Adjustable Bench Power Supply - Redux

Finished up the front panel supply board today.
Looking good Jason. I'm waiting with baited breath for you to try your short circuit tests....... (I'll buy you a steak dinner if you can do those tests without your blood pressure rising about double...:)

There must be some way to combine the 2 styles of PS like is done on some power amps, changing the charging voltage to the main caps, before the pass regulator on the fly.
There is indeed a way... Now you are getting into an area I explored with my current bench power supply I am in the middle of building. You can do this in the analog domain but it's kinda complicated in my opinion, requiring kinda precise tuning of several R/C circuits.

Rather, I thought I might do it in the digital domain using an Arduino or even a Raspberry PI. The idea was to sample the input and output voltages, take the difference, add about 20V (so that the charge on the caps feeding the pass element was always about 20V more than the requested output voltage). Then based on that difference voltage calculated, determine the conduction angle of a rectified sine wave that would charge the PS cap to that value. Then turn on an SCR for the correct number of microseconds that would allow rectified B+ to flow to charge up the caps to the that value, then turn it off at the bottom of the rectified wave cycle. I'm pretty confident this would work quite well. I got all the way down the path, far enough along to design the software algorithm that would do the job.

But when it came down to prototyping it, I moved away from that approach because I felt a microcontroller was just another part that would eventually break, and I didn't want to have to worry about servicing the power supply in 10 years...so I back tracked and went a different approach.

I don't know a whole lot about switch mode power supplies, but I think you could also do it that way--vary the width of the duty cycle to get that cap charged to the proper value so you have minimum voltage drop across your pass device.
 
switchers aren't inherently evil as long as they are filtered properly. Thats really where the bad reputation comes from. The DC supply I use for mobile electronics work is a 40 amp switcher and its perfectly quiet. I've run all sorts of radio equipment off of it, no problems with RF leakage or other nonsense problems.
 
Layout for the screen supply board (will use jumper wires for the top leads of those axial film caps). Trying to keep it fairly compact while giving the power resistor room to breathe.
 

Attachments

  • IMG_5003.JPG
    IMG_5003.JPG
    44.6 KB · Views: 21
Screen supply board is working pretty well -- I left it hooked up to a 2K load to burn in while I made / served dinner. The MOSFET at about 53C on the heat sink (with a mica insulator and thermal grease). But my output voltage is reading higher than simulated. The cathode of my 120V Zener is reading 128V (which is within tolerance, but I suppose I should have tested all of the ones in my inventory for the closest match), and so my output voltage into this load (which ends up being 62.5mA) is 125V, 9V higher than my target of 116V.

Now maybe this won't make much of a difference. OTOH, I'd like to understand what is going on and correct if I can. I'm dropping 20V across the 2.2K resistor in series with the 120V Zener, and that's 9mA, so should be plenty of current to get the Zener to behave as expected. Is this just a tolerance issue?
 

Attachments

  • IMG_5005.JPG
    IMG_5005.JPG
    62.4 KB · Views: 12
I think so. Mine measured a little high also. I used an OnSemi 5W 120V Zener (1N5380), and it seems to deliver a higher voltage than specified at the current being used. (in my case that was about 12 mA). But it should be fine. That screen voltage will run the 6550's just a little hotter when the supply is in constant current mode, but still it's within tolerance of the tubes.

So I'd leave it be if it were me.
 
I think so. Mine measured a little high also. I used an OnSemi 5W 120V Zener (1N5380), and it seems to deliver a higher voltage than specified at the current being used. (in my case that was about 12 mA). But it should be fine. That screen voltage will run the 6550's just a little hotter when the supply is in constant current mode, but still it's within tolerance of the tubes.

So I'd leave it be if it were me.

Ok, I'll call it good enough, then. (That's the specific 120V Zener I'm using as well.)
 
Ok, over the past couple of evenings, built up the C- supply board. Definitely the most complicated of the sub-assemblies. Simulation told me that, worst case, the first MJE5730 dissipates 500mW, so I put it on a smaller heat sink than the second one (which dissipates 2.2W worst case).

IMG_5028.JPG

Testing the board went without a hitch, and the fine adjustment afforded by the 10-turn pot is nice.

IMG_5023.JPG

So, now I have the panel supply board, the 6550 screen supply board, and the bias supply board. Next step -- the heart of the beast -- the LM317 sub-assembly.

IMG_5026.JPG IMG_5027.JPG IMG_5025.JPG
 
Looking good. My C- Supply is adjustable (no load conditions) from -108V to 0V, and (from memory) from about -98V to 0V under a 30 mA load. You should now be able to test short circuit conditions on your C- supply. Better to know now if it's going to go up in smoke before you get it all installed.

