considerations for a high power forward converter

[Powermax]

I did some quick napkin math with my old flyback transformer cores to see how power I can push through them, and I think I can push upwards to 1kW through them! With f=100kHz, k=0.7 (which is generous for hand-wound xformer), J=^4.5A/_mm2 Aw=5cm^2 Bmax=0.5T. Anyway, so I started simulating the thing in LTspice. I don't really know what the permeability of the core is, I'll need to measure it, I just went with 470uH for the 24 turn primary as a wild guess for now

The idea is to convert 36V to 50V to 120Vac 60Hz. I chose to go with a forward converter over a resonant converter because I wanted to create sine wave by means of regulating the output voltage to follow the curve, and wasn't sure if a resonant converter could have a large output voltage range. Can the output be adjusted from 170V to close to zero? (A forward converter this should be straightforward by simply changing the duty cycle)

I figure a half bridge on the primary is best, a full bridge uses more expensive FETs and could result in a DC bias on the very low resistance primary which would hurt efficiency and might cause eventual saturation of the core? But I'm afraid the size of the DC blocking capacitors may need to be huge to tolerate possibly up to 60A_rms at 100khz AC. Can a bunch of parallel MLCC capacitors survive this??

Sadly the output is just above the range of schottky diodes so either I need to deal with the power dissipation of ultra-fast PN diodes, or synchronous FETs. Thankfully it looks like synchronous FETs should be easy enough to implement since they are in-phase with the primary side MOSFETs, and it appears the current steadily increases during conduction also.

Although in my simulation I found that the output voltage really collapses if I add anything but 1 for the coupling coefficient. Why is it so? I guess the high power demands on the secondary makes the flux really reluctant to go through it and find another path not through the core Is this going to be an issue that limits me in real life?

[T3sl4co1l]

I did some quick napkin math with my old flyback transformer cores to see how power I can push through them, and I think I can push upwards to 1kW through them! With f=100kHz, k=0.7 (which is generous for hand-wound xformer), J=^4.5A/_mm2 Aw=5cm^2 Bmax=0.5T. Anyway, so I started simulating the thing in LTspice. I don't really know what the permeability of the core is, I'll need to measure it, I just went with 470uH for the 24 turn primary as a wild guess for now

So, 500 mm^2 A_e?  That's quite a large core!  Are you sure?  (Just that, I don't think I've seen any quite that large; mind, it's not much bigger than those I have seen, and I think they're available about that big.  Just double checking.)

J can probably be larger; although, if you're going to stuff the winding area full of copper, a lower J may be wise.

Hmm, that's not much winding area either, is that just a regular ER or whatever shape?  Not a UR like salvaged from a CRT flyback?  The latter is the first thing I think of with such a description, but maybe you aren't meaning it in that way.

Bmax is definitely not that high, if it's any ferrite; 0.3 is safer.  Or probably 0.1-0.2T is more reasonable at that frequency, just because of core losses?

In any case, the power level sounds reasonable, yeah.

The idea is to convert 36V to 50V to 120Vac 60Hz. I chose to go with a forward converter over a resonant converter because I wanted to create sine wave by means of regulating the output voltage to follow the curve, and wasn't sure if a resonant converter could have a large output voltage range. Can the output be adjusted from 170V to close to zero? (A forward converter this should be straightforward by simply changing the duty cycle)

The load resistance being ~constant, helps.  Resonant might not do it with low distortion or noise (because it has to pulse at low duty cycle to reach the low power levels near zero crossing), but if you don't mind that, sure, I would think so.

Of course, anything that violates that assumption, won't behave as nicely.  A typical rectifier load (most SMPS) won't draw any current below peak voltage, so the waveform will just kinda... sit there, until it's forcibly flipped on the next half-cycle.  (A synchronous power stage is handy here, as the output filter capacitor can be discharged, that reactive energy being recycled into the supply -- a passive rectified forward converter will just hover, or at worst, will have to be "down programmed" with a bleeder resistor to hit zero -- losing that reactive energy as heat.)

Better behaved loads, should be fine.  Resistors, inductive loads (to some extent), and supplies with active PFC.

I figure a half bridge on the primary is best, a full bridge uses more expensive FETs and could result in a DC bias on the very low resistance primary which would hurt efficiency and might cause eventual saturation of the core? But I'm afraid the size of the DC blocking capacitors may need to be huge to tolerate possibly up to 60A_rms at 100khz AC. Can a bunch of parallel MLCC capacitors survive this??

Would recommend full bridge, or push-pull, actually.  The switching impedance is higher that way.

Low switching impedance means stray inductance is that much more critical.

By switching impedance, I mean, more or less, the peak switch voltage (just after the instant of turn-off) divided by the peak switch current (just after the instant of turn-on, or just before turn-off, depending on load phase).

