Components of Optical
Networks
Based on:
Rajiv Ramaswami, Kumar N. Sivarajan, “Optical Networks – A
Practical Perspective 2nd Edition,” 2001 October, Morgan
Kaufman Publishers
Optical Components
! Couplers
! Isolators/Circulators
! Multiplexers/Filters
! Optical Amplifiers
! Transmitters (lasers,LEDs)
! Detectors (receivers)
! Switches
! Wavelength Converters
Couplers
! Directional coupler is used to combine
and split optical signals.
Input-A Output-C
(Fibre-1)
Input-B Coupling length Output-D
(Fibre-2)
! C=αA+(1-α)B D=αB+(1-α)A
! α is called the coupling ratio
Couplers
! Can be:
! Wavelength independent (α is independent
of the wavelengths)
! Wavelength selective (α is wavelength
dependent)
Wavelength Independent
Couplers
! 3dB coupler (α=0.5) can be used to
build n n couplers (e.g., n n star
x x
coupler)
! Can also be used used to tap off a small
portion of light (α=0.9−0.95), e.g., for
monitoring purposes.
4x4 Star Coupler
11 11
C C
22 22
33 33
C C
44 44
Power is split equally among the outputs…
Wavelength Dependent
Coupler
! Used to combine signals at 1310nm and
1550nm (two different bands) together
without loss. It may only have one
output.
! Can be used to separate optical signals
of different bands.
! Also used for mixing the pumping signal
for EDFA.
Isolators
! Allows “light-flow” in one direction but
blocks it in the other (nonreciprocal
device).
! Used after EDFA and lasers to improve
performance.
! Insertion loss: loss in forward direction
(~1dB)
! Isolation: loss in reverse direction
(~40-50dB)
Circulators
! Isolator with multiple ports (e.g., 3, 4).
! Helpful in constructing optical add/drop
elements.
Filters
! Essential to selecting (dropping)
wavelengths from the fibre
(demultiplexing).
λ1, λ2, λ3, λ4
λ1
λ2, λ3, λ4 Wavelength Filter
! Filters can be tuneable
Multiplexers
! Essential to combine wavelengths onto
one fibre.
λ1
λ2 λ1, λ2, λ3, λ4
λ3 Wavelength MUX
λ4
Static Wavelength
Crossconnects
! The crossconnect pattern is established at
manufacturing and cannot be changed
dynamically. Example:
λ1
λ2
λ1, λ2, λ3, λ4 Λ1, λ2, λ3, Λ4
λ3
λ4
Λ1, Λ2, Λ3, Λ4 λ1, Λ2, Λ3, λ4
Optical Amplifiers
! Since optical signals are attenuated by fibre and
insertion losses of other components, signals
may become too week to be detected.
! It is possible to do 1R regeneration optically
(with all its benefits and drawbacks).
! Furthermore optical amplifiers have large gain
bandwidths (one regenerator is enough for the
entire band).
! But, they introduce additional noise (noise that
accumulates). (Gain should be also flat over the
entire band and insensitive to the input signal).
Optical Amplifiers
! Types of optical amplifiers:
! Erbium-doped fibre amplifiers (silica fibre
doped with Erbium ions. Laser pumps are
needed)
! Raman amplifiers (using the Raman scattering
non-linear effect. Needs laser pump source
but it can be used al all bands).
! Semiconductor optical amplifiers (earlier
technology, using p-n junctions of
semiconductors to amplify light. No pump
source is needed).
Transmitters - LEDs
! Light Emitting Diodes are inexpensive
but have a wide spectrum and work
only in the ~800nm range.
! LEDs cannot be directly modulated at
high data rates.
Transmitters - Lasers
! Lasers are used as transmitters and pump
sources for EDFA and Raman amplification
(higher power is required).
! Semiconductor lasers are the most commonly
used, essentially they are semiconductor
optical amplifiers with positive feedback (very
effective and efficient).
! Erbium lasers are EDFAs positively fed back
but need pumping source (semiconductor
laser)
Transmitters - Lasers
! Need to produce high output (0-10dBm)
! Have to have narrow spectral
bandwidth (for WDM)
! Have to be stabile in the transmitting
wavelength (lifetime drifting needs to
be small)
! Need to be easily modulated.