I'm much more careful with 500V on the HV regulator, but on the C- regulator, I have accidentally shorted those leads a few times. Very glad for the current limiting built in there...
 
You should now be able to test short circuit conditions on your C- supply. Better to know now if it's going to go up in smoke before you get it all installed.

Ah, good point, I forgot to do a short-circuit test. I should be able to get to that this afternoon.
 
Ah, good point, I forgot to do a short-circuit test. I should be able to get to that this afternoon.

Finally got around to doing the current limit test. Test as follows: Output of C- board connected to a 1K 10W resistor in series with my Fluke set to read DC current. Keithley connected across the output terminals to measure DC volts. Pot was set to minimum, AC power applied, and pot slowly turned towards maximum while observing the voltage and current readings.

As expected, current tracked voltage precisely -- -6V -> 6mA. As I started approaching -24V, the pot became slightly less responsive... more turns required to get the voltage to move. With pot at max, I get -28V and 28mA, as opposed to the unloaded -120V or so. After removing AC power and verifying the voltage had dropped to 0V, I felt the power transistors and they were warm to touch, but not hot.

So, it seems the current limiting function is working correctly!
 

Attachments

  • IMG_5038.JPG
    IMG_5038.JPG
    91.3 KB · Views: 4
Seems to be working. Correct, it's not brick wall current limiting on the C- design. It's a little gradual as you noted. (You could make a different design that does do brick wall current limiting but I didn't see the need.) At this point, you should be able to short the leads directly together from a starting point of any dialed in C- voltage, and not cause anything to blow up, and you should be able to do that repeatedly.
 
It occurred to me last night that one of the failure modes of this power supply design could most likely be eliminated by using a mechanical limit instead of an electrical limit.

Let me 'splain.

One of the failure modes I was concerned about in post #30 was the failure of the voltage adjust pot due to too much current through it. The pot I spec'd was a single turn 250K unit, with 2W dissipation (i.e., the biggest unit I could source for reasonable cost). The problem I saw was that if you have a large cap only load, say 820 uF, and your starting condition is a fully charged cap (480V), and then if you very quickly turn the voltage adjust pot down to zero (if you are quick on the draw, you can make that rotation in what...1/2 a second). Because the load cap has no other discharge path, it will discharge through the adjust circuit (see post #30 for detailed description of the problem).

This is only an issue with a very sudden or abrupt change of selected output voltage. Since the supply can react extremely quickly--much quicker than the cap can discharge, the conditions are setup so that the full voltage change between voltage on the cap and the new selected output voltage must discharge through the supply's adjust circuit. This can be catastrophic for the pot and will overheat it and blow it up under the right conditions (I KNOW because I blew up several before I was keen to this problem). So to solve it I put in place "current backflow preventer"--a series diode that would not allow the cap to reverse discharge through the supply's adjust circut.

The side effect of this "fix" is that supply regulation is not nearly as sharp as it could be without the diode. The diode always drops about 1 volt across it, so the best regulation you will see on a fluctuating load is about 1V. Terrible considering what the supply is capable of.

As a (possibly) better solution, what if you were limited in how fast you could turn down the voltage adjust pot? Let's say you could physically only turn it 25K at a time. For a 250K adjust pot, a 25K difference in adjust pot value would adjust the output by 1/10th of 480V, or 48V (assuming linear relationship). In other words, the most voltage difference the supply could ever see under this condition would be a 48V change at any one time. With a large cap only load and the backflow prevention diode removed, instantaneous current discharge back through the adjust circuit would then be 48V over whatever the equivalent adjust circuit resistance was at the time.

In this scenario we need to consider both the steady state DC conditions and the dV/dt (fluctuating) condition. Let's take two boundary condition use cases:

Use case 1:
  • Adjust pot is set at 250K, and supply is currently delivering 480V.
  • You quickly move the adjust pot 25K downwards (so its new value is now 225K), which adjusts the output voltage to be 480V - 48V = 432V.
  • Instantaneous current change through the adjust circuit (cap discharge through the adjust circuit) would be no more than 48V/225K = 0.213 mA. Wattage dissipation of the pot would be 10 mW.
  • Steady state current through through the adjust circuit would be 435V/225K = 1.93 mA. Wattage dissipation of the pot would be 0.84 watts
With a 2 watt pot, the pot is easily able to handle these conditions.

