Big linear power supply and some design questions

[Da_Stier]

Hi all,

I started to build a big linear power supply a while back.
I have some design goals in mind, mostly these:

- completely analog / linear, absolutely no switching components / digital circuitry
- should provide around 9V to 15V DC
- should be regulated
- the transformer I had has 1000VA so that's my max. power

I started by preparing a old enclosure from a piece of HP UMTS testgear, I had laying around.
For a first size check I collected most of the big components.



I wanted to use what I had so I patched the old front and backpanels up with some new sheetmetal and went from there.









I also prepared the heatsink for the pass transistors with some nice bus bars for load sharing.
I used a .47Ohm resistor on each emitter and a 1Ohm resistor on each base.




Next I built a small 24V support supply for all the internal "logic" and sequencing.
This supply also contains a 30Ohm precharge resistor to softstart the big transformer.
I tried it without it and it immediately blew the breaker.






The current state looks like this:



The power path is basically done with the regulation loop of the main pass element being the next thing to do.
This is were I ran into a bit of an issue.

The 7 parallel pass transistors (just simple 2N3055s) have a combined hfe of around 30.
I want to use a similar circuit as with the support powersupply, meaning I want to use the big pass element in combination with a LM317.
Does the current gain for this configuration still apply?
I would think it still should, so when delivering around 80A at the output I would need around 2.6A at the base.

Since the LM317 can't supply this amount of current by itself, I would need an intermediate stage, correct?






Thanks and greetings,
Michael

[rikkitikkitavi]

Nice. A few feedback points to check up on:

A slight remainder that the power supply cant provide 1000W, but when you approach about 600W or so the RMS current on the transformer will reach the thermal limit of the windings, which corresponds to the 1000VA rating. Transformer VA ratings are limited by thermal loss in the windings, ie I^2*R

This is due to the charging current peaks of the capacitor bank. You can with inductors shape the current and reach about 750-800W but those would be bulky. Or a more complex active rectifier system but that probably is overdoing it.

Of course it will take quite a while to heat the transformer so for shorter periods it is possible to draw more power.

Also, I think you need to have some kind of electronic overcurrent protection. Slightest short and you will, even with 7 2N3055 blow the transistors.


I dont really see the layout, are the heatsinks not force cooled?

In that case , have you sized them properly. And if they are not installed with the fins outside I really think you need a fan to ventilate the case. Just the transformer and rectifier will shed 50W or so.

[AstRii]

You will surely need a current amplifier to drive all those transistors.

Also every transistor is slightly different so you may want to consider implementing some current balancing.

[rikkitikkitavi]

Low Hfe like 2N3055 are normally not that different in hfe so I think that ordinary emitter resistors should suffice.

Say 0,47 ohm apiece?

I have an old linear power supply, 30V 50A that utilizes 4X8 2N3055 in parallell on a force cooled heat sink and that used 0,47 ohms. It uses the older 2N3055H however. It is now dismantled, the transformer (30VAC, 400V input ) was used for an arc oven experiment but now it is just dead weight together with the insanely large inductor and the heatsinks and transistors are keept in the "good to save box". The supply was broken. probably due to the fact that the caps was datecoded 1978 and some electronic looked pretty bad. I have plenty of serversupplies 48V for other experiments at a twentieth of the weight.

[klugesmith]

Da_Stier's OP has text and pictures showing 0.47 Ω emitter resistors.
I came to question the resistor mechanical attachment with plastic tie-wraps. Want to be sure the plastic doesn't lose too much strength at the maximum resistor temperature.

[Da_Stier]

Hi all,

first of all, thank you very much for all the great feedback and tips, you did mention a few things, that I didn't consider so far. 

So let me address a few points made:

A slight remainder that the power supply cant provide 1000W, but when you approach about 600W or so the RMS current on the transformer will reach the thermal limit of the windings, which corresponds to the 1000VA rating. Transformer VA ratings are limited by thermal loss in the windings, ie I^2*R

This is due to the charging current peaks of the capacitor bank. You can with inductors shape the current and reach about 750-800W but those would be bulky. Or a more complex active rectifier system but that probably is overdoing it.

Of course it will take quite a while to heat the transformer so for shorter periods it is possible to draw more power.

