PUNiSH3RFor my GPSDO project I’ve opted for a Piezo 2940210 VCOCXO at 10 Mhz. There is a dearth of information on this particular module available online.
(A notation notation: I utilize “E” notation to indicate orders of magnitude; thus 1.5E-6 is in fact 0.0000015 and 7.4E3 is 7400)
Via the information available at N4IQT.com, the following specifications apply to this OCXO module:
Output Specifications
- 10 Mhz Square Wave
- 2Vp-p±10% (2.053V in this case) into 50Ω
- Load of 50W±5%
- Harmonics <-25dBc and Spurious signals <-60dBc
OCXO Stability
- Ambient Temperature Stability of <±2E-9 from -30C to 60C, Referenced at 25C
- Daily Aging Stability of <±1E-9 after the first 30 days, <±5E-10 after the first 90 days
- Yearly Aging Stability of <±1.5E-7
- 10 Year Aging Stability of <±4E-7
- Input Voltage Stability of <±5E-10 for a ±2% Voltage Change
- Short Term Stability of <±1E-10 per second
- Load Stability of <±1E-9 for a ±5% Load Change
- Phase Noise @10Hz, <-105dBc; @100Hz, <-125dBc; @1kHz, <-145dBc
Pinout
- Pin 1: Ground
- Pin 2: EFC Voltage (0-5)
- Pin 3: ??? (See Below)
- Pin 4: +24V
- Pin 5: ??? (See Below)
- Pin 6: 10MHz Output
And here is where I began to run into conflicting information:
- Heater Voltage (Pin 5) of either +15V or +24V
- What does Pin 3 output? Is it a VRef? A TTL level oven monitor?
- Is the Output (Pin 6) 2Vp-p as listed in the spec, or 4Vp-p?
- What is the EFC Sensitivity in Hertz per Volt? Is it linear?
- What is the overall draw on the 24V supply?
And so, not being one to let others do the heavy lifting, I broke out the test gear and started measuring…
Pin 5, Heater Voltage and Pin 3 Output
So, is it +24V or +15V? I breadboarded a 7815 and a 7805 – for the
heater and EFC, respectively – and brought the +15v to Pin 5. I routed
Pin 3′s output through a 50k precision 10-turn potentiometer as a voltage divider to apply a variable voltage to the EFC pin, Pin 2.
Upon
applying 24 volts, the oscillator began operation and Pin 3 showed
5.12V. So far so good; the pot dividing the pin 3 voltage was adjusted
to put 2.5V to the EFC pin and the oven was left to settle. About five
minutes in, the frequency began to drop very quickly and I discovered
that the EFC voltage had inexplicably dropped to 200mV in an instant.
Pin 3 measured <0.5V at this time. Hmmm…
In an attempt to lower the time spent on this endeavor, I routed Pin 3 to the board and wired in an LED and resistor to act as an indicator. I also brought the 7805 output to the voltage divider pot and rebalanced it to 2.500V. After allowing the oven to cool slightly, I reapplied power and the Pin 3 LED immediately illuminated, only to fully extinguish less than a minute later. So. Pin 3 is in fact a TTL level oven indicator and NOT a voltage reference.
Timing the oven runtime on a coldstart, the 15V input is fully sufficient to get the oven up to temperature and keep it there indefinitely. As much as I would like to, I won’t be simplifying my power supply and running it on 24V. I do, however, wonder about that Pin 3. After the initial warmup, it does not go high again. Ever. But I know that oven has to be running intermittently to maintain the temp. Curious.
Pin 6, Output Level
Pin 6 does indeed output 2Vp-p
Supply Draw
At 24.0V, the OCXO draws 280mA during warmup and 170mA during constant operation. Keep in mind this is with the 15V and 5V regulators taking their source from the 24V supply as well, so there’s a reasonable bit of that wasted as heat.
EFC Sensitivity: A Primer
Ah, the real meat-and-potatoes of this investigation. A critical piece of information for the disciplining of a VCXO is the sensitivity of the EFC input. The units are in Hertz per Volt (which does not need to reduce to (s^2*A)/(m^2*kg) for those of you obsessed with dimensional analysis…).
A good example to work with are the HP 10811A and 10544A models, both of which have an ‘S’ (Sensitivity) of 1E-8 per volt. What does this mean? Well, the value is referenced to the output of the oscillator (10MHz or 1E7 Hertz), so in plain language, a change of one volt on the EFC equates to a change of 0.1 Hz in the output frequency.
