Stacking LT3045 for higher output voltage

[matthuszagh]

I'm designing a power supply that receives AC mains and provides +26VDC for a load. I'd like use LT3045's as the final regulator in order to get the performance benefits of that regulator.

My plan was to do the following (see attached image). I'll use a transformer with 2x 24VAC output windings. These windings will be (full-bridge) rectified and filtered and each connected to an LM350. The output of each LM350 goes to an LT3045, each configured for +13VDC output. The ground terminal of one LT3045 will be connected to the positive output of the other, such that the voltage difference between the ground terminal of the 1st and positive output of the 2nd will be +26V. Anything wrong with this configuration, or anything in particular to be aware of? Or, is there a better way to do this? Or, just other ways I might consider?

I'll actually need to power 3 or 4 +26VDC loads, and a single LT3045 can only supply one load, so each LM350 output would actually be connected to 3 (or 4) LT3045s, each with the configuration described previously.

[David Hess]

What you are doing will work fine for increasing the output voltage, however if one side goes into current limit, the other side may drive its output in reverse.  To prevent this, each output should have a power diode connected across its output to prevent reverse voltage.

[jbb]

How about an LT3045 to make +13V and an LT3094 to make -13V.  Or an LT3097 to make +13V and -13V at the same time?

[David Hess]

How about an LT3045 to make +13V and an LT3094 to make -13V.  Or an LT3097 to make +13V and -13V at the same time?

There is no need for different positive and negative regulators when two floating sources are available.

[matthuszagh]

What you are doing will work fine for increasing the output voltage, however if one side goes into current limit, the other side may drive its output in reverse.  To prevent this, each output should have a power diode connected across its output to prevent reverse voltage.

Thanks David. Would you mind providing a little more detail about the situation here? If the load shorts, what would cause one of the regulators to have a reverse voltage applied across its output? I looked around a bit for information on this but came up mostly empty-handed. I found this post (https://electronics.stackexchange.com/a/444904) but didn't really understand the explanation. Thanks!

[magic]

I don't understand this diagram. Isn't it the case that you wanted to connect the upper pair of LT3045 to the upper LM350 and GND1 and the lower pair of LT3045 to the lower LM350 and GND2?

Also, I suppose LT3045 are not rated for 30V or so input voltage or you would simply use a single one. But here's a problem: the output of the lower LT3045 in each pair starts at 0V when its load capacitor is discharged and then the upper LT3045 in the pair sees full input voltage across its GND-VIN. There is also problems if the output is shorted.

[matthuszagh]

I don't understand this diagram. Isn't it the case that you wanted to connect the upper pair of LT3045 to the upper LM350 and GND1 and the lower pair of LT3045 to the lower LM350 and GND2?

No, the upper pair of LT3045s (which power a single load, collectively) cannot have the same ground potential. The output of one is used as the ground of the other. The numbers by the ground nodes signify separate grounds. But, the ground of the upper LT3045 powering the first load should have no problem being at the same potential as the upper LT3045 of the 2nd load. The diagram accurately reflects my intention.

Also, I suppose LT3045 are not rated for 30V or so input voltage or you would simply use a single one.

Correct, they're not. But even if they were I would still use a pre-regulator. The LT3045s aren't designed for a lot of heat dissipation, so the LM350s are there to dissipate most of the heat and provide input voltages to the LT3045s that give them good regulation and high PSRR (vout + 2V) while minimizing their power dissipation.

But here's a problem: the output of the lower LT3045 in each pair starts at 0V when its load capacitor is discharged and then the upper LT3045 in the pair sees full input voltage across its GND-VIN.

I'm not convinced this is true (I could be wrong). The LM350 will regulate between its output and gnd1. gnd1 is also the LT3045 ground so it should try to keep the voltage across the input of LT3045 at its setpoint. At power on the capacitors across both LT3045s will be at 0V so they'll both need to ramp up. In order for the situation you describe to happen, the LM350 would have to increase its output well above its target. Why would it do that?

There is also problems if the output is shorted.

Yeah David Hess was alluding to this too, I think. It sounds like the solution is diodes across the outputs. But I'd very much like to understand this, so if you could provide any more details that would be much appreciated.

[magic]

Something is still not quite right because two LT3045 have their ground pins shorted together by GND2 and their outputs shorted together by GND1. They will be fighting each other.

If they are supposed to be two completely independent, floating 26V rails, you need to break those connections.
If they are supposed to be a pair of ±13V rails supplying separate loads, connection through GND1 may remain.
If they are supposed to be a pair of +26V rails supplying separate loads, connection through GND2 may remain.
In each case, the connections between LM350 and LT3045 will probably need to be rearranged a little.

I would recommend using the ground symbol only for true ground and drawing explicit wires for everything else. This will make it easier to spot problems.

[matthuszagh]

Something is still not quite right because two LT3045 have their ground pins shorted together by GND2 and their outputs shorted together by GND1. They will be fighting each other.

Ah, damn you're right. Thanks for pointing this out and saving me some headache.

