Designing and building an OCXO and GPSDO
WB6BNQ wrote:
The first is a series of Application notes from Agilent (old hp test
div) called AN-200. A total of 5 App notes comprise the AN-200 series.
If you go to the following Web page and enter AN-200 at the top of the
page in the search box, you will get a return of all the AN-200 booklets
in PDF that can be downloaded. The BIG one is AN-200-2, but it would be
to your advantage to collect all of them. You need to paste in the
entire link below if your browser doesn't see the whole thing when
clicking on it.
Hm, nothing found on Agilent's site, but Google found it (search term:
"agilent.com AN-200").
Lots of interesting stuff there, from quartz oscillator basics through
microwave signal generation and measurement...
That should keep you busy for a while. A lot of stuff on the WEB, some
good and some not so good, just take it with a grain of salt ! NIST
(the old NBS) has several things worth reading, however, most of that
deals with the measurement process or is a rigorous mathematical
analysis of one thing or another.
Well I was looking at the manual for the ELV OCXO400
(http://www.elv-downloads.de/service/manuals/OCXO400/OCXO400_KM_G_010502.pdf)
as a partial reference, and the disassembly photos for one of the HP
dual-oven oscillators (can't find the link at the moment).
I dismissed the ELV's mechanical design based on the fact that it was two
BC337 (read: low power) transistors mounted on each side of the crystal, with
a cheap thermistor on top. No real thermal load at all -- as soon as the
transistors turn on, the temperature is likely to shoot WAY, WAY up, then
drop FAST when the transistors are turned off. The comparator-based
thermostat-type temperature control isn't likely to make that much easier
either.
Apparently the oscillator circuit is quite badly designed too (based on
information at http://glowbug.nl/projects/M51.html). Badly biased output
transistors and mismatched xtal load capacitance mainly.
Second it does not take much to get parts in 10^-7 range. Temperature
compensated crystal oscillators easily handle that level. With care, a
crystal oscillator in a well designed circuit can reach parts in 10^-8
with a bit-bang oven control. HP did that in the late 1950's. From
that point the difficulty is logarithmic.
It certainly seems that way...
Nice to know it's not quite as hard as it looks. The ELV is specced at
2x10^-8 stability and uses an AT-cut HC49U fundamental-mode parallel-resonant
crystal. I don't have the specs for the crystal cut used for my 10MHz xtals,
but I do know the base specifications are almost identical to the ELV unit's
crystal.
I should probably mention the main reason I wanted to use the PIC control --
so I can adjust the oven parameters (desired temperature, PID variables, and
so on) in real time, and read off the oven status at the same time (current
temperature, target temperature, last warmup time, hours run since last power
cycle, total crystal age, starting temperature, crystal oscillator state, and
so on). Obviously the turning point for each crystal isn't going to be
exactly 50.0000000... Celsius, so being able to hook up a high-resolution
counter and figure out what the exact turning point is and program the
thermal management kit with that would be useful.
In truth this is more a research and learning exercise. It'd just be nice to
end up with something that compares reasonably well to a commercial OCXO,
that I can use as a basic frequency reference. Then later on I'll decide if I
really need anything better.
Thanks,
Philip
I dismissed the ELV's mechanical design based on the fact that it was two
BC337 (read: low power) transistors mounted on each side of the crystal, with
a cheap thermistor on top. No real thermal load at all -- as soon as the
transistors turn on, the temperature is likely to shoot WAY, WAY up, then
drop FAST when the transistors are turned off. The comparator-based
thermostat-type temperature control isn't likely to make that much easier
either.
Thermal gradients in the vicinity of the crystal will be quite high.
Apparently the oscillator circuit is quite badly designed too (based on
information at http://glowbug.nl/projects/M51.html). Badly biased output
transistors and mismatched xtal load capacitance mainly.
Not to mention the low isolation buffer amplifier with relatively high
phase noise.
Extracting the signal through the crystal is a good method for ensuring
a low phase noise floor.
Ignore the misguided comments in the literature and on the web that say
otherwise.
Their calculations indicate a phase noise floor 20dB or more higher than
that achieved in practice.
This is the result of applying an standard equation to a situation where
the assumptions implicit in its derivation don't hold.
Second it does not take much to get parts in 10^-7 range. Temperature
compensated crystal oscillators easily handle that level. With care, a
crystal oscillator in a well designed circuit can reach parts in 10^-8
with a bit-bang oven control. HP did that in the late 1950's. From
that point the difficulty is logarithmic.
It certainly seems that way...
Nice to know it's not quite as hard as it looks. The ELV is specced at
2x10^-8 stability and uses an AT-cut HC49U fundamental-mode parallel-resonant
crystal. I don't have the specs for the crystal cut used for my 10MHz xtals,
but I do know the base specifications are almost identical to the ELV unit's
crystal.