PP has the downside that the loop inductance includes leakage between halves of the primary.  A good winding design is required.  This isn't hard, it just has to be designed in.  The bridges have the advantage that it's all on board, between transistors and supply bypass caps, controlled by placement and layout.


Yes, MLCCs are definitely fine here; but you may be embarrassed by the sticker price.  Using the numbers from your example, 60A rms at 100kHz and >36V supply, suppose the voltage drop should be less than 10%, i.e., < 1.8V (half because half bridge -- see how the low switching impedance stacks things against you?).  1.8V / 60A = 30mohm, and 30mohm at 100kHz is 53uF.  53uF is required under bias, and keep in mind type 2 dielectrics lose capacitance under bias.  Probably the current rating dominates anyway, so that you need a good 20 or so parts, rated 3A each; they might be 10uF 50V 1210 X7R, or something like that.

You will probably find a stack of electrolytics is more palatable, and doesn't take up too much space.  Combine that with the ~15A rating needed for full bridge or PP, and you should be sitting pretty nice.

Sadly the output is just above the range of schottky diodes so either I need to deal with the power dissipation of ultrafast PN diodes, or synchronous FETs. Thankfully it looks like synchronous FETs should be easy enough to implement since they are in-phase with the primary side MOSFETs.

Silicon schottky are available up to 200 or 300V, and SiC starting at 600V or so.  200V would be marginal for a 160V bus, I don't think I'd like that.  PN diodes aren't terrible, and I'd consider it.  Synchronous might be a bother, because of the high voltages and isolation (isolated drivers? bootstrapping?).  Would be worthwhile for a more expensive but higher efficiency unit.  SiC is great, but a bit on the pricey and lossy side (drops a few volts, give or take part selection*).

*Most of the voltage drop for SiC is due to internal resistance.  You do stand to save some conduction loss using oversized diodes, or multiple in parallel.  This isn't so much the case for Si diodes, where internal resistance is small, and so less is saved by using bigger or parallel diodes; parallel Si also have current sharing issues.

Although in my simulation I found that the output voltage really collapses if I add anything but 1 for the coupling coefficient. Why is it so? I guess the high power demands on the secondary makes the flux really reluctant to go through it and find another path not through the core  Is this going to be an issue that limits me in real life?

For a forward converter, k > 0.99 or so is a good idea.  Leakage inductance draws reactive energy corresponding to load current.  This costs output voltage, and causes reactive primary current (which is clamped in a bridge, so that's fine here; it's all switching loss in a 1-switch converter though!).

The value of k corresponds to the bandwidth you're coupling with the transformer; a conventional forward converter needs nice square waves, i.e., harmonics many times Fsw, so, wide bandwidth.  Very roughly speaking, say for 100 harmonics (or 10MHz BW out of a 100kHz Fsw), you want k = 1 - (1/100) = 0.99.

For a resonant converter, the leakage is intentional, and some amount may be implemented in the transformer, whether by winding positioning on the bobbin, or by clever arrangement of the core.  Any remainder has to be taken up by a separate inductor.

The resonant converter using lower bandwidth (few harmonics beyond Fsw), means it can use a lower k.  I don't think this particular fact is all that useful, I think it's probably a better idea to describe it in terms of an impedance coupling network (the series and parallel inductances, resonating with capacitance, have an impedance transformation effect by themselves, on top of the transformer ratio).  But this is at least consequential to that, I think.

I guess it's kind of convenient that -- a very simple transformer winding schedule might give too much leakage for a forward converter, but too little for resonant; there's no way to reduce it for the forward converter (other than by rewinding it entirely), but by adding external inductance in series, you can get the perfect value for resonant.

I would recommend sticking with the plain forward converter, for beginners purposes.  Not that resonant is much harder (it's actually pretty straightforward once a few additional facts are understood), just that it's the simpler case.

Tim

[Powermax]

Thanks T3sl4co1l

Yeah the figures I used for the transformer are probably wrong, my math was messy and probably put the wrong numbers in the OP. The core has 4cm x 1.5cm winding area and maybe about 1cm^2 cross-sectional area. old-school flyback transformers have massive winding area! Oh and sorry, I meant 50mT or 500 Gauss!  I will need to measure this, too! That's what I did my math with.

That pretty much echos my understanding of this problem thus far! My simulation likes to draw peaks of like 200A on the primary and I had to use 100uF ideal capacitors since I found 1uF wasn't enough  . 50uF is probably more reasonable though! I didn't consider electrolytic I feared they would just explode under those conditions, or many small ones would take up a hell a lot of space. What about film polypropylene capacitors?

I think this part alone makes the LLC converter more appealing, since the series resonant capacitor can be small and also serves the function to remove DC bias! Needing the harmonics for the forward converter may be a problem, flyback cores are designed to work at 30kHz or so, with a ramp wave. I should note that I do plan to remove the gap as necessary.

