Oven Controlled Crystal Oscillator

    Precision Frequency Measurement: What would you do if you were stuck in an underground bunker with no GPS signal and your life depended on making a precise frequency measurement? An improbable scenario, perhaps, but as the Boy Scout motto says, “Be prepared!”

    In previous project write-ups I described a GPS-disciplined frequency counter and, subsequently, a rebuild using surface mount components. At the time of carrying out these projects, and writing about them, I was only dimly aware of something called an ‘oven-controlled crystal oscillator’ (OCXO). Some time later I read about OCXOs, and their applications.¹  To appreciate the potential usefulness of OCXOs, the keyword is stability. Once fully heated the OCXO shown in the photo above produced a nearly constant output frequency, certainly less variable than would be expected from an ordinary crystal oscillator in a typical application environment.


    When first powered-on, the OSC5A2B02 draws about half an amp at 5 volts. As its temperature increases, current drain gradually decreases, settling to about 200 mA. Although I had thought when purchasing a couple of used (reclaimed) OCXOs on eBay, that their frequency should be adjustable—within a limited range—I did not immediately see how to do this. However, I realized that from the standpoint of deriving a time interval for counting a test signal, the exact OCXO frequency doesn’t matter, as long as it does not change appreciably.

    Sometimes I do not see the thing that is staring in my face. —In this case it was the pin labeled Vref in the photo, and referred to as Vc on the OSC5A2B02 datasheet. Biasing this pin with 2 volts ± 2 volts corresponds to a tuning range of -1 ppm to +2 ppm (i.e., -10 to +20 Hz at 10 MHz). By the time I got around to experimenting with the Vref pin I had invested a great deal of time in other investigations, some of which were of potential use, others dead-ends. (I will return to this point.) In any case, as the photos accompanying this paragraph illustrate, by adjusting the bias to approximately 1.8 volts it was possible to tune the heated oscillator to a frequency of 10 MHz exactly. Furthermore, with the OCXO temperature stabilized at approximately 200 mA current draw, its output frequency remained constant for a considerable time afterwards.

    While the GPS pulse-per-second (pps) is accurate to better than a billionth of a second, the temperature stability of the OCXO type that I purchased is claimed ≤ ± 10 parts per billion (ppb) and, from direct observation, as noted above, the OCXO frequency remains very nearly constant (within 1/4 hertz) for a long time after it has fully heated. To apply the OCXO’s frequency stability to counting or measuring a test frequency, an obvious approach would be to derive a timing interval from it, analogous to the GPS pulse per second. Clearly, a one-second pulse could be obtained by dividing the OCXO frequency by ten million (10⁷). The quotient would be a time interval that is extremely close to 1 second, though not as precise as the GPS. Thus, it would be possible to modify and extend the GPS-disciplined frequency counter to use an OCXO time base if needed.

    Dividing the crystal frequency by 10⁷ seemed straightforward, but involved a small quirk. I had purchased type CD4017BE decade counters with the idea of using seven of them to divide the frequency from 10 MHz to 1 Hz. However, the specified maximum clock input frequency for these ICs was 2.5 MHz at Vdd = 5 volts. Through testing I ascertained that the 4017s counted or divided correctly up to an input frequency greater than 10 MHz. Even so, it would surely be better to adhere to the published specification, which meant that faster devices should be used for the first divide-by-10 stage. Therefore, for the first stage of division (i.e., from 10 MHz to 1 MHz) I experimented with a couple of counting circuits from the Internet before running across what seemed a superior solution in a paper titled Clock Dividers Made Easy, by Mohit Arora of ST Microelectronics Ltd. At the top of page 12 is a divide-by-5 circuit that uses three ‘D’ flip-flops, augmented by two AND gates, one OR gate, and four NOTs (inverters). When constructed on a breadboard, this circuit, adapted to match ICs that were on hand, performed as promised (my drawing below).
 

    The divide-by-5 circuit consumes three D-flip flops, or one and one-half TTL type 7474 ICs, leaving one flip-flop available for the divide-by-2 part (bottom of drawing). Before adding this first stage I had already soldered six CD4017 IC sockets and three female headers on a prototype circuit board, and had left no room for anything else, intending that the first divide-by-10 stage would occupy a daughter board. But then I decided to try to squeeze the five additional ICs onto the same board. Somewhat to my surprise, they were able to fit, though just barely.²


    To make room for the divider board, the frequency counter enclosure also required a redesign. The revised enclosure consisted of three levels and a top panel. The bottom level housed the original SMD frequency counter board. The divider substrate (above) was the second level. The third level (annotated below) had the binary LEDs panel on the front. The power jack, SMA jack (signal to be measured), a 4-pin female DIN connector (for the GPS or OCXO plug-in), as well as a small reset button and SPST toggle switch are on the back. For this version I did not bring out connections for the input signal pre-amplifier. If needed, an external preamplifier could be used. In the photo below the top panel has a rotary encoder in place of the original GPS frequency counter’s normally closed pushbutton. I will explain what that is about shortly.

 
    In the annotated photo above, the GPS or OCXO toggle switch would more accurately be labeled ‘10 MHz - 1 Hz’ because the alternative time base is not necessarily an OCXO, but rather an approximately 10 MHz stable clock. Similarly the pulse-per-second signal is not necessarily from a GPS. As was previously mentioned, prior to figuring out how to tune the OCXO to exactly 10 MHz, I had experimented with stable frequencies in the vicinity of 10 MHz, using both the purchased OCXOs and one that I had made using resistors to heat a standard 10 MHz crystal. To facilitate this part of the exploration I modified the microcontroller code, first to include a pre-programmed calibration constant, and subsequently to permit runtime calibration. That is why the top panel has a rotary encoder instead of just an SPST pushbutton.

