[time-nuts] Mains frequency

[Magnus Danielson]

Tom,

On 11/17/2013 03:02 AM, Tom Van Baak wrote:
> Charles, et al.
>
> I think we agree. Just to clarify...
>
> I rely on no hardware and no software filters when I use a time-stamping counter such as a sub-nanosecond Pendulum CNT-9x or sub-microsecond picPET. An electrical zero-crossing happens when it happens. If you "filter" you're just trying to change history: spikes are spikes; noise is noise; history is history. Deal with it. Record it, don't filter it away.
>
> The beauty of the time-stamping method is that you capture any and all positive zero-crossings. If there is "noise" all it does is create unexpected and obvious artificial too-early or too-late samples -- which are trivial to analyze or eliminate in software.
>
> Some call them "outliers" and ignore them. This is correct. However, if one "filters" or "averages" them, you give validity they may not deserve. Bogus data should be eliminated by *logic*, not attenuated with pseudo-analog *filtering*.
>
> You can either focus on the signal, or the noise. That's two separate plots. An extraneous time-stamp happens to me a couple times a month; they are easy to spot and ignore. Similarly, a couple times a year I might miss a 60 Hz sample; these are also easy to spot and repair. For best time & frequency results, never "divide by =60"; instead "decimate by ~1 second".
Standard wide-bandwidth counters isn't really ideal for signals like this.

When you measure the mains signal, nominally 60 Hz in this case, spikes
etc. is noise which is local and not of interest when comparing over a
large area. Inter-area oscillations have much slower properties.
If you go the time-stamping way, you *should* remove such noise.

Removing or padding over it by logic will mean dropping data, which is
not helpful.

Turns out that professional gear for this does not do time-stamping in
this regard. Rather, they I-Q demodulate the signal with a reference
signal at the nominal rate, low-pass filter it and pay attention to
details of filtering like group-delay and compensation thereof. It's a
rather wise approach for the type of conditions you have.

Cheers,
Magnus