[time-nuts] Questions about TAC frontend, and some measurements

Bruce Griffiths bruce.griffiths at xtra.co.nz
Wed Dec 26 18:57:50 UTC 2012

Fabio Eboli wrote:
> Hello, hope you all had a happy Christmas.
> Back to the topic.
> Bob Camp asked:
>> Hi
>> One very simple question - how good would it do if you just did it 
>> all with logic gates? Tri-state buffers and things like that….
>> Now that you are up to a 100 to 200 ns long pulse, a lot of the 
>> fiddly stuff about "can't get a 2 ns pulse through it" goes away.
>> I'm not suggesting you tear up what you have. It's just something 
>> else to try and compare
>> Bob
> Bob, are you hinting to something like the last mail from Bruce?
> I.e. to use a tristate buffer to charge the capacitor?
> If not can you explicit what are you thinking? :)
> Thanks also to Alan Melia and Tom Miller for the details about
> bjt saturation .
> Bruce, about the tempco of the current generators,
> There is the led in series with a BE junction.
> The blue leds should have a tempco in mV per °C similar
> to th BE junction, dont know the red ones.
> Would it be better to use something like a 4.7 or 5.1V
> zener? If I remember correctly these zener voltage
> shuld cancel most of the BE tempco.
> And what about a TL431 instead of the led+bjt?
The BJT is essential to ensure that the current source has a high output 
Opamps dont help as most have insufficient gain at high frequencies.
The input capacitance of an opamp input connected directly to the 
current source emitter isnt conducive to a high output impedance either.
Its better to drive the emitter of the second transistor with an npn 
emitter follower thats part of a servo loop using a suitably decoupled 
opamp (series resistor from the current source emitter to the inverting 
input of the opamp) to set the emitter current of the current source 
> The Avago diodes are pretty costly :)
> Is that circuit working like the internals of ECL logic
> families?
Yes apart from the lack of emitter follower outputs.
Its called current mode logic.
>> The simplest (lowest part count and least number of power supplies)
>> consists of a tristate buffer driving an RC circuit.
>> The PPS signal is connected directly to the buffer input whilst the
>> output of the PPS synchroniser (at least 2 stages to minimise the
>> probability of metastabilty at the synchroniser output) drives the
>> buffer tristate control input.
> A 2 stage syncronizer is composed of 3 FF?
No 2 FF, the first FF is just a convenient means of stretching narrow 
pulses and ensuring that the synchroniser input pulse width has the 
required duration.
> I.e. clock in parallel to 3 FF, PPS to the
> first D, Q from the first to D of the second,
> same from the second to the thid, and Q from
> the third to out. Let's assume that the inputs
> from PPS and 10MHz are fast enough, what can still
> generate metastability? Setup time violation?
>> The RC network starts charging when the PPS signal goes high and
>> stops when the synchroniser output goes high.
>> The capacitor charging is nonlinear but this is easily corrected in 
>> software.
>> The capacitor is connected between the input of a capacitive charge
>> redistribution ADC and ground.
>> Software correction for the effect of charging the charge ADC input
>> capacitance is also required.
> I see you are stressing the fact of using a capacitive charge
> redistribution adc. I dont know much about the internals
> of the ADC devices, can you suggest a partnumber for an example?
Almost any modern SAR ADC such as the LTC1282 and later.
>> Suitable fast single gate tristate drives are readily available.
>> With low tempco resistors and capacitors the TAC gain tempco can be
>> 200pmm/C or less.
>> The only disadvantages are the increased software complexity and the
>> need for an extra bit of ADC resolution to maintain TAC resolution.
> The 3-state buffer + R-C seem an elegant solution for a microcontroller
> based thing, I'v given an eye to logic buffers, and seem that all
> suggest that the Hi-Z state leackage current is not very well
> specified, but something around 1uA, that means that cap's voltage after
> the pulse can rapidly (and unpredictabily?)change due to leackage.
> I imagine also that the leackage of the buffer will vary with 
> temperature.
Kasper Pedersen has used this technique. http://n1.taur.dk/gpsdo2a.pdf
However he only used a single stage synchroniser which is far from ideal.
> The ADC of the micro is pretty fast, I shuld check the datasheet
> but I remember around 1uS per conversion, what would happen connecting
> directly the micro ADC to the charged cap? And sync the ADC to sample
> immediately (few uS) after the pulse. Could the loading from the
> s/h capacitance be corrected in fw?
The best way is to have the ADC in sample mode whilst the capacitor is 
being charged.
Wait sufficient time at the end of the charging process for the ADC 
sampler to settle and then trigger a conversion.
If possible, its best for this conversion trigger to be generated by the 
synchroniser (use a shift register 1st 2 stages for the synchroniser 
prper and the following stages used to generate the required delay.
The effect of the sampling capacitance can be corrected in software to a 
first approximation it merely changes the TAC gain.
>> Bruce
> By the way, I updated my miserable schematic, I tried a simple
> mod to avoid the saturation of the switches. Only because I had
> it already built: http://pastebin.com/9VHkhmSv
> Now I'm chasing the origin of the drift variation, logging
> the temperatures and voltages. More on this as soon as I
> have some data.
> Thank you all,
> Fabio.

More information about the time-nuts mailing list