Bootstrapped Cascode Stability

Hi

I'm trying to debug an amp using a bootstrapped cascode as conceptually shown in PAD112_Rev_C.pdf; the input pair is a LSK389 (at about 3 mA Id), the rest (cascode, bootstrapping follower (at 2 mA Ic), drain load and current sources) is made up of BC550C/BC560C.

In inverting mode the amplifier works like a charm. However in noninverting configurations I get oscillation which is either just conveniently visible in the 20 nS setting on the scope or above. It appears that adding series resistance/a ferrite bead in series with the noninverting input helps things but it's not a complete cure.

In any case bypassing the cascode transistors completely eliminates the troubles; increasing the compensation doesn't help, which makes me believe that it must occur as a local instability in the bootstrapping circuit. Things are on a PCB with relatively tight layout, and the used transistor are not that fast, so I wonder where's the problem.

Any ideas what to try here? Thanks!

Samuel

Note that each half of the bootstrapped cascode diff pair is very similar in effect to the Baxandall superpair, the sort of folded darlington, also existing in versions studied by Hawksford (seemingly unaware of Baxandall), and all of these have a tendency to have a negative input impedance at some frequencies.

There is a thread in diyaudio about the Baxandall etc.* with a lot of speculations and sims about stabilization. I wound up concluding that jcx's suggestion of a little output C loading probably worked the best, at least when looked at in isolation. It is more than just slugging down the circuit with a C---it has a disproportionate effect on response.

How exactly this fits into the differential bootstrapped cascode is a little different, since in that case each half is loaded at the source by the other source. It's possible that just a little C to common can neutralize the negative input C component.

Another thought: if the oscillation is associated with parasitic L the likely culprit according to a course given at UCLA by an old Tek guy Siegfried (?) Knorr, whose course notes were kindly copied to me by R. V. "Balu" Balakrishnan, is that L in the base lead of a common-base stage. But I suspect what you are seeing is a more fundamental interaction, given the short traces and reported frequencies.


*http://www.diyaudio.com/forums/showthread.php?s=&threadid=25172&perpage=25&pagenumber=1

Thanks for mentioning the negative input capacity--totally forgot about that one! Thought about the base L of the cascode though but expected it to be unlikely to be the problem as the BC550C has pretty high rbb' which might dampen things enough, but who knows. Back to breadboard in a week...

Samuel

It turned out that the only problem was a bad cable... The shield was lousy enough to cause instability when the output cable came close to the input cable for a high-gain application. :evil:

However in the mean time another Q emerged: I'm somewhat short on common-mode input range and measurements indicate that noninverting distortion improves when I lower the source-drain operating voltage. So how do I determine the lowest possible source-drain voltage which will not suddenly cause a saturated input pair in production?

I feel like I should know this already, but FET parameters are sometimes still somewhat misterious for me...

Samuel

The pinchoff voltage and the threshold voltage are in a way two sides of the same coin. Without going into a complete explanation based on the physics, generally you can operate reasonably well down to magnitudes of drain-source voltages roughly comparable to the on threshold voltage magnitude.

The considerations, though, will also include the substantial increase in drain-gate capacitance and the increase in drain-source conductance. But the bootstrapped arrangement does a pretty good job of mitigating those effects. However, for really large drain conductances, the voltage noise of the cascode transistor will start to contribute to the total noise. Normally its contribution is small and dominated by shot noise in base current (for a bipolar).

To make matters worse for prediction at least, the sim models I've seen so far can be not terribly accurate. As I mentioned elsewhere, there's a contribution to be made here by some energetic person willing to improve the models.

From a noise standpoint I recall one of the nuclear science preamp people saying that a FET of choice (probably a process 50-ish device like a 2N4416) had no degradation in thermal channel noise performance down to Vds of 3V. This was of interest since he noted that gate current was well beyond expectations at as low as 5-6V at the optimally low temperatures for thermal channel noise, attributed to impact ionization of the cold lattice by the "hot" electrons. Not a consideration for audio, unless we start putting cooled FETs in condenser mics.

Practically, I'd suppose that 2.5-3V would be o.k. for a ~SK170 device, but as JR says YMMV.

Thanks; I'm currently running them considerably lower at 1.6 V, so perhaps I'll need to change that. I didn't observe increased noise though but will check again.

Samuel

The noise contribution will probably be small compared to external feedback component noise and maybe other things, but I wanted to mention it for completeness. Normally one could ignore the voltage noise component of the cascode common-control-electrode device, since it is able to get traction only when the driving source has got some fair conductance.

I've seen that one condenser mic schematic in here that looks as if the input FET was running at an awfully low Vds, and apparently it works o.k.

Im dealing with something very similar right now.

Ive built a discrete gm amp loosly based on the CA3080 circuit. I recently got a bunch of diodes Inc's matched duals DMMT3904, 3906 in sot363 to try out, So im using one of those as the input diff pair. the diff amp is bootstrap-cascoded with MMBF4391 JFETs in the collectors. there is a local oscillation at very high frequency, I cant see it at all on my 100mHz scope, but I can measure a large (two orders of magnitude) increase in second harmonic distortion at 1kHz when it is happening. This is independent of closed loop gain (from 6dB to 60dB), inverting or non-inverting. increasing the compensation cap has no effect. Im getting about 6V of Vce across the 3904, which is maybe a little bit high. but I think the real problem may be the relative high speed of the 3904 versus the 4391? damping one of the 4391 drains with 200pF to ground seems to work (the one that connects to one of the second stage current mirrors), but i dont know how optimal that is. Does anyone think there is any point in adding small B-C caps on each transistor of the input pair to slow it down?

Aside: Ive been working at it for a while, and am finally getting good results making prototype PCBs with the help of a 30W CO2 laser cutter/engraver we got at work (we use it to engrave front panel graphics). I coat the copper with black spray paint (hi-temp grill paint) then burn off the paint with the laser. from there etch with ferric chloride. i'm currently achieving 12 mil trace and space RELIABLY, which is good enough to run traces up between the pads of SOT363's (!). unlike press-n-peel, it is very consistent, if I get the right parameters EVERY trace is perfect, no rogue pads falling off or random shorts. the laser can't cut the copper directly, but it can slice right thru 63 mil fiberglass, and also drill holes. my time to prototype just went WAY down.

mike p

[quote author="mikep"]... but I think the real problem may be the relative high speed of the 3904 versus the 4391? damping one of the 4391 drains with 200pF to ground seems to work (the one that connects to one of the second stage current mirrors), but i dont know how optimal that is. Does anyone think there is any point in adding small B-C caps on each transistor of the input pair to slow it down?
mike p[/quote]

Doubt that the 4391 is too slow.

I can conjecture about what you circuit looks like, but of course I don't know precisely. However, what sims tend to say is that a small C at a control electrode to common has the desired effect of compensating for the tendency of these cascoded stages to have a negative input impedance.

Another property I've noticed in breadboard is the ability to oscillate with a high-Q input resonator, which can be just an unterminated cable. A small R in series or a ferrite bead should quell.

Just to make sure: you've verified that the instability disappears with the cascodes removed? How do things behave with high/low source impedances?

Samuel