Ulrich's Plotter is good for this (use the overlapping ADEV algorithm - although TOTDEV and Theo1 are even better estimators of the Allan deviation) All going well, you will see a minimum in the Allan deviation versus tau plot. In most cases the value of tau at the minimum will be relatively close to the value of tau at the Allan intercept.

This estimate is consistent with the fact that LO parasitic radiation from TV systems on boats have been known to jam GPS for distances of several kilometers.

It's just a first estimation of what seems fair for me.

I guess I should add that this process is called Integration estimating with finite sums but I thought you would have already known that so just gave the abbreviated term.

Is your unknown signal likely to be quiet enough to estimate it's frequency this fast? Jitter, spurs, and noise are all directly related. The better the accuracy you are after, the lower jitter your unknown signal needs to be. Going crazy to get a 0.01 ps floor isn't useful if the signal has 100 ps cycle to cycle jitter on it.

I list a few values: 8E-8 @ 1 s 1E-9 @ 10 s 2E-10 @ 100 s 5E-11 @ 1000 s 2E-12 @ 1 d 3E-13 @ 10 d 7E-14 @ 100 d The diagram has no log sub scale so the readings are my estimate.

Most commercial users set their probability at around 30 nS, and experience virtually no estimates out of that boundary. The 30 nS error can be reduced to a better number if position is accurately known and the receiver "knows" that it is stationary - but still 5% of the time you can get noisy / degraded time and still be "in spec.".

A note on ensembles is that NTP actually features ensemble calculations, as it is able to estimate the noise, do weighting of various sources etc. Inspired by the work done at NIST. I'm not completely sure that NTP will work well with unlocked frequency sources, but I mention it so people can look in their NTP books and read up a bit.

I get as close as I can easily.....then allow the unknown to stabilise and do an estimate of the error and stick a label with thay value on the front until the next check is required eg "2 in 10^9 high 6thApr13" ......would not suit an avid "time-nut" but then the equipment is often barely "time-nut quality" :-)) Alan G3NYK . ----- Original Message ---

My goal is to estimate the stability of a rubidium at 0.1s,1s and 10s. I have a GPSDO, some picDiv and one Pulse Generator. I believe the OCXO in the GPSDO (a HP 58530A) is better in stability than the Rubidium for shorts taus even if it has not a clear view of the sky (it doesn't see 6 satellites all the time) so I locked the 53131A counter to the GPSDO.