Yeah, I am aware of that. The use case for the supply will be my HAM radios. They each only draw like 1A at most during receive but up to 20A on transmit.
The transmit duty cycle will be very low and there will be at most two radios transmitting so if I get around 400W out in the end, it will be plenty.
The rectifier is actually something I might need to take a closer look at, some basic tests showed a quite large and load dependent voltage drop across my bridge rectifier but I will need to invest some more time into that issue.

Also, I think you need to have some kind of electronic overcurrent protection. Slightest short and you will, even with 7 2N3055 blow the transistors.
I dont really see the layout, are the heatsinks not force cooled?
In that case , have you sized them properly. And if they are not installed with the fins outside I really think you need a fan to ventilate the case. Just the transformer and rectifier will shed 50W or so.

I plan on adding a current limit that will throttle the pass transistors to prevent current peaks.
For now I monitor the current by hand but that is definately on my to do list.
The heatsinks have no fans for now but I already added holes to mount some if necessary.
I hope that the low duty cycle on high output power is enough to keep the heat under control but this is once again a topic that I will need to see about.

You will surely need a current amplifier to drive all those transistors.

Also every transistor is slightly different so you may want to consider implementing some current balancing.

Thanks for the confirmation.
I guess I will put the output transistors in a darlington configuration with an intermediate stage.
One nice benefit might be that the reduced bandwith of such a configuration might help to prevent oscillation?
.... not sure about that though. 

Low Hfe like 2N3055 are normally not that different in hfe so I think that ordinary emitter resistors should suffice.

Say 0,47 ohm apiece?

I have an old linear power supply, 30V 50A that utilizes 4X8 2N3055 in parallell on a force cooled heat sink and that used 0,47 ohms. It uses the older 2N3055H however. It is now dismantled, the transformer (30VAC, 400V input ) was used for an arc oven experiment but now it is just dead weight together with the insanely large inductor and the heatsinks and transistors are keept in the "good to save box". The supply was broken. probably due to the fact that the caps was datecoded 1978 and some electronic looked pretty bad. I have plenty of serversupplies 48V for other experiments at a twentieth of the weight.

I tested my seven transistors (actually all 15 I bought) and they were all pretty similar with an hfe of around 54 to 58 each.
I acutally used .47Ohms right now but I might change to something a smaller.
One issue I have right now is that the LM317 configuration has no feedback from the actual output so the voltage drop across the resistors doesn't get compensated.
With .47Ohms I have: (using 40A as an example current)

U = (.47Ohm * 40A) / 7 = 2.6V

which is way to much, especially since I don't have that much headroom for the regulation.
(which might be quite a big oversight, once again, I will see about that)

I currently plan to reduce the emitter resistors to propably 10mOhms each as well as to change the regulator from a LM317 design to an op-amp design with actual feedback.

Da_Stier's OP has text and pictures showing 0.47 Ω emitter resistors.
I came to question the resistor mechanical attachment with plastic tie-wraps. Want to be sure the plastic doesn't lose too much strength at the maximum resistor temperature.

Thanks for pointing out the zipties, they were actually intended to hold them on during soldering but somehow managed to stay there.
I guess it is fine for now but before putting the lid on in the end I will propably replace them with some steel wire.


I will report back as soon as I got some more measurements done. 





Greetings,
Michael

[petespaco]

Maybe a bit off topic, but if one doesn't need a high degree of regulation, one might simply skip the whole linear regulator thing!
Here are some thoughts on regulation that I got from a rewound MOT transformer that would feed rectifiers and capacitors:

/>
Most of you are too young to remember, but back in "the olde days"  (1940's, for instance), we got by without fancy regulators on a lot of equipment.

Pete Stanaitis
---------------

 

[Twospoons]

I'd be a little concerned if you drop your emitter resistor too much then you sacrifice current sharing among your BJTs.
You might consider using one as a master and using opamps to slave the other transistors to it.  Conceptual sch attached.

[rikkitikkitavi]

I actually totaly missed you had 0,47 ohms already, and planning to use it momentarily.

But it is good that you considererd duty in the heat loss factor.

[Da_Stier]

Hi all,

I have some more updates.

Maybe a bit off topic, but if one doesn't need a high degree of regulation, one might simply skip the whole linear regulator thing!
Here are some thoughts on regulation that I got from a rewound MOT transformer that would feed rectifiers and capacitors:

/>
Most of you are too young to remember, but back in "the olde days"  (1940's, for instance), we got by without fancy regulators on a lot of equipment.

