General Purpose Power Supply Design

[NewBeginner]

Bingo, I'm not as silly as I look 

BTW, I'm not sure why everyone thinks my supply is locked into using the "exotic" LT3080. You can substitute for an LM317, but then it only goes down to 1.2V which is not a big deal in most cases. Just have to offset the drive voltage by 1.2V. And that aspect can be fixed if you really want.

Dave.

Dave, I know this is an old post but I have one question:
if, in your design, you would replace the LT3080 with an LM317 wouldn't that affect the current limiting feature (especially at low set voltages / low set currents) of your design? Since even if you bring the ADJ pin of the LM317 to ground the output will still be around 1.25V.

I'm a beginner so I may be wrong (please don't throw rocks at me  ).


Thank you

[Kevin.D]

 Well spotted bug ,your right I wreckon . Short that out , and it will  just supply the max current until it folds back .

[Kevin.D]

Yes I just tried it ,thats exactly what happens . Thats a severe limitiation of using this current control method then (i.e it doesnt work .) with 3 terminal regulators that dont go down to 0 V .

[NewBeginner]

Thank you for testing this. It is good to know .

[Thor-Arne]

I suppose I should wake up this thread again since this is a really interesting design. 

I've been playing around in LT Spice with a design using BC337, MJE2955 transistors and 1N5819 Schottky diodes (as I have these parts) based on the Mark 3 design with the overshoot protection.

As I can see the circuit is quite sensitive to the parts used, this causes a lot of oscillation when the load (I put in a 500mA load) and the current reference is set at the same value.

When the current reference is set to 750mA I get the same result as with the original design, and at 250mA LT Spice claims that the output is -1.7V.

I must say that neither LT Spice or PSU design is my strongest points, so I really struggle to see what's actually going on here. 

I've put together the circuit on a pref-board, and I can confirm the analysis in LT Spice.

Disregarding the current limiting issue, the regulating is just perfect, with the input at 12.11V I can regulate from 0V to 11.99V.
I didn't spend that much time testing this, but it seems like the circuit should be able to regulate up to 1A with these parts without getting too hot (assuming a pre-regulator supplying 2V above the output voltage).

Attached the .asc file if anyone wants to play with it.

[amspire]

The high gain and high speed of the 2N2222A transistors is pretty vital. The  2955 is slow and low gain. It will wreck the phase margin and it will need a higher current driver - which causes a host of other design issues.

The current transistors add negligible phase shift till over 100khz which is well past the 0dB loop gain frequency. The 2955 will be adding phase shift right at the wrong point causing instability and ringing on transient load changes.


Richard.

[Thor-Arne]

I expected you'd say something like that, you mentioned that the speed of the 2N2222 was important in a earlier post.

Regardless, I decided to put it together with the parts I had laying around anyway. I need to play with these tings to get a grip on it, and it's not like it's expensive (less than USD 3 for what I put together).

I think I'll order some 2N2222's next time, need to check the power dissipation and such with those.
I really want to pursue this design as I have a nice bunch of toroid transformers that I want to use.

BTW: Is TO-18 or TO-92 best suited to this design?

[SeanB]

Buy 2N2219A's instead. Same die but in a bigger case and can dissipate more heat. As well you can get push on heatsinks to fit them, so you can do 1W of power dissipation as well.

[Thor-Arne]

Hm, Ideally I'd want to use TO-92.
TO-18 cost 6 times more than TO-92, and the 2N2219A is 9 times more.
A quick check on RS. I can get TO-18 from Tayda to about the same price as TO-92 from RS.

[amspire]

The TO92 packages are fairly good for power and very cheap. Should be able to get a bunch at well under 10c each.

There is a really nice surface mount package - the SOT223 versions such as the PZT2222A it is rated at 1.15W but I usually divide numbers like that by 2 at least.

http://www.nxp.com/documents/data_sheet/PZT2222A.pdf

[c4757p]

the SOT223 versions such as the PZT2222A it is rated at 1.15W but I usually divide numbers like that by 2 at least.

Yeah, I'd be pretty hesitant to dissipate 1W in a SOT-223.

[Thor-Arne]

Think I need to consider using SMD parts for this design, but I generally only have through-hole parts here. So for testing the regulator I'll go with that.
I haven't really looked much at the pre-regulator, and it hasn't been discussed here either. But I noticed that Dave used a LT1935 in one of his simulations, this can only be found as SMD. Yet another reason for going SMD.

The PZT2222A is about 50% more expensive than the TO92, not that it matters, the BOM for this is really low cost.

Btw, the PZT2222AT1G from ON Semi is rated at 1.5W.

[amspire]

Btw, the PZT2222AT1G from ON Semi is rated at 1.5W.

