[time-nuts] Designing and building an OCXO and GPSDO

[Philip Pemberton]

Bruce Griffiths wrote:
 > If you are serious forget the fancy digital or semiconductor
 > temperature sensors they aren't good enough.

I was intending to use the slow Dallas chips as a calibration reference
(out-of-box they're usually quite accurate) and for testing. Is there 
any particular reason the analog-output (Microchip TC1047A) sensors are 
no good?

 > For the best performance, unless you use a bridge oscillator circuit
 > of some type, you will need to control the temperature of all the
 > oscillator components as well.

I did have a "plan B" -- a hollow metal box with a metal sheet soldered
inside at the half-way point. The crystal and oscillator circuitry would
be mounted in the bottom half, and the temperature control in the top
half. The temperature sensor is a three-pin SOT23 (about the size of a
grain of rice) and the ground is a single pin on one side. I was
thinking about mounting the sensor directly on the copper, using a small
piece of Kapton tape to stop the sensor's Vout or Vcc shorting against
the grounded copper sheet.

That would leave two hollow air-filled cavities for the control
circuitry and hold the temperature of that reasonably close to that of
that of the crystal (minus a few degrees).

 > An analog bridge using an RTD or an NTC thermistor can have much
 > better stability.

That sounds about right.. I was going to use a Pt100 or Pt1000 RTD, but
couldn't find any decent information on them other than the resistance
being 100R or 1kR at 25C -- even the manufacturer's datasheets were
somewhat thin on information.

 > If you use an appropriate high resolution sigma delta ADC it can
 > reverse the bridge excitation polarity as part of the measurement
 > sequence and give you most of the benefits of an AC bridge with fewer
 > devices and lower cost.

IIRC the A/D on the PIC is a 10-bit successive-approximation type with a
built-in sample-and-hold (though other types have 12-bit converters).

That's a measurement range of 1024 counts, which with the 4.096V
reference provides a resolution of 4mV, or 1/2.5 of a degree C per
count. 4V is actually the minimum reference voltage the A/D can accept.

Sensor output is ((degrees_c * 10) + 500) mV.

 > I would build a room temperature version first for debugging.

I'm planning to do that anyway. I've got a few 10MHz room-temp crystals
of a similar spec to the oven crystals that I can use, and I can
probably use the same parts in the prototype oven for testing.

 > To minimise the phase noise contributed by the varicap the EFC range
 > should be as small as is practical.

That's the part that's going to need "a bit" of experimentation I think :)

 > A very low noise power supply is also required for good performance.
 > A modified version (uses 2 transistors and larger capacitors) of
 > Wenzels active supply filter can be used to reduce the power supply
 > noise by 30-40dB for frequencies above 1Hz or so.
 > http://www.wenzel.com/documents/finesse.html
 >
 > I can provide circuit schematics if you are interested.

That would be great, if it's not too much trouble.

Thanks,
Phil.