32V 8A Regulated PSU Schematics

[Shredly]

Originally posted by otherside
Hey guys,

Some people here suggested as better Darington power transistor and other simple power transistor. I was reviewing the design today and noticed that if a normal transistor is used the current flowing from em to base will flow through the regulator, which means that the resistor will not limit the current and basically will not work. It that right or am i off course?

Remember that the current flowing out of the base (in a PNP, conventional current) controls the current flowing out of the collector. The current gain is variously known as h(fe) or DC Beta; this is the ratio of current out the collector to current out the base. For the MJ2955, this is between 20 and 70, in most regimes of Vbe and Ic. It can decline as low as 5, but that takes unusual and heavily loaded circumstances. What that means is that the current that the regulator will have to sink is 20 to 70 times smaller than the current that the transistor will source out its collector.

A PNP Darlington uses the emitter of the "front" transistor to sink the base current of the "back" transistor; the emitter current is the sum, of course, of the collector and base currents. Thus, by using a Darlington pair, you multiply the two current gains together; an MJE2955 Darlingtoned with an MJ2955 therefore shows a current gain of between 25 and 4900, depending on the precise current and voltage drop regime it is running in. This means that (supposing the current gain is 25, the worst case) for each amp out the collector of the back transistor, 40mA would have to be sunk; at 10A, this would be 400mA; in the most likely case (supposing it is 400) for each amp that the "back" or "driver" transistor sources out its collector, the base of the "front" transistor will need to source 2.5mA, or 0.0025A. per amp, giving 25mA for 10A, or (supposing it is 4900) 204uA, or 0.000204A, per amp, giving 2.04mA for 10A.

Since most of the 3-terminal regulators available can handle an amp or more, this should not represent a problem; however, if the transistor were not Darlingtoned, at a gain of 20, with a 10A collector current, 500mA would need to be sunk, and in the worst possible conditions, with a gain of 5, two amps would be required to be sourced out of the base of the transistor, which is well beyond the capabilities of the general run of 3-terminal regulators; high-current regulators would be required to run worst-case without a Darlington pair.

So the real design-level question is, do you need the extra current gain to keep your regulator within tolerance? If so, use a Darlington; if not, don't. Remember that there are two Vbe drops in a Darlington, so if you use Eva's diode trick, you must use two diodes in series with a Darlington. The reason this intrudes on the design is that it makes the minimum voltage drop across the BE junctions of the two transistors twice the standard drop- which may exceed the brownout of the regulator and shut down the transistors while the regulator is still working, placing the entire load on it, instead of the transistors, and potentially blowing the regulator. And this extra voltage drop not only harms efficiency, but also increases the total dissipation of the supply, which means you have more heat to get rid of.

However, your design goal of 5V drop across the regulation circuit is more than sufficient to handle this; the worst-case Vbe(on) is 1.5V for MJE2955 and MJ2955 transistors, so that's 3V maximum. Just be sure you keep this in mind.

Your choice of the LM338 means that you probably don't need to Darlington the transistors. A minimum beta of 5 means that there will always be plenty of demand, and if you use Eva's trick, then the regulator will shut down at 10A. This may well be considerably more than you need, not to mention more than your transformer can handle, and you might not even want to use the outboard transistor; see below.

Originally posted by otherside
Also since i finally landed on the amp design (bridge/parallel) the final voltage is 30V. Peak current for each LM4780 is 4.5A so a total of 9A. The toroidal would have to be at 25V = 35V DC. I Consider the following approaches sorted by cost.

1. One 500VA tr, two bridges and regulation circuits, that would be 10A per circuit. The TO-3 LM338 is far more expensive (7GBP)than a power transistor (1GBP) so one way to go is use two in parallel to spread the dissipation. At this case the dissipation will be max 5V x 10A = 50W, (Even one could handle it well) heatsinked of course.

2. Same as before transformer, four bridges and regulation circuits, that will be 5A per circuit. Again use LM338 and one power transistor per circuit. Max dissipation will be 25W.

3. Two 300VA tr (to give a little more room). Each one on two bridges and circuits with LM338 and one power transistor. Dissipation will be at max 6A x 5V = 30W.

