For crystal oscillators that are pulled, setting the initial condition of the tune voltage for minimum capacitance(max neg gm) will give speed up. Other techniques we have used with success in the past include ALC (amplitude level control) where a larger current is used during the initial startup which is reduced as swing increases.

The answer  probably depends on your jitter tolerance to start with, and continuing onward...after whatever "speed up" method settles out.
The xtal Q limits your startup time and settling to a given jitter level.
Any kicks or AGC loop must settle before jitter is constant, or you might have a momentarily large amount of jitter.

You may consider a ring oscillator  to jump start some non-critical circuits and switch over to a Crystal after a given count.

Wave

search "quick start oscillator" on google scholar and you will get a few hits. The one I was thinking of is by Blanchard. He uses an oscillator that is trimmed to about the correct frequency to drive the crystal itself until it was at the right amplitude. I was never crazy about this idea, but I think AMI (now ON semi) still uses it.

I have removed the tuning capacitor to minimize the startup time, but I had an issue. I think it was that the oscillator was sensitive to noise which was causing glitches and double edges which was causing my PLL to go ape which prevented the micro from loading in the correct trim. This did not impress the customer! But it was a cool idea.

It may have been something else on top of that... that was 7-8 years ago.


rg

Hi,

My simulation experience proves that the kick-start oscillator is a failed idea. It takes almost the same time as that without kick to start the crystal oscillator in SPICE simulations.

I guess it is due to the status/current through Ls, the inductor in the equivalent circuit of a crystal, is not near the expected value. In other word, the current through Ls is probably near zero at the imediate moment that the kick is stopped. If the kick is stopped when the current throught Ls is near its maximum value, perhaps the crystal oscillator will start faster.


Best regards,
Yawei

loose-electron wrote on Jun 13^th, 2009, 10:24am:


Interesting. It never seemed like a real robust solution so I believe what you are saying.

Referring back to my comment about the problems I had when reducing the capacitance... I had removed all of the cap which allowed the parasitics to couple in noise. If you leave a small amount there you should be fine. So that seems like an obvious.

As you probably know, Vittoz's paper has a dynamic current source to start the osc up quickly. I think that paper also covers the start-up time issue w/ Q. There is another paper covering start-up requirements (Berkeley or Stanford?) which shows that it won't start-up if Q is too low. Maybe that is why lowering Q isn't done.

All these solutions have the issue of recovery from the transient when you revert to the steady state configuration. Good luck!

loose-electron wrote on Jun 18^th, 2009, 6:47pm:


Hi Jerry,

While this sounds like a good way of atleast speeding things up a bit, won't it also suffer from the same fundamental issue that Yawei already pointed out? I imagine that the amplitude of the signal coming out of the crystal won't really change that fast as you sweep the slope of the VCO control voltage.

Basically, the high Q of the crystal means that it does not lose much energy quickly (or gain much energy), i.e. the crystal has enormous inertia. Whatever you do, it takes an almighty effort to force an amplitude change. I don't see how you can get around that.

In my opinion, you can only speed up things in the sense that you avoid using suboptimal methods which themselves would slow down the process of the crystal reaching its steady-state. So, using the amplitude regulation schemes like in Vittoz's paper may be a way of speeding things.

Best regards,

Vivek

Jerry,

Assume the kick-start works if the you find the "kick frequency", it would have been very complex to implement such an architecture. For a high Q crystal oscillator, the bandwidth is very narrow, say 100 Hz. The VCO tuning range must cover PTV variation. At the same time, the frequency sweeping must be slow to find the accurate "kick frequency". Perhaps it takes longer time to find the "kick frequency".

BTW, what's the spec for the start-up time?


Best regards,
Yawei