Mosfet Transformer Power Amp Performance

[Bo Deadly]

Consider the following circuit:

Can someone recommend how I might get a little better linearity out of this?

Ultimately the intent is to explore the "compression" effects of a single supply non-complementary push-pull by playing with the supply impedance. And I also want to explore dynamic bias to produce "explansion". But that's a different discussion. For now I'm just trying to understand what sort of performance I can hope to get from a circuit like this. Specifically, I'm trying to get the maximum linearity with this circuit topology.

Also I am seeking the lowest quiescent current possible vs the maximum power. Meaning ideally it should have very low quiescent current but when there is significant signal the current drawn from the supply is many times greater. So if I could make that ratio 10:1 and get sub-1% linearity, that would be fantastic.

If I look at the waveform visually it seems like as I adjust the bias it goes from horizontal crossover distortion to a triangle wave to vertical crossover distortion. So what is causing the triangle wave effect and how do I minimize it?

UPDATE 1:

For posterity here is the latest circuit  and SNR / THD results:

Currently, with the circuit as shown and a ~27.5 mA bias Iq(R1), according to my QA400 analyzer I get:

SNR: 62 dB
THD: 0.28 %

Not bad for a guitar amp. But it might just be the limit of what I can get out of this old transformer. The harmonics responsible look fairly harmless:

As I raise the input level however the distortion goes up dramatically. I just did the test at a particular level where the THD was minimized (-10dB on the QA400). As the level gets higher, the waveform visually has the horizontal typical crossover distortion.

I also have to correct my comment below about the drain resistor improving performance - it does not. If the output level is adjusted to be a particular level, changing the drain resistor from 2.6 to 1.6 to 1.0 has little effect on THD.

I hope to explore a dynamic bias (adding voltage to servo+ or subtracting current from the servo-) such that the bias is increased if the signal goes above a certain threshold. My thought is that this would reduce said distortion and basically make a sort of expander or noise gate since gain increases with bias. Then I might be able to push the whole thing into clipping without going through the crossover distortion.

UPDATE 2:

The servo bias in the above circuit does not work. There is significant DC shift across the common drain resistor as the signal level is increased such that it causes the bias to shift lower. Even with a relatively slow servo, if the signal is sustained, it will start to bias off the transistors and cause serious crossover distortion. If I use a simple fixed bias the circuit is much more consistent at all signal levels. Of course it should use a proper temperature compensation circuit but I think I've played with this enough for now ...

 

[moamps]

IMO, you need to implement some kind of the NFB in the circuit to get lower THD.
The cause of higher THD at higher Iq can be mismatch of the FET's Iq, and consequently the core saturation.

 

[Bo Deadly]

Interesting point about the FET mismatch. I will try different FETs and see what effect it has on THD.

One thing that I have noticed dramatically improves things is if the drain resistor is smaller. I'm hoping to get it down to less than 1R and use a servo to drive the bias.

 

[Samuel Groner]

Giving each transistor its own source resistor is one easy thing to consider. Making the resistor value equal to the inverse of the transconductance of the transistor at quiescent current might be a good starting point for minimum crossover distortion (for BJTs, this condition is pretty accurate, for FETs it's a rough estimate only).

Samuel

 

[abbey road d enfer]

Can someone recommend how I might get a little better linearity out of this?

Unfortunately, a circuit that exploits the direct characteristic of a FET cannot be linearized AND retain the desired dependance on voltage and current operating point. Applying NFB would reduce thd but make the circuit more or less impervious to its environment.

If I look at the waveform visually it seems like as I adjust the bias it goes from horizontal crossover distortion to a triangle wave

Isn't it simply that the gain increases as quiescent current increases?

to vertical crossover distortion.

Looks to me like the power rail sagging does this. Check with an o'scope.

 

[Bo Deadly]

Can someone recommend how I might get a little better linearity out of this?

Unfortunately, a circuit that exploits the direct characteristic of a FET cannot be linearized AND retain the desired dependance on voltage and current operating point. Applying NFB would reduce thd but make the circuit more or less impervious to its environment.

If I look at the waveform visually it seems like as I adjust the bias it goes from horizontal crossover distortion to a triangle wave

Isn't it simply that the gain increases as quiescent current increases?

Yeah, that makes sense. I suppose a tube is actually a little better for this sort of thing because it's transfer function follows a 1.5 power law instead of the square law of a transistor.

I would like to avoid NFB for the moment but if I do want to do that, I suppose I could put an op amp in front of the DRV135 and have 2 feedback paths, one local and one global. A little global NFB might have a dramatic effect on the sound.