Transient problems on the 0-30V 0-3A supply

[liquibyte]

I need some opinions on whether or not the soft start in this circuit will work.  I know we have several threads where this design has been talked about but I thought that if this works it might be a good design for us newbies.  Note that I put back the base resistor and diode on the BD139 from the original design and I'm not sure that it's entirely necessary but it does change the way things go at the output, especially if a capacitive load is there.  The parts I'm really wondering about are the PFET soft start circuitry and if what I've done is worth trying out or if I'm way off base here.  Anyway, if some of you gurus could take a gander and comment, I'd certainly appreciate it.  Much thanks go to homebrew for his contribution of the original spice simulation.

Edit: I forgot to add the TLE2141 models to the zip, sorry.

[Strada916]

Very hard to see with a black background.

[liquibyte]

Very hard to see with a black background.

Right click, open in new tab.  Just clicking on it doesn't make it as big as it really is.

[Andy Watson]

Very hard to see with a black background.

Right click, open in new tab.  Just clicking on it doesn't make it as big as it really is.

That really doesn't help much. However, after dragging it into gimp and inverting the colour scheme:

Whatever the purpose of M1, Q6 and associated circuit is, it is a complete overkill. You would do better to install some voltage regulation to provide a few volts more than the maximum output voltage.

Examine the operation of Q2, apply some of the reasoning that prompted the design of M1 and Q6 etc, to provide a proper power-on delay and power-off detection. You might want to include some resistance between the collector of Q2 and the output of U2.

Aren't the TLE2141 rail-to-rail in operation ? Once you've sorted out the operation of Q2 you can probably get rid of the negative rail.

[liquibyte]

That really doesn't help much. However, after dragging it into gimp and inverting the colour scheme:

Whatever the purpose of M1, Q6 and associated circuit is, it is a complete overkill. You would do better to install some voltage regulation to provide a few volts more than the maximum output voltage.

Examine the operation of Q2, apply some of the reasoning that prompted the design of M1 and Q6 etc, to provide a proper power-on delay and power-off detection. You might want to include some resistance between the collector of Q2 and the output of U2.

Aren't the TLE2141 rail-to-rail in operation ? Once you've sorted out the operation of Q2 you can probably get rid of the negative rail.

Sorry folks, I tend to prefer dark colors because I'm usually up late and it's easy on these old eyes.

Q2 was put back in the circuit by me to prevent a shutdown transient that was happening.  There have been a couple of redesigns that used regulators but I was trying to understand the impetus behind the original circuit as it was published.

M1 and Q6 are also additions by me to prevent a startup transient of over 6 volts on the output.  The M1 circuit acts as a barrier to the transient.  C12 and R13 ramp the gate voltage up until after the transient passes.  D15 keeps the gate source voltage below 20 volts.  I had a problem that if C12 was charged it kept the mosfet on until it drained.  Q6, R15, R17, and R18 helps in that regard by draining the cap out after power off.  the Q6 circuit hasn't been tested in the real world yet which is why I was asking because the M1 circuit has been.  I tried several variations of the Q2 circuit here but couldn't get it to work right so I took a stab at trying the Q1 circuit there and it seems to work.

I'll agree that it's kind of hackish but I'm trying to fix issues that weren't discovered with the design until recently.  To be honest, I'm still learning and this is my attempt.  I'm sure I'll eventually work out a better regulation scheme but I'm trying to fix what I've got to work with right now.

[liquibyte]

wow the soft start is becoming very ... intricate lol ... 

i am curious what is V2 suppose to simulate? inrush?

fyi, i find that the H_fe of Q6 affects the PMOS greatly, so it maybe important to use the actual BJT you are going to build, or just sub in a large 10turn 100k somewhere

I've tried several options when it came to soft starts, reading more documentation than I've ever wanted to with nothing really working until this.  I'll agree that it is kind of involved but it seems to work.