They achieve this power by using 1.6 x 1.6 inches of copper on a PCB at 25 degC ambient. That is a lot of space. I think a 20mm x 20mm board heatsink at 60 degC ambient is more realistic for a working design, and I would not go beyond 0.7W and 200mA per transistor. Above 200mA, the transistor gain drops off quickly.

[Thor-Arne]

I see. Even for 20*20mm it's a lot of space, this is multiplied by the number of transistors also.
Which in turn increases the board cost, many things to take into consideration.

I was hoping to get 1A out of this design, that would mean 125mA for each transistor if using 8. The voltage drop is another issue, you said 2V above the output this gives 0,25W for each transistor (correct?).

[gxti]

Thanks so much for bumping this thread, I'd totally missed it. I've been tinkering with my own 5A lab supply concept, first with a dual LT3083 and now exploring a discrete configuration. The problem is I have only a very small, foggy clue about compensation and stability -- the big downside to getting into electronics with no education. Ironically this thread seems to have convinced me that I need to kick the whole thing in the ass and go back to a simpler, feed-forward design using the LT3083 and just driving the set pin, not worry about the voltage dropped across the (post-regulator) shunt, and lose the isolation and STM32F3 which was handling ADC/DAC duties. My original motivation was to make a moderate-current supply with fairly precise current monitoring for digital applications, and worrying about 10mV here and there is not in line with that application.

Still, I do want to make a nice lab supply board that I can stack to get multiple channels, and I may be happy with 1A for that. I'm really interested in the flyback converter in particular as it would solve the problem of needing the bulk input power to be isolated which was one of the issues I had been putting off in the first design. I've you've developed it any further I'd love to hear about it.

[Thor-Arne]

This design is made to be scalable by adding more regulator boards. In my case I'm limiting myself  to getting 1A because I have quite a few 24V/1A toroidal transformers that I want to use.

The pre-regulator hasn't been discussed here yet, so who knows how that's going to scale.

There's been some discussion on the control circuit, and so far it seems like it will be very accurate. I don't remember all the details, but it's in the thread somewhere.

I have another lab-psu I'm working on to, but that is on hold for now since I ran into some issues I need to figure out.

Anyway, can't get enough psu's. So I'm going for this one to. 

[Rerouter]

gtxi, care to keep us posted on the progress you make with a regulated isolated flyback, i was planning a similar approach myself for a 5A discrete varient, but couldn't nut out the feedback mechanism,

[amspire]

I am tidying the circuit up with a pre-regulator, and with the circuitry for parallel and serial tracking operation.  Sorry for the delays - it is just a part time task and I am playing with different pre-regulators right now.

I think I will add a overtemp protection to it as well.

Richard.

[IanB]

The high gain and high speed of the 2N2222A transistors is pretty vital. The  2955 is slow and low gain. It will wreck the phase margin and it will need a higher current driver - which causes a host of other design issues.

The current transistors add negligible phase shift till over 100khz which is well past the 0dB loop gain frequency. The 2955 will be adding phase shift right at the wrong point causing instability and ringing on transient load changes.

You know, it would be brilliant if someone like Shahriar could do a video tutorial on how you do these kinds of phase margin stability calculations and how you tie them in to the circuit topology and datasheet parameters of the components used (hint, hint  ).

I know the same basic theory applies to the stability of general feedback control systems, but it's not something I've ever had to use in my career thus far. So it's fascinating to see this stuff used to solve real engineering design problems!

(So yes, when you are introduced to stuff at university and you ask yourself, when will I ever need to use this in my job? The answer is, maybe, for some of it, you won't. But you won't know exactly what until after you have retired. So you'd better learn all of it, just in case    It could be that Bode plots are not just an academic exercise, but a lifesaver!)

[BravoV]

You know, it would be brilliant if someone like Shahriar could do a video tutorial on how you do these kinds of phase margin stability calculations and how you tie them in to the circuit topology and datasheet parameters of the components used (hint, hint  ).

+1, I've been expecting one our resident expert to bring this topic too.

What scared me off as an hobbyist is when it comes to this compensation stuff, even at some datasheets from high profile companies mentioned that they have to walk back & forth from the component bins to the prototype circuit to find the right combination  , yes, this is literally written at their datasheet. 

[Thor-Arne]

I am tidying the circuit up with a pre-regulator, and with the circuitry for parallel and serial tracking operation.  Sorry for the delays - it is just a part time task and I am playing with different pre-regulators right now.

I think I will add a overtemp protection to it as well.

Richard.

No worries, take your time.
It's not my intention to nag you to get this done.

Perhaps the overtemp protection is best placed together with the mcu? DS18B20 perhaps?

What about voltage references, without checking the datasheet I'd expect the internal reference in the  mcu isn't that stable. Perhaps LM336 would be suited?

