HV SS Regulators?

[mhuss]

There are already about 1uf of film caps across the two outputs. I imagine it would be worse without them.

[R.G.]

Actually, I meant just across the zeners themselves.
Reverse-avalanched junctions make noise. Many classic noise sources start with a zener run at low current and then amplify that. Feeding a zener from a very low current source can make this even worse. I worried a bit about that when I was going through the design.

The short version is that the MOSFETs are probably buffering the zener voltage, including any noise. This makes for a very low-impedance output voltage, but apparently that is enough to let the low impedance noise drive the 1uF caps. I'm a little surprised at that part of it.

My suggestion of caps across the zeners themselves puts filtering on the high impedance loop across the zeners themselves, avoiding the buffering effect of the MOSFETs. At such high impedances, a given capacitance has much more effect.

There's a long harange on zeners, impedances, the conflicting requirements on high voltage zeners, etc. that I haven't typed here. I will if you're interested.

As an emergency stopgap, there is a known-quiet high voltage regulator scheme that could be used, but it's more complex, using LM317 regulators with high voltage buffers floating on the output to keep the 30V-max ICs from dying. This works, but it has more special cases to worry about.

[mhuss]

I added two foil caps, 1uF across the lower zeners and 2.2uF across the upper. I got an irregular sawtooth wave ~60hz at both outputs (see attachment, 5ms, 20mv x10). Looks like PS ripple, but I would expect the Zeners to keep the voltage across the caps fairly constant? A good old RC preamp filter has less ripple. The ripple is of course appearing at the amp outputs as well.

I removed the lower cap, and got hash riding on the sawtooth.

I reattached the lower and removed the upper, and just got hash (no sawtooth).

[loudthud]

To clear up any confusion;

Vertical type MOSFETs like the IRF 'HEXFET' types have the Drain connected to the case or heatsink tab. I'm not aware of any exceptions.

Some but not all Lateral Type MOSFETs like the famous Hitachi audio types have the Source connected to the case or heatsink tab. You must check the data sheet.

[R.G.]

Oscillating?? Yipes!
I'll go do some thinking and model running. This has to be some kind of interaction between the current source and the zeners.

Next shoot-from-the-hip possible cure: leave off the caps and make the gate-to-source resistors smaller. This runs the actual zeners at higher currents and may cut the noise.

The oscillation looks like the CCS is ramping up the caps at a constant rate, but its' not clear to me yet why they discharge for the sudden drop part of the sawtooth.

I'll get to work on the floating regulator version if you like. Or you could use the 5W zener verion on that board with some patching and jumpering. I'm composing my "sorry I mucked up the design message" just in case.

@Loudthud: Yep, good to remember. These are the drain-on-the-tab types.

[mhuss]

How much smaller? Need to keep the Zener current << (rated power/voltage), but I'm not sure how to calculate this with the Zeners already dropping approximately the amount of voltage between the MOSFET source and drain.

[R.G.]

How much smaller? Need to keep the Zener current << (rated power/voltage), but I'm not sure how to calculate this with the Zeners already dropping approximately the amount of voltage between the MOSFET source and drain.

You're hitting some of the asterisks and footnotes to the design I mentioned. Let's dig in.
The MOSFETs I suggested are FPQ2N80's, and they specify a Vgs threshold of conduction of 3 - 5V. Any given device may be anywhere in that range. When a device starts conducting, it increases current at about 1-2 amperes per volt of enhancement above threshold, and we're talking milliamperes, so for all intents and purposes, you can just pick a "typical" threshold and assume that is what you'll get. I picked 4V. It might be +/- 1V on the final result.

The MOSFET doesn't conduct until the zeners from drain to gate conduct enough current to raise the voltage across the gate-to-source resistor by the threshold voltage. T2+68ners will be just starting to conduct, and the Vgs will be the threshold, so the composite MOSFET-zener will have a zener voltage of the zener(s) from drain to gate plus one Vgs threshold. For the 130V zener (one 62V and one 68V) the actual MOSFET voltage drain to source will be 62V+68V+4V = 134V +/-1V for the fog surrounding the real Vthreshold.

The compound zener circuit is the zeners plus the gate-source resistor, paralleled by the MOSFET. The zeners start conducting at 62+68V (in this example) and have to let through a current of I = Vthreshold/Rgs before the MOSFET starts conducting. When the MOSFET conducts, it eats any current necessary to keep its gate-source voltage in range, so when the MOSFET kicks in, the zener currents stop increasing. So the gate-source resistor sets the reference zener currents. You pick a resistor that makes the actual zeners happy, and the MOSFET does the hard work.

I think I picked Rgs too high, trying to keep the zeners from dissipating much power. IIRC correctly, I picked 1/2W devices, so the 68V zener can conduct 0.5/68V = 0.007A, 7ma or less to keep it from over-dissipating. So the lowest resistor you could use there might be for a 5V threshold MOSFET, and that's 5V/0.007 = 714 ohms. I'm thinking I overdid it and use a massively higher value, putting the zeners in a high-noise region.

Notice that the zeners let current through before the MOSFET conducts. When the MOSFET conducts, the zeners run at a constant current and voltage, the MOSFET soaking up everything more than that.

For a zener to operate well, you have to know the max zener current before it burns up, and the minimum zener current to keep it conducting, and in some cases, more than the minimum to reduce noise, as here. To regulate, the zener must always have some current flowing through it. If the load eats all the available current, the zener's current goes to zero and voltage drops. If the load current out of the zener drops to zero, the zener must be able to eat all the available current without burning up.

In this circuit, the Rgs resistor sets the cut-in current for the MOSFET. If you want to zoot up the current in the zeners to lower noise, decrease Rgs. Don't go less than 714 ohms, or it will cook the 68V zeners. But you could go down to as little as 2K, for instance, without killing the zeners. They will get hot, though. Maybe drop in some 10Ks and see what happens, and decrease in steps, watching to see if noise goes down as current goes up.

[mhuss]

Actually, the 62 & 68s are 1w3 Zeners, so the microamps they're getting is likely in prime 'random noise generator' territory. I'll try increasing the current.

[R.G.]

OK then. Max current in a 1.3W 68V zener is going to be 1.3/68 = 19ma. Absolute minimum Rgs is then 5V/19ma = 261 ohms. It's a conservative idea to not go above half the power dissipation, so that would make the minimum resistance twice that, about 520 ohms. You could comfortably go as low as 1K.

[mhuss]

I put 4k7 in parallel with the 220ks, and that seems to have solved the noise problem. (whew). The MOSFETs are running around 50C with the chassis open, so hopefully they will be OK. The Zeners are barely warm to the touch. Burning in the system now...

Thanks for all the help, I really appreciate it.

[R.G.]

Waaa-hoooo! Great! I was beginning to think I'd led you down the garden path or up the creek.

The MOSFETs are probably OK at 50C. Those suckers are OK with junction temps up to about 175C.

You're quite welcome. I enjoyed ginning it up.

[mhuss]

The owner is happy with the repair, he says it sounds great and he has had no problems. Thanks again for loaning your expertise!

[R.G.]

You're welcome. It was a fun design exercise.