Yep, V2 simulates the voltage spike I was measuring.  I let it repeat 3 times for good measure so I could have a point to start M1 up after.

The H_fe on the BC556B is min 180 to max 460 with typical 290 according to the datasheet.  I've never really understood this value and how it affects things but I suppose I should look into it.  After all, I've got more time in this circuit than I'd care to admit so I might as well keep on going until I get it completely done.  If you run the simulation and take out the 556 and three associated resistors, you'll see what I encountered when I built it.  It works if the cap is completely drained but if you power off and then back on, the transient presents itself at the output again.  I tried bleeders, NPN's and finally this PNP circuit and it seems to work but I'm not sure if it's, number one, a good idea and, number two, doing nicely in the sim but won't work in reality.

[liquibyte]

for hfe, being un-pro at circuit theory ... i went looking for alot of spice of BJT with diff characteristics to plug them into simulation. eg : pbss4xxx and 5xxx series have some very high hfe BJT (>700). i added my sim pic up there ... that 100R resistor is eating alot of current from the PNP base when @ high current

(too bad my ghetto circuit theory = build rough ghetto circuits lol) ...

by certain combination of your R18, you can adjust current limit ... now you have adjustable max current (of some sort)

You're right, 15.3415W accross that resistor.  Back to the drawing board to do some rearranging of the connections to see if I can get the power down to an acceptable level.

[liquibyte]

I did some tweaking and I got the power on the resistors down.  I can't seem to get the gate source voltage up above the bare minimum with the current configuration and I can't seem to get the transistor to make the voltage fall off at a decent rate that's better than seconds.  The mosfet I'm actually using is an IRF9540 that has a Vgs of -2.0 to -4.0.  I go above 4 but it takes 17 seconds to fall below 2 which is not really acceptable for my purpose.

[liquibyte]

You might want to include some resistance between the collector of Q2 and the output of U2.

This was a good bit of advice.  Sometimes I think I've gotten in over my head with this design.  I was getting a small transient at shutdown on the output of U2 and a 1K resistor helped that in the simulation.

[liquibyte]

I'm starting to think that I'm approaching this from the wrong direction but I've got something that's slightly workable and the power situation is tolerable if I use a TO-220 transistor that's rated for it.  I'm not happy with the time to get from turn off to Vgs <2V but I should probably take the advice that's been given and figure out why this is making it to the output in the first place.  I really need to start simulating some old Tek and HP supplies to see how things work there.  This is a useful tool though, I have to admit.  The more I use it, the more I like it.

[liquibyte]

you need to break apart your circuit into sub parts to simulate. that will help narrow down i think

Actually, it simulates really well.  I've been noticing some odd things when it comes to the inverting and non-inverting inputs and the outputs of the op amps.  I've been doing some research on different ways to make things stable and have had a little success but I'm taking my time with it.  I've also been working on a few other designs to get ideas.  Honestly, it could just be that this is not that good of a design and I may never be able to get it working the way it should.  I'm having fun with it though, it's a good learning experience.

[liquibyte]

I think I may be on to something finally.  I've checked the power across the soft start components and things seem to be within tolerances there.  You can replace the TLE2141's if you want but they work fine if you keep the TLE2141.301.sub file in the same folder as the 0-30V-0-3A.asc file.  I got this circuit out of some research I was doing on soft starts to the primaries of the transformers and modified it for my purposes.  The original circuit had a pot in series with R13 to adjust the delay but it didn't want to work in this context so I deleted it.  I'm still not sure why it wasn't adjusting the time delay but I'll keep playing with it until I figure it out.  I'll admit that the mosfet and pnp might not be ideal but I used the ones that were standard to LTSpice and had the kinds of specs I wanted.  I did check the datasheets on them and, to my newbie mind, they seemed fine for this instance.  Please give criticism and suggestions if you have them.