Pete Stanaitis
---------------

Thanks Pete, as you wrote, not that useful for what I am doing but very interesting to see some measurements done.
How would you go about the rectification?
I think most rectifiers would have a voltage drop dependent on the current and therefore the output voltage would change.
I'm not sure how much it would matter, maybe the effect could just be ignored in an application like this?

I'd be a little concerned if you drop your emitter resistor too much then you sacrifice current sharing among your BJTs.
You might consider using one as a master and using opamps to slave the other transistors to it.  Conceptual sch attached.

Thanks, Twospoons.
Is there some formula / literature, concering the value of the emitter resistors?
I took a look at my books but the only thing I found for now is for amplifier applications, not really in a pass element scenario.
I guess it also depends a bit on the output impedance you need for your supply.

Regarding your suggestion, I think in this configuration each op-amp needs to supply the complete base current for a pass transistor, correct?
It looks like a very interesting configuration, I might take a closer look at it in LT-Spice.

I did draw up what I was thinking about doing:



For driving the pass transistors, I would get the feedback connected after the pass transistors, of corse:



In reality I would most likely divide the feedback path by two or three and use a TL431 reference for the setpoint potentiometer.




Greetings,
Michael

[klugesmith]

Very interesting.
The resistors whose values are in milliohms invite a comment for readers new to circuit simulation.

AFAIK, in SPICE and its derivatives there's no case sensitivity.
10m and 10M both mean 10 milli-whatevers.
To get 10 mega-whatevers you need to say 10Meg or 10MEG or 10meg.

I bet that goes back to the original SPICE program in about 1970.
Written in something like FORTRAN, that has only 26 alphabetical characters.

Since the beginning, a value with U or u suffix means micro.  In some recent schematic-styled front ends I've seen µ (lowercase mu) appear. Never looked to see how that is rendered in the actual text-file SPICE deck.  Can anyone here tell us more about that?   The name SPICE deck is a holdover from punched-card days.

[Da_Stier]

Very interesting.
The resistors whose values are in milliohms invite a comment for readers new to circuit simulation.

AFAIK, in SPICE and its derivatives there's no case sensitivity.
10m and 10M both mean 10 milli-whatevers.
To get 10 mega-whatevers you need to say 10Meg or 10MEG or 10meg.

I bet that goes back to the original SPICE program in about 1970.
Written in something like FORTRAN, that has only 26 alphabetical characters.

Since the beginning, a value with U or u suffix means micro.  In some recent schematic-styled front ends I've seen µ (lowercase mu) appear. Never looked to see how that is rendered in the actual text-file SPICE deck.  Can anyone here tell us more about that?   The name SPICE deck is a holdover from punched-card days.

Funnyly enough, I actually give SPICE introduction corses as part of my dayjob. While those corses aim more  at giving a general overview (what is SPICE, where did it come from, what is it used for and what now and how do you use LTSpice as an example), I still go over the basic history.

I normally start all the way at the CANCER program and Ronald Rohrer developing SPICE with his students as a lab class.
I am pretty sure that you are right about the missing case sensitivity being a relic from the punch card / FORTRAN times.

I took a quick look at how the milli is handled in the .asc file.
It is basically as one would expect a "m".



I tried to take a look at the "M" as well as the "u" and "µ".
LTSpice actually doesn't accept the "M" any more and replaces it with a "m".
The "u" is automatically replaced with "µ" and displayed like this in the .asc file:




Greetings,
Michael

[davekni]

One nice benefit might be that the reduced bandwith of such a configuration might help to prevent oscillation?
.... not sure about that though. 

Achieving stable regulation feedback without oscillation is often the most difficult part of supply design, both linear and switching.  You need to have one dominant pole (dominant R/C low-pass).  Open-loop bandwidth needs to drop to unity due to that pole at a frequency where all other delays add up to well less than 90 degrees of phase shift.  Difference from 90 degrees is called phase margin.  In general, any reduction in bandwidth of any stage except intended dominant pole will reduce stability.  Of course, when there are embedded local feedback loops, those need to be stable as well.

The tighter your regulation requirements are, the harder it will be to achieve stability.  If you can tolerate larger voltage change with load and/or line input voltage, design will be simpler.  Don't know your actual requirements here.

I used a .47Ohm resistor on each emitter and a 1Ohm resistor on each base.

