Amplifier topology subjective effects

[jcx]

how much does the gain change if R92 is replaced with a current source? If it increases > 2-4x then voltage controlled Gm of Q56 with a higher input voltage from increased input impedance at Q58 base is a better explaination of the gain

[jcx]

Oops, that's not going to work... the input mpedance increases by the same factor the current division at Q56 base changes

you certainaly can be limited by current gain and input i is going to be related to output i by the device current gain (and any current division along the way) but input v times Gm*RL product will also be consistant

Gm is much more predictable for BJT than current gain and Gm based modeling shows commonality of operating principle with FETs where current gain is a fairly useless concept

[traderbam]

Peufeu,
What do you think the VAS atage is trying to do?
I like your earlier thinking about treating the input as a current node. So what about the output too? Is it really a voltage amp or is it a current amp which is being highjacked to form an integrator to keep the system stable?
Can you justify having 4 non-linear junctions in the signal path?

Say the maximum output of the amp is 20A. The max input current is 50uA. This is a current gain of 400k. If each transistor has a current gain of 100 (say) then you need a mimimum of 3 devices. You have used 4 just for the VAS stage.

Food for thought.

[andy_c]

Peufeu,

I see I've caused some confusion. I have no scanner, so I tried to describe the circuit in words. Bad idea! . I tried sketching out the symbolic circuit with the LTSpice schematic capture and it came out okay. I've shown it below. My thought was not to model the circuit at a device level, but rather to have the simplest possible "macro model" (for lack of a better term) for whatever combination of devices are used in the Vas. My hope was to get a result simple enough that I could obtain some intuitive information from it. So the term "gm" I used is not the device-level transconductance, but rather the y21 value of the composite network. It's only equal to the device gm for a standard CE amp with no emitter resistor.

I do realize that the circuit is very oversimplified, but what was important to me was the core ideas rather than ultimate accuracy.

Regarding the input impedance issue, you mentioned that the important thing was beta. But since hie=(beta + 1)*re, increasing beta has the effect of increasing the input impedance of a common emitter amplifier. So I think we're talking about the same thing, but just thinking about the problem in different ways.

You should be able to verify my loop gain expression using the circuit below.

[peufeu]

About jcx'x post :

I was about to write a stupidity :

"The VAS input impedance is rather low here because no emitter resistor is used."

Then I checked in the Simulator. Remember the input diff stage + current mirror has an output impedance of about 100K (roughly...).

A simple transistor, and a cascoded transistor, both have an input Z of about 1K (Ic = 10mA, BC557C) at DC, falling first order after the dominant pole because of Cdom.

This 1K is small relative to the 100K so the "current source inptu stage" is valid.

However when using a super pair which has a lot more current gain, VAS input impedance raised to about 100K (which is about the same as the input stage output impedance, so it may be masking the real VAS impedance).

Conclusion : As we increase VAS gain, its input impedance increases too, and magnifies any nonlinearity in the input stage output impedance.

Does it matter ? I'm not sure really...

Originally posted by jcx
Gm is much more predictable for BJT than current gain and Gm based modeling shows commonality of operating principle with FETs where current gain is a fairly useless concept [/B]

Yes, but in this case, if you model using voltage, can you tell me the output impedance of the input stage ? Not easy to calculate the output impedance of two collectors back-to-back...

[peufeu]

There's an easy way out though : put a resistor in parallel with Cdom and this will add feedback at DC too...

Problem is you need about 100 Megohms.

[traderbam]

The level of sophistication is growing!

Why do you need dc feedback around the VAS amp?
I mean, are you trying to make this stage set the voltage gain of the system or are you intending to have global feedback do that?

According to Hugh the more feedback (as Cdom) the worse the thing sounds - so why apply even more feedback with a resistor?

[peufeu]

What is the OL gain at DC ? It's actually impossible to answer this question, as it depends on so many parameters in the VAS, including transistor tempeartures, PCB insulation resistance, air hygrometry, DC leakage in Cdom... all of which can be signal-dependant.

Using a resistor would remove these uncertainties. Does it sound better ? I don't know. You'd gotta try.

[andy_c]

Since we were talking about the Hawksford cascode, I thought I'd mention an alternative that does a similar thing. I'd been playing around with simulations of distortion with the Hawksford cascode, and the distortion numbers looked very good. But the behavior of the amp in clipping was not good - it had a tendency to "bounce" off the rail rather than maintain a constant voltage in clipping. I tried using a Baker clamp to improve the clipping behavior, but it doesn't work right when integrated with the bias circuitry for the base of the common base amp. It ends up not working like a standard Baker clamp should, yet increases the distortion considerably as well.

Then I remembered a circuit that Wyn Palmer of Analog Devices used in the AD846 op-amp. I've shown a variant below. I've represented the ampllifier's output stage as an unity gain VCVS. Assuming Q13 and Q9 are matched, the collector currents of Q13 and Q9 will be equal. The collector-base voltages of Q14 and Q10 are nearly equal but for a Vbe drop. So assuming Q14 and Q10 are matched, the instantaneous base currents of Q14 and Q10 are nearly equal. The "cloned" base current of Q10 gets injected into the emitter of Q10, which increases and linearizes the output impedance of Q10, while also making the composite circuit's current gain closer to unity.

The disadvantage of this circuit is of course the increased complexity and need to match devices. The advantage is that the base of Q10 is now at a fixed DC voltage. Assuming the bias voltage source for the base of Q10 can provide the instantaneous current necessary to pull it out of saturation quickly, clipping will be very clean. With the ideal voltage source V5, simulation shows the clipping behavior to be about as good as it gets. Simulated distortion numbers are nearly the same as the Hawksford cascode. So after endless experimenting with the simulator, I think I've finally found the topology I want to use for my amp!

[peufeu]

If you replace Q10 with a Baxandall Super Pair, and get rid of q13/q14, would you get about the same effects without having to match devices ?