[time-nuts] Weird TEC data

Scott Newell newell at cei.net
Sun Aug 7 00:51:55 UTC 2011


At 07:13 PM 8/6/2011, Hal Murray wrote:

>newell at cei.net said:
> > The general shape and bumps in the plots track nicely, but I'm  wondering
> > why there's so many cycles difference after 36 hours.
>
>How are you collecting the data?  What's the time between samples?

First off, your TEC data was an inspiration.  I thank you sir.


The embedded box sends a string with every sample, so 60Hz.  The 
computer timestamps the arrival and logs the time, count, and 
frequency data to a file.  (I'm actually generating two files now, 
one with every sample, one downsampled 60:1.  Other neato stuff too: 
sync start, auto file rotation at top of day, etc.)


>One possibility is that one system is picking up extra clock ticks.  If your
>data is dense enough, that should be pretty obvious if you zoom in on the

I think that's what happened to the setup at location "tock".  Friday 
I threw iron at the situation--a consumer grade UPS in front of a 
Sola line conditioner.  Both sites got new embedded firmware with the 
new detection scheme.  (Wait for edge, start 15.something ms hardware 
timer, send data to computer, wait for timer expiration, 
repeat).  That should reject a lot of noise.


>Glitches like that are easy to spot if you plot the frequency.  It's the
>difference in counts divided by the difference in times between a pair of
>samples.  At 10 second sampling rate, you get 600 counts per sample.  601
>counts turns into 60.1 Hz.  That 0.1 Hz is well above the noise.  (at least
>with my setup)

I'm sorta doing this now.  The logging program has a FIFO for the 
count and timestamp, so I'm calculating the frequency over the last 
1s, 60s, 600s, and 3600s intervals.  Every minute or so the frequency 
data is written to a file for Munin to plot (so that I have a live 
picture to look at).  But yes, the data is in the file, so I can also 
gnuplot it.


55779 data is downloading now, so I'll put some graphs up in an hour 
or so.  I can already tell from the Munin plots over the last 24 
hours that the two sites are tracking very closely.  (Right now, tock 
shows -25.13 cycles at 00:35, and sparc is at -24.87.)


There's another PC at site "sparc" doing the webcam thing.  Still 
using the Windows XP powertoy to timelapse, so the intervals aren't 
exactly 60s.  If I can get a few hundred cycles of error, it should 
be obvious from the picture .  (There's an old 60 Hz clock and a WWVB 
clock in the frame, so the measurement is limited by the flashing 
color, maybe +/- 0.5s or so.)


Both site's computers have Debian supported audio hardware, so it's 
tempting to add that into the mix...


-- 
newell  N5TNL 





More information about the Time-nuts_lists.febo.com mailing list