[liquibyte]

I've run into a small problem but basically things are working almost perfect with the simulation.  Either the mosfet or the transistor, or both, in the soft start is not behaving properly as it (they) switch.  I've tried several techniques recommended for taming these issues but I'm truly out of my element here.  Sometimes I'll do something that seems to work but then affects another part in a negative way.  Meh.  Sorry about the large size of the plot window but it's the only way I could get all of the divisions to show up right.  Dual monitors are nice to have, I'm now wishing I had four.

I meant to say that this is happening with the current pot turned way down and the voltage pot all the way up.  I was thinking that it was because of the load but I'm not sure because it's supposed to be current limiting.

[Andy Watson]

I've run into a small problem but basically things are working almost perfect with the simulation.

Call me cynical if you like but isn't this where you started, three threads ago? In the process of bringing up the voltage, the control section passes through an undesirable state. Can you identify how slowing down the voltage rise is going to change this?

Also, in the present configuration, Q2 will probably not appreciate having 40+ volts shoved the wrong way up its base-emitter junction.

[liquibyte]

I've run into a small problem but basically things are working almost perfect with the simulation.

Call me cynical if you like but isn't this where you started, three threads ago? In the process of bringing up the voltage, the control section passes through an undesirable state. Can you identify how slowing down the voltage rise is going to change this?

Also, in the present configuration, Q2 will probably not appreciate having 40+ volts shoved the wrong way up its base-emitter junction.

No, completely different approach to the way I'm going about it.  I've been trying to work out a way to delay the full on state with a reasonable soft start that works.  I came up with something that only halfway worked with a PFET but the charging cap ramped the gate voltage up stayed charged after power off and I couldn't figure out how to discharge it without affecting the circuit in undesirable ways.  Having said that, the part that did work, worked well.  If I discharged the cap and turned on the power supply, the spike was not measurable.

As far as Q2 goes, I've got one happily living soldered to the underside of one of my supplies and it hasn't failed yet and has eliminated the problem of the shutdown transient.  Thanks for pointing it out though,  I should probably try and play around with that side of the circuit to see if I can come up with something better there.  Some of you guys go way over our heads so that's why this is in Beginners, because I'm not always entirely sure if the things I'm doing are good, bad, or otherwise crazy.  Trying to solve the issues with this supply has me learning but I'm not trying to earn a PHD with it, just figure out how to make it work well enough that the several hundred I've sunk into it wasn't entirely wasted.  If I keep playing with it, along the way I may just discover why it doesn't work all that well and come up with something that does.

I know this is in several threads and I've been trying to get help to solve the problems and have gotten some good suggestions.  Sometimes though I just get "this is a bad design" with no other indication as to why or what to change to make it not so bad.  There's a thread where newbies are asking for a good first build and a sticky to go along with it because there's too many bad designs out there and they (we) don't know where to start.  This was my first build besides the ubiquitous 317/337 design and I'm rather proud that my first boards worked with the exception of the fact that I got the footprint of Q1 backwards.  Small problem, boards worked.  I've decided that since this is a much hated design and no one has really put forward a design that has been asked for repeatedly, that I'm going to eventually fix it and make it work well enough that it gets used without too much derision.

[liquibyte]

Some hacks you could try short out the pass BJTs base (keep it open/off) for a suitable time or the reference or put a PFET on the positive output (SSR). Use a timer or a RC to hold off the PFET for 250mS or however long it takes for  all the voltages to stabilise.

Does the overshoot just occur at start-up or anytime when the voltage is adjusted?

I'll try doing that and see how it performs.

The overshoot just happens at start up, otherwise the circuit performs admirably from the limited testing that I've done with it.  Without Q2, the output spikes at shutdown as well.  I haven't tried to short the output yet but others have and have reported that it works fine.  Current limiting works well from what I've observed.  I've done some rudimentary resistive load testing but I don't have a proper load yet so nothing too fancy there.  I really need to put together something along those lines.

