to be 80 MHz, corresponding to 12.5 ns interference time. Thus, one should
be able to record the optical beat interference when it falls within the photodetectors'
bandwidths.
3.3
Experimental Verification
An experiment was conducted to compare the result with the theoretical
calculadon. In the experiment, two NEC single mode distributed feed-back (DFB) lasers
with built-in isolators operating at 1.55 pm were used. The lasers are designed for up to
1.2 Gbps long distance transmission system with RIN less than -135 dB/Hz. The builtin isolators ensured that no injection locking occurred between the two lasers in the
experiment Laser 1 was intensity modulated by a subcarrier /, at 896.5 MHz from a
voltage controlled oscillator (VCO) and Laser 2 with a subcarrier / , at 620.9 MHz. The
VCOs have the tuning range 600 MHz - 1 GHz. The choice of the particular subcarrier
frequencies was based on the fact that the second and higher harmonics were at their
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
75
lowest power levels. The subcarriers' power levels were chosen so that their modulation
depths ml and m2 were set to be 3.3% and 3.0%, as these were assumed to be small in
Section 3.2. The modulated optical carriers were mixed using a 2 X 2 coupler and one of
the outputs of the coupler was fed to a high speed photodetector while the other output
was connected to an optical spectrum analyzer, as shown in Figure 3.4. The RF output
of the photodetector was observed on the RF spectrum analyzer. The wavelengths of the
lasers were controlled by temperature controllers. Thus, by changing the temperature of
Laser I, its wavelength was tuned, resulting in a different beat interference over the dc to
12 GHz region where the optical beat interference was measured, since the beat
interference center frequency is the difference in the wavelengths of the two lasers.
temperature
controller
VCOl
high speed
photodetector
DFB laser 1
2
spectrum
analyzer
x 2 coupler
V C 02—^ DFB laser2
spectrum
analyzer
temperature
controller
Figure 3.4 Block diagram of experimental setup.
The RF spectrum analyzer resolution bandwidth was set to be 3 MHz. This
corresponded to a baseband B = 1.5 MHz; thereby, the SIR was read right off the RF
spectrum analyzer. The interference bandwidth at 3 dB was measured to be 80 MHz,
which suggested a linewidth of 40 MHz for each laser, which was usually the case for a
DFB laser. Figure 3.5 shows two cases of beat interference. In Figure 3.5(a) the two
lasers have the same wavelength and the beat interference is thus centered at dc, while
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
76
Figure 3.5(b) shows a case where the beat interference is at 1.23 GHz. The SIR for /,
of 896.5 MHz was taken from the RF spectrum analyzer at different interference
frequencies and plotted as cross points in Figure 3.6. The solid line comes from the
theoretical model using Equations 3.11, 3.12 and 3.13 in Section 3.2, where again, the
parameters were used as
VF .003
ft£F «3£
10
1.8204 « :
SPANFtU
RES 8* 3 Mtt
RSF -36 «Bb
10 aS/
ftTTEK0 <8
Stf> ftUTO
VF .003
*
f
Figure 3.5 Spectra of microwave subcarriers with the presence of optical
beat interference, which is centered at (a) dc, and (b) 1.2 GHz.
m, = 3.3%, m2 = 3.0%, /, = 896.5 MHz, / , = 620.5 MHz, and A/ = 80 MHz.
It is clearly shown from Figure 3.6 that the model matches the experimental data very
well. According to the theoretical model, when the optical beat interference is right on top
of the subcarrier, the signal is about 18 dB less than the noise, causing the subcarrier
outage. At this point, from the RF spectrum analyzer, the subcarrier was so totally
immersed in the beat interference noise that the signal power could not be measured;
thereby the SIR was set to be zero.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
77
40 —I
30-
20-
n
X
i io ca
3
®
cn
n0 —
-
10 -
-
200
2
4
6
8
Wavelength difference (GHz)
10
12
Figure 3.6 Signal to interference ratio versus optical frequency for subcarrier/i.
