MOSFET Follies Clairification

[Steve Conner]

I think RG is right, even a single TO220 MOSFET should be capable of destroying your power tube grids.

I mean, you can buy MOSFETs that are capable of driving the speaker themselves, without any of this power tubes and OPT business.

On a more serious note, all of these circuits need some way of limiting the maximum grid current to avoid burning out the grids, MOSFETs and grid stopper resistors. And the limit has to be graceful, if your motivation for using AB2 is "softer clipping".

[defaced]

I strongly recommend a 12v Zener between gate and source of each MOSFET. The gate really is insulated, and if it gets more than about +/-30V away from the source, the device will fail sooner or later for "no apparent reason".

Some fets have the Zener built in, but not all of them, and the datasheet doesn't always say. In a linear circuit with enough voltage floating around to damage the gate, I just always use the zener, no exceptions.

Some applications need two zeners in series back to back, so the gate can swing down to -12v, but I doubt this is one of them.

A very good point and a definite omission from the circuit. This follower circuit pre-dates my understanding of gate/source breakdown by about a year and a half: http://music-electronics-forum.com/t25452/

[Gingertube]

Guys,
Late to this thread which is a little unfortunate as I have a fair amount of experience with these.
1st - Negative rail for the source follower. Rule of thumb is 3 times the bias voltage. In a Class AB Amp output tube grid swings from the bias voltage up to 0V BUT down to tube cut off. This voltage swing must be accommodated across the source follower load resistor.
2nd - You only need enough operating current to charge/discharge the Miller Capacitance at the output tube grid. For a guitar amp where you don't need response to tens of kiloherz AND the tubes are run in pentode mode (low Miller Capacitance) then you will find that 2mA is plenty. For Triode mode outputs or Ultralinear connection you will want maybe 3 to 5 mA (Guidance - for Ultralinear connected EL84 in a HiFi Amp I run 3.2mA, mind you that is for response flat to >60kHz).
3rd - The protection zener (cathode to mosfet gate, anode to mosfet source) is absolutely required. This should go directly across the mosfet pins.
4th - MOSFETs are like output tubes, areb prone to parasitic oscillation, gate stopper resistor in the mosfet gate feed is required, 1K will do with the resistor body as close to the mosfet gate pin as possible.
5th - You can limit the power dissipation in the MOSFET by using a low voltage +ve rail on the mosfet drain. This supply can be chosen to limit other problems. You will note that the mosfet source pin can swing up to (and maybe a volt or two above) 0V, the reverse capacitance from drain back to gate, Crss, varies with the voltage across the mosfet and so Crss is modulated by (varies with) the audio signal at the source. The change in Crss with drain to source voltage flattens out above Vds = 20V - so choose the drain supply voltage to maintain a minimum of say 25V across the mosfet when the source swings full positive (approx 0V at source). A +30V rail for the drain is fine. This supply needs good decoupling. You can of course use the existing B+, all that does is dissipate more heat in the mosfet, GOOD decoupling of the drain supply is still required. A bad +ve rail to the drain is audible as an agressive (harsh) edge to the top end.
6th - to keep device capacitance low use as small a mosfet as you can, IRF820 are NOT what I would choose. When ever possible I use ZVN0545A mentioned in the Mosfet Follies article but it has only a 600mW power rating. Find something with a say 2 Watt rating and low Crss.
7th - Keep temperature rise under control with heat sinks or a TO220 bolted to the chassis or similar. Mosfets run quite happily up to very high temperatures but gate capacitance doubles for every 8? (IIRC) degrees C rise. Keep the running temp below 100 degrees C.

If you don't mind sand in the audio path, which you obviously don't if you are putting in mosfet source followers then why not go the "Whole Hog" and use current source loads for the source followers. I've used simple "Ring of Two" transistor current sources with good results.

I hope this is of some help.

Cheers,
Ian

[Steve Conner]

Good stuff, but I have some questions:

5: How do you determine what is "full positive"? Power tube grid current doesn't limit anything any more. Clipping will either be by the PI running out of drive, or the MOSFETs bashing into their positive rail. The stuff we're working on here is for guitar amps, it has to handle massive overdrive without burning out or sounding bad.

6: Small MOSFETs are all very well, but if you want to drive grid current they need some power handling capability. This ties into the previous point, how much grid current are you planning to run and what measures you have in place to limit it. I have seen builders who never thought about that at all, then hit a big power chord and wondered why their grid stopper resistors exploded.

I've seen some cheap FETs in TO220 packages capable of switching in 5ns, which implies very low capacitances, but I forget the part number.

7: Are you sure?! Aren't you getting confused with gate leakage current in JFETs or something?

[exclamationmark]

Good stuff, but I have some questions:

5: How do you determine what is "full positive"? Power tube grid current doesn't limit anything any more. Clipping will either be by the PI running out of drive, or the MOSFETs bashing into their positive rail.

"Full positive" is defined as "when your grids melt" . Generally you have an idea on the max voltage you want your grids to get to, but achieving this limit in practice is sort of difficult. Perhaps the best solution is a compromise, where you size the grid stopper large enough to limit to voltage to a sensible number, but not large enough as to induce square wave clipping.

A lot of problems are caused by the differences in individual tubes. For example, if you carefully design your circuit to never go above a certain grid voltage (through PI clipping, or gain limitation), what happens when you replace tubes? The bias of new tubes can shift by 5 volts or more, which means your carefully chosen limit is shifted by +/- 5 volts, which is a huge amount in grid current land!

[Steve Conner]

Generally you have an idea on the max voltage you want your grids to get to, but achieving this limit in practice is sort of difficult.

