Continue to Site

# 32Hz stable discrete square wave oscillator, how to?

Status
Not open for further replies.

#### neazoi

Hi,
My application requires a about 32Hz reference oscillator (an OCXO accuracy will do), preferably square wave, but not essential.

My question is, how to derive this frequency using the less amount of components, but only using discrete components, not ICs at all (that is a requirement).

Some ideas:

1. Start with a watch crystal at 32.768kHz because these will require least stages of division. The division stages can also be discrete transistor flip flops. You will need 10 divide-by-2 stages to get from 32.878kHz down to 32Hz.

2. Another idea, to divide by a factor greater than 2 in one single stage, would be transistor phantastron divider circuits (transistor equivalent of an old valve/tube circuit). You might then only need 3 divide-by-10 circuits, instead of 10 divide-by-2 circuits.
But I cannot find any such circuits or how do they work.

3. Start off with TWO 32.768kHz oscillators and add capacitance so that one frequency is offset from the other by 32Hz. Then mix them together and select the difference frequency. I think the two-oscillators idea would have less components (just mixers, not flip-flops) but also less stability with temperature. The ppm drift of the crystals is multiplied by 1024, compared to the case where the frequency of a single crystal is divided. But if an oven is used for both oscillators, should the frequency be stable enough?

4. Any other ideas...?

Last edited:

My question is, how to derive this frequency using the less amount of components, but only using discrete components, not ICs at all (that is a requirement).

Why do you try to turn back the clock and create a complicated solution, instead of using one simple cheap IC?

1. Start with a watch crystal at 32.768kHz because these will require least stages of division. The division stages can also be discrete transistor flip flops. You will need 10 divide-by-2 stages to get from 32.878kHz down to 32Hz.

One CD4020 binary counter IC for 0,30€ will do binary division up to 14 stages.

Without ICs I would suggest the two crystal approach but make sure the entire circuit is kept at the same temperature to minimize drift. I suggest two buffered oscillators and a mixer, if you try mixing the two oscillator outputs directly you will probably find they injection lock to each other. At 32Hz difference you can probably pull watch crystals far enough but there is no reason why higher frequencies can't be used.

Brian.

neazoi

### neazoi

Points: 2
Without ICs I would suggest the two crystal approach but make sure the entire circuit is kept at the same temperature to minimize drift. I suggest two buffered oscillators and a mixer, if you try mixing the two oscillator outputs directly you will probably find they injection lock to each other. At 32Hz difference you can probably pull watch crystals far enough but there is no reason why higher frequencies can't be used.

Brian.

I suspected that the mixing is far simpler, despite I have seen some neat designs like the one in the second page of the attached pdf, but the output frequency depends somehow on the capacitor charge/discharge curve and the buffer threshold and it is not guaranteed to be stable. And thanks about the buffering of the oscillators to prevent locking, this is very valid at such low frequency difference indeed.

Back in the mixing solution, initially, I has thinking of ovenizing the two oscillators, to keep their temperature stable, to minimize drift.

Later on, I thought that the crystal oscillators may not be ovenized if these were made matched. If they can be matched, any temperature differences will be seen by both oscillators and the output frequency of the mixing would be stable independent of temperature variations.

Since matching could be proven tricky, later on, I wondered if a "weird" dual oscillator design exists, in which both oscillators use the same resonator, be it crystal or ceramic, with the ability to vary one of the two oscillator's frequency by an external L or C. Then any temperature variations on the crystal would automatically affect both oscillators and no output drift would be produced on the mixer. Does any such design exists?

#### Attachments

• dividers.pdf
34.2 KB · Views: 106

I suspected that the mixing is far simpler, despite I have seen some neat designs like the one in the second page of the attached pdf, but the output frequency depends somehow on the capacitor charge/discharge curve and the buffer threshold and it is not guaranteed to be stable.

... and you will get rather bad jitter/phase noise behaviour. What a waste of signal quality that you get from the initial OCXO!

Why don't you go for the digital divider solution that gives a very clean and stable output signal? This is what the professionals do when they want the best possible signal quality.

Why don't you go for the digital divider solution that gives a very clean and stable output signal? This is what the professionals do when they want the best possible signal quality.

Thanks but as I said I would like to find out how it can be satisfactorily made discrete

Hi,

maybe an 8 pin microcontroller is the smallest solution.
XTAL, some Cs, software only to initalize timer/counter for waveform generation.

After initialisation you could send the controller into sleep mode. No processing power needed.

Klaus

To get OCXO accuracy, you need at least an OCXO and a frequency divider that doesn't increase the drift and jitter.

So if you rely on a discrete solution (why, is it a competition?) make the transistor FF frequency divider. Other divider circuits like synchronized relaxation oscillator are possible but need adjustment and probably add jitter.

Everything else, e.g. the mixer idea, is missing the stated accuracy by orders of magnitude.

neazoi

### neazoi

Points: 2
To get OCXO accuracy, you need at least an OCXO and a frequency divider that doesn't increase the drift and jitter.
Everything else, e.g. the mixer idea, is missing the stated accuracy by orders of magnitude.

But why two oscillators placed inside the same oven and mixed to create a 32Hz signal, would not work satisfactorily?
Even more relaxed, if the oscillators components are quite matched this can lead to reasonable stabilitywithout even an oven.

