[time-nuts] Low cost synchronization

David Forbes dforbes at dakotacom.net
Sat Aug 20 17:30:41 EDT 2005

At 4:07 PM -0500 8/20/05, Mike Ciholas wrote:
>One wonders if you can build some sort of long term reception
>processing that would pick out the signal from the noise.  Since
>you know what you *should* be getting, you can overlay multiple
>minutes of reception to cancel out the noise.  I wonder how much
>processing that will take.  Would it be possible to recover
>enough signal fro the noise to make VLF receivable worldwide?


This is what I do for a living - integrating radio signals over long 
time periods to get a signal from what looks like pure noise.

The basic formula is that the S/N increases by the square root of the 
integration time, as long as the sample phase remains correct. This 
is the limiting factor - the PPM of the fob's timebase determines 
this and will be another watch crystal which is good to maybe 1PPM.

There are two ways to proceed: One is to try to actually learn the 
time of day form WWVB, and the other is to just keep your local clock 
synchronized to the one second pulses from WWVB.

The standard time signals such as WWVB use a one-bit-per-second 
coding scheme where the signal is amplitude modulated, so you'd need 
to measure the phase of this modulation to perhaps 5% of a cycle, or 
0.05 seconds. Your 1PPM oscillator will be off by .05 seconds in 
50,000 seconds or 14 hours. So you could integrate data for 14 hours. 
That should give you sqrt(50,000) improved S/N, or 200 times better 
S/N ratio (23dB).

If you just want to maintain your oscillator's 1PPS phase against 
WWVB, then you just need to do a pulsar synchronization technique of 
splitting a second into perhaps 20 bins and accumulating the signal 
strength during each bin. You are looking for the rising edge of the 
carrier which indicates the top of the second. That is, a bin with 
very little signal followed by one with lots of signal is when the 
pulse starts.

If you want to learn the time, then the next problem is that you're 
integrating data which is changing, so you have to deal with that by 
some fancy integration tricks such as using separate bins for 1 and 0 
level bits.

I hope this makes some sense. If not, then read up on pulsar 
detection and timing techniques.

--David Forbes, Tucson, AZ

More information about the time-nuts mailing list