Trouble with bipolar pass element in linear supply

[xliijoe]

Dear all,

I've been designing and building a lab power supply for fun and as a learning experience. I'm at a point where it's 90% great but 10% a failure and I was hoping to get some advise from you people.

It's a 1-60V 4A linear supply, it uses one or both of the 34V secondary taps on the transformer in order to limit the heat loss in the pass element (although at 4A a 30V drop is obviously still a considerable heat output).

Parts of the supply are over engineered (like PWM fan control), some parts may have simpler and cheaper solutions (the 2x5 digit display logic most notably), please understand that this is a learning experience not a design optimised for mass production.

As pass element(s) I use two BDW39C Darlington transistors; at 100V rating they should be sufficient for the max. 90V drop or so that they see - at 10A rating they should be far above the 2A they each should see at max load. The reason I use two is in for heat dissipation onto the heatsink - a single TO220 will not be happy dissipating 120W usually.

Everything works now, at least almost. After running the supply with 4-10V out drawing less than 1A, for maybe 10 minutes or so (heatsink clearly warm, but not too hot to touch), suddenly the pass elements (Q9+Q10) weld themselves shut and I get 40-some Volt on the output (unless it was in the high range in which case I get nearly 90V).

I was looking at "Safe Operating Area" for the pass elements since I'm guessing that's the problem - but I can't find good plots for heatsinked operation.

So my questions to this forum are of the sort:
1) Is there something fundamentally wrong with what I'm trying to do here?
2) Can I recover from this by picking a different pass element?  Any suggestions?
3) Is it simply because I'm out of the Safe Operating Area or are there other things I should investigate?

Schematic in PDF attached (along with my rambling comments, apologies in advance) - please focus on the "Linear output stage" box.

Thank you all very much,

 / jakob

[coromonadalix]

hum
Why did you go all the way like this

Mosfet  pass transistors  a/d d/a  stuff etc .......   why did you not use an full linear model

Sorry when i see this  it's kinda over engineered ??

Switched xformer secondaries is the best,  but sorry  i have a hard time dealing with such an  cutted/many parts schematic like this ...  pls redo an complete section  ??

Normally  switched secondaries means you have a floating supply line for the low power section who doesn't change with xformer switching

Did you tried mosfets  wanting to have the lowest RDS value and try to have less heat ?

You wanted an gate protection with an optocoupler ?   did you design a circuit without it ?   im not sure if the opto help in this case, linearity of this device etc ...

I'm not bashing your work ... 

[srb1954]

Dear all,

I've been designing and building a lab power supply for fun and as a learning experience. I'm at a point where it's 90% great but 10% a failure and I was hoping to get some advise from you people.

It's a 1-60V 4A linear supply, it uses one or both of the 34V secondary taps on the transformer in order to limit the heat loss in the pass element (although at 4A a 30V drop is obviously still a considerable heat output).

Parts of the supply are over engineered (like PWM fan control), some parts may have simpler and cheaper solutions (the 2x5 digit display logic most notably), please understand that this is a learning experience not a design optimised for mass production.

As pass element(s) I use two BDW39C Darlington transistors; at 100V rating they should be sufficient for the max. 90V drop or so that they see - at 10A rating they should be far above the 2A they each should see at max load.

If you accidentally short the output you will momentarily get the full 90V across the series pass transistor at which point the BDW93C is only good for 150mA collector current accordingly to the SOAR curve. If your current limit is set to more than 2 x 150mA then the transistors may go into second breakdown and will short out. Even if you are operating on the lower 40V range the BDW93C is only good for a little over 1A per transistor before potentially going into the second breakdown area. The second breakdown derating on these transistors is fairly severe as evidenced by the extent of the second breakdown limit (the steeper portion of the derating curve) You may need to get some much beefier transistors to handle this level of power and voltage.

The reason I use two is in for heat dissipation onto the heatsink - a single TO220 will not be happy dissipating 120W usually.

