i'm on a roll... the kgsshv

[Nebby]

On a purely semantic level, how can you call a design that uses only one type of device complementary?

I like your idea, luvdunhill; it makes more sense I think.

[K3cT]

Sorry to interrupt guys but are we going to open a separate group buy thread for the boards and kits? I'm definitely interested for one of the board and one of the cannot-get-in-Mouser kit.

If I have experience building the β22, this should be doable right? Other than extreme voltages we are going to work with of course.

[kevin gilmore]

To manage things the group buy of boards will be a seperate thread.

Probably do board runs of 20 sets at a time.

You can do a bit of handwaving and say that a bridge circuit built from strictly one kind of

part is complementary because when measured across the bridge, the rise and fall times

are identical.

[K3cT]

All right, that settles it then. Thanks for organizing this Kevin.

[kevin gilmore]

This is the power supply i'm planning on going with

http://gilmore.chem.northwestern.edu/kgsshvps3.jpg

it uses the active battery from the T2 as the regulator.

And is adjustable for exact voltages.

It is the most efficient (least amount of excess heat) of the 3 supplies. (2 watts vs 5 watts vs 9 watts)

And it spreads the pass transistor heat on 2 heatsinks.

It is almost as good for regulation over temperature as the T2 supply. (.2% vs .05%)

and very low in noise. (about 5mv peak to peak for 1% AC input ripple)

It should come about the same money.

Schematic later.

Edited September 18, 2010 by kevin gilmore

[luvdunhill]

Very nice!

[MASantos]

Kevin, could you elaborate on this active battery regulator? I've started following the T2 thread recently and haven't read about this.

[Craig Sawyers]

Kevin, could you elaborate on this active battery regulator? I've started following the T2 thread recently and haven't read about this.

Think of it as a completely floating (ie not necessarily ground referenced) zenerless ideal voltage source. So very low noise, because high voltage zeners (in fact anything >6.2V) operates in avalanche mode, not pure zener, and is an excellent wide-band noise generator. Basic idea shown in the T2 schematic, which can be found with a bit of searching on the T2 thread.

6.2V mentioned above actually has a relevance to zener calibration standard voltage sources. Because this voltage is the transition between zener action and avalanche, and they have opposite temp coeffs. So 6.2-ish volts is exceptionally temperature insensitive over a narrow range. So such sources have zeners which are designed with the insensitivity at room temperature (20C), or at an elevated oven temperature (30C or so). All of which is completely irrelevant to any high voltage design useful for electrostatic phones.

[kevin gilmore]

like this

http://gilmore.chem.northwestern.edu/kgsshvpower3.pdf

The 4 pass fets are all isolated case... Same thing with the 2sc4686a

Sure would be nice to come up with an isolated version of something

for the bias supply. It would make assembly idiot proof.

You can of course go nuts on the 10 volt reference chip, some of the linear technology

chips are 10 times better, and 20 times the cost. Problem is distributor availability on those.

The issues with the zeners is absolutely real. This is why the KGBH supply, and the cavalli supplies

drift a few percent over temperature. The KGBH supply is still absolutely the best for AC noise

rejection, but i want something that is better than a 1% line/load supply.

[Kerry]

I love this idea

I modeled this in LTSpice and was playing around with the voltages, etc (wanted to see if it would work for the BH as well). I noticed when I reduced the voltage gap between the regulated and unregulated that I need to drop the values of R1 and R9 (I'm guessing to increase the current on the collector of Q3 feeding the pass fets) to prevent an oscillation. Did you see the same thing and do you think this is a good idea?

Also, did you consider replacing the 2SC3381 with something less expensive (not that it's terrible)? I see on the silk that you were allowing for two individual devices.

[Craig Sawyers]

Also, did you consider replacing the 2SC3381 with something less expensive (not that it's terrible)? I see on the silk that you were allowing for two individual devices.

The 2SC3381's are used as current mirrors. For these to work properly there needs to be good matching and thermal tracking between Vbe drops. The best chance of that is to use a dual device with both transistors on a single piece of silicon.

[kevin gilmore]

Yep, works for the BH, but you are wasting a bit of power because the unregulated rail needs to be 600 volts to

make the bias source work. Now if you want to get really silly you can replace R1/R28 with a 1ma CRD

(in parallel with a 100v zener to protect it on turn on) and get the ripple down into the 50 microvolt range.

I have not tried this yet, but i have ordered the diodes.

