Please comment: 10A 13.8V linear power supply

[thinkfat]

Hi,

this was about the first project I did when I started going into Electronics as a hobby (again), I had an old 10A power supply that I used for HAM radio stuff previously but it failed, and since I wasn't happy with it anyway (very susceptible to RFI) I decided to do something with the case and the transformer. The original PSU used that standard-issue LM723+2N3055 design, but wiring was a mess and I figured some of the RFI problems came from that. I did the schematics, simulation and PCB with EasyEDA.

I don't quite remember if I had an underlying idea for the design, I think I cobbled it together from bits I found online and from application notes on the LM723.

The board is in 2oz copper, nowadays I'd probably just tin-coat the traces that need higher carrying capacity.

I only recently got around to assembling the PCB after having it laying around for months (years, even) and it seems to work, but I could not test it fully because the heatsink is not yet mounted. Anyway, it basically works, as far as I can see, but still I'm thinking about possible improvements.

Looking forward to your comments.

[Kleinstein]

Current sharing just with the base resistors is not as accurate as the more normal emitter resistors. The resistors at the collector side do not make much sense.

The OP amplifier part for the current sensing is also odd - the amplifier should work with gain larger 1. This way the resistors are less critical in effecting the CMRR.
I think one should rethink the current sensing together with the position of the 0.1 Ohms resistors in the power stage.

[David Hess]

You might get some ideas from checking the service manual schematics for the older Astron linear power supplies which are available online.  They also use the 723 but with an all NPN output stage.

Current sharing just with the base resistors is not as accurate as the more normal emitter resistors. The resistors at the collector side do not make much sense.

This is the first thing I noticed.  Move the collector resistors to the emitter side.

[duak]

I agree with Kleinstein about R23 to R26 - they should be in the emitter leads instead.  I think you will find that the PNP pass transistors will not share current equally and some will get much hotter than the rest.  Worst case, one or more will probably fail shorted.

I'm not quite sure how stable the current limit will be.  I suppose the voltage between VCC and SENS is a function of output current but it will also be affected by temperature.  I expect that the current limit will increase as the temperature of the pass transistors increases because their betas increase.  On the other hand, their VBEs decrease at the same time somewhat compensating for the increase.  Have you run temperature tests on the model?

One of neat things about the 723 regulator is that when the integrated current limiter is used with the sense resistor between the pass transistor(s) and the output connection it responds very quickly and protects the pass transistors and sometime the load too.  In this circuit, a delay is imposed by the opamp.  It should't be a big problem, but an accidental direct short can damage a pass transistor faster than the current limiter can respond.

I would add a high current diode across the output connections to prevent the supply from seeing negative output voltages.  These can occur when connecting supplies in series.

If you put an LED between the two CLIM  tags it should light when the supply is in current limit.  You might need a high efficiency LED because the current will probably be quite low.  It's probably a good idea to also add a 1K resistor from U2-2 (CL) to ground to handle any leakage current and also provide more current to the LED when in current limit.

I'd also add a thermal switch on the heat sink to shut off the supply if it gets too hot.

R18 has to dissipate at least 200 mW.  Is it sized correctly, say 1/2 W?

[thinkfat]

Thanks for the comments!

I already noticed an imbalance in heat dissipation, the misplaced load sharing resistors explains that. I wasn't able to do a real temperature simulation, the transistor model I found for the power transistors doesn't have that (or I didn't notice).

Funny thing, those base resistors were not really meant for load sharing, I just put them to lower the power dissipation of the TIP31C a bit. Ah well, I didn't know better.

The opamp circuit for the current limiting was added because I couldn't figure out a way to use the built-in limiting circuit of the LM723 without it. It seems it's just a transistor that pulls down the input of the internal output stage, the limit pin being the base of that transistor.

The current limiting is not very clever altogether, but that's probably in the nature of the LM723, it will just reduce the output voltage once you get beyond the set-point to keep the current stable, and that will of course increase power dissipated in the pass transistors. I'll have to think about a protection there - temperature sensor on the heat sink probably, as you suggested.

I guess it's a good opportunity to redo the design, I wanted to move it to Kicad anyway and to figure out how simulation works there. I'm also thinking about the mechanical construction, I think I'll flip the pass transistors around so that I can mount the PCB onto the heatsink, with the transistors under the PCB. Would make for a more compact design.
 

[David Hess]

The opamp circuit for the current limiting was added because I couldn't figure out a way to use the built-in limiting circuit of the LM723 without it. It seems it's just a transistor that pulls down the input of the internal output stage, the limit pin being the base of that transistor.

It is commonly called "Vbe current limiting" and it is intended as a rough short circuit current limit with limited accuracy.

