Mixed linear and switch-mode topology in a lab power supply?

[technix]

The MC34063 is not really suited to drive MOSFETs. Adding a gate driver makes things more complicated. It is also limited to relatively low frequency and thus needs large inductors and capacitors. So the cheap chip may not save much.

For the UC384x (no need for the 14 Pin version) is mainly the question if you can live with a negative side buck converter (negative buck -works well with N-MOSFET) - if yes this is a viable low cost solution. The difference to the 34063 is that it include a gate driver for an N_Mosfet at the GND side.

A positive side buck converter either needs a P-MOSFET or a special boots-trapped high side driver for an N-MOSFET and still has limits. There are reasons for more fancy and more expensive controllers - so a cheap controller will have some downside like more noise, less efficiency.

For learning it could be OK to use older more standard chips, but this also works with lower power - so the first version really should be lower power. Designing an SMPS is not that easy - you can well burn a few MOSFETs / diodes / caps. For a 10 A version this could even blow the board.

The reason I started off at 10A is because of chips like LM2596 rendering the point of rolling a buck at the current of a few amps moot. I have previously designed a few boosts using MCU (a self-powering boost) or MC34063 (the MOSFET gate drive for my M102v5 board) and those worked very well. Also I have some previous experience of IRF4905 MOSFET in LDO application and that worked well too.

I am looking at alternatives on TI's product page and this caught my eyes: TL5001. Its datasheet have shown using it as a high-side buck driving a PMOS, and I believe my existing way of converting a PMOS drive with OC output into synchronized rectifier (the diode and pull up for gate drive) should work with my 30V input. A con point for that chip: it is more expensive than ATtiny85 from some sources.

I am editing the circuit now, replacing the MC34063 with that TL5001 and add a relay switching for the input windings. My transformer have dual 12V windings and so far my design puts them in series. Now I am adding a relay so if the output voltage of the SMPS stage goes lower than 10V the two windings are paralleled instead of being winded in series.

Also, do not take resistor values for granted, as I will recalculate them after the schematics are done.

[T3sl4co1l]

Well, toss in a current transformer and an additional error amp, and you can do as well as anyone would ever hope: a proper average-current-mode buck controller.  But the reference circuit they give?  Yeah, stick with 34063...

Tim

[technix]

Well, toss in a current transformer and an additional error amp, and you can do as well as anyone would ever hope: a proper average-current-mode buck controller.  But the reference circuit they give?  Yeah, stick with 34063...

Tim

It seem to me that you have an unquenchale hatred toward MC34063.

[T3sl4co1l]

Eh?

[technix]

This is it, no more argument on which chip to use.

Main rail uses LM3477 (this puppy have built-in NMOS bootstrap circuitry) with IRF540 as the external switch. I need good efficiency but don't need good ripple suppression on the primary SMPS as the linear section immediately follows, with 2.5V headroom.

Auxiliaries are the domain of LM2596 and MC34063. I don't need good efficiency on those rails as they only drive internal circuitry (5V logic from LM2596-5.0, 12V op amps, relays and cooling fan from LM2596-12, 3.3V BTLE module from MC34063 and -12V op amps from MC34063). The 5V rail is the secondary output but LM2596-5.0 on its own is already good enough.

[T3sl4co1l]

Main rail uses LM3477 (this puppy have built-in NMOS bootstrap circuitry)

Ooh, this one looks good!

Tim

[technix]

Main rail uses LM3477 (this puppy have built-in NMOS bootstrap circuitry)

Ooh, this one looks good!

Tim

Reworked the SMPS section again. Compensation computation got complicated since I am using a tracking regulator instead of a normal resistive divider for feedback. The 4.5uH inductor is self wound.

The two AC12V inputs can be switched between series and parallel using a relay. SMPS output of 15V is the threshold. Some hysteresis is added.

[T3sl4co1l]

Probably the filter cap can be a whole lot smaller too.

Hrm, also that inductor will get preposterously hot. You need ferrite at that frequency and ripple%.  mu=75 powdered irons are crappy even at 50kHz.  They're cheap and that's it...

