Startable Oscillator

You need to say a LOT more about what you're doing. If you didn't care about phase, you'd just use a switch to gate it. So, how much do you care about phase? Distortion of the first cycle? Single shot or repetitive? Precisely repetitive or randomly enabled? Do you care about the phase when the oscillator stops? Fixed or variable oscillator frequency?

How much delay is the "known startup delay"? picoseconds? fortnights?

At ~3GHz, each picosecond of uncertainty in the enable signal represents a degree of phase error. Just getting the enable signal coupled to the oscillator is non-trivial. Think about the risetime of your enable signal. How much power supply noise does it take to shift the enable threshold by a picosecond?

It's tempting to think you can have a fast oscillator triggered by slow enable logic. You might find that the bandwitdh of the whole system needs to be greater than the oscillator frequency... a lot greater if you need tight phase control. And we haven't even started talking about getting the oscillator started.

There's a thing called a "gated restart oscillator". You don't start it, you let it free run. At the trigger point, you stop it and restart in the desired phase. Think of it as forcing a discontinuous change in phase. IIRC, it works particularly well with ring oscillators. Can be made to work inside a PLL. Don't know if anybody has tried to build one at the frequencies you're talking about.

Do you have a plan-B? It's far easier to resynchronize the enable with a free-running oscillator and use that to gate the other parts of the system if at all possible.

The "few Ghz" part makes it tough; it's fairly easy at lower frequencies.

Two techniques:

1. Build an LC or equivalent oscillator. Stop it by applying a resistive load to the tank, and also pump some initial-condition current into the inductor. To start it, just remove both suddenly. Done right, you get a perfect sine-wave start and flat-amplitude oscillation ever after. The "suddenly" part (1/4 the oscillation period or less is best) gets hard at GHz frequencies, where a cycle is only hundreds of picoseconds.

I've done this with discrete LCs, in the 50 MHz ballpark, and with coaxial ceramic resonators, around 500 MHz. Even at 500 MHz the initial-kick circuit gets tricky, but then there are some really fast ECL parts around nowadays.

1. Use a gated delay-line oscillator, a really fast ECL or CML NAND gate with one input connected to the output through a delay trace, the other input being the gate. Also difficult at GHz speeds. Some digital delay generators (HP, BNC) used this in the 100 MHz ballpark, and both Tek and HP sampling scopes used gated delay-line oscillators as their triggerable time base counters. Usually a varicap fine-tunes the frequency.

Try to find an online schematic for the hp 5370A/B counter or the

5359A hp time systhesizer. They use gated delay-line oscillators.

Look for patents by David Chu of Hewlett Packard. He did a lot of cool stuff.

What frequency do you need? What's the application?

John

That is known in some circles as a "stop oscillator", where the OP is suggesting a "start oscillator." There's really not much difference. If you want low jitter, you have to stop the oscillator long enough that all ringing from previous oscillation has died down into the noise, at which point it has the same initial conditions as the start oscillator. [1]

The motivation for the stop oscillator is that you can keep it warm and fine-tune it (ie phase lock it) between triggers so that it's dead-on center frequency when you kill/restart it. The disadvantage is that the start time is increased by the kill time, and any jitter in the kill one-shot adds to oscillator jitter.

See the HP instruments and patents mentioned elsewhere.

John

[1] for an LC oscillator having 50 ohm L and C reactances at resonance, slapping a roughly 30 ohm resistor across it is optimal (I have the exact value around here somewhere), in which case you can quench it to essentially zero residual ringing in under one cycle.

On a sunny day (Sun, 06 Sep 2009 09:12:56 -0700) it happened John Larkin wrote in :

For low frequencies it is easily done with a multivibrator. This I used as a TV pattern egenerator, start at end of H pulse, produce vertical lines every few cm. Always in sync and phase. I am not sure if one can make a GHz MVB though without using exotic materials. Tunnel diodes? Josephson switch?

Ring oscillators can go that fast, and are easily gated with one two-input gate. Not too easy in a built-up circuit, of course.

Cheers

Phil Hobbs

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net

n

erials.

SAW resonators are widely used in 300-500MHz area. Some are even close to GHz. But you have to deal with the GHz transistors and logics,

And the fact that a SAW resonator is a fairly high Q device, and so will have a long turn-on transient.

