Current feedback single-ended input stage + push-pull VAS?

[fryingpan]

Would such a thing be possible? Talking about a power amp, of course. (And with either a cap at the output or a DC servo). Neither Douglas Self nor Bob Cordell talk about current feedback at any point in their books.

Why, I hear you say? Because this way I can use Slone's soft clipping VAS (complementary cascoded VAS, with current essentially set by a divider between the power rail and output), and with current feedback you should have less problems with stability.

[PRR]

Looks like a good question for diyaudio.com. Both Doug and Bob post there intermittently, and several theories of clipping get discussed.

You will get a LOT more reaction if you can draw it. Words are weak things.

[Rob Strand]

You will get a LOT more reaction if you can draw it. Words are weak things.

Especially:

Slone's soft clipping VAS

I don't have Slone's book.

I wasn't sure if "divider between the power rail and output" was the SOA protection.    But that's the output stage not the VAS.

[merlinb]

Emitter degeneration is current feedback, is that not what you mean?

[fryingpan]

What I mean is the following input stage:



married to the following VAS from Slone's "favourite amp":



The CFB input stage is a single transistor with the negative input coming in through the emitter (a low impedance node, therefore it works through current, not voltage). It can't correct for DC, therefore a DC servo becomes essential, unless you AC couple the speaker with the power amp through a capacitor.

The VAS is right there in the middle of the second picture, showing a double cascode (where the compensation also is). The aim is to have the clipping happening inside the power amp, not before (which means that, coupled with a higher output impedance through mixed mode feedback, it can better approximate the behaviour of a tube amp).

[amptramp]

There is an application note for a current feedback amplifier here:

https://www.analog.com/media/en/technical-documentation/application-notes/58052492001115525484056221917334an211.pdf

This is a 70 watt amplifier using IGBT devices rather than MOSFET's for the output stage and the distortion numbers are very low - 0.001% at a 50 watt output with a maximum output of 70 watts and low intermodulation distortion as well.  There is a thorough explanation of how the amplifier works.

Current feedback is a different world.  The things we ordinarily do to stabilize voltage feedback amplifiers like a capacitor across the feedback resistor to limit frequency response tend to destabilize current feedback amplifiers.

[antonis]

What I mean is the following input stage:

If by "input stage" you refer on Q1 (and relevant circuitry) you could try it, as long as you retain the global NFB circuit..

P.S.
Are you trying just to simplify Slone's favorite amp circuitry or you aim to something else..??

[fryingpan]

What I mean is the following input stage:

If by "input stage" you refer on Q1 (and relevant circuitry) you could try it, as long as you retain the global NFB circuit..

P.S.
Are you trying just to simplify Slone's favorite amp circuitry or you aim to something else..??

Well, yeah. Slone's favourite amp is kinda complex and overkill for guitar, I feel. Considering that I'm playing with the idea of a 20W @ 8 ohm (3% THD), 35W @ 8 ohm (6-8% THD), with as soft a knee I can get into clipping, and that I may as well use an output capacitor, the DC offset inherent in this kind of input stage becomes mostly a non-issue. The idea is to use lateral MOSFETs at the output for easy biasing (I've never made a power amp, and the inherent ease of use and robustness of lateral MOSFETs appeals to me), mixed mode feedback and, possibly, a DC servo in place of the output capacitor. I'm not aiming at high feedback (considering the distortion figures I've just mentioned). A push-pull VAS should be enough to drive the power MOSFETs without a driver (which is usually recommended for MOSFETs even though they don't strictly need them to "fight" against their high capacitance). I would also implement some current limiting (but I don't think I will implement V-I limiting), the only thing I'm really worried about is crossover distortion which is harder to cure on MOSFETs.

[Clint Eastwood]

Interesting. I hope you build it and let us know how it sounds 

[antonis]

May I ask about intended supply voltage and kind (single or bi-polar..)..??

[fryingpan]

May I ask about intended supply voltage and kind (single or bi-polar..)..??

Either +/-36V (ish) or 70V (ish). If dual, a DC servo would be more appropriate obviously.

