Class-A Mosfet Source Follower

So---125mA (4V peak, 32 ohms) + say another 20mA?

[quote author="bcarso"]So---125mA (4V peak, 32 ohms) + say another 20mA?[/quote]

Yes.

In the original schematic from sbeach's link output DC is -4V. Current through FET is 11/250 = 44 mA. Maximal negative voltage swing will be (11/(100+150+32))*32=1.25V, i.e. 3 times less with 2.5 times more of a voltage from the power supply. I don't speak of distortions now, you may use SPICE model to check...

I'm not really interested in the original thread design, which I agree is not likely particularly high performance. I wanted to get the parameters of yours from the other thread so as to be able to simulate. Thank you for providing that information.

I'll switch over to your original thread eventually.

[quote author="bcarso"]I'm not really interested in the original thread design, which I agree is not likely particularly high performance. I wanted to get the parameters of yours from the other thread so as to be able to simulate. Thank you for providing that information.

I'll switch over to your original thread eventually.[/quote]

Ok! I hope you'll have some interesting results!

> I found a class-A MOSFET current boosting circuit on a headphone-oriented web site

In fact, you found it at HeadWize, where its flaws have been noted, but the webmaster is a busy student and has not found time to correct, update, or remove it.

> Don't you need to bias the MOSFET gate by applying the appropriate voltage for the device? That's the reason for the trimpot...

What is the voltage at the opamp's output????

To simplify: Ground voltage, plus opamp's small error voltage times the generally low DC gain of the opamp's feedback network. The opamp's unloaded error is maybe 1mV, the gain is 2 to 30, the output will be within 30mV of ground voltage.

How much trim of Gate voltage does the "bias network" give?

Note that my simplified analysis of opamp output didn't consider the "bias network", because "usually" the load on an opamp (within limits clearly hinted in the specs) does NOT affect its output. To add this correction: we may say the opamp has a small but non-zero output impedance, reflecting the fact that we have to push the input harder to make the output do real work. This may be as high as 300 ohms for TL071, but under 50 ohms for nearly all other general purpose opamps. Say 100 ohms. And say we jam the "bias network" all the way to one side and break the other side. We have about 100K pulling to 15V, 100 ohms pulling to zero. This extreme "de-biasing" has shifted the opamp's output only 15mV before NFB, and thousands of times less with the NFB in operation.

The plan was stolen from some scheme withOUT the op-amp, where we really did need a DC reference for the FET Gate. What is the ideal Gate voltage? As Wavebourn says: start from your output relations! The resistor-coupled power stage can be done several ways: maximum power, maximum voltage, maximum efficiency. A simple analysis shows that the cathode (or whatever) should sit either at 1/2 or 2/3 the total supply voltage, so zero or +5V for this plan as drawn. The grid offset determines the grid voltage; for modern MOSFETs at very low (for their size) current it will be 1V to 3V higher, +1V to +8V depending what device and what you optimized.

The exact output voltage is not critical. We are cap-coupled to the load. Large "errors" of output voltage will reduce output power and voltage and efficiency, but 1 or 2 or even 3V error is small compared to our 30V supply.

But this 121K-5K-118K string gives -0.123V to +0.49V gate bias. What possible difference does this make within a 30V framework and MOSFETs with 1V to 3V gate-source offset?

> resembles a dish made from good salad, good gamburger, blueberry jam, cookies, sugar and pepper, all taken from good cookbook, but mixed together.

Certainly the "FET bias" came from a different cookbook page than the direct-coupled op-amp. And the resistor values don't give enough swing to make any sense, like putting a eighth-dash of parsley in 55 gallons of chili.

Part of the art of elegant engineering is knowing what to put in and what parts are not needed in the new combination.

> Would it be possible or advisable to include the current booster MOSFET within the opamp's feedback loop?

Given the AC coupling: dubious.

As Wavebourn says, there's a lot of NFB in the naked cathode follower. "100%" is not a good indicator though; 100% of what? Well, of the same device and load working as a voltage amp. The IRF150 at those conditions would have a gain of maybe 10, so it really has 20dB NFB. But its flaws are, up to clipping, fairly simple and inoffensive. Such "crude" schemes often sound fine.

