Will this work as audio amplifier?

[lumanauw]

While trying to understand how a differential pair works, I came up with this idea. The differential pair Q3 and Q4 can be dragged up close to the rail, because the CCS1 can be so low in voltage requirement. If the differential is close to the rails, it can perform as VAS. So Q3 and Q4 is VAS differential. Q4 will drive output stage, but it will need signal depended current change in Q3.

The input is fed to Q1. Q1 is only biased about 156uA (very low) to get sensitive operation. I dont know about noise figure, maybe bad. Q1 drives 25k resistor that fed the signal to Q3.

The problem is, how to get current fluctuation on Q1 that is dependent on input signal. If the emitors of Q1 is connected to CCS, it wont work, because Q1 will only form a common base transistor. So, the emitors of Q1 will need fixed value for reference. This reference is formed by Q2 (from its base). Q2's base is driven by a servo, I assumed about -1V2, so the output is steady at ground voltage.

Q2 is the escape path for the current in Q1+Q3 and works also for "reference" for Q1.
I can use Q1 directly to Q4's emitor, without 25k and Q3 so it will perform folded cascode. But I wanted to asked about this particular design.

I know the dissipation in the transistors are big, but it can be tackled by small heatsinks. The output also will drop some significant voltages from the rail.

1. Whats the effect that the input stage is running in very low bias (156uA at Q1)? Will it have bad noise figure (audible noise)?

2. What will be the harmonic properties in this configuration. Will it have bigger 2nd harmonic than ordinary 3 stage design, because input is only fed to 1 transistor (Q1)? Differential at input stage tends to cancel 2nd harmonic, dont they?

[lumanauw]

What is the mechanism of 2nd order canceling in differential pair?

[Richard C]

Is Q4 base supposed to be tied to +50V?

[lumanauw]

ARGHHHHH.....,
You are right. I draw wrong. Q4's base tied to the bottom of zener. It biased 3V9 from 50V rail.
Q1 and Q2 are sort of differential, but using 1Nch and 1Pch, so Q2's base will sit on -1V2.

[MikeB]

Originally posted by lumanauw
What is the mechanism of 2nd order canceling in differential pair?

If i understood it correctly, odd harmonics are symetrical distortions,
even harmonics asymetrical distortions.
As a diffpair works symetrical, (1st transistor closes, 2nd opens
exactly the same amount) it can only produce odd harmonics.
It doesn't cancel K2, it simply can't create them.

If you add a currentmirror, this looks completely different...

Mike

[thanh]

MikeB! Do you want to say that Lumanauw's diff input Q1 and Q2 can cancel odd order harmonics?

[lumanauw]

The reason I put Q1 and Q2 that way is to advoid 2nd harmonic cancelation (because wanted to try how "nice" it is to have 2nd harmonic not canceled, some sort of experiment, if it works).

If I put Q1 and Q2 same type Nch transistor, it will perform differential.
From how I look at differential, it always have 3 current path, 1 is downward (usually to CCS) and 2 upward to each diff transistor.
In the configuration I make, I use 1 Nch and 1Pch, with both emitors attached to each other, like differential. The difference is that in this configuration, it will only 2 current path (from C of Q1 to C of Q2), not 3 like ordinary differential.

But the consequence is that we have to put a reference voltage in base of Q2, so it will have signal depended current change flowing from Q1 and Q2----> needed to activate the VAS differential of Q3 and Q4, to perform audio amplifier.

The drawings I attach if for DC coupled power amp. If we put cap in the input (before 1k resistor attached to base of Q1) then base of Q1 will need another resistor to ground, for putting ground reference in base of Q1 (so it will be 3 of them, 1=1k=input resistor, 2=100k=feedback resistor, and 3=for ground stand point).

If we have #3 resistor attached, there will be current flowing from collector of Q1 to collector of Q2, when the base of Q2 is about <-1V2 (>of 2 VBE) from ground (because Q1's resistor is about ground voltage by R#3, and Q2's base is modulated about -1V2 by the servo, or by a trimpot. I use servo, because the change in Vbe because temperature will add, not cancel each other like differential, because we use 1Nch and 1Pch. In differential this thermal Vbe change will cancel each other--->DC offset can be maintained without servo).
Q1 and Q2 will perform some kind of signal depended current switch.

[MikeB]

I've spent some hours yesterday with spice, in theory it should
be possible to build a diff amp (or something else) that cancels
out 3rd harmonics, but its very difficult. Had no real luck yet...
The problem is, that even harmonics are by nature shifted 90deg
in phase, so adding (subtracting) complementary signals automatically
cancel them (-90deg <-> +90deg -> 180deg = negative).

The odd harmonics are not phaseshifted, so the idea might be to
somehow "reverse" the transfercharacteristic of a bjt.

I believe there is a reason, why most famous ClassA-Amps don't
use diffamps. (like zenquito,hiraga,JLH etc.)

The simplest solution yet is to use CFP in the diffamp, it does a
great job in reducing harmonics, but does not cancel them.
(CFP "only" linearizes, reducing thd of the diffamp to ~0.02%,
without GM-degenerating, and boosting the gain to 1:1500)

At this moment i'm trying a circuit with CFP in the diffamp, i think
it should be functional this evening. I can't wait hearing it...
But maybe there is a reason, why nearly noone uses CFP in the
diffamp ? (Except steven and some cyrus-amp)

Mike

[lumanauw]

I also wonders about this CFP differential. MikeB, since you have spice, could you trace how low bias current can to in the first transistor in CFP differential? The Cyrus is about 100-150uA, but is that can be made lower?
Doug Self stated that the optimal value for this CFP front resistor is 2k2 (that gives about 270uA bias). He said if the resistor is bigger than this, the noise figure will increase.
What happens if this resistor is 10k or 22k? What noise will be generated? Bias so low=bad noise?

[lumanauw]

This one uses 4k7. How big is the limit of this R? 10k? 100k?