And what about ADC? The internal ones in the mcu doesn't have enough bits?
I suppose one could calculate the voltage, but the amps drawn needs to be measured.

[gxti]

gtxi, care to keep us posted on the progress you make with a regulated isolated flyback, i was planning a similar approach myself for a 5A discrete varient, but couldn't nut out the feedback mechanism,

I've never made an isolated converter in my life. I once tried to make a non-isolated mains converter straight out of a Power Integrations datasheet and it wouldn't start. So I'm not the guy to be making something like that right now, especially something with jellybean parts that will be around for 20 years. But I've been playing in LTspice anyway to see if I can get a better understanding of how isolated converters work.

Sorry for the delays - it is just a part time task and I am playing with different pre-regulators right now.

Me too. I'm not rushing you either, I'm just excited about the ideas that have been presented so far.

[HackedFridgeMagnet]

gxti: Im not 100% sure but I think a forward converter would be more appropriate for the isolated pre regulator than a flyback converter.

Would it be better done as three stages.
1. An off the shelf mains to 48vdc. Which probably would be a forward converter.
2. non isolated buck pre regulator.
3. linear regulated stage, with current limiting.

[gxti]

What you describe is essentially what I was originally targeting, but buying a separate power supply module for each channel would add up quickly. What piqued my interest here is the idea of having one off-the-shelf main supply (perhaps even an ATX supply) and low-voltage, low-cost isolators for each channel. You might be right about forward vs flyback, again I have no idea what I'm doing and that includes choosing a suitable topology.

The way I envision it (and here I go, overdesigning already) each channel card would have two transformers. One main with a variable output that tracks the output voltage. And one housekeeping supply to power the digital bits. Perhaps +/-5V with post-regulators so it could also provide a negative supply to the opamps. The digital portion would include an inexpensive microcontroller to act as a DAC for the voltage and current set points (as discussed in this thread), and to monitor all the unregulated converter outputs where a built-in 8 or 10 bit ADC is adequate. For precise voltage and current monitoring you could either use a popular discrete ADC or something like INA226 that can monitor a current shunt and voltage directly and very precisely, with a I2C interface. The microcontroller would then interface with the master over isolated SPI so it can easily select each channel card.

There are few interesting variations here in terms of where the parts go. For example, the housekeeping converters of all the channel cards might be clocked from a single bus by the master, since they all operate with the same voltage and current loads. The channel card microcontroller might drive its own main supply converter through the isolation barrier, one benefit of having a separate housekeeping supply. It might be a bad idea though, depends on how well inexpensive high-speed digital isolators perform in terms of jitter etc. Worst case the master micro can drive it but it wouldn't be as responsive because it would be receiving feedback through the serial interface.

Another variant is to eliminate the channel card microcontroller, use an I2C isolator to bring out the INA226 measurements, and drive both converters and the PWM DAC from the master. But you lose the housekeeping measurements and you have to deal with giving the INA226s unique addresses since they'd be on the same bus and couldn't negotiate an address automatically. Plus, as discussed here already, micromanaging one PWM channel is difficult enough. Doing it for 4 channels while also working the user interface or even Ethernet is stretching things a lot although a Cortex-M3 could probably pull it off.

So yeah, there's some wild speculation for you. Hopefully you fine folks can refine and focus it a bit.

[amspire]

Last time I worked on the supply last year, I was trying to make a switching isolated converter - the idea being it was easy to have, say, one earthed ATX supply powering multiple isolated supplies. It would have been fantastic if successful as it would make it really easy to use almost any kind of power source to power an isolated supply. It would have also made it really easy to have multiple tracking supplies in series to make dual supplies and higher voltage supplies.

I was working on something like 3 to 15V input, and 4 to 30V output at 1A.

I had no problem making a small and very cheap (about $4 in parts) prototype converter with a wide input and output range and good power capability. The problem I ran into was that with cheap solutions for transformers and other components, I was getting far too much leakage inductance in the transformer leading to too much waste energy turning into noise on the output.  With a wide ranging converter, it was easy to get peak currents in the transformer of up to 20A for a 1A output supply, and so it doesn't take a big leakage inductance to generate some pretty powerful noise. I just do not want a supply that sends your scope nuts when you use it to power a circuit. I want to be able to put my DS1105E on 2mV/division and get no artifacts generated by the supply visible when monitoring a test circuit.

It is much easier making a low noise non-isolated switching converter as you eliminated the leakage inductance problem particular to transformers, but I haven't worked on any non-isolated converter designs.

For now, I will stick to a linear pre-regulator first so I get a very low noise output.

I am still very interested in any successes anyone else has with a switching regulator. It would be great to eliminate the heat and inefficiency of a linear regulator.

Richard.