[liquibyte]

I think I finally got it and I hope I haven't made any mistakes.  If anyone notices anything, let me know.  The big issue was where U2's non-inverting input connected through D12 to U1's output.  I put a diode between them and the spikes went away and the current control works beautifully if you make sure the load is adjusted right.  I think I'm finally understanding how to approach fixing this circuit and not needing the soft start circuitry.  Of course, I could be wrong but I'm going to keep plugging along and see if it can be done.

[Andy Watson]

and the current control works beautifully

Really? I'd be surprised if it worked at all.

[macboy]

I think I finally got it and I hope I haven't made any mistakes.  If anyone notices anything, let me know.  The big issue was where U2's non-inverting input connected through D12 to U1's output.  I put a diode between them and the spikes went away and the current control works beautifully if you make sure the load is adjusted right.  I think I'm finally understanding how to approach fixing this circuit and not needing the soft start circuitry.  Of course, I could be wrong but I'm going to keep plugging along and see if it can be done.

There is no "IF". The current control either works until all load conditions, or it does not work.
If you short the outputs of the supply and crank the V pot, the current control should provide perfect control over the current from 0 (-ish) to the max. Same for any other load: a 10 ohm load with a 10 volt output will draw 1 amp, but you should be able to just twist the current down to any level from 1 A to zero and see the output voltage drop accordingly.

[liquibyte]

I think I finally got it and I hope I haven't made any mistakes.  If anyone notices anything, let me know.  The big issue was where U2's non-inverting input connected through D12 to U1's output.  I put a diode between them and the spikes went away and the current control works beautifully if you make sure the load is adjusted right.  I think I'm finally understanding how to approach fixing this circuit and not needing the soft start circuitry.  Of course, I could be wrong but I'm going to keep plugging along and see if it can be done.

There is no "IF". The current control either works until all load conditions, or it does not work.
If you short the outputs of the supply and crank the V pot, the current control should provide perfect control over the current from 0 (-ish) to the max. Same for any other load: a 10 ohm load with a 10 volt output will draw 1 amp, but you should be able to just twist the current down to any level from 1 A to zero and see the output voltage drop accordingly.

You're correct and it works.  The diode seems to be all that's needed.  I've taken out the 2N4401 circuit and the soft start crap and it's stabilized the simulation.  Next up is to see if it actually works in circuit.

and the current control works beautifully

Really? I'd be surprised if it worked at all.

Then why don't you build one and report back to us how it doesn't work?  I've got quite a few engineers that have tried to help but exactly zero have offered to build and test this.  I really don't see how what I'm doing is so wrong to you.  I haven't seen any empirical evidence on your part that disproves what I've been trying.

[Andy Watson]

Then why don't you build one and report back to us how it doesn't work?

I don't need to build, nor fry the output stage to see that D17 renders D12 (and hence the current control) inoperative. You may as well not populate U1 and its associated components - you've prevented it from functioning.

[liquibyte]

Then why don't you build one and report back to us how it doesn't work?

I don't need to build, nor fry the output stage to see that D17 renders D12 (and hence the current control) inoperative. You may as well not populate U1 and its associated components - you've prevented it from functioning.

Well, it's back to the soft start again then since I can't seem to get anyone that knows more than just the bare basics to build this.  It's too bad because while I get good advice from all over, I can never get testing results from someone that knows how to do it right.

[liquibyte]

How does this look?  I moved the transistor from the output of the op amp to the gate of the mosfet in the soft start circuit.

[liquibyte]

well im lost in your circuit lol ... what does Q6 do now? it looks like a device to ... short something from the op amp, but theres a diode above it ... 

