Say you need to communicate one digital signal from box A to box B 15.2m (50 ft) away. You choose a single-ended, 3.3V, 50Ω line driver running on RG-58 coax at 1000 Mbaud (1 nsec/bit), with a rise/fall time of 250 psec.
The response of this system appears in Figure 1. the figure shows the actual eye pattern, as simulation predicts, using solid lines. With a dashed line, it depicts the ideal transmitted waveform, assuming no skin-effect distortion or attenuation. The transmitted data pattern is "...1111010001111 ...."
The figure shows the worst-case high and low receiver thresholds for single-ended 3.3V JEDEC LVTTL at 2.0 and 0.8V, respectively. Notice that the low-side threshold fails to catch the first negative-going excursion, causing a bit error. Even when the LVTTL receiver does properly interpret the data, you can expect a fair amount of jitter in the received waveform.
If you instead select a differential receiver and a differential cabling system, the receiver thresholds more nearly center in the middle of the data pattern, because differential receivers are commonly specified with more accurate switching thresholds than ordinary single-ended logic.
For example, the chart shows the differential-receiver thresholds for LVDS logic. These thresholds still properly discriminate the data, even in the face of severe pulse distortion. In general, for the same amount of transmission-line distortion, a differential receiver generates less jitter than a single-ended receiver. The generally better threshold tolerances available in differential receivers, and not the differential architecture itself, provide this advantage.
In Figure 1's example, you could improve the single-ended system performance by using a differential receiver with its negative input terminal tied to a stable and accurate source of 1.65V. That simple change creates a single-ended receiver with much better control over the input threshold than the JEDEC 3.3V LVTTL thresholds.
It is best to tie the 50Ω termination at the end of the coaxial cable to the same 1.65V source you use for the receiver threshold. That makes the dc attenuation in the cable symmetrically affect both high and low logic levels, keeping the received signal perfectly centered.
EDN magazine takes this excerpt from the forthcoming Prentice Hall publication,High-Speed Signal Propagation: More Black Magic , by Howard Johnson, ISBN 013084408X, February 2003. Adapted by permission of Pearson Education Inc, Upper Saddle River, NJ.