Everything works now, at least almost. After running the supply with 4-10V out drawing less than 1A, for maybe 10 minutes or so (heatsink clearly warm, but not too hot to touch), suddenly the pass elements (Q9+Q10) weld themselves shut and I get 40-some Volt on the output (unless it was in the high range in which case I get nearly 90V).

I was looking at "Safe Operating Area" for the pass elements since I'm guessing that's the problem - but I can't find good plots for heatsinked operation.

You need to create your own derating curve for the case of transistors on a finite heatsink. The published curve is for the case where the transistor case is maintained at 25C on an infinite heatsink.

Work out the total thermal resistance (theta) from junction to ambient of your transistors, junction-case + case-heatsink + heatsink-ambient, and work out the maximum power that can be dissipated P_dmax = (T_jmax -T_amb)/Theta. Remember that the effective heatsink-ambient theta will be doubled from the data sheet specification if you have two transistors sharing one heatsink.

Using this maximum power figure plot a constant power derating line on the SOAR diagram. This will be parallel to the 45deg line on the diagram but closer to the origin. Check that the transistor operating point is always within this line and also within the second breakdown lines if these still infringe on the allowable operating area. If not, you will need to specify some larger transistors.

Lastly, I would perhaps look at some slightly larger emitter resistors, perhaps 0R22, on the series pass transistors to ensure that the current is shared more evenly between the two transistors

[xliijoe]

Ah, yeah I didn't explain this in detail. The mosfets are simply instead of a relay for switching between the lower (~40V) input to the higher (~80V) input for the pass elements.  So, as the user increases the desired output voltage past around 30V, the 'loss-of-regulation' circuity will detect that the drop over the bipolar pass elements is getting low, and therefore switch to the higher (~80V) input.  A relay would have been much simpler I agree - this part of the circuit works though.

The trouble is with the bipolar pass transistors; I opted for bipolar because I wanted big nice "tolerant" devices for linear regulation.

Thanks!

[dmills]

Second breakdown probably.
I would be looking at a couple of much butcher and more modern devices, something in TO247 or such, maybe a couple of MJL3281s or such which will each handle a few amps with 40V across them.

Note these are also much faster devices, but you will want a driver transistor as these are not Darlington's.

[xliijoe]

Thank you for the helpful replies!

I was worried there was something more fundamental wrong that I was not aware of (I don't know what I don't know).

It looks fairly obvious (now that it has been pointed out to me - thanks!) that I was driving the BJTs way out of their SOA on even rather mild loads and most certainly in any case involving something resembling a short circuit of the outputs.

I've ordered some 2SD2560 replacements; they won't fit in the PCB but I can jury-rig them to make it work; looks like they will suffer 3A at 100Vce for 10ms and at least give the current limiter a fighting chance before they weld themselves closed.

Also I ordered some 0R22 emitter resistors as suggested.

I hope to try out this change over the weekend and I'll report back here how it goes.

For now, thank you for your help - it is much appreciated!

[David Hess]

30 volts across the transistors is right at the point where the secondary breakdown considerations limit the maximum safe current below that of the power dissipation, and this limit decreases with increasing temperature.

You have to calculate your own safe operating area curves based on the calculated junction temperatures based on the measured case temperature or calculated junction to ambient thermal resistance.

The solution is to use either higher power devices or more of them.  60 watts is still high for continuous operation of a TO-220 packaged device.

[Benta]

Your schematic is indecipherable, being broken up into separate blocks connected through labels and with dotted lines around each (this seems to be the fashion on the web these days). 
Working my way into it would take hours (tip: schematics are for clarity and readability) including redrawing it, so I'll pass.
Good luck. 

[floobydust]

Yes it's no fun looking for the net labels on that style of schematic 

OP you also have to confirm your current limiter is fast - not something too slow which overstresses the pass transistors. Did you measure that?
C30 100nF (A.3) might be too much.

