|Originally published December, 1997|
|¿ 1997, 2005 Carlo Kopp|
The optical fibre is without doubt one of the defining technologies of the last decades of this century. It has brought about revolutionary changes in long haul communications, and if single mode fibres are considered, fibre bandwidth is limited mainly by the equipment at either end of the link. It is therefore curious to note the lack of enthusiasm for fibre in the LAN community.
Is it an issue of cost, or an issue of technological timidness ? That is an excellent question to pose, and in this feature we will take a closer look at copper and fibre technology to see whether there is any substance behind the tortuously slow proliferation of optical fibre into modern LANs.
When discussing fibre the staring point must be what fibre replaces, which is the trusty old copper cable. Copper cables come in every possible variety, but for the purposes of high speed data transmission, two types of cable are of most interest. These are the twisted pair and the coaxial cable.
Conceptually, any cable be it electrical or optical, is a waveguide. We feed an electromagnetic wave into it, and a electromagnetic wave comes out of the other end. Unlike radio broadcasting, this wave is (at least in theory) mostly confined to the cable.
What are the metrics via which we can judge the worth of a cable ?
The first of these is attenuation, which is basically a measure of how much energy we get out of one end, for how much energy we put it. In a copper cable, made up of metal and a dielectric material, the deciding factor in high speed cables is dielectric loss and skin effect loss, the latter due to this interesting propensity of high frequency waves to confine themselves into the surface rather than the bulk of the material.
If we get a cheap copper cable, the skin effect loss can be high since Copper is a good, but not exceptional conductor. If we use the cable long enough, the surface will being to oxidise and cuprous oxides being what they are, the performance of the cable will degrade. That is why many more expensive cables used a more expensive plating on the cable wires. Probably the best material for this purpose is gold, which is both highly conductive and doesn't corrode easily. So if he want a really high performance copper cable, one step to take is to gold plate the conductors (or cladding/core in coax).
Dielectric loss is another interesting subject area, since dielectrics can have either very good or quite poor performance at high frequencies. The humble thin-net coax we all love and know is most often made with a solid polyethylene dielectric, which unfortunately makes it quite useless if we want to un beyond 50 MHz or so, particularly over a distance. Air is usually the dielectric of choice for faster applications. In practice this means air filled coaxial cables with circular polyethylene spacers every so many centimetres, or a polyethylene foam, especially with a twisted pair cable.
What are the drawbacks of a foam or air dielectric cable ? The first is that the plastic will oxidise and deteriorate with age, increasing losses, the second is that moisture can and usually does find its way into the cable, also increasing losses and often also providing an electrolyte for corrosion in the cable.
Any more warts we can think of ? Well there is this little issue of electrical thermal noise, which arises in any conductive material and increases with temperature. If you are trying to fish a very faint microvolt signal out of a cable, having it embedded in a similar amount of noise is not going to do good things for your error rates.
Another naughty property of most copper cables when transmitting wideband digital data is a frequency dependency of cable phase , which added to the frequency dependent (ie increasing) attenuation loss of the cable will cause often very nasty shape distortion of signals. We feed a nice trapezoidal stream of pulses into a cable at one end, and get nasty looking, much fainter, mush out of the other end of the cable. This basically sets a bandwidth limit on any Copper cable, since the high frequency components of the pulse stream are what carry its shape and allow us to distinguish ones and zeroes at the other end. Poor bandwidth translates into smearing of pulses (especially trailing edges), which in turn contaminate following pulses, increasing the error rate.
What other vices does copper have ? Well, the is the wonderful phenomenon of crosstalk in twisted pair cables, a traditional headache in the telephone business. Run several twisted pairs in a bundle and you will very quickly find that you get both inductive and capacitive coupling between separate twisted pairs. Basically energy will leak between the cables. If you are trying to detect a faint electrical signal in one cable, while transmitting a hefty signal in its neighbour, you may find yourself in the situation where what you are trying to detect is weaker than what is leaking from the neighbouring cable. This incidently can also be a problem with coaxial cables used for bidirectional transmission (eg cable modems on cable TV coax).