Use case 2:
  • Adjust pot is set to 50K, and supply is currently delivering 96V output
  • You quickly move the adjust pot 25K downwards (so its new value is now 25K), which adjusts the output voltage to be 48V.
  • Instantaneous current change through the adjust circuit (cap discharge through the adjust circuit) would be 48V/25K = 1.92 mA, Wattage dissipation of the pot would be 92.16 mW.
  • Steady state current through adjust circuit would be 48V/25K = 1.93 mA. Wattage dissipation of the pot would be 92 mW.

Again with a 2 watt pot, the pot is easily able to handle these conditions.

So there is no reason in my estimation that the back flow preventer diode is needed, IFF we can somehow limit max change in the adjust pot to 25K at a time. And if we can eliminate that series diode, output regulation improves substantially to the spec of what the 317 chip is capable of.

Is there a way to limit adjust pot change to only 25K at a time? The answer is YES, and it's so simple I can't believe I didn't see this before now. The method is to use a 10 turn precision pot! A human hand is only capable (on a good day) of making a single turn in (let's say) 1/2 of a second. You then need to reposition your hand on the knob to turn it for another turn. About the fastest you can do that is ~1 second intervals. (unless you're Neal Peart or something).

Anyway, I speculate that there is enough time for the cap only load to dissipate 48V through the adjust circuit that this condition I've been worried about should not be a problem, if using a 10 turn adjust pot. And that removes the need for the "backflow preventer" diode, and thus substantially increases regulation capability.

Furthermore, I would probably fit a round knob on the pot shaft so that you will not be tempted to be able to turn it faster than say one full turn every 1/2 second. (A chicken head knob might be able to be turned faster if you use your finger and push it around).

Just a thought anyway...
 
Is there a way to limit adjust pot change to only 25K at a time? The answer is YES, and it's so simple I can't believe I didn't see this before now. The method is to use a 10 turn precision pot!

...and perhaps at the same time eliminate the need for the fine adjustment control!
 
FINALLY got around to finishing the main regulator board (garage had become a construction staging area again for a while). As a basic smoke test of the regulator function, I shorted across the terminals for the voltage adjustment pots to bring it down to the minimum voltage output... 100K || 1.3K -> 1283, which should be 9.9V out of the LM317. Hooked it up to my Heathkit power supply with my Fluke meter on the input voltage and my Keithley on the output voltage. The LM317 snaps into regulation promptly at 11.6V, providing 9.77V on the output, which is just about what I would expect. Regulation is right on the money as I increase the input voltage right up until ~35V in...and going past that, right on cue, the Vishay SA22A TVS diode does its job and conducts around the LM317 at 36.1V, protecting it from the over-voltage situation. (FWIW, I'm using the HV version of the TI LM317, which IIRC gives me 57V across the LM317... So, this circuit is built conservatively).

Now, onto the 6550 board!
 

Attachments

  • IMG_9318.JPG
    IMG_9318.JPG
    105.9 KB · Views: 13
It occurred to me last night that one of the failure modes of this power supply design could most likely be eliminated by using a mechanical limit instead of an electrical limit.

Let me 'splain.

One of the failure modes I was concerned about in post #30 was the failure of the voltage adjust pot due to too much current through it. The pot I spec'd was a single turn 250K unit, with 2W dissipation (i.e., the biggest unit I could source for reasonable cost). The problem I saw was that if you have a large cap only load, say 820 uF, and your starting condition is a fully charged cap (480V), and then if you very quickly turn the voltage adjust pot down to zero (if you are quick on the draw, you can make that rotation in what...1/2 a second). Because the load cap has no other discharge path, it will discharge through the adjust circuit (see post #30 for detailed description of the problem).

This is only an issue with a very sudden or abrupt change of selected output voltage. Since the supply can react extremely quickly--much quicker than the cap can discharge, the conditions are setup so that the full voltage change between voltage on the cap and the new selected output voltage must discharge through the supply's adjust circuit. This can be catastrophic for the pot and will overheat it and blow it up under the right conditions (I KNOW because I blew up several before I was keen to this problem). So to solve it I put in place "current backflow preventer"--a series diode that would not allow the cap to reverse discharge through the supply's adjust circut.

The side effect of this "fix" is that supply regulation is not nearly as sharp as it could be without the diode. The diode always drops about 1 volt across it, so the best regulation you will see on a fluctuating load is about 1V. Terrible considering what the supply is capable of.