Also, is that feedback actually going to work?  It seems to have ludicrous voltage gain off the output node, which isn't a good sign.

Tim

[technix]

Probably the filter cap can be a whole lot smaller too.

Hrm, also that inductor will get preposterously hot. You need ferrite at that frequency and ripple%.  mu=75 powdered irons are crappy even at 50kHz.  They're cheap and that's it...

Also, is that feedback actually going to work?  It seems to have ludicrous voltage gain off the output node, which isn't a good sign.

Tim

How much output filtering do you suggest? 470uF?

From the datasheet of my ferrite it works up to 1MHz. I do have a few mu=4 ones but those 50MHz ones toroidal requires 80-100 turns. Or maybe I can reuse a few of those "3R6"s I took out of a dead Core 2 Duo motherboard?

The feedback is intended to make the output of the SMPS track the voltage of the output of the linear voltage with a constant offset of 2.5V. So here I need a differential voltage gain of 0.5 or so.

[Kleinstein]

There might need to be an additional capacitor right before the regulator. There is quite some ripple at the input side and no real need to pass this through the relays. At least have a look at the ripple current ratings for the caps. For EMI reasons one might also need some filtering on the input side.

If space/ weight is not at a premium, a slightly larger (both value and mechanical size) inductor is often working better.

With a switched mode regulator, is there a real need to switch the transformer ?
The relays and the double rectifier also give some extra loss. So I am not that sure the slightly better SMPS performance at lower voltage is worth it. Just using a good inductor, diode and FET might be easier.

The current feedback circuit looks like having gain.  Just the pot at the collector side tends to be rather nonlinear and temperature sensitive.  To get rid of the gain and make it more temperature stable one usually has a resistor at the emitter of the transistor too. The two resistors at emitter and collector set the gain and voltage difference for the linear regulator. Usually a gain of something like 0.5 should be about right. Depending on the SMPS chip, you may need a capacitive coupled direct feedback too, especially if an LC filter before the linear stage is used.

[technix]

There might need to be an additional capacitor right before the regulator. There is quite some ripple at the input side and no real need to pass this through the relays. At least have a look at the ripple current ratings for the caps. For EMI reasons one might also need some filtering on the input side.

The input comes from a mains rated transformer. How can I do EMI filtering after such an transformer?

If I used mains-side switching (something the UC3842 mentioned above really masters at) I would definitely include EMI filtering (and maybe active PFC too)

If space/ weight is not at a premium, a slightly larger (both value and mechanical size) inductor is often working better.

It is indeed not at premium, but the SMPS controller chip have a tight tolerance for inductance (I can go from 3.7uH to 6.0uH, so somewhere in the middle I got 4.7uH)

With a switched mode regulator, is there a real need to switch the transformer ?
The relays and the double rectifier also give some extra loss. So I am not that sure the slightly better SMPS performance at lower voltage is worth it. Just using a good inductor, diode and FET might be easier.

It is not the volts, it is the current. This is not a 2-phase SMPS so I have to switch the windings to get more current if needed. The double rectifier may indeed be a problem but fusing can be problematic (as each winding have its own fuse)

The current feedback circuit looks like having gain.  Just the pot at the collector side tends to be rather nonlinear and temperature sensitive.  To get rid of the gain and make it more temperature stable one usually has a resistor at the emitter of the transistor too. The two resistors at emitter and collector set the gain and voltage difference for the linear regulator. Usually a gain of something like 0.5 should be about right. Depending on the SMPS chip, you may need a capacitive coupled direct feedback too, especially if an LC filter before the linear stage is used.

I did think about emitter-side resistors but I need the ability to adjust the tracking anywhere between 1.3V and 3V no matter what voltage the output is. When I go for 1.3V tracking the emitter resistor really will give me headroom issues when the output is approaching zero.

[T3sl4co1l]

How much output filtering do you suggest? 470uF?

Depends on the compensation components, but those can change too. Probably a minimum of 22uF would be a good starting point: which has the advantage of being very fast, so you can have quick response and lower 60Hz ripple.

From the datasheet of my ferrite it works up to 1MHz.

Oh, is it actually a ferrite?  What's the datasheet?