If they are supposed to be two completely independent, floating 26V rails, you need to break those connections.
If they are supposed to be a pair of ±13V rails supplying separate loads, connection through GND1 may remain.
If they are supposed to be a pair of +26V rails supplying separate loads, connection through GND2 may remain.
In each case, the connections between LM350 and LT3045 will probably need to be rearranged a little.

"If they are supposed to be a pair of +26V rails supplying separate loads, connection through GND2 may remain." This is what I'm trying to achieve. I can keep the GND2 connection, but the upper LT3045 and LM350 will need the same ground connection (GND1) so I don't know how to break that. The LT3045s can be paralleled (described in the datasheet), so I could power all loads from the same set of paralleled LT3045s. However, I explicitly used the former method to provide greater noise immunity between the power supplies (the loads are very stable OCXOs and I need to avoid injection locking). There are other ways to achieve the noise immunity at 10 MHz (where I need it), but the LT3045 has 50 dB of PSRR at 10 MHz and I'd like to make use of that if I can.

[magic]

Note that the "lower" LT3045 are doing nothing except providing a stable 13V reference voltage for the "upper" ones and sinking their quiescent current (BTW, are these regulators able to sink current? I'm not sure.) They could be replaced with only one and the preregulator supplying them doesn't need to be very powerful at all.

The two secondaries of the transformer and two LM350 could power each of the "upper" LT3045 independently, maybe crosstalk would be lower this way.

edit
Sorry, I was thinking about something other than your schematic again.
In your case the "lower" are working as hard as the "upper", but the problem is that the "lower" are shorted and the "upper" share a single LM350 supply.
Perhaps it would make more sense to utilize the the two transformer windings to produce two completely independent supplies. 24V AC may be enough for 26V DC output without series connections.

[PCB.Wiz]

There are other ways to achieve the noise immunity at 10 MHz (where I need it), but the LT3045 has 50 dB of PSRR at 10 MHz and I'd like to make use of that if I can.

That number only matters, if you have significant 10MHz on the power supply input.

However, I explicitly used the former method to provide greater noise immunity between the power supplies (the loads are very stable OCXOs and I need to avoid injection locking)

Separate regulators and grounds is the best way to ensure separation.
What are the exact loads, and load lines, you need in Amps/Volts x How many ?

I notice the LT3045-1 has a VIOC function added, to assist an upstream tracking regulator.
They give a switching regulator example.


Most of the high performance comes from the LT3045 having a filtered reference, with a tracking follower, so you could also consider a low noise higher voltage opamp + power device as a tracking follower ?

[chuckb]

Overall this approach looks good. A few points

Diodes on the outputs are required because the current limit points of the two regulators are not identical.
If the top regulator limits at 400ma and the lower limits at 500ma during a short the upper regulator will have 100ma driven through the chip backwards. The difference in current limits will not be this bad but that’s the idea. If one chip overheats and shuts down then all the current from the other regulator will be driven through it. So it needs a diode to bypass the current around the chip.

The lower regulator needs a 3-4 ma resistive load on it. The upper regulator has a typical 2.2 ma operating current. That current needs to go somewhere. If the lower regulator can not SINK the current the mid- point voltage (13V) will rise and go out of regulation. At 500ma output currents the operating current can rise to 25ma (base current of the upper pass transistor) but I think that will be lost in the 500ma passing through the lower regulator.

Depending on the load requirements you may want an overall output capacitor across the 26v output. With the two 10ufd output caps in series the load will see 5uf of cap. You may need more cap to support your load.

If you think the LT3045s will be running hot you could reduce the input voltage to each regulator to 14V or less. At very low voltage they need a minimum of 2V input for all the electronics to wakeup and operate. At higher voltage they can operate with a lower minimum in-out voltage, down to 0.3V.

Invite AnalogTodd (on this forum) to review the design. He designed the chip while at Linear Tech.
Good luck!

[David Hess]

What you are doing will work fine for increasing the output voltage, however if one side goes into current limit, the other side may drive its output in reverse.  To prevent this, each output should have a power diode connected across its output to prevent reverse voltage.

Thanks David. Would you mind providing a little more detail about the situation here? If the load shorts, what would cause one of the regulators to have a reverse voltage applied across its output? I looked around a bit for information on this but came up mostly empty-handed. I found this post (https://electronics.stackexchange.com/a/444904) but didn't really understand the explanation. Thanks!

The outputs of the separate regulators are in series, so the current through them is always the same, however their individual current limits will differ slightly, so as the current increases, one of them will limit its output current first.  When this happens, the output voltage of that regulator will fall until it reaches zero, and the other regulator, which has not limited its current, will drive the output of the regulator which is limiting its current below zero.

Regulators generally do not react well to having their output driven negative, so placing a power diode across each output is used to prevent each output from going negative by more than one diode voltage drop, which is generally safe.  Bipolar tracking regulators, which routinely operate in series by design, have these diodes installed across their outputs for exactly this reason.

