[time-nuts] Short term 10MHz source
Magnus Danielson
magnus at rubidium.dyndns.org
Wed Jan 9 00:03:23 UTC 2019
John,
Recall that the quantization is really a form of time-stamp value for
the channel in it's relation to the time-base. It's a systematic pattern
in the time-base clock and it is phase-locked to the time-base phase.
Let's just think of our time-base as being a 10 MHz. That 100 ns cycle
is then interpolated to the coarse-count clock, so let's assume a 100
MHz coarse-clock which is phase-locked to the 10 MHz time-base, now you
have 10 ns steps which interpolates the 100 ns. However, since we want a
higher resolution, we then use a time-interpolation technique of some
sort to get higher resolution, let's say we create a 50 ps resolution
which then interpolates over the coard-counter's 10 ns, that gives
10000/50 = 20000/100 = 200 steps interpolation steps. If we for the
moment assume that this interpolation is perfectly linear, they will be
very neatly distributed exactly 50 ps appart in relation to the 10 ns
and hence the 100 ns of the time-base.
Now, as we have two of these channels, if both of them is asynchronous
to the time-base, over time you will sweep over the full range of these
for both clocks, and you get an average of them and the statistics of
them smoooth out.
If we now let one of the channels actually be either directly be a
delayed version of the time-base clock or a low-jitter generated version
of the time-base but at some other frequency (such as 1 Hz of PPS) as
you compare these back to the time-base, they have a relatively static
relationship in the systematic context and what only remains is the pure
random phase-noise relationship of WPM and FPM, and essentially you can
now have a situation where the noise is low enough that you hit almost
all your triggers in the same interpolation bin on that channel, with a
few hits in the side-bins. For this all of a sudden the systematic noise
of that channel is almost or completely gone for all practical purposes.
As you now is measuring the other channel which is asynchronous, you
only see the asynchronous noise of that channel.
If you have a synthesized signal which does not hit the same spot in the
interpolation bin on each occasion you may either have a scenario of
hitting only a few spots for simple frequency rations or you end up
sweeping the full range. The more you go towards the full range, the
more the signal acts like a asynchronous signal. This is what creates
the funky resolution plot vs. frequency as you sweep over the fractional
frequency relationships between an input channel and time-base for a
normal frequency setup. It's a very systematic feature. As you sweep
over many points of the interpolators phase error, non-linearities of
the quantization step compared to the phase of the time-base will expose
themselves more and more.
Now, if we now use a delay cable to insert the same signal into the
start and stop, we can have a scenario where both start and stop
interpolators hit about the same bin if they are fed from the time-base
(or a source with a frequency very near the timebase) and then what is
measured will look very much better than the actual noise floor, unless
the signal is rise-time limited and cause severe trigger jitter on both
channels, at which time you hit a range of interpolator bins. Anyway,
with good signal conditions your baseline measurement will give you
over-confident values for your baseline measurement than would be the
case for one or both of these being asynchronous to the timeb-base.
For such a cable setup to give reasonable baseline it needs to be done
either with a source which is clearly asynchronous or has a fractional
rate which within the datacollection range sweeps the full range more or
less equally well.
So, to conclude, the quantization noise that we have is very systematic
in its nature, and depending on your measurement setup it can play out
very differently for you. I've only showed some of the scenarios here,
but you can play out more scenarios which have interesting effects.
You just have to be very careful with this.
Does this make sense?
Cheers,
Magnus
On 1/8/19 10:07 PM, John Ackermann N8UR wrote:
> You've mentioned this before, and I'm having trouble getting my head
> around it. I may have this all wrong, but isn't the quantization simply
> the resolution of the device? Your histogram shows humps about 50
> picoseconds apart, but that's the resolution of the counter.
>
> Wouldn't any counter that has 50 picosecond resolution look the same?
> Its data output is quantized by its resolution. For example, the 5334A
> with 2ns resolution will always show bins no less than 2ns apart.
>
> What am I missing?
>
> John
> ----
>
> On 1/8/19 3:34 PM, Tom Van Baak wrote:
>> John,
>>
>> Your hunch is correct. For most modern TIC devices, measurement noise
>> is white. That is, if you do a self-test using a common DUT & REF you
>> get a nice clean Gaussian plot. But the TICC is not like that. The
>> TICC is based on a ring counter and so there is a *high* degree of
>> quantization. This is not bad, per se, but it does impact how one
>> should perform, or interpret, a noise floor test.