Transmitters – Tunable Lasers
! Highly desirable components:
! For a 100-channel WDM system 100 types of
conventional lasers have to be stocked (extensive
inventory)
! Key elements of reconfigurable optical networks
(less lasers than wavelengths; switching
times:~ms)
! Also essential for for efficient optical packet
switched networks (switching times: ~ns)
! Still in research labs but soon to be available.
Detectors - Receivers
! For O-E conversion.
Optical
signal Optical Photo- Front-End Decision Data
Amplifier detector Amplifier Circuit
O E
! Tunable receivers are available.
Wavelength Converters
! Converts data on optical signal for incoming
wavelength to an other outgoing wavelength.
! Can be used:
! between legacy and WDM systems
(transponder).
! Within the network to improve on utilization.
! Between boundaries of networks belonging to
different carriers, who do not coordinate
assignment of wavelengths.
Wavelength Converters
! Can be classified:
! Fixed-input, fixed-output
! Variable-input, fixed-output
! Fixed-input, variable-output
! Variable-input, variable-output
! Can be:
! Optoelectronic (2R/3R OEO conversion, usually VIFO)
! Optical grating (VIFO or VIVO)
! Interferometric (Optical 2R)
! Wave mixing (truly transparent, λ−s have to be close)
Switches
Switches
! Applications:
! Provisioning
! Protection switching
! Packet switching
! External modulation
Provisioning Switching
! Provisioning of lightpaths.
! Switches are used inside wavelength
crossconnects to reconfigure them to support
new lightpaths.
! These switches replace manual patch-panels,
thus requiring additional control software but
enable easy and fast reconfiguration.
! Switching times of a couple of ms are
acceptable.
! Challenge is to realize large switches.
Protection Switching
! Switch the entire traffic of a primary fibre to
another fibre in case the primary fibre fails.
! Switching time is in order of couple of ms.
(The entire protection operation should be
done in a couple 10ms).
! Switch sizes may vary from 2 ports up to
several thousand ports (when used in
wavelength crossconnects)
Packet Switching
! High-speed optical packet switching
switches, switching on a packet-by-
packet basis.
! Switching times should be as small as
couple of ns. (at 10Gbps 53bytes
correspond to 42ns).
! This is the switching technology of the
future…
External Modulator Switches
! To turn off and on the laser beam after
the laser transmitter (to reduce laser’s
spectral bandwidth, thus to reduce
chromatic dispersion).
! Switching time is around 10ps (rise and
fall time) for a 10Gbps signal (1bit time
= 100ps, switching time has to be at
least an order of magnitude less).
Switches - comparison
Application Switching time Number of
ports
Provisioning 1-10ms >1000
Protection 1-10ms 2-1000
Packet 1ns >100
switching
External 10ps 1
modulation
Parameters of Switches
! Extinction ratio: output power on state/output
power off state (40-50dB for mechanical
switches, 10-25dB for high-speed
modulators).
! Insertion loss: loss should be uniform over all
paths (determined by the architecture mainly
not the technology).
! Crosstalk: calculated by the output power of
all input ports not switched to that output
port.
Parameters of Switches
! Polarization dependence should be negligible.
! Latching: switching remains intact even if
power supply is turned off.
! Monitoring: switching state should be
monitorable.
! Reliability: long-term history is desired. Short
term reliability is tested by switching through
states a couple million times. In provisioning
although it is important that the switch
remains capable of switching even after
spending years at a given state.
Large Optical Switches
! Number of ports: n*100-n*1000
(couple of fibres carrying several tens
to hundreds of wavelengths).
! Properties of large optical switches:
! Number of switching elements required.
! Loss uniformity.
! Number of crossovers.
! Blocking characteristics.
Number of Switching Elements
! Large switches are built up by multiple
small switch elements (e.g. 2*2 or 1*n
elements).
! Cost and complexity depend on the
number of switching elements.
Loss Uniformity
! The problem of loss uniformity (as
mentioned before) is exacerbated for
large switches.