Much is often said about the superior resilience of copper cables to mistreatment such as bending and kinking, especially in comparison with fibre. This is alas a canard of truly gross proportions when talking about high speed cables. Coaxial cables do not handle kinks well at all, since these distort the geometry of the cable and change its characteristic impedance. Any change in cable impedance causes reflections of the signal when it hits the impedance discontinuity, and if we are trying to carry high speed data, it produces more low level hash to be sorted from the real signal. This is true of both coaxial and twisted pair cables, and shows up very clearly if you apply a Time Domain Reflectometer to any so abused cable. The author less than fondly recalls a sudden increase in error rate on a LAN cable which was later traced to an electrician having crushed the cable with a cable tie, in attempting to attach a mains power cable to a bundle of data cables, these being the arguably more resilient twisted pair as well.
Impedance variations can cause other forms of misbehaviour in copper cables. One important issue with copper cables, whether used for LANs or busses, is the problem of proper cable termination. The load at either end of the cable should ideally have an identical impedance to the cable, for all frequencies on interest. If the impedance differs, a reflection of the aforementioned variety will occur. As a result, link performance will be impaired. It is therefore no surprise that copper Fibre Channel cables, running at 1 Gigabit/s, employ often very clever matching networks at both ends. These usually serve to "equalise" the shape distortion in the cable, as well as to avoid impedance mismatch reflections.
Another problem with copper cables is their potential for "radiofrequency environmental pollution", using trendy language for good old RFI/EMI. Connectors on coax often leak and radiate the signal, at a low level, from the cable. With twisted pair cables this type of behaviour can be much nastier, since a kink in the cable producing an impedance discontinuity can also cause the cable to behave like an antenna, and radiate part of the signal carried. So if you don't care about being eavesdropped from the building next door, spare some thought for neighbourhood TV viewers.
While the "RF pollution" problem is as yet not significant in
Australia, it has become a big issue in the US and Europe, with the US
having a significant RF electrical noise threshold in most higher
density urban areas. While much of the problem is being blamed on leaky
cable TV connectors, we can rest assured that the proliferation of 100
Mbit/s twisted pair LANs has not helped in the slightest.
So far we have considered only a benign operating environment, where security and resilience to malicious attack have not been issues. Coaxial cables can be easily tapped into by eavesdroppers, and twisted pair can be inductively eavesdropped without leaving any physical trace. Of course, a truly malicious soul can simply couple a Tazer stun gun or shortwave transmitter into your copper cable and toast every device connected to it.
If we however don't care about polluting the neighbours with RF emissions, and are not fussed about the neighbourhood info-terrorist catching our passwords, or frying our machines, lets look at the issue of costs.
Copper is cheap, so we are told, much cheaper than optical fibres. Go to any number of sites and we will find the local LAN made up of segments of 10-Base-5 Ethernet cable, mixed with segments of 10-Base-2 Thin-net cable, 10-Base-T twisted pair cable, and often one or more varieties of 100-Base-T or proprietary clone 100 MBit/s twisted pair cables. Need more bandwidth or extra stations, lets drag a few more cables through the ceilings. Add yet another interface card to the hub or switch.
So as much as many in the LAN community love their Copper cables, it is without any doubt that there will be a good measure of "create-oneself-a-job" in adhering to copper based networks, moreso with the impending to Gigabit speed LANs.
Having explored the vices of the established copper infrastructure, let us now turn to the optical fibre and explore its characteristics a little.
Optical fibres are commonly referred to as light pipes, but this is a somewhat crude analogy. Fibres are optical waveguides, and come in a number of varieties with characteristics no less diverse than their copper cousins, and in turn quite precisely tuned for specific applications.
The basic idea behind any fibre cable is that of using a core and cladding with slightly different refractive indices, achieved by doping the core and cladding glass with different additives. When light impinges from the core on to the boundary between the core and cladding, it is reflected back into the core, and is in effect trapped. Shine light into the core of the cable and it will be guided along in a fashion very similar to a radar waveguide.
The key to good optical fibres is high purity doped glass, devoid of structural imperfections, absorbent contaminants and geometrical imperfections. Top of the line single mode fibres can carry Gigabit speed optical signals for tens to hundreds of kilometres.
The simplest type of optical fibre is the Step Index (SI) fibre, which is considered to be obsolete in the late nineties, but can still be found in relatively undemanding applications. In an SI fibre, a thick core is surrounded by a thin cladding, the most common variety being the 100/140 micron fibre. The 100 micron diameter core is surrounded by a 140 micron diameter cladding, in effect a 20 micron thick layer of cladding glass.