[PCB.Wiz]

Anything wrong with this configuration, or anything in particular to be aware of? Or, is there a better way to do this? Or, just other ways I might consider?

That circuit is somewhat doomed, with many issues.

The LT3045 can be thought of as a smart PNP, which means the lower artificial ground moving stage, needs to sink that base current of  2.2mA~25mA, depending on load.
ie the lower stage here needs to be a sinking regulator, or a positive regulator with a minimum sinking load.

The other problem is practical - the LT3045 has a precision current source, allowing a simple single resistor set point, but to make two floating units track, you need 2 variable resistors. Ugh.

You can voltage drive the SET pin, to better manage this, but you are now using the LT3045 as a rather costly 'zero offset' PNP transistor.

It may be simpler, and much cheaper, to start with a low noise reference and low noise, > 26V opamps with added transistor power output stage.

[magic]

This is why I advise creating a complete and unambiguous schematic. On the current one, everybody sees what he wants or expects to see

For the record, I think OP's intent was to create two 13V sources powered by separate windings and connect them in series for 26V. Therefore, ground current of the "upper" regulator returns immediately to the negative output of the "upper" rectifier through GND1 and doesn't need to be sunk by the lower regulator.

But I too got it wrong at first. And the above approach (if true) is still problematic for other reasons.

[matthuszagh]

The LT3045 can be thought of as a smart PNP, which means the lower artificial ground moving stage, needs to sink that base current of  2.2mA~25mA, depending on load.
ie the lower stage here needs to be a sinking regulator, or a positive regulator with a minimum sinking load.

chuckb already pointed this out. I should be able to address that by adding a resistive load to the lower regulator.

The other problem is practical - the LT3045 has a precision current source, allowing a simple single resistor set point, but to make two floating units track, you need 2 variable resistors. Ugh.

Who said anything about needing to making the floating units track? magic did point out that the lower units (as shown in the first diagram) will fight as their grounds and outputs are shorted together but will have slightly different set points. But, to address that I think I can parallel all lower units (procedure for this is described in the datasheet). So, all lower unit set pins have their currents flow through the same set resistor and the outputs are connected together after passing through very small ballast resistors (just short lengths of trace). The upper LT3045s, on the other hand, drive different loads and are not in competition. I'm not sure why you feel 2 variable resistors are needed.

[matthuszagh]

This is why I advise creating a complete and unambiguous schematic. On the current one, everybody sees what he wants or expects to see

Yeah I will provide a complete schematic, but I haven't gotten around to it yet.

[PCB.Wiz]

But, to address that I think I can parallel all lower units (procedure for this is described in the datasheet). So, all lower unit set pins have their currents flow through the same set resistor and the outputs are connected together after passing through very small ballast resistors (just short lengths of trace).

Hmm.. I thought you were striving for isolation here ?

Note you do not need 'all lower units', a common higher virtual ground, needs only sink 25mA per upper LT3045. Lower units do not pass the full load current only the PNP base current.

The upper LT3045s, on the other hand, drive different loads and are not in competition. I'm not sure why you feel 2 variable resistors are needed.

With no complete circuit, I was assuming you wanted this adjustable. That's usually why someone chooses a LT3045 
If the whole thing is fixed 26.00V, no adjustments, of course, yes it can be simpler.

[Zero999]

It sounds expensive to use two LT3045s.

Why not add a series transistor and a zener to form a cascode?

Here's an example using the LM317. The same idea will work with the LT3045.

[magic]

It won't, because LT3045 is not a floating regulator. The absolute maximum input rating is referenced to its GND pin, which is normally grounded.

Perhaps it could work if you also bootstrap GND and related pins to a voltage slightly less than OUT, while leaving Rset connected to actual ground.

[cncjerry]

I am a big fan of the lt3045 using it to clean up supplies and making power supplies up to 2amps.  I've had great help from LT in the past.  I would find someone at LT to help.  But given all the circuits in the datasheet, the one they dont have is a series floating supply.

[Zero999]

It won't, because LT3045 is not a floating regulator. The absolute maximum input rating is referenced to its GND pin, which is normally grounded.

I missed that. 

Perhaps it could work if you also bootstrap GND and related pins to a voltage slightly less than OUT, while leaving Rset connected to actual ground.

Perhaps an ordinary zener diode would do. Another diode would need to be connected between the output and the floating ground, to pull it down, in case of a short on the output.

[PCB.Wiz]

Perhaps it could work if you also bootstrap GND and related pins to a voltage slightly less than OUT, while leaving Rset connected to actual ground.

Perhaps an ordinary zener diode would do. Another diode would need to be connected between the output and the floating ground, to pull it down, in case of a short on the output.

Yes, a PNP current sink with base zener (~5v6 or above) referenced to the output, would give an elevated offset tracking GND point.
Data is unclear on how much GND pin decoupling is needed for stability, so a cap and decent schottky diode for shorted output handling would be added here.
With up to 500mW in the PNP, it would need SOT223 or DPAK ballpark packaging.