! Can be measured, e.g., by counting the
minimum and maximum number of
switch elements in the optical path for
different inputs/outputs.
Number of Crossovers
! Some switches are manufactured on a single
substrate on a single(!) layer. If paths of
waveguides cross it will introduce two
undesirable effects:
! Power Loss and
! Crosstalk.
! Thus crossovers have to be minimized
(eliminated).
! It is not an issue with free space propagation,
i.e., with MEMS.
Switch Blocking Characteristics
! Switches can be of two types:
! Non-blocking: an unused input can be
connected to any unused output, thus
every (possible) interconnection pattern
can be realized.
! Blocking: some interconnection patterns
cannot be realized (e.g., there is no way of
connecting input fibre one to output fibre 6
on wavelength 3).
Non-blocking Switches
! Can be further characterized into:
! Wide-sense non-blocking: every input can be
connected to every output without rerouting any
ongoing connections but employing some kind of
sophistication for establishing connections to
begin with.
! Strict-sense non-blocking: any input can be
connected to any output without rerouting and
without added sophistication.
! Rearrangeably non-blocking: ongoing connections
may be interrupted and rerouted which may not
be acceptable but uses fewer switching elements
and more sophisticated control is needed).
Basic Switch Architectures
Non-block. Number Max. Loss Min. Loss
type Switches
Crossbar 2x2 Wide n2 2n-1 1
Clos Strict
2x2
4 2n1.5 5 2n − 5 3
Spanke 1xn Strict 2n 2 2
Beneš 2x2 Rearr. n(2log2n-1)/2 2log2n-1 2log2n-1
Spanke-Beneš Rearr. n(n-1)/2 n n/2
2x2
Crossbar Switch
! No crossovers but not loss uniform
! Used in Clos switches
Clos Switch
! Used in practice for large switches.
! 3 parameters: m,k, and p.
! 1st and 3rd stage have k (m*k) switches, 2nd stage has
p (k*k) switches.
! If p>=2m-1 then switch is strictly non-b, thus usually
p=2m-1.
! Individual switches are usually designed by crossbar
switches.
! Loss uniformity is better than with crossbar.
! Number of switching elements is less than that of a
crossbar.
Clos Switch
! A three stage 1024 port switch:
Spanke Switch
! Becoming more and more popular.
! n*n switch is established by n (1*n) switches
and n (n*1) switches.
! These elements can be built (e.g., using
MEMS technology).
! Only 2n switches are needed (linear!) and all
paths cross through only 2 elements.
! Loss uniformity can be achieved and
insertion loss is small.
Spanke Switch
Beneš
! Rearrangeably non-blocking and very efficient
in the number of 2*2 switches (waveguide
crossovers are needed)
Spanke-Beneš
! Rearrangeably non-blocking, efficient in the number
of 2*2 switches (no waveguide crossovers – n-stage
planar architecture). Not loss uniform.
Optical Switching Technologies
! Bulk Mechanical Switches
! Micro-Electro-Mechanical System (MEMS)
! Bubble-Based Waveguide Switch
! Liquid Crystal Switches
! Electro-Optic Switches
! Thermo-Optic Switches
! Semiconductor Optical Amplifier Switches
! Electro-Holographic Switches
Mechanical Switches
! Examples include:
! Moving mirrors in and out of the optical path
! Bending or stretching fibre in a coupler changing
the α value of coupling.
! Low insertion loss, low crosstalk and relatively
inexpensive and well suited for crossbars.
Switching times of few ms and little port
counts (small crossconnects, protection,
provisioning). Long term reliability? Can be
cascaded but there are better ways…
MEMS Switches
! Small mechanical devices on silicon
substrates. In optical networking MEMS refers
to small mirrors of a few hundred
micrometers. Several of these mirrors can be
put on one substrate with common
semiconductor manufacturing techniques.
! Mirrors can be digital (only two positions – for
crossbar architectures) or analogue (several
positions – for 1*n switches) controlled.
Pop-up (Digital) MEMS
Pop-up (Digital) MEMS
! Practical size: 32x32
A.S. Morris, “In Search of Transparent Networks,”
IEEE Spectrum, October 2001
Analogue Beam Steering Mirror
! Also called a Gimbel mirror or 3D mirror.
