Bob Cordell's Power amplifier book

I am not sure what your point is. I agree that Cdom reduces OLG and therefore does not permit as great a reduction in crossover distortion at high frequencies as at low. I am not saying that crossover distortion is caused by Cdom, and of course Cdom is widely used because it is stable.
Crossover distortion can be reduced if the OS can be included in the Miller scheme. That is why I suggested transitional Miller seems about the optimum.
Regarding the use of cascode, indeed it can cause stability issues. ALthough it gives apparently a higher frequency response it adds a pole. That can be compensated for by adding a small capacitor roughly that of Cjc as a local Miller to restore the response to a single pole. But the reduction in Early effect distortion is probably worth the effort, particularly if additional supply rails are added. It does depend on the degree to which local feedback is incorporated into the VAS stage. If, as Bob suggests, a 10:1 is used then I would agree that that is probably enough to largely eliminate that distortion.
BUt any local feedback while it may linearise that stage reduces OLG again which means less feedback to reduce output stage distortion.
 
I am not sure what your point is. I agree that Cdom reduces OLG and therefore does not permit as great a reduction in crossover distortion at high frequencies as at low. I am not saying that crossover distortion is caused by Cdom, and of course Cdom is widely used because it is stable.
Crossover distortion can be reduced if the OS can be included in the Miller scheme. That is why I suggested transitional Miller seems about the optimum.
Regarding the use of cascode, indeed it can cause stability issues. ALthough it gives apparently a higher frequency response it adds a pole. That can be compensated for by adding a small capacitor roughly that of Cjc as a local Miller to restore the response to a single pole. But the reduction in Early effect distortion is probably worth the effort, particularly if additional supply rails are added. It does depend on the degree to which local feedback is incorporated into the VAS stage. If, as Bob suggests, a 10:1 is used then I would agree that that is probably enough to largely eliminate that distortion.
BUt any local feedback while it may linearise that stage reduces OLG again which means less feedback to reduce output stage distortion.

These are good points, but I have just one caveat. While local feedback uses some OLG, in a modern design there is a huge amount of OLG to go around. In fact, over most of the audio band with simple Miller compensation, OLG available for global feedback is only limited by the compensation roll-off that has to bring loop gain down to zero before too much lagging phase shift is introduced. This gain crossover frequency usually occurs in the region of 500 kHz to 2 MHz.

If loop gain falls to unity at 2 MHz at 6 dB/octave due to compensation, available distortion-reducing loop gain at 20 kHz is 40 dB, virtually independent of local degeneration in the IPS or VAS in a modern design using a current mirror load for the IPS, a 2T VAS and maybe a 3-EF output stage.

I use 10:1 degeneration in both the (bipolar) IPS and VAS and am never lacking in open-loop gain.

Cheers,
Bob
 
There is a side effect to degenerate the VAS.

There is a pole in IPS (I am referring to long tail pair configuration with current mirror). The pole is directly related to the input impedance of VAS. Degenerating VAS will increase input impedance of VAS, which causes the pole move to a lower frequency. It lowers the phase margin of global nfb.
 
Last edited:
Hi Mark,

That's not my experience. Distortion due to the nonlinear Cob of the VAS, is not reduced by Cdom. One might reason that Cob is "swamped" by the much larger value of Cdom, but the distortion will be still there, undiminished.

Cheers, E.


Hmm, that's not the received wisdom - nor does it agree with the equations for a transimpedance stage if you look at linearity of voltage out to current in, which is a ratio - adding Cdom reduces the (transimpedance) gain so the input current has to increase, which by definition swamps the capacitance-induced error current for the same signal swing.


Although its so hot today my brain is struggling!


A quick simulation could settle the matter I think.
 
There is a side effect to degenerate the VAS.

There is a pole in IPS (I am referring to long tail pair configuration with current mirror). The pole is directly related to the input impedance of VAS. Degenerating VAS will increase input impedance of VAS, which causes the pole move to a lower frequency. It lowers the phase margin of global nfb.

There is a much greater side effect when you install an emitter follower between the IPS and the VAS. This increases the input impedance of the (EF + VAS) by a factor of 200X (the Beta of the EF). And if you install a MOSFET follower (instead of a BJT follower) between the IPS and the VAS, the input impedance increases even more dramatically. Nevertheless, amplifier designers include followers routinely, and the resulting amps operate quite beautifully.




_
 
Last edited:
Hi Mark,

That's not my experience. Distortion due to the nonlinear Cob of the VAS, is not reduced by Cdom. One might reason that Cob is "swamped" by the much larger value of Cdom, but the distortion will be still there, undiminished.

Cheers, E.

