[time-nuts] Commercial software defined radio for clock metrology
kb8tq at n1k.org
Sat Jul 30 11:21:32 EDT 2016
There are a *lot* of papers out there on downsampling and ADEV. Just about any / every technique known
has been tried and evaluated. The only “correct” answer is to throw away the samples (decimation). Anything
else you do will give you subtle (or not so subtle) issues. That said, there are standard filtering approaches that impact the first
point, but don’t do much past that. Sam Stein published a few papers on the why and the how of the conventional
approach (if you are going to filter). It has an impact, but at least you know what it is likely to do.
> On Jul 29, 2016, at 11:44 PM, Kevin Rosenberg <kevin at rosenberg.net> wrote:
> Hi Bob,
> You have a good point. That leads to the question is what is the “best” measurement technique when you are sampling at a more smaller interval than the desired tau?
> SDRs sample at high rates. The slowest the USRP N2x0 can sample is just under 200Ksps. For easy math, let’s assume we sample at 1Msps but we want to record only 1sps for a long-term measurement. What’s best way to handle the 1e6 to 1 ratio of available samples to desired samples? One method is to discard 999,999 samples and just record the phase difference with a true tau of 1 sec. The other is to take a window of 1e6 samples and output the average phase difference over that 1 second window. Is your point that averaging samples that are more frequent than the tau will overestimate stability at the tau? If using averaged data, would it be “less lying” to multiply the ADEV by the sqrt of the length of the averaging window?
> I’d appreciate your thoughts on the subject,
>> On Jul 29, 2016, at 6:51 PM, Bob Camp <kb8tq at n1k.org> wrote:
>> Keep in mind that if you apply pre-filtering, an ADEV plot is lying to you ….
>>> On Jul 29, 2016, at 6:58 PM, Kevin Rosenberg <kevin at rosenberg.net> wrote:
>>> Thanks for your very useful paper Oscillator Metrology with SDRs. I
>>> created a C++ program and checked residuals using a 10 MHz clock split
>>> to the A and B channels of a LFRX and BasicRX boards and sampled at 1
>>> Mhz. Using boxcar averaging of 1000 samples at 1 kHz, I was impressed
>>> by the low noise floor approaching that of my Timepod which was
>>> several times the cost. I included the Allan Deviation without
>>> averaging showing the sqrt(1000) increase in noise floor without the
>>> I had a question about your experience. You mentioned using a input
>>> signal near the maximum of the USRP’s ADC to get the best SNR. I
>>> reviewed the schematics and application notes. I found a maximum Vpp
>>> mentioned of 3.3V. I was wondering what voltage you were using to
>>> drive the USRPs. When I go above 1.5-2 Vpp, I start getting signal
>>> distortions and not much increase in the amplitude.
>>> Many thanks for publishing your work in this area.
>>> time-nuts mailing list -- time-nuts at febo.com
>>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>> and follow the instructions there.
>> time-nuts mailing list -- time-nuts at febo.com
>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>> and follow the instructions there.
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
More information about the time-nuts