LM317 power supply diagnostics

[microbug]

Good - I was right  .
I've been using a 100 ohm 5W resistor which is fine, but power dissipation is above 4W so it gets hot and can only be used for so long (a couple of minutes) before I feel I need to take it out because it'll melt my breadboard . I could use 1k resistors, but my 1k resistors are only 0.25W and the heat dissipation again would be 0.4W; too high. I'm considering using a couple of the same 100 ohm resistors in series.
I was going to say that my V_be multiplier wasn't working with the LM317, but I thought to check the LM317 pinout just to be sure and I had got IN and OUT mixed up. I'm glad I didn't damage the transistor; I haven't yet put in a diode across the V_be multiplier. Having switched them around, it now works as expected, although I have not tested with any load yet. I'll try putting 100uf caps back in for -UNREG as well as doing some on the current controller tomorrow, but right now it's getting late.
I like the V_be multiplier because I can precisely trim the output to 0v, although the accuracy is only as good as my pot - it can vary from 5 to 40 ohms when you turn it to its minimum, but that only equates to a few tens of millivolts of change so I'm not too fussed. I'll have a load switch in the final design so I don't need to be able to get the output to exactly 0v, just as close as possible
Thanks for helping!

[Mark Hennessy]

In my prototype, a 2k2 resistor feeding a green LED is enough to make the rectifiers behave themselves.

Of course, there is a useful trick (forgive me if you already know) - you could take 4 of your 1k resistors and put them in a series-parallel combination, which will give you a 1k, 1W resistor. But as I say, 2k2 seems to do the trick for me.

The VBE multiplier can be improved as things progress - ultimately we can put a resistor in series with the preset to reduce the range of adjustment, while still providing enough range to cover the expected deviation. Of course, the VBE multiplier is temperature-sensitive, so some drift is to be expected, though it'll only really noticeable if you fit digital metering...

[microbug]

In the final design I was thinking about having a fixed 5v output from a 7805 regulator. This would mainly be for powering 5v logic, but would also be useful internally for providing a reference for setting the current limit using a resistor divider and driving a power LED. Although I know there is no need to use a separate regulator for this a 5v fixed output would be useful anyway so why not put one in . The power LED and small amount of current for the current limit would probably be more than enough for rectifier stability based on your setup.
I was planning to put a panel meter into the final design, so perhaps a potential solution would be to set the V_be multiplier to slightly under so the output can always be hand adjusted. I will probably have a front panel board with binding posts and a switch to choose display options for the panel meter so the V_be trimpot could go on there - the board would probably be linked by an IDC connector + ribbon cable so an extra two connections is fine. I'll have to check my enclosure to see whether there would be space for two panel meters (current and voltage), but I'll get on with the rest of the prototyping first.

[microbug]

I checked how well the -UNREG rail works with both caps at 100uF of capacitance: it's fine, but there are a few volts of ripple; it works better with C_series at 470uF (the other cap is fine at 100uF). I don't want too much ripple on -UNREG because the -1.25V rail will inherit that ripple and I'll get it on the output as the ADJ pin is driven from the -1.25V rail. On a related matter, I tested the output of the LM317 and it works fine up to 500mA, but at 500mA +UNREG is pulled down to about 17V with several volts of ripple. Should I be concerned about this?
The 2k2 resistor and LED keep the rectifier in check for me as well.
I'm also going over the design, in particular the current amplifier and limiting. What are the advantages and disadvantages of doing current measurement high side or low side? I have seen in several places that low side current measurement is better, but why? Also, I was thinking about using either:
- some sort of zero-drift op amp to cut down on input offsets (OPA335)
- a purpose-built current sense amplifier.
What would be the pros and cons of each?

[microbug]

Just to let you know, I've switched from Eagle to Upverter (online) for my EDA tool, so you can now find the schematic at https://upverter.com/microbug/d177dc3e638e9332/LM317-linear-power-supply/.
I forgot to say in my last post that I have switched to the LM833 for my op amp; it has a much lower input offset voltage (max 2mV, typical 0.15mV). Current set / sense now works using 1A=100mV to help with offset problems; the feedback resistor is now 10k so even with the LM833's large input bias current of 300nA, the offset is lower (300na * 10k = 3mv offset compared to 5mv previously).
EDIT:
The schematic is saved every time I edit it, so if you're viewing it while I'm editing it might be incomplete.
EDIT 2 (11:36AM 22/12):
OK, finished editing.

