♫♪ My little cheap Circlophone© ♫♪

[Elvee]

Originally Posted by kenpeter

. Some of the stoppers clearly are
not stopping much, and could go if the main
problem with stabilty was solved. It goes full
triangle by 40KHz, but even thats not got it...

Those Taylors are made not to ever shut off.
I don't know if thats part of the problem?

You are using an "inverted" logic compared to the Circlophone: detection of lack of quiescent current with subsequent ORing of + and - informations.
That's the path I tried to follow for months, when I was developing the Circlophone: it seemed natural and logical, and avoided the saturation issues of the other option.
I could arrive at extraordinarily good static or quasi static results, but the dynamic behavior was a nightmare.
In the end, I gave a try to the dual version, and miracle: the dynamic problems were gone.
There was a degradation in the static linearity, but it was acceptable, and the Circlophone was born.
Now, I am not saying all systems based on the lack of current are doomed, I have made successful attempts in the opposite direction, but for some reason I don't grasp, the "positive" variant seems more responsive

Originally Posted by kenpeter

Actually, a diode bridge between gates seems to fix the clipping instability.
Limits gate voltage difference to 1.4V, now hard clips and recovers clean.
Guess this also limits the maximum error stored in gate charges, thus the
quick recovery?

Still can't get this thing past 40K. If I ease up on comp, goes banannas.
Lessening the open loop gain helps stable bandwith, but hurts distortion.
Like for example: source resistors, or using a BJT limited by beta. Going
other way, to a MOSFET with even higher gain, stability is unobtanium.

Maybe I am pushing too few transistors for way too much open loop gain?
Feels like xquad again, though I doubt any negative resistance is at fault.

The stability margins of both loops seem rather marginal, some form of compensation would be needed to make the thing safe.

Originally Posted by kenpeter

And one more nutty unstable distraction for Elvee to ponder...
Same old same old, but new and improved with over and under.
Yeah verily, er whatever!

Interesting, but I fail to see the advantages of this configuration: ideally, we are looking for an optimum differential gain and linearity, and a minimal interaction between common and differential modes.

[kenpeter]

Evequad what I has in mind, more or less?
Anyways, yes it works. No, not any better.
This might have been worth exploring, but
I am ready to concede it is not fully baked.

Weird thing here: Gain is max when emitter
resistors (and diodes) are abused up top, to
match the ones on the bottom. Not just the
distortion goes down, open gain goes up!!!
Bigger or smaller resistors lose gain, why?

But open loop gain here is not comparable
to a plain old xquad, nor even the CFP way.
This Evequad seems only just barely gainful
enough to get the open loop job done, and
I don't really understand it well enough to
know what stability means in this contex.

[heinz]

Another iteration of powerflux pcb ...

[kenpeter]

Bandwidth is up, and behaving well for both reasonable
(soft clipped) and unreasonable (hard clipped) overloads
at the input. Both input transistors stay on til the clipping
nonsense gets way beyond even that...

I was just trying to keep these transistors on, when I
noticed a problem with diodes from input to output...
If you don't attenuate that feedback, it can be mis-
interpreted as a positive and aggravate the clipping,
maybe even latch.... Give it anything less than unity,
and the problem does not occur.

So basically, anytime the output can't properly follow
the input, the input is attenuated to a smaller copy of
the distorted output. Prevents all sorts of wacky evils
when faced with the impossible.

Thats when I got this dumb flash that maybe this is
also where my comp should go? YESSS.... that worked!
Finally could loosen the screw on those output transistors.
40KHz lookin' siney with just a little bit of triangle, and
tips not bent (phase shifted?) too much forward nor back.

[kenpeter]

More foolishness from the same bag...
Drop either rail to 12V and watch it shut off.

---

Do you ever get the impression that an amplifier
trying to play music is like a guy at the hardware
store trying to juggle an egg with vice grip pliers?

[terranigma]

I powered up new pcb but it shows 300mV level dc-offset. If I bypass input cap and shortcut input to ground, It is falling to 50mV level and decreasing to 1mV slowly in a minute. But starting to increase again just after. I have no idea where to look at. Any advice please?

[terranigma]

Here some voltage measurements referring to ground.

[kenpeter]

Looks to me: 7V too much Zener, or 7V too little power supply?
You could try a cascode at output potential, rather than a zener.
Q5 Q6 won't have any excuse to see different collector voltage.

D10 or D11 might be faulty?

One of your thresholds is maybe higher than the other?
Why is there only .1V drop across Q7?

[terranigma]

Originally Posted by kenpeter

D10 or D11 might be faulty?

Spot on!

Thank you Ken, after replacing D10-11 it seems it is ok now. Let me see how it performs.

[terranigma]

Originally Posted by kenpeter

Looks to me: 7V too much Zener, or 7V too little power supply?
You could try a cascode at output potential, rather than a zener.
Q5 Q6 won't have any excuse to see different collector voltage.

My previous builds were having same issue and Elvee told me that it is not possible everytime to getting exactly half of the total supply voltage with zeners. This voltage difference doesn't affect sonic performance of amplifier according to him, if I remember correctly. If you can find an opportunity, it may involve us if you clarify with some schematic that what you mean with "cascode at output potential, rather than a zener".

Regards.