I should probably mention the main reason I wanted to use the PIC control --
so I can adjust the oven parameters (desired temperature, PID variables, and
so on) in real time, and read off the oven status at the same time (current
temperature, target temperature, last warmup time, hours run since last power
cycle, total crystal age, starting temperature, crystal oscillator state, and
so on). Obviously the turning point for each crystal isn't going to be
exactly 50.0000000... Celsius, so being able to hook up a high-resolution
counter and figure out what the exact turning point is and program the
thermal management kit with that would be useful.
You'll need a really stable reference for the counter to get close to
the turning point.
Don't ramp the temperature too quickly as AT crystals are also sensitive
to thermal transients.
In truth this is more a research and learning exercise. It'd just be nice to
end up with something that compares reasonably well to a commercial OCXO,
that I can use as a basic frequency reference. Then later on I'll decide if I
really need anything better.
Thanks,
Bruce
Philip
The attached circuit schematic is for a 10MHz oscillator using a
fundamental mode AT cut crystal.
The problem with using a common base buffer with a fundamental mode
crystal is that the crystal ESR is rather low so that the real component
of the CB stage input impedance is not much less than that of the
crystal. This reduces the crystal loaded Q significantly unless the
common base stage is run at a high collector current which increases the
CB stage output current noise.
The crystal buffer shown here (Q103+Q104) reduces the real part of the
input buffer stage input impedance to a value much less than the ESR of
the crystal without using a large collector current.
The oscillator transistor is in a Colpitts configuration. The transistor
is only on for part of the cycle. The resistor (R102) connected between
the emitter and C101+C102 junction reduces the phase noise of Q101 when
it is on. The crystal current is determined by the dc collector current
of Q101 and can be varied by adjusting R103.
R103 is adjusted to set the crystal current to about 1.8mA (+13dBm
output in a 50 ohm load.)
The output stage uses an 18V supply so that it can drive an open circuit
load without saturating the output transistor.
This supply voltage can be reduced if a differential CB output stage is
used OR
-
The output device is allowed to saturate when driving a high
impedance load -
The output transformer turns ratio is reduced
-
A lower output than +13dBm is acceptable.
Since the phase noise floor is determined by the phase noise floor of
the buffer amplifier some care has been taken to ensure that the buffer
amplifier phase noise floor is low.
Power supply noise will increase the phase noise of the amplifier and
oscillator so a low noise power supply is essential.
At low offset frequencies the crystal and the oscillator stage start to
contribute to the phase noise.
Thus the oscillator phase noise also needs to be low for low phase
noise close to the carrier.
It can be advantageous to use transistors with low collector base
capacitances as their PM and AM noise can be made very low with RF feedback.
The major source of flicker PM and flicker AM noise in an amplifier is
modulation of the various device capacitances by low frequency noise
currents or voltages.
Thus low noise power supplies and low noise components should be used.
Only use metal film or thin film resistors.
No thick film, carbon resistor, metal oxide resistor should be used as
they have significant flicker noise when a dc current is flowing through
them.
All coupling caps should be NP0/C0G or equivalent.
Bruce
Hi Philip,
On Sun, 2008-08-10 at 22:45 +0100, Philip Pemberton wrote:
As far as temperature regulation goes, I'm going to use a PIC
microcontroller (one of the 8-pin chips with an A/D converter) to monitor the
temperature of the crystal, and use a PID loop to control the two power
transistors to maintain a temperature of 50C +/- 2 Celsius (the accuracy spec
of the temperature sensor). I also have other higher-accuracy sensors (Dallas
DS18S20 and DS18B20) that I can calibrate with; these are accurate to around
half a degree Celsius with a resolution of 0.5C.
Here is an example of a kind of synthetic TCXO using Dallas temp sensors.
http://www.ijs.si/time/temp-compensation/
The resolution of the DS18x20-sensors are better than 0.5C.
--
Björn
Attached schematic is for a variant of the Wenzel circuit using a JFET
high input impedance buffer to drive a cascaded pair of CB stages.
L104 is approximately resonant with C111 at 3MHz so that either an RF
choke and series resistor or a relative;y low value resistor (10K) can
be used to set the JFET gate voltage.
The impedance of C104 at 3MHz is about 530 ohms so a 1mA crystal current
will result in about 530 mV rms across C104.
An additional inductor can be used in series with the crystal to
resonate out the input capacitance of the oscillator stage if required,
this will allow a greater tuning range to be achieved.
Inductor L104 needs to have a reasonably high Q ( > 100 @ 3MHz) to
avoid degrading the loaded Q significantly.
The drain current of JFET Q103 is determined by the collector current of
Q104 this minimises the effect of JFET parameter variations from part to
part on its drain current.
The RF voltage at the source of Q103 can be amplified (a widebandwidth
opamp could be used) and used to drive a detector diode for AGC purposes.
Alternatively a common base stage connected to the other end of the
secondary of T101 could be used an RF current source to drive a full
wave diode detector with a low load resistance.
This will largely eliminate the effect of the detector diode forward
voltage drops on the detected output. This detector dc load current can
than be compared with a reference current and the dc collector current
of Q101 adjusted to regulate the crystal current.
Bruce
