Open Drain Lines

I have a question to ask: in my current design I have multiple long transmission lines, each driven on both ends by an open drain driver with an external pull-up resistor. Each driver can drive the line active low (then the current flows through the pull-up and the driver) or completely turn off (hi-z), so that the line is only pulled up by the resistor. Also each driver can switch into an input configuration. Finally, both ends can drive the line low at the same time, or both turn off, or both become inputs, plus all the combinations there of. The frequency is 10 MHZ, and the allowable driver current is 8 mA.

I have a choice between using one pull-up resistor per line, located somewhere in the middle between the two ends, or two double-valued resistors at each end. I think the second option is better for line termination reasons, but I cannot explain why.

Michael Korkin

Thanks for your interest in High-Speed Digital Design.

If your drivers can pump out enough current, the best strategy, from a signal integrity perspective, is to place a pull-up at each end of the line. The pull-up impedance should match the characteristic impedance of your line.

This configuration is called a double-ended bus termination.

The advantage of this arrangement is simple to understand: regardless of which end is driving the line, the other end will always have a good, solid termination. You'll get very little ringing, except for one peculiar case.

The peculiar case is this: If both drivers simultaneously go low, you get a very LOW impedance at both ends of the line, which is a ringy-dingy situation. Fortunately, in this case, with both drivers switched LO, neither side is looking at the line conditions (they are both, in effect, blinded while driving), so who cares if the line rattles around for a while. Also, if you draw out a complete time-space diagram for the system you will note that while the ringing is apparent near the center of the transmission line, there is almost none at the ends (where the low-impedance drivers are located). If yours is a truly point-to-point configuration, with no receiver taps strung along the middle of the line, then this ringing will be of no consequence.

If your drivers can't put out enough current to drive two terminators at once, then we have to consider various "second- best" strategies. None of these second-best ideas will be perfect. They will all involve some amount of ringing, delay, non-monotonic response, or slow risetimes.

Here are some things that help

  • Slower drivers always help. You're only going 10 MHz, so you can afford to use very low drivers.
  • Try a small resistance in series with the transceiver (like 39 ohms), and a VERY light pull-up (like 1K). You'll get a fast, crisp pull-down, and a v-e-e-e-r-r-r-y-y-y slow pull-up. At least it's monotonic.
  • The higher the impedance of the connecting trace, the larger you can make your termination resistors, and the easier the whole structure will be to drive. Use as skinny a trace as you can fabricate.
  • Some termination, any termination, at each end of the line is better than nothing. If you can't drive two 50-ohm terminations, try using two 100-ohms terminations.

Things that don't work

If you must use the wrong value of termination, putting a single value in the middle of the line DOES NOT work as well as splitting it up at both ends. With a single termination in the center, there is a mode of operation that remains completely undamped. To see this mode, think of a child's see- saw. It's pinned in the center, but the two ends can wave up and down opposite each other. That's exactly what happens with the center terminator. If one end has been driving for some time, and then switches OFF, you end up with a wave pattern on the wire that oscillates for a long time, dying down very slowly. It's better to take the center resistor, divide it into two parallel resistors of twice the value, and move one to each end.

Conclusion

If you can really drive two terminations, the double-ended structure gives you ideal performance, with no overshoot or undershoot, and very--very fast, accurate rising and falling edges.

Best Regards,
Dr. Howard Johnson