What insures voltage balance on stacked pairs of PS Filter Caps?

[nevetslab]

We all are used to seeing a pair of high voltage power supply caps wired in series with ballast resistors across each, usually on the first filter stage of the amplifier. It's normally done to achieve sufficiently high votage rating for cap values chosen, where a single cap doesn't have high enough voltage rating. I normally have a look at those during service work, to see that half the supply voltage is across each cap in series, as it should be. Usually they are.

In the present instance, where I'm servicing a vintage Hewlett-Packard 712C High Voltage Power supply, perfect for tube amp work (0-500VDC @ 0-200mA, 0 to -150VDC Bias, 6.3VAC CT @ 15A heater supply), with date codes from 1971 (unit came from Sylvania). I'm inside servicing the Constant Current Comparator circuit, and recapped most of the aged electrolytics. The Output caps following the series voltage regulator used a pair of 80uF/300V caps, twist-tab style, and I fitted in a pair of 100uF/400V parts, with a minor lead spacing adjustment & O-Rings glued in for mechanical stability.

After finding a small current surge when I'd switch on the HV supply, the output would overshoot (to 500V from a 300V setting!), then settle down to desired setting after a few seconds, completely uncharacteristic of lab supplies! After checking a few things, I measured across the 100k ballast resistors of the two 100uF caps and found a huge imbalance (225VDC & 75VDC for a 300VDC setting), under no load on the output. Pulled the caps, checked them on one of the two bridges I have here, reading 92uF & 90uF, same DF reading. Close enough match, but whose capacitive reactance difference shouldn't be enough to cause that much imbalance. Brand new parts.

I looked at the original parts I had removed. One cap was almost half the value of the other @ 120Hz, and at 1kHz, 47nF instead of around 80uF. So, no sense in putting those caps back into place to see that order is restored. I won't go into my pondering on why, after installing a pair of 47uF caps temporarily has now yielded no output and spun me off into assessing the SCR control circuitry on the front end, and the series regulator and other control loops feeding in the system. My new headache.

I'm wondering what insures proper voltage balance. Now, I did state NO LOAD on the output of the supply, while observing this gross imbalance. Is that the key to having proper balance on the front end stacked buss caps? In that instance, you DO have charge current flowing thru them. Not so in my case where they're merely smoothing filters (no load), and would have a little bit under DC load. Makes me go hmmmmmm..........

[52 Bill]

After checking a few things, I measured across the 100k ballast resistors of the two 100uF caps and found a huge imbalance (225VDC & 75VDC for a 300VDC setting), under no load on the output.

Check the value of the two 100K ballast resistors. If they are close, the voltage should divide evenly across the two filter caps.

[nevetslab]

Ballast resistors are both equal, 100k. I agree with you that they should divide evenly.....were this at a first filter stage. At some point, I'll wire those same caps up as in input filter to see that it DOES behave normal (balancing). Checked the two caps on my GR 1617 Cap Bridge with 200VDC bias, look fine, very low leakage. Cap & DF agree with the GenRad DigiBridge. Nothing like swapping them out with another new pair, closely matched, and get hit upside the head with NO OUTPUT. LOL! Restoration will come with the price of good education on this one.

It's really more of an emperical question....what would cause such an imbalance? I normally only find that condition when one of the parts, from years of use or preamature failure, has partially failed, with accompanying high ripple. Here, I'm downstream from charging current issues.

[tedmich]

FYI there are specific formulae for calculating the values for series cap equalizing resistors, although most amp designers use 200k-300k automatically. I can't remember the math but the calculated R values came out much lower than 220k if I remember correctly.
heres a method:
http://www.illinoiscapacitor.com/pdf..._resistors.pdf

[Pedro Vecino]

I d´ont know the reason but is a effect that I have seen many times. I use in some classic amps old German high quality electrolytics (NOS) and I appreciated that sound is different than using modern ones. Sometimes I found decompensation in large voltages and what I do is match them. I also have seen the following effect: they need time to work and after several ignitions the tendency is a slowly approaching once under load.

[R.G.]

Real capacitors can be modeled very accurately by a "perfect" cap surrounded by resistors and inductors to dirty it up.

For DC conditions and low frequency AC, you can model them pretty accurately with a perfect cap and two resistors - one resistor in series to model the internal resistances of the cap and one in parallel to model the leakage.

When you have two caps in series and apply a voltage, the voltage initially divides by the inverse of the ratio of the actual capacitances - that is, the bigger capacitance can eat more charge per volt than the smaller one, and the smaller cap gets proportionately more voltage. The current into the combination is the time constant of the series combined capacitance and the ESR's in series, so the initial current is V/(esr1+esr2).

Of course, the caps charge up with the time constant (esr1+esr2)* (C1*C2)/(C1+C2). After about five time constants, all the charge-up is over, as the charging current drops exponentially.

But the current into the leakage resistance simply follows Ohm's law, and the current in the leakages rises linearly with the voltage across each cap. When the ESR-limited charging current drops below the current for the DC leakage, the voltage distribution changes to being set by the leakage resistances, and the actual capacitances no longer matter. The DC voltage across each cap divides as set by the leakage resistance.

Bleeder caps across series capacitors make for grossly large "leakages" and make the actual internal DC leakage not matter. The voltage on series caps paralleled by bleeder resistors becomes the voltage set by the resistor ratio as long as the bleeder resistors are low enough to make the DC leakage resistance inside the caps insignificant. This can be taken by rule of thumb to be 1/10 the maximum specified internal leakage resistance.

So if you do your bleeder string right, the capacitance doesn't matter, the bleeders set the DC. The bleeder value simply has to be less than 1/10 the equivalent internal leakage resistance at end of life. This is usually quite large.

