One of
the interesting pieces of technology which has grown in popularity
during recent years is the optical free space datalink. In this month's
feature Carlo Kopp explores technology issues in modern optical datalink
design.
Conceptual Design
In the simplest of terms an optical datalink comprises a
modulated optical source, an optical package to focus the beam, a
transmission path through the atmosphere, a receiving optical package,
and optical detector, receiver and demodulator. A digital modulation is
imposed at the transmitter end, propagates through the channel and upon
reception the digital data stream is reconstituted for use.
In concept the idea seems very simple, but in practice
the engineering of such links is often anything but simple. The writer
in a previous life completed most of the design of a 32 Megabit/s
optical datalink, and can comment from direct experience.
Modern optical datalinks have been in the market for
about one and a half decades at this time, since high power LEDs and
semiconductor lasers became available at reasonable costs. Data rates
have followed Moore's Law, insofar as early designs ran at speeds of the
order of a Megabit/s, and contemporary products can deliver Gigabits/s.
While the speed performance of datalinks, driven by
receiver and transmitter technology, has grown and arguably will
continue to grow, the physics of the optical hardware and of
propagation through the atmosphere have not changed. In many respects
these are the key constraints to what optical datalinks will be able to
achieve. Much marketing literature for datalink products conveniently
glosses over what frequently is the biggest show stopper in the game.
To better understand the limitations and the future
potential of optical datalinks, we will explore in more detail the
design issues in transmitters, receivers and take a brief look at
propagation impairments to datalink operation.
Datalink
Transmitters
The transmitter is in many respects the simplest part of
any datalink design. A typical datalink, apart from any router or
bridging hardware which may be embedded in the same package, comprises
a line interface, power supply, modulator, optical driver, optical
source and some optical arrangement for focussing the output from the
source into a tightly collimated beam.
While the line interface is arguably trivial, for many
optical datalink designs, the design can be quite demanding. Because
optical datalinks are often located in relatively inhospitable
environments like rooftops, or upper portions of building structures,
they must be built to survive and reliably operate under harsh
conditions of high temperatures, low temperatures, rapidly changing
temperatures, often high humidity, nearby lightning strikes and
interference from mobile telephones and other reasonably strong RF
sources. Often the line interface cables must double up as remote DC
power feeds for the equipment proper.
A robust design typically uses a shielded cable, twisted
pair for low speed designs, coax for high speed designs and preferably
optical fibre for very fast designs. Where copper is used, a good high
voltage isolation scheme is advisable.
The data format going up the line interface (or down at
the receiver end) can be either in the format which goes over the
optical channel, to reduce the complexity of the hardware in the actual
optical head, or it can be another arbitrary format designed about the
line interface. Synchronous streams with an embedded clock are usually
preferred.
Data arriving at the line interface of the optical head
is used to drive the optical transmitter proper.
Transmitter designs vary widely depending upon the bit
rates in use and the type of optical source being used.
The simplest designs, operating at low speeds, will use
a high efficiency, high power Light Emitting Diode or LED. LEDs are
non-coherent sources, which produce a frequently fairly wide infrared
colour spectrum, usually centred around 0.8 microns with a Gaussian
spread in colour. Peak output powers can be up to Watts, and some
designs may employ ganged LEDs in an array to increase output power
levels. LEDs are highly capacitive loads for driver circuits, and
usually have to be driven very hard to deliver the required optical
output. As a result, the drive circuits usually pulse the LED with a
high current spike. Most LEDs do like this type of treatment and will
usually experience a decline in deliverable optical peak power over the
life of the LED. Therefore, if your elderly datalink on the rooftop is
beginning to show an increased bit error rate, odds are the LED has
followed the inevitable aging curve and is delivering a fraction of the
optical power it did when new. Time for an overhaul and new LED
installation.
Newer and faster designs typically employ solid state
lasers, a favoured colour is the eye-safe 1.55 microns. Lasers designed
for industrial automation are frequently a favoured component in
datalink designs.
