4:Wpm - 4:30pm (Invited)
`oc 2.2
`
`Technologies and Architectures for
`Multiwavelength Optical Cross-connects
`
`W. J. Tomlinson
`Bellcore (Rm 3X-369)
`331 Newman Springs Road
`Red Bank, NJ 07701
`Email: wjt@cc.bellcore.com
`
`Summary
`
`Wavelength-selective optical cross-
`connects are expected to be key enabling
`elements
`for multiwavelength
`optical
`networking.
`In designing such cross-
`connects, there are significant interactions
`and
`tradeoffs
`between
`the
`internal
`architectures of the cross-connects and the
`hardware technologies used to implement
`them. With some switch and multiplexer
`technologies, it is necessary to use dilated
`architectures, which increase the complexity
`of the cross-connect, in order to meet system
`In
`requirements for crosstalk rejection.
`cross-connects
`requiring
`wavelength
`translation, different internal architectures
`require different
`types of wavelength
`translation elements.
`A
`for
`requirement
`critical
`multiwavelength cross-connects
`is high
`crosstalk rejection. In simple point-to-point
`multiwavelength
`systems, wavelength-
`selective components with a crosstalk
`rejection of -15 dB are sufficient (provided
`that all the signals have similar intensities).
`In multiwavelength
`optical
`networks,
`involving wavelength reuse, it is possible to
`have crosstalk between two channels at the
`same wavelength. In this case the signals can
`interfere coherently, and to eliminate the
`this
`coherent
`crosstalk,
`effects
`of
`multiwavelength
`cross-connects
`with
`crosstalk rejections of 35 dB or more are
`
`required.
`for
`requirement
`key
`Another
`for
`components
`wavelength-selective
`multiwavelength optical networking is that
`the useable channel bandwidth (over which
`the loss and crosstalk rejection requirements
`are met) be as wide as possible. The useable
`bandwidth of such components has a major
`impact on
`the
`requirements
`for
`the
`wavelength stability of laser sources. It also
`effects the number of wavelength-selective
`components that can be concatenated in a
`network, which
`impacts
`the optimum
`network architecture and
`the maximum
`In most components
`network size.
`the
`characteristic
`that
`limits
`the useable
`bandwidth is crosstalk rejection, not insertion
`loss.
`
`a
`of
`functionality
`basic
`The
`multiwavelength optical cross-connect is to
`take signals from N input fibers, each of
`which carries up to M different signals on
`different wavelength channels, and to route
`those NxM signals to N output fibers. The
`in such cross-
`probability of blocking
`connects can be reduced by including the
`capability to translate an input signal on one
`wavelength channel to a different wavelength
`channel, but the need for this capability a
`controversial issue.
`
`53
`
`FINISAR 1012
`
`

`

`to date it is
`the best devices reported
`necessary to use dilated architectures to
`achieve sufficient crosstalk rejection, and
`interactions between channels may require
`additional dilation.
`wavelength
`for
`Technologies
`translation are at a much earlier state than
`switch technologies, although there have
`been some impressive demonstrations of
`prototype devices. Devices that accept an
`input at any arbitrary wavelength, and
`translate it to a fixed output wavelength,
`would be used at the output of a cross-
`connect. Most of
`the experimentally-
`demonstrated devices are of
`this
`type.
`Devices with a fixed input wavelength and a
`variable output wavelength would be used at
`the input to a cross-connect. Devices that
`could accept an arbitrary input wavelength,
`and
`translate it
`to an arbitrary output
`wavelength could be located in the middle of
`a space-division switch fabric, and would
`simplify
`that
`fabric, but such devices
`currently appear to be the most difficult to
`realize.
`In summary, there is no clear winner
`for cross-connect technology, and there are
`many opportunities for further research and
`development.
`
`cross-connect
`the
`of
`Many
`architectures currently being studied make
`use of wavelength demultiplexers on each of
`the N input fibers, a set of M (one for each
`wavelength) NxN
`space-division
`(non
`wavelength selective) switches, followed by
`a wavelength multiplexer for each of the N
`output
`fibers.
`Since
`the wavelength
`de/multiplexers are used in series, a crosstalk
`rejection of -18-20 dB per component is
`sufficient, and can be achieved by most
`available technologies. Multiplexers using
`multilayer dielectric filters generally provide
`the largest useable bandwidths, but it is
`difficult to achieve channel spacings of <2
`nm with this technology.
`switches,
`For
`the
`space-division
`micro-optic devices using mechanical motion
`of prisms or other beam deflectors provide
`excellent crosstalk
`rejection and
`low
`insertion loss, but do not scale gracefully to
`larger-dimension
`switches.
`Multiple
`switch elements can be
`electro-optic
`integrated on a single substrate (e.g. lithium
`niobate), but the required crosstalk rejection
`can only be achieved by using dilated switch
`architectures, which require many more
`switch elements. Semiconductor switch
`elements, with switchable optical amplifiers,
`may be able to provide sufficient crosstalk
`rejection, but the technology needs further
`development to establish its capabilities. A
`recently-described technology, using liquid-
`crystal switch elements, appears to provide
`the required crosstalk rejection, but in a
`bulk- op tic s configuration.
`Using acousto-optic effects (in lithium
`niobate waveguides) it is possible to make a
`2x2 switch element that can simultaneously
`provide "independent" switching of multiple
`wavelength channels. In principle, a cross-
`connect using this technology would require
`many
`fewer switch elements
`than
`the
`approached described above. However, with
`
`54
`
`