[time-nuts] 60 Hz power glitch, US West coast (Silicon Valley)

[Hal Murray]

Tom Harris <celephicus at gmail.com> said:
> I'm just surprised that you get such results with a cheap transformer. 

I think you are missing the decimal point.  (Or maybe I'm not being nutty 
enough.)

I'm counting cycles and grabbing the time with a PC running Linux.

I haven't carefully looked at the error budget.  60 Hz is 16.6 ms per cycle.  
I'm pretty sure the clock on that PC is good to a fraction of a ms.  It's not 
good to the microsecond level.  As tvb said, the "time" from the power grid 
wanders over seconds during the day.  That's hundreds of cycles, so a tiny 
fraction of a cycle is lost in the noise.

There is also noise on the PPS capture path - interrupts disabled and cache 
loads and such.  I don't have good numbers for the 60 Hz case, but should 
have some data on the 1 PPS path as used by ntpd.  I'll see if I can come up 
with an interesting graph.


One thing that is interesting.  If the measurement system misses a pulse 
(interrupts disabled for too long or whatever), there is an obvious big 
glitch in the frequency.  A missing pulse turns 60 Hz into 59.9 Hz.  It's 
rare that I see anything that far off.   As you can see from the graph, the 
frequency usually doesn't change much on the scale of 10s of seconds, but 
there is a lot of noise at the 0.01 Hz level.  I don't know if that's from 
the capture system or in the raw data.

Similarly, if noise on the hardware turns into an extra pulse, that stands 
out in the other direction.  So I'm pretty sure the pulse count is correct.


-- 
These are my opinions.  I hate spam.