How do you make armstrong oscillators self-starting?

[Whales]

This oscillator works but I have to manually start it by momentarily shorting the mosfet's drain and gate together:



Q: is there common terminology for an 'oscillator starter' circuit?  I feel like I'm missing a keyword in my research.

I've tried a few combos of diodes, capacitors and resistors (to put a startup pulse into the LC/gate); with no luck.  I thought I had something working but it was unreliable. 

[T3sl4co1l]

Why does the gate have to be DC grounded?

Tim

[Whales]

Are you suggesting connecting the coil from the gate to +VCC instead of ground? 



That doesn't oscillate, surprisingly (edit: even with coil polarity on the transformer swapped).

I thought this would work -- perhaps the DC current saturates the inductor & ruins the circuit gain?  I'll keep poking.

[Whales]

Hooray, this works brilliantly   Self-starts and all.  Thankyou Teslacoil.



The Vthresh for these 2n7000's is somewhere around 1.6-2V, so the 3V (half of VCC) starting point is still high enough to work.  Once started however it's DC current limiting (thanks to the resistors) but not AC limited (short through the 100u capacitor to GND). 

EDIT: apparently this mod has made the oscillator 20% faster   The 1nF is a film cap, so I presume it's my transformer saturating & therefore reducing in effective inductance.  Who said they needed varicaps to make a VCO?

I still think the reason why the mod in previous post didn't work is due to excessive DC saturating the transformer, but I might be wrong.  I'll follow down some paths that use less quiescent than the divider, see if they work as well.

[Whales]

Saturation theory likely confirmed: if I remove the lower resistor in the voltage divider then the circuit does start, but runs for about half a second before stopping.  I suspect this is the 100u charging up until a point that too much DC is being put through the inductor.

My original plans to oneshot a pulse in (using capacitors + diodes) might actually work, I was probably just using too small of capacitors.  To fight the 30uH inductor up to more than Vthresh of the mosfet I probably need 30uF+ of capacitance.

[Zero999]

Saturation theory likely confirmed: if I remove the lower resistor in the voltage divider then the circuit does start, but runs for about half a second before stopping.  I suspect this is the 100u charging up until a point that too much DC is being put through the inductor.

I don't see how it can be due to saturation, as there's no DC path through the coil, on the gate side of the transformer.

It's probably due to the MOSFET being biased hard on, when the capacitor fully charges.

[Ian.M]

Move the current limiting resistor to be in series with the Source, and put a decoupling cap across it so it doesn't kill the AC gain.  Adjust the gate bias potential divider till oscillation just starts.   The MOSFET will conduct during  positive going half cycles (at the gate), raising the average drain voltage and thus reducing the effective Vgs bias as oscillation amplitude increases.

[Whales]

Doh, I was imagining a BJT instead of a FET.  Indeed it couldn't saturate at all from DC current flow (there is none, except to the non-pictured load).  Vgate can however rise high enough that the mosfet is saturated on.

Thankyou both

[Ian.M]

Worse than that, if relatively strongly coupled (but not near unity, when transformer action will limit the voltage) and Q of the tank circuit is high, it can easily develop enough amplitude to exceed the Vgs rating of the MOSFET!
You may need to add some sort of non-linear damping to limit the amplitude, e.g. a Schottky diode + series resistor from the gate to the positive supply.

LTspice sim attached.   Note that the gate bias voltage is quite low, the coupling is *VERY* weak^# and I gave both coils 1 ohm resistance to lower the Q.

# Think: air cored coil for the tank, with another air cored coil at a near right-angle to it in the drain circuit, adjust the angle of the latter till oscillation is sustained.  However, see T3sl4co1l's comments below on resonant voltage multiplication when weakly coupled.   

[Whales]

Ian: wow yours is hot, around 30V p-p on Vgs:



My gate only swings a few volts p-p in real life, the positive regen of my circuit is somewhat limited by the resistor on the drain.  I've noticed that when my power supply drops a bit I need a smaller R there for the oscillations to sustain.

I wanted to go back to my SPICE sims to verify that they showed only a few volts p-p oscillation too, but instead I found that my model is completely wrong.  Increasing the simulation time beyond just a few cycles reveals it doesn't have anywhere enough regen to keep going for even a 10Kohm load  



The circuit works in real life quite well, so I'm going to have to work out what I'm missing in my simulation.  Tweaking the drain resistor to be lower completely kills all oscillation, as does tweaking the resistor divider too much either way.   EDIT: my IRL load is just another mosfet gate, not a 10K resistor a bit different, so maybe I have weak regen but still enough.  Time for sleep, I'll look tomorrow.

