Re: [time-nuts] Question about effect of spurious frequency modulation on Allan Deviation
I am thinking that I should simply make a very simple notch filter that suppresses the spurious line by 10 or 20 dB. I can then do an A/B comparison of the measured ADEV with and without the filter. If there is no change, I'm done. Otherwise, I add filtering in steps until the ADEV reaches an asymptote. Rick On 8/6/2018 6:39 PM, John Miles wrote: Again, this is analogous to the 5071A having 1 MHz spurs of about -90 dBc, yet the ADEV is not much different than an open loop 10811. How high would the 1 MHz spurs have to be to affect ADEV? With ADEV, you have to consider the measurement bandwidth before you can ask questions like this, much less answer them. :) 1 MHz spurs are far outside the restricted measurement bandwidth that would be desirable when measuring a high-performance source like a 5071A. I suppose there might be some influence if you use a DC-to-daylight sampling process such as a counter, but the instrument noise floor will probably obscure any interferers in the -90 dBc range. With a TimePod or similar device, a general rule of thumb is that a spur a few Hz away from a 10 MHz carrier at -120 dBc will cause ripple in the 1E-11 to 1E-12 range. This coincides with typical levels of coupling between nearby RG-58 cables -- see page 38 of http://www.miles.io/TimePod_5330A_user_manual.pdf for instance. That particular situation wouldn't have shown up on a traditional counter-based measurement, but others might. For instance, in your 1 MHz example, I doubt you'd see any effect at exactly 1 MHz, but if your spurs are at 1.01 MHz and your counter is capable of 1-ns single-shot resolution, I can imagine that there will be some corruption. The exact numbers will depend on so many factors that it's almost easier to set up a test and observe the effects firsthand. -- john, KE5FX Miles Design LLC ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] Question about effect of spurious frequency modulation on Allan Deviation
Surely all that's required is a simplistic worst case analysis. Just assume that the value of Tau is always the worst possible so the full effect of the modulation is always seen. If the worst case phase modulation is say phi then the worst case ADEV (Tau) will be proportional to phi*T/Tau where T is the nominal signal period. i.e. ADEV(Tau) < constant*phi*T/Tau Bruce > On 07 August 2018 at 10:37 "Richard (Rick) Karlquist" > wrote: > > > The discussion has gotten off track. > I probably didn't make myself clear. > > When I talked about cleaning up the signal, I meant > exactly that, cleaning up the signal using a filter. > As opposed to cleaning up the measurement after the > fact. I just need to know how well I need to filter > the signal to meet a certain ADEV measurement level. > > The FM I have is not "slow". The rate > is in the MHz range. Can Stable 32 simulate that? > > Again, this is analogous to the 5071A having 1 MHz > spurs of about -90 dBc, yet the ADEV is not much > different than an open loop 10811. How high would > the 1 MHz spurs have to be to affect ADEV? > > Thanks. > > Rick > > On 8/6/2018 1:56 PM, Tom Van Baak wrote: > >> If you know the *source* of the bias you're trying to remove (i.e. you > >> know it's a sinusoidal frequency modulation), I don't know that it's any > >> different than removing long term drift. > > > > The TimeLab 'n' command (apply notch filter to phase records) is > > specifically for this purpose. JohnM added it when he ran into an H-maser > > which suffered from some sort of consistent periodic modulation. It spoiled > > the ADEV plots, but it did so in a deterministic manner. In a case like > > this you either debug the root cause of the h/w problem and make the > > repair, or just "repair it" in s/w. Like you say, it's the same concept > > (and danger) as removing linear drift or other deterministic / model-able > > effects. > > > > To explore this for yourself, you can use Stable32 to generate synthetic > > phase/frequency data with your choice of noise and modulation. Then use > > TimeLab to view the raw data and to apply the notch filter. > > > > I've written up some quick examples here: www.leapsecond.com/pages/adev-fm/ > > > > The plots dramatically show what effect slow FM can have on an ADEV plot. > > It also shows how well the TimeLab notch filter works. If you don't have > > time to look at that page, I've attached one plot to whet your appetite. > > > > /tvb > > > > > > > > ___ > > time-nuts mailing list -- time-nuts@lists.febo.com > > To unsubscribe, go to > > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > > and follow the instructions there. > > > > ___ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] Question about effect of spurious frequency modulation on Allan Deviation
