An independent research program that measured LAN channel return loss and studied its echo and distortion noise effects has unearthed some interesting information.
March 1, 2001
In 1997, the Institute of Electrical and Electronics Engineers, Inc. (IEEE) 802.3 Ethernet group began defining the chip characteristics necessary to run Gigabit Ethernet (1000Base-T) on four pair UTP copper. Because the group wanted Gigabit Ethernet to signal properly on the worldwide installed base of Category 5 copper networks, it asked the Telecommunication Industry Association’s (TIA’s) copper systems committee (TR-42) to define the minimal electrical characteristics for these Category 5 premise networks. The group needed the common test parametres and a new test characteristic: return loss (RL). Up until that time, the TIA’s 568A local-are network (LAN) standard had not specified RL.
Acting independently in 1997, Quabbin Wire & Cable Co., Inc. began measuring LAN channel RL and studying its echo and distortion noise effects; that program continues today. The technical information that is the basis for this article is a result of the company’s research. The data has also been contributed to the TIA for its benefit and presented at a recent BICSI conference.
The TIA has now quantified RL for Category 5e LAN components and also for links and channels. The newest hand-held field testers readily provide pass-fail measurements.
What is return loss (RL) and why was it previously ignored? To understand the answer to this question, you must comprehend how the different versions of Ethernet transmit signals. The commonly used 10Base-T Ethernet and most versions of the speedier 100Base-T use unidirectional signaling. As unidirectional signaling sends data on one pair and receives on another, signal echo or RL noise within a given pair never interferes with a receiver. The chip set’s receivers do not have to compensate for a noise they will never hear. RL is simply signal echo within a pair. (It should not to be confused with signal crosstalk, which is signal noise between pairs.).
To enable Gigabit Ethernet to signal at 1000 Megabits per second (Mbps), the IEEE decided to use bi-directional signaling on all four pairs and to implement five level voltage encoding. Each of the four pairs would be simultaneously sending and receiving data with a transceiver chip at each end. Since the ends of each pair are now sending and receiving, signal echo or RL noise matters. Also, since Gigabit Ethernet transmits more data in the same unit time, the signals are inherently more sensitive to noises in general.
Echo noise produced by RL is relatively easy to understand, and much has been written about it. (For information, refer to www.quabbin.com/tech_briefs/ whatisrl.html.) However, RL has another effect that is not widely understood: signal distortion. Transmitted pulse distortion is especially important for all higher speed signaling schemes, such as ATM and 100Base-T, even though they use unidirectional signaling.
SIGNAL DISTORTION EFFECTS
RL is created by discontinuities and impedance mismatches within a network. Discontinuities occur at bent or distorted pairs, plug-to-jack physical connections, and at the interface between a pair’s conductors and a connector. Impedance mismatches are usually within a pair (along its length) or between connected pairs in different cables. All of the components in a network channel must be designed for 100-Ohm impedance with very little variance. If a plug and jack or a patch cable and horizontal cable are not closely matched to the 100-Ohm target, some of the transmitted signal gets reflected and RL is created. An ideal or perfect channel would have none of these defects and, therefore, no RL noise — only signal attenuation.
RL distorts digital signaling by causing the transmitted signal pulses to spread out in time. In the worst case, the pulses actually interfere with, or spread into, adjacent pulses, termed “inter-symbol interference” or ISI. The chip sets for 100Base-T and 1000Base-T are designed to accept a certain amount of ISI and to equalize the distorted signal, but too much overwhelms them. Networks with too much ISI have bit errors at the receiver, require data retransmissions, and effectively slow down or even crash.
A mathematical model was developed for a channel to prove and understand this RL/signal distortion effect. The advantage of having a mathematical model or algorithm is that it can be modified to run “what-if” simulations, similar to a spreadsheet calculation. The equation defines the transfer function of a channel, including terms for input voltage, output voltage, attenuation, propagation constant, length, reflection coefficient and transmission coefficient. (A detailed discussion of the equation may be reviewed at www.Quabbin.com/tech_briefs/rlndt.html).
Three different channels were modeled, each with the same physical geometry. Each consisted of a 90-metre horizontal cable, a two-metre equipment cord, a two-metre patch cord, and a two-metre work area cord.
Channel Model A is an ideal or perfect channel, impossible except in a simulation. It has attenuation but no RL because it has been designed with no impedance mismatches or discontinuities. The transfer function for Model A is shown in Figure 1. Since there are no impedance mismatches or RL, the curve is smooth.
Channel Model B has only about 2 Ohms difference between all cords and the horizontal cable. The connections are also very good, with 18-dB RL at 100 MHz. Model B is quite close to actual channel solutions now being sold as “Category 6”. The transfer function for this very good channel is also shown in Figure 1. (Note that the curve shows some roughness but is quite similar to the ideal Model A).