[Mark Hennessy]

Of course you're going to get ripple on unregulated rails - but beware that while increasing smoothing capacitance will reduce the ripple, it will shorten the conduction angle and increase the amplitude of the charging pulses, and hence I-squared-R losses in the diodes and transformer windings.

A few volts of ripple going into the regulator is fine. Of course, the lowest voltage combined with the drop-put voltage of the regulator determines the highest regulated voltage you can get from the regulator.

You're correct to say that any ripple on the -1.25V rail will be transferred directly to the output, so it's good to reduce the ripple there if possible. In my prototype, I took the more elegant approach of splitting the resistor feeding it (R4 in your schematic) in half, and decoupling the mid-point to ground. That has much more effect on the ripple than simply increasing C1 or C2. Of course, an even more elegant approach would be to replace R4 with a current source, but that's increasing the component count somewhat, and as any LM317-based supply is never going to be lab-grade while the LM317 is doing the voltage regulation, I'd draw the line at that point.

The basic problem is that the VBE-multiplier has a reasonably high internal impedance, so the ripple current is turned into a ripple voltage. Better references are available, but they cost more and aren't adjustable either side of 1.25V (as that's the band-gap voltage). You can improve on the simple circuit by adding another transistor - one of the best texts I can think of regarding VBE-multipliers is Douglas Self's audio power amplifier book.

Of course, it is possible to use a better reference with a higher voltage, and add a trim in series with the voltage adjust pot (e.g. TL431 - cheap as chips!). But my experience of the LM317 suggests that it's not really worth it; the VBE-multiplier is obviously temperature-sensitive, but so is the LM317. Sadly, the drift of the LM317 isn't a simple straight line, else it might be possible to compensate for it with the -1.25V reference.

Another way to reduce noise: place a capacitor between ADJUST and ground. And C3 becomes redundant then...

Looking at your schematic, I wouldn't bother using 1N5401s in all those positions. Perhaps D1-4 only, if you're planning to take more than 1A (which I definitely wouldn't for thermal reasons). 1N4002 or something is fine.

Regarding the current control, I used a single 741. You don't need a differential amplifier if you're doing low-side sensing. I used 0.1 ohms as a current shunt. Yes, it affects the output impedance, but this isn't a lab-grade PSU. As to whether you do high-side or low-side, it depends on the topology of the rest of the regulator, but in general, high-side requires a differential amplifier that can swing the whole range of the output of the supply, and its CMRR becomes an issue. Why bother? And as to the other possibilities you mentioned, the answer is the same - why bother? A simple "jellybean" op-amp does the job perfectly. If you don't mind me saying, you're doing the classic thing of throwing a lot of complications at a really simple problem because you don't yet fully realise how simple all this really is.

C7 is wrong (I think I mentioned this before) - it needs to go around the current-control op-amp.

I moved the red LED because the op-amp couldn't go negative enough to pull the output all the way to ground (I decided not to increase the negative rail). As the CC indication is useful (essential IMHO), I used a simple transistor to do it. Note how I shared parts with the power LED... A dual-colour LED would work well there, changing from green to red when in CC mode.

My prototype is by no means fully ready for production, but it's getting there...

More in a bit...

[microbug]

I'll have a look at 'Audio Power Amplifier Design Handbook' by Douglas Self.

I'll integrate the changes you made in your schematic to mine (I will likely make a PCB from this eventually so I'll need my own schematic. Re-drawing your (or others') schematics also helps to clarify the purpose of each component to me as I have to think before I place a part). I noticed that you used a 240 ohm resistor between OUT and ADJ and put another trimpot in parallel with the voltage adjust pot; I'll work out the values for that. The output voltage was stable for me with a 330 ohm resistor in the limited testing I did but it won't hurt to lower that.

I was using 1N5401s on my schematic (1N4001s in prototype) because I might want to draw at least an amp out of the power supply and I thought that perhaps the diodes wouldn't be able to handle this. If they can function at larger currents for a short amount of time (feeding the smoothing cap), I'm happy to use 1N4001s.

Thanks for clarifying the current control section. I have got a 0.1 ohm shunt, so I'll use that. I realise that there is no need for a fancy op amp for the current control; I was more concerned with the current sense amplifier but I imagine that's not needed any more. I might just try a standard non-inverting amplifier with the 0.01 ohm shunt (I just copied the differential amplifier from the original schematic which was not drawn by me), but it's probably more trouble than it's worth for my purposes. There is also the issue of contact resistance in the breadboard for me when prototyping, which is not insignificant. I'm sure I was throwing too many complications at the problem and I'm grateful for your letting me know.
Question: why have you put 10k resistors on the inputs of the current control op amp? Is it because you needed high input impedance for the inverting input (attached via a 100pF cap to the output) and put another resistor on the other input to balance it out?

