inductor tempco

My Colpitts will be amplitude limited by clipping, either the c-b junction of the BFT25 or by the startup circuit. Either way, the gain (emitter current) determines how hard it clips, and that is easily tuned. Amplitude stabilizes in a couple of cycles. I have no idea how clipping (as opposed to some delicate AGC mechanism) affects things like phase noise.

Spice is not useful in simulating this, and I don't have the budget or the learning time to use fancy RF tools. It's solder-and-fiddle time, I guess. I really need to do that anyhow, for tempcos and such.

The actual design needs to leave lots of knobs to turn, to allow tweaking.

Resonant time bases (like an LC tank) provide better phase noise and stability characteristics than non-resonant solutions (even, I'm pretty sure, your 1/4-wave line).

Not knowing what JL is planning I can't say which approach is better, but if he needs good stability and phase noise characteristics, than building something around a good LC oscillator is the way to go.

If you waste some current, it can also be the solution: instead of an instant-start by applied voltage spike, you can 'idle' the tank by switching a high regulated current through the coil. When you unclamp the coil and the loop amplification takes over, it's already got stored energy in I-squared-L, so it doesn't need to accelerate up to full amplitude.

Like this (a tutorial by a friend of mine from grad school, Mark Rodwell):

As he says on the first page, "...a well designed LC oscillator like the Colpitts must be current limited: that is, we allow the active device to go into cutoff for part of the period of oscillation. This reduces the effective gm of the device, averaged over the period, and thus gives the needed gain compression. It also benefits phase noise as we will later see. The simulated drain current for a common collector Colpitts oscillator below illustrates that the device can be in cutoff for much of the period."

"The opposite of current limiting is called voltage limiting. In that case, clipping limits the amplitude of oscillation. This is undesirable because it generates a lot of harmonic distortion plus the low impedance state when the device is driven into saturation or the ohmic region kills the tank Q for part of the cycle and is bad for FM or phase noise."

IOW E-B cutoff is your friend if you want a simple oscillator with low phase noise.

The tank gain is only about 1.1, which is probably fine. I initially misread the schematic--I was thinking of a CC Colpitts, which is what I usually make. In your CB case, I should have read the tank the other way up, i.e. a gain of 10.

Cheers

Phil Hobbs

Right, that's how you do it.

Cheers

Phil Hobbs

A 1/4 wave line is a resonant solution.

Cheers

Phil Hobbs

I don't think it is the way that Bill is proposing to use it.

When properly terminated for the job a 1/4-wave line is resonant, but Bill's talking about using it as nothing more than a 90 degree delay line, not as a resonant element.

Cliff: your clock appears to be about a half day behind everyone...

Tim

Seven Transistor Labs, LLC Electrical Engineering Consultation and Contract Design Website:

The amplitude at the BFT25A's base is about 40mV, based on the capacitive divider ratio. Without having Spice'd it, there should be enough gain headroom to double the output by increasing the divider ratio, which will probably increase the signal purity also (more tank energy, same base bleed).

Clifford Heath.

Ah, right you are, thanks. I was thinking about JL's fave coaxial resonators.

Cheers

Phil Hobbs

Interesting, it's sort of DC bootstrapped, so despite looking like a follower, it'll actually have high loop gain!

Be careful it doesn't also oscillate, I suppose (cap on base and emitter!).

The offset voltage needs to be smaller though. Or the oscillator side needs to be biased up slightly. Otherwise it's too close to the Vcb limit anyway. Well, using a schottky diode would help I think, so it limits to more like 0.5-0.6Vpk.

Running the oscillator at much higher Vce would also save fT, but that's not a problem with an already blazing fast transistor. The flatter Ccb might be handy I guess. Or heck, use a slower/fatter one, and modulate Vee for frequency control? :-p

Tim

Conversely, if it were a voltage mode amplifier (transresistance or pure voltage gain), voltage clamping would be preferable -- and the resonant tank would have to be series resonant, not parallel. (The output would be current through the tank, rather than voltage across it.)

But transistors are gm devices first and foremost, so the parallel resonant and gm limited operation is preferable.

Tim

It'll be less than 5% change.

Tim

You

It's a class C oscillator and the make-up current (small compared with the standing current at Q around 100) is a Dirac spike, with all the harmonics of the fundamental up to limit imposed by the width of the spike (or the ba ndwidth of the transistor).

Why not? You have to plug in real values for the inductor (but Spice now in cludes a bunch of Wirth inductors with real values for the parallel and ser ies resistances and capacitances). It doesn't have Spice models for fast tr ansistors, but they are on offer on manufacturer's web-sites.

Not to mention lacking the enthusiasm, either.

Or be done properly in the first place, to minimise the post-assembly tweak ing.

The 1/4 wave line wasn't part of the tank. It's just one way of getting a quadrature signal that you can feed back to tweak the frequency. It's the kind of thing you can build into the printed circuit layout if you have enough space.

He eventually got around to revealing the "instant start-up" aspect (which isn't impossible - as has been pointed out - but isn't something I've thought about in any detail).

e wave oscillator you could use a pair of Analog Devices fast multipliers - the AD834, AD835 or ADL5391 are all fast enough for a 100MHz oscillator.

the amplitude where you wanted it, and the other would provide adjustable ( positive or negative) quadrature input to allow you to pull the frequency u p or down. At 100MHz, half a metre of coax could provide the quadrature com ponent to be fed into the second multiplier. A coax delay line isn't a broa d-band solution, but sufficiently broad-band for something you might otherw ise fine-tune with a varicap.

but it would be easier to explain to customers than the traditional approac h.

And thousands of dollars worth of tweaking.

In low volume production, what you pay for the components over the commerci al life of the device is dwarfed by what you spend on designing, debugging and documenting what you build.

A more expensive, but more predictable device, in a more designable circuit , can often come out cheaper over-all, particularly if you can get it right first time.

Thanks. For some reason this Linux virtual machine doesn't synchronize via VMWare, even though it's configured to. I'll have another look at it.

Sure, the idle initial condition is current through the inductor, applied from the impedance that best damps the tank. I need to start and stop the oscillator ASAP. We've done that lots of times, with LCs and with coaxial ceramic resonators.

I've done similar instant-start oscillators around 600 MHz, using 1/4 wave coaxial ceramic resonators. They start out as pulse-propagating "digital" delay-line oscillators and, as the corners round off, gradually transition to sine waves. That can be mitigated some by starting them softly, launching a slower initial edge.

CCRs are great for acuracy and tempco and Q, but they start around 500 MHz, are sort of expensive, and tend to have Zo values around 10 ohms, so need a lot of initial-condition current, which implies low oscillation amplitude. 500 MHz is too high for our FPGAs, so we'd need a divider too.

Real coaxial cables are a mess. They are big and ugly to install on a board. Teflon has a phase change right around room temp, so frequency tempco is awful. Polyethylene melts when you solder the braid.