The current limiting is not very clever altogether, but that's probably in the nature of the LM723, it will just reduce the output voltage once you get beyond the set-point to keep the current stable, and that will of course increase power dissipated in the pass transistors. I'll have to think about a protection there - temperature sensor on the heat sink probably, as you suggested.

In fixed designs the common solution is called "foldback current limiting" which reduces the current limit as the output voltage falls.  Integrated voltage regulators commonly implement this internally.  With Vbe current limiting like the 723 uses, this takes a couple extra resistors.

[NiHaoMike]

Since the high current is only needed during transmit, have you considered building a mixed linear/switching supply? Easiest way is to use a buck converter and add a current limited linear regulator in parallel set to a slightly higher voltage.

[m3vuv]

you try using a switching supply near a radio,great if you just want noise!

[xavier60]

you try using a switching supply near a radio,great if you just want noise!

The linear regulator in parallel being set to a slightly higher voltage would cause the switch mode path to be inactive during RX.

[thinkfat]

Since the high current is only needed during transmit, have you considered building a mixed linear/switching supply? Easiest way is to use a buck converter and add a current limited linear regulator in parallel set to a slightly higher voltage.

Thanks for the suggestion. I didn't consider it back then and the power supply is not primarily for HAM radio use any more.

[thinkfat]

I'm not quite sure how stable the current limit will be.  I suppose the voltage between VCC and SENS is a function of output current but it will also be affected by temperature.  I expect that the current limit will increase as the temperature of the pass transistors increases because their betas increase.  On the other hand, their VBEs decrease at the same time somewhat compensating for the increase.  Have you run temperature tests on the model?

I've meanwhile redone the design in KiCad and I've run a temperature sweep with ngspice and the current limiting drifts a lot with temperature. I've since then moved the "SENS" net to one of the load sharing resistors which are now at the emitters of the pass transistors, but I found that by far the worst problem is the crude instrumentation amplifier I implemented with the TLC2272. I guess it's the input leakage current that's causing the problem. Also, the fact that the inputs are not symmetric is definitely not helping either.

Updated schematic attached.

[thinkfat]

You might get some ideas from checking the service manual schematics for the older Astron linear power supplies which are available online.  They also use the 723 but with an all NPN output stage.

I've looked at some schematics now for the Astron supplies, currently looking at the RS-12A (http://www.repeater-builder.com/astron/pix/rs12a-1983-11-02.jpg).

Interestingly, they're not using any shunt resistor for the current limiting, instead they're using the voltage at the base of the power transistors referenced to the regulated output voltage. What do you think about that? Shouldn't their V_be also be quite temperature dependent? Do you see any compensation for that?

[duak]

In 1974 I built and still have a power supply based on the attached circuit.  It was from the Popular Electronics January 1974  issue: https://www.americanradiohistory.com/Archive-Poptronics/70s/1974/Poptronics-1974-01.pdf

It uses a current sense resistor in the unregulated side of the pass transistor Q4.  Q1 detects overcurrent and reduces drive to the pass element by turning the 723 current limit transistor on (pin 2) .  The threshold is determined by the VBE of Q1 so it is temperature dependant.  A good thing is that it reduces current at high temperatues.

Back to the OP's new circuit - the function that Q1 performs in the Poptronics schematic could be improved by closing the loop around it with an opamp.  The opamp will have to operate with its inputs at VCC so its V+ is connected to VCC and its V- is connected to GND through a dropping resistor.  The zener diode is stil connected across the opamp to limit the voltage.

The stability of the current limit can be improved by averaging the voltages across the sense resistors.

[David Hess]

You might get some ideas from checking the service manual schematics for the older Astron linear power supplies which are available online.  They also use the 723 but with an all NPN output stage.

I've looked at some schematics now for the Astron supplies, currently looking at the RS-12A (http://www.repeater-builder.com/astron/pix/rs12a-1983-11-02.jpg).

Interestingly, they're not using any shunt resistor for the current limiting, instead they're using the voltage at the base of the power transistors referenced to the regulated output voltage. What do you think about that? Shouldn't their V_be also be quite temperature dependent? Do you see any compensation for that?

The emitter series resistors add to that voltage so the temperature coefficient of Vbe is reduced.  The current limit is for fault protection and not particularly accurate.

I think that circuit also protects the driver transistor from excessive current.

Note that as the output votlage decreases, the voltage across R4 also decreases so the current limits earlier.  This is foldback current limiting.  So with an output short, the current limit lowest.

[thinkfat]

In 1974 I built and still have a power supply based on the attached circuit.  It was from the Popular Electronics January 1974  issue: https://www.americanradiohistory.com/Archive-Poptronics/70s/1974/Poptronics-1974-01.pdf

It uses a current sense resistor in the unregulated side of the pass transistor Q4.  Q1 detects overcurrent and reduces drive to the pass element by turning the 723 current limit transistor on (pin 2) .  The threshold is determined by the VBE of Q1 so it is temperature dependant.  A good thing is that it reduces current at high temperatues.