Why add resistors to bases?  Doubt they cause any problems, but I'm also not thinking of any benefit.  (For very high frequency designs, base resistance can lead to oscillation.  Makes emitter appear inductive, which can resonate with capacitance of emitter load/wiring.)

Is there some formula / literature, concering the value of the emitter resistors?

Here's a rough idea of how to calculate.  Depends on heat sinking, thermal resistance from one transistor to another.  (In other words, if one transistor dissipates 1 watt and others 0 watts, how much warmer does the 1-watt transistor junction get?)  To first-order conservative estimate, use total thermal resistance of one transistor.  I'll use 3C/watt for this example.
Also depends on Vce, highest at 9V output.  I'll presume Vce max is 10V (19V input).  May be higher at low power, but current sharing isn't critical at low power.
Vbe changes roughly -2.2mV/C.  2.2mV/10mOhm = 220mA.  220mA * 10V = 2.2W.  2.2W * 3C/W = 6.6C.  So a 1C temperature increase causes a 6.6C increase.  Thermal runaway, by a factor of 6.6.  Increasing resistor to 66mOhm would make marginal current balancing.  I'd suggest 0.1 ohms (as well as good heat sinking).  I think 2N3055 is specified as 1.5C/W junction to case.  Even with wonderful mounting, 2C/W is likely best achievable.  That's why I used 3C/W in above example.

Hope your design goes well!

[Twospoons]

Heres a BJT only way of improving current sharing. In this sim I've deliberately unbalanced the power transistor by including a 200mV offset in Vbe in Q5, which is the sort of thing that happens in thermal runaway.  But in this case the current is regulated by Q1 and Q4, which only have a few mA of Ic and so wont heat up.  Q3 and Q5 currents end up matched to ~1%.
This of course depends on Q1 and Q4 being well matched thermally.  Q2 and Q6 provide the current gain to drive the power transistors.

Again this is really only a concept - you would need to check all those BJTs are suitable for the job. (spice lets you do all sorts of things that would normally release the magic smoke)

[Da_Stier]

Hi guys,

thanks David and Twospoons, you both provided some really great information.
I think I have enogh input now to continue with the regulator as well as knowing where to look / what to look for, when I need more info.

I will propably start with my circuit first but I will definately also do a simulation for your version.

The whole project started as a learning experience, using mostly parts I already had or could get locally.
I think I might have to do some more serious calculations now to get it going well.



Thanks guys and greetings,
Michael

[Da_Stier]

Oh I also never posted the current schematic of the powerpath, so here you go.

[petespaco]

I think most rectifiers would have a voltage drop dependent on the current and therefore the output voltage would change.

I am probably "fighting over my weight" here, but----
I often use full wave bridge rectifiers like the KBPC3510,
But I have always considered that they have a constant forward voltage drop of about 0.7 volts per diode, no matter what the current is.

[Da_Stier]

I think most rectifiers would have a voltage drop dependent on the current and therefore the output voltage would change.

I am probably "fighting over my weight" here, but----
I often use full wave bridge rectifiers like the KBPC3510,
But I have always considered that they have a constant forward voltage drop of about 0.7 volts per diode, no matter what the current is.

Thanks for the insight.
You are right for a certain range of currents however if you have a very high range (like a couple hundred mA to a couple 10s of Amps) even bridge rectiviers vary quite drastically.
I looked for something similar like the KBPC3510 you mentioned on Digikey and took a look at the first result I got.

This is the forward voltage drop vs current curve.



So depending on the application you might have 2V difference after rectification which might or might not be ok. 




Greetings,
Michael

[Anders Mikkelsen]

Also consider that the peak rectifier current will be a lot higher than the average output current, since the capacitors are only charged for a small part of the mains cycle. This brings you closer to the 3 V mark for the total rectifier dropout.

[davekni]

Oh I also never posted the current schematic of the powerpath, so here you go.

You can eliminate the separate current shunt by using emitter resistors for sensing current.  If only need is over-current protection, monitoring voltage across a single emitter resistor might be accurate enough.  To increase accuracy, use 7 resistors to average voltage of the 7 emitters.  Wide range of values should work as long as all 7 are the same.  For example, if resistors are 70 Ohms, one from each emitter to a common point, impedance of that sense point is 10 Ohms (10.014 to be precise and include emitter resistors).  Compare that voltage to output node (common right side of all 7 emitter resistors in your schematic).