• These commercial threads are for private transactions. diyAudio.com provides these forums for the convenience of our members, but makes no warranty nor assumes any responsibility. We do not vet any members, use of this facility is at your own risk. Customers can post any issues in those threads as long as it is done in a civil manner. All diyAudio rules about conduct apply and will be enforced.

21st Century Maida Regulator

I have been experimenting with solid state regulators for use in tube amps for some time. I want to make one that's suitable for experimentors which imposes a new set of constraints. You need an adjustable output over a fairly wide range, and it needs to be stable for all loads including a short.

That's a really tall order. I'd take a look at how it's done in the HP6209B. The first stage of regulation is done in the rectifier (SCR-based). That's then followed up with a linear regulator. This minimizes the amount of power dissipated in the linear regulator. Clever design... I have one. It's a rock solid supply. It does put out a bit of 120 Hz crap resulting from the synchronous rectification, but otherwise, it's a really nice supply.

The one that I could never make right with the Maida design is the ability to handle an accidental (or in the Tubelab sense deliberate) short circuit on the output when the regulator is powered up and operational. Have you tried this. If not, and you do try it, shield yourself from mosfet schrapnel.

No kidding!! I have not tried this yet. I figured I'd do all the non-destructive testing first... :) But I haven't been able to build a regulator that would survive a short circuit either -- certainly not an indefinite short circuit. My experience with my previous Maida regulator designs have been similar to yours. My expectations of this "21st Century Maida" are no different. I'm hoping that it will at least survive a momentary short circuit (hence the addition of the Tranzorb across the regulator), but I'll be happy if it survives start-up into my 47 uF, 630 V Solen polypropylene caps (5 mOhm ESR!).

~Tom
 
Last edited:
My guess is that below -80dB ripple rejection is going to become less relevant than Zout.

I guess I should complete my sentence... My goal is to have the best ripple rejection I can get while maintaining a low output impedance. How's that? :)

You're right. Zout does need to be low across the audio band.

I would suggest looking at some of the fast transient response regulators from LLTC and not overly damping the device for low noise. Some of the LLTC regs allow for remote load sensing and this is a MAJOR IMPROVEMENT!

Remote sensing makes a huge improvement in Zout, that's for sure.

Before picking the LT3080, I did glance through a couple of other ICs. But I don't recall seeing a floating regulator with remote sensing capability. Am I mistaken?

~Tom
 
That's a really tall order.

Tell me about it. I had about half a coffee cup full of blasted mosfets from my attempts. Place the coffee cup upside down over the mosfet before testing.

I have been using an old HP6448B for my current high powered amp experiments. It is a 60 Hz switcher with current limiting....but the reaction time of the limiter is hundreds of milliseconds AND there is a large electrolytic inside the unit across the output terminals. It is rated for 600 volts and 1.5 amps, and turned up all the way it makes 650 volts and 1.7 amps. I am sure that a tube short would set an OPT on fire before the limiter could work, and it isn't big enough any more! I have reached the end of the power supply extracting 525 watts from Pete Millett's 18 WPC big red board.

I am working on an even bigger amp so I am currently using an unregulated setup with some big transformers, a big variac, diodes and caps, big caps. Like 2 600uF 500 V polyprops in series, just a few milliohms ESR. It makes 750 volts and trips the bench breaker at just over 2 amps. Now, I want to make a mosfet regulator with a fail safe current limiter. Fuses explode like mosfets at this power level, so the coffee mug is back.

I experimented with the mosfet Maida and it kinda works but doesn't like a big cap across the output, but that is not wanted for safety reasons. I have been experimenting with other circuits, but I think the key is the CCS property of pentodes and mosfets. If you fix the grid or gate voltage you get a CCS. For the Fuji mosfets I am using (for now, I have 18 left out of 50) a 5.1 volt zener from source to gate limits the drain current to 2 amps. I never got it to work right in the Maida circuit, but that was 2 years ago. You still need a method of limiting dissipation. For now I am just setting the variac to limit the voltage across the fet to 100 volts max during normal operation. There is a .05 ohm resistor in the return lead for current sensing and fault detection. I am still working on a better solution.
 
I have reached the end of the power supply extracting 525 watts from Pete Millett's 18 WPC big red board.

Nice... :D

If you fix the grid or gate voltage you get a CCS. For the Fuji mosfets I am using (for now, I have 18 left out of 50) a 5.1 volt zener from source to gate limits the drain current to 2 amps. I never got it to work right in the Maida circuit, but that was 2 years ago. You still need a method of limiting dissipation.