I know I said about oven stability in my post #1, but a 2-3Hz stability will suffice, so the thing can be relaxed a bit.
Trying to implement FF dividers using discrete components will lead to a very large circuit, but simple mixing could potentially do an acceptable job with much fewer components.

So if you rely on a discrete solution (why, is it a competition?)

That's what I would also like to understand. Why discrete if integrated divider can give better performance? What is the benefit of using separately packaged transistors if we can have multiple closely matched transistors in one package?

I know I said about oven stability in my post #1, but a 2-3Hz stability will suffice, so the thing can be relaxed a bit.

I read OXCO accuracy as better than 1 ppm, usually better than 0.1 ppm. 2-3 Hz is worlds apart, at least factor 10000 worse.

2-3 Hz can be easily achieved with a LC or RC oscillator.

volker@muehlhaus

### volker@muehlhaus

Points: 2
For minimum component count and still good accuracy Klaus's suggestion of an 8-bit micro is best. I'm thinking of something simple like a PIC with on-board DAC or better still, an external resistor ladder DAC. It will produce a pretty good 32Hz sine wave with far better accuracy than 2-3Hz even using it's internal oscillator and with a crystal is will be better than 1Hz. One IC, two capacitors and eight resistors is all you need to produce a 256 step sine wave.

If a square wave is all you need you can use something like a PIC10F200 and one capacitor which together costs less than a watch crystal.

Brian.

I would take a 1 MHz OCXO, divide it by 31250 in an HMC983 divider. if you need a square wave out, just divide by 15625 and then follow up with a flip flop.

OR I would just use a DDS.

Hi,
My application requires a about 32Hz reference oscillator (an OCXO accuracy will do), preferably square wave, but not essential.

My question is, how to derive this frequency using the less amount of components, but only using discrete components, not ICs at all (that is a requirement).
O/k integrated circuits are totally out for some reason ?
Least components....

How about a very simple LC oscillator.
Big pot core, thousands of turns, several Henries very easy to do.
A decent fairly stable sine wave should not be too hard

O/k integrated circuits are totally out for some reason ?
Least components....

How about a very simple LC oscillator.
Big pot core, thousands of turns, several Henries very easy to do.
A decent fairly stable sine wave should not be too hard

There is not always a reason for doing something, but how you could do it is interesting.
Going discrete explores the possibilities of how could this be done that way, leading to many learning procedures.
I am aware of all these solutions with dividers and micros and these are much better. At 32Hz a modern micro could produce a very good sinewave.

The LC does not seem to be a reference oscillator at first, but at 32Hz I suspect even an LC could be fairly stable, maybe.
What if I use the sound blaster? will that be stable enough? Of course this requires a PC.
Other ways? After all, it is an audio frequency.
Maybe some kind of bridge circuit with auto correcting mechanism based on the imbalance of the bridge?

Just to put some very rough numbers on it, ten Henries and about 2.5uF will get you close to 32Hz.
Both are very practical values, and a simple JFET oscillator would probably work out fairtly well.

If you were dead keen, adding a small proportion of negative temperature coefficient ceramic capacitors might get you a quite a respectable drift versus temperature characteristic, over a reasonable ambient range.

Not exactly equal to a caesium time reference standard, but perhaps not too bad if you go to a bit of trouble to build and test it properly set it all up.....

Last edited:
neazoi

### neazoi

Points: 2
Just to put some very rough numbers on it, ten Henries and 2.5uF will get you close to 32Hz.
Both are very practical values, and a simple JFET oscillator would probably work out pretty well.

If you were dead keen, adding a suitable small proportion of NPO ceramic capacitors might get you a quite a respectable drift versus temperature characteristic over a reasonable ambient range.

Not exactly equal to a caesium time reference standard, but perhaps not too bad if you go to a bit of trouble to build and test it properly set it all up.....

And you can ovenize the whole thing? How good will that be in a LC?

If you are going to that amount of trouble, a conventional quartz crystal and a digital divider might be the way to go.

Another possibility for an ultra simple 32 Hz oscillator might be a tuning fork oscillator. Made of Invar that would have zero temperature coefficient.

neazoi

### neazoi

Points: 2
If you are going to that amount of trouble, a conventional quartz crystal and a digital divider might be the way to go.

Another possibility for an ultra simple 32 Hz oscillator might be a tuning fork oscillator. Made of Invar that would have zero temperature coefficient.

A simple proportional oven is not that much trouble when compared to the complexity of the internals of a digital divider. Try to implement a digital divider out of transistors and you know what I mean.
If an ovenized LC leads to a stable output frequency, that would be a very simple solution. But I am not aware how effective would be the temperature stability on the LC, or if the LC stability would be affected as well by other factors apart from the temperature.

A tuning fork electronic oscillator? That is new to me, and although it would be much more complicated, thank you for sharing such a wonderful information!

Should be quite possible.
A bit of mucking around, but posssible.

Design the LC oscillator to be suitably temperature compensated for a reasonably wide range either side of the highest expected ambient temperature.
Then oven control it up to that temperature.
Even if the oven cycles back and forth over a few degrees, frequency should still be pretty stable.

From memory, oven controlled crystals operate around 45C.
That is pretty warm, but should be above any reasonable ambient.
Should not need a lot of power or insulation to keep warm.

neazoi

Points: 2