Both taking into account Vds(sat), which means that about 10V per device are simply necessary to use the power MOSFETs. Although I don't quite understand the idea behind higher rails for the preceding circuitry (which I won't do).

[antonis]

May I ask about intended supply voltage and kind (single or bi-polar..)..??

Either +/-36V (ish) or 70V (ish).

If so, I'd suggest MPSA92 for Q1..
(and a bias voltage divider bootstrap configuration..)

[PRR]

The CFB input stage is a single transistor.... The VAS is right there in the middle of the second picture, showing a double...

Your single transistor has just 1 output. This Vas has 2 inputs. Sure you can tie the extra input to a dummy node, but a Vas already has a LOT of constraints (or failure modes) and I don't think this will be easy to patch-up so that it survives stage-abuse.

But study-up with Dan Meyer, Doug Self, Bob Cordell (who has a Klever Klipper?), et al, and maybe you can advance the state of tone.

[fryingpan]

The CFB input stage is a single transistor.... The VAS is right there in the middle of the second picture, showing a double...

Your single transistor has just 1 output. This Vas has 2 inputs. Sure you can tie the extra input to a dummy node, but a Vas already has a LOT of constraints (or failure modes) and I don't think this will be easy to patch-up so that it survives stage-abuse.

But study-up with Dan Meyer, Doug Self, Bob Cordell (who has a Klever Klipper?), et al, and maybe you can advance the state of tone.

In his power amp book, Self dedicates a whole chapter to the push-pull VAS and he does talk about the possibility of feeding the two inputs from one output. Performance is not stellar by hi-fi standards, but it could work for guitar? But sure, I could buffer the input stage's output, which should not cause too much concern for stability provided that the passband of the amp is kept low. (I don't think I need flatness up to 100kHz. Considering that a guitar amp has, at most, 10kHz of passband, the input signal rolls-off at about 5kHz, even considering gross distortion of about 25%, and 35Vp, we can safely assume that about 1V/us is all we need as slew rate, so even some strong compensation shouldn't cause problems). Or instead of a buffer, a current mirror.

[Clint Eastwood]

You could use a symmetrical input stage like the Hiraga class a amp, or like his 'monstre' with jfets.


 

[Rob Strand]

The VAS is right there in the middle of the second picture, showing a double cascode (where the compensation also is). The aim is to have the clipping happening inside the power amp, not before (which means that, coupled with a higher output impedance through mixed mode feedback, it can better approximate the behaviour of a tube amp

Thanks that makes more sense.

The way I see the VAS working is simply to reduce the VCE voltage seen by each cascode transistor by half.    For the most part it won't do much as the current through each transistor is the same.   It's possible something could happen when the output stage hits the rails.   It doesn't seem inherently soft to me although something may show up in the details.   (It's a cascode with a variable base bias voltage.)

Back in the old days when power transistor voltages weren't so high there were a few power amplifiers which connected two sets of output stage transistors in series.  Each side/polarity of the output transistors was made up of two transistor in series.  One of the output stage transistors used a divider very similar to the VAS.

The idea is still alive here in a different form:
https://electronics.stackexchange.com/questions/285253/connecting-transistors-in-series

As for grafting the Slone cascode onto the simple Elliot sound amplifier.  It would be very easy to do for the negative swing, which has the VAS transistor.   The question is what to do with the other polarity.   For a fully symmetric design it's pretty obvious to mirror each polarity - that makes the circuit much more complex.   For a "current source" made up of two resistors and a bootstrap cap like the Elliot sound circuit (the VAS load) it can be done but perhaps doesn't make a lot of sense.   If you replaced the two resistors and bootstrap cap with a solid-state/BJT current source then it makes a little more sense.

I'm not sure if the author (Slone) makes any claims that the VAS is a soft clipper?

Here's the output transistor version applied to an audio power amp,
https://www.tubecad.com/2023/01/blog0575.htm

[fryingpan]

There you go.