I do have a few mu=4 ones but those 50MHz ones toroidal requires 80-100 turns. Or maybe I can reuse a few of those "3R6"s I took out of a dead Core 2 Duo motherboard?

Those might be good; in that service, they would've been at a similar frequency (or higher), dropping 12V to ~1V at piles-of-amps.  You probably want a bit more inductance though ("3R6" = 3.6uH).

The feedback is intended to make the output of the SMPS track the voltage of the output of the linear voltage with a constant offset of 2.5V. So here I need a differential voltage gain of 0.5 or so.

Intended yes, but beware that an intention-aware circuit has yet to be created.    A transistor wired like that will have a voltage gain of about 20, increasing your loop gain considerably.  If nothing else, your compensation RC will have to be very different from usual.

Better to have the SMPS programmed from the same reference as the analog output, merely with an offset added in.

Programming is very simple as you only need a resistor from REF to FB.  Think of FB as an inverting input, with a fixed precision offset voltage; or as the voltage sense divider from two voltages (jointly from OUT to FB to GND, and REF to FB to GND).

The downside is, you don't get the SMPS output tracking the analog output under current limit conditions.

Tim

[technix]

How much output filtering do you suggest? 470uF?

Depends on the compensation components, but those can change too. Probably a minimum of 22uF would be a good starting point: which has the advantage of being very fast, so you can have quick response and lower 60Hz ripple.

From the datasheet of my ferrite it works up to 1MHz.

Oh, is it actually a ferrite?  What's the datasheet?

Toroid core. Yellow/white -26 iron power core.

I do have a few mu=4 ones but those 50MHz ones toroidal requires 80-100 turns. Or maybe I can reuse a few of those "3R6"s I took out of a dead Core 2 Duo motherboard?

Those might be good; in that service, they would've been at a similar frequency (or higher), dropping 12V to ~1V at piles-of-amps.  You probably want a bit more inductance though ("3R6" = 3.6uH).

Rechecked, those says R36 so bummer.

The feedback is intended to make the output of the SMPS track the voltage of the output of the linear voltage with a constant offset of 2.5V. So here I need a differential voltage gain of 0.5 or so.

Intended yes, but beware that an intention-aware circuit has yet to be created.    A transistor wired like that will have a voltage gain of about 20, increasing your loop gain considerably.  If nothing else, your compensation RC will have to be very different from usual.

Better to have the SMPS programmed from the same reference as the analog output, merely with an offset added in.

Programming is very simple as you only need a resistor from REF to FB.  Think of FB as an inverting input, with a fixed precision offset voltage; or as the voltage sense divider from two voltages (jointly from OUT to FB to GND, and REF to FB to GND).

The downside is, you don't get the SMPS output tracking the analog output under current limit conditions.

Tim

How do I track it without introducing gain then? I think I will have to compromise and allow the SMPS to have bad tracking when the output voltage drops lower.

[Kleinstein]

for the tracking Feedback, the single transistor should work reasonably well: with equal resistors at emitter an collector you get a gain of about 1 and an difference voltage of about 0.6 V + 1.3 V = 1.9 V.  With a smaller resistor at the FB pin one would get  2.5-3 V at a gain of about 0.5. An additional RC combination can provide direct AC feedback from the voltage if needed. So no real problem here. Only a drop of less than 1.9 V might cause trouble, but I see no need or use for less than 2 V as this also gives little room for ripple and fast load changes.

The first step would be testing the SMPS alone, thus with normal Feedback.

Iron powder is not a good choice at 500 kHz - So you might need to think about a different core.

[T3sl4co1l]

Toroid core. Yellow/white -26 iron power core.

Ok yeah, mix #26 pretty much useless for > 10% ripple at >50kHz. It makes a bad resistor and an even worse inductor at >100s kHz...

Tim

[technix]

for the tracking Feedback, the single transistor should work reasonably well: with equal resistors at emitter an collector you get a gain of about 1 and an difference voltage of about 0.6 V + 1.3 V = 1.9 V.  With a smaller resistor at the FB pin one would get  2.5-3 V at a gain of about 0.5. An additional RC combination can provide direct AC feedback from the voltage if needed. So no real problem here. Only a drop of less than 1.9 V might cause trouble, but I see no need or use for less than 2 V as this also gives little room for ripple and fast load changes.

