AVR-Chat

On Feb 19, 2009, at 7:26 AM, David VanHorn wrote:

> Soapbox on..
>
> The Feb 19th issue of EDN has an article that I found very  
> interesting.
> Many of you have heard me speak on crystal oscillators, and how the
> loading caps and drive levels need to be checked before committing to
> a design.
>
> This article references a train crash in Fremont CA, tracked down to a
> bad crystal oscillator on a control board.
> "An inappropriate value for the oscillators tank capacitor overdrove
> the crystal, causing the part to jump to an overtone frequency.
> As a result, the train sped up rather than slowed down as it
> approached a station, and the resultant crash caused many injuries."
>
> The article does not elaborate, but I'd guess that we'd be looking at
> something like a 22pF crystal with 22pF caps, and I'd bet that almost
> 100% of the oscillators on the boards they built "worked". ("It works"
> is not a viable defence of bad design.)
>
> http://www.eetasia.com/ARTICLES/2000SEP/2000SEP07_AMD_AN.PDF?SOURCES=DOWNLOAD
>
> "Crystal manufacturers specify a load capacitance number.
> This number is the load seen by the crystal which is the SERIES
> combination of C1 and C2, including all parasitics (PCB and holder)."
>
> http://www.foxonline.com/pdfs/osctheoryoper.pdf
>
> "For good, reliable circuit operation, it is recommended that the
> negative resistance be a minimum of five times the specified maximum
> resistance value of the crystal unit."
>
> Repeat after me: "The crystal determines the cap values, NOT the
> processor data sheet."
> "Check the drive level against the crystal
> data sheet."
> "Check the oscillation margin. 5x is acceptable"
> "Verify that the CKOPT fuse is set for
> rail-to-rail mode."
>
> Soapbox off..
>
> It also occurs to me that a second independent xtal osc, like a 32khz
> fork crystal, could have been used as a sanity check, measuring its
> period against the processors timer.
> You'd get some timer value per cycle, and if that was out of bounds,
> you could force a "fail-safe" condition. It would be extremely
> unlikely for both oscillators to fail in the same way, at the same
> time. For bonus points, use a series-mode crystal since your
> processor almost certainly uses parallel mode.
>
> -- 
>
> "The very powerful and the very stupid have one thing in common.  
> Instead of
> altering their views to fit the facts, they alter the facts to fit  
> their
> views... which can be very uncomfortable if you happen to be one of  
> the
> facts that needs altering." Doctor Who, Face of Evil
>
> 
Drive level is really important for 32KHz  crystals. If it is too  
high, frequency changes in one direction with supply voltage and if  
too low, changes the other. Results are not so obvious in a system  
with well regulated supply. But, if it is an unregulated battery  
powered system, it can cause a lot of grief.

I have also campaigned for proper design of crystal oscillators but  
have often gotten the push-back of "Well, won't it work with the  
values I have?" To which I answer "Well, Most of the time, probably.  
But I would not stake my life on it."

That train accident in Fremont is somewhat legendary. As I recall, it  
was supposed to be a "hi-rel" unit but had failure modes (the  
oscillator frequency) that nobody had considered. It seems to me that  
the Professional Engineer who signed off on it lost his job over it,  
and maybe some others. BART (Bay Area Rapid Transit) was grounded  
until the cause was fixed and it happened just after the system was  
started up. So, it really disrupted transportation throughout much of  
the San Francisco - Oakland area.

Jim Wagner
Oregon Research Electronics

[Non-text portions of this message have been removed]