! Control is difficult sophisticated servo
control is needed (high voltages with mV
scales).
! Suited to realize Spanke architectures
with hundreds to thousands of ports.
! Challenges: control, reliability, stability to
temperature, humidity and vibration
Analogue Beam Steering Mirror
Analogue Beam Steering Mirror
Analogue Beam Steering Mirror
A.S. Morris, “In Search of Transparent Networks,”
IEEE Spectrum, October 2001
Bubble-Based Waveguide Switch
! Technology is similar to what is used in
inkjet printers!
! Trenches are filled with index matching
fluid (that can be vaporized or moved).
! Small crossbar switches (32x32) can be
built efficiently with switching times of
>10ms.
Bubble-Based Waveguide Switch
Non-switched signal
Switched signal
A.S. Morris, “In Search of Transparent Networks,”
IEEE Spectrum, October 2001
Liquid Crystal Switches
! Make use of polarization effects. LC
cells can be used to rotate polarization
(or not). Different polarizations of the
same signal then can be used to cancel
each other out.
! Can be produced. in volume with low
cost (32x32 size)
Liquid Crystal Switches
Electro-Optic Switches
! Based on directional couplers with
varying coupling ratio (α) using a
voltage.
! Made with lithium niobate (usually)
! Capable of rapid changes (<1ns).
! High loss, high complexity and high
polarization dependent loss.
Electroholographic Switches
! A switch matrix is created by ferroelectric
crystals (e.g., lithium niobate)(basically
electro-optic switches).
! Rows correspond to fibers, while columns
correspond to wavelengths(!)
! They claim it is going to be fast enough for
photonic packet switching.
! Do not scale well, PDL, sophisticated focusing
is needed, etc.
Electroholographic Switches
A.S. Morris, “In Search of Transparent Networks,”
IEEE Spectrum, October 2001
Large Electronic Switches
! Today most practical crossconnects still use
OEO switching elements.
! Clos architecture is preferred (strict non-
blocking).
! Cost is mainly determined by the number of
OEO conversions not the switching fabric.
! High data rate streams may be spliced into
lower rate parallel streams. But today 64*64
crossbar ICs operating at 2.5Gbps are
available (dissipates 25W !!).
Large Electronic Switches
! A 1024x1024 switch needs about 100 such
ICs => power dissipation is around 25kW
(cooling is needed)(with 3D MEMS it would
be ~3kW and would be significantly more
compact).
! Connections between boards and racks at
these high speeds becomes a problem =>
not scalable. (Can be done optically with less
dissipation and interference while at a longer
range.)
Categorization of Optical
Switching Techniques
Reconfigurable Switches
! Switches may be called routers, cross-
connects or ADMs (add-drop
multiplexers).
! Optical switches keep the data stream
in the optical domain, although they still
may be controlled by electronics.
! Currently these equipment are common
in a backbone (core) network.
Switching Techniques in
Networks
! Circuit Switching
! Packet Switching
! Burst Switching
Circuit Switching
! 3 steps:
1. Circuit set-up
2. Data transfer
3. Circuit tear down
! Good for constant bit rate.
! Circuit set up time has to be significantly less than
data transfer time.
! No processing needed at the intermediate nodes,
once a circuit is established, thus does not heavily
rely on fast switches nor does it need to buffer data
(no delay jitter).
! Routing is part of circuit set-up.
Fast Circuit Switching
! First step does the routing but does not set-
up a circuit.
! Circuit set-up (or tear down) is done at the
transmission of the bursts by employing a
short control message.
! Circuit set-up can also be a “one-way”
process, where no acknowledgement from
the network is needed prior sending the
burst.
Packet Switching
! Data is sent w/o setting up a circuit.
! Usually employs distributed routing control.
! Store-and-forward mechanism needs buffering
and dynamically (re)configures the switches.
! Good for bursty traffic => allows statistical
sharing. Requires buffers and quick switching.
! Message switching: large sized packets packets
are assembled together at the end-nodes.
Requires larger buffers and message delay will
be high(er).