When designing a WDM network or an Optical ATM switch such as THESEUS,
one has to take into account the optical beat interference of different optical carriers to
achieve an acceptable biterror rate (BER) or SIR. Schwartz [13] showed that to achieve
10"9 BER, one needs to achieve a 20 dB signal to noise ratio in the presence of Gaussian
noise. In this case, to achieve 20 dB SIR, one needs to keep the optical carriers at least 4
GHz apart, which corresponds to 0.32 A, only to transmit a less than 1 GHz microwave
subcarrier and in turn to transmit 1.5 Mbps data, indicating a rather poor spectral
efficiency.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
78
3.4
Improvements of THESEUS over Optical Beat Interference
There are several approaches for THESEUS or WDM networks to overcome the
optical beat interference. One is based on the spread spectrum method. As it has been
established in Section 2.2 that when the interference time is shorter than the
photodetector response time, the photodetector will just perform time averaging and not
observe the beat interference signal, but will instead detect a white noise over its entire
bandwidth. Thus the microwave carriers might be able to be recovered. There are two
ways to decrease the interference time. One is to decrease the coherent time of each laser
by adding extra phase noise. A phase noise generator or simply a phase modulator could
be placed before the optical carrier combines with the others at the star coupler. The
phase noise generator could also be placed before or after the optical carrier is modulated
by a microwave subcarrier. Banat and Kavehrad [14] have theoretically and
mathematically simulated a case where an independent pseudorandom driven phase
modulator was placed before the star coupler with the modulation frequency higher than
the photodetector bandwidth; that is, the coherent time or, in turn, the interference time is
shorter than the photodetector response time. It was shown that a dramatic reduction in
the optical beat interference can be achieved with appropriate phase modulation signals.
The other way is to place a phase noise generator right between the optical filter and the
photodetector to directly decrease the interference length. This is probably the better way
to broaden the optical beat interference spectrum, because it broadens the optical
linewidth after the optical carriers are filtered by the optical channel filter, while the first
way might lead to adjacent optical channel crosstalk if the linewidth of each optical carrier
is broadened so much that it exceeds the channel filter bandwidth.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
79
Another approach is to randomize the polarization of the optical carriers. In
Section 2.2, it was assumed that the carriers have the same polarization in deriving the
model. If there are only two optical carriers at the photodetector with their polarization
perpendicular to each other, there would be no optical beat interference. For more general
cases, if there are more than two optical carriers, they would interfere with each other
along the two orthogonal axes. Thus, by randomizing the polarization, the optical beat
interference noise could be decreased. This method can be used in connection with the
previous spread spectrum method, since the phase modulator changes both phase and
polarization as mentioned by Banat and Kavehrad [14].
In the derivation in Section 3.2, it was assumed the modulation depth is small. If
one increases the modulation depth, a better signal to interference ratio might be
achieved, as claimed by Wood and Shankaranarayanan [6 ], where the modulation depth
is 1.8, even though it would increase the interference which has large sidelobes. This
matter needs further careful investigation.
output 11
output
interface
input 11
buffer &
header
processor
c o p y
network
buffer &
header
processor
input NM
Batcher
sorting
network
tra p
network
WDM/
SCM
network
output
interface
output NM
Figure 3.7 Block diagram of modified version of THESEUS.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
80
Finally, one can avoid the optical beat interference from a system point of view.
Here a modified version of THESEUS is proposed. The switch consists of a header
processor for every input port, a copy network, a Batcher sorting network, a trap
network and the WDM / SCM network, as shown in Figure 3.7. Assume that every
input port is synchronized or the incoming cells are buffered to be synchronized at the
buffer & header processor. The cell header is read, a table look-up is performed, the cell
header is updated, and an output address tag is added to the cell. The tag contains a
physical layer address as nm instead of VCI and VPI of the ATM layer, where n is used
to tune the laser wavelength and m is used to select microwave subcarriers. The cells are
then sent to copy network to copy those cells for multicasting. The new cell copies have
updated cell headers and destination tags also. The Batcher sorting network sorts all
these cells in a nondecreasing manner according to the tag address. The trap network
compares the destination tag with the one above it, and sends those with the same
address back to the Batcher sorting network to be sent in the next cycle, thereby solving
the destination contention. Then, the trap network compares the most significant digits of
the physical layer address, n, bundles those which have the same partial address n
together, sends them to the first M input ports of the WDM I SCM network, and groups
those with higher number rc+1 to the next M input ports, and so forth. The number of
cells in the group is guaranteed to be equal to or less then M, since all the contended cells
have been already trapped and sent back to the Batcher sorting network. The WDM /
SCM network is slightly different from the original THESEUS, such that there is only
one laser for every M input ports, since all the cells in these input ports have the same
partial address n. These M cells or less select their microwave subcarriers according to
another part of destination address m. The modulated subcarriers are then combined
together via M to 1 power combiner and modulated on a single optical carrier. Therefore
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
81
it is guaranteed that every photodetector detects only one optical carrier after the optical
filter. There is no optical beat interference at all.
3.5
Summary
A simple, intuitive theoretical model to explain and calculate the heterodyne
optical beat interference has been developed. A new concept of interference time and
interference length has been introduced. An experimental confirmation has been
conducted. The experimental data matches the theoretical model very well. It shows that
4 GHz or 0.32 A optical bandwidth is needed to transmit 1.5 Mbps data using 896.5
MHz subcarrier in this case, indicating very poor spectral efficiency. Several approaches
for THESEUS or WDM / SCM networks to overcome the optical beat interference have
been proposed based mainly on the proposed model.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
82
R eferences
[1]
G. Magyar and L. Mandel, “Interference fringes produced by superposition of two
independent maser light beams,” Nature, vol. 198, pp. 255-256, 1963.
