Optical Transport in the Core Network

RF over fiber has been made possible by an enabling technology—linear lasers. In the
core network, optical transport bandwidth quantity and quality is improving as new
enabling technologies became available, particularly linear optical amplifiers, optical
filters, and in the longer term, optical memory. In parallel, as we highlighted earlier,
optical DSPs have been proposed that support baseband processing at optical speeds.
As lasers become more accurate (the ability of a laser to produce a discrete frequency),
channel spacing can reduce.
Anarrowband channel in the optical domain is 25 GHz. As our ability to synthesize
optical frequencies increases, and as the frequency accuracy and stability of new optical
devices increase, capacity increases.
Wavelength division multiplexers and multiple-carrier generation frequency management
techniques are effectively delivering an order of magnitude increase in optical
capacity every decade. Splitting light (for example, in a prism) is potentially easier
than carrying a packet in an electronic buffer delivering some interesting multiplexing
opportunities.

When we define optical bandwidth quality, many of the performance metrics are
not dissimilar to RF bandwidth quality metrics. Dispersion loss over distance increases
as optical frequency increases. Higher-frequency, smaller-wavelength optical channels
will therefore have less quality than lower-frequency, longer-wavelength channels.
Quality of service can also be defined as how physically secure the fiber is—how difficult
is it to tap the optical bit stream.
At present, we use electronic switching at the (optical) network edge. In the longer
term, we might justify moving optical switching closer to the user/consumer using
Multi-Protocol Label Lambda Switching (MPL λS) to provide differentiated quality of
service.
Wavelength-division multiplexing provides the ability to configure optical bandwidth
to respond to different QoS requirements at an aggregated traffic level and to provide
multiple optical routing trajectories for resilience and redundancy, giving good
restoration capabilities. At present, the only software-reconfigurable network element is
the Optical Layer Cross Connect (OLXC), but in the longer term, we will have tunable
lasers and receivers supporting reconfigurable optical add/drop multiplexers.
Multiplexing options include electrical or optical time-division multiplexing to
combine input channels into a single wavelength or adaptation grouping in which
groups of wavelengths are added or dropped as a group (depending on laser tunability
and optical channel spacing).
As bit rates increase (2.5 to 10 to 40 Gbps), power has to increase. Optical networks
are power-limited rather than bandwidth-limited, just in the same way that radio networks
are power-limited rather than bandwidth-limited. In an optical network, higher
power creates problems with impairments and nonlinearities.
Linear impairments are independent of signal power and affect wavelengths individually.
Typical impairments include polarization dispersion, chromatic dispersion,
and amplifier spontaneous emission. Nonlinear impairments increase as power
increases and generate dispersion and cross talk.
As the number of wavelengths increases, the blocking probability of higher priority
traffic classes increases. We need to start considering using offset time-based access/
priority control to deliver differentiated quality of service. This, however, depends on
our ability to provide fast switches. It is hard to get a mechanically tuned grating to
switch at less than a millisecond, so this becomes a constraint. (There is no point in
having bandwidth available if you cannot provide access to the bandwidth.)
If we can increase the number of optical frequencies (that is, reduce channel spacing),
we can reduce the bit rate per optical stream and make switching at either end of
the pipe slightly easier. In radio network terms, this is rather like discovering the benefits
of using narrowband RF channels. It is also analogous to the way we used OFDM
in wireless LANs and digital TV to subdivide the frequency spectrum and slow the bit
rate (to improve intersymbol interference, which in turn is analogous to chromatic or
polarization dispersion).
Arrayed waveguide gratings are becoming available that can discriminate between
40 × 100 GHz optical channels and, in the longer term, 80 × 50 GHz or 160 × 25 GHz
optical channels. If we can switch optically, we can reduce power consumption by
about 75 percent compared to electrical switching and deliver a size footprint reduction
of 75 percent (glass and air replacing silicon and copper). However, optical switching
needs a new generation of enabling technologies.

Presently the options include the following:
 Micro-electromechanical devices (MEMS)
 Liquid crystal devices
 Electro- or thermo-optic devices
 Bubble switching (inkjet technology)
We can, for example, use MEMS to build lots of tiny microcells on a silicon chip. Tilting
the mirrors routes the optical data streams between, potentially, several thousand
input and output fibers. The trouble is MEMS don’t work fast enough for packet
switching; we can only use them to reroute around a failed fiber path—that is, for
restoration, reconfiguration, or protection (to drop the loading from a compromised
light path for example). If we compare optical switching with the existing optical/electrical/
optical switching we have today, we can say that an optical switch is fast but stupid,
and an optical/electrical/optical switch is smart but slow.
Figures 13.20 and 13.21 show superconductors as a potential halfway house. The
example is a 10 Gbps switch proposed by Conductus taking in an optical signal (refer
to the right side of Figure 13.20, processing it through a photodetector, performing
switching, and then amplifying prior to reconverting to the optical domain via a laser
diode. The photo detector, switch, driver, and GaAs pre-amplifier are all supercooled.
This adds an intermediate layer between the optical layer and electrical switch layer,
that is, a superconductor routing layer (see Figure 13.21).
Alternatively, we might try and do everything in the optical domain, but if we
wanted to use IP packet routing, we need the ability to buffer—to give us time to read
routing instructions and to smooth bursty traffic—and, at present, we do not have optical
RAM. If we are trying to multiplex lots of narrowband optical channels into a single
fiber (to relax routing/switching performance), we also begin to lose power in the
combining process. Broadband coupling is very lossy (typically 4 dB for a 2-channel
multiplex and 13 dB for a 16-channel multiplex), and narrowband filters are bulky and
expensive. Demultiplexers are also very hard to design and suffer from high insertion
loss and poor sensitivity.