[Thunderer]

As pass element(s) I use two BDW39C Darlington transistors; at 100V rating they should be sufficient for the max. 90V drop or so that they see - at 10A rating they should be far above the 2A they each should see at max load. The reason I use two is in for heat dissipation onto the heatsink - a single TO220 will not be happy dissipating 120W usually.

Everything works now, at least almost. After running the supply with 4-10V out drawing less than 1A, for maybe 10 minutes or so (heatsink clearly warm, but not too hot to touch), suddenly the pass elements (Q9+Q10) weld themselves shut and I get 40-some Volt on the output (unless it was in the high range in which case I get nearly 90V).

I was looking at "Safe Operating Area" for the pass elements since I'm guessing that's the problem - but I can't find good plots for heatsinked operation.
 / jakob

Hej Jakob, few words about your transistor. The mighty BDW

1. If after the fault you get 40V at the output, I guess you have at least 40Vcollectors to gnd. By reading the SOA, at about 40V Vce the maximum current is just under 1A. It is true, you have 2 in parallel. And you roughly get about 20W dissipation on each if the output is 5V @ 1A.
2. Reading the Pd derating, if the case gets to about 75*C, you get ideally to dissipate about 40W. All this while considering a perfect thermal coupling between the case and the heatsink.

Reading all above you may say you should be safe (and you are in the range for the given case Vout=5A @ 1A), but any device failure will affect its brother device and proceed to failure.

What about the case of 20Vout and 4A (still with 40Vcollectors to gnd)? You get about 40W (ideally) dissipated on each BDW. But the SOA says at 20V you barely go to 4A. Up to the limit in everything.

You should implement the SOA of the power supply in the control of the DACs, based on output voltage to limit the current, while measuring the case temperature and applying derating. And I would consider having a single BDW (in my calculations), for safety reasons. Anyway, in time, you will understand you do not need an universal power supply. You will need several, some for low currents (very precise and fine tuning) and some for high currents (kind of a coarse tuning).

Good luck!

PS: I liked the way you integrated all, looks like built to fit a certain enclosure. But, it is preferable with a modular design, you can always update or fix faulty modules. I speak from experience.

[jonpaul]

use scope and check BJT and FETs for HF parasitic oscillatons

may need beads or series res near base/ gates

j

[Kleinstein]

The  2SD2560 are still a bit on the small side, when used for 2 A.  It may by just inside the SOA when at 25 C, but with higher temperature the SAO also shifts to lower power / lower voltage.

The easy fix may be to reduce the maximum current (e.g. to 2.5 or 3 A instead of 4 A). This may also help with another common point, where a design can go wrong a load a part too much:
The AC current loading the transformer is considerably higher than the DC current drawn after rectifier and filter capacitors. So 4 A would need a transformer rated for some 6-7 A, not a 4 A  rated one.
More on the high side with torroidial transformers and rather large filter capacitors.

[xliijoe]

Your schematic is indecipherable, being broken up into separate blocks connected through labels and with dotted lines around each (this seems to be the fashion on the web these days). 
Working my way into it would take hours (tip: schematics are for clarity and readability) including redrawing it, so I'll pass.
Good luck. 

I am a software engineer not an electronics engineer; believe me I am under no illusion that my schematic looks as if it had been done by someone who actually knew what he was doing.  That said, I do this to learn.  If you have something you feel are good examples of schematics, I'll go right ahead and read them.

Thank you for the feedback!

[blackdog]

Hi xliijoe

Another forum member also said it, this is a very poorly drawn diagram.
You make it virtually unreadable to yourself and others who would want to read it.

Here is an example of how to draw better.
The signal goes from right to left and at the end through a link with an arrow to a small audio amplifier.
This diagram is intended for monitoring the 230V system here in the Netherlands.


.
As for your Power Stage, this too has already been pointed out, you have not taken into account the SOA and that scraps your bdw93c transistors.