BTW: a BJT has higher THD as a volt-amp, but very much higher gain, maybe 100 instead of 10. When used as a cathode follower, the 40dB NFB will put the BJT's THD numbers very low.

Do note that all resistor-coupled "power amps" have a maximum efficiency of 6%. This is fine for 1mW output, but gets you into heatsinks for even 50mW output. Changing the 150 ohm resistor to a current source gets you to 25% efficiency. Push-pull A gets you to 50%, but you lose the simple bias-scheme.

There is a very good reason to NOT close the feedback loop globally. You can't buy small MOSFETs. The Gate capacitance, hung on the opamp's output, destabilizes the opamp and the whole thing becomes a radio transmitter. There is no simple answer, because at the frequency that any good opamp can reach, the input capacitance of a large Gate is lower than the load impedance, so we need an opamp which can drive the load directly, so why are we putting the MOSFET in there?

A TIP32 in the same place is stable, until the BJT gets close to cut-off. That's going to be a problem with any wide-band gain stage feeding an output device which is smacked to maximum power and thus swings near zero current and zero bandwidth.

The root problem is: a NFB loop has to have a very smooth single-pole rolloff (or very tricky pole/zero compensation). But all multi-stage amps have multiple poles. You pick the one where you paid the most: the output stage. You leave that alone. You pick an earlier stage, usually the input stage, and clobber that so its gain rolls smoothly to unity overall gain before the output stage starts to droop. Note that a source or emitter follower does NOT have a large bandwidth: voltage gain stays unity but the input impedance falls. A MOSFET has gate capacitance to a supply rail and its input impedance falls to zero at high frequency. A BJT's Hfe falls to unity at Ft. And any "power" device (gotta be "power" because of the 6%-max efficiency) you can find has a lower bandwidth than the OPA134's own output stage, the one is was designed to be stable with.

BTW, why is it drawn with + and - supplies? Both the input and output are AC coupled, and the feedback could be too. Single-supply is always cheaper, and would eliminate the apparent ambiguity about what polarity the output cap needs to be.

There is a reason to want the global feedback. Use a bipolar supply and use the opamp to force the MOSFET's/BJT's output to near-Zero DC voltage, eliminate the costly and dubious-sounding output cap. If you wander back to Headwise and search on "Tori amp", several years back, you may find such a plan. If you find the one with a MOSFET, keep going.... I'm sure the BJT version is better. Others report it sounds fine. It should sound very much like the raw op-amp used, only at higher current and power levels. However a much older yet similar plan made for live recording is called "HotBox" for very good reason. I ended up with heatsinks from a 70 watt class AB amplifier. This is a terrible power amp plan, no matter how much strawberry jam you smear around it.

But don't let people tell you why it sucks. It will drive headphones exactly as drawn. Do try fiddling the "bias network" and seeing what it does. Take it out.... what happens? Try closing the global feedback. It will probably oscillate ultrasonically, but even so you may see the opamp force the output DC to near-Zero instead of a couple volts negative. The audio will be low-quality while the loop oscillates hypersonically, though this may not be grossly obvious.

Noticed another blunder in the original schematic: the decoupling Rs from the power supply are shown as 100 ohms each. But the the 150 ohm MOSFET source resistor is after the lower 100 ohm! As the lectroid says in Buckaroo Banzai, Very Baaad Design.

Available MOSFETs do tend to be large, and it is a shame we as designers/experimenters can't get intermediate geometries to play with. Even the 2N7000 is a bit clumsy for many apps. I recently had to "fix" a FET limiter for a client's subcontractor that used a 7000, much as I wanted to tear it out and do something with a JFET or another variable-gain element altogether. It was possible to get decent performance out of the 7000, but took some work and additional parts.

One finesse on the big MOSFET gate capacitance however: The gate-source component of it does get mostly removed to an extent as an input load by the bootstrapping of the source follower configuration. However, the gate-drain C is there, and it is also rather large, especially at low drain-gate voltages. And use it in a grounded-source voltage amplifier and Miller effect takes its toll and really slows things down.