Hi Edmond,

I think you are referring to a 1T VAS, where Cob is essentially in parallel with Cdom. In a 2T VAS, the EF up front breaks this path for Cob of the VAS transistor.

Cheers,
Bob
 
There is a much greater side effect when you install an emitter follower between the IPS and the VAS. This increases the input impedance of the (EF + VAS) by a factor of 200X (the Beta of the EF). And if you install a MOSFET follower (instead of a BJT follower) between the IPS and the VAS, the input impedance increases even more dramatically. Nevertheless, amplifier designers include followers routinely, and the resulting amps operate quite beautifully.




_

There is "pole splitting" in play.
It is rather tricky to talk about gain for BJT, voltage gain or current gain.

Let me try it again, here.
We assuming the current goes through bjt base is very very small that you can ignore, comparing the current goes through miller cap. Thus, the only input impedance you see before VAS is about 1/gm. (gm is the entire gm of VAS)

Actually adding a EF does not change gm. Thus, input impedance remains same.
Adding a degeneration resistor to vas decreases gm. Thus, increase input impedance.


BTW, I believe (EF + VAS) is a more superior configuration.
 
Last edited:
A quick simulation

Hmm, that's not the received wisdom - nor does it agree with the equations for a transimpedance stage if you look at linearity of voltage out to current in, which is a ratio - adding Cdom reduces the (transimpedance) gain so the input current has to increase, which by definition swamps the capacitance-induced error current for the same signal swing.
Although its so hot today my brain is struggling!
A quick simulation could settle the matter I think.

Hi Mark,

Before I put my comment, I made sure that this -rather counterintuitive- statement does hold by simulating the VAS distortion.

Cheers, E.
 
Hi Edmond,
I think you are referring to a 1T VAS, where Cob is essentially in parallel with Cdom. In a 2T VAS, the EF up front breaks this path for Cob of the VAS transistor.
Cheers,
Bob

Hi Bob,

Indeed, I was referring to a 1T VAS. To make the story complete, I've also simmed a 2T VAS (in isolation!). As expected, the distortion (due the nonlinear Cob) was much lower, about 1ppm. But adding Cdom had the opposite effect: THD rose to about 1.5ppm.

Cheers, E.
 
Hi Bob,

Indeed, I was referring to a 1T VAS. To make the story complete, I've also simmed a 2T VAS (in isolation!). As expected, the distortion (due the nonlinear Cob) was much lower, about 1ppm. But adding Cdom had the opposite effect: THD rose to about 1.5ppm.

Cheers, E.
You measured the distortion of the VAS by itself, or the overall distortion?
This may be a consequence of loading interaction between distortion mechanisms, which gets complicated and which I specifically was not considering here. The VAS in isolation should be linearized because you are adding more linear feedback (assuming the cap is NP0 of course 🙂
 
You measured the distortion of the VAS by itself, or the overall distortion?
Hi Mark,

I "measured" (read: simulated) the overall distortion, but since the rest of amplifier (input- and output stage) comprises ideal components, the only source of distortion is the VAS. See picture for the 2T VAS.
This may be a consequence of loading interaction between distortion mechanisms, which gets complicated and which I specifically was not considering here. The VAS in isolation should be linearized because you are adding more linear feedback (assuming the cap is NP0 of course 🙂
True. But at the same time the gain of the VAS is decreased by the same amount. As a result, the overall distortion is still the same, or even (slightly) worse.
As Bod already suggested: "when Cdom was added was a result of Cdom's loading of the 2T VAS, increasing required current swing a bit."
IOW, the input stage has to deliver more current, which -inevitably- will lead to a little more distortion.

Cheers, E.
 

Attachments

  • VAS.jpg
    VAS.jpg
    30.1 KB · Views: 246
Cdom linearizes the current in -> voltage out characteristic of the VAS where it is dominant (which is not a given).

However increasing Cdom will not reduce the distortion of the amplifier barring edge cases.

This trope of "Cdom has a linearizing effect" is taken way, way too far. It is intuitive, yes, and it's very very wrong in the way everyone seems to use it here. Like some kind of virus.

Cdom just adds to the current through Cob, it does not eliminate it in any way, and it also adds whatever distortion is present at the VAS output.
 
Last edited:
I don't completely agree with some of the terms you are using here. I think it adds confusion to the discussion. What people have been discussing here about Cdom has not been way, way off.

One also has to be very very careful about whether one is talking about a 2T VAS or a 1T VAS.

The Miller feedback around a 2T VAS is simply shunt feedback, and it does provide linearization in the frequency bands where its local loop gain is adequate. This means that there can be exceptions at very low frequencies, but those exceptions can add to confusion if they are given too much prominence in the discussion.

Cheers,
Bob