[time-nuts] Syncing Computer to Datum 9700 Programmable Time System

[Bruce Griffiths]

Hal Murray wrote:
>> I noticed all the wonderful outputs on the back, and I'd like to take
>> the 1PPS signal and use it as a timing standard for my IBM workstation
>> running XP Pro SP2, thereby making it the home time server for the
>> rest of our computer network.  I'm assuming that it would have to go
>> into one of the available serial ports, but is there an off-the-shelf
>> cable w/software to enable this, or am I heading into uncharted
>> territory? 
>>     
>
> On the hardware...
>
> The usual approach is to feed the PPS signal into one of the modem control 
> signals.  (I think it's usually DCD, but be sure to double check that.)
>
>   
DCD is the usual choice.
> Some people claim you need a TTL to RS-232 level shifter.  A raw TTL signal 
> works just fine with almost all RS-232 input chips, including whatever is 
> used on all the systems I've worked with.
>   
If you want good noise margins and to avoid the ringing that can occur
when driving
high capacitance cables of indeterminate length with TTL, a TTL to RS232
level translator is desirable.
> The normal level shifter includes an inverter so the two approaches end up 
> with the opposite polarity.  (Most software has an option to select 
> rising/falling edge.)
>
> I don't know of any off-the-shelf cables, but it's pretty simple to make your 
> own.  (at least if you use the no-level-shifter approach and have a soldering 
> iron)
>
>
> On the software...
>
> I don't know much about Windows.
>
> There is a version of the reference implementation of ntpd compiled for 
> Windows and packaged with an installer.  I don't know if that includes any of 
> the refclock drivers which you probably need to use the PPS signal.  
> (probably not)
>   http://www.meinberg.de/english/sw/ntp.htm
>
> To get great time, you need some extra code in the kernel.  The key idea is 
> that the interrupt routine grabs the time when DCD changes.  User code can 
> ask for that time and use it for updating drift and offset.  The net result 
> is much more accurate (less jitter) than if the user code asked for the time 
> after it got woken up via an interrupt and such.
>
> A PPS signal doesn't tell you which second the pulse corresponds to.  So you 
> need something else for that.  Most boxes that generate PPS signals also 
> generate an ASCII string with the date/time in it.  If not, you can probably 
> get it from NTP servers on the net.
>
>
> So I wouldn't call it "uncharted", but it doesn't sound like mature 
> technology.
>   
Performance with windows is so poor that its hardly worth the effort
except as a learning exercise.

Bruce