The 3rd solution is far more expensive than the other two but my concern is mostly adjustment of regulation since component tollerance come in effect. The four regulation circuit might have slight differences in the output, i'm not sure what effect will that have on the amps and generally in the sound.

So what do you guys think is the best solution?

Thank you very much for your patience.

This is why I suggested variable resistors for the ADJ pins of the regulators. This way, you can adjust each regulator as close as possible to minimize the differential.

500VA should be sufficient to provide 14A total at 35V (caution: this is a rough rule of thumb, and you have to derate for the transformer losses- the rule of thumb for that is 10%, which will leave you with 12.6A), that would be 6.3A on each side (negative and positive) to ground; note that this needs to be a balanced load, so check that the amplifier ICs will present such a load. In general, they do. This should be more than sufficient to run them; you're not going to derate the transformer output amperage to 5A, that would be ridiculous, it's going to be in the neighborhood of 6-7A depending on the efficiency of the transformer, 6.3A by rule of thumb, so there should be plenty. You might find that a pair of LM338s is sufficient; since they can deliver 5A each, and the demand is 4.5A on each rail. This means that your amplifiers will never demand more than the regulators can supply, and the regulators will cut out well before your transformer is endangered, so this is a safe circuit (you don't EVER want to blow a transformer- it can present dangerous (LINE!) voltages because it shorts as it melts, it can start fires, and it will certainly make garbage of your amplifier by toasting everything in the casing at molten metal temperatures- in general, transformers have fusible links to prevent this, but NEVER count on them to save your life or your equipment).

If you decide you need more current handling, I'd go with the two 300VA transformers, which will give you 8.57A on each leg, for a total of 17A, and the outboard transistor to handle extra current; but be warned, your transformer then becomes the weakest link (conservatively, your regulator will handle 10A, but the transformer will only handle 7.7A or so by rule of thumb), and you'll want to be absolutely certain that you never demand more current than the transformer can send. That means including a fuse (I'd say 7.5A or less, try a standard fuse first and only use a slow-blow if it blows at startup) in the secondary circuits of the transformers to ensure they never see more than they can handle; this is mandatory if your regulator won't shut down before the transformer exceeds its capacity, optional if the regulator shuts down before the transformer is overloaded as with the previous circuit.

AndrewT will probably chip in with precise calculations for the losses in the transformers, and how to derate the total current demand using them. Before you build, you should use these calculations to be absolutely certain you won't overload the transformer(s).

As far as the rectifiers go, I'd overrate their continuous current rating by 100% in a regulated supply. They're cheap, and this will give headroom to charge the caps. Remember that during startup, large caps will present only their ESR across the transformer, and this will be further lessened by the load, so while you don't have to rate the transformer for this (luckily- they'd be half a meter across and weigh twenty kilograms otherwise), you do have to ensure that the pulse current rating of the rectifiers is sufficient to charge the caps without frying them. You don't need a continuous current rating of enough to charge the caps. Most modern regulators' data sheets give pulse rating as well as continuous rating. Just be sure they will handle this important task.

Finally, poohbah recently turned me on to a nice little power supply simulator you can find on Duncan's Amp Pages, called PSUDII; while I have issues with simulators, this one seems pretty good to me, and I've verified its calculations on paper, so I think it's trustworthy. A google of PSUDII should turn it up for you. It will make a nice final check on the ripple current ratings for your big reservoir caps, as well as a check on the pulse current rating for your rectifiers.

[otherside]

Wow what a post, thank you very much for your time i appreciate it!

Ok let see a couple of things

First of all once again i missed something on the calculations, what i was refering to regarding my amp is per channel, that removes the possibility of having 2 trans/channel =4, in terms of physical size and cost its starting to get extreme. So total max demand is 9A / rail / channel, summing up as 18A over the two rails /channel. I think a headroom of 2A to cover loses it should be enough, so we come again to same design more or less for a 500VA trans. The regulation circuit must be able to handle maximum of 10A at 30V / rail.

One solution is to use 2 x LM338 on TO-3 Package. This option has two very important issues. First of all the cost, 8 x LM338K is around 60GBP which is a lot, comparing to simple transistors. The second issue is mechanical, the heatsinks which i already have (300mm x 100mm x 40mm) will accomodate both regulation and 2 xLM4780s / channel. The TO-3 Package is very difficult to fit nicely since the fins of the heatsink are continoous in vertical, so it's almost impossible to cut the fins in order to fit the TO-3 (outside->inside). Furthermore i'll have to use the TO-220 pack.