For my circuit at relatively low energy levels (compared to the ones you're dealing with anyway), the soft start of the regulator will actually save the cascode device when starting up into a capacitive load. In Mike Maida's original circuit, implementing softstart for higher output voltages was a royal pain. And without the softstart, you're practically guaranteed to hit the SOA limit on the cascode during start-up. However, with the LT3080, it's just a matter of adding a low voltage cap. Yay. This is how my regulator is capable of starting up with 47 uF on the output as long as there is enough resistive load to take up the current that flows through the zener diode (D2).

Currently, I'm only running 1 mA in the resistive divider forming the feedback network. That's a bit on the light side. I need at least 3~500 uA plus the worst case current through the zener diode, D2. This way the regulator should be able to start up into a purely capacitive load. But I still need to get some more resistors and verify this in the lab...

I may try your trick of limiting the Vgs of the cascode. But given that the drain current is some exponential function of Vgs, I doubt it can be used as a reliable protection device. It may work as a one-off deal, but I'm afraid repeatability will be an issue due to part-to-part variation. Then again, it might just work well enough...

~Tom
 
It may work as a one-off deal, but I'm afraid repeatability will be an issue due to part-to-part variation.

I have seen anywhere from 1.5 to 3 amps on different fets from the same batch. The worse part is the temp variation. So far it works to save a fet rated at 7 amps continuous, 25 amps peak. Without the diode the fet just explodes. It also lets the fuse blow, not explode.

It's all a temporary solution for now. I am working on a big amp breadboard for learning purposes. Then I'll get back to the power supply. I am thinking buck converter....with a zener across the gate of the switch fet. Linear after that. Otherwize I'll need a really big heat sink, which I do have.
 
SMPS on breadboard are recipes for disaster.

SMPS on breadboard at 750 volts are called fireballs.

I am making a "breadboard" amplifier. It is a two layer monster that uses several PC boards and two slabs of plywood. The power supply is currently all unregulated supplies with transformers, diodes, caps, and variacs. It takes up the entire lower level and I can barely lift it. I needed something simple and effective so I could get started with the fun stuff.

The variable regulated supply is an entirely different project and will be all PC board construction. The HV supply may use a buck converter for the first stage just to lower the dissipation. The second stage will be a linear regulator.
 
LT3070 has a "physical" sense input, LT1963 has a "meta-physical" sense input if you look into the datasheet.

Do you actually have this working in a high-voltage setup?

The LT1963 requires at least 1 mA (1.5 mA worst case) to regulate. Add the zener current for the cascode bias, some margin, and you'll have 2.5~3.5 mA flowing in the feedback resistors. Still better than the original Maida with its 10 mA, but getting kinda high...

But after our conversation about output impedance the other day, I got to thinking... How low output impedance is actually needed? Given that the DC resistance of a typical OPT is around 100 ohm, having Zout of 1 ohm or even 10 ohm is probably OK.

I'm currently working to build an output impedance test rig. I should have measurements by the weekend, assuming the rig works.

~Tom
 
Last edited:
I haven't run this with an LT1963A in HV, but I'm willing to give it a try. I just have to finish up with another project. I've been itching to do HV phase-gain plots anyway.

With a hat-tip to Mr. Yaniger -- I think that the Impasse and Master's Noise preamps sound so great is that he uses a depletion MOSFET pair as the current source, almost immediately adjacent to the business end, and after the Maida regulator which is on an umbelical anyway. walt Jung described the use of these in AX and the articles are archived on his website.
 
Arent we drifting a little off-topic?
I liked the schematic Tom proposed, and all his design parameters seem still valid to me.
Low current to minimise dissipation,excellent ripple reduction, low enough output impedance and so on...
Short circuit proof isnt very important to me, a normal c l c supply isnt short circuit proof either.
Looks to me like all he needs is a way to ensure stability and proper startup on a somewhat larger buffer cap. I didnt like the 30k parallel resistor to ensure startup, its eating 5 watts all the time on 400v. isnt there anything you can do with a series resistor, or another path to load that capacitor?
 
Arent we drifting a little off-topic?

Yeah....... Back to the regular scheduled program. :) Every once in a while, the sidebar conversations are quite fruitful, though. But it's a balance.

I liked the schematic Tom proposed, and all his design parameters seem still valid to me.