[Ian.M]

R1 is a 10 K load effectively shunting/damping the tank, and with the 1K drain resistor limiting the energy that can be put back into the tank, I'm not surprised its marginal.    You need to measure the loop gain in the sim.  I suggest Tien's method.  See https://sites.google.com/site/frankwiedmann/loopgain
and in LTspice: examples\Educational\LoopGain2.asc

N.B. as SPICE AC analysis uses an ideal linearised at its operation point model for all active components, you'll probably have to mess with the bias voltages as they *WILL* affect the loop gain.  If the large signal bias shift causes the loop gain to drop below unity, with a significant delay, the oscillator will squeg.   

Edit: fixed broken link

[T3sl4co1l]

K = 0.01?  Yeah, you'll get Q multiplication at the gate.  OP is talking about "transformers" which presumably have K > 0.9... but who knows, no such parameters have been mentioned here yet.

Anyway, yeah, biasing.  A divider works more or less, but won't work over a wide voltage and temperature range.  Better is a diode-strapped MOSFET, so that the oscillator drain draws a current proportional to the bias current into the "diode", and the ratio of channel widths.  Downside: this isn't practical to do with discrete parts (2N7000s), it's almost exclusively an IC thing.  SPICE will behave correctly though (indeed, SPICE assumes transistors are identical; indeed, it was written for simulating ICs).

The solution for a real discrete implementation, is to make a "Vbe multiplier" (except Vgs(th) in this case), i.e. apply bias current to drain (source grounded), and wire a voltage divider from drain to gate to GND.  This way, drain current equals the bias current (less a small current into the resistor divider: use relatively large values so this current is small), and therefore Vgs is held at whatever voltage corresponds to that drain current.  Then: bias the oscillator from another voltage divider, so that its Vgs can be offset/adjusted relative to that reference.  If the transistors were matched, the two divider ratios would also be matched, but in practice they will have different Vgs(th) and gain, and the adjustable range allows you to account for this.  Now the two transistors will track with temperature and be independent of supply voltage, or at least, better than a fixed divider will.

Tim

[Andy Watson]

Q: is there common terminology for an 'oscillator starter' circuit?  I feel like I'm missing a keyword in my research.

It is a slight tangent but try looking up "squegging" or " squegging oscillator"

[mawyatt]

You could combine an Armstrong oscillator with a Colpitts oscillator and get an Arm-Pits Oscillator.

Phase Noise stinks tho

Best,

[Whales]

(Bumping this topic to give a few closing notes, because I'm about to open another oscillator SPICE topic)

It turns out that most (if not all) of my sims, including those pictured above, were not valid.  If I added a tiny load (eg 100K) and increased the simulation time period then I would see the oscillations die out.  I think the sim's initial DC conditions were good enough to get the L and C oscillating, but the circuit didn't have enough regen to keep them going.

Or in other words 'just because it works on the bench doesn't mean you know enough to immediate SPICE sim it'.  Enough armstrong for me for now.


> It is a slight tangent but try looking up "squegging" or " squegging oscillator"

Hey, I've accidentally made those before.  I assumed my little battery supply was collapsing and returning every few seconds.  Now that I know this is a totally valid concept I feel slightly chuffed 

[Dave]

Small tweak and it's self-starting with very little offset by the divider.
You will need to ensure that the power rail snaps up quickly enough. A regulator with divider configurable enable would work nicely here.

[Zero999]

(Bumping this topic to give a few closing notes, because I'm about to open another oscillator SPICE topic)

It turns out that most (if not all) of my sims, including those pictured above, were not valid.  If I added a tiny load (eg 100K) and increased the simulation time period then I would see the oscillations die out.  I think the sim's initial DC conditions were good enough to get the L and C oscillating, but the circuit didn't have enough regen to keep them going.

Or in other words 'just because it works on the bench doesn't mean you know enough to immediate SPICE sim it'.  Enough armstrong for me for now.

Presumably you left the series resistance field blank, in which case it defaults to some impossibly near zero value, so no wonder it oscillated. For a realistic simulation, set the series resistance field to something sensible.