I figured that by using a full bridge on the primary, I could increase the turns and decrease the current, that would help a good bit. But I'm concerned about saturation if the pulse period isn't perfectly symmetrical. I'd be also afraid of something causing the control of the fets to latch up and conduct DC  I suppose the other approach is to use a split primary with lo-side FETs either end but then leakage inductance is going to cause massive overshoots too.

I would like this inverter to be able to power inductive loads, too. My intent is eventually to power a 3 phase 1/2HP motor or a computer power supply as an online UPS.

Another benefit of using an active rectification could be that I could double up on the FETs, 2 in series on each end, and maybe control them such the output voltage can be inverted every half cycle! This has proven to be a bit difficult to implement, though. and brings me up to a total of 6 to 8 FETs, yikes! And those FETs have to be fast and double the voltage rating of the output, too, vs using a typical H bridge.

[T3sl4co1l]

That pretty much echos my understanding of this problem thus far! My simulation likes to draw peaks of like 200A on the primary and I had to use 100uF ideal capacitors since I found 1uF wasn't enough  . 50uF is probably more reasonable though! I didn't consider electrolytic I feared they would just explode under those conditions, or many small ones would take up a hell a lot of space. What about film polypropylene capacitors?

You uh, might want to check the startup conditions, and also the control methods (guessing none at all right now?) there.

Here is a good example of how to do it right:
https://www.seventransistorlabs.com/Images/Flashlight2Sch.png
https://www.seventransistorlabs.com/Images/Flashlight2_Schematic.png
This is only a boost circuit, but the same control scheme applies perfectly well to buck, forward, etc.

The key is the control circuit monitors the inductor current.  Anything can happen to the circuit around it: supply and load voltages can be anything, doesn't matter, as long as the inductor current is known, the output power will be limited and the switch current will remain safe.  (Obviously for a boost, the output voltage can't be much below the input, but that is an exception beyond our control.  Everything else, we have total control.)

This happens to be a battery-powered flashlight, so the inductor current can be sensed at the low side, through the battery.  If high side sensing is needed, a current shunt resistor and current-sense amplifier can be used, or a Hall effect sensor (which is also isolated, so can sense secondary side current as well).

The control is fully discrete, so implements everything that you should expect to see in a controller's block diagram.  Top left (IC2A and such) is the ramp oscillator; IC2B is the PWM modulator (compares a PWM setpoint voltage, to the ramp, thus generating PWM output); IC1 is the gate driver (here just a logic buffer, no need for anything fancy); and IC3B is the current error amplifier.  INSP is the current setpoint, and IC3B controls PWM such that the feedback signal IBATT is balanced with INSP.  As INSP voltage goes up, IBATT voltage falls, i.e., inductor current goes up.  C11, C15 and R20 set the rate at which the error amp responds, so that it can be tuned to a stable response over all source and load conditions.

What good is current?  We want voltage!

Well, even if you wanted regulated current, you wouldn't quite have it yet, because this controls input current, not output.  In the boost converter, the output current is less, by the duty cycle (give or take).  So you'd have to multiply by that to know it, which... can be done, that's not too awful to build a circuit to solve.  But we can measure it even easier, and that's simply what ILED is doing.  Since this is powering LEDs, a constant-current output is desirable: this gives a fixed brightness regardless of battery voltage, until the battery is so low that it simply can't provide enough power at all.

For a constant-voltage output, simply wire ILED to a voltage divider on the output: then IC3A adjusts the current setpoint to compensate for changes in load voltage.  The output capacitors (C3, 4, 7, deliver load current in the short term, and after some time constant (set by C10, C13, R19, values chosen again for stability) the converter takes up the load.

This is better than simply wiring IC3A to IC2B (i.e., voltage feedback to PWM), because if the voltage is very low, it will demand 100% PWM (i.e., IC3A's output saturates to +V).  But a boost converter doesn't deliver any voltage until it switches off at least once... it'll just latch on, and burn itself to pieces.   Okay fine, so just limit it so it doesn't go to 100% PWM -- divide it down so it only goes to, say, 70 or 80 or 80% PWM.  Then it keeps switching, so the output will keep rising while "full throttle" is being delivered.  Ah, but how much throttle is it really doing?  How do we know if it's drawing 1A or 10?  We have no clue.  The inductor current is a free variable, it just does whatever it does.

Instead if we have an inner loop controlling inductor current, then it simply goes to whatever we set it to.  If IC3A saturates, it demands, well, whatever 3-4V corresponds to at INSP.  (The resistor divider R16-R22 translates this to a smaller (negative) voltage at IBATT, and R1-R2 convert this voltage to a current.  So, about 7A it seems.  Hmm, that's quite a lot for a 18650, I might've been rather optimistic with these original component values...)

