High-speed data transmission requires new testing of installed Category 5 cabling.
Paul Kish / nordx/cdt Inc.
The basic corporate networking platform today is evolving from 10- to 100-megabit-per-second Ethernet for horizontal distribution. The demand for more bandwidth is driven by the need to deliver large amounts of different types of information. It is also stimulated by the relatively low cost--less than 100 dollars per port--of dual-rate, 10/100-Mbit/sec Ethernet hubs and network interface cards.
Next-generation Gigabit Ethernet networks--1000Base-sx/lx and 1000Base-T--are just around the corner and will become available this year, first on fiber and later on copper.
Advances in copper cable
Two key trends have spurred advances in unshielded twisted-pair (utp) cabling technology for data networks that make gigabit networking over copper possible. The first is the global acceptance of a common standard for telecommunications cabling in commercial buildings, first published as eia/tia-568 by the Electronic Industries Association and Telecommunications Industry Association (eia/tia--Arlington, VA) in 1991 and later followed by iso/iec-11801 of the International Organization of Standardization and the International Electrotechnical Commission (iso/iec--Geneva). The eia/tia-568 standard specified the contents of a structured cabling system. Subsequent telecommunications systems bulletins tsb-36 and tsb-40 specified performance requirements for Category 3, 4, and 5 cables and connecting hardware. These requirements were recently incorporated into a revision of the original standard, tia/eia-568a.
The worldwide recognition of this standard has opened up the Category 5 cable market, which is growing at a rate of 30% per year. The name "Category 5" has become synonymous with "lan cable," and the medium has superseded coaxial cable and all types of proprietary data cabling. Most data applications today are supported over utp Category 5 cable, which has an available bandwidth of 100 megahertz.
The second key trend is competition in the marketplace. This competition has led to many innovations in the design and manufacture of utp cables and connecting hardware. The pace of innovation, in fact, is ahead of the evolution of cabling standards. As a result, there are currently many choices of products with improved transmission performance compared to the minimum requirements for a Category 5 channel. There is also much confusion in the marketplace about the meaning and significance of various performance claims made about these extended-frequency products.
Here we will look critically at how the electrical parameters of these cables affect the performance of a local area network. Also, the use of sophisticated digital signal-processing technology for Gigabit Ethernet (or 1000Base-T, in its copper implementation) changes our perspective on traditional measures of performance, such as attenuation-to-crosstalk ratio.
The tia is moving rapidly to develop two additional specifications--one for installed-base Category 5 cabling systems and another for enhanced-performance Category 5--both of which are intended to address the Gigabit Ethernet application. It is expected that the two standards will be published early this year, to coincide with the release of the 1000Base-T standard by the Institute of Electrical and Electronics Engineers (New York).
The two tia specifications will be published as addenda to the tia/eia-568a standard. Most of the installed base of Category 5 cabling should meet the minimum far-end crosstalk (fext) and return-loss requirements that must be met to implement 1000Base-T, but this installed base will have to be verified in the field for conformance. Enhanced-performance Category 5 cabling will be the preferred wiring solution for Gigabit Ethernet in new installations because it will provide a higher performance margin.
It should be noted that the new tia transmission requirements for enhanced-performance Category 5 cabling will be specified up to a frequency of 100 MHz only. Contrary to some misleading claims in the marketplace, Gigabit Ethernet (1000Base-T) is designed to operate over a frequency range of 100 MHz, just as is the case with 100Base-TX. Likewise, to ensure system compatibility, both 100Base-TX and 1000Base-T will be designed to use the same distribution of spectral energy.
More and more bandwidth
Usable bandwidth over a twisted-pair channel is that portion of the frequency range where signal strength exceeds noise. Signal-to-noise ratio (snr), measured in decibels, is the parameter used to specify 100-meter utp channels. If the snr is positive, the signal will be recognized at the receiver. Available bandwidth is measured at the point at which noise and signal levels are equal.
Note that it is possible to extend the bandwidth of a medium to include the range where the noise level is greater than the signal level. However, it would require very sophisticated encoding algorithms to transmit successfully in this range. The added complexity of the electronics required for this encoding is not justified in practice, since coding efficiency would be very low--less than one bit per hertz.
The controlling noise source for most local area networks today is near-end crosstalk (next), which is interference between a transmitting and a receiving pair. In the absence of other noise sources, snr is the same as attenuation-to-crosstalk ratio, measured in decibels.
A fundamental relationship exists between the bandwidth of a channel, expressed in megahertz, and the information-carrying capacity or data rate of that channel, expressed in megabits per second. Bandwidth can be likened to the width of a highway, measured by the number of lanes of traffic it carries. Data rate can be likened to traffic flow, or the number of vehicles passing a particular point in a given time.
