[time-nuts] have 10MHz need 19.5Mhz

[Jim Lux]

On 6/2/13 11:59 AM, Chris Albertson wrote:
> Recent talk about NTP servers.  It seems the limit to their accuracy is the
> quality of the crystal that drives the CPU clock.  Most of them make really
> good thermometers.  I'd like to try and replace the crystal on a Raspberry
> Pi with a signal derived from a time nut quality 10MHz standard.
>
> The Pi uses a crystal (not a TTL can, a real two lead crystal and a pair of
> 47pf caps) Both leads of the crystal attach to a pair of pins on an IC.   I
> figure I can unsolder the crystal and inject a balanced 19.5MHz signal
> directly to the IC's pins.

One pin is an input, the other an output most likely. The usual scheme 
has an amplifier on chip to make an oscillator with the crystal between 
in and out.

Drive just the input.

(if you can't find out from the data sheet, you can just try em both to 
see which one works. or, heck, drive both through a suitable resistor. 
The amplifier output won't care)



    I know the ARM CPU just might work on a 20MHz
> clock or maybe 15MHz but the video likely would not.  I'm going to have to
> supply 19.5MHz.
>
> The question is the best way to get from 10MHz to 19.5MHz.  I care only
> about long term (tens of seconds to days) stability.
>
DDS eval board is your friend.  Any of the analog devices parts would 
probably work. the older 985x ones are cheap, but draw a lot of power.