Question about 32.768KHz Oscillators........

[Pld4me]

1 hz = 32.768 khz * 2^15

While I understand that they are divisible by powers of 2, I find it hard to believe that they are used exclusively to divide down to 1Hz (2^15).

I would think that with inherant silicon delays in whatever you use to divide the signal down to 1Hz that it would be wiser to just use a 1Hz oscillator.

Can someone shed some light on this for me?

 

[unkarc]

divide frequency crystal 32768hz to 1hz

The problem is to design a 1Hz oscillator with the stability you can get from an average 32.768kHz oscillator that is divided to 1Hz.
With the divison ratio applied you can actually increase the stability of the needed 1Hz signal by as much as the divison ratio without using very precise oscillator hardware for a 1Hz oscillator.
Say the 32.768kHz oscillator changes its frequency by 10Hz (this happens easily by ambient temperature change) then the divison to 1Hz 'corrects' this to a very small change (let me not calculate it) and think this over to a 1Hz oscillator.

unkarc

 

[Pld4me]

Excellent explanation unkarc!

This has been puzzling me for awhile, and I could not find a reasonable explanation for it.

Best regards,

Pld4me

 

[Kral]

Pld4me,
32,768 Hz is a compromise between physical size and drive requirements. A 1 Hz crystal would be huge, and very difficult to drive.
Regards,
Kral

 

[jasmin_123]

It seems to me, Kral, that a huge 1-Hz crystal would also have a smaller Q and this would reduce the oscillator frequency stability.

 

[JohnJ]

Think of it this way....you are averaging any short-term errors (i.e. jitter) over 32768 clocks. Any repetative errors tend to go away.

Also, at this point-in-time, since just about everyone uses this value, the volume the part is made in tends to improve it's quality.

Before 32kHz, there was (in the US) the 3.579545MHz US TV color crystal. The sheer volume this crystal was made in (every US TV), the better and more accurate they made it at the same price. There are still time-clock chips around using that frequency since it was the first (that I know of) crystal with any reasonable accuracy made for < $1.

I seem to remember the first sub-1MHz crystals also being huge. The 132kHz crystal we used one place I worked was bigger than 1 inch long. Now they put them in our $1 wrist watches (my 2 year old $5 watch hasn't needed to be set since I bought it.)

 

[jasmin_123]

Jitter is a feature of logic gates, not the oscillator. Make an analog oscillator (not Pierce)
and you will have no jitter, only phase noise (if you stabilize the amplitude).
At 1 Hz, phase noise may even be smaller than at 32 kHz. However, at 1 Hz,
the flicker noise of the amplifier and/or the zero-detector can ruin the stability
of frequency.

PS: $50.000 double-crystal watches can have a 0.5 s/year accuracy.