As a (possibly) better solution, what if you were limited in how fast you could turn down the voltage adjust pot? Let's say you could physically only turn it 25K at a time. For a 250K adjust pot, a 25K difference in adjust pot value would adjust the output by 1/10th of 480V, or 48V (assuming linear relationship). In other words, the most voltage difference the supply could ever see under this condition would be a 48V change at any one time. With a large cap only load and the backflow prevention diode removed, instantaneous current discharge back through the adjust circuit would then be 48V over whatever the equivalent adjust circuit resistance was at the time.

In this scenario we need to consider both the steady state DC conditions and the dV/dt (fluctuating) condition. Let's take two boundary condition use cases:

Use case 1:
  • Adjust pot is set at 250K, and supply is currently delivering 480V.
  • You quickly move the adjust pot 25K downwards (so its new value is now 225K), which adjusts the output voltage to be 480V - 48V = 432V.
  • Instantaneous current change through the adjust circuit (cap discharge through the adjust circuit) would be no more than 48V/225K = 0.213 mA. Wattage dissipation of the pot would be 10 mW.
  • Steady state current through through the adjust circuit would be 435V/225K = 1.93 mA. Wattage dissipation of the pot would be 0.84 watts
With a 2 watt pot, the pot is easily able to handle these conditions.

Use case 2:
  • Adjust pot is set to 50K, and supply is currently delivering 96V output
  • You quickly move the adjust pot 25K downwards (so its new value is now 25K), which adjusts the output voltage to be 48V.
  • Instantaneous current change through the adjust circuit (cap discharge through the adjust circuit) would be 48V/25K = 1.92 mA, Wattage dissipation of the pot would be 92.16 mW.
  • Steady state current through adjust circuit would be 48V/25K = 1.93 mA. Wattage dissipation of the pot would be 92 mW.

Again with a 2 watt pot, the pot is easily able to handle these conditions.

So there is no reason in my estimation that the back flow preventer diode is needed, IFF we can somehow limit max change in the adjust pot to 25K at a time. And if we can eliminate that series diode, output regulation improves substantially to the spec of what the 317 chip is capable of.

Is there a way to limit adjust pot change to only 25K at a time? The answer is YES, and it's so simple I can't believe I didn't see this before now. The method is to use a 10 turn precision pot! A human hand is only capable (on a good day) of making a single turn in (let's say) 1/2 of a second. You then need to reposition your hand on the knob to turn it for another turn. About the fastest you can do that is ~1 second intervals. (unless you're Neal Peart or something).

Anyway, I speculate that there is enough time for the cap only load to dissipate 48V through the adjust circuit that this condition I've been worried about should not be a problem, if using a 10 turn adjust pot. And that removes the need for the "backflow preventer" diode, and thus substantially increases regulation capability.

Furthermore, I would probably fit a round knob on the pot shaft so that you will not be tempted to be able to turn it faster than say one full turn every 1/2 second. (A chicken head knob might be able to be turned faster if you use your finger and push it around).

Just a thought anyway...

I didn't look into it so this might be much ado about nothing.
The condition would be created by turning pot too quickly.
What if you went to a multi turn pot
(Assuming one available)
 
If you go with a multi-turn potentiometer, a turns counter on the shaft (if you have room) can be handy. It allows one to preset the potentiometer and they include a brake (the lever on the side) to prevent accidental movement of the potentiometer.


upload_2018-2-21_12-46-36.png


If you can't find a multi-turn potentiometer to meet you needs there are planetary, vernier drive for a standard potentiometer. The can be had in various turns ratios.

Back in the day I used a Heathkit grid dip meter (actually a tunnel diode dipper) and it was difficult to set the frequency. The tuning function was direct drive right to the tuning capacitor. I added a vernier drive to it, and it worked a lot better.

upload_2018-2-21_12-55-53.png
 
I forgot about those.
I had somebody slide a SSB CB nearly 40 years ago. It had a vernier dial marked with 10,20,30 marks
 
What if you went to a multi turn pot
Yeah, that's what I was suggesting also.

If you go with a multi-turn potentiometer, a turns counter on the shaft (if you have room) can be handy.
For some reason single turn pots have a lot more friction in the movement than do the multi-turn ones so it's easy to mis-adjust them. The kind of pot you referenced might be a reasonable option if there is some friction in bearing/movement mechanism.
 
The brake on a turns counter can be applied gradually to increase the forced needed to turn the knob.

There are some vernier drives that have brakes that can be used to gradually increase force needed to turn the knob.

To be clear I did not recommend any specific potentiometer, I just recommended various things to be used with your potentiometer of choice.

BTW, some various turns counters and vernier drives had viscous damping for that smooth feel.
 
Back
Top Bottom