How will "enhanced" or "next generation" premise systems perform at frequencies beyond 10Base-T Ethernet? With network speeds getting higher, revision of the 568A standard, and dispute over "Level 6" and "Level 7," this is a huge question for the datacom industry.

Testing systems to 350 MHz shows how well components match. If they don't closely match the 100-ohm system impedance target, they won't perform well as a system. To find out exactly how various system components, working together as a network, will perform under high-frequency conditions, we did some special testing. The results are quite revealing and lead to the following recommendations:

  • Look at end-to-end performance of a system. It's not enough to look at individual components, because a marginal component can have a huge negative effect on an entire system at high frequencies.

  • It's possible (though not easy) to assemble premise systems that will support protocols requiring the very best high-frequency performance.

  • A high-performance patch cord is critical to assure the optimum high-frequency performance of any premise hardware system.

    Currently, there's no formal specification test or pass/fail criteria that define what you should compare the gathered test data to. However, our tests demonstrate the importance of component stability. For example, some manufacturers have wall plates that perform much better than their patching system, and vice versa. Others have extremely good control of system components from both ends. The following are a few problem areas:

  • Impedance at high frequencies.

    A cable pair's impedance varies slightly along its length. This is due to small variations in manufacturing, such as concentricity, diameter, quality of twisting, and wall thickness. These small impedance variations, along each pair, create signal losses due to reflections of the energy transmitted along the pair. These losses are more significant at higher frequencies because as the wavelengths become shorter, small variations become more "visible" in the network. Impedance stability at low frequencies is easier to achieve but gets more difficult as frequency increases.

    Manufacturers make high-performance or extended frequency cables according to tight manufacturing tolerance controls. This stabilizes their impedance and minimizes the reflected losses in the network.

  • System Return Loss (RL). Return loss measures all the relative reflected power lost in a network. Low return loss is one measure of the quality of a system. RL, or echo, has an effect on pulse shape and therefore system error rate. System RL is so important for Gigabit Ethernet performance that it's one of the new test criteria now defined for existing Cat. 5, Cat. 5E, and the new Level 6 cabling. As with impedance, a network's RL becomes more important at higher frequencies.

    RL performance is a critical parameter for successful Gigabit Ethernet use because Gigabit Ethernet uses bidirectional transmission on each pair. From this, you'll see any reflected energy as noise at the receiver. Echo cancelation is the method you use to minimize this noise. You do this by complex integrated circuits, trained to cancel the reflected energy. Without cancellation, a worst-case Cat. 5 installation would not support Gigabit Ethernet or any other high-speed applications using bidirectional pair transmission.

Here's how we test for high-frequency performance. To measure the high-frequency performance of next generation networks, we use a realistic LAN premise hardware system. Here, a network analyzer connects to 2 m of patch cord (representing work area cabling), which then terminates through a wall plate to a 90-m run of horizontal cable. The horizontal cable connects to a patching system (normally located in a wiring closet). The last portion of the network is another 2 m of patch cord, which normally goes to a hub, bridge, or router. For the test, this last length of patching cord terminates back to the network analyzer. This arrangement allows testing in both directions through the network. It also approximates a long length basic link, as defined in TIA/EIA 568A, TSB 67.

Test results. After evaluating more than 20 premise hardware systems and various combinations of patch and horizontal cabling (of almost every major manufacturer), we observed some interesting results:

  • There's a difference between hardware systems, especially above 50 MHz.

  • A 2-m length of patch cord has a large effect on system performance, especially on RL and impedance.

  • All premise systems, regardless of their quality, benefit from using the best available patch cable.

This last item, regarding high-frequency network applications, was unexpected. For obvious reasons, we shared the specific results for each system tested only with its manufacturer.

With a lower quality patch cable, the impedance stability, which ideally should be close to the 100-ohm target line, begins to break down at about 50 MHz. Above this level, there's a wide variation in the measured impedance.

When we tested the high-performance hardware system with a high-performance patch cord, we found the impedance became remarkably smoother, staying relatively flat out through 350 MHz. The RL for this same system meets the reference target level. We didn't expect 2 m of patch cord could have such an impact on a network.

We also tested the impedance stability of the lower performance hardware system. With the low performance patch cable, the impedance begins to break down at around 30 MHz. Installing a high-performance patch cable in this low-performance hardware system improves its performance: The cable impedance is relatively smooth out to about 100 MHz.

Measuring RL for the lower quality system in combination with the lower performance patch cable shows performance to the reference line that extends out to only 10 MHz. Beyond 10 MHz, RL deteriorates rapidly. However, RL improves significantly for this system by changing the patch cords, with performance above the target level to about 50 MHz.

Skew. A recent addendum to the TIA/EIA 568A standard adds new test requirements for propagation delay and delay skew. Delay is a measurement of the time it takes for a signal to propagate through a length of cable. Because each pair has a different degree of twist, their electrical lengths vary, and the signaling delays are different. Delay skew is a measurement of the signaling delay difference from the fastest pair to the slowest. The skew measurements for the best hardware system combined with 2 m of low-performance patch cord exceeds the TIA/EIA 568A specified maximum allowable network skew of 45 nanoseconds (nsec). Again, simply changing the patch cord provides a significant improvement, with a performance of less than 25 nsec of skew at the highest frequencies. This is a much less skew or timing delay difference pair to pair.

ACR. ACR, or attenuation-to-crosstalk ratio, is roughly equivalent to the signal-to-noise ratio in a circuit. This is true because crosstalk is the most significant source of noise, and attenuation measures signal strength. In our testing, however, we didn't plot ordinary ACR. Instead, we measured power sum ACR. That is, we calculated the ACR using the more difficult power sum crosstalk measurement. Our testing shows the network has positive ACR past 200 MHz, which is impressive. It also meets 10 db power sum ACR at 160 MHz. This is good performance for a network that represents a near maximum length configuration. (This performance level has been promoted by a national distributor. More data and additional testing is available at Quabbin's website, www.quabbin.com.)




Sidebar: The Current Situation re: Network Speeds

Several new LAN operating protocols have just come to market, or are coming to market right now. All exceed the 10 Mbps transmission rate of the 10Base-T networks found in what seems like every office in America. Here are the new network protocols.

Asynchronous Transfer Mode (ATM). This network runs with signaling rates of 155 Mbps, with only a small amount of signal energy in the range above 100 MHz. ATM sends data in fixed length packets rather than in synchronized time slots. This type of packet technology is a superior technology. This technology is the wave of the future. Our current ATM scheme, however, will not be the final version of it.

100Base-T. This is the new Ethernet protocol that increases data rates from the current 10 Mbps to 100 Mbps. Many networks have already installed dual-rated Ethernet insert cards, which are capable of running Ethernet at 100 Mbps. There are, however, questions as to whether the networks will operate acceptably at ten times their current speed. Under the new protocol, you would use all four pairs in the cabling system rather than the two pairs used under the current scheme. This requires a significant change to the test methods for crosstalk; you now must deal with both near end and far end crosstalk.

Gigabit Ethernet (1000Base-T). This is proposed to run over Cat. 5 or better copper wire. A new Cat. 6 is being designed for protocols beyond Gigabit signaling.