Note that IC3B is allowed to command fully 0% to 100% PWM.  There's nothing wrong with leaving the transistor on for an extended period of time -- it's controlled by inductor current, so the only thing that can happen is the inductor current just isn't rising, and, well, the switch can handle the current it's designed for, so it's not going to smoke or anything.  Anyway, this could only happen if the supply voltage were quite low, which can't happen, so it would only stay on for, eh, a few cycle or something like that.  Long enough to ramp up the current to the setpoint.

(This is also acceptable behavior on a boost or forward converter with current sensing in series with the transistor: while the transistor is on, inductor current is known.  This control method (average current mode control) isn't so suitable with such a connection however (the current is not known while the transistor is off!), something to keep in mind.)


Obviously, for a full-wave forward converter, you'll have alternating switches, which needs a different PWM modulator; and you'll have a secondary side inductor, for which a Hall effect sensor is probably a good idea (isolated current sensing).

A TL494 can be used as a mostly-all-in-one block.  It even has two error amps, though, they're wired in parallel rather than cascade, so I suggest disabling one, and using an external error amp to regulate voltage.

Which, again due to isolation, should be located on the secondary side.  A typical solution, then, would use a TL494 on the primary side, to regulate secondary current (sensed with a Hall effect sensor).  Its setpoint is driven by an optoisolator, which is driven by a TL431 or similar voltage regulator IC.  (The TL431 is typically drawn as an adjustable zener, but it's actually a three-terminal op-amp, with a conspicuously large, yet suspiciously stable, input offset voltage.  Thus, you use it just like an error amp, with compensation RC across it, and that closes your voltage loop.)

This can all be drawn out in the simulator as well, indeed you can put in the above schematics, and use stock models, and you should be able to get it working.  Then you can replace certain parts (say the voltage reference, or current setpoint) with VPULSE sources, and observe the step response for example.

I figured that by using a full bridge on the primary, I could increase the turns and decrease the current, that would help a good bit. But I'm concerned about saturation if the pulse period isn't perfectly symmetrical. I'd be also afraid of something causing the control of the fets to latch up and conduct DC  I suppose the other approach is to use a split primary with lo-side FETs either end but then leakage inductance is going to cause massive overshoots too.

Well, just don't do that...

In the full bridge, you can still use a coupling cap if you like.  It can even be rated lower voltage, since it's not expected to have full supply across it.  (Ah, which makes type 2 ceramics actually rather attractive, as they have maximum capacitance near zero voltage.  Who needs derating?!)  Not so much in push-pull, where matched pulse widths are required, and some mitigation is had by reducing the transformer inductance a bit (by increasing the air gap) and limiting the maximum duty cycle (per switch) to somewhat less than 50%.

(This allows some dead time, during which the transformer's built-up flux imbalance, manifest as unbalanced primary current flow (inductance is the ratio between flux and current, H == V.s / A), is able to speed up or slow down the voltage transition from one transistor to the other (commutation).  Thus, lower primary inductance increases the current flow for a given imbalance, forcing the waveform to be faster or slower on one side or the other.  You don't want to lower the inductance too much, as that increases reactive energy storage: energy drawn from, and returned to, the supply, without performing useful work (output power).)

Tim

[mag_therm]

Hi Powermax
I am doing some similar hobby inverters, but with the push-pull topologies.
Here is some data that may give some reference points you may be able to scale for your design.
 and for your queries about flux density and capacitor temperature.

Using various surplus gapless ferrite cores of E and toroid, I found that at 100% duty cycle,
it is best to limit the peak flux density to be about 0.15 Tesla.
(That is, a swing of about 0.3T )
I am running these at either 35 kHz or 70 kHz.
Here is a trace today of an inverter primary current (top) and the rectified output voltage (bottom)
( first try at image)

The load is a 100 watt Hg Lamp (hot)  . There is no DC filter after the secondary rectifier. The load impedance has a lot of jitter.
The inverter is in current regulation, the op amp regulator has a corner frequency of 15 Hz
The input is 24V DC at 6 Amp. The output is about 100V DC average.
The inverter is running at about 50% duty, so the peak DC capacitor current is about 14 Amp ( per the top trace).
 Peak output voltage is about 200 V ( the lower trace).
The period is 14 usec: 71.5 kHz

The DC filter is an old can type 490 uF 85 V ( 25mm dia by 70 high)  electro in parallel with a 220nf leaded polyester.
The electro  is running at 45 C ( 20 C ambient)
The transformer core diameter is 13mm and the primary has 5 turns (bifilar) of awg 24  for each half of the push-pull.
That gives a peak flux density of 0.127 T at 100% duty. At 50% duty, core is barely warm after 1 hour
Note that the primary current shape is similar to secondary voltage, indication low magnetising current.