For the current limit LED, I think I'll either put it on a separate resistor (when both LEDs are on, the current going to each will drop - the difference in brightness is the sort of thing that would get on my nerves) or perhaps use a dual colour LED like you suggested. I believe Dave did a video that applies to the latter about the LED on his Hakko 888 soldering station.

Typing 'conduction angle' into Google Images brings up a labelled waveform from your website as the first entry!

[Mark Hennessy]

I was using 1N5401s on my schematic (1N4001s in prototype) because I might want to draw at least an amp out of the power supply and I thought that perhaps the diodes wouldn't be able to handle this. If they can function at larger currents for a short amount of time (feeding the smoothing cap), I'm happy to use 1N4001s.

To be honest, I really think that you're pushing your luck to expect more than an amp from this. Here's why:

Worst-case power dissipation is when the output is shorted with the current limit turned to max. Perhaps there is ~20 volts across the regulator with 1 amp flowing, giving 20 watts.

To determine the junction temperature rise, we calculate the total thermal resistance from junction to free air. Perhaps we'll have a good heat sink that has a thermal resistance of 1 watt per degree C. So, in theory, the rise is only 20 degrees, right?

But, we have to also consider the thermal resistance of the case to the heat sink, which could be 1C/W again. OK, so the rise could be 40C. Fine so far...

But the killer here is the thermal resistance between the junction and the case. Which is 5 degrees per watt. So now we have 7C/W in total, so the total rise is 140 degrees C. Ambient might be 25C, so the junction temperature is 165C. Well over the maximum allowed temperature of 125C!

Even if the heat sink was "perfect", the thermal resistance inside the IC, which we can't do anything about, will be responsible for a 100C rise.

OK, so you probably wouldn't run it continuously into a short, but you never know what might happen accidentally. So, there are a few approaches:

1. Re-specify your goals. Make it 500mA instead, and that brings things back into the allowable limit. That'll be my starting point when I write all this up as an article on my website... Don't forget that you'll be able to connect your two units in parallel (or even offer a "parallel" mode, like the Thurlby units do).

2. Add a big power transistor or two around the unit, as the datasheet suggests. These will have a lower thermal resistance, solving the problem. Been there, done that. It works.

3. Use a thermal cut-out to protect the LM317. This is pragmatic, as it saves over-specifying the heat sinking which probably won't be needed 75% of the time, but protects everything against "accidents". I do this a lot, and recommend it, even if you do think everything will be OK (vents get blocked!). The simple bi-metallic units are pretty good for the money.

There are other ideas, but all a bit complex at this stage. Oh, and if you rely on the in-built thermal protection, don't expect it to be reliable. Well, that's my experience, at least

For the current limit LED, I think I'll either put it on a separate resistor (when both LEDs are on, the current going to each will drop - the difference in brightness is the sort of thing that would get on my nerves) or perhaps use a dual colour LED like you suggested. I believe Dave did a video that applies to the latter about the LED on his Hakko 888 soldering station.

Ah well - that's another trick. I was going to explain it previously, but instead left you a clue with the word "changing"...

A red LED has a Vf of 1.6V typically. Whereas it's ~2V for green. So when the red LED comes on, the green LED goes off. Neat 

Although if you like the sound of that, do check with the LEDs you have. The ones on my prototype do appear to have a similar Vf, meaning that the green one does indeed dim, and I agree that is a bit naff. If it was a dual red/green LED, it would go yellow, which isn't so bad.

Typing 'conduction angle' into Google Images brings up a labelled waveform from your website as the first entry!

So it does! Lol!

Question: why have you put 10k resistors on the inputs of the current control op amp? Is it because you needed high input impedance for the inverting input (attached via a 100pF cap to the output) and put another resistor on the other input to balance it out?

As you mentioned it first, I'll plug another article on my site - the one about op-amps.

In summary, op-amps draw a tiny DC current into each of their inputs, and each input takes very nearly the same current. This current is turned into a voltage by the net resistance "seen" by this current, and if this total resistance is different for each input, then a voltage difference exists at the op-amp inputs, which of course becomes an offset at the output.

While we're looking at this part of the circuit, I'll point out another "trick". Note the 10k current pot - which was the first one I found - is working into a low impedance (1k1). This will "distort" the law of the pot, so that it's no longer linear. Which in turn gives you more control over the current at lower settings. Yes, you could buy a linear pot, but they have certain drawbacks. And my prototype uses 10k for voltage (with 3k9 in parallel), and it's always best to use identical components where possible.