Datagram Based Packet
Switching
! Packet header (control information) and
payload are sent over the same
channel. Headers are “glued” to the
payloads. (E.g., IP)
Virtual Circuit based Packet
Switching
! Two (3) phases:
1. Setting up a VC (or routing)
2. Sending packets over the VC (switching)
3. (tearing down the VC)
! Setting up a VC does not require dedicated
bandwidth – just an entry in the routing tables of
intermediate nodes along the selected path. These
entries map the VC identifier (sent together with
each packet) to the route it has to go on to. This
table lookup is easier than making a routing
decision. (E.g., ATM has this capability)
Multiprotocol Label Switching
(MPLS)
! [Callon, 1997] Is currently standardized by
IETF.
! Similar to VC packet switching.
! Label-switched paths (LSPs) are established
(instead of VCs). Routing decision is made
only once at the establishment of the LSP.
! MPLS can handle different traffic types
(packets belonging to the same source and
destination pairs can have different LSPs
depending o their importance – and routing
can consider these metrics).
Multiprotocol Label Switching
(MPLS)
! The establishment of LSPs can be:
! control driven (performed by the network
according to its topology)
! Or data-driven (e.g., the first couple of
packets are routed by IP at each node, but
when the destination receives a given
amount of packets it triggers the
establishment of an LSP (e.g., for one TCP
stream).
Burst Switching
! Packets to the same destination are
assembled into bursts, that are sent over
the network with one connection set-up
(routing).
! One-way reservation: no acknowledgment
from the network is needed, thus pre-
transmission delay of the burst is reduced.
Burst Switching
! Three variations for bandwidth releasing:
! Tell-and-go (TAG): as soon as burst is
transmitted the sender sends an explicit
release message to tear down the circuit (like
circuit switching).
! Reserve-a-fixed-duration (RFD): each set-up
request specifies the duration for the circuit.
! In-band-terminator (IBT): a burst contains a
header and tail (terminator) to indicate the
end of the burst. (like packet-switching)
Burst Switching
! Let T be the time between the issuance of the
set-up request and the transmission begin of
data.
! T can be shorter with on-way reservation and
TAF or RFD than the time required to set up all
immediate switches.
! Although, if T is too small, buffering is needed.
Virtual cut-trough is the technique where if the
next hop is established the burst can be sent
right away (even when it is still being
received).
Switching in Optical Networks
Wavelength Routing
! Basically circuit switching in an optical network.
! All-optical path (no OEO conversions are
needed) and is established before data can be
sent.
! They provide high speed high-bandwidth
“pipes”.
! Lightpaths may be dynamically established.
! Not efficient for bursty Internet traffic.
Optical Packet/Label Switching
! Data remains in the optical domain
while headers may be processed
electronically (or optically – not mature
yet).
! Since limited optical processing is
available , VC based packet switching is
more popular than datagram (NO
OPTICAL RAM).
Optical Burst Switching (OBS)
! Provides temporary solution between
label and and wavelength switching.
! It is difficult to optically recognize ends
of bursts, thus OBS is likely to be TAG
and RFD.
RFD based OBS - JET
! Just Enough Time (JET) [qiao, 1997]
! Offset time between message and control-
message is greater than equal to the sum of
set-up times for the switches involved.
! Burst is buffered at the source => no buffers
or delay lines (FDL) are needed.
! If the requested bandwidth is not available
the burst is blocked (and will be dropped if no
buffering is available).
RFD based OBS - JET
T = ∑h =1 δ ( h)
H
T burst
S
t
1
t
δ1
2
t
δ2
D
t
δ3
TAG based OBS vs. JET
! Explicit tear down signal is used.
! Since loss of tear down signal would
result in wasting the bandwidth, each
source is required to periodically refresh
ongoing reservations (timeouts).
! JET is more bandwidth efficient.
pJET for Differentiated Services
! Prioritized JET, where two classes of bursts
are distinguished:
! Best-effort
! Real-time
! For real-time bursts, T is expanded by an
offset, enabling the network to make
reservations way ahead of time.
! The selection of the offset is a trade off
between independence of the two types of
burst handlings and induced delay for real
time traffic.