The first step would be testing the SMPS alone, thus with normal Feedback.

Iron powder is not a good choice at 500 kHz - So you might need to think about a different core.

Different core material is hard to come by, and at 10A inductor heat is hard to deal with. I am really in the middle of reworking the circuit again but either go back to MC34063 (which have a lower switching frequency hence lower requirement at inductor) or go up to LM2642, a 2-phase synchronized rectifying controller so I can split the current into 2 inductors and 2 sets of switching elements (also lower frequency at 300kHz).

Or maybe it would be better for me to scratch the entire idea of mains transformer - nonisolated SMPS - linear regulator, and use mains-side switching (UC3842 controller + IRG4BC20KD IGBT as switching element) + flyback regulator instead. Maybe in that way I can tackle both the power delivery issue and magnetics issues together.

[Kleinstein]

Having a classical transformer and switched mode regulator after that, still has the advantage of lower common mode noise. The common mode coupling if a flyback converter is hard to filter, and the linear stage does not help at all with this.
A flyback converter is also more sensitive to the magnetics and more difficult / dangerous to test/measure. It is somewhat attractive as is might be cheaper than iron transformer.

Using a dual phase regulator might be interesting as this reduces ripple current to the input side capacitors, but it also makes the layout more complicated. With the still rather high output voltage (full power only at more than about 12 V)  I don't think using synchronous rectification makes much sense. The lower frequency makes some sense, as this simplifies the layout and choice of parts, though it requires a larger inductor - 300 kHz are still not that low.

[technix]

Having a classical transformer and switched mode regulator after that, still has the advantage of lower common mode noise. The common mode coupling if a flyback converter is hard to filter, and the linear stage does not help at all with this.
A flyback converter is also more sensitive to the magnetics and more difficult / dangerous to test/measure. It is somewhat attractive as is might be cheaper than iron transformer.

Using a dual phase regulator might be interesting as this reduces ripple current to the input side capacitors, but it also makes the layout more complicated. With the still rather high output voltage (full power only at more than about 12 V)  I don't think using synchronous rectification makes much sense. The lower frequency makes some sense, as this simplifies the layout and choice of parts, though it requires a larger inductor - 300 kHz are still not that low.

LM2642 requires synchronized rectification anyway, or I have to switch to yet another controller chip. The good thing about '2642 is that it have 2 identical controller units operating at 180 degree phase difference inside, which can be configured into a 2-phase buck, so I can get both 30V/5A (with phase 2 turned off) and 12V/10A (both phases on) with a single 24V/5A winding on the transformer.

If I step frequency down even more my next option is sadly ATtiny25 with SMPS firmware, using its 250kHz PWM signal. Then it is down to ye olde MC34063 with external FET...

[Kleinstein]

The advantage of a 2 phase controller is reducing the ripple current and possibly faster response at the same clock frequency. There is also an advantage with very high currents, when parallel FETs would be needed anyway. The input current seen by the transformer is about the same for a 1 or 2 phase controller. So it's more about less requirements for the input side capacitors (and if present filter) and less noise. 2 phases are especially interesting when the voltage is reduces to about half, but still work with a variable voltage.

It would be rather unusual to have 2 phases using different input voltages. The 2 phase controller actually works very good from 30 V to 10 V.  So I would still use both phases powered from about 30 V. Using a lower input voltage might be an advantage for less than 5-8 V output, but this is already at not so high power - so not a problem for the transformer, but only more demanding for the filter.  Still I think classical 2 phase operation would be a larger advantage than using 2 input voltages but only single phase operation.

With an 5 A (AC) rated transformer you would not get 5 A at the full voltage anyway (this would require PFC at least), but more like a 3 A maximum. With the switched mode this limit will gradually go up at lower voltage - giving an about constant power and than level off at a maximum current set by the SMPS. So maybe 3 A at 30 V, 5 A at 17 V 8 A at 10 V and 10 A below 8 V.