Oh, and the resistor values shown won't get you to 1 amp - perhaps only a third of that. I haven't optimised them yet - it was just what was conveniently to hand.

[Mark Hennessy]

A bit of development

...or, why it's important to prototype, and give it plenty of time for ideas for ideas to form and develop

I was planning to let this unfold gradually, but knowing how busy I'm going to be this week, I'll write it all down down now.

Earlier I mentioned that the VBE-multiplier wasn't perfect, but had the virtue of being adjustable. We also mentioned ripple across the VBE-multiplier, saying that we needed to clean up the supply feeding it...

Initially, I split the 2k2 resistor into two, and decoupled the mid-point. This is fine - it works well, and has a certain elegance. I could have left it there, but no-one should add electrolytic capacitors to a design unless they really have to. So, what else?

Instead of the capacitor, a zener diode would be good. Perhaps use 10 volts or thereabouts. But hang about - don't we already have a zener diode nearby, feeding the op-amp? Well yes - let's use that.

If so, we need to maintain the current into the VBE-multiplier (~10mA), and that requires a resistor of around 430 ohms. And to ensure that there is enough current for the op-amp and the zener, maybe the resistor feeding that from -UNREG should come down a little as well - perhaps make this something between 470 and 680 ohms. OK, that'll work...


A better reference?

I mentioned better references earlier. The VBE-multiplier has a temperature coefficient - a normal PN junction such as we find in the base-emitter junction of a bipolar transistor is around -2mV per degree C, and we're doubling that to get ~1.25V. So a not-untypical 10 degree rise will shift the output voltage by around 40mV. Quite noticeable if we fit digital metering with 10mV resolution. Even just placing a finger on the transistor shifts the output voltage by ~30mV!

We could fit a 1.25V bandgap reference, and hope that we don't need to trim it. An LM4040 is nice, not too expensive, but it possibly a little OTT here, given that it has a 100ppm/C tempco. Save that one for the next project!

As it's hard to find an adjustable 1.25V reference, what else can we do? Does the reference itself have to be adjustable, or can something else be adjusted? Well, yes.

Imagine we had a 2.5V reference - these are readily available (the humble TL431 can be salvaged from most junk SMPSs). But now, when we turn our voltage control to minimum, the output voltage would go below zero volts: to -1.25 volts in theory. Not ideal. If only we could put back the 1.25 volts somehow...

But remember, the "adjust" current from the LM317 is a fixed 5.2mA (1.25/240). So if we installed a 240 ohm resistor between the voltage potentiometer and the -2.5V reference, then this resistor would always have 1.25 volts across it. And, importantly, it could be tweaked with a preset if required.

Developing this idea further, remember that we installed a 79L05 earlier when we cleaned up the -1.25V rail. Can we use that?

Looking at the datasheet, we see that the 79L05 has a similar tempco than the LM317. It'll be better in practice as it won't see such a large range of temperatures. And the tempco of the LM317 is multiplied by the output voltage setting, of course. So, my initial thought is that the -5V rail is indeed good enough to use as a reference.

Now, we need to drop 3.75 volts across the padding resistor, which should be 720 ohm. A 680 plus 100 ohm preset will do nicely. On my prototype, I needed 560 plus 100 ohms.

The prototype behaves pretty well. Output noise is nice and low, and the drift isn't too bad. Holding the 79L05 causes the output to drift by about 2-3 mV, which is 10 times better than the VBE-multiplier. The LM317 output falls by about 50mV as it heats up. And that's in addition to the load regulation losses caused by the current shunt and the IC itself. As I've said, LM317 designs aren't lab-grade 

The attached schematic shows a welcome reduction in complexity around the negative side of things, and all with readily available cheap components. Can't be bad

[microbug]

I pretty much did all of the three approaches you suggested:
- I have added a power transistor to my design to help with thermal issues. You can see it on the Upverter page where the schematic is hosted. I'll also add a thermal cut-out to the heatsink as a safeguard.
- I decided to redefine the supply as being 0V-15V at a max of 1.5A (my transformer can only do 2A so I don't want to have a max current higher than that).
- I'll add a thermal cut-out to the heatsink as an additional safeguard in the final design. I hadn't heard of these before, but they seem to be a good idea! If I were to have two supplies in the same enclosure, where would I put a thermal cut-out (or would it be better to have two, or some other combination)? I think I could put it on the mains input, the ones on Farnell say they're rated to mains voltages.

