Startable Oscillator

You need to tell us more about your requirements. A few GHz of what, fixed or adjustable? Standard like 2.5GHz or nonstandard? Even 1% accuracy needs careful plannings and adjustments.

The frequency is fixed, but the actual value is not decided yet. (depends on the circuit driven by the oscillator) Likely candidates are:

Beeing able to slightly tune the frequency for calibration might be beneficial but is probably not necessary.

Kolja

If all you need is 12 cycles or so, I'd definitely be looking at an LC. Set the comparator threshold at zero, so the frequency and duty cycle don't wander around as the ringing decays, and use John's approach with a resistor in series with the output--that way the transistor hoists the output to a reliable logic level in the quiescent state.

Cheers

Phil Hobbs

On a sunny day (Tue, 8 Sep 2009 07:13:18 -0700 (PDT)) it happened Kolja wrote in :

You could ping a lower frequency crystal, and add a couple of frequency multipliers. With frequency multipliers I mean transistors biased to work in the non-liner region with a LC in the collector tuned to 3 x f or so. A 58.64197531 crystal, followed by 4 3x multipliers would give you 4.57 GHz. Takes about 5 transistors and 4 tuned circuits, and a specially ordered xtal per frequency Could solder it together in an hour or 2.

OK, that's not real hard.

A coax ceramic resonator in the low GHz range can have a Q in the thousands, so will ring for a long time. One could drive it from a fast ECL gate through a resistor, and pipe the resulting ring into a fast comparator to make LVDS. Startup prop delay (trigger in to first edge out) would be way below 1 ns. The resonators unfortunately are very low-impedance devices, acting like, typically, 10 ohm shorted transmission lines, so the resulting circuits tend to be power hogs.

Contact me by email (jjlarkin/atsign/highlandtechnology/dotcom) and we'll see if we can help.

John

In all this discussion -- why not just keep the oscillator running, and gate the output? Your signal will come one quickly and with predictable delay (and start up envelope, if that matters), and you don't have to mess with designing oddball oscillators at difficult frequencies.

I have, on occasion, seen requirements for the first half-cycle of a gated oscillator to be accurately 50%.

I have schemes that I use for RFID tags, but they're not as fast as the OP wants.

Actually, Larkin's "pre-charged" LC (or coaxial resonator) scheme sounds quite promising.

...Jim Thompson

All true except for the last statement. Free running or not, he might still need multiple oddball oscillators. But they are easier to deal with.

I'm not arguing that my way will solve all of life's problems -- just pointing out that he may be giving himself unnecessary grief by not thinking through his requirements.

(Of course, if he uses a gated oscillator and it's the Wrong Idea then that won't be at all good either).

multipliers.

region

per frequency

That would take forever to start up, both in the XO and the multipliers, and it would be very difficult to get a clean, phase-coherent clock out.

I think what he wants is...

______________________________ | trig ________|

___ ___ ___ | | | | | | out _________| |___| |___| |___ etc

with a minimum delay from trig to the first clock edge.

John

On a sunny day (Tue, 08 Sep 2009 12:00:25 -0700) it happened John Larkin wrote in :

multipliers.

region

per frequency

Yes, but as Tim? mentioned perhaps he could keep the osc running.

My MVB idea... too high a freq I think. I like your tuned circuit switch too.

Maybe if the OP told us what exactly he wants to use it for, then it would be easier, maybe what is needed is totally different.

For certain values of "roughly". ;-P

For certain values of "essentially". ;-D

Cheers! Rich

I was quoting from memory. I ran the sims in 1998 and just looked them up today in my notebook:

If the damping rsistor is 0.52 x Xl, you're below 20 PPM amplitude in one oscillation period.

Does that better meet your high standards?

John

It works a peach to use an LC "pre-charged" with a current. Since I do lots of very fast CML/PECL stuff... sometimes up to 5GHz, I'll stash that into my bag of tricks. Thanks, John L!

...Jim Thompson

Exactly.

Not actually minimum delay, but a very precise delay. I assume that precise delays are easier to achieve for small delays, but we can handle any delay as long as it is well known.

All edges of the oszillation must occure at known points in time relative to the rising edge of the trigger. They are preferably evenly spaced, but that is not essential.

We cannot use a gated oscillator, because than the edges are at random positions relativ to the trigger.

We know that the approach with the phase splitter/ S&H/Mixer is used by commercial oscilloscopes and works very well. It is however overkill for our application. It is rather larger and complicated. Also: it's patented.

Kolja

How about feeding the trigger directly to one input of a very fast XOR gate and using a 100-200 ps delayed version of the trigger on the other input to generate a 100-200 ps glitch at the output. Feed this pulse to the 2.5 - 5 GHz oscillator tank circuit and the oscillator would start to oscillate at full amplitude with a known phase. A longer pulse might be acceptable, if some useful harmonics will hit the resonant frequency.

The oscillator amplifying element would initially have to be in class C to prevent any oscillation due to thermal noise. Also some precaution is needed to prevent oscillation due to the power up transient.

After all, there is not much difference between an oscillator that does not oscillate or an amplifier that oscillates :-)

Paul

Oh, come on, I was just joshing, a la "for certain values of 'about'".

Sorry if I spoiled your day.

Thanks, Rich

Not at all. I got to look up my sim data and refine "roughly" by a full digit. I hope Sylvia approves.

John

John, Your "force some initial-condition current" got me to thinking, so I conjured this up quickly...

A real tacky tank, nothing like a ceramic T-line resonator, but the same concept applies.

(In real-life I'd also use AGC on the oscillator for spectral purity.)

Might be a PECL open-collector part around you could pull off-the-shelf to try this?

...Jim Thompson