Exactly, so let's hear some discussion on how (if!) you all have achieved it in practice. It's all right for Morgan Jones, he designs hi-fi amps and doesn't have to worry about overdrive.

For instance, the original Ampeg SVT used a pair of back-to-back diodes to limit the input to the power amp. That seems dreadful (ugh! silicon!) but in the context of this discussion it actually makes a lot of sense.

Maybe the cathode followers in the SVT were AC coupled, but that just shifts the problem from burning things out to blocking distortion.

My ultimate solution would be a Vactrol-based compressor that ran off screen and control grid current. Like Peavey's DDT for tubes.

[Steve Conner]

I got the part number for those MOSFETs, IPP50R520CP from Infineon.

[Gingertube]

Steve,
"How do you determine what is "full positive" drive?" - Good question, I wish I had a good answer for you. There are three things working for you:
- the input impedance of the output tube in "grid rectification" mode
- the grid stop on the output tube (which adds to the above).
- The output impedance of the source follower.
Unfortunately all I have is guesses for the relative values of these.

The attached schematic is NOT a Guitar Amp but the HIFI Amp I designed and built for myself - it is simply to shows an example of what I was talking about. This one ran 1.7mA idle in the source followers.
The 3.2mA mentioned in my 1st post was a mistake , that is what I ran in a 6SL7 + 2 x 6V6 version of this circuit.

It also has at least one more "novel" design feature which is not relevant to this thread, so don't get to hung up about those 47K comming of the output tube anodes.

Cheers,
Ian

[R.G.]

to keep device capacitance low use as small a mosfet as you can, IRF820 are NOT what I would choose.

Good point, and good description. When I did the original work on this, the IRF820 was the smallest-geometry MOSFET that I could find that would withstand the voltage and was reasonably available. It's not what I'd choose today, either. It's been some years.

When ever possible I use ZVN0545A mentioned in the Mosfet Follies article but it has only a 600mW power rating. Find something with a say 2 Watt rating and low Crss.

That was the problem with the newer and smaller Zetex devices if I remember correctly - hard to get enough power dissipation. Paralleling would probably help, but that also doubles the gate-drain and gate-source capacitances. I remember being more worried about the gate-source capacitance until I'd convinced myself of the dilutive effect of follower operation.

There is probably not a good single device that is a perfect match for the application. Perhaps a cascoded JFET/MOSFET. One day I'll get back to it.

If you don't mind sand in the audio path, which you obviously don't if you are putting in mosfet source followers then why not go the "Whole Hog" and use current source loads for the source followers. I've used simple "Ring of Two" transistor current sources with good results.

I took quite a bit of flack from the tube purists when I put that out. Not so much here, but I got some interesting emails about. A current source for the follower loads would make sense to get you some independence from the negative supply.

The power supply for this is perhaps the single biggest issue with it. It needs a supply that's not "organic" to the existing tube guitar amp supply, so only people motivated enough to derive the right supply for it will go to the trouble.

[Gingertube]

R.G.
Thanks - that point about the -ve rail isolation when using a current source load on the source follower (rather than a resistor) is often overlooked. It allows use of a much simpler -ve rail power supply.
See my other post below the one you split from to see the schematic of what I did in the HiFi Power Amp if interested. It has the current source loaded source followers. I posted that circuit over at DIYAudio and it has been built by about 50 folk around the world with no compaints.
Cheers,
Ian

[Tage]

These are the Mosfets I'm using for my source followers, STMicroelectronics STP3NK90Z. They also have nice and low capacitance and 7ns on time switching. ALthough these were reccomended in another thread, I ordered a few and it turns out they work for this application as well. I'm just wondering about the concern over their capacitance, is it just a HF roll off concern? or does the input capacitance effect the overall performance with mosfets as source followers?

[Gingertube]

Tage,
Don't get too hung up on the device capacitance. In a source follower the source follows the gate (hence the name) so effectively you are not changing the voltage across the gate capacitance at all. The reverse capacitance form drain to gate Crss is more significant and that was addressed above. It just makes good sense to use a device with as low a capacitance as practical.
Cheers,
Ian

[Tage]

Thanks Gingertube, I actually have just gotten a friends old EE textbooks and am reading up on Mosfet construction and applications right now. Very interesting stuff, but pretty dry reading

[R.G.]

Tage,
In a source follower the source follows the gate (hence the name) so effectively you are not changing the voltage across the gate capacitance at all.

Yep. I spent a long time contemplating my navel before I realized that the capacitance is effectively reduced by the transconductance; the gate-source capacitance has to be charged to change the gate-source voltage, but the source voltage follows the gate nearly perfectly. The "nearly" is the imperfection caused by the non-infinite transconductance of the device.

The gain of a source follower is incrementally (gm*Rst)/(1+gmRst), which produces an effective gain of nearly unity. Rst is the parallel combination of the real source resistance, the internal r0, and the body-effect Gm, gmb, in parallel.

The gate-source voltage changes by a fraction of the signal equal to 1-[(gm*Rst)/(1+gmRst)], or with some manipulation, 1/(1+gmRst). The term gm*Rst is large compared to one, and gets larger as gm and Rst get larger.

Since the gate-source voltage changes only this fraction of the signal voltage, the charge needed to change the capacitance voltage (and hence the current resulting from a signal) is effectively reduced by the same fraction. This looks to the signal input like the gate-source capacitance is smaller.

[defaced]

So as the young pup trying to learn, what does using a constant current source provide over a simple resistor as the MOSFET source load?