I need to extend a memory-mapped bus into another enclosure and thought that a bidirectional LVDS implementation with serial/ deserializer pairs at each end might work. Does anyone have any experience or guidance on such a setup?
LVDS has a really low common mode voltage tolerance, so if you use it between enclosures, make very sure you have an excellent grounding scheme and control of return currents between enclosures. Generally not a good application of LVDS without some means of improving CMV tolerance.
Erm... one (single-ended) application of LVDS I know of shoves a 25MHz clock down some ribbon cable... 10m of it. Seems to work. LVDS, being a current loop, tolerates this kind of (ab)use better than most.
To the OP re bidirectional: it's not clear to me whether you expect to run two loops (one outgoing, one incoming), or shove data down one pair of wires in boith directions. If the latter, I'm not clear on where you'd put the receiver load - at both ends? If so, expect to see half the voltage across each receiver.
I was contemplating the former as shown in National's application (Fig
2.4):
formatting link
It also has the precaution about lower voltages, but it predicts a 10m limit (my app is much shorter) and cautions against a noisy environment (one enclosure includes an rf transmitter).
Further to earlier response: re-reading this carefully, I'd agree if we were talking about normal (TTL-esque) digital signals. But with LVDS, it
*is* a differential current loop. A twisted pair is ideal, otherwise I'd put the two conductors side by side in a ribbon with a guard ground either side.
If you're going to use SERDES on the LVDS lines, I would assume your bit rate is fairly high. In this case I would advise against the bidirectional wire idea. Save yourself the headaches and use one set of wires for each direction. Certainly it is possible to use the circuit of figure 2.4, but you won't be able to change direction quickly, in addition to the reduced signal strength mentioned. The point of SERDES is to reduce the number of pairs significantly, so it shouldn't cost too much to assign the additional pairs to run independent one-way data paths.
How about a hybrid solution? I don't want to run so many wires (I don't have room for such a big connector). Could I gang up 4 signals per pair without a lot of overhead, and still keep it bidirectional?
Most of the LVDS receivers that we've tested, including the Xilinx Spartan3 parts, seem to behave like excellent, very fast rail-to-rail comparators, at a fraction of the price of an officialy-designated comparator of similar performance. They do generally have a deliberate input offset, in the 10's of millivolts sort of range. Most of them make good zero-crossing detectors, signal on one input pin and ground on the other.
Still, one wouldn't want to drive any input hard much beyond its local rails.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.