    Practical constraints: The original GPS-disciplined frequency counter, with its decorative binary LED display (photo right), consumed all ATmega328P digital and analog I/O pins, except digital pins 0 and 1, which were reserved for the serial interface. The ‘Request measurement’ pushbutton was a normally closed type, because that is what happened to be handy when I constructed the prototype. I puzzled over how the design could be extended to allow substituting a rotary encoder in order to permit runtime calibration adjustment of an imprecise time base. If a rotary encoder were to be used, its integral pushbutton switch would be of the normally open type. And which MPU pins could be assigned to the clock and data signals? It seemed there was no choice but to use MPU pins 0 and 1 for the latter. In truth, there was the choice of eliminating the decorative binary-remainder LEDs, but I like flashing LEDs.

    Solution: The table below shows how a single pushbutton was adapted to perform multiple functions associated with incorporating a switchable OCXO time base. The trick was to distinguish between a short and long press. In order to accommodate the normally-open rotary encoder button, I used a spare TTL inverter on the ‘divider’ board, and left the button sensing part of the MPU program unmodified.³


    Software: As previously remarked, the justification for relying on a stable crystal oscillator at a known frequency near 10 MHz is that the exact value of the counting interval can be easily computed. It is the oscillator’s actual frequency in Hz divided by 10⁷. This fact makes it possible to correct a test count and to obtain a very good frequency measurement. Thus the microcontroller code was modified to accommodate editing a custom (imprecise) time base ‘constant’, and to store the correction in EEPROM. In GPS mode, there is no correction to the counted frequency, while in custom time base mode the 4-second count gets adjusted for the oscillator’s true frequency. Implementing this correction was tricky, as 4-second count numbers are big, and Arduino C++ does not like big numbers. Code for correcting the raw count (below) was a bit of a guess, but it appears to work. The complete revised sketch, including OCXO support, may be examined or downloaded here.


    Kitchen Clock: That is my nickname for a clock that is based on a 32,768 Hz crystal (2¹⁵ Hz). At such a low frequency the derived one-second interval cannot be very exact. On the other hand it would be fun to calibrate such a clock and see how good a frequency measurement can be obtained from it. For this exercise, I first repeated a measurement of the 10 MHz OCXO using the GPS time base. It was 10 MHz exactly (stays there for hours).  Next I unplugged the GPS and connected the output of the kitchen clock in its place. Once again, my prediction bit the dust. The measured frequency of the OCXO was 10,000,006 Hz, just 6 Hz high. At this point, just for fun, I reset the frequency counter to the ‘custom time base’ startup mode, and calibrated the kitchen clock time base. After a couple of tweaks, the OCXO measured 10 MHz exactly.⁴ Of course, the kitchen clock will drift—I say it will—I am sure it will. Hmm.

   Afterthoughts: The more useful part of this project was the alternative measurement time base, derived from a 10 MHz OCXO by frequency division, and not dependent on a GPS. This was simply a hardware addition (OCXO plus divide board and switch). The time base calibration part of the project is likely of less interest in the present context. OCXOs are not expensive or difficult to obtain. The couple that I purchased cost less than $10 (for both). Nevertheless, adding a calibration feature, without changing the basic measurement interface, was an instructive exercise.

    As I thought about it, switching the OCXO power supply with a relay was unnecessary. The OCXO should remain powered-on and fully heated when taking measurements. Also, for fine-tuning the oscillator to exactly 10 MHz, a multi-turn potentiometer is easier to adjust than the trimmer that was used with the prototype. So, for the other OCXO I cut a smaller board—about 1/3 the size of the original, and used a multi-turn 100K potentiometer. In the vicinity of the 10 MHz target frequency, each full turn corresponds to approximately 2 Hz increase or decrease.

    Evidently it is possible for an LCD backlight to burn out. This has never happened to me, and I think it must be a rare occurrence. Even so, it bothered me that the LCD remained illuminated for hours on end, or overnight, while experimenting with the OCXO time base. To add an inactivity timeout was a simple software change. The timeout value (INACTIVITY_TIMEOUT) is specified in milliseconds. For example 10 minutes is 600,000 milliseconds. Pressing the button during timeout turns the backlight on and has no other effect. To schedule a measurement, or any other function, it is necessary to press the button again after the display has been illuminated. This software-only revision (version 3.2) may be examined or downloaded here.

    It is good to have options. Be prepared!

    Demo: OCXO.mp4

   Breaking news: [July 2024] The eleven-chip 10 MHz to 1 Hz divider described in the preceding paragraphs can be replaced by one tiny 8-pin IC. For details see:  https://www.lloydm.net/Demos/ocxoFC.html.




1. For example, I learned about this instructive project by DL2KHP, which incorporated both GPS and OCXO, from Email correspondence with another ham radio operator.

2. I exercised this attempt on paper before soldering additional sockets on the populated circuit board! The drawing below is rotated 180° with respect to the circuit board photo in the main text.


4. This exercise is reproduced in the video demo with similar results.

---

Project descriptions on this page are intended for entertainment only. The author makes no claim as to the accuracy or completeness of the information presented. In no event will the author be liable for any damages, lost effort, inability to carry out a similar project, or to reproduce a claimed result, or anything else relating to a decision to use the information on this page.