[jbb]

Have you done power electronics / higher voltage electronics before? This is absolutely doable, but there will be some safety hazards (electric shock, flying chunks of plastic after explosion etc.)

Look out for any capacitor operated at >50V. They can stay charged for bloody ages and try to kill you. Remember to fit bleed resistors to your circuit.

As it happens, I have played this game before; using a forward converter to make a voltage envelope at 100 Hz (or 120 Hz if you’re in a 60 Hz region) and a slow H bridge to flip every other cycle negative.

My conclusion was: wrong tool for the job :-(

Issues:
* even assuming a nice resistive load, the peak power (which sets inductor saturation etc.) is twice the average power. So you build a bigger converter than necessary
* the pulsating power ripple (100 Hz or 120 Hz depending on region) gets sucked out of your energy source. You might then need a big input filter cap to keep it away from your batteries
* the transformer output goes into a diode bridge and filter inductor. There’s nothing there to claim the voltage so you can get really big voltage spikes when the transformer coupling (k) factor is less than 1. I suggest modelling k=0.99 for a start. Some kind of clamp or snubber is likely required
* the system doesn’t deal well with capacitive loads, as Teslacoil (?) previously mentioned

On resonant converters: they have numerous advantages but aren’t really good for wide wide output voltage range.

If I were to build an inverter, I would go to a two-stage. An isolating DC/DC converter which delivers average power (not peak power), a DC link cap to handle that 100 Hz ripple, and a DC to AC inverter.


For a first attempt, you want something that will work without too much angst. Maybe a forward converter for DC DC, then an IGBT H bridge for DC AC (switching frequency around 5-20kHz) is a good start.

Forward converter could use a silicon Schottky or Silicon Carbide (SiC) Schottky rectifier diode. Don’t try for synchronous rectification yet; it’s conplicated!

Why IGBT for H bridge? Because IGBTs don’t have a body diode. You can get them with separate, good, diodes in the same package. Yes, the conduction losses are higher.

Why not MOSFET for H bridge? Because MOSFETs >200V have horrible body diodes with high reverse recovery charge (Qrr). This drives up switching losses and can even blow up your H bridge (ask me how I know...)

If you then want to do better (smaller, higher efficiency, more bragging rights) you can then get into clever stuff. Options would include stuff like an LLC DCDC stage, maybe a Gallium Nitride (GaN, allows much higher frequency) inverter, maybe a multilevel inverter (can be tricky but a lot of fun for nerds like me).

You will blow some stuff up, so I thoroughly recommend isolated gate drivers, current sensors (I like LEM LAH series) and voltage sensors.

[NiHaoMike]

As it happens, I have played this game before; using a forward converter to make a voltage envelope at 100 Hz (or 120 Hz if you’re in a 60 Hz region) and a slow H bridge to flip every other cycle negative.

My conclusion was: wrong tool for the job :-(

Issues:
* even assuming a nice resistive load, the peak power (which sets inductor saturation etc.) is twice the average power. So you build a bigger converter than necessary
* the pulsating power ripple (100 Hz or 120 Hz depending on region) gets sucked out of your energy source. You might then need a big input filter cap to keep it away from your batteries
* the transformer output goes into a diode bridge and filter inductor. There’s nothing there to claim the voltage so you can get really big voltage spikes when the transformer coupling (k) factor is less than 1. I suggest modelling k=0.99 for a start. Some kind of clamp or snubber is likely required
* the system doesn’t deal well with capacitive loads, as Teslacoil (?) previously mentioned

I have seen one inverter that had no real energy storage between the output H bridge and DC/DC supplying it, it was a modified sine inverter in a UPS designed only for switching power supply loads so handling a heavy capacitive or inductive load wasn't a concern. The DC/DC was effectively soft started on every transition so it had less EMI output than most modified sine inverters that more or less hard switch a capacitor bank to the output. Since it was modified sine rather than pure sine, the peak to average power ratio is lower and overall likely was in fact cheaper than a more conventional design.

[T3sl4co1l]

I once did one in the usual way, DC-DC into H-bridge; but with current limiting into big inductors, so the slew rate is limited (and presumably the EMI isn't terrifyingly bad).



The circuit is... aha, I did make scans of the notes, here you go:



The DN2450 supply limiter isn't really necessary for my purposes, but it does provide simple load dump protection.

Primary end-to-end leakage inductance is critical, as the note suggests.  I did a multisection SPSPS windup, using 24AWG on a EE33 size core and former.  The secondary is 20 turns per layer, 20-40-20 by section, 80 turns total.  The primary is 6 strands wound in a flat multifilar pattern, 4 turns, brought out as 3+3 strands (the start of three strands is the start of the winding; the ends of those join to the start of the remaining three for the CT; and the ends of those strands are the end of the winding).  Both primary sections are wound in this way, and both sections are wired in parallel.

