Interesting TCXO (32768 Hz) oscillator

I recently came up with this unusual design for a 32768 Hz crystal oscillator.

It uses the reverse breakdown of an NPN transistor with a timing cap and resistor as an oscillator. I found that using a crystal as shown, I can lock the frequency quite accurately.

The two green LED's act as some temperature compensation. I also tried red ones, but the compensation was not adequate. When tested in a lab oven from 0-40° C it only drifted about 0.4 Hz. The crystal was nothing special, like $1 for a bag of 10 eBay special. Not all common NPN types work in this circuit. I tested a MPSA18 low-noise audio transistor, with no success. Looking at the breakdown curve on a curve tracer also showed no noticeable negative-resistance region.

I constructed the first three stages on an IC socket, but later cut out the NPN to try other types. That is the device hanging on the side in the picture.

The counter was locked to the lab Rb. GPS frequency standard during the tests. I am not sure if any type of these crystals will work in the circuit. More tests needed.

The 1 sec jitter also averaged under 100 pS as shown.

P1 is a 10 k pot.

Really nice circuit. I can't say that I fully understand it.

One question though: the initial accuracy is provided by the 32.768Khz crystal, correct?
The remaining circuitry is for stabilization??


What brand is your counter?

The value of C1 was selected so that free-running oscillation is near the wanted frequency. The frequency is locked in by the crystal, so the quality of the crystal will have an effect. The rest of the circuitry on the left is for temperature compensation. The frequency drifts low with an increase in temperature. The tempco of the LED's manage to cancel that out quite effectively.

I still plan on measuring the actual crystal current and power dissipation when time permits.

The counter is a Stanford Research Systems - SR620. Great counter.

Watch crystals have typically a quadratic frequency error characteristic with turn-over point around 25 °C. It's not obvious at first sight how the circuit would generate a quadratic compensation function.

FvM, yes it needs more testing to understand fully how the compensation seems to work that well. I must mention that the total drift over the temperature range included negative and positive errors to make up the 0.4 Hz result. I think that the behavior of the transistor breakdown with temperature also introduces a variable that may be responsible for the results.