I believe I finally have my frequency measurement process refined and have described it here: https://drive.google.com/open?id=1luVumTygkvfnDsZvGZGSsJA75IwMKmJ8 Comments, corrections and criticism are welcome. Regards, Mark W7MLG

Hi Mark, On 2019-01-09 08:27, Mark Goldberg wrote: > I believe I finally have my frequency measurement process refined and have > described it here: > > https://drive.google.com/open?id=1luVumTygkvfnDsZvGZGSsJA75IwMKmJ8 > > Comments, corrections and criticism are welcome. I have not seen anyone comment on this post yet. It's an interesting approach and in general, this is one of the ways we can expect that frequency/stability measurements to be done these days, by sampling the RF and analyze it. It would be interesting to figure out why you have a "bump" there, so I wonder what part does that. For one thing, if the frequency estimation algorithm you depend on does averaging/least-square style of algorithms, the lower taus will be significantly lower than expected. TvB has some pretty good plots from experiments illustrating this. Another thing, one reason to get a "bump" is due to high Q in a PLL circuit. However, I'd be careful to judge it to be any of these without more careful look on the data. It can be useful to alternate views to figure things out. For instance, swapping between MDEV and TDEV could give you some hints. Similarly swapping between frequency and phase does the same in raw-data view. For phase/frequency plots, it may be worthwhile to average the data using the +/- keys in TimeLab, as the filtering away high-frequency noise may make it easier to see lower-frequency variations. In general, I recommend you to have a third reference to play around with and measure. As one fools around with different combinations one learns to see which artefacts follows which device. Do keep up the investigations. Try different approaches and learn. Cheers, Magnus

On Sun, Jan 20, 2019 at 5:10 PM Magnus Danielson <magnus@rubidium.se> wrote: > Hi Mark, > > I have not seen anyone comment on this post yet. > > It's an interesting approach and in general, this is one of the ways we > can expect that frequency/stability measurements to be done these days, > by sampling the RF and analyze it. > > It would be interesting to figure out why you have a "bump" there, so I > wonder what part does that. > The bump in the later plot, with the Bodnar measured against the Trueposition is almost certainly due to a PLL. There is the Bodnar and also the Wenzel and I don't know exactly what each is doing and what all the time constants are. I have no clue why there is a bump on the first wraparound plots at about Tau=1000s. > > For one thing, if the frequency estimation algorithm you depend on does > averaging/least-square style of algorithms, the lower taus will be > significantly lower than expected. TvB has some pretty good plots from > experiments illustrating this. > > In my various experiments I could see when it was averaging and the lower Taus did decrease. I chose such that this did not happen. I reviewed TvB's paper at http://www.leapsecond.com/pages/adev-avg/ many times and saw most of the effects described there, resolution, noise and averaging. > Another thing, one reason to get a "bump" is due to high Q in a PLL > circuit. > > However, I'd be careful to judge it to be any of these without more > careful look on the data. > > It can be useful to alternate views to figure things out. For instance, > swapping between MDEV and TDEV could give you some hints. Similarly > swapping between frequency and phase does the same in raw-data view. > > For phase/frequency plots, it may be worthwhile to average the data > using the +/- keys in TimeLab, as the filtering away high-frequency > noise may make it easier to see lower-frequency variations. > I've done this ad nauseam, eventually finding some anomaly that happened at various times of the day. > > In general, I recommend you to have a third reference to play around > with and measure. As one fools around with different combinations one > learns to see which artefacts follows which device. > I'm pretty sure the Bodnar is the least stable device I have, as it is based on a TCXO and it does have some issues that I know Leo has improved, but I can't spare the time to send it back to him right now. I don't have another device to use that will give me the 80 MHz I need for the SDR clock that also gives me good phase noise. Both the Trueposition and Bodnar/Wenzel are pretty good, at least better than what I am measuring. > > Do keep up the investigations. Try different approaches and learn. > > At this time, I am going to use what I have. I think it is the best approach I could come up with, using the available equipment. Regards, Mark