I would like to raise another small question regarding AC termination: Surely there are specific requirements on the RC time constant of the termination circuit. Should it be "greater than... (what) or smaller than (what)"? Why?
Could AC termination be combined in Thevenin termination, i.e., pull-up resistor to VCC and series R-C to GND?
Thanks for your interest in High-Speed Digital Design.
The promise of an AC terminator is the idea that maybe, just maybe, there is a value of C big enough to make a good termination, but at the same time small enough to not draw much current from the source.
For a truly long line, one where the ratio of (line delay)/(signal risetime) is greater than 1, there is no overlap between these two criteria. FOR RANDOM DATA, by the time you have made C big enough to act as a good termination, you have also increased the peak I(OH) and I(OL) levels required of the driver to as much as twice what would have been required had you used a split-end terminator (or a series terminator, which requires the same PEAK I(OH) and I(OL) as an end terminator).
If your driver cannot source enough current, then during the initial line roundtrip time (at least two risetimes) PLUS the capacitor charging time, the signal will not meet V(OH) (or V(OL)).
Don't combine the AC-terminator with RANDOM DATA and a truly long line.
If the ratio of (line delay)/(signal risetime) is more moderate, say only about 1/3 to 1/2, then you will see the AC terminator provide some useful benefits. In this circumstance it will act to somewhat reduce ringing behavior (although it never completely eliminates it) at the expense of a somewhat increased PEAK driver current. Messing with the resistor value hardly helps. Using R(AC) = Z0 is the best choice.
If the line carries data that is DC balanced (equal numbers of ones and zeros, like a clock, or a Manchester-coded data signal), then the situation changes. With DC-balanced data you are free to pick any capacitor BIG ENOUGH to hold its charge throughout the bit period. For example, on a fast clock line I might use a .1 uF bypass capacitor. Such a capacitor will charge up to precisely Vcc/2 and stay there. The termination therefore presents a Thevenin equivalent circuit that looks like Z0 ohms going to a battery of Vcc/2 volts. That is a nice end terminator. As far as the signaling is concerned, it works just like a split terminator with 2*Z0 going to Vcc and 2*Z0 going to ground. As far as power dissipation is concerned, it absorbs only half the power of the split termination. That's one reason people like it (for clocks).
Dr. Howard Johnson