[2] C. Desem, “Optical interference in lightwave subcarrier multiplexing systems
employing multiple optical carriers,” Electron. Lett., vol. 24, pp. 50-52, Jan.
1988.
[3] C. Desem, “Optical interference in subcarrier multiplexed systems with multiple
optical carriers,” IEEE J. Select. Areas Commun., vol. 8 , pp. 1290-1295, Sept.,
1990.
[4] C. Desem, “Measurement of optical interference due to multiple optical carriers in
subcarrier multiplexing,” IEEE Photon. Technol. Lett., vol. 3, pp. 387-389, April
1991.
[5] N. K. Shankaranarayanan, S. D. Elby, and K. Y. Lau, “WDMA/SubcarrierFDMA lightwave networks: Limitations due to optical beat interference,” J.
Lightwave Technol., vol. 9, pp. 931-943, July 1991.
[6 ] T. H. Wood and N. K. Shankaranarayanan, “Operation of a passive optical
network with subcarrier multiplexing in the presence of optical beat interference,”
J. Lightwave Technol., vol. 11, pp. 1632-1640, Oct., 1993.
[7] W. Xin, N. Antoniades, T. S. Stem, and E. S. Yang, “Optical beat interference of
two independent semiconductor lasers,” Bulletin o f The American Physical
Society, vol. 42, no. 2, pp. 943, April 1995.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
83
[8 ]
W. Xin, N. Antoniades, T. S. Stem, and E. S. Yang, “Heterodyne optical beat
interference limitations on wavelength division multiplexed networks,” 1995 IEEE
/ LEOS Summer Topical Meeting Digest on RF Optoelectronics, August 7-11,
1995, Keystone, CO, pp. 68-69.
[9]
P. A. M. Dirac, The Principles of Quantum Mechanics, 4th ed. (revised), Oxford:
Clarendon Press, 1958, pp. 9.
[10] M. Nazarathy, W. V. Sorin, D. M. Baney, and S. A. Newton, “Spectral analysis
of optical mixing measurements,” J. Lightwave Technol., vol. 7, pp. 1083-1096,
July 1989.
[11] A. Yariv, Quanmm Electronics, 3rd ed., New York: John Wiley & Sons, Inc.,
1989, pp. 596.
[12] C. H. Henry, “Phase noise in semiconductor lasers,” J. Lightwave Technol., vol.
LT-4, pp. 298-311, Mar. 1986.
[13] M. Schwartz, Information Transmission, Modulation, and Noise, 4th ed. New
York: McGraw-Hill, Inc., 1990, pp. 428.
[14] M. M. Banat and M. Kavehrad, “Reduction of optical beat interference in
SCM/WDMA networks using pseudorandom phase modulation,” J. Lightwave
Technol., vol. 12, pp. 1863-1868, Oct. 1994.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
84
Chapter 4
4.1
Device Requirements for THESEUS
Introduction
To implement high speed WDM / SCM networks or optical ATM switches such
as THESEUS, one puts some stringent requirements on the devices used in the systems.
In the case of THESEUS, High quality microwave sources that provide narrow
linewidth, single frequency, low phase noise, and low harmonic microwave subcarriers
and local oscillators for reception are needed. Also needed are high channel isolation and
high speed microwave switches to select subcarriers. A variety of high speed, low power
and high power, linear microwave amplifiers and attenuators; and a variety of high
quality power dividers and combiners are needed. To modulate and demodulate baseband
data, high quality modulators and demodulators for different modulation schemes are
needed. In the ASK scheme of THESEUS, high quality liner mixers are needed. Also
needed are all kinds of microwave filters. For the optical part, one needs high tuning
speed, high modulation bandwidth tunable lasers or laser arrays with high linearity [ 1],
In some cases, one might need high quality optical amplifiers such as erbium doped fiber
amplifiers (EDFAs) to meet power budget requirements [2]. To put more optical carriers
in the networks or switches, and in turn, to enlarge the size of the networks or switches,
very narrow bandwidth optical filters are needed. On the receiving end, one needs high
speed photodetector with high sensitivity and responsivity, which can be easily
integrated with the silicon based integrated circuits. This chapter primarily concentrates
on photodetectors, especially on the material requirements for high speed photodetectors
[3].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
85
4.2
Time*Resolved Reflectivity Measurement
The development of high speed long wavelength photodetectors has been driven
by the needs of lightwave telecommunication networks. The response of a photodetector
is determined by the optical pulse shape, carrier transit time, and carrier lifetime of the
semiconductor used for the photodetector. The optical pulses generated by the collidingpulse-mode-locked (CPM) dye lasers can be as short as 60 fs, which is negligible
compared to the duration of the photodetector output electrical pulses. The carrier transit
time, which is the time for photo-generated electrons traveling from cathode to anode,
can be decreased by reducing the electrode spacing. But the reduction of the electrode
spacing is limited by the fabrication technology and by the fact that the electrode spacing
cannot be reduced to too small a size, in order to maintain a certain active detection area
out of the illuminated area, in turn, to achieve a high sensitivity, since the interdigitated
electrodes cannot be made too narrow. Therefore, the major factor which affects the
detecting speed is the semiconductor free carrier lifetime, which is the time for the photo­
generated electrons from their generation to recombination. By introducing traps or
recombination centers via ion implantation, one can decrease the carrier lifetime [3-9].