The 100/140 SI fibre is often found in short haul and LAN applications, since it can couple a large amount of light easily into the fibre. The ability of a fibre to couple light in is termed its Numerical Aperture (NA), a measure of the solid angle into which light is coupled into the fibre.
The drawback of the large NA SI fibre is poor bandwidth, resulting from the physics of propagation within the fibre, an issue which has an important bearing on how newer types of fibre are built.
Consider a large NA SI 100/140 micron "multimode" fibre, into which we blast a pulse of light, which has a nice uniform spherical wavefront when it hits the face of the fibre (ie the light is produced by an "ideal" point source). Some of the light waves impinge on the core/cladding boundary at shallow angles, some at even shallower angles. The shallower the angle, the lesser number of times the light "ray" is reflected, and the lesser distance it has to travel to reach the other end of the fibre. The result of this is that light coupled in around the middle of the core travels through the fibre in less time than light coupled in around the edges of the core. The result of this is that the pulse of light we fired into the fibre spreads out in time the further it travels, exhibiting similar naughty behaviour to our old friend, the copper cable. This effect is properly termed "dispersion", as the pulse disperses in time with increasing distance. That is why the large NA SI fibre is today only used for applications like connecting keyboards to to machines, or peripherals to local hosts, since its bandwidth is quite poor.
By the early eighties, the large NA SI fibre was pretty much superceded by the much more cleverly designed Graded Index (GI) fibre. In a Graded Index fibre there is no sharp transition between the core and the cladding, as a result of which we do not get a sharp bounce but rather a bending of the light beam. By cleverly choosing the refractive index profile of a GI fibre, it can be made to have much lower dispersion than a SI fibre, so much so that useful bandwidths of Gigabits/s can be achieved over hundreds of metres.
The advantage of the GI fibre is that it still retains a respectable Numerical Aperture, and can easily capture light and couple light in and out of a receiver or transmitter. It was designed for the LAN environment, to provide a relatively cheap interface yet retain good dispersion (bandwidth) performance. GI fibres for LAN applications typically have a 62.5/125 micron geometry, also 50/125 micron fibres can still be found.
The fastest and longest ranging fibres are the Single Mode types (SM). A single mode fibre uses a very small core, with a diameter of only several wavelengths of the light it is to carry. As with a radar waveguide, this produces an interesting effect, in that the fibre will only carry a single mode of lightwave propagation. As a result, there is no dispersion effect of any substance in a SM fibre, and it has an extremely high bandwidth as a result. Dispersion effects if anything result from differential propagation delay resulting from wavelength dependency of propagation velocity, sometimes called colour dispersion. This effect is circumvented by the use of Single Mode lasers to drive the fibre, since all of the energy of the transmitter goes out at a single wavelength.
It is the SM fibre which is most often touted as the example of what can be done with fibres, and certainly if used properly, SM fibres can indeed carry multiple Gigabits/s over hundreds of kilometres without any repeaters.
Optical receivers are a fascinating subject within themselves (the author having designed some over the years), and in the most simple of terms involve the use of a detector diode of some variety (eg PIN diode or avalanche diode), and a wideband amplifier. Such receivers are usually so sensitive to RFI that they must be whole enclosed in shielded metal boxes. Modern "receivers" are usually integrated designs, in a metal casing with a DIP footprint, containing the detector element and the amplifier, and sometimes also a Peltier cooler element to reduce noise.
Optical transmitters are another area which can be discussed at great length, the simplest designs being infrared Light Emitting Diodes (LEDs) which are most often limited in speeds to tens of Megabits/s, and semiconductor lasers, which are much faster but also much more finicky to drive and cool. Letting a semiconductor laser overheat is not a good idea, since you quickly end up with a puddle of GaAsP in your transmitter casing. Lasers vary significantly in cost and performance, with cheap and basic designs for LAN applications running between tens and hundreds of dollars unit cost, and sophisticated cooled single mode long haul communications Lasers costing hundreds to thousands of dollars per unit. Typical modern designs are yet again packaged into metal casings with Peltier coolers, DIP footprints and embedded driver circuits.
If you plan to design yourself a fibre LAN interface, you need only specify the transmitter and receiver characteristics, and purchase and drop modules directly on to your printed circuit board (see the Fibre Channel feature for details).
Interconnection technology and connectors for fibre applications are now very mature, with a wide range of connector designs available.