Channel Model C has relatively large impedance mismatches. The patch cords are 108 Ohms, 88 Ohms and 88 Ohms, respectively, and the horizontal cable is 108 Ohms. The connections consist of one at 10 dB and two at 18 dB RL. Model C is representative of a typical older legacy channel that meets Category 5 requirements and is now running 10Base-T Ethernet. It does not meet the requirements of Category 5e. Model C’s transfer function exhibits the most roughness. (See Figure 1)
An input data pulse was applied to each model. The pulse equation was converted from time domain to frequency domain and multiplied by the transfer function, and then the output pulse was calculated (converted back to time domain). As you can imagine, the calculation is complex. (A more precise explanation is available at the web address listed earlier). It was then possible to directly compare input to output for the three models. The input pulse is shown in Figure 2. The pulse has a rise time of about 3.4 nanoseconds (Ns) and a width of about 8 Ns. It represents a generic high frequency pulse that is actually quite similar to that used in 100Base-T signaling.
The three simulated outputs can be seen in Figure 3. Ideal Model A’s output pulse represents a target that any network should hope to approach. The pulse rise-time is very good and the pulse trailing edge shows spread, due to attenuation alone. Model B’s output is also very good — virtually identical to Model A. However, Model C’s output is not nearly as good. There is also a bump in the trailing edge that occurs well outside the 8 Ns pulse width of the input pulse. The receiver could interpret this as a bit-error.
The models were then used to generate “eye-pattern” diagrams, which can show the effectiveness of real channels processing high frequency data pulses. Three-level voltage pulses were applied, which is very similar to 100Base-T encoding. (Fast Ethernet or 100Base-T is a complex multi-level signaling protocol to which many networks are currently upgrading). A random data sequence was applied to the transfer functions and the output shifted in time, generating the pattern. The size of the “eye” openings represents the receiver-input quality, with the top “eyes” being positive voltage response, and the lower eyes being negative voltages. Open, or large, eyes are better, as they are equ
ivalent to less ISI.
The eye-pattern charts of Models B and C are shown in Figure 4. The Model B channel shows healthy receiver input, even when transmitting a relatively complex, higher frequency pulse. The positive and negative eye openings are quite open. Channel B provides a strong signal input for the receiver chip to process. Model C’s (Category 5 legacy channel) eye-pattern is not nearly as good. The negative voltage eyes are not wide open and the positive eyes are almost shut. This channel’s relatively poorer receiver input is due to the bump in the output pulse, higher channel RL and increased ISI.
IMPROVING A CHANNEL
If a channel’s transfer function is relatively smooth, the chips in the network interface card (NIC) or hub can easily compensate for the ISI and distorted signal quality by using a relatively simple equalizer function that is the opposite of the cable’s attenuation. When the channel function is as rough as Model C’s, compensation becomes much more inefficient. Most equalizers can not fully compensate, resulting in data and timing errors, re-transmissions, and therefore a slower channel. Complex 100Base-T signaling would process more quickly and efficiently on channel Model B. Since there are so many legacy networks migrating to 100Base-T, this raises an important question: Can Model C be easily and economically modified to make it more efficient?
Quabbin Wire’s earlier actual measurement studies showed proof that channel RL performance is influenced by the quality of patch cords used in the channel (see www.quabbin.com/tech_briefs/premise.html). Changing cords is also a relatively economical “fix”, which the TIA’s TSB-95 document recommends you try if you experience network difficulties.
Changing cords in the channel models is also simple. Therefore, the better cords used in Model B were mathematically substituted into legacy Model C, making no change in Model C’s horizontal cable or connections. Doing so resulted in a much improved eye-pattern diagram. Figure 5 shows that Model C is now approximately equal to Model B. The Model C channel has improved and become more efficient when transmitting Fast Ethernet or 100Base-T signals. Due to the reduced RL noise and ISI in the channel, the active gear chip sets can more accurately process the transmitted signal.
THE LAST WORD
The modeling simulation proved that typical legacy channels with impedance mismatches and large amounts of RL noise perform much worse when transmitting higher frequency protocols. This is true even if they are transmitting unidirectional protocols such as 100Base-T or ATM. Their chip-sets struggle to find the distorted transmitted signal in the noise caused by ISI. Transmitted bi-directional protocols (such as 1000Base-T) are even more affected by RL noise. The signals are degraded twice: first by ISI and signal distortion, and second by echo noise at both ends of each pair.
The good news is that channel modeling has also shown that higher quality patch cords actually improve legacy or marginal networks. Better cords result in less channel RL noise, allowing the active gear to perform functions more quickly and efficiently. Most LANs worldwide are four pair UTP installations rated Category 5. Therefore, this relatively inexpensive and easy “fix” will enable hundreds of thousands of networks to efficiently migrate to protocols beyond 10Base-T. CS
Thomas R. Russell is VP of Technical Marketing at Quabbin Wire & Cable Co., Inc. in Ware, Massachusetts. Mr. Russell is a Mechanical Engineer who has worked for more than 30 years selling, marketing and developing low voltage electronic and instrumentation cable. He is an active participant in the TIA’s TR 42.7 Copper Cabling Systems Committee, helping to revise the TIA/EIA 568A document and develop Category 5e and 6 standards.