Several methods have been used to determine the carrier lifetime of
semiconductors. The time-resolved reflectivity measurement has the advantage of being
all-optical and contactless [10]. According to Drude's model, at low excitation levels, the
change in the index of refraction of a semiconductor is proportional to the carrier density
given by [ 10 ]
A _
A n
—
2 m 0e2 Kr
___
£m * co
")
e
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(A
1\
1
>
86
where no is the index of refraction of the semiconductor, e is the background dielectric
constant, m* is the electron effective mass, a) is the laser frequency, and Ne is the carrier
density. In the time-resolved reflectivity measurement, a laser beam, called the pump
beam, illuminates the sample and generates the free carriers. These excess free carriers
cause a change in the index of refraction and subsequently a change in the reflectivity. A
delayed laser beam, called the probe beam, is focused onto the illuminated area of the
sample, and its reflected beam is detected. By measuring the intensity of the reflected
probe beam, one can measure the change in the index of refraction.
The dynamic response of the carrier density induced by the laser pump beam is
described by
where R is the reflectivity of the pump beam, a is the absorption coefficient, Ipump{t) is
the pump beam intensity, and t is the carrier lifetime. The first term on the right side of
Equation (4.2) is the carrier density induced by the laser pulse; the second term is the
decay due to recombination [10]. The solution of Equation (4.2) is
W
' 1)
(4.3)
The change in the reflectivity is proportional to the change in the index of refraction. This
can be shown by considering a simple case of normal incidence of a laser beam into a
semiconductor from air. The reflectivity is given by [12]
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
87
where n is the index of refraction of the semiconductor. By taking the derivative, one
gets the change in the reflectivity as
&R = 4 n - 1 , An.
(« + 1)
From Equation (4.1), the change in the index of refraction is proportional to the carrier
density. Thus the change in the reflection of the probe beam from the sample has the
form of
(4-4)
where Iprobe(t'-t) is the intensity of the probe beam delayed t seconds. Since the pump
and probe laser pulses are so short compared to the carrier lifetime r, IpumP(t) and
IProbe(t'-t) can be treated as 5-functions as
= W W and1 p r o b e d “0 =
-0,
where I^p o and IpmbeQ are the peak intensities of the pump and probe beams,
respectively, if one carries out the integration, Equation (4.3) becomes
N.
(4'5>
and Equation (4.4) becomes
~
NM-
(4.6)
Substituting Equation (4.5) into Equation (4.6), the change in the reflection of the probe
beam is given by
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
88
(4.7)
Therefore, by measuring the reflected probe laser intensity, one can measure the carrier
lifetime r.
4.3
C a rrie r Lifetime of Silicon G erm anium
Silicon germanium has drawn much attention recently, because it can be operated
in the long wavelength region of 1.3 to 1.5 Jim by changing its composition [13,14].
Figure 4.1 shows that with 40 to 90% of germanium composition, silicon germanium
can be operated at 1.55 (im wavelength range. Also, it can be easily integrated into the
well developed Si technology.
1.2
U N S T R A IN E D
BULK A L L O Y
i/ i
o
o
E
x
«Z9
lli
M
STRAIN-SPLIT V.B
(CALC)
-
0.6
o
Si
0.2
0.4
0.6
Ge FRACTION , x
0.8
2.0
10
Ge
Figure 4.1 Summary of energy gap values for Si/Si,.xGex strained-layer
superlattices on Si(OOl) and unstrained bulk alloy [14].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
89
The carrier lifetime of silicon germanium was measured. The samples used in the
experiment were grown by the molecular beam epitaxy (MBE). The schematic structure
of the sample is shown in Figure 4.2. A 10 period Si/Sii.xGex strained layer superlattice
(SLS) with a total thickness of 0.5 pm was deposited on top of a 3000 A silicon buffer
layer which was grown on an n-type silicon wafer. The fraction of germanium in the
strained layer superlattice varies from 0 to 50%. A 0.5 pm relaxed SiojGeoj was grown
and then capped with 30 A silicon.