Idealy i'd like to get off by using one LM338 (TO-220) and one TO-3P transistor which will handle 90% the current (~8A). Power dissipation and safe operating area covered of course.

Anyway

500VA should be sufficient to provide 14A total at 35V (caution: this is a rough rule of thumb, and you have to derate for the transformer losses- the rule of thumb for that is 10%, which will leave you with 12.6A), that would be 6.3A on each side (negative and positive) to ground;

The transformer will be 500VA @ 25Vac so that's 20A, with 10%derate is 18A and that's 9A/rail, which is pretty much what will be demanded at absolutely maximum power, (unlikely this will ever happen).

As far as the rectifiers go, I'd overrate their continuous current rating by 100% in a regulated supply. They're cheap, and this will give headroom to charge the caps. Remember that during startup, large caps will present only their ESR across the transformer, and this will be further lessened by the load, so while you don't have to rate the transformer for this (luckily- they'd be half a meter across and weigh twenty kilograms otherwise), you do have to ensure that the pulse current rating of the rectifiers is sufficient to charge the caps without frying them. You don't need a continuous current rating of enough to charge the caps.

I agree 100%, the ones that i'm considering have cont. load of 25A and peak 300A for short pulse, and will be mounted on the chassis which is aluminium.

Finally, poohbah recently turned me on to a nice little power supply simulator you can find on Duncan's Amp Pages, called PSUDII; while I have issues with simulators, this one seems pretty good to me, and I've verified its calculations on paper, so I think it's trustworthy.

I've tried to use it without much success, i'll try again at some point.

[otherside]

Hey guys,

Just got most of the components for the first tests of the psu.

The first approach was based on all of your ideas.

http://www.diyaudio.com/forums/attac...amp=1152561229

Initially i didn't use the diode before the LM338 just to do some testing and i have to say the diode does it's job very well. For the moment all tests are below 2 amps since i don't have at home a load that can take more than that.

I ended up using TIP2955 on a TO-3P package

I'm a little bit concerned about the emmiter resistor. Even if an equal current is shared between the LM338 and the transistor the two input resistors will have to be able to handle quite some wattage. If i calculate the wattage by I^2 * R for 5 amps a 25W resistor must be used... that's huge! (whilst the voltage drop I*V would be 5V also huge). So much smaller ohm values i guess should be used.

I think a 0.1Ohm on the LM338 and a 0.025Ohm on the emmiter should be fine. (somewhere close 0.025 doesn't exist)

What you guys think?

[otherside]

OK, testing continues

The 0.1/0.025 resistor configuration doesn't work. Basically it will work ok for the first few amps but not fore more. The voltage drop across the resistor-diode it's not enough to push the 2955 to allow large current. The drop should be around 1.8V for 8A and this is not possible with the small value resistors. I'm pretty much stuck at this point with this configuration because using larger resistors as i mentioned will mean more dissipation, more watt rated resistors, problems with voltage drop etc etc.

I did some experiments with 0.5ohm on the regulator and 0.1 on the emmiter still it didn't seem to work ok.

Another one i tested was the original one, no diode no current sharing resistors just one resistor in series with the LM338 and across it the 2955, this seemed to work almost ok, on 7A load the 2955 was taking around 4,5A and the rest from the LM338, this with a 1ohm resistor. I need to do a bit more testing on that to determine it is really working ok.

Using the simple transistor there is a problem with limiting the max current. This can be implemented using this technique (attached schematic). From "The art of electronics"

Searching around for other solutions i noticed a linear regulator controller from TI the UC3834. It can be a bit complex to build but it looks a very neat solution. I received today the samples and i'll test it tomorow. Using this device someone can virtually get as much current as the pass transistor can deliver. I'll let you know tomorow how it goes with the tests.

[otherside]

Hey guys,

I did several tests with the LM338 the LM137 and the UC3834.

