[time-nuts] Termination talk

Magnus Danielson magnus at rubidium.dyndns.org
Fri Dec 7 22:42:44 UTC 2012


On 11/28/2012 11:52 PM, M. Simon wrote:
> I was going to post an anecdote about termination to the list and then thought that the piece would make a great column. It did. My magazine featured it in one of its daily mailings.
>
> If you want to check it out: http://www.ecnmag.com/blogs/2012/11/long-lines-pcb

... and again rise-time is your enemy, unless you know what you are doing.

Another war-story was in the early days of the company, when we made the 
first boards and had issues on the 1,0625 GBd serial links giving 
significant bit-errors. They where fighting the issue and it dragged 
out. One morning I came to the office, took my morning coffee and 
strolled into the lab, picked up a board and just looked at it, just for 
fun. I was following the traces and looked at it, and then suddenly I 
realized that someone had located the terminating resistors of the PECL 
pair at the wrong end of the transmission line. The designer didn't 
think of it as a problem, as the 50 ohm load to ground was there, so 
what is the problem? I chose not to take the fight, but I did leave him 
with the notion that I was not convinced by his argument.
He kept fighting the issue for several days, until he out of desperation 
tried my advice, and out of his big surprise it did work. Only then he 
made the comment that there might be something in that "Black Magic" 
book that I had bought and distributed.

It can be OK to run without or with the wrong termination, if you know 
the situation will remain stable and you verify that it behaves the way 
you expect it to do, but if not, do your terminations seriously, 
consider both source and destination termination and reflections.

And *do* recall that a frequency might require a certain 
slew-rate/rise-time, but the rise-time triggers the reflections and 
couplings every time it occurs, regardless of the rate. So, design for 
the rise/fall-time you have, and not the frequency.

Cheers,
Magnus



More information about the time-nuts mailing list