One way to increase the flow of traffic on a highway is to widen it. Another way is to improve the road surface and eliminate traffic bottlenecks. Similarly, it is possible to support a higher data rate for a channel of given bandwidth by using a more elaborate line code. These more elaborate encoding techniques can pack more bits of information into each hertz of available bandwidth, but they also require a higher snr.
The relationship between bandwidth and information- carrying capacity was discovered by Claude Shannon and published in a famous paper in 1948. The maximum information-carrying capacity of a noisy channel, according to Shannon, is limited by bandwidth, signal power, and noise power as linked together in an equation.
Shannon`s equation also provides an intuitive way of estimating the information-carrying capacity for any channel. If, for instance, a 100-MHz channel is divided into 10-MHz "bars," or frequency bands, we can use the equation to calculate maximum information-carrying capacity in bits per second for each bar. These values can then be added to give the aggregate capacity over the available bandwidth.
If we perform these calculations for a worst-case Category 5 channel, Shannon`s information-carrying limit works out to about 600 Mbits/sec. So where did we go wrong? This limit is significantly less than the more-than-1000 Mbits/sec required for 1000Base-T. The problem is that our assumptions about the channel model were oversimplified. The 1000Base-T protocol will not be implemented using a simple 2-pair, next-limited channel model. In fact, 1000Base-T will have to use some innovative and unconventional techniques to have a fighting chance of achieving so high a data rate over the installed base of Category 5 cabling.
To quote the forthcoming standard, "1000Base-T will implement a dual-duplex transmission with 5-level Pulse Amplitude Modulation [pam] coding, 4-dimensional 8-state Trellis Forward Error Correction encoding, pulse shaping and signal equalization for operation over 4-pair Category 5 cabling. All four pairs will be used for parallel dual-duplex transmission. Each transceiver will also implement complex digital signal processing for next cancellation and echo cancellation."
To put this in simpler terms, the information is divided into four parts, so about 250 Mbits/sec will be transmitted over each pair. Each pair, then, can be considered a separate channel carrying 250 Mbits/sec of information as an encoded signal, using the available bandwidth of approximately 100 MHz. Some of the added noise caused by multiple disturbers and signal reflections is canceled out at the receiver.
Will this scheme work over Category 5 cable? Maybe, but there are still some unknowns that could affect gigabit operation using this medium. For instance, signal reflections caused by mismatched components and external noise in the environment can be a problem for certain worst-case Category 5 installations.
So, does that mean that enhanced Category 5 cable should be installed? And what about Category 6 cable? The main benefits of enhanced Category 5 cabling are additional signal-to-noise margin and improved transmission parameters. As to Category 6, the iso/iec proposed a bandwidth of 200 MHz for this medium last September, but the standards work has only just begun and is not expected to be finalized until the year 2000. Even the specifications for the various components of a Category 6 channel are unclear at this point.
In addition, there are different ways of achieving a cabling solution having a bandwidth of at least 200 MHz. We have already seen some proprietary approaches, which may not have an enduring lifespan in the marketplace. A more cost-effective approach might be to use standards-based components.
Other parameters for enhanced cabling
In addition to next, there are two other transmission parameters that affect the performance of gigabit systems: return loss and equal-level far-end crosstalk (elfext).
Return loss is important because each of the four pairs in a utp cable will carry information in both directions, just as is the case with an analog telephone line. Any impedance mismatch between components will result in signal reflections, or echoes, that appear as noise at the receiver. Although this noise is partially canceled in the equipment, it remains a significant contributor to the overall noise budget.
The proposed return-loss requirements for an enhanced Category 5 channel are:
1 to 20 MHz: 17 dB
20 to 100 MHz: 17 - 10*log (f/20) dB
For return loss, there is a difference in performance between using an interconnect and a crossconnect in the telecommunications closet (TC). When the test setup was designed, the worst-case configuration turned out to be one in which the far-end termination is within 15 meters of the TC. Test results indicate that connector impedance mismatch dominates at high frequencies, while the impedance mismatch between the patch cord and the cable dominates at low frequencies. The return loss of the channel is 4.5 dB better for an interconnect than it is for a crossconnect using the same connecting hardware. So, everything else being equal, using an interconnect provides better performance than a crossconnect in enhanced Category 5 installations.
More so than return loss, fext may be unfamiliar to many in the cabling industry. Pair-to-pair fext is the crosstalk noise caused by a far-end transmitter on a neighboring pair as it interferes with the signal on a given pair. The fext noise level is measured in decibels relative to the magnitude of the transmitted signal. elfext is a measurement of the same noise, but the result is expressed in decibels relative to the magnitude of the received signal. Other ways to think of elfext are as fext minus attenuation, as the attenuation-to-far-end-crosstalk ratio, or as the attenuation-to-crosstalk ratio due to fext noise.