As you can see, the author makes direct claims that the cascode transistors will soft clip when the output approaches the supply rails. I think he gets the "maths" wrong, since the base voltage is not a function of the difference between the two voltages (supply and output) but rather the mean.

I don't understand why the single output of the input stage cannot be fed into both "signal" transistors in the VAS. Is it a matter of current?

[Rob Strand]

As you can see, the author makes direct claims that the cascode transistors will soft clip when the output approaches the supply rails. I think he gets the "maths" wrong, since the base voltage is not a function of the difference between the two voltages (supply and output) but rather the mean.

Interesting.   I might have to have a closer look at it.  His waveform looks very soft.  I just can see the VAS doing that (I could be wrong as something could show up in the details.)

The base voltage derived from a divider so the base voltage is the instantaneous.   If you were to put a cap across one of the divider resistors then it would start to see the mean;  I think the tubecad link has an example.

I don't understand why the single output of the input stage cannot be fed into both "signal" transistors in the VAS. Is it a matter of current?

It will work. .. however, *if* the VAS is softening the clipping then it will only soften when it approaches 0V across the VAS transistor.   If you want soft clipping on both polarities then you need the effect to occur on both rails and that requires a cascode ckt for each polarity (that why I spoke about different ways to implement it).

I have a feeling the soft clipping has something to do with the MOSFETs and not the VAS.   When the VAS swings close to the rail it pulls the gate close to the rail.  However in order that the MOSFET passes current that requires the drain to sit below the gate voltage (+swing/N-channel case).  When you do that the VDS of the MOSFET is quite low and the MOSFET starts to operate in the triode region giving it softer clipping.

When the old Hitachi MOSFETs came out around 1980 those amps sound nice when clipped because it had a softer character.   I think some of the Trace Elliot bass amps had Hitachi circuits.    The sound was due to the MOSFETs.

[fryingpan]

As you can see, the author makes direct claims that the cascode transistors will soft clip when the output approaches the supply rails. I think he gets the "maths" wrong, since the base voltage is not a function of the difference between the two voltages (supply and output) but rather the mean.

Interesting.   I might have to have a closer look at it.  His waveform looks very soft.  I just can see the VAS doing that (I could be wrong as something could show up in the details.)

The base voltage derived from a divider so the base voltage is the instantaneous.   If you were to put a cap across one of the divider resistors then it would start to see the mean;  I think the tubecad link has an example.

I don't understand why the single output of the input stage cannot be fed into both "signal" transistors in the VAS. Is it a matter of current?

It will work. .. however, *if* the VAS is softening the clipping then it will only soften when it approaches 0V across the VAS transistor.   If you want soft clipping on both polarities then you need the effect to occur on both rails and that requires a cascode ckt for each polarity (that why I spoke about different ways to implement it).

I have a feeling the soft clipping has something to do with the MOSFETs and not the VAS.   When the VAS swings close to the rail it pulls the gate close to the rail.  However in order that the MOSFET passes current that requires the drain to sit below the gate voltage (+swing/N-channel case).  When you do that the VDS of the MOSFET is quite low and the MOSFET starts to operate in the triode region giving it softer clipping.

When the old Hitachi MOSFETs came out around 1980 those amps sound nice when clipped because it had a softer character.   I think some of the Trace Elliot bass amps had Hitachi circuits.    The sound was due to the MOSFETs.

Oh, but I would implement the VAS as a push-pull with a cascode on both polarities (two PNPs at the top rail, two NPNs at the bottom rail), it's just that I wouldn't implement the input stage as a symmetric pair (it seems overkill to me).

Also, with the mean I meant the arithmetic mean of the two instantaneous voltages. You have a voltage divider between V+ and the output. It's V+/2 when the output is nil, and it gets linearly closer to V+ as the voltage increases. In order to actually track half the difference you would need to divide the sum of V+ and Voutput inverted.

[Rob Strand]

Perhaps a little crude but I did this basic simulation and it's not showing soft clipping,

The difference signal is x100.


I also tried changing the bootstrapped resistor current source on the VAS to a transistor current source.   I added a second cascode off the BJT current source.  That didn't show any soft clipping either.