Thank you. I like it as well. If I can make it start up into a capacitive load (I think that's possible) and the output impedance is reasonable, then I think we have a winner.

Looks to me like all he needs is a way to ensure stability and proper startup on a somewhat larger buffer cap.

Agreed.

isnt there anything you can do with a series resistor, or another path to load that capacitor?

That's what I need to find out. My simulation says it should be possible to start up into a 47 uF capacitor with 5 mOhm ESR. I have parts arriving in the mail on Friday so I can test it out in the lab over the weekend. I should also have my output impedance test rig ready by then. Hopefully, it'll work (sim says it should).

I think it'll definitely be possible to start up with an RC on the output of the regulator. I.e. cap with series resistor. But that defeats part of the purpose of having the regulator in the first place as it increases the output impedance.

~Tom
 
Last edited:
This is what I mean by remote sense for the LT1963:

If you disconnect R4,5 from Vout, and instead connect them to the sensitive B+ node at the amplifier you'll effect the same change. When you do your boards you can just leave a connection to Vout which may be jumpered (for "local sense") or "remote sensed" by the user.

I think C5 @100nF may damp the response of the LT3080 too much.
 

Attachments

  • RemoteSense.gif
    RemoteSense.gif
    4 KB · Views: 2,382
If you disconnect R4,5 from Vout, and instead connect them to the sensitive B+ node at the amplifier you'll effect the same change. When you do your boards you can just leave a connection to Vout which may be jumpered (for "local sense") or "remote sensed" by the user.

I think you'll end up shooting yourself in the foot if you route R4, R5 all the way to the load. Here's why:

The LT3080 is a floating regulator. It develops its voltage reference by sourcing 10 uA through the SET pin. This develops 1.00 V across R7. Vout is servoed by the error amp of the regulator, hence, the voltage on the output pin is the same as that of the SET pin. This means, 1.00 V will be across the combination of R5||(R4+R6). With the values in the schematic, 1 mA will flow in the R4~R6 combo, resulting in 1.01 mA (1 mA + 10 uA) flowing in R9. This sets the output voltage of the regulator.

If you route R4, R5 to the load as you suggest, all you will end up doing is increasing the resistance of the R5||(R4+R6) combo, thus, decreasing the output voltage by some amount. And as the voltage drop across the routing will be load current dependent, you've effectively shot yourself in the foot. It will result in higher output impedance (and possibility for instability and/or noise/EMI injection).

I won't go into a discussion of LT1963 vs LT3080. I made my justifications for choosing the LT3080 in Post #1. Feel free to disagree or comment, but I think my justifications are pretty reasonable.

I've attached the schematic to this post so readers don't have to flip back to Post #1 for the parts designators.

I think C5 @100nF may damp the response of the LT3080 too much.

I arrived at the values for R8, C5 through extensive simulation in LTspice. Those values turned out to be a good compromise between stability and fast transient response. I may play with it in the lab, but right now I'm taking the "if it ain't broken, don't fix it" approach.

~Tom
 

Attachments

  • MaidaReg_1p0_Schematic.png
    MaidaReg_1p0_Schematic.png
    48.8 KB · Views: 2,578
Last edited:
You can count on me to get a couple boards when they're done!

My affinity for the LT1963 owes to its performance in a recent listening test with a bunch of folks (we used the LT3015 on the negative rail). I wouldn't sacrifice PSRR or output impedance to get a low noise figure -- just my 2 centimes. I have a bunch of the LT3080 and just haven't deployed it yet.

btw, my HP3577 does nice phase-gain plots but it is really unhappy with almost any DC on its inputs!
 
You can count on me to get a couple boards when they're done!

Sweet! Thanks.

My affinity for the LT1963 owes to its performance in a recent listening test with a bunch of folks (we used the LT3015 on the negative rail).

This in a low voltage application?

I think you'll like the LT3080 as well. At least I found my 300B amp sounding cleaner than it did with my previous regulator. I think it may have to do with less hum on the amp outputs, but I have no measurements to confirm this. I was never able to pick up any 60/120 Hz IMD products on my spec an. But it is as if there's more quiet space in the sound stage. It was really quite a pleasant surprise.

btw, my HP3577 does nice phase-gain plots but it is really unhappy with almost any DC on its inputs!

Yeah... RF gear is notorious for disliking DC. External protection diodes and DC coupling are definitely called for.

I've done extensive op-amp characterization at work using network analyzers as well. I'd love to get an HP 3577 for my home lab... But right now, my HP 3562A does the trick (at least up to 100 kHz).