The idea to set the current limit through R15 is a bit crude, though. The wiper contact has to bear the full load current.

[duak]

If I wanted variable current limiting, I would use an opamp as I described comparing a variable voltage (corresponding to the current limit) to the voltage drop across the pass transistor emitter resistors.  By referencing this function to VCC rather than GND, I don't have to worry about matching the resistors around U2A nor the Common Mode Rejection Ratio of the opamp.  It should be possible to have a relatively stable current limit range down to 10 mA - certainly to 100 mA.
 

[David Hess]

If I wanted variable current limiting, I would use an opamp as I described comparing a variable voltage (corresponding to the current limit) to the voltage drop across the pass transistor emitter resistors.  By referencing this function to VCC rather than GND, I don't have to worry about matching the resistors around U2A nor the Common Mode Rejection Ratio of the opamp.  It should be possible to have a relatively stable current limit range down to 10 mA - certainly to 100 mA.

That is how I would do it.

In the 1970s, Tektronix did it that way in their discrete transistor regulators using differential pairs for the error amplifier and current limit instead of operational amplifiers and later they converted these designs to use operational amplifiers just like Duak describes as shown below.

[thinkfat]

That's a rather interesting design. It seems their reference is a stabilized +50V supply through a 9:1 voltage divider, which U114A provides to the base of the driver transistor through a diode. U114B keeps the base positively biased when the current limit is not reached. U114A pulls the base down to the reference voltage (+0.6V). When the current limit is reached, U114B pulls the base down and the diode isolates the output of U114A. The foldback is through the voltage divider on the negative input of U114B, which the +5V rail couples into.

For my own design, I tried sensing referenced to the unregulated power supply rail before. I don't think it worked very well. There's a significant 100Hz ripple on the VCC rail at higher loads which will probably play all kinds of funnies with the current limiter.

[thinkfat]

Allright, updated the schematics. I think I'm almost happy with this version now. It's got a nice foldback mechanism that actually pulls down the short circuit current to 0 once it goes over the trip point. Drawback - it's got an interesting bootstrap problem:

if you start it under full load, it will go into foldback immediately, due to the ramp-up of the different voltages. If I want to avoid that, I need to implement some kind of delay switch that would only connect the load after approx. 20ms or so. If anyone has a suggestion how to do that easily, please chime in.

[thinkfat]

Here's some screenshots of the pcb, managed to get it all into a 100x80 mm form factor.

Of course I realized after I launched the PCB that 16V as an analog supply rail, connecting to the V+ input of the UA723 might be a bit tight. The error amplifier might not have enough headroom to drive the output stage. In the simulation it works, but I have no idea how accurate the spice model is with these boundaries. On the other hand I cannot go higher than that because the TLC2272 doesn't like more than 16V difference at its supply rails.

Anyway, if it doesn't work I can still cut the VCCA trace to V+ and wire it up to VCC instead. It'll still work just the load regulation will not be as nice. The simulation shows more ringing on VOUT.

[duak]

Foldback limiting is a good idea to protect the regulator but is not appropriate if the load can draw a high current at low voltage during normal operation.  I see it as a way to use a smaller pass element and heatsink.

Some ideas to fix this are:
 - add some capacitance to the part of the current limiter that is sensitive to the voltage drop across the pass transistors.  This will allow the regulator to supply full current for short periods of time when the output voltage is below nominal.  I don't have time now to figure out how to do this with the existing circuit.
 - add a temperature sensitive element thermally connected to the heat sink that varies the limit current to limit dissipation in the pass transistors.  This is done in most monolithic regulators.

[thinkfat]

PCBs are in and I've spend a day assembling and testing. The load sharing works fine, I get a roughly equal heat dissipation on all the pass transistors. The foldback circuit is working fine, I just have a slight dimensioning problem and it cuts off too early. As expected, bootstrapping is quite peculiar. I added a 1µF capacitor in parallel to R17 to help it start up. If the filter caps on the VCC rail are not discharged, it won't start up if VOUT is at 0V, too. That means, if the overcurrent protection trips, you need to switch off the power supply, wait for all the caps being discharged before you power it up again. So I'll add a bleed resistor of 500 Ohms or so to the capacitor bank. A 2W chip resistor in 2512 package should be fine here.

The 2k resistor in front of the 16V Zener diode is waaay to small. 70mW is seriously too much power for a SOT-23 to dissipate. Also, the 1k emitter resistor of Q6 is nonsense. It needn't be there at all.

Another point is the load regulation, it's much slower than in the simulation. I might experiment with the value of C9 and see if it improves.