[time-nuts] GPS clock to RT-Linux for pulsar machine
dforbes at dakotacom.net
Fri Jul 1 17:55:42 EDT 2005
Mike Nolan wrote:
> For every application I've used, we use ntp to choose the second, and use
> a sampler with a trigger input that gets 1 PPS ticks from the maser.
> At 100 us, you might get away with using the GPS 1PPS as a trigger,
> but even that would make me nervous. But I think you need a trigger on
> the sampler if you care about timing.
> Even that way we've had trouble: our datataking under some circumstances
> slew ntp by a huge amount (like 500 PPM). That can get it off by seconds
> before ntp can react.
Thanks for the info. I plan to store timestamps with all the data
samples, so any ntp drift can be spotted and accounted for by
comparing it to the stored 1PPS samples.
An alternative I've considered is to feed an NMEA stream from GPS into
the PC's serial port so that I am not relying on ntp for my timestamps.
I don't anticipate CPU interrupt overload trouble since the quantity
of data is so low, only ~20K words per second. At 80 GHz you don't
need to de-disperse the pulse, so one data value per polarization
channel is sufficient.
I may design the interface hardware to synchronize the ADC clock to
1PPS as you suggest, since it would increase the phase accuracy. It's
easy to do.
More information about the time-nuts