Unlike the LED, which is a simple and robust component
which can be driven by a simple high current transistor switch, lasers
are much more finicky devices. A semiconductor laser is in simplest
terms a refined type of LED which incorporates a Fabry-Perot optical
resonant cavity in the design of the die. Current flowing through the
device causes LED-like light emission, which bounces back and forth
between the mirrors of the cavity. At some threshold current level and
thus light intensity, stimulated emission or lasing occurs, and the
device outputs a highly coherent high power light emission.
The threshold current of the laser is a parameter which
varies with the temperature and the age of the device. In a typical
pulsed laser circuit design, a continuous current, which is just below
the threshold current, passes through the laser to keep it biased just
below the point at which it lases. To initiate lasing, a signal current
is then added to the bias current to push the laser into full power
emission. To turn the beam off, the signal current is removed.
In practice lasers typically require a dedicated power
control loop, which senses the optical power level out of the laser and
controls the bias current. This compensates for device temperature and
age. Since lasers can dissipate respectable amounts of power, and
optical heads can be installed on very hot roofs, a good power control
loop will include an over-temperature crowbar to prevent the device from
be converted into a little pool of molten Gallium Arsenide.
Like LEDs, lasers are capacitive loads, albeit usually
smaller. This can complicate the design of driver circuits considerably,
if speeds of hundreds of Megabits/s are required in pulsed operation.
Designs running at high speeds are usually very difficult to operate in
a simple pulsed mode, and may instead employ a more complex analogue
modulation which is imposed by varying a drive current through the
laser. A good laser has an electrical to optical transfer function
(i.e. relationship between lasing current and output optical power)
which is fairly linear, so a modest modulation depth may produce very
little distortion in the modulation. Since lasers usually do not have
to share bandwidth with other services in adjacent bands, second and
third harmonic distortion is not the show stopper it is in radio
broadcast and communications.
In principle, a laser transmitter can exploit any
technology used in optical fibre communications, with the caveat that
higher power levels must be produced.
The modulated light produced by the LED or laser must
then be focussed by suitable optics into a beam which can be aimed at
the receiver, which is typically hundreds of yards away, or perhaps
even a kilometer or further away, depending on the design. Various
optical designs are possible, although in practice the most common
approach is some lensing arrangement. An LED with a large emission
angle is commonly focussed using a large lens which acts as a
collimator. Lasers usually produce very narrow coherent beams, and some
designs will get by with little more than a small cylindrical lensing
package attached to the laser device proper. If the system uses an EDFA
optically pumped amplifier, then a suitable lensing arrangement to
capture the light leaving the end of the fibre will be required.
The width of the beam produced is an issue in its own
right. The optical head which is bolted to a mast on a roof, or
attached to the side of a building, is ideally perfectly rigid. In
reality it is never perfectly rigid, since masts and even tall
buildings to sway in the wind, and almost may carry some structural low
frequency vibrations. Therefore the line-of-sight of the optical head
will drift with the mechanical movement of the mast or building.
If the beam produced had an infinitesimal width then a
deflection of X degrees at hundreds of yards would simply cause the beam
to drift off its aimpoint and away from the intended receiver, causing
the signal to be lost entirely.
Therefore the optical designer faces an inherent
contradiction - the narrower the beam the better in terms of optical
power delivered to the receiver, yet the narrower the beam the greater
the difficulty in keeping it properly pointed at the receiver. This
problem is not unique to rooftop datalinks - military lasers installed
on jets, helicopters and vehicles, used for guiding bombs and missiles,
frequently incur huge cost overheads in the provision of highly accurate
beam stabilisation hardware to avoid this sightline jitter problem.
Since rate gyros and servo driven mirrors are much too
expensive for rooftop datalinks, the usual solution is to make the
beamwidth such that it is wide enough to maintain receiver in the beam
even if the sightline of the optical head does drift around by a degree
or so. The price to be paid is in optical power delivered, since the
wider the beam, the larger the spot produced by the optics at the
distance of the receiver and the smaller the proportion of the
transmitted power coupled into the receiver optics. Indeed this is one
of the big limitations to optical datalink range.