For your impression, think about 20 to 25 watts of dissipation for these transistors if you keep them cool enough and the voltage across them is not too high.

Better to equip your power stage with good transistors that are meant to dissipate sufficiently.

TO3 types, MJ15003, MJ15004, MJ15024, MJ15024, these are transistors that are solid enough for your application and not too fast.
Fast is good, but also requires special wiring techniques.
Drive these power transistors with a BD139/140.

My preferred configuration is the Power transistor configuration called a Sziklai Darlington.

Some information of Darlington pairs
https://www.electronics-tutorials.ws/transistor/darlington-transistor.html

You then need a PNP Power transistor and an NPN driver.
The PNP Power transistor gets a 68 Ohm BE resistor.
The BD139 driver gets a 100 Ohm resistor directly at its base to suppress generation phenomena.

Kind regards,
Bram

[Kleinstein]

Darlington transistors are somewhat convenient, but they need more (maybe 2x) emitter resistance for currentsharing. In addition the datasheet Ft values have little value - judging the speed is a bit tricky.
If more transistors are needed in parallel I would prefer separate transistors, with 1 smaller transistor driving 2 or 3 normal transistors in parallel. The BD139 may already be a bit on the small side for a driver and reach power / SOA limits.

A Sziklai type darlington replcement can get tricky and unstable on it's own. If not needed, as in the floating regulator ciruit shown, I would avoid it.


For the circuit diagram it helps to use labeled signals not too often and especially for signals that are easy to understant, like supplies or maybe digital control signals from the µC, but usually not for the analog signal paths.

[blackdog]

Hi Kleinstein,

For the configuration indicated here, the transistors I indicated are good enough for most applications.
But the BD139 can be replaced for a D44H11

Many people are indicated that a Sziklai Darlinton is unstable.
Then I can also say that many opamp's are unstable when directly driving a cable,
so that's why we should just run every opamp stage with an LM741 or an LM324.... 

I do not indicate for nothing that a small series resistor should be included directly at the base of the BD139.

Should you have to work with long wiring, which I would not do with a power supply that will deliver 4 to 5 Ampere, you can optionally add also a 2.2 to 10 Ohms in series with the emitter of the BD139/D44H11.

The Sziklai Darlinton is much better in its features than the standard Darlington.
If anyone wants to know more about that, check out the: The Art of Electronics the X Capters

Kind regards,
Bram

[David Hess]

The Sziklai pair requires a little more care to prevent local oscillation, and is a little less tolerant of a widely varying load impedance.  None of this would prevent me from using it and I often have.

[Benta]

I am a software engineer not an electronics engineer; believe me I am under no illusion that my schematic looks as if it had been done by someone who actually knew what he was doing.  That said, I do this to learn.  If you have something you feel are good examples of schematics, I'll go right ahead and read them.
Thank you for the feedback!

@xliijoe, I wasn't trying to put you down, I commend your efforts here.
But starting out on a new thing, it's often important that someone stops you going off in the wrong direction from the start (otherwise it quickly gets inborn).

Everyone has her/his own style when doing schematic artwork (yes, it's really artwork, just as PCB layout is artwork). This can be symbol styles, adding colours or shading etc. But the main goal is to make the circuit understandable to yourself and to other people.
There are a few basic rules that should be followed:

The main signal path runs from left to right through the schematic. Input: left, output: right.
Supply voltages are stacked from top to bottom, most positive voltage at the top.
Symbols are connected using connecting lines (and buses), NOT using labels.
Feedback paths usually run right to left. This simplifies routing and is immediately understandable.
Symbols are drawn as shapes with signals belonging together also being grouped together (sequentially, if indicated). Symbols are NOT drawn as a DIP/SO-16 or PLCC-44 packages with pins 1...16/44 (or whatever).
Global labels are used to get signals into and out of the schematic, not for internal connections (there are exceptions, but they are rare).