--
www.wescottdesign.com

Agreed. But, and I don't know one way or another, can a circuit be arranged so that the Q can be degraded synchronously (with the start signal) and then formed back towards high Q along a ramp in the 10ns period allowed? Just asking.

Jon

That's hard to do with resonators, 'cause they're inherently high Q -- try to yank them around too much and you find yourself conversing with the parasitic components instead of the intended effect.

'sides, the OP (remember him?) is only asking for better than 1% frequency -- there's other ways to get there from here than with a SAW.

OTOH, the OP just said "specified" delay, and didn't say what it was -- so an oscillator based on a SAW with some sort of fast AGC (or diode clipping?) may work.

--
www.wescottdesign.com

It's easy with an LC or a delay line or a coaxial ceramic resonator. All are "simple" elements that you can couple tightly to. LCs can have Qs in the hundreds, coax resonators in the thousands, but you can start either instantly and stop it in about one cycle.

+5-------------+------+ | | | | L C | | | | | | +------+----->

| | | R1 | | | c ttl----R2----b npn gate e | gnd

This starts oscillating instantly when the npn turns off, and the oscillation dies quickly when it turns back on, if you get R1 right.

Of course, it needs the usual additional gain element to sustain the oscillations long-term.

SAWs and quartz crystals and such are complex gadgets with the equivalent of many coupled nodes inside and weak coupling to the outside world, so it's hard to start or stop them quickly.

John

If the phase is important, and the frequency over a GHz, you might want to use a laser pulse to turn on a photoconductive switch. Photoconductors are slow to turn off, but that shouldn't bother this application.

Another possibility (but a tad pricey) is to get a couple of fast track/hold amplifiers and multipliers (Gilbert cells).

[output] :=3D=3Dsin(w*t - phi) =3D sin(w *t) cos(phi) - cos(w*t) sin(phi)

so if you have a master oscillator with multiphase outputs (sin(w*t) and cos(w*t), and sum the track/hold of cosine times the sine, and the negative of track/hold of sine times the cosine, the sum is sin(w*t - w*t) =3D 0 when the track/hold amplifiers are tracking, and becomes a known-phase sinewave when they hold.

c l

If power consumption is not an issue, you can switch it with a low impendence circuit. The particular SAW device I am looking at has open circuit Q of 11000, but 50ohm loaded Q of 1400.

Umpteen years ago (30??) there was an ECL 2-input NAND gate that could be used as a part of an oscillator; the other input allowed one to gate the oscillator. Startup was *zero* in that the first swing of the output _was_ the start of the sine wave. Mind you this was not in the giggles region, just rather high meggiehoits region, but you may use the same principle even if you have to use discretes to make a gated oscillator.

As John Larkin regularly reminds us, ECL has gotten faster in recent years. ON Semiconductor has some quite fast stuff

Farnell stock the MC100EP05D. John Larkin may know if you can actually buy the NB7L86M-D and its other 8GHz relatives.

-- Bill Sloman, Nijmegen

We use a few of the "NB" parts. They are available and they work. Single-gate parts run around $30 each.

The slower el/ep parts are easier to use (available in SO8) and run in the $3 to $7 range, still single gates. Micrel and Arizona Microtek second-source some of them, too.

Analog Devices makes some very fast comparators that can be persuaded to do logic-type functions.

John

If you can leave it running to stablize it, then gate it, a Gunn diode comes to mind.

Steve

The sustaining gain element isn't the real issue here. What's hard is to damp the resonator and force some initial-condition current into it, then turn it loose to ring when it's triggered. If it has a reasonable Q, the ringing will damp slowly, so just a little added negative resistance sort of gain will keep it from drooping. A fast transistor or a MMIC will do that part.

If you can get your chores done in a modest number of cycles, you can just use the ring and not need a gain element at all.

Range markers on tube radars were often generated by such a triggered-ringing circuit.

John

Yes, but multivibrators have more jitter than a good LC or - better yet - a ceramic coaxial resonator. LCs can also be temperature compensated to a couple of 10s of PPM, and the ceramic gizmos are that good out of the box.

John

That would probably do the trick. A dozen oszillations or so that have enough amplitude to be reliably converted into an LVDS-Signal would be perfect.

Any idea on what to read to learn how to get a ceramic resonator to ring?

Or a pointer to an engineering service that would design and simulate such a circuit for us? We will only design this once, so it is not strictly necessary, that I understand how to build this. (Although I am really curious and love to learn)

Regards,

Kolja