Optical Propagation
At this point our notional optical datalink has put a
spot of laser (or LED) light over the receiver and its immediate
surrounds at some nominal distance of several hundred yards, or perhaps
even the odd kilometre. Our receiver, which we are yet to discuss in
detail, captures some fraction of the impinging light and converts it
into an electrical signal for further processing.
How does the atmosphere affect the behaviour of our
transmitted beam? The simple answer is in various deleterious ways!
Let us first assume that we have a nice hot dry summer
day, with no pesky raindrops or fog droplets to complicate life. We are
shining our laser beam close to parallel with the earth's surface, since
the curvature of the earth is negligible at these ranges and thus
refraction due to curvature is also negligible.
As the earth's surface heats under the sun, we get
thermals, or heated air rising upward. Assume our beam is travelling
above the roofs of various houses, commercial buildings, carparks and
other such artifacts of industrial age civilisation. What happens as a
result?
Different surfaces heat at different rates, and also
transfer heat to the surrounding air at different rates. The result is
that the columns of rising air above each of our artifacts will be at
slight different temperatures, which also means slightly different
densities and thus refractive indices. Since the boundaries between
these pockets or cells of air will be continuously undulating, as all
thermals do, we end up with a time variant change in the shape of the
boundary.
Geometrical optics tells us that light impinging on a
boundary between two areas with different refractive indices will bend.
For pockets of air with small differences in temperature, this bending
may only be a tiny fraction of a degree. But our beamwidth may only be
of the order of a degree. What this means in practical terms is that
our beam will be randomly zig-zagging along the general transmission
path it is following, the zig-zags varying in time with the behaviour
of the thermals.
The consequences of this are twofold. The first is that
the beam will become defocussed, reducing power at the receiver
slightly. The second is that the length of the transmission path
changes, and multiple transmission paths may exist for various parts of
the beam. The former effect can be compensated by pumping more power
out of the transmitter, but the latter cannot be easily handled (unless
we borrow the adaptive mirror technology the US Air Force is putting
into its multi-MegaWatt burn the ballistic missile out of the sky 747
mounted Chemical Oxygen Iodine laser). Why does it matter? If we aim to
pump Gigabits through our datalink beam, this effect may cause us
considerable heartache since it is no different in principle to the
modal dispersion effects we see in a step index or graded index optical
fibre. Except, unlike the fibre, the severity of the dispersion will
vary with the time of day and day in the year, and weather on the day,
since it is a result of local atmospheric temperature changes.
This is a non-issue for the older generation optical
datalinks, but will become a serious impediment once people attempt to
pump multiple Gigabits/s or more through a free space link with
distances of kilometres. How serious an impediment will depend on the
distance and local atmospheric environment, needless to say this problem
is not one which is trivially calculated or modelled.
To be trivial, though, let us hypothesis that half of
our beam gets delayed relative to its sibling by 0.01%, over a 1 km
transmission path. Let's assume the light seen by the receiver is a mix
of 50% from each half. Then the receiver gets a mix of equal
proportions, time shifted by 1/3 of a nanosecond. A bit cell at 3 GHz
occupies about - 1/3 nanosecond.
Dispersion effects due to atmospheric density pockets
will be an issue for transmission of multi-Gigabit/s data streams over
optical datalinks, in the same manner as modal dispersion is becoming
an issue in fibre LANs. How severe an issue will depend on the speed of
the link and the distances involved, as well as the local atmospheric
environment.
Is this the only headache the datalink designer faces?
Fog is without doubt the big show-stopper for optical
datalinks, since the water droplets are similar in size to the
wavelength and the their density is sufficiently high to completely
scatter all of the beam energy in a matter of metres. If you live in a
foggy neighbourhood, don't bother with an optical link unless the
downtime doesn't matter.
Areas which might experience high levels of atmospheric
pollution with dust and aerosols are also not the best environment for
an optical datalink. While the density may not be such as to blind the
link, it may be enough to cause a power loss and thus increased bit
error rate.
Rain is another headache for optical datalinks. Rain
droplets are much larger than the wavelength of the beam and thus will
cause most of the energy impinging on them to be scattered and lost.
