[time-nuts] Advantages & Disadvantages of the TPLL Method

Bruce Griffiths bruce.griffiths at xtra.co.nz
Sat Jun 12 22:55:24 UTC 2010


Magnus Danielson wrote:
> On 06/12/2010 11:29 PM, Bruce Griffiths wrote:
>> WarrenS wrote:
>>> subject: Advantages& Disadvantages of the TPLL Method.
>>>
>>> Here is a new and unique Idea that may be useful for many.
>>> Rather than focusing on what some members may or may not already know,
>>> or how good or bad one specific working BB configuration is.
>>> How about focusing on what the TPLL method can and can not do well.
>>> If someone will make a place to post and compile a couple of list,
>>> I can start it off with what I've learned so far:
>>>
>>>
>>> DISADVANTAGES of the TPLL method:
>>> -------------------------------------------------------
>>> #1) The TPLL method is limited by it's reference OSC.
>> This isn't necessarily correct, one could use a pair of tight PLL loops
>> and use correlation techniques to reduce the contribution of the
>> reference oscillator noise.
>
> True. The same technique is being used for LPLL phase noise 
> measurements. The reference oscillator will still be a limit, but 
> wither you can go below the reference oscillator noise or not is what 
> makes the difference. Such a setup costs about twice of a 
> single-channel TPLL. Usually there is two ADC channels available.
>
Yes the cost of the reference oscillator dominates the system cost, the 
additional $10 (omitting the cost of the phase detector) to implement 
the tight PLL is relatively insignificant.
> The cross-correlation processing isn't too hard to achieve and is 
> efficiently performed using FFTs and a little support-processing. FFTW 
> is a good tool to toss the FFT processing to. The remaining wrapping 
> is in a few ten lines of codes or so. Going down the FFT path will 
> give the frequency plot for free, getting it back into the time-domain 
> cost extra.
>
If one is calculating the FFT then it is possible to calculate ADEV 
directly from the FFT (of the frequency samples) with little additional 
effort, for the relevant formulae see:

http://hal.archives-ouvertes.fr/docs/00/37/63/05/PDF/alaa_p1_v4a.pdf

Note such processing doesn't increase the cost of the system as one 
needs a PC to calculate frequency stability measures, unless one 
wants/needs to do it in real time.

One disadvantage of a tight PLL system is that finite EFC range and EFC 
non linearity may preclude its application to noisier sources.
Linearising the EFC transfer function will help but the reference 
oscillator EFC range will ultimately provide an upper limit to the 
measurable noise.



>>> The ref osc (or the DUT) needs to have an Analog&/or Digital EFC
>>> control input with a bandwidth that is wider than the desired Tau0
>>>
>>> #2) It basically measures Freq and not Phase differences, and few
>>> understand how and why it works so well or it's many advantages.
>> This is not true, there is no inherent SNR advantage in measuring
>> frequency changes as opposed to measuring phase differences.
>> When the phase measurement system and the frequency measurement systems
>> being compared have the same noise bandwidth then the measurement floors
>> are comparable.
>> For example, the TSC5120A is a narrow band system based on measuring
>> phase differences with a comparable or lower noise floor than your
>> implementation of the tight PLL.
>>
>> The common technique of using a time interval counter to measure the
>> phase difference between 2 RF signals once ever second or so is a
>> wideband technique with severe undersampling, consequently the system
>> noise floor is much higher than for narrow bandwidth techniques. If the
>> phase difference between the 2 signals were measured more frequently and
>> digitally low pass filtered the noise will be much lower.
>
> Using time-stamping counters at high rate would be possible if being 
> able to cope with the rate of samples. You want a frontend to do that 
> if you want to run continously.
>
> As for digital filtering. When doing measurements in the 0,1 - 1000 s 
> range for the G.813 measurements, a 10 Hz low-pass filter is being 
> required.
>
>> Since one has to calculate average frequency from the frequency samples
>> by integration/averaging this is mathematically equivalent to
>> reconstructing the phase change between the start and end of the
>> averaging time (Tau0).
>
> Depends on the details. Some counters (SR620 for instance) can have 
> biases for frequency data which their time-difference measures do not 
> have. A TPLL does not suffer from that particular problem, as it 
> cranks out its frequency estimation by a different method.
Yes, but I thought that we were calculating the required averages from 
the frequency (EFC) samples by approximating the required integrals.



>
>> One effect of undersampling is to convert (in the sampled data) a
>> proportion of any flicker phase noise (and other non white phase noise
>> components) to white phase noise.
>> The effect of this is to change the ADEV vs Tau plots from their true
>> shape.
>
> Care to hand a reference or two for this statement?
References for the whitening effect of undersampling:
http://www.obs-besancon.fr/tf/publis/metrologia98a.pdf
http://www.obs-besancon.fr/tf/publis/metrologia98b.pdf

The change in shape of the ADEV vs Tau plot is a consequence of the 
whitening of the phase noise.


>
> Regardless, care must be taken to ensure high enough bandwidth 
> compared to the tau for the measurements not to be affected.
>
> Cheers,
> Magnus
>
Bruce




More information about the time-nuts mailing list