[time-nuts] GPS clock to RT-Linux for pulsar machine

[David Forbes]

Matt Ettus wrote:

>>Don't do it that way, you will get in trouble with the interrupt
>>latency.
> 
> 
> I have to echo this advice.  While RTLinux and other RTOS'es have
> great interrupt latency performance, using those interrupts to trigger
> sampling on an ADC is a bad idea.  It won't work at all if you are
> trying to sample faster than about 1-10 kHz, and the jitter is
> horrible even below that.  Even a great RTOS will only get you about
> 10us repeatability.  It is always best to use a real oscillator to
> trigger sampling.  For some reason, interrupt-based sampling seems
> very popular in the radio astronomy world.
> 
> Matt

Matt,

Thanks for the good advice, but I said that I was planning on using 
the maser for the ADC sampling clock, and the interrupt only for 
getting 1PPS into the computer.

After pondering all the quick replies, I think I'll just run the 1PPS 
signal straight into an analog channel on the ADC card. All the $300 
ADC cards have a 16-input mux so it's cheap to do that. It's not quite 
as memory-efficient as using a digital input, but it's quick and 
effective. I can discard the unused bits when I write to disk.

It's funny how the typical middle-aged engineer brain wants a one-bit 
digital input for a digital signal, but these days wasting 15 bits of 
analog data for one digital bit just isn't a problem.

As for radio astronomers and interrupts, my group is pretty good about 
using the appropriate method to get data into the computer. I do shake 
my head in wonder occasionally when I read how things used to be done, 
though.