Whether enough energy penetrates for the link to remain open will
depend on the rainfall rate across the beam. If the raindrops are large
and sparse, as in a summer shower, odds are some impairment to bit
error rate will occur but the link will remain up. A nice run of winter
drizzle from a Stratus or Altocumulus layer will most likely produce
the same effect as fog will. The author recalls a certain 256 kilobit/s
datalink which resolutely refused to run in rainy weather.
In practical terms, the propagation environment is the
big limiter to optical datalink reliability and performance. Therefore
if you live in a hot and humid climatic environment, this is not the
best choice in technology. On the other hand, if you live in a climatic
environment which is relatively dry and cold, and not hampered by fog,
then optical datalinks can produce excellent uptime.
For potential users of optical datalinks, some careful
study of the satellite literature on regional rainfall and fog
behaviour might just be a very good idea before investing heavily in
this technology.
Optical Receivers
No discussion of an optical datalink would be complete
without some exploration of the behaviour of receivers. Like
transmitters, the same caveats concerning optics, optical head
mechanical stability, line interfaces, and environmental conditions
apply.
In a receiver, a light gathering optical system, very
frequently a lens package, focusses the incoming light on to a PIN diode
or avalanche diode optical detector. In turn it produces a faint
electrical current which is amplified by a front end receiver and then
handled accordingly to demodulate the waveform and extract the data
stream. The same technology base used with fibre links can be exploited
quite effectively here.
Unlike a fibre system, a free space link has one very
nasty problem to cope with - background emission current.
The noise performance of a top end optical receiver
depends mostly upon the noise produced in the detector element and the
front end receiver. In a detector element, the dominant noise sources
are thermal effects and shot noise. Shot noise is produced mostly by
the tiny trickle of dark current, essentially leakage through the
photodiode structure of the detector. In a fibre system, designers
invest much effort into suppressing shot noise and may even cool down
the detector element with a Peltier thermo-electric refrigerator to get
better performance.
In a free space optical datalink, the light impinging on
the detector comprises the received signal from the transmitter, plus
whatever infrared light is emitted and reflected by the objects or
scenery within the field of view of the receiver optics. The easiest way
to visualise this is to imagine putting your eye up to the receiver
optics, where the detector is placed. You would see what the detector
sees, in effect the same image as seen through a telescope with similar
optical strength - the transmitter optical head and its surrounds.
The electrical current produced in the detector is thus
a sum of the signal current and the background current. If the scene is
very bright and reflects enough, the transmitter may be completely
swamped.
Since the background emissions and illumination around
the transmitter are not modulated, the result is a largely DC current
through the detector. In effect this current is no different from dark
current, except it is much greater in magnitude. Accordingly it produces
a lot more shot noise than dark current does. Therefore the same
detector used in a free space link will never match the noise
performance in a fibre application.
Are there any tricks a designer can play to beat this
problem? Infrared optical filters, the same technology used in
heat-seeking missiles, have been and continue to be widely used for
this purpose. Typically a rare earth doped glass filter with an
additional multilayer interference filter can knock out most of the
infrared outside the immediate colour of the transmitter. As a result,
a very substantial proportion of the infrared background can be removed
and the background current reduced to a manageable level, but it can
never be eliminated.
Pushing the Envelope
The big thrust in optical datalink technology today is
exploitation of recently devised technologies used in fibre
applications. These are techniques such as direct optical amplification
using spools of doped, optically pumped fibre, and Wavelength Division
Multiplexing (WDM).
Both of these technologies will provide important gains
in achievable transmitter and receiver bit rates. However, the
idiosyncrasies of the optical and transmission environment will not be
beaten that easily. The basic physics which result in beam jitter,
signal time dispersion and background current are implicit parts of free
space transmission and cannot be easily managed or constrained.
For the end user the important caveat is to match the
datalink technology with the local environment. If some careful thought
is applied, optical datalinks can be an exceptionally useful adjunct to
fibre or copper links. However, the predictions often seen in the
prospectus' of some vendors are likely to remain marketeers' fantasies
for the foreseeable future.