One could consider turning of one phase at light loads (e.g. less than 0.1 - 0.5 A) - some controllers include this feature, but it's not absolutely needed. With a linear regulator to follow one could even consider turning the switching regulator fully off (e.g. FET turned on, so full output) at very light loads to get a low noise version for low currents. However this does not work with bootstrapped N-MOSFETs for a positive buck converter - it would need an extra P-MOSFET to bypass the regulator.

[technix]

The advantage of a 2 phase controller is reducing the ripple current and possibly faster response at the same clock frequency. There is also an advantage with very high currents, when parallel FETs would be needed anyway. The input current seen by the transformer is about the same for a 1 or 2 phase controller. So it's more about less requirements for the input side capacitors (and if present filter) and less noise. 2 phases are especially interesting when the voltage is reduces to about half, but still work with a variable voltage.

It would be rather unusual to have 2 phases using different input voltages. The 2 phase controller actually works very good from 30 V to 10 V.  So I would still use both phases powered from about 30 V. Using a lower input voltage might be an advantage for less than 5-8 V output, but this is already at not so high power - so not a problem for the transformer, but only more demanding for the filter.  Still I think classical 2 phase operation would be a larger advantage than using 2 input voltages but only single phase operation.

With an 5 A (AC) rated transformer you would not get 5 A at the full voltage anyway (this would require PFC at least), but more like a 3 A maximum. With the switched mode this limit will gradually go up at lower voltage - giving an about constant power and than level off at a maximum current set by the SMPS. So maybe 3 A at 30 V, 5 A at 17 V 8 A at 10 V and 10 A below 8 V.

One could consider turning of one phase at light loads (e.g. less than 0.1 - 0.5 A) - some controllers include this feature, but it's not absolutely needed. With a linear regulator to follow one could even consider turning the switching regulator fully off (e.g. FET turned on, so full output) at very light loads to get a low noise version for low currents. However this does not work with bootstrapped N-MOSFETs for a positive buck converter - it would need an extra P-MOSFET to bypass the regulator.

Don't worry about the PFC issue - I have a power cap of 50W along with a current cap of 10A.

I will research into that LM2642-based 2-phase buck pre-regulator (and leave LM3477 for some other projects in the future.) LM2642's two phases are capable of individual operation including running at different output voltages and being switched off, which is a feature I am very willing to exploit. So I am trying out this:

* LM2642 controller (300kHz 2-phase, bootstrapped high side N-channel switching, synchronous rectifier)
* Power FETs: 4x IRF7470 (at less than 0.5W dissipation I am trusting this SO-8 MOSFET without heatsinks)
* SMPS stage can be bypassed using a single IRF7404 (when the dissipation on the linear pass transistor is lower than 10W, and the transistor's temperature is lower than 50 degrees Celsius: that is, since the input voltage is 34V rectified, 30V 1A gets straight linear regulator, but 5V 0.5A will use the SMPS)
* Each of the two phases have a maximum current of 5A.
* When the output current is lower than 1A, phase 2 (the one without the power good output) of the LM2642 controller will be turned off.
* Microcontroller: STM32 + ADuM3160? Or the good old ATmega328P? Bluetooth go external anyway.
* Control interfaces: Keypad & screen (good ol' HD44780 20x4 LCD,) Bluetooth Low Energy + iPhone app, for the STM32 I can add native USB too.

[Kleinstein]

The LM2642 is specified for 30/32 V max supply. So it can easily blow with a 34 V input. Even the LM3477 with a maximum of 36 V is rather close.

The gate drivers of the LM2642 are not very strong to drive large FETs - so the losses might be to high to get 5 A from each channel, especially with FETs that are not effectively cooled.

The IRF7404 is only a 20 V FET, but at least 40 V, better more would be needed.

For the controlling µC, USB integrated in the µC does not help very much, as the µC is likely connected to the output. As isolation on USB level is rather difficult, I would prefer UART level for isolation and an extra UART-USB chip if needed. As the cheap LCDs often want 5 V signals, a 8 Bit µC running on 5 V might be easier. A AVRmega88 might not have enough pins - so you might need a mega164 or similar.