The range of EFC input on the HP units is -5V to +5V, thus the control range is (5-(-5))*(1E7*1E-8)=1, or simply one Hertz. Assuming you are using a DAC to drive the EFC, you can then use the number of DAC input bits to determine to smallest possible change in voltage and, by correlation, the smallest possible change in output frequency.
Continuing to work with the HP units and assuming an 18-bit DAC (which is what I’m using for this project), the smallest possible change in EFC voltage would be equal to 10/(2^18-1) volts, or 3.815E-5, or 38.15μV. We can then apply this voltage to the EFC and net a change in output frequency of 0.1*3.815E-5= 3.518 μHz. Referencing this back to the scale of the oscillator, a one bit change in the DAC output equates to a 3.815E-13 change in output.
Why does this all matter? Well, the method of oscillator disciplining will determine a ‘floor’ beyond which the method cannot have a higher stability. In this particular case, a GPSDO PLL, the stability floor is in the neighborhood of 5E-12 within a reasonably short (1E5 to 1E6 seconds) sample period (as a datapoint that’s five parts per trillion, which would be the scalar equivalent of measuring the circumference of the Earth in inches (a little over 1.577 billion) to within eight thousandths (±0.004″)). We need the ‘step’ size – 3.815E-13 in the above example – to be far smaller than that to effectively maximize the utilization of the disciplined stability. If the step size were exactly the same as the minimum stability, the system would continuously oscillate around the stability floor of the disciplining circuit, wasting all your hard work in making such a stable reference.
If you find yourself intrigued by the concepts of precision timing and wish to know more about accuracy, precision, stability and timing in general, check out Tom Van Baak’s awesome website, LeapSecond.com
You can also search for “Allan Variance” to investigate the measurement of stable timing sources; I have glossed over many of the details in this writeup and it really is a fascinating study.
EFC Sensitivity: Piezo 2940210 OCXO
As you can see below, my initial results showed a sensitivity of 4.6 Hz/Volt (seen as the leading coefficient of the linear approximation), or 4.6E-7 with reference to the output frequency. That and the aging of the unit has driven the center frequency of the EFC to about 4.34V (it should be ~2.5V, the center of the range) which will require ‘trimming’: an adjustment to the crystal oscillator circuit to change the output frequency without changing the EFC voltage.
The OCXO has presumably been sitting, disconnected for a long while. As such, this value is prone to drift (currently at around 0.25 Hz/Hour) until the crystal “aging” settles down. Until that time, the OCXO cannot be trimmed effectively for constant operation. However, this is the lesser of two potential issues…
If the full range of 0V to 5V is made available to the disciplining circuit, the scale of the sensitivity of the OCXO becomes problematic. From the above equations, the 1-bit change in output frequency is a whopping 4.6*((5-0)/2^18-1), or 87.74 μHz. As compared to a 10811A, that’s twenty five times larger! Using a reference to the output frequency, that puts the sensitivity at 8.774E-12 (!!!) which is far beyond what we need it to be if we’re bothering to develop a control system with a stability floor of 5E-12.
So what’s to be done? There are two obvious answers, assuming you don’t want to buy a new crystal oven. One is free(ish) and one is very much not free. The latter answer is buy a better (read: more bits) DAC. Easy, right? Spend anywhere from $2 to $20 and be done with it. Right?
Wrong. More bits = new chip = more i/o = more expensive control system = new microcontroller = new program. So the standard application of MOAR in this case is not a good thing… In fact it’s so bad, it means a ground-up redesign of the system, which is a major wombat in my mind. Rather, let’s consider a change in control schema…
The free(ish) option is to change the range of the EFC control voltage. Just because the OCXO can take 0V to 5V doesn’t mean we have to give it that. To investigate this, we turn the above equations ‘inside out’ and solve for the range, rather than the step size. So what if we wanted the 3.518 μHz step size of the HP oven? At 4.6 Hz/volt this would equate to an EFC voltage change of 3.518E-6/4.6 = 7.648E-7 volts, or 764.8 nV. Applying the same equations again, the full scale of the EFC is almost exactly 0.200 volts. Centering this around 2.5 volts nets a range of 2.4 to 2.6 volts. See? Easy.
I’ll hopefully be expanding on this as the GPSDO controller portion of this project develops. Thanks for the read!