One might also do some trials based on comparing ADEV results between a clean carrier signal and ones corrupted with varying degrees of FM (for example), to get a feel for the problem. If nothing else, one ought to be able to get some feel for the sensitivities involved. Dana K8YUM On Mon, Aug 6, 2018 at 8:17 AM, jimlux wrote: > On 8/5/18 11:22 PM, Poul-Henning Kamp wrote: > >> >> In message <84a802ff-88f1-5f50-1f79-71d8ba3c4...@rubidium.dyndns.org>, >> Magnus D >> anielson writes: >> >> What does exists is a formula for how a single sine spur would produce >>> ADEV. A FM deviation with low enough modulation index creates two >>> side-bands of opposite sign but same amplitude. >>> >> >> I find the easiest way to wrap my head around this is to think >> about measuring Adev by timing zero-crossings. >> > > >> Depending on the modulation signal, there may be moments where the >> zero-crossing is "where it should be", for instance if the modulation >> is sine or triangular, but not if it is a signed square wave. >> >> > > > What about doing some sort of fit to the measurement data before > calculating the ADEV? Similar to removing a linear ramp. > > basically you'd solve for the three sine parameters (f, phase, > amp/deviation), then remove that from the data, then run the ADEV > calculation. > > > > ___ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to http://lists.febo.com/mailman/ > listinfo/time-nuts_lists.febo.com > and follow the instructions there. > ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] Question about effect of spurious frequency modulation on Allan Deviation
On 8/5/18 11:22 PM, Poul-Henning Kamp wrote: In message <84a802ff-88f1-5f50-1f79-71d8ba3c4...@rubidium.dyndns.org>, Magnus D anielson writes: What does exists is a formula for how a single sine spur would produce ADEV. A FM deviation with low enough modulation index creates two side-bands of opposite sign but same amplitude. I find the easiest way to wrap my head around this is to think about measuring Adev by timing zero-crossings. Depending on the modulation signal, there may be moments where the zero-crossing is "where it should be", for instance if the modulation is sine or triangular, but not if it is a signed square wave. What about doing some sort of fit to the measurement data before calculating the ADEV? Similar to removing a linear ramp. basically you'd solve for the three sine parameters (f, phase, amp/deviation), then remove that from the data, then run the ADEV calculation. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] Question about effect of spurious frequency modulation on Allan Deviation
In message <84a802ff-88f1-5f50-1f79-71d8ba3c4...@rubidium.dyndns.org>, Magnus D anielson writes: >What does exists is a formula for how a single sine spur would produce >ADEV. A FM deviation with low enough modulation index creates two >side-bands of opposite sign but same amplitude. I find the easiest way to wrap my head around this is to think about measuring Adev by timing zero-crossings. If we had a perfect zero-crossing detector, perfect AM would not matter, because perfect AM does not move the zero-crossings. But in theory at least one electron, and in practice many, must run the opposite direction before we can detect the zero-crossing, so even perfect AM matters in practice. (Interesting detail: Measuring at high impedance may be smarter than at 50 Ohm impedance) For FM or PM there is no loophole: They move the zero-crossings and that's that. Depending on the modulation signal, there may be moments where the zero-crossing is "where it should be", for instance if the modulation is sine or triangular, but not if it is a signed square wave. -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 p...@freebsd.org | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] Question about effect of spurious frequency modulation on Allan Deviation
Hi Rick, On 08/06/2018 07:12 AM, Richard (Rick) Karlquist wrote: > I need to measure ADEV on a source that has spurious > sine wave frequency modulation on it. I am looking for a > formula that would tell me ADEV, vs FM deviation and > carrier frequency. I'm not sure if modulation frequency > or tau matter. I am hoping to determine how much > I need to clean up the signal order to get down to a particular > ADEV measurement threshold. > > I tried a literature search but didn't turn up anything > I could use. > > I remember when I worked on the output section of the 5071A > we specified 100 kHz and 1 MHz spurs to be down -80 dBc. > These didn't seem to affect the ADEV of the 10 MHz outputs, > but I can't prove why this should be the case. Does the > f-sub-h measurement bandwidth come into play? > > Thanks. What does exists is a formula for how a single sine spur would produce ADEV. A FM deviation with low enough modulation index creates two side-bands of opposite sign but same amplitude. The modulation index, which is just another aspect of frequency deviation, is direct steering the amplitude of these sidebands through Bessel polynomial, but for low deviations this is linear. The modulation frequency will care, and it will depend on tau. The actual sample rate of the ADEV will interact with the sideband "spurs" and add onto each other. The additional ADEV comes on top of any other ADEV response, so it is clearly a factor on ADEV polution. Without doing the math, expect the double amplitude to that of the single sideband sine of the same amplitude itself. Cheers, Magnus ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.