[technix]

The LM2642 is specified for 30/32 V max supply. So it can easily blow with a 34 V input. Even the LM3477 with a maximum of 36 V is rather close.

Since LM3477 is a no-go due to inductor nonavailability and obviously MC34063 is being hated on passionately, I have to dial down the input transformer and the maximum output voltage. AC20V/5A input 25V/10A/50W max output, or AC18V/5A input 20V/10A/50W max output.

The gate drivers of the LM2642 are not very strong to drive large FETs - so the losses might be to high to get 5 A from each channel, especially with FETs that are not effectively cooled.

IRF7470 are fairly small, logic level FETs. LM2642's requirement of the use of logic-level FETs is kind of a barrier here, and I don't really want to put IRL540 out here. Also how do I cool SO-8 FETs?

The IRF7404 is only a 20 V FET, but at least 40 V, better more would be needed.

Then I'll just use a relay to switch it. Ha even less loss.

For the controlling µC, USB integrated in the µC does not help very much, as the µC is likely connected to the output. As isolation on USB level is rather difficult, I would prefer UART level for isolation and an extra UART-USB chip if needed. As the cheap LCDs often want 5 V signals, a 8 Bit µC running on 5 V might be easier. A AVRmega88 might not have enough pins - so you might need a mega164 or similar.

Cheap LCD may want 5V but it is fairly simple to isolate that - PCF8574. Those I2C GPIO extenders can work with a 3.3V I2C bus while itself being powered by 5V. Or maybe I can just run it headless and rely entirely on Bluetooth for control.

I am going to try the circuit with ATmega328P first, and if I find that lacking I am going to switch over to ATmega644 or maybe ATmega128. (Or maybe I should give PIC16F1939 or PIC24FV32KA304 a shot too?)

[Kleinstein]

There are P_channel MOSFETs for more than 20 V. A PNP might work as well, as the mode without the switched mode would be useful mainly for low currents. Some drop is not a problem as the voltage is too high in most cases anyway. So no reason to use a relay.  A relay is bulky and slow.

For the µC I see not need for large processing power and lots of memory. With some extra controls you may just need more than the about 20 IO pins available with the Atmel mega??8 series.

The LM2642 curves go up to 5 A per channel with a 20 V FET - so it might be difficult to reach that level with a 40 V FET. It might work - but I would not take is for granted. With only a 50 W power limit, there may not be that much use in planing for up to 10 A anyway. You lose something like 2-3 V for the linear regulator, maybe 0.5 V for the switched mode part - so 10 A would be available at very low voltages (e.g. less than 1-2 V) only.

[technix]

There are P_channel MOSFETs for more than 20 V. A PNP might work as well, as the mode without the switched mode would be useful mainly for low currents. Some drop is not a problem as the voltage is too high in most cases anyway. So no reason to use a relay.  A relay is bulky and slow.

IRF7240. This ought to make it. Another SO-8 business though.

For the µC I see not need for large processing power and lots of memory. With some extra controls you may just need more than the about 20 IO pins available with the Atmel mega??8 series.

Since those pin- and code-compatible parts cost about the same I went for the larger one so I can have some extra room for calibration data and software features.

The LM2642 curves go up to 5 A per channel with a 20 V FET - so it might be difficult to reach that level with a 40 V FET. It might work - but I would not take is for granted. With only a 50 W power limit, there may not be that much use in planing for up to 10 A anyway. You lose something like 2-3 V for the linear regulator, maybe 0.5 V for the switched mode part - so 10 A would be available at very low voltages (e.g. less than 1-2 V) only.

The 50W power limit is on the output power. So the 10A current is only available when the output voltage is less than or equal to 5V. At maximum output voltage (20/25V) the current is limited to 2.5/2.0A.

[technix]

</thread>

I am splitting this project in two: M103v1-3 with same structure but a reduced current ratings (3A max so LM2596-ADJ can work, the 30V voltage cap and 50W output power cap still stands) and M103v2-10 with same ratings but uses mains-side switching (two UC3842 forward converters, one outputting auxiliary rails and one outputting primary rail, with active PFC.) This will result in simpler power inductor selection for me (as I am budget and supplier restricted) and easier design.