~Tom
 
Good news:

I made some tweaks. The regulator now starts up into a 47 uF capacitive load with no external resistive load. That's a 47 uF polypropylene cap with about 5 mOhm of ESR. So practically start-up into a dead short. I tried this several times with 560 V DC on the regulator input. The regulator starts up to 400 V (my target output voltage) every time.

I also tried my "loose tube socket" test. I use light bulbs for the test. Four 25 W bulbs in series draw almost exactly 200 mA. By partially unscrewing one of the bulbs and jiggling it, I'm able to simulate a load with high dV/dt - just like an output tube that's a little loose in the socket. Again, the regulator survives. It actually nicely goes into soft start and smoothly regulates back to 400 V DC.

As promised, I built up an output impedance test rig, but I'm not getting results that make sense. It seems to work fine with a low voltage supply, but on high-voltage supplies - like the present regulator - I'm getting weird results. In fact I'm measuring the exact same output impedance for my previous regulator as I do for this regulator and my HP 6209B. This makes no sense at all. So I'm afraid the AC measurements of the output impedance will have to wait.

That said, it is, of course, always possible to measure the output impedance at DC by varying the load. By measuring the voltage drop as I apply my light bulb load, I arrive at the following: For 0 mA output current, I measure an output voltage of 400.00 V. At 188 mA load current, the output voltage is 399.81 V. This translates to 1.01 ohm of output impedance. This includes my hook-up wires (about 120 cm worth total length). So not bad...

Another test I tried was to let the regulator startup without load and then apply my lightbulb load. Recall, that the cold filaments of the light bulbs represent nearly a dead short circuit. In reality, they measure about 180 ohm, so they should result in an inrush current of roughly 2.2 A. Again, the regulator takes this abuse without issue.

So let's summarize:
  • 400 V, 200 mA design target met.
  • Start-up into up to at least 47 uF purely capacitive load
  • Start-up into harsh load (2.2 A inrush light bulb load)
  • Soft start (10 second start-up with resistive load, 1 second with no load)
  • Handles high dV/dt current pulses
  • Well-tamed transient response
  • Output voltage adjustment range: ~15 % of nominal output voltage

I'm liking it...

~Tom
 

Attachments

  • MaidaReg_1p0_Schematic.png
    MaidaReg_1p0_Schematic.png
    40.6 KB · Views: 2,637
Last edited:
I have two methods to evaluate power supplies.

First, a very simple method of observing the output impedance of a power supply as the load "frequency" is varried. Since square waves are used there is no single frequency load.

I place a resistive load on the supply equal to the minimum desired load. I use a big mosfet to switch in another resistor with the desired maximum load. (source on supply negative, drain to big resistor the other end of which is connected to supply positive, function generator wired between source and gate.) You can then vary the frequency and watch the supply output with a scope. This one will catch weird stuff and frequency dependent transient anomalies.

The second started out as a variation of the old "alternator whine test" we used with mobile police radios at work. The original test used a large filament transformer (to avoid core saturation) with the secondary wired in series with the radio under test. The power source was a car battery. The primary was driven with an old tube type HP audio oscillator (model 200AB). This setup could put about 1 volt P-P of sine wave ripple on the 12 volt supply lines. It was used to evaluate the PSRR of the two way radio since the car battery was assumed to be a short at AC. The big filament transformer limited the test to frequencies below about 1500Hz but that was good enough.

I decided to build a similar setup to torture power supplies and test PSRR of amplifier circuits, but I used an SE OPT driven by a audio power amp. For testing power supplies place a motor run cap across the load. for testing PSRR of a circuit, the cap goes across the power supply. After blowing up a chip amp I found that the turn on transient stuffs a spike through the OPT, back up the speaker output of the amp used to generate the test signal, so I switched to a tube amp, one of my SSE's.

After a bit of head scratching it dawned on me that the two OPT's were redundant and weren't needed at all. My latest power supply / circuit tester is simply an KT88 triode SE amp without the OPT. I connect the plate of the output tube through a variable resistor to the power supply being tested. One scope probe goes on either side of the resistor. Drive the amp with sine waves, triangle waves, square waves, and even music. Adjust the slider on the variable resistor until the AC voltage on the plate is twice the AC voltage on the power supply output and the resistor value will be a decent approximation of the output impedance of the supply. For testing the PSRR of a circuit, connect it directly to the plate of the output tube.