________30 A Si________
5000 A unstrained Si.sGe.5
12 A Si
475 A SiJGe.5
SLS 10 periods
25 A Si
475 A Sii-xGex
3000 A Si buffer
n - Si substrate
Figure 4.2 Schematic structure of SiGe. x in Sii-xGex varies
from 0.0 to 0.5 in 10 period strained layer superlattice.
Figure 4.3 shows the rocking curve of (004) reflection of the sample. The higher
peak on the right is from the silicon substrate. The SiGe peak on the left shows that a
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
90
relaxed crystalline film has been grown, but the broad peak indicates that there might still
be some residual strain in the film. Another possible explanation for this broad peak is
that the graded strained layer superlattice peaks are overlapping with the relaxed
Sio.sGeo.5 peak. Figure 4.4 shows the depth concentration profiles of copper as well as
germanium obtained from Secondary Ion Mass Spectroscopy (SIMS). It shows that the
copper of 1016 to 1017 c m 3 has been doped throughout the entire SiGe growth.
1000-
800600-
t&
400-
*
200-
t
33
34
35
36
37
Angle (degree)
Figure 4.3 Rocking curve of (004) reflection of SiGe. The higher peak on the right
is from the Si substrate and the lower one is from the epi layers.
The set up of the time-resolved reflectivity measurement is shown in Figure 4.5.
The CPM dye laser produces 60 fs pulses at a wavelength of 620 nm and a repetition rate
of 100 MHz. These pulses are split into two beams. The first one is for pumping, and
the second one is delayed and further split into two beams as a probe beam and a
reference beam. The pump beam and probe beam are perpendicularly polarized in order
to avoid interference between the two beams at the sample. The average power of the
pump beam is 4 mW, corresponding to 40 pJ per pulse, and the probe beam has an
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
91
average power of 2 mW. The pump beam is focused onto the sample in a spot with a
diameter of 15 (im, while the probe beam is focused onto the same spot with a diameter
of 5 nm. The reflection of the probe beam is collected by a lens and detected by the
differential detector. The differential detector also detects the reference beam, and the
difference between the two beam is taken as the output The output signals from the
differential detector are fed into the current preamplifier and then into the lock-in
amplifier. The pump beam is chopped and synchronized with the lock-in amplifier by the
triggering signal. The computer controls the delay of the probe beam and stores the
differential detector output signal which is displayed on the screen.
Six samples numbered from 0 to 5 were used in the experiment Samples 1 to 5
were implanted with oxygen ions through the entire SiGe region at different dosages,
while sample 0 was undamaged. Table I shows the ion implantation energy and dosage
for these samples, as well as the carrier lifetime measured from the experiment.
The experimental results are shown in Figure 4.6. The vertical axes correspond
to the changes in the reflectivity, and the horizontal axes represent the time delays of the
probe beam. The solid lines are the experimental data, and the dashed lines are the
exponential fitting curves. The unimplanted Sample 0 data shows that the carrier lifetime
x = 8 ps. The Hall mobility for this sample was also measured. The mobility is 1,170
cm2/V-s at 300 K and 10,340 cm2/V-s at 77 K. For Sample 1, the multiple oxygen ion
implantation dosage and energy are 1 x 1012 cnr2/100 keV, 8 x 10" cnr2/200 keV, 6 x
1011 cnr2/300 keV, 4 x 1011 cnr2/400 keV. The fitting curve has two time scales with
their weight percentage as, x, = 4.5 ps, 61% and x2 = 8 ps, 39%. The average carrier
lifetime x = 5.9 ps. Sample 2 has X\ = 3.5 ps, 69%; x2 = 8 ps, 31%; and average carrier
lifetime x = 4.9 ps with implantation dosage and energy as 5 x 1012 cm*2/100 keV, 4 x
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
92
1012 cm-2/200 keV, 3 x 1012 cnr2/300 keV, 2 x 1012 cm-2/400 keV. Sample 3 has x, =
2.5 ps, 78%; x2 = 8 ps, 22%; and average x = 3.7 ps. Samples 4 and 5 have one carrier
lifetime as x equal to 1.5 and 1 ps, respectively.
m
,
oo
I8
8»-»
§
m
o«
u
iH
K
2
S:
<
0.0
o.2
0.4
o .i
o.a
i.o
1.2
1.4
APPROXIMATE DEPTH (microns1
Figure 4.4 SIMS depth concentration profiles of Cu and Ge.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
93
CPM Laser
V
stepping motor
controlled stage
Probe Beam
A
7
Ref Beam
Chopper
Pump Beam
Differential
Detector
Computer &
Displayer
Preamplifier
Triggering
Lock-in Amp
Figure 4.5 Set up for time-resolved reflectivity measurement.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Sample #
Oxygen Ion Implantation Dosage and Energy
Carrier Lifetime
0
unimplanted
t = 8 ps.
