[time-nuts] ADEV vs MDEV - using sound card

Bob Camp lists at cq.nu
Sun Feb 7 01:19:39 UTC 2010


Hi

Take the outputs of the ADC's, bandpass filter them, do a straight line fit to the 100 or so points nearest each zero crossing. Report the result to the PC for each zero crossing. Not much data to the PC. Easy to do it all in a fairly small FPGA.

Staying down at low frequencies opens up the range of offset oscillators available. The higher you go, the tougher it will be to select a good reference. 

Bob


On Feb 6, 2010, at 7:57 PM, Bruce Griffiths wrote:

> Bob Camp wrote:
>> Hi
>> 
>> My main concern with the low frequency pole in the sound card is the quality of the R/C used. You can certainly model what ever you have. If they used an aluminum electrolytic for the "C" it may not be the same next time you check it ....
>> 
>>   
> One should at least calibrate the effect before and after each run.
> It would be even better to embed such a calibration within each run.
>> On a 10 Hz system, a 1 Hz pole is probably not an issue. It might get in the way with a 1 Hz beat note.
>> 
>> Another thing I have only seen in passing: "Sigma Delta's have poor low frequency noise characteristics". I haven't dug into it to see if that's really true or not. If you buy your own ADC's, you certainly would not be restricted to a Sigma Delta.
>> 
>>   
> The input noise spectrum for my sound card does indeed rise significantly at low frequencies.
> A beat note of around 100Hz or 1KHz note would be more suitable than lower frequencies.
> With a Costas receiver the equivalent timing noise isn't strongly dependent on the beat frequency.
>> Even with a cheap pre-built FPGA board, you could look into higher sample rates than a conventional sound card. You would drop back to 16 bits, but it might be worth it.
>> 
>> Bob
>> 
>> 
>>   
> 1 bit of a 16 bit accurate ADC corresponds to a timing resolution of around 0.48ps referred to a 10MHz mixer input signal.
> The beat frequency amplitude has to be amplified to just below the ADC full scale input.
> 
> Bruce
>> On Feb 6, 2010, at 6:46 PM, Bruce Griffiths wrote:
>> 
>>   
>>> Even better is to toss out the mixers and sample the RF signals directly.
>>> However suitable ADCs cost $US100 or more each.
>>> To which one has to add an FPGA and an interface to a PC with sufficient throughput to handle the down converted I + Q samples.
>>> 
>>> Bob Camp wrote:
>>>     
>>>> Hi
>>>> 
>>>> You probably could put a couple of cheap DAC's
>>>>       
>>> (ADCs are preferable as it avoids having to implement the conversion logic plus comparator required when using a DAC.)
>>> 
>>>     
>>>> on a board with a FPGA and reduce the data on the fly. I'd guess that would be be in the same $100 range as a half way decent sound card. Clock the DAC's off of a 10 MHz reference and eliminate the cal issue.
>>>> 
>>>> If you are down around 10 Hz or worse yet 1 Hz, the AC coupling of the sound card will get in the way, even with a bandpass approach. You really don't know what they may have in there at the low end. Build it yourself and that stuff's not an issue.
>>>> 
>>>> Bob
>>>> 
>>>> 
>>>>       
>>> My sound card has a 1Hz cutoff  RC high pass input filter plus an internal high pass digital filter.
>>> Its not too difficult to measure the sound card frequency response using a white noise source for example.
>>> 
>>> Bruce
>>>     
>>>> On Feb 6, 2010, at 6:12 PM, Bruce Griffiths wrote:
>>>> 
>>>> 
>>>>       
>>>>> If one has a high end sound card then it could be used to implement the bandpass filter and replace the zero crossing detector.
>>>>> It may be necessary to insert a pilot tone to calibrate the sound card sampling clock frequency.
>>>>> A noise floor of about 1E-13/Tau should be achievable.
>>>>> This simplifies the DMTD system by replacing the zero crossing detector with a low gain linear preamp.
>>>>> 
>>>>> If one analyses the resultant data off line then one can also try out different techniques such as a Costas receiver rather than a simple bandpass filter plus zero crossing detector.
>>>>> However 1000 seconds of data for 2 channels of 24 bit samples at 192KSPS will result in a file with a size of at least 1.15GB.
>>>>> 
>>>>> Bruce
>>>>> 
>>>>> 
>>>>> Bruce Griffiths wrote:
>>>>> 
>>>>>         
>>>>>> If one were to use a bandpass filter with a Q of 10 to filter the beat frequency output of the mixer, then if the input frequency is 10MHz and the filter component tempco is 100ppm/C then the resultant phase shift tempco is about 16ps/C referred to the mixer input frequency.
>>>>>> 
>>>>>> This phase shift tempco is certainly low enough not to have significant impact when measuring the frequency stability of a typical 10811A  if the temperature fluctuations are kept small enough during the run.
>>>>>> 