There is another overwhelmingly more important thing to be done with bleeder strings. Bleeders guarantee that the power filter caps will go to 0V in some amount of time, and so they help with safety considerations in servicing the amp. All high voltage caps need bleeders that pull them down to safe levels after some time passes. We were forced to use "below 42V peak in 30 seconds" as a safety measure. The value of resistor that will do this will also automatically satisfy voltage distribution issues for stacked caps.

[Chuck H]

I can't speak to this issue specifically, but...

How old are the "new" caps you installed??? Just because they've never been installed in a circuit doesn't guarantee one of them isn't all crunchy and dry inside. And in this state they may test good but fail to perform in real, higher stress working conditions. It sounds by your description like one cap is taking longer to charge and stabilize. I've had trouble with caps from my drawers that were "new" but of uncertain age. In the days of real electronic stores most US cap manufacturers had on shelf policies similar to produce. That is, date codes were used, stock was to be rotated (new inventory loaded to the back so the older stock moves first), and anything past an expiration date was to be removed and not sold. I no longer stock electrolytic caps. I buy them fresh for each project and I never have a problem. Obviously a repair shop in a remote locale couldn't get away with this because lead times on the repairs would suffer too much. But in that case the inventory should be managed. This doesn't even consider that you may have a cap that's bad for reasons other than age too. But my bet is on a bad cap.

[soundguruman]

"I'm wondering what insures proper voltage balance."

A constant load, which is enough to stabilize the voltage.

"If they are close, the voltage should divide evenly across the two filter caps."

It doesn't work out that way in the real world. Only in theory.

[Chuck H]

Thank you helpy helperton. What do you suppose caused the imbalance in this case. With his balanced cap values and balanced resistors?

[g1]

Geez Chuck, maybe you're being unfair, are you actually expecting him to read the thread before commenting?

[Enzo]

OMG, the original "dim bulb tester."

[Jazz P Bass]

Thank you helpy helperton. What do you suppose caused the imbalance in this case. With his balanced cap values and balanced resistors?

Helpy Helperton.
Thanks for the chuckle, Chuck.

[gbono]

Here is a site which might help with your HP power supply issues - Yahoo Groups

[nevetslab]

Found the problem

Some good information presented here in the threads. I just got back to the computer this morning to find the replies to my last post. I looked at the tech paper provided by tedmich from Illinois Capacitors, and realized I also had the Tech Notes for Electrolytic Caps from Nichicon in my database that I didn’t consult.

The caps in question are new stock Nichicon VR series 100uF/400V Radial Lead, purchased from Mouser. No date codes, so probably not much more than 1 yr shelf life, if that, though I don’t know what their inventory practice is on parts having shelf life issues.

On their data sheet, they spec their leakage current for CV values > 1000 (they don’t factor in the 10^-6 exponent here), I = 0.04CV + 100uA for > 1 min application., which works out to 1.7mA. Both caps with 400V applied measure 1mA leakage current, and around 500uA with 250V applied. In the Illinois Capacitor tech note the Ballast resistor is calculated as R =10/C, where C is in uF & R is in Meg. R = 10/100 = 100k. That’s the value HP selected for their circuit.

Also in the Illinois tech note is a calculation of the caps’ DC resistance of the leakage current. In the HP supply application, the series caps are not run higher than 500V total across the pair, or 250V across each, ASSUMING balanced voltage. That DCR is 250V/500uA or 500k. Illinois' more accurate method of selecting the ballast resistor value would be 1/10 of this value, or about 51k, with a power rating of 2W. HP uses 100k 2W in the circuit, so all in the ballpark.

I haven’t swept the impedance curve on these, but where the impedance bottoms out in the frequency vs impedance curve, then becoming inductive, that point normally coincides with the ESR value. So, at the moment, I don’t have an ESR or min impedance value on these.

Now, sitting with egg on my face, I FOUND THE PROBLEM. As all too often is the case, it was something stupid. The original Sprague metal can caps have the construction similar to twist-tab units, three support terminals total (all three teriminals are NEG, part of the CAN). HP’s PCB layout for the upper cap only had connections to the (+) & (-) where the (+) terminal is between two of three support terminals, no connections to these outside support terminals. On the lower cap, (+) in series with upper's (-), one of the support terminals NOT opposite the (+) terminal IS THE (-) OUTPUT TERMINAL TO THE POWER SUPPLY (V-)!! With the replacement caps I installed, I missed that connection, didn’t follow it out, and ASSUMED it got to the supply NEG terminal. IT DIDN’T!!. The ballast resistors are across the two caps, though. Jumpering that lower (-) connection to P/S V-, the supply is running correctly, with voltage balance within a couple volts of each other at 500VDC output. No overshoot when the HV is turned on, as should be, and works as it should. Sigh……..don’t it figure.

ASSUMPTION IS THE MOTHER OF ALL F**KUPS!

[nevetslab]

Conductive O-Rings

Conductive O-Rings! I thought I was thru. I had posted prior to removing the pair of 47uF/450V caps, and then re-installed the Nichicon VR series 100uF/400V caps, where I had O-rings glued to the bottom to allow re-alignment of the leads to fit the PCB pattern. I never thought to check that the black rubber O-rings were even partially conductive. TURNS OUT THEY WERE! So, besides having missed a crucial connection, I also introduced an additional leakage path with the O-0-Rings. After cutting a couple pieces of Fralock -.015" fishpaper insulators & fitting them between the plated copper traces of the top side of the PCB and the O-rings on the bottom of these caps, NOW I have it working correctly with the pair of 100uF/400V caps.