Compared to copper high speed cables, optical fibres are well behaved as a transmission environment. They are immune to RFI and crosstalk, do not radiate interference, are quite hard to tap, since they are glass, they also cannot propagate damaging electrical signals (accidental or intentional) into user equipment.
Since modern GI and SM fibres are extremely fast compared to copper cables, they are much more durable as well as well-behaved cable for Gigabit networking.
While fibre has many significant performance and robustness advantages over copper LAN cables, fibre also has its quirks. Fibres, just like high speed electrical cables, do not like being bent, crushed or kinked. Usually a fibre cable is protected against such mistreatment by using cables with very rigid plastic structures to protect the glass fibre from mechanical stress. The weakest link is usually the fibre patchcord from the wall socket to the host.
Connectors can also be troublesome, although much less than in the early days of the SMA connector. The issue with connectors is typically that of preventing dust, lint and other forms of typical office dirt from getting into the connector.
On site fibre cable terminations seem to be the biggest cause of unhappiness in the user community, since fibre does demand a lot more care in attaching connectors, than older copper cables did. While newer connector designs for LAN applications typically allow for cleaved terminations, a huge advance over the ground and polished terminations of the eighties, putting a connector on to the end of a fibre is still something you can't contract the neighbourhood electrician to do, just as the same party may not do a spectacular job of laying fibre cables.
Putting Optical Fibre into the LAN
Optical fibre variants of LAN equipment have been available since the mid eighties. Some sites boldly bit the bullet and moved to fibre, and feedback the author has had is generally very favourable. Typically there is a ramp-up period during which the LAN maintainers must be trained in the technology, taught how to diagnose cable problems, use optical test equipment, and install connectors. Once this initial transient cost is absorbed, the cost differential is solely in equipment adaptors, patchcords and cables.
A decade ago the arguments for fibre were primarily in electrical robustness - the ability to survive neighbourhood lightning strikes, electrical ground faults in buildings, and nasty power surges and spikes. From a performance perspective, copper cables were easily adequate for the purpose, from a handling, installation and maintenance perspective, copper cables were much cheaper.
We are now confronted by the next move in the LAN technology base, going up from 100 Mbit/s to 1 Gigabit/s speeds.
What this will mean for the LAN user base is that much of the existing copper cabling base will be quite useless, since the cables are too slow electrically. The choices will be to move to faster coaxial copper cables (twinax), shielded twisted pairs on shorter cable runs or to optical fibre. The cost differential between fibre and copper cables, connectors and components will drop significantly, since high quality high speed copper cable is not that cheap to make, nor are the components. Moreover, faster copper cables will need to be handled more delicately than fibre, unless they too are made with stiffening structures in the cable to prevent kinks and bends.
Copper will still retain some price advantage in interface adaptors, but these will be much less robust than existing lower speed products, since faster transistors have smaller die areas. We will see the cost of LAN electrical hardware into the cost bracket of RF microwave hardware, as will be the case with installation and support costs. If you trust your neighbourhood electrician to do your Gigabit/s LAN copper cabling, since he did an admirable job of terminating your 10-Base-T cables on to your Krone blocks, think again ! Going to site-wide Gigabit/s LANs will involve some sticker shock, regardless of whether copper or fibre is used.
We can expect to see increasing pressure over the next decade to control RF emissions from LANs, whether this will come from the ever zealous green movement or from consumers unhappy with RFI ruining their TV and radio reception, remains to be seen. Other pressures may also develop, such as insurers mandating the use to fibre to protect against info-terrorism and eavesdropping, something which may become a major issue for the finance sector and government.
Therefore the strategic decision to go to copper or fibre in a Gigabit/s LAN upgrade will be a fairly sensitive one. Some cost savings may be achieved with copper, with some loss in robustness, and the risk that five or ten years downstream the whole lot has to be replaced with fibre, either to handle the step up to 2 Gigabit/s, or to satisfy other constraints.
From a technical perspective it is curious that so few LANs have moved to fibre so far, since such users have a much better long term investment, capable of handling many generations of networking equipment. Clearly the short term cost imperatives of minimising upfront expenditure appear to be dominating the marketplace.
The future clearly belongs to fibre, and the longer the user base dithers on making this jump, the greater the long term costs they will incur.
|$Revision: 1.1 $|
|Last Updated: Sun Apr 24 11:22:45 GMT 2005|
|Artwork and text ¿ 2005 Carlo Kopp|