Oscillator AGC supply voltage sensitivity

I'm assuming this is a discrete-transistor design, not an IC design: if it's IC, the answer may be different.

I can only regurgitate a comment made by Wes Hayward (designer of Tek Spectrum Analyzers, and of innumerable amateur radio projects):

Instead of using AGC, it's a good idea to design the oscillator so that it current limits. Make it with one (not THREE for god's sake!) active element, and bias that active element so that it never saturates. Control the oscillator amplitude by setting the active element's collector current.

With four transistors to play with you should be able to build an oscillator and follow it with enough buffer stages so that what comes out is always at the same amplitude, even with a 2:1 or 3:1 variation in oscillator amplitude.

Short answer: replace R5 with a current source.

It's a CA3046 transistor array (implemented with LM3046 in fact). But you're right to ask; the original design was by Jim Thompson :). I was looking for something that has excellent spectral purity -

It's only one chip, but yeah, I take your point. My original design (that triggered JT's) was a Colpitts circuit, but I had trouble getting the AGC to work reliably and to start fast. I've had another play with it today and have figured out where it was going wrong. The latest of that is attached (also with AGC). However, you'll see that the collector voltage still falls below the base - is this the saturation you said to avoid?

Clifford Heath

I'm in the midst of moving myself from temporary lodging to our new house, so have no time right now to play with your circuit, but study this for now...

Might have some time tonight to look at your circuit.

Another suggestion, take AGC pick-off from Q1 collector, but that may add other problems :-( ...Jim Thompson

The reason that Hayward doesn't like AGC is because it adds to the phase noise of the oscillator (strictly speaking it adds to the amplitude noise, but at some point there'll be some phase shifting that turns it into phase noise).

If your goal is to get a nice pure sine wave, and if the consequent extra noise isn't an issue, then AGC may be the way to go.

I'd investigate variations of the Colpitts circuit and see if there's a circuit that tends to be fairly pure, and then see if I could follow it with a bandpass or lowpass filter -- but I'd also look into doing it with AGC.

In the realm between the collector voltage equaling the base voltage and the collector voltage being one diode drop down from the base voltage, the definition of "saturation" is kinda slippery. The more drop from base to collector, the lower the impedance will be. When you get to one diode drop (and collector-base junctions tend to have a lower diode drop than 0.6V) then you're definitely saturated.

Note that a saturating collector-base junction is, by itself, a source of harmonics. So even if you didn't want to avoid it because of phase noise issues, you want to avoid it for your purposes.

Whatever else you do, make sure that you get the thing so that it works over all expected temperature ranges and component variations. Oscillators are persnikity things.

With your circuit as shown, I don't think the AGC is doing anything at all. Rather, it looks like the amplitude is limited because Q1 is saturating.

In a normal common-collector Colpitts, you don't have the collector resistor (R3 in your schematic). If I set R3 very small, then the AGC does something -- it slams down to zero volts, which means its out of regulation. If I increase R6 so that the AGC has a chance to work, then the oscillator squegs. But, I can fix that with C4. Then starting gets slower.

Try this. It's a proper grounded-collector Collpits. You can see that AGC is really working because the bias line does get pulled down, but never gets pulled all the way down to zero. Starting is a hair slower, because the higher emitter resistance limits transistor current (and hence gain) -- you may be able to influence this a bit by dinking with the bias network.

This is in no way optimized or even checked to make sure that it's not just a fixture for burning up LM3046 chips -- that's your job.

Be sure to check it with a variety of inductor Q's -- I'd make sure that it works at the manufacturer's advertised Q, and a healthy range around that, from no loss to about 1/3 the manufacturer's Q.

Also be sure to check its operation with a pick-off in place -- this thing is no use to you if it works great in isolation but can't cough up a signal. Fortunately, you've got transistors to use for your buffer amp.

And finally -- why am I doing this? And why am I doing it for free? Shouldn't I be doing paying work?

Note that the AGC voltage is developed by comparing the emitter voltage with the AGC transistor's diode drop. That diode drop changes with temperature, so the oscillator's output voltage will change with temperature, too. If that's an issue, you need to find a different way to develop your AGC (if it's critical, you probably want an op-amp in there somewhere).

Note, too, that the AGC voltage comes from a combination of the oscillator's DC current and the actual oscillation voltage value -- and most of the action is from changing DC current. Again, if the oscillator's p-p voltage matters, then you should measure it properly, then use an op-amp based AGC circuit to actually control it.

Done right, you'd use a pair of transistors (oscillator + buffer), a diode (to measure output), and a dual op-amp.

It'd be good to know how to run such a range of tests using LTSpice.

Perhaps Jim can offer advice on how to automate that kind of modeling?

It's not critical. I just want two things: to stabilise the amplitude fairly quickly, and to keep following stages linear. If you disconnect the AGC in that Colpitts circuit, you'll see that the amplitude is still stabilising after 100us, instead of closing within 30us, because C1/C2 are still charging to the emitter's DC point.

The purity was dramatically improved by the addition of the base capacitor, which also allowed removing a much larger cap in series with the inductor. That cap was also a cause of slow startup, because it takes a while to reach its final DC state of charge.

I will be adding varactors to tune the oscillator, which is why I don't want to rely on post-filtering (though the varactors add more non-linearity themselves).

Regarding extra phase noise from adding AGC, surely that depends on the kind of filtering applied? Here I have a simple R/C combination, but it could go 2nd order.

Thanks for your thoughts. Oh, I just saw your final response. I'll send this, then play with your proposal and respond to that.

Clifford Heath.

I think the AGC transistor's BE diode is turning on hard enough to turn Q1 full on. I played with different resister values and the AGC seems to be doing something, but there's definitely two methods of action.

Right - I had that at one time, and I'm sure I had it regulating too, but it's fiddly as you say.

I had this topology at one time but didn't get it as clean as you have it. Probably discounted it because its slower to start, but the payoff is there - impressively clean result.

If it's any consolation, it's not my job either :) I'm just a hobbyist learning how to design DF gear, in the hope of attracting some of the very fit Sydney geeks into a radio hobby, through sprint ARDF.

Good advice, will do.

Clifford Heath.