The requirements for elfext are still under discussion in the tia`s TR-41.8.1 Systems Task Group, with work proceeding to measure the contribution of connecting hardware to the elfext of the overall channel.
Power-sum fext is the combined crosstalk noise caused by far-end transmitters on neighboring pairs interfering with the signal on a given pair. The combined interference is calculated using a power-summation equation, as is power-sum next, which is the combined crosstalk noise caused by near-end transmitters on neighboring pairs interfering with the signal on a given pair.
All of these parameters affect the performance of an enhanced Category 5 channel, but it can be shown graphically that next remains the dominant source of interference at high frequencies. Nevertheless, fext noise is still significant for 1000Base-T, since in the current implementation of the protocol, it is not cancelable and must be taken into account as part of the overall noise budget.
Usable bandwidth is measured as the frequency range over which signal strength exceeds the noise level. If available bandwidth is broken down into bars, Shannon`s equation can be used to determine total available information-carrying capacity over the entire frequency range of a medium.
An enhanced Category 5 channel is affected by many electrical parameters, but in this graph, attenuation and channel power-sum near-end crosstalk and far-end crosstalk are the limiting factors on performance.
WILL BANDWIDTH REQUIREMENTS CONTINUE TO GROW?
In the last five years, we have witnessed an incredible increase in the power and speed of computer applications. Just three years ago, for example, I upgraded to a 486 personal computer considered to be state-of-the-art at the time and easily able to handle my most demanding applications. Then Microsoft introduced Windows 95 and a host of refinements to the operating system to support multi-threaded processing, realistic full-motion video, 24-bit true-color imaging, 3D graphic rendering, and other sophisticated applications. My Windows directory was bloated to more than 120 megabytes for the operating system alone, and my computer became overburdened and sluggish. I gave up the struggle to rejuvenate the old PC and purchased a state-of-the-art Pentium-based machine with all the latest bells and whistles. The new system "flies"--for today, at least--and, incredibly, all that horsepower cost much less than what I originally paid for my first ibm AT computer--now a museum piece collecting dust in my attic.
But the story of my 486 PC is not yet over. I decided to donate it, fully loaded with Windows 3.11, vintage games and software, to a family friend. The gift was much appreciated, and I felt good about it until recently, when the youngest member of my friend`s family received the latest Disney cd-rom game as a birthday present from his mother. I was asked to load the game--and you can imagine the disillusionment in the young boy`s face when his favorite Disney characters were arrested, moving in slow motion in a fabulously sculpted 3D landscape. The game was all but unusable, because my 486 PC is already obsolete by today`s standards.
So, what does this story have to do with data cabling and networking? The answer to this question lies in the fact that computer networks are becoming the lifelines of today`s enterprises. More and more workers are employed in the business of producing and processing information, and they are using powerful computer applications to do so. The amount and diversity of information are growing at a staggering rate. Success in a business operation depends on managing this information, as well as dealing with rapid changes in technology and meeting diverse customer needs in a highly competitive global economy. The effectiveness with which one manages information can control the cost of doing business.
This business climate has led to the unprecedented growth of the World Wide Web, as well as of corporate intranets, and these sources of information and vehicles for delivering it produce a huge demand for bandwidth, which affects data networking and the cabling that supports it.
Initially, Gigabit Ethernet will provide needed relief for congestion on corporate backbone networks, but the technology will eventually migrate from the backbone to the desktop. What killer application will demand that sort of bandwidth to the desk? No single application stands out, and the demand will not come about suddenly, but will evolve slowly, as has been the case in the past.
Consider, for instance, this imaginary scenario for the turn of the century. I will realize, in time, that my current high-end Pentium-based PC is becoming sluggish and restrictive. I finally yield to the inevitable and purchase a state-of-the-art machine--let`s call it a 500-MHz Gamma II processor with 512 megabytes of "eram" (enhanced random-access memory) a 100-gigabyte optical-disk storage system, and the latest AI-2000 operating system. At work, I can plug into a worldwide hyperlinked network, where I can quickly access, filter, organize, and process multiple forms of information from many sources simultaneously. Any networking speed of less than one gigabit per second seems terribly inhibiting. And that 7-year-old son--now 12--of my friend will spurn my offer of a 200-MHz Pentium-based computer, because, he says, it does not have enough horsepower to cope with the latest virtual-reality adventure from simworld Holobytes.
Paul Kish is ibdn product manager for nordx/cdt Inc. (Pointe-Claire, QC, Canada), and chairman of tia`s TR-41.8 subcommittee.