1
lxlO 12 cm-2/100keV, 8x10“ cnr 2/200keV,
Xi=4.5 ps, 61%,
6xl0 11 cnr2/300keV, 4xlOn cm-2/400keV.
T2= 8 ps, 39%.
5xl0 12 cm*2/100keV, 4xl0 12 cm-2/200keV,
Xi=3.5 ps, 69%,
3xl0 12 cm-2/300keV, 2xl0 12 cm-2/400keV.
t 2= 8 ps, 31%.
IxIO 13 cm-2/100keV, 8xl0 12 cm-2/200keV,
ti=2.5 ps, 78%;
6xl0 12 cnr2/300keV, 4xl0 12 cnr2/400keV.
x2= 8 ps, 22 %.
5xI0 13 cm-2/100keV, 4xl0 13 cm 2/200keV,
x=1.5 ps.
2
3
4
3xl0 13 cm-2/300keV, 2xl0 ‘3 cm-2/400keV.
5
lxlO 14 cm*2/100keV, 8xl0 13 cm-2/200keV,
x= 1.0 ps.
6xl0 13 cnr2/300keV, 4xl0 13 cm-2/400keV.
Table 4.1 Oxygen ion implantation versus carrier lifetime.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
95
t=2jp*,78*
nm
olafloo
snnubon
10
T
un
eD
elay(p
a)
T
uneD
elay(p»)
■ m
t=4.5pc.6l%
*■0.3-
sm
ulaiion
stm
uatoo
<-0.4-
4
T
im
eD
elay(ps)
m
6
8
T
un
eD
elayfps)
t=3Jps.69*
^=8ps,31%
sum
lanoo
am
nlation
n p « i|
T
uneD
elay(ps)
Figure 4.6
T
un
eD
elay(ps)
Time-resolved measurement for Samples 0 - 5 .
From Table 4.1 shows that the carrier lifetime decreases as the implantation
dosage increases, from 8 ps of unimplanted sample to 1 ps of high dosage sample. At the
intermediate dosage levels, the carrier lifetime has two time constants as in Samples 1, 2
and 3. But the shorter lifetime Ti decreases and its weight percentage increases with the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
96
implantation dosage. This phenomena can be explained as follows: When the trapping
centers introduced by the ion implantation are saturated by the photo-generated carriers,
the excess carriers do not experience the trapping effect of the oxygen implant, leading
to a long carrier lifetime z2 (~8 ps) equivalent to the carrier lifetime of the undamaged
sample. The contribution of x2 to the average carrier lifetime is inversely proportional to
the defect density or the ion implantation dosage. Figure 4.7 is the plot the carrier lifetime
or average carrier lifetime T verses the total oxygen ion implantation dosage which is the
sum of the four doses at four energy levels, respectively. This shows that at low dosage
levels the carrier lifetime drops sharply with the implantation dosage, while at high
dosage levels the change in the carrier lifetime tends to saturate. It is also observed that at
high implantation dosages the signal-to-noise ratio is higher than that at low dosages.
This is because the absorption coefficient a increases monotonically with the ion
implantation dosage [10]. According to Equation (4.5), the photo-generated carrier
density is higher at a high dosage. Therefore, the change in the reflected probe beam is
more dramatic than it would be at low dosages.
P. M. Fauchet et al. [15] have used a similar technique which measures the
transmission instead of reflection and reported 6.25 ps carrier lifetime of amorphous
hydrogenated silicon germanium (a-Sio^GeojiH). The above experimental results show
that the carrier lifetime of relaxed Sio.sGeo.5 is 8 ps, which indicates that the copper
dopants act as recombination centers in the epi layer. The rocking curve of Figure 4.3
shows that the epi layer is a single crystal. Based on this, the relaxed Sio.sGeoj has a
higher mobility than does amorphous silicon germanium, which makes the relaxed
Sio^Geo^ more attractive for high speed applications.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
97
8
H
7H
•I
65 “
J 4.<5
U
2-
0
50
100
150
200
250
Oxygen Ion Implantation Dosage (x 1012 cm'2)
F igure 4.7 Carrier lifetime vs. total oxygen ion implantation dosage.
The carrier lifetime of 800 A copper doped strained layer Sio.75Geo.25 was also
measured to be 600 fs. This time scale corresponds to a bandwidth of 1.7 THz.