I'm still tweaking it a bit but the Q5, R26, and C13 circuit acts as a switch to the gate of Q6 activating after a delay.  The Q7, R27, and R28 circuit shunts the power to ground after switch off avoiding the transient there.  Several people have told me that the transistor on the output of the op amp was a bad idea so I moved it on a hunch and it worked.  Like I said, I'm still working on things and have added bleeder resistors to help drain the caps.  On this schematic, there's a much lower value across the main filter cap that I was thinking of trying out at 20W to see how things worked.  Most of the problems I've been having have been due to inrush and then staying charged.  Right now this is just a mental exercise but I was still hoping for a bit of advice.  I've been testing this one and at 3A it turns on just under 200ms and at 10mA it comes on at just over 100ms.  The caps take around 7.6-7.7 seconds to drain below 2V so there's that but it's plotting much better than it was now that I've moved the transistor.  I'm learning slowly, but like you said, it's kind of an R&D experiment.

I meant to add that I've bumped up some of the wattages on the resistors in favor of lower values as well.  The 1N752 D5 diode might give simulation troubles because I've got a few extra models I downloaded to have more parts to work with.  You can get the parts I installed here.  I'm attaching the updated model as well.  I've tried to pay attention to all of the limits of parts according to datasheets and such but I may still be missing something.

[liquibyte]

I think I've finally come up with something workable.  Power off has always been the big issue with me on this because if you turn back on the caps are still charged up and letting the transient pass.  I may have missed something but from what I've looked at things are looking good.  The PNP could probably be something completely different because I don't think it consumes all that much power but I've worked with the models I have so as long as I pick a part that's similar but rated for what I'm seeing, I should be good to go.  I'll leave it to anyone that wants to see the waveforms to run it in LTSpice for themselves.  I'm going to build this version and see how it goes when I have the funds.  If anyone sees anything terribly off, please let me know.

Edit:  I think D12 needs to be a 4001.

Edit again:  I uploaded the wrong .asc file, sorry folks.

[liquibyte]

I'm going to be using an LM35 based circuit to monitor the 0.47R sense resistors which are the components that get the hottest when under full load.  Based on tests I've done with the supplies I have built, the 2N3055's and the sense resistors themselves are the only components that get hot enough to require active cooling so that's what I'm going to do.  Everything else that gets fairly warm is on a decent heat sink.  The circuit itself adjusts the fan speed based on the temp it's measuring so that with little or no load, the fans are fairly quiet.  The simulation fairly accurately mimics the real circuit as far as power goes so I'm confident in the decisions I've made.

[Andy Watson]

I can never get testing results from someone that knows how to do it right.

May be, just may be, "someone that knows how to do it right" wouldn't get as far a testing it before going back to the drawing board. Just a thought

[liquibyte]

I can never get testing results from someone that knows how to do it right.

May be, just may be, "someone that knows how to do it right" wouldn't get as far a testing it before going back to the drawing board. Just a thought

Why?  Because it's not an overly complicated switcher?  Maybe because it doesn't have a microcontroller?  It's funny that you think along those lines but don't really offer why or advice to something better.  There's lots of that that goes on here and, frankly, if everyone thought like this then how would we learn.  So, you say it's bad and that even on paper its flaws can be pointed out.  Care to?  I've gotten a few ideas here on what to look for to fix this issue and I've actually managed a solution I think that can work around the flaws.  I came to these conclusions based on the advice I did get and yet I still won't stop messing with this because a few people keep telling me it's a bad design.  Why?  See the pic (thanks DH).  I see similarities in that design here and this was a production unit.  Sure, it got superceded but somone with a degree designed it and it sold.  Are you telling me that the company that produced that design was full of incompetent engineers?  All I want to do is understand what is going on enough that I can either work around it or correct the flaws so what I see now isn't happening anymore.

I want to thank everyone that has helped me to try and understand this circuit.  I can't say that I could do a better design on my own but I can say that I have learned a ton from all of the conversations on this board and from those members in particular.  Tim and David especially have a way of explaining things that make me spend hours doing research so that I grasp better what's going on.  I've fully caught the bug and, for what it's worth, I won't give up on this until it's completed.  That's why I come here.

[liquibyte]

I'm sure that some of the more experienced folks have been having a laugh at my mistake.  I failed to account for the cap holding the relay open which basically rendered it non-functional in this regard.  A couple of resistors and a transistor fixed that.