>>>>>> The effect of using a bandpass filter with too narrow a bandwidth is to artificially reduce ADEV for small Tau, so it may be prudent to use a higher beat frequency that 1Hz or even 10Hz and not calculate ADEV for Tau less than say 10(??) times the beat frequency period. A trade off between this and the effect of aliasing is required.
>>>>>> 
>>>>>> Bruce
>>>>>> 
>>>>>> Bob Camp wrote:
>>>>>> 
>>>>>>           
>>>>>>> Hi
>>>>>>> 
>>>>>>> With most 10811 range oscillators  the impact of a simple bandpass filter is low enough to not be a major issue. That's for normal lab temperatures with the circuitry in a conventional die cast  box. No guarantee if you open the window and let the fresh air blow in during the run.
>>>>>>> 
>>>>>>> That's true with a heterodyne. I can see no obvious reason it would not be true on DMTD.
>>>>>>> 
>>>>>>> Bob
>>>>>>> 
>>>>>>> 
>>>>>>> On Feb 6, 2010, at 5:12 PM, Bruce Griffiths wrote:
>>>>>>> 
>>>>>>> 
>>>>>>>             
>>>>>>>> The only major issue with DMTD systems is that they undersample the phase fluctuations and hence are subject to aliasing effects.
>>>>>>>> The low pass filter has to have a bandwidth of the same order as the beat frequency or the beat frequency signal will be significantly attenuated.
>>>>>>>> Since the phase is only sampled once per beat frequency period the phase fluctuations are undersampled.
>>>>>>>> Various attempts to use both zero crossings have not been successful.
>>>>>>>> 
>>>>>>>> In principle if one can overcome the increased phase shift tempco associated with a bandpass filter, using a bandpass filter can in principle ensure that the phase fluctuations are oversampled.
>>>>>>>> 
>>>>>>>> 
>>>>>>>> Bruce
>>>>>>>> 
>>>>>>>> Bob Camp wrote:
>>>>>>>> 
>>>>>>>>               
>>>>>>>>> Hi
>>>>>>>>> 
>>>>>>>>> A straight heterodyne system will get you to the floor of most 10811's with a very simple (2 stage) limiter. As with the DMTD, the counter requirements aren't really all that severe.
>>>>>>>>> 
>>>>>>>>> Bob
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> On Feb 6, 2010, at 4:24 PM, WarrenS wrote:
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>>                 
>>>>>>>>>> 
>>>>>>>>>>                   
>>>>>>>>>>> "It's possible / likely for injection lock ... to be a problem ..."
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>>>                     
>>>>>>>>>> Something I certainly worried about and tested for.
>>>>>>>>>> What I found (for MY case) is that injection lock is NOT a problem.
>>>>>>>>>> The reason being is that unlike most other ways, where the two OSC have to be completely independent,
>>>>>>>>>> The tight loop approach forces the Two Osc to "Lock with something like 60 + db gain,
>>>>>>>>>> so a little stray -80db injection lock coupling that would very much limit other systems has
>>>>>>>>>> no measurable effect at e-13. Just one of the neat little side effects that make the tight loop approach so simple.
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>>                   
>>>>>>>>>>> "then a part in 10^14 is going to be at the 100 of nanovolts level."
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>>>                     
>>>>>>>>>> For that example, just need to put a simple discrete 100 to 1 resistor divider
>>>>>>>>>> in-between the control voltage and the EFC and now you have a nice workable 10uv.
>>>>>>>>>> BUT the bigger point is, probable not needed, cause you are NOT going to do any better than the stability of the OSC with a grounded shorted EFC input.
>>>>>>>>>> 
>>>>>>>>>> as you said and I agree is so true:
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>>                   
>>>>>>>>>>> "There is no perfect way to do any of this, only a lot of compromises ... you need to watch out for".
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>>>                     
>>>>>>>>>> But you did not offer any easier way to do it, which is what the original request was for and my answer addressed.
>>>>>>>>>> This is the cheapest easiest way BY FAR to get high performance, at low tau, ADEV numbers that I've seen.
>>>>>>>>>> 
>>>>>>>>>> ws
>>>>>>>>>> ***************
>>>>>>>>>> 
>>>>>>>>>> ----- Original Message ----- From: "Bob Camp"<lists at cq.nu>
>>>>>>>>>> To: "Discussion of precise time and frequency measurement"<time-nuts at febo.com>
>>>>>>>>>> Sent: Saturday, February 06, 2010 12:09 PM
>>>>>>>>>> Subject: Re: [time-nuts] ADEV vs MDEV
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>>                   
>>>>>>>>>>> Hi
>>>>>>>>>>> 
>>>>>>>>>>> It's possible / likely to injection lock with the tight loop approach and get data that's much better than reality. A lot depends on the specific oscillators under test and the buffers (if any) between the oscillators and mixer.
>>>>>>>>>>> 