>>
>> Take a look at: http://leapsecond.com/pages/ticc/
>>
>> And in particular:
>> http://leapsecond.com/pages/ticc/ticc-log5342-hist-1.gif
>>
>> What this means is that a noise floor measurement made with the same
>> chA and chB and REF could be quite wrong. This is not an indictment
>> against the TICC. I have several, and use them all. But because of the
>> strong quantization effects you can't just feed in a common DUT and
>> REF and expect that to represent all possible real-life phase
>> measurements. Those quantized "camel humps" are really quite extreme
>> with the TICC.
>>
>> /tvb
>>
>> ----- Original Message -----
>> From: "John Ackermann N8UR" <jra at febo.com>
>> To: <time-nuts at lists.febo.com>
>> Sent: Tuesday, January 08, 2019 10:48 AM
>> Subject: Re: [time-nuts] Short term 10MHz source
>>
>>
>>> Hi Luciano --
>>>
>>> Thanks for posting that. There's a subtle point about the noise floor
>>> that's forever been on my list of things to investigate: the noise floor
>>> should be lower in timestamp mode than in time interval (A->B) mode.
>>>
>>> That's because in timestamp mode there is jitter contribution only from
>>> a single measurement, whereas in time interval mode there is a
>>> measurement from each channel so you have two jitter components. So a
>>> guess is that the floor should be about sqrt(2) lower in timestamp mode.
>>> Someday I will test that theory.
>>>
>>> John
>>> ----
>>> On 1/8/19 12:38 PM, timeok at timeok.it wrote:
>>>>
>>>> .gif of the TICC noise floor.
>>>> Luciano
>>>>
>>>>
>>>> Da "time-nuts" time-nuts-bounces at lists.febo.com
>>>> A time-nuts at lists.febo.com
>>>> Cc
>>>> Data Tue, 8 Jan 2019 18:31:02 +0100
>>>> Oggetto Re: [time-nuts] Short term 10MHz source
>>>>
>>>> Hi Paul,
>>>> here the TICC noise floor.
>>>> Regarding the GPS/TICC versus a good Rubidium standard like the
>>>> HP5065A , you cannot apreciate the Rubidium ADEV stability lower
>>>> than 10Kseconds.
>>>> Luciano
>>>>
>>>>
>>>> Da "time-nuts" time-nuts-bounces at lists.febo.com
>>>> A "Discussion of precise time and frequency measurement"
>>>> time-nuts at lists.febo.com
>>>> Cc
>>>> Data Sat, 5 Jan 2019 12:35:26 -0000
>>>> Oggetto Re: [time-nuts] Short term 10MHz source
>>>> Hi All sorry for a new be question but what is a TICC regards
>>>> Paul B UK
>>>>
>>>> -----Original Message-----
>>>> From: time-nuts [mailto:time-nuts-bounces at lists.febo.com] On
>>>> Behalf Of Chris
>>>> Burford
>>>> Sent: 02 January 2019 03:56
>>>> To: Time Nuts List
>>>> Subject: [time-nuts] Short term 10MHz source
>>>>
>>>> I have a situation in which I have access to a GPSDO 10MHz
>>>> source but for
>>>> only about 10-12 hours at a time. My current residence does not
>>>> allow a
>>>> permanent GPS antenna therefore I am limited in its use.
>>>>
>>>> I do realise that the long term stability of the GPSDO is
>>>> somewhat superior
>>>> to a Rubidium source. I'm planning on using my TICC to validate
>>>> both my
>>>> GPSDO and RFS. I'm aware that such a short "power on" period is
>>>> somewhat
>>>> counterproductive but I have no other options. I'd like to know
>>>> if a 6-8
>>>> hour window for the GPSDO is sufficient for use as a 10MHz
>>>> source for the
>>>> TICC.
>>>>
>>>> I appreciate any and all comments.
>>>>
>>>> Regards, Chris
>>
>>
>> _______________________________________________
>> time-nuts mailing list -- time-nuts at 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 at lists.febo.com
> To unsubscribe, go to
> http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.
More information about the Time-nuts_lists.febo.com
mailing list