4.4
Sum m ary
In conclusion, the photo-generated carrier lifetime of 8 picoseconds for relaxed
copper doped Sio.sGeo.5 grown by MBE was measured by using time-resolved
reflectivity measurement. This time scale, comparable to that of amorphous hydrogenated
silicon germanium (a-Sio.sGeos:H), indicates that copper dopants are effective trapping
centers in SiGe. Together with its high mobility, this material Sio.sGeo.5 :Cu is a good
candidate for high speed optoelectronic devices. With the oxygen ion implantation, a
carrier lifetime as short as 1 ps was realized. Strained layer SiGe was measured to have a
600 fs carrier lifetime. The time scale of 1 ps to 600 fs corresponds to 1 to 1.7 THz
bandwidth of photodetectors. This huge bandwidth of silicon germanium, together with
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
98
the fact that it is easy to be integrated with well developed silicon technology, offers
tremendous opportunities to develop new optoelectronic devices and systems.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
99
R eferences
[1]
C. E. Zah et al., “Monolithic integration of multiwavelength compressive-strained
multiquantum-well distributed-feedback laser array with star coupler and optical
amplifiers,” Electron. Lett., vol. 28, pp. 2361-2362, Dec. 1992.
[2] E. Desurvire, Erbium-Doped Fiber Amplifiers: Principles and Applications, New
York: John Wiley & Sons, Inc., 1994.
[3] W. Xin, H. K. Liou, E. S. Yang, L. Xu, and S. H. Xin, “Picosecond carrier
lifetime in relaxed silicon germanium and low temperature gallium arsenide,”
Bulletin of The American Physical Society, vol. 40, no. 1, pp. 802-803, March
1995.
[4] P. R. Smith, D. H. Auston, A. M. Johnson, and W. M. Augustyniak,
“Picosecond photoconductivity in radiation-damaged silicon-on sapphire films,”
Appl. Phys. Lett., vol. 38, pp. 47-50, 1981.
[5] A. G. Foyt, F. J. Leonberger, and R. C. Williamson, “Picosecond InP
optoelectronic switches,” Appl. Phys. Lett., vol. 40, pp. 447-449, 1982.
[6 ] D. H. Auston and P. R. Smith, “Picosecond optical electronic sampling:
Characterization of high-speed photodetectors,” Appl. Phys. Lett., vol. 41, pp.
599-601, 1982.
[7] P. M. Downey and B. Schwartz, “Picosecond photoresponse in 3He+ bombarded
InP photoconductors,” Appl. Phys. Lett., vol. 44, pp. 207-209, 1984.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[8 ] R. J. Manningand J. R. Hill, “Photoconductive response times of Si-on-sapphire
damaged with Si28 ions," Appl. Phys., B 38, pp. 17-21, 1985.
[9]
Th. Pfeiffer,
J. Kuhl, E. O. Gobel, and L. Palmetshofer, ‘Temperature
dependence of the picosecond carrier relaxation in silicon-irradiated silicon-onsapphire films," J. Appl. Phys., vol. 62, pp. 1850-1855, 1987.
[10] F. E. Doany, D. Grischkowsky, and C.-C. Chi, “Carrier lifetime versus ionimplantation dose in silicon on sapphire,” Appl. Phys. Lett., vol. 50, pp. 460-462,
1987.
[11] G. P. Agrawal, and N. K. Dutta, Long-wavelength Semiconductor Lasers. New
York: Van Nostrand Reinhold Company, 1986, pp. 225.
[12] M. V. Klein and T. E. Furtak, Optics, 2nd ed. New York: John Wiley & Sons,
Inc., 1986.
[13] R. Braunstein, A. T. Mooie, and F. Herman, “Intrinsic optical absorption in
Germanium-Silicon alloys,” Phys. Rev., vol. 109, pp. 695-710, 1958.
[14] R.
People,
“Physics
and
applications
of
GexSii_x/Si
strained
layer
heterostructures” IEEE J. Quantum Electron., vol. QE-22 no. 9, pp. 1696-1710,
1986.
[15] P. M. Fauchet, K. A. Young, W. L. Nighan, Jr., and C M. Fortmann,
“Picosecond carrier dynamics in a-Sio.5Geo.5 :H measured with a free-electron
laser,” IEEE J. Quantum Electron., vol. 27, no. 12, pp. 2714-2717, 1991.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
101
Chapter 5 Conclusion
S.l
Summary
A new multirate ATM switching architecture: Terabit Hybrid Opto-Electronics
Self-routing Ultrafast Switch (THESEUS) has been proposed. The switch is based on
wavelength division multiplexing (WDM) / microwave subcarrier multiplexing (SCM)
techniques with ASK modulation scheme for the microwave subcarriers and direct
intensity modulation for the optical carriers. It is intended for high speed ATM networks
with simplified protocols and less processing overhead. Its size can be increased to as
large as 1000 x 1000 without internal blocking.