>>>>>>>>>>> If your OCVCXO has a tuning slope of 0.1 ppm / volt then a part in 10^14 is going to be at the 100 of nanovolts level. Certainly not impossible, but it does present it's own set of issues. Lab gear to do it is available, but not all that common. DC offsets and their temperature coefficients along with thermocouple effects could make things exciting.
>>>>>>>>>>> 
>>>>>>>>>>> There is no perfect way to do any of this, only a lot of compromises here or there. Each approach has stuff you need to watch out for.
>>>>>>>>>>> 
>>>>>>>>>>> Bob
>>>>>>>>>>> 
>>>>>>>>>>> --------------------------------------------------
>>>>>>>>>>> From: "WarrenS"<warrensjmail-one at yahoo.com>
>>>>>>>>>>> Sent: Saturday, February 06, 2010 2:19 PM
>>>>>>>>>>> To: "Discussion of precise time and frequency measurement"<time-nuts at febo.com>
>>>>>>>>>>> Subject: Re: [time-nuts] ADEV vs MDEV
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>>>                     
>>>>>>>>>>>> Peat said:
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>>                       
>>>>>>>>>>>>> I would appreciate any comments or observations on the topic of apparatus with demonstrated stability measurements.
>>>>>>>>>>>>> My motivation is to discover the SIMPLEST scheme for making stability measurements at the 1E-13 in 1s  performance level.
>>>>>>>>>>>>> 
>>>>>>>>>>>>> 
>>>>>>>>>>>>>                         
>>>>>>>>>>>> If you accept that the measurement is going to limited by the Reference Osc,
>>>>>>>>>>>> for Low COST and SIMPLE, with the ability to measure ADEVs at that level,
>>>>>>>>>>>> Can't beat a simple analog version of  NIST's "Tight Phase-Lock Loop Method of measuring Freq stability".
>>>>>>>>>>>> http://tf.nist.gov/phase/Properties/one.htm#oneone    Fig 1.7
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>> By replacing the "Voltage to freq converter, Freq counter&     Printer with a Radio shack type PC data logging DVM,
>>>>>>>>>>>> It can be up and running from scratch in under an Hr, with no high end test equipment needed.
>>>>>>>>>>>> If you want performance that exceeds the best of most DMTD at low Tau it takes a little more work
>>>>>>>>>>>> and a higher speed oversampling ADC data logger and a good offset voltage.
>>>>>>>>>>>> 
>>>>>>>>>>>> I must add this is not a popular solution (Or a general Purpose one) but
>>>>>>>>>>>> IF  you know analog and have a GOOD osc with EFC to use for the reference,
>>>>>>>>>>>> as far as I've been able to determine it is the BEST SIMPLE answer that allows High performance.
>>>>>>>>>>>> Limited by My HP10811 Ref OSC, I'm getting better than 1e-12 in 0.1 sec (at 30 Hz Bandwidth)
>>>>>>>>>>>> 
>>>>>>>>>>>> Basic modified NIST Block Diag attached:
>>>>>>>>>>>> The NIST paper sums it up quite nicely:
>>>>>>>>>>>> 'It is not difficult to achieve a sensitivity of a part in e14 per Hz resolution
>>>>>>>>>>>> so one has excellent precision capabilities with this system.'
>>>>>>>>>>>> 
>>>>>>>>>>>> This does not address your other question of ADEV vs MDEV,
>>>>>>>>>>>> What I've described is just a simple way to get the Low cost, GOOD Raw data.
>>>>>>>>>>>> What you then do with that Data is a different subject.
>>>>>>>>>>>> 
>>>>>>>>>>>> You can run the raw data thru one of the many ADEV programs out there, 'Plotter' being my choice.
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>> Have fun
>>>>>>>>>>>> ws
>>>>>>>>>>>> 
>>>>>>>>>>>> *************
>>>>>>>>>>>> 
>>>>>>>>>>>> [time-nuts] ADEV vs MDEV
>>>>>>>>>>>> Pete Rawson peterawson at earthlink.net
>>>>>>>>>>>> Sat Feb 6 03:59:18 UTC 2010
>>>>>>>>>>>> 
>>>>>>>>>>>> Efforts are underway to develop a low cost DMTD apparatus with
>>>>>>>>>>>> demonstrated stability measurements of 1E-13 in 1s. It seems that
>>>>>>>>>>>> existing TI counters can reach this goal in 10s. (using MDEV estimate
>>>>>>>>>>>> or 100+s. using ADEV estimate). The question is; does the MDEV tool
>>>>>>>>>>>> provide an appropriate measure of stability in this time range, or is
>>>>>>>>>>>> the ADEV estimate a more correct answer?
>>>>>>>>>>>> 
>>>>>>>>>>>> The TI performance I'm referring to is the 20-25 ps, single shot TI,
>>>>>>>>>>>> typical for theHP5370A/B, the SR620 or the CNT81/91. I have data
>>>>>>>>>>>> from my CNT81showing MDEV<     1E-13 in 10s. and I believe the
>>>>>>>>>>>> other counters behave similarly.
>>>>>>>>>>>> 
>>>>>>>>>>>> I would appreciate any comments or observations on this topic.
>>>>>>>>>>>> My motivation is to discover the simplest scheme for making
>>>>>>>>>>>> stability measurements at this performance level; this is NOT
>>>>>>>>>>>> even close to the state-of-the-art, but can still be useful.
>>>>>>>>>>>> 
>>>>>>>>>>>> Pete Rawson
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>>                       
>>>>>         
>>> 
> 
> 
> 
> _______________________________________________
> 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_lists.febo.com mailing list