So, the primary is like a ribbon cable, where the wires are alternating between left and right ends of the CT winding.

(For a 24V version: divide the primary into four windings of 4t each.  So each winding uses three alternate strands in each section.  Connect pairs of these in series, section one to section two.  Connect these in series again, to get the CT, start and end of the full 8+8t winding.)

This gives a very low impedance to the primary winding, ensuring low leakage, reducing switching loss and eliminating a snubber.

The secondary inductor I think was a few mH, seems I didn't write down.  Well, most of the component values, really, are left as an exercise to the student.

Note that the 598 is wired as a primary side current regulator (which is not ideal, but it's something).  One error amp regulates primary switch ground-return current.  The other error amp is actually used for OVLO, disabling the output for supply over 20V -- a simple way to ensure the transistors survive up to maximum rated voltage, again for load dump.

The kinda-floating notes at the bottom are illustrations of the OVLO network (since it got rather cramped up there), the aux supply feedback circuit, temperature sensing (an additional LM393 provides this) and layout for the aux feedback (this circuit was hand-cut on copper clad, after the above breadboard circuit was proven to work).

Also, not shown, a UC3843 flyback supply, which, I forget if it's powered by raw 12V, or the DN2450.  The main requirement is providing 12V to the isolated side.




The "modified sine" generator and current limiting.  Gate drivers are IR2101 I think.  This uses the TL494 at low frequency and fixed duty cycle, and uses an open-collector hack to chop the high-side switches to implement current limiting.  Low-side remain on in their respective phases, so that current can be measured in the ground-return path.  It would be adequate to sense just one return current, but the split paths remains symmetrical so I might as well take advantage of the dual LM393.

Current limiting is hysteretic, so more power can be drawn just under the current limit, than above it -- once the 2A limit is hit, current has to dither back to 1A and so on.  The sound of additional switching edges also makes it scream louder (a combination of electrostriction in the filter caps, and magnetostriction in the filter chokes which are plain old #26 powdered iron).

FETs are FS12KM I think.  Some old salvaged part, 250V 5A or something like that.  Modest Qg, doesn't get hot under load.  Notice capacity is only 1 or 2A, i.e. 100-200W or thereabouts.  Nothing impressive; enough to be useful.

Output filter not shown, but it's just a normal-mode filter: for each side, a series choke, a parallel cap to ground, and then a little more cap between outputs.  A second stage LC, smaller values, ensures low EMI (note, haven't measured how much yet).

One of the amusing consequences of the (audibly) noisy filter, is CPU monitoring, much faster than Task Manager displays it.  Namely: when running my laptop and charger off the inverter, the CPU/GPU current draw goes straight through internal converters, through the charger, to the mains; there's very little delay between increased power use, and increased noise from the filter components.

Tim

[jbb]

Thanks for sharing those. Did you have any trouble with voltage spikes in the rectifier after the step-up transformer?

[Powermax]

Here is a good example of how to do it right:
[url]https://www.seventransistorlabs.com/Images/Flashlight2Sch.png[/url]
[url]https://www.seventransistorlabs.com/Images/Flashlight2_Schematic.png[/url]
This is only a boost circuit, but the same control scheme applies perfectly well to buck, forward, etc.

The key is the control circuit monitors the inductor current.  Anything can happen to the circuit around it: supply and load voltages can be anything, doesn't matter, as long as the inductor current is known, the output power will be limited and the switch current will remain safe.  (Obviously for a boost, the output voltage can't be much below the input, but that is an exception beyond our control.  Everything else, we have total control.)

This happens to be a battery-powered flashlight, so the inductor current can be sensed at the low side, through the battery.  If high side sensing is needed, a current shunt resistor and current-sense amplifier can be used, or a Hall effect sensor (which is also isolated, so can sense secondary side current as well).

The control is fully discrete, so implements everything that you should expect to see in a controller's block diagram.  Top left (IC2A and such) is the ramp oscillator; IC2B is the PWM modulator (compares a PWM setpoint voltage, to the ramp, thus generating PWM output); IC1 is the gate driver (here just a logic buffer, no need for anything fancy); and IC3B is the current error amplifier.  INSP is the current setpoint, and IC3B controls PWM such that the feedback signal IBATT is balanced with INSP.  As INSP voltage goes up, IBATT voltage falls, i.e., inductor current goes up.  C11, C15 and R20 set the rate at which the error amp responds, so that it can be tuned to a stable response over all source and load conditions.

What good is current?  We want voltage!

Well, even if you wanted regulated current, you wouldn't quite have it yet, because this controls input current, not output.  In the boost converter, the output current is less, by the duty cycle (give or take).  So you'd have to multiply by that to know it, which... can be done, that's not too awful to build a circuit to solve.  But we can measure it even easier, and that's simply what ILED is doing.  Since this is powering LEDs, a constant-current output is desirable: this gives a fixed brightness regardless of battery voltage, until the battery is so low that it simply can't provide enough power at all.

For a constant-voltage output, simply wire ILED to a voltage divider on the output: then IC3A adjusts the current setpoint to compensate for changes in load voltage.  The output capacitors (C3, 4, 7, deliver load current in the short term, and after some time constant (set by C10, C13, R19, values chosen again for stability) the converter takes up the load.

This is better than simply wiring IC3A to IC2B (i.e., voltage feedback to PWM), because if the voltage is very low, it will demand 100% PWM (i.e., IC3A's output saturates to +V).  But a boost converter doesn't deliver any voltage until it switches off at least once... it'll just latch on, and burn itself to pieces.   Okay fine, so just limit it so it doesn't go to 100% PWM -- divide it down so it only goes to, say, 70 or 80 or 80% PWM.  Then it keeps switching, so the output will keep rising while "full throttle" is being delivered.  Ah, but how much throttle is it really doing?  How do we know if it's drawing 1A or 10?  We have no clue.  The inductor current is a free variable, it just does whatever it does.

Instead if we have an inner loop controlling inductor current, then it simply goes to whatever we set it to.  If IC3A saturates, it demands, well, whatever 3-4V corresponds to at INSP.  (The resistor divider R16-R22 translates this to a smaller (negative) voltage at IBATT, and R1-R2 convert this voltage to a current.  So, about 7A it seems.  Hmm, that's quite a lot for a 18650, I might've been rather optimistic with these original component values...)

Note that IC3B is allowed to command fully 0% to 100% PWM.  There's nothing wrong with leaving the transistor on for an extended period of time -- it's controlled by inductor current, so the only thing that can happen is the inductor current just isn't rising, and, well, the switch can handle the current it's designed for, so it's not going to smoke or anything.  Anyway, this could only happen if the supply voltage were quite low, which can't happen, so it would only stay on for, eh, a few cycle or something like that.  Long enough to ramp up the current to the setpoint.

(This is also acceptable behavior on a boost or forward converter with current sensing in series with the transistor: while the transistor is on, inductor current is known.  This control method (average current mode control) isn't so suitable with such a connection however (the current is not known while the transistor is off!), something to keep in mind.)

Sounds like you are describing a current mode controller lol. I read up on those a little while back when exploring how to power an XHP70.2 LED from a couple 18650's and choosing a boost converter for the job. I'm familar with it and yes it is generally more popular than PWM control with a butt-ton of compensation to slow the loop down enough to be stable.

Obviously, for a full-wave forward converter, you'll have alternating switches, which needs a different PWM modulator; and you'll have a secondary side inductor, for which a Hall effect sensor is probably a good idea (isolated current sensing).

A TL494 can be used as a mostly-all-in-one block.  It even has two error amps, though, they're wired in parallel rather than cascade, so I suggest disabling one, and using an external error amp to regulate voltage.

Which, again due to isolation, should be located on the secondary side.  A typical solution, then, would use a TL494 on the primary side, to regulate secondary current (sensed with a Hall effect sensor).  Its setpoint is driven by an optoisolator, which is driven by a TL431 or similar voltage regulator IC.  (The TL431 is typically drawn as an adjustable zener, but it's actually a three-terminal op-amp, with a conspicuously large, yet suspiciously stable, input offset voltage.  Thus, you use it just like an error amp, with compensation RC across it, and that closes your voltage loop.)

This can all be drawn out in the simulator as well, indeed you can put in the above schematics, and use stock models, and you should be able to get it working.  Then you can replace certain parts (say the voltage reference, or current setpoint) with VPULSE sources, and observe the step response for example.

I actually ended up taking inspiration from the venerable TL494 to implement the alternate switching transistors to drive the forward converter, in the simulation. I like the clever use of a D flip flip to divide the clock by 2 and the use of logic for driving the outputs, the use of an OR gate to combine several error amps into a single PWM output of the highest duty cycle (which results in the lowest duty cycle out on the output, easier to analyze  the demorgan's AND equivalent with a NAND and AND gates following the Dflop.)    and yes, just like the 494, the 431 is another one of those old but venerable voltage references which always come in handy!

Well, just don't do that...

In the full bridge, you can still use a coupling cap if you like.  It can even be rated lower voltage, since it's not expected to have full supply across it.  (Ah, which makes type 2 ceramics actually rather attractive, as they have maximum capacitance near zero voltage.  Who needs derating?!)  Not so much in push-pull, where matched pulse widths are required, and some mitigation is had by reducing the transformer inductance a bit (by increasing the air gap) and limiting the maximum duty cycle (per switch) to somewhat less than 50%.

(This allows some dead time, during which the transformer's built-up flux imbalance, manifest as unbalanced primary current flow (inductance is the ratio between flux and current, H == V.s / A), is able to speed up or slow down the voltage transition from one transistor to the other (commutation).  Thus, lower primary inductance increases the current flow for a given imbalance, forcing the waveform to be faster or slower on one side or the other.  You don't want to lower the inductance too much, as that increases reactive energy storage: energy drawn from, and returned to, the supply, without performing useful work (output power).)

Tim

Oh! Of course, I didn't consider that. Having (ideally) no DC bias on the capacitors does help a lot, class 2 and 3 MLCC's can apparently lose 90% of their capacitance with the full DC rating across them. Although it is still probably a good idea to have them rated for the supply voltage(?)

[Powermax]

Have you done power electronics / higher voltage electronics before? This is absolutely doable, but there will be some safety hazards (electric shock, flying chunks of plastic after explosion etc.)

Look out for any capacitor operated at >50V. They can stay charged for bloody ages and try to kill you. Remember to fit bleed resistors to your circuit.

I have taken a few classes (I am an EE) and built a couple small fly-back transformer drivers and tesla coil stuff. Although i'm no expert in the field of power electronics. I want to change that. 
And yes, big capacitors the size of coke cans with like 10 millifarads 60V ratings are a hella lot of fun when shorted, as are 400V mains filter caps  I think the most fun one I did was with UVA solar car where I demonstrated the danger of not bleading the charge on the motor controller after removing power. The end of the screwdriver was just gone. Ears ringing even with ear protection. 😬

As it happens, I have played this game before; using a forward converter to make a voltage envelope at 100 Hz (or 120 Hz if you’re in a 60 Hz region) and a slow H bridge to flip every other cycle negative.

My conclusion was: wrong tool for the job :-(

Issues:
* even assuming a nice resistive load, the peak power (which sets inductor saturation etc.) is twice the average power. So you build a bigger converter than necessary
* the pulsating power ripple (100 Hz or 120 Hz depending on region) gets sucked out of your energy source. You might then need a big input filter cap to keep it away from your batteries
* the transformer output goes into a diode bridge and filter inductor. There’s nothing there to claim the voltage so you can get really big voltage spikes when the transformer coupling (k) factor is less than 1. I suggest modelling k=0.99 for a start. Some kind of clamp or snubber is likely required
* the system doesn’t deal well with capacitive loads, as Teslacoil (?) previously mentioned

On resonant converters: they have numerous advantages but aren’t really good for wide wide output voltage range.

Damn, that was the exact approach I was gonna take. And I can't seem to get the control loop to work correctly in the simulation.

But I have begun to discover those problems myself as well. I am currently just analyzing continuous peak power, which is closer to 2kW and as soon as I add even the tiniest bit of K less than 1, I immediatly see lots of problems, lots of ringing, but thankfully less problems with huge nanosecond long current spikes from the random ultrafast diodes selected for the output.

If I were to build an inverter, I would go to a two-stage. An isolating DC/DC converter which delivers average power (not peak power), a DC link cap to handle that 100 Hz ripple, and a DC to AC inverter.

For a first attempt, you want something that will work without too much angst. Maybe a forward converter for DC DC, then an IGBT H bridge for DC AC (switching frequency around 5-20kHz) is a good start.

Forward converter could use a silicon Schottky or Silicon Carbide (SiC) Schottky rectifier diode. Don’t try for synchronous rectification yet; it’s conplicated!

Why IGBT for H bridge? Because IGBTs don’t have a body diode. You can get them with separate, good, diodes in the same package. Yes, the conduction losses are higher.

Why not MOSFET for H bridge? Because MOSFETs >200V have horrible body diodes with high reverse recovery charge (Qrr). This drives up switching losses and can even blow up your H bridge (ask me how I know...)

If you then want to do better (smaller, higher efficiency, more bragging rights) you can then get into clever stuff. Options would include stuff like an LLC DCDC stage, maybe a Gallium Nitride (GaN, allows much higher frequency) inverter, maybe a multilevel inverter (can be tricky but a lot of fun for nerds like me).

You will blow some stuff up, so I thoroughly recommend isolated gate drivers, current sensors (I like LEM LAH series) and voltage sensors.

For by transistors I got some HY1920P's from LCSC. They are meant for my half bridge tesla coil (which is a sort of LLC converter I guess  ) although for low power test they should be good.  synchronous rectification is hard! and expensive too, since you need more FETs and more isolated gate drivers. I have seen some really nice implementations where the gate driver is powered via a large series of photodiode acting as a tiny integrated solar cell. Those are awesome! Not cheap though.

For now I just need to characterize my core now that I'm back home again.