Using the LM regulator device the main problem is the drop-out voltage. Starting with the output of the large cap with no load voltage is around 38V, and the regulated output very clean at 30V. As the load increases input voltage drops to 32V mean (30-34 with the ripple). I tried many different configurations with darlington, simple power, diodes, without diodes, pretty much everything. Even the non darlington power transistors require more than 2V drop e-b to provide 8A, and again there is a problem of sinking the base current when hfe starts to get low. Anyway at this point of full load 9.5A (only resistive load) the output drops to around 24-25V with the ripple being around 1.5V. which is a lot down to the target.

Another problem if the foldback method for current limiting the circuit. The LM338 has quite some tolerance on the limit of the current and this causes the foldback to work very slowly, by the time the current is limited the transistor and tranformer will be gone... which is not good. Simply the LM338 does not cut at 5A but around 7-9A which is pretty much impossible to design the foldback to work this way.

Anyway assume that using an opamp and a couple of component a current limit can be implemented nicely still the drop-out is too much. The only good point raised from the LM devices was the output ripple. Even on full load the ripple was relatively small comparing to the 5V p-p input ripple.

OK the UC3834 device now. I have to admit it's an excellent device, however it has some drawbacks as well. This device is a linear regulator controller, it can handle itself around 200mA that are used to source the external transistor. It has a lot of safety around it overvoltage-undervoltage, current monitor, foldback method, crawbar output, fault and reset and stuff like that. I tried out some basic configuration to see what comes up.

I used TIP147 which is a darlington, can handle around 90W and max 10A. On first tests i was impressed with the very low drop-out on high load, Testing the current limit without a foldback proved dangerous. I blew up one transistor because the current was limited, due to resistive load the voltage droped, dissipation went high and ... dead in less than 2 secs. Continuing the tests without the current limit proved some points.

With full load the output didn't drop at all! at least the mean lavue, because what went up was the output ripple. As the load increases the output ripple becomes the same as the input ripple maybe a bit less. This is with a 1000uF cap at the output. On lower loads the ripple was quite good. With 1A load output ripple was around 80mV p-p which is more than the LM338 but still not enough and considering that the cap will not be half a meter (oscilloscope probes) away but right next to IC this will not even be visible. The LM4870 has itself quite good ripple rejection... at least from what i hear.

So the question of the day. Is it better to have less ripple and higher drop out or more ripple and higher voltage. I honestly don't know ;-).

Tell me what you think guys.

[bigamp]

building the same thing mine uses 2 tx not ct a 36v ac . dc out is 52v per rail filtered.working on dropping to 35-40vdc ouput. tried using a tip3055 and tip2955 pass transitor with zener diodes
no load it works fine one short to ground and it blows the diodes and transistor still working on it

[pacificblue]

Originally posted by otherside
The LM4870 has itself quite good ripple rejection... at least from what i hear.

Why don't you just look it up? The datasheet is no well-hidden secret. On page 12 there is the graph. PSRR on the negative rail looks like 95 dB for 100 Hz and more than 110 dB on the positive rail.

Originally posted by otherside
So the question of the day. Is it better to have less ripple and higher drop out or more ripple and higher voltage.

Go for more ripple based on the knowledge that the LM4780 will be better off with a well-designed unregulated power supply anyhow.

Or make a listening test and hear, which version sounds better with the LM4780. If you cannot hear a difference, choose the version that has the higher minimum voltage (dropped voltage - ripple).

[h_a]

Interesting thread!

Maybe I missed it - really a lot of text here - but why don't you use a mosfet as passdevice? Probably due to the higher dissipation in the base resistor?

A mosfet would be a nice solution as it has no gate current, low rds-on and it's also very fast.

Have fun, Hannes

[h_a]

I quickly estimated a few numbers: using a 3.5 Ohm power resistor one can even use the current limiting feature of the LM-regulator (about 15 A total output current in this case with a vertical power mosfet).

Of course power dissipation for the resistor is always high and about 14 W at maximum current.*

Still that sounds very promising to me and I'm tempted to use it in my amp.

Do you see any drawbacks or other flaws that I missed?

High power dissipation is not an issue Performance wise it should be up to the LT1083 IMHO.

Have fun, Hannes

*as my amp is heavy biased at idle, the pass fet is always on.

EDIT: finally got the flaw, bad drop-out voltage. Ouch, about 8V - except one uses some funky japanese fets.