A testbed to implement THESEUS has been successfully constructed. Multirate
of 1.5, 6.3, 44.7, and 139 Mbps binary data streams have been switched through the
testbed. Good receptions with bit error rate of less than 10”9 at the output ports of the
switch have been realized with small modulation indices for the ASK modulation of the
binary data streams.
For a small ATM switch or using this architecture for a small computer
communication network such as a local area network, the microwave subcarriers can be
used as output buffers. This scenario has been analyzed by using a slotted ALOHA
access scheme and immediate-first-transmission protocol. It has been shown that the
microwave subcarrier throughput or the channel utilization can be reached to 50% at
heavy traffic load for the group assignment of the subcarriers. It has been shown that the
group assignment of the subcarrier has a better throughput and cell delay than the random
assignment.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
102
The heterodyne optical beat interference limitations on WDM networks have been
both theoretically and experimentally investigated. A simple, intuitive theoretical model
of the optical beat interference has been developed. A new concept of interference time
and interference length has been introduced. The experimental data shows a good match
with the model. According to the model, a number of approaches to overcome the optical
beat interference have been proposed
Some of the device requirements for THESEUS and WDM / SCM networks,
especially the high speed photodetector and its semiconductor material requirements,
have been discussed. The relaxed silicon germanium as a new photodetector material has
been investigated. With the oxygen ion implantation, the photo-generated electron carrier
lifetime of the silicon germanium has been reduced to as short as 1 ps, corresponding to
1 THz bandwidth of photodetectors. It was also found that strained layer silicon
germanium has the carrier lifetime of 600 fs. This huge bandwidth, together with the fact
that it is easy to be integrated with well developed silicon technology, offers tremendous
opportunities to develop new optoelectronic devices and systems.
5.2
Suggestions for Future Work
The THESEUS testbed still requires further work to implement the tunable
transmitters and fixed receivers. It has to be decided whether to use distributed Bragg
reflector (DBR) lasers or DFB laser arrays, since the former need D/A converters as part
of their tuning circuits and the latter need switches to turn one or some of the lasers on at
one time. It needs to be investigated how the tuning range, speed of DBR lasers and
switching range, speed of DFB laser arrays would affect system performance. For
example, if DFB laser arrays were used in THESEUS, one could share one laser array
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
103
for M microwave subcarriers. For those subcarriers intended for the same optical carrier,
one could combine them together to modulate one wavelength, thereby limiting the
optical beat interference. For those intended for different optical carriers, one could just
turn the corresponding lasers on and modulate them individually. This depends on how
the laser arrays are designed and fabricated.
The nonlinearity in the system has to be improved. It has been shown that the
nonlinearity starts to affect the performance at the high data rate of 139 Mbps or OC-3
links. To use THESEUS for a higher data rate such as OC-12 links of 622 Mbps, one
should look into other modulation schemes such as frequency-shift keying (FSK) and
phase-shift keying (PSK).
To further the system performance evaluation, one could use a network simulator
such as OPNET to simulate the system even in the presence of optical beat interference.
This would give us better clues to improve the system design.
As mentioned in Chapter 4, several proposed approaches to overcome the optical
beat interference need more vigorous investigation.
The optical approaches are presented as follows:
Using an optical phase noise generator to broaden the optical beat interference
noise spectrum, in other words, to spread the interference noise energy all over the
frequency spectrum, one might be able to achieve a reasonable signal to interference ratio
to recover transmitted data. This approach is based on the fact that the interference time
should be reduced to shorter than the photodetector response time. The location of the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
104
optical phase noise generator within the system would have profound effects on system
performance.
The other approach is to randomize the polarization of the optical carriers. The
necessary condition for optical beat interference to occur is that the two optical carriers
should have the same polarization. Thus by randomizing the optical carrier polarization,
one can reduce the interference noise. This approach can be combined with the previous
approach, since the change in polarization would also cause a change in the optical
phase.
The third approach is to change the microwave subcarrier modulation depth. It
has been shown that the increase the modulation depth would yield a better signal to
interference ratio.
The system approach is to use a copy network, a Batcher sorting network and a
trap network before the WDM / SCM network to ensure that every receiver detects only
one optical carrier. More analysis study such as using OPNET to simulate this new
architecture is needed before reaching a decision to implement it
Silicon germanium is shown to be a promising candidate for high speed
photodetectors. Further investigation of this material and possible fabrication of a
prototype photodetectors are suggested.
This future work awaits us and will require a great deal of effort from all of us.
And the future looks promising and rewarding.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Документ
Категория
Без категории
Просмотров
0
Размер файла
4 117 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа