♫♪ My little cheap Circlophone© ♫♪

[Elvee]

Originally Posted by Tony

ok, how about input stage linearity, do you really need very high open loop gain?

This variant has been studied to meet the requests of Daniel: high output power on 8 ohm (~150W), with a high input sensitivity. This requires a closed loop gain of more than 200x. Without a sufficiently high open-loop gain, performances will be seriously degraded.

Now one could question the philosophy of it all.

What we see is that an increase in the input gain provides an equivalent improvement in linearity.
This seems logical, but in fact it simply means that the gain supplement provided by the CFP is clean enough to be used "raw", for example to reduce the errors of other parts of the amplifier (here, the output stage with its clumsy 2N3055's).
So, why would you want to linearize something that is already better than the context it is going to be used in?

I think that's simply sensible resource management. Now, you have the right not to like high open loop gains for ideological or other reasons, but that's completely different.
As Renardson says: negative feedback works, it delivers the goods.


Note that just for the fun of it, I have designed amplifiers and output stages not relying on global NFB, but these are exercises in style: they prove you can completely dispense with NFB and yet achieve ppm level distortion:
UniGaBuf, a follower cut out to be a leader
The Tringlinator: a MOS-based Tringlotron amplifier

[Elvee]

Originally Posted by terranigma

Do seem 2N5858 (80V, 200Mhz, 12pF, 750mW) another candidate as Q5/Q6 despite its very high fT?

No, with 12pF there will be enough self-compensation, that should be OK.

If I can find some samples, I'll try it just to make sure.

Supply regulation.. could circlophone benefit discrete regulators like LM338K? Is it worth?

No, I don't think so: as with many decent amplifiers, good reservoir capacitors are sufficient to ensure good quality.
The PSRR is not huge, and it could in principle benefit from a (very) good electronic filter (gyrator), but it would have to be really excellent to make a difference and not to degrade other aspects.

The output impedance in particular would need to be extremely low, even at high frequencies.

[kenpeter]

Originally Posted by Elvee

Removing R1 has no benefit for the differential gain, but it increases hugely the common mode gain, which is problematic

True .01% of the time, during tiny class A crossing only.
Rest of the time in near B? R1 degens open gain for sure.

Same thinking applies to comp, but you got at least one
voltage moving at all times, so the other has something
to comp against. (I had trouble comping complimentary
version, because both drive voltages fold on the rails)

B ain't exactly "differential" like an equal and opposite
that would cancel in R1.

-------


Noo, wait... maybe you are right? This is confusing...

-

Nope, I was right. In B, one emitter (Q5 or Q6) should
be stuck at constant current, not at constant voltage.
Voltage decided entirely by R1 and opposite emitter.
R1 would be degenerating "differential" gain, because
it is ping-pong single ended half cycle for all practical
purpose... Very little common mode about it.

[kenpeter]

If the input pair collectors were both pegged 1 drop
from the bottom rail, comp across those collectors
would have very little effect. It is only because R1
allows one or the other to move that it works at all.
This is more likely why removing R1 makes unstable.

A resistor in each collector circuit would provide the
same voltage swing across the comp, without degen.
Then maybe you don't need to CFP the input pair for
more open loop gain, you already have it.

[Elvee]

Originally Posted by kenpeter

True .01% of the time, during tiny class A crossing only.
Rest of the time in near B? R1 degens open gain for sure.

That's a bit like saying the brakes on a car are used 1% of the driving time, therefore caring about them is a waste of time.

Same thinking applies to comp, but you got at least one
voltage moving at all times, so the other has something
to comp against. (I had trouble comping complimentary
version, because both drive voltages fold on the rails)

B ain't exactly "differential" like an equal and opposite
that would cancel in R1.

-------


Noo, wait... maybe you are right? This is confusing...

-

Nope, I was right. In B, one emitter (Q5 or Q6) should
be stuck at constant current, not at constant voltage.
Voltage decided entirely by R1 and opposite emitter.
R1 would be degenerating "differential" gain, because
it is ping-pong single ended half cycle for all practical
purpose... Very little common mode about it.

I think you are too smart at analyzing the circuit: you "know" how it works, and you use that knowledge to draw conclusions, but that prior knowledge is preconception.
You have to play the game fairly, work the way a simulator does.
The dynamic inter-emitter resistance remains the same, with or without R1. Since the operating current in Q5 and Q6 is 13mA, that resistance is ~=2*26*13^(-0.9)~=5ohm
That resistance sets the transconductance of the stage in differential mode.

The common mode transconductance with R1 is a bit smaller than 1/R1 when present, and 1/1.25 when it is absent.
That's a big difference, and that's the only thing that matters for the circuit.
The common mode exists regardless of the circuits operating conditions, and it has to be made stable, and preferably independent of the differential loop.
Without R1, the mode mixing is maximal.

Then maybe you don't need to CFP the input pair for
more open loop gain, you already have it.

The initial purpose of the CFP was to decrease the bias current, increased loop gain is just a bonus. It could be thrown away with some degen, but why not use it constructively.

[danielwritesbac]

Originally Posted by Elvee

This variant has been studied to meet the requests of Daniel: high output power on 8 ohm (~150W), with a high input sensitivity. This requires a closed loop gain of more than 200x. Without a sufficiently high open-loop gain, performances will be seriously degraded.

Now one could question the philosophy of it all.. . .

Maybe I don't need 200x? So, I have 22k feedback resistor on a 45w to 8R amplifier and for 75w to 8R, that doesn't need much more gain. So, is there a suitable way to do just a little bit more, like maybe some fine tuning or an especially suited device selection?

With 22k feedback, I did get two discrepancies:
55w effortlessly and much more on dynamic peaks.
Dramatic decrease in temperature--It is efficient and cool.

From this point of view, it seems that the people who are using fans might have too much watts to the heatsink instead of the speaker.

[odysseybmx414]

Changing the gain does not alter quiescent current, or power dissipation

[kenpeter]

Together with common mode feedback, one emitter must be at constant current.
Simulation confirms this. Effective emitter resistance is, pluggin' numbers from
the graph into my calc. Taking out the tilt and going by the hump in the middle.
oh about 1.1K??? R1 clearly becomes significant to the differential by way of the
other emitter. The other one, not jammed to constant current by a logic threshold,
maybe a few ohms per your numbers.

[AJT]

This requires a closed loop gain of more than 200x.

ok, got you, just that i am used at looking at amps with gains in the 20~40 range...... thanks Elvee.....

anybody built one yet?

[odysseybmx414]

Elvee, damn you! Now I have to build another Circlophone...