I attach two examples, one analog and one digital. These are the "rough-and-tumble" type of schematics I draw to get the best functional overview. Power flags etc. are missing, but can be added as needed when it moves on to ERC, simulation etc.
The point is, that placing/moving the symbols make for best routing, readability, and understanding.

[xliijoe]

Hi all,

First of all thank you so much for all the helpful replies!

Just giving a progress update to let you know this project is still alive.

I replaced the BDW93Cs with 2SD2560s and the plan is to de-rate the supply a bit since I don't need 4A anyway (good advice on having different supplies for different needs).  I also upped the balancing resistors from 0.1Ω to 0.22Ω and compensated the current sense logic accordingly.  Finally the dampening capacitor (C30) on the current limiting circuit was changed from 100nF to 10nF to allow for faster reaction.

With these changes I did some load testing successfully without having the BJTs weld themselves together and I measured the response time on the current limiter, since it's clear that the SOA does not leave a lot of room if the input voltage on the pass elements is in the high 80-90V range and the output is short circuited.

Looks like the current limiter takes about 150µs to limit in a milder scenario where I don't completely short-circuit the output. From the datasheet SOA the BJTs will each suffer 4A for 10ms at 100Vce; I'm not sure how to model a short-circuit scenario though. In simple theory I'm dumping the 4400µF @90V through the BJTs and a set of 0.22Ω resistors (so that's 400A each); in reality there is induction and resistance in the wires and capacitance on the output that's short-circuited too (and clearly I'd never see close to 400A). Are there any rules of thumb to apply here?

I'm attaching a photo of how I jury-rigged the new BJTs to the PCB as well as the response time plot with two signals; the current sense signal (voltage over balancing resistors) in blue ("rigol green" I suppose) and the current limiter output signal in yellow (normally high when not limiting), to show limiter response time.  Please don't mind the significant noise on those signals - the way I attached the probes I somehow picked up the PWM output for the fan/display and by the looks of it a half dozen radio stations. I could even see the current limiting diode putting on a dim light upon attaching the probe - and that's not normal, therefore I suspect the heavy noise isn't there usually when the probes are not present.

Anyway, I am currently working on the fan control (you can see the NTC leads on the heatsink in the picture) and I will be doing some more short-circuit validation. Next big step, assuming short-circuit will work, is putting the thing in a box and being done with it

Cheers,

[Kleinstein]

On a short the power transisors may well have to dump more power than just the main filter capacitors. The relays take some time to turn down the input supply.
On the positive side BJTs are somewhat limited on how much current they are willing to provide, as the gain goes down with very high current and the base current is usually limited. In this case maybe 20 mA from the OP-amp and a high current gain of maybe 1000 for darlington transistor. Ideally there could an additional fixed fast current limit (e.g. with an extra NPN transistor at the current sharing resistor(s) deverting the base current).

It looks a bit like there is some oscillation or maybe just ripple in the CC phase / high current phase.
Chances are C30 could be made even smaller (e.g. 1 nF or even 100 pF range) and maybe a series resistor of some 1-10 K could help with stability.

[xliijoe]

Hi all

I felt I should post a followup since this story has a reasonably happy ending.

After jurry-rigging in some larger pass elements and a few other changes, I got the whole thing put together in a steel cabinet I made for the purpose (learning at multiple levels here - it's all good fun).

I now have a reasonably well behaved lab power supply that fits my current needs. It's not perfect; there is still output oscillation when it's in constant-current mode and the CC circuit is surprisingly sensitive to noise from the load (so driving a cheap commutated motor will light up the CC LED even if average draw is 10% of the set limit). But it's "good enough" this time around and to make it a lot better I need to re-design much of it really. Besides, if it was perfect I would have had to make a prettier case for it too. Therefore I'm calling it done - it works, and I built it (with a bit of help - thanks again!).

Cheers all,
  / jakob