Modifying Power One Supplies for the Bench

[H713]

There has been some interest in the bench power supply I put together a while back using a pair of Power One HD24-4.8A linear power supplies.

First, a little background on the capabilities of the finished project and why I chose to go this route. I must apologize for the quality of the images- it's difficult to get good images of the inside of this supply as it is pretty cramped inside.

My primary interest has been audio. As many people into audio will note, most commercial bench power supplies are very poorly suited for audio work. Most audio circuits require both positive and negative voltage rails, and they need to be very quiet- some audio circuits are very sensitive to noise, particularly single-ended designs like those found in Neve modules. For these reasons, switchers are out unless they are of very high quality. When weighing my options for a bench power supply to replace my aging, unreliable and difficult to service early 70s NJE power supplies, I was unable to come up with any solution, new or used, that would meet my needs. Everything suitable was at least $300-$400 US, even used, (add another zero for a new one), and nearly everything that could supply the 2-3A needed for testing power amplifiers was a 3U rackmount.

The process for modifying is relatively simple. Normal Power One Linear supplies have around five volts of adjustability with the internal trim pots. This is a single-sided PCB (and not an amazing one either), so take care in desoldering these pots. The connections for them will be brought out to the pots on the front panel of your power supply.

To increase the adjustability range, resistors R12 and R14, which along with the pot form a voltage divider, need to be changed. At some point I had a schematic drawn up for this part of the supply, however that drawing has disappeared. It is not a difficult circuit to reverse-engineer if you want to do so.

A little about potentiometers for this application:
I had a pair of concentric wirewound Clarostat pots on hand (very precise), which come out to be a 5K pot (with the fine and coarse adjustment in series). As these are difficult to find these days, I suggest using a 5 or 10k 10 turn. I do not recommend using cheap carbon pots, as you will find it extremely difficult to set the voltage accurately- they are extremely twitchy. Good carbon pots like the old Clarostat and Allen-Bradley potentiometers are better, but they are still nowhere close to a Helipot or concentric wirewound. Also, pay attention to how much power you are dissipating here. It is certainly possible to exceed the ratings of a pot (especially small ones) in this application.

The values of R12 and R14 will depend on the potentiometer chosen, however for a 5K pot, the following applies:

R12: 500R
R14: 1.2K
Adjustment Range: ~10-32 VDC.

Many will consider this range to be unacceptable. By tweaking the values of R12 and R14, it is possible to move this range around a bit. I cared more about being able to output 32V than I did about being able to output less than 10V. The lowest it these power supplies can be easily coaxed into outputting is about 7.5 volts. It's been a while since I did this project, however I seem to remember being unable to get the full 7.5-32V range with a 5K pot without exceeding the dissipation rating of said pot. With a 10k pot, this would not be an issue, however you would have to recalculate (or do it by trial and error) the values of R12 and R14.

It should be noted that this is the inverting input of the LM723, so increasing the voltage on pin 4 of the LM723 will correspond to a decrease in output voltage.

Lastly, current limit. Don't expect the world here, this is a protection mechanism for the power supply, not a CCS. I never reverse-engineered the whole power supply, so I'm not completely sure how this aspect works. What I can say is that with the current limit pot at 0 ohms, the power supply current limits at about 250 mA, and with the pot open (removed), there is no current limit. More to the fact, it is extremely inaccurate. I used a 1K wirewound pot for this (500 ohms would be better), and even still about the only thing the current limit is good for is minimizing how much damage you can do.

A load switch is MANDATORY for a power supply like this- with 13,000 uF of capacitance, you can expect a good 30 seconds or more for it to bleed off. I suggest connecting the sense lead as close to the output jack as possible, but before the load switch.

The other modification I made is one that I don't consider mandatory unless you plan to push the current capabilities. By default, Power One Linear supplies use TO-3 package transistors (usually 2N3055s) with a mica insulator and no thermal compound. I struggle to understand the reason for this, as mica spacers have incredibly poor thermal transfer without heatsink grease. Perhaps sil-pads weren't available when the supplies I bought were made. While I had all the transistors removed to modify the board, I added some thermal grease to the mica washers. This is how mica washers are supposed to be used.

Depending on which model power one linear you use (the F24-12A being the biggest offender here) and the power switch you choose, you may want to build a soft-start circuit to deal with the inrush current from those filter caps.

What I wound up with, after about 6 hours of work total (most of that was wiring up the panel meters and associated switches, which are completely optional), I ended up with a power supply better suited to my application than anything I could buy for less than $300-$400. Sure I could have designed something myself- adding pass transistors to an LM317 is not difficult, and it's not like the 723 is a difficult IC to design around, but to do so would have taken considerably more time, and unless you have a large stock of transformers like I do, you can expect to spend about $40 on a power transformer, which is the same as what the two power one linear supplies cost me on eBay. Again, this is not a good solution for everyone- most people want their power supplies to go below 10 volts. For audio, I think this is a fairly decent option. Just to make it completely clear, this is not supposed to be a general-purpose lab supply.

I am a bit embarrassed by the wiring job inside. There is a lot on a relatively small front panel, and I never expected this to be seen by anyone but myself. The two 500R resistors inside the heatshrink is purely to drop the voltage for the pilot light- ugly, but free. Expect to spend about $50-$75 if you shop around and buy some used parts (especially the power supplies), and it is a very easy build. I expect the cost should be quite reasonable in other countries as well, since these power supplies are quite common in industry as well as research labs.

[duak]

When the company I was working for downsized and I was packaged, I ended up with a number of Power One supplies that would have gone to the landfill.  A pity to waste them.  Some had been put in cabinets for bench use although they were not made adjustable.  I agree with your changes and upgrades.  I've got to say that for lab use, watching the cooling is a good idea - at least a fan to stir up the air if they're inside a sealed cabinet! 

Good tip about the mica insulators - I guess they were trying to minimize costs.  I first ran into silpads in the mid-70's.  Even the best were never as good as mica or Kapton with thermal grease but there were much less messy.

By adding one or two resistors, the current limiter could be modified to consider the voltage across the pass transistor as well and introduce current foldback ie., power iimiting.  Since the 723 regulator is mounted close to the pass transistors, the thermal coupling tends to reduce the maximum current at higher temperatures.

I built a 723 based lab supply in the 70's.  The load current meter had 10, 100 and 1000 mA settings.  One thing I should have done was also switch the maximum current limit setting to match the current range of the meter to prevent it from being pinned when the current was too high for it.  The selector switch even had the extra contacts.