[Kerry]

Cool

This is what I was thinking for the power supply and bias. I used the voltage doubler from the original BH PS and just bumped some of the cap values to accomodate the load required by the zener string. With this I can use the same transformer from the original BH design. That should keep the heat from the pass fet around 7.5W +/- 3W or so.

Here's the schematic (just the positive rail and bias):

image removed at the request of Kerry

I ran it through LTSpice and got this:

image removed at the request of Kerry

Any thoughts?

I'm not above being silly; but what's a CRD?

Edited November 24, 2010 by deepak
image removal at request of poster

[luvdunhill]

I'm not above being silly; but what's a CRD?

current regulative diode. Basically a JFET with a trimmed source resistor for the lazy

[justin]

current regulative diode. Basically a JFET with a trimmed source resistor for the lazy

but tends to be closer tolerance than just taking a bunch of JFETs and using the same resistor

[nattonrice]

current regulator diode = crd

... and I'm too slow hehe

[kevin gilmore]

That is not oscillation, that is the charge/discharge of the main caps.

What you have to do is remove the other traces and expand the green.

Also look at the drive voltage on the base of the reg transistor.

Sure looks like it works right to me.

Fact is this power supply design is at the -100db level for noise/ripple.

Not easy to do for high voltages. John curl has enough of a hard

time doing that for +/-30v.

Also that version of the voltage doubler might be more than 900 volts

which is a bit much for the ixys part.

This is exactly what i want, a discussion of design by informed people.

Doing the circuit boards is the easy part.

2.6K load resistor?? Thats quite high in power...

Edited September 20, 2010 by kevin gilmore

[Pars]

Yep, works for the BH, but you are wasting a bit of power because the unregulated rail needs to be 600 volts to

make the bias source work. Now if you want to get really silly you can replace R1/R28 with a 1ma CRD

(in parallel with a 100v zener to protect it on turn on) and get the ripple down into the 50 microvolt range.

I have not tried this yet, but i have ordered the diodes.

Sounds like a Mike Elliot (Counterpoint) trick. Watch the CRD voltage limits... the 1N stuff is usually 100V, but the J5xx series is only 50V. Elliot uses a 91V zener for protection also.

[Kerry]

That is not oscillation, that is the charge/discharge of the main caps.

Also that version of the voltage doubler might be more than 900 volts

which is a bit much for the ixys part.

2.6K load resistor?? Thats quite high in power...

The posted files did not have the oscillation issue, so your right, everything worked correctly.

I left in some of the adjustment for the Bias voltage (pre regulation) with R25, R21. You can keep this voltage in a safe range (not for humans) at around 700V. These values could be adjusted for different PS uses (KGHB, KHSSHV, etc).

The 2.6K load was based on the BH 400V / 150mA.

Watch the CRD voltage limits... the 1N stuff is usually 100V, but the J5xx series is only 50V. Elliot uses a 91V zener for protection also.

I've added a 100V zener for protection and tested in spice. It works well. I could lower it to 91 volts.

Here is the updated schematic and LTSpice with the addition of the CRD (thanks for the help on this ) and protection zener.

I like this version much better because it will work with a variety of input voltages without adjustment. The only trick is that it will only regulate over about an 85V range, which should be fine. In my case, for a 400V output, that is 415V to 500V. lower and it doesn't work (420V is more stable) and higher and the zener will kick in - I tested this in LTSpice and it works well. I think that by adding back R5 (referencing schematic per my first post) and adjusting for different voltage inputs you could increase this range.

[digger945]

Sounds like a Mike Elliot ...

I thought of Ti. Sigma22. A little bit different arrangement for ~12uV unloaded.

[Pars]

I thought of Ti. Sigma22. A little bit different arrangement for ~12uV unloaded.

I was referring to the protecting the CRD from overvoltage with a zener trick.

[kevin gilmore]

current version of board, luvdunhill wanted heatsinks on the +/-15 to power something

http://gilmore.chem.northwestern.edu/kgsshvps7a.jpg

will add the crd's once i test that it works as i believe it will.

[kevin gilmore]

more better yet.

http://gilmore.chem.northwestern.edu/kgsshvps7c.jpg

no expensive crd's, the 10m90s works just fine.

small heatsinks probably not necessary, but there is room for it.

noise and ripple about 200 microvolts peak to peak with full load.

[ujamerstand]

On the second page there was recommendations on alpha and TKD pots. Now I've tried contacting alpha about the four channel pot but got no answers. I rather not spend $250 @ thel-audio for a quad cp-2500 either... Does anyone know of a place to buy these 4 channel pots? (other than familygate's quad RK27)

[justin]

i may be buying some TKD quad CP-2500. group buy?