I also worked out resistor values for the min / max voltage adjustment trimmers, the max voltage is trimmable from 13V to 18V.
I'm a little concerned about the transient response on the LM317 with a power transistor, but I haven't tested any of this yet so I'll have to see. At the time of writing the schematic is incomplete but the things I talked about here (except the thermal cut-out) are already implemented. I will need to get a few more parts (namely the power transistor) so I won't be able to test the whole thing for some time (Christmas postage times).

Yesterday I saw this on eBay - while I don't wish my finished product to have a tracking option (for complexity reasons) it would be nice to have a series / parallel / independent switch like you suggested. It would need to be rated to a decent current, or there would need to be a control board with a couple of relays.

EDIT: I'm going to stick with a 7805 as the reference by which the current limit is set (as opposed to your design which uses +UNREG); +UNREG can vary from 15V up to 30V and will change the value of the current limit when it does so. I'll keep the LM741 on +UNREG though; it might need to go higher than 5V.

Have a merry Christmas!

[megajocke]

Of course you're going to get ripple on unregulated rails - but beware that while increasing smoothing capacitance will reduce the ripple, it will shorten the conduction angle and increase the amplitude of the charging pulses, and hence I-squared-R losses in the diodes and transformer windings.

It might seem so - but only if the transformer has negligible resistance and leakage inductance, which is typically not the case. Your example waveform is quite representative of what happens in the input filter of low-power SMPS where the source impedance is comparatively low, but power supplies with transformers usually have waveforms looking more like the attached waveform with considerable flattening due to the transformer impedance. For very low capacitance values you can see changes in the conduction angle with changed capacitor value, but for reasonable capacitor values you typically don't see much change as the capacitor value is varied.

The attached waveform shows a simulation using measured transformer parameters from a 2x20 V 6 A toroidal transformer. Source resistance is 0.14 ohms (inductance was negligible) and no-load peak voltage 25.9 V (this is at low line voltage of -10%) . The simulation is for a single secondary (assuming equal loading of the other secondary), 7 mF (left) and 70 mF (right) of smoothing and 3 A DC load current. This makes the AC RMS current 5.99 A for 7 mF or 6.08 A for 70 mF, right on the transformer thermal limit. Peak-to-peak ripple is roughly 15 % (left) or 1.5 % (right). There isn't really much of a difference there.

Measuring the current when putting the components together in reality showed results very close to the simulated values.

[microbug]

I've ordered the parts I need for the latest version of the design. They'll be here tomorrow so the problems will probably start again at that time! I've been quite busy over Christmas and was saving up for a parts order, so it's been a while, but I assure you I am still alive.


Sent from my iPod touch using Tapatalk

[microbug]

Up until now all has been fine. The constant voltage section, current limit LED and -5V rail were fine. I connected the LM741, but didn't realise I had the pinout wrong: I had confused the power pins. I managed to blow both my 79L05s (and an electrolytic cap; that was exciting!) so I'll get replacements from Maplin at the weekend. I (probably) can't do it before then so the next post could be a while.

[microbug]

Almost everything is now working. The current limiting is semi-functional, but it seems that the resistor values I chose (see here) only get the current limit up to 88mV which equates to a max limit of 880mA. Despite this, the current limit seems to kick in at around 30mA which is obviously useless.

I'll do some more testing to try to work out why this is. I am using an LM833 instead of the 741 because the 741 was dead after the previous incident.

- microbug

[microbug]

All appears to be working as expected (the current limit goes to about 1.2A with my current (no pun intended) setup). I think the current limit is functioning ok, but the contact resistance of my breadboard (~5r from the rectifier to the shunt) means that I can't test it. As my breadboard is generally rubbish (cheapie from eBay), I will buy a nice Wisher one from Farnell. Thanks for your patience!

[microbug]

It works! The current limiting is correct and I have tested the current capacity up to 200mA (so far).

What is the next step? I would like to test the transient load capability with and without the power transistor - I will make an order from Farnell with some N-Channel MOSFETs, I have 100ohm power resistors which I can put in parallel (should I buy some smaller values as well?). I don't have a pulse generator but maybe I could do something with a 555 timer? Alternatively I can look out on eBay for a used one.

Let me know what you think,
microbug

[bance]

Sounds like you could use Jay_Diddy_B's dynamic load.....

https://www.eevblog.com/forum/projects/dynamic-electronic-load-project/msg288313/#msg288313

HTH Steve.

[microbug]

That looks ideal! I'll make one as soon as possible (probably a while).