Fresh fix for *that* Slone amp (ie yet another symetric topology)

[godfrey]

I got to thinking about this after seeing some of the grief over here: HEEEELLLPPP : M. Randy Slone Mirror Image Topology Construction - Troubles

The problem with that kind of topology (with or without mirrors), is that if the VAS has a decent amount of current gain, then the DC biasing gets horribly twitchy and virtually impossible to stabilise.

The circuit below solves the problem by using cross-coupled feedback to stabilise the idling current, while simultaneously boosting the effective current gain of the VAS. Here's how it works...

Any increase in the current through the VAS devices (Q5 and Q8) causes a decrease in the current through Q7 and Q10, which supply the tail current for the LTPs. So the common-mode idling current is stabilised by negative feedback.

The picture is totally different for differential-mode signal current though. An increase in Q5's current results in a decrease in Q8's current, which results in an increase in Q5's current etc... So the signal current is subject to positive feedback, which effectively boosts the VAS current gain.

So much for the theory but how well does it work?
For the circuit as shown, the input devices idle at about 0.8mA each and the VAS idles at about 9mA. Changing any one of the resistors by 5% causes at worst a 1 to 2 mA change in VAS idling current.

The surprise (for me anyway) was that changing both voltage references from 2.4 to 2.5 volts only increased the idling current to about 11.5 mA. Without thinking about it too hard I'd expected worse in that respect.

As a final test, I tried unbalancing the input by replacing one of the 2n2219 with a 2n2369. This resulted in a change of about 3 to 4 mA in the idling current, as well as the expected hefty dose of input offset voltage.

So, in a nutshell, you'd have to try pretty hard to really screw it up.

More later...

[godfrey]

A belated disclaimer:
I don't recommend actually using the transistors shown in the sim. Since my model library doesn't have anything decent and I'm too lazy to download any, I just grabbed the first jellybeans that came to hand.

Meh - they're good enough for a proof-of-concept, and this is supposed to be about topology not device selection anyway.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Back to the circuit...
To avoid weirdness (like negative input impedance) it's probably a good idea to keep the feedback factor below 100%. For the circuit shown it's about 40%. More feedback does improve the performance though, so it's worth aiming a lot closer to 100%.

A nice option is to reduce Vs1 and Vs2 from 2.4 to 1.7 volts, and reduce R7 and R10 to 560 ohms. This gives a useful increase in voltage gain (or decrease in output impedance, depending on the load and how you look at it). Reducing the resistors further to 470 ohms gives even better performance and is probably close to optimum, but idling current increases to about 15mA if the voltage references remain at 1.7V (a convenient value as it allows for simple LED biasing).

A somewhat odd side-effect of the cross-coupled feedback is that when driving a fairly heavy load (e.g. the 10k shown), the signal current through Q1 and Q3 is a lot lower than that through Q2 and Q4. So there's a load-dependent difference between the input impedances of the inverting and non-inverting inputs. At higher frequencies where the miller caps are rolling off the voltage gain, the differences disappear.

[Olivier73]

Hi Godfrey, this is nice thinking. I will implement the idea in the simulator and see what rolls out. You read the grief in my story :-)
Indeed a not so funny adventure but hey, I learned a lot !
I'll come back to you if i got some results...
cheers
olivier