Stabilizing TL494-based converter

[leblanc]

I'm trying to build a very low cost offline UPS and I'm testing out some architectures. The UPS will need to be pure sine wave. I'm hoping I can figure out an architecture that doesn't use a DSP/MCU/etc. The UPS is only a hundred watts or so.

Right now I'm focusing on the inverter portion.

What I'm testing out is a using a phase shift oscillator (e.g. bubba) to generate a 60Hz sinusoidal reference signal, which I can then pass to a switching converter (TL494) as reference and create a an AC supply, which then goes to a 60Hz filter and then transformer. I want to have my 48V battery middle, and then one side of my transformer primary is going to be connected to my AC supply output, and the other side is going to be connected to battery middle (i.e. 24V), so that my transformer primary sees a 48V peak-to-peak. Then my secondary is just AC line and neutral. If I can connect my my battery middle to AC neutral, I might consider a centre-tapped auto-transformer.

I plan on taking the feedback before the 60Hz filter, instead of after the transformer, because I expect the filter+transformer is going to give me trouble.

I'm looking at the TL494 because it's cheap, its error amp inputs can go all the way up to Vin-2V, it exposes both EA inputs, and offers a pretty flexible output (uncommitted transistor output). If there are any (cheap) alternatives for the TL494, that may also be of interest. Maybe the SG3525.

I had issues stabilizing it in LTSpice, so I simplified it to just fixed Vout=24V and Vin=48V, to see if I could even get that to stabilize, but I can't seem to get it to stabilize. I like to follow International Rectifier's AN-1162 (https://www.infineon.com/dgdl/an-1162.pdf?fileId=5546d462533600a40153559a8e17111a) for compensator design.

[T3sl4co1l]

For starters, use a voltage divider and the internal VREF; your hint is that, as shown, the error amp cannot fully control itself, say in response to a step change; its output voltage range is a bit under 5V.  (This isn't perfectly obvious from the datasheet values; the test voltages for output sink/source (FEEDBACK) can be taken as a recommended range.  The appnote, which takes apart the chip showing internal implementation, shows this in full detail.)  It's interesting that the input range is -0.3 ... VCC-2 so up to 39V at ratings, but I don't think it's useful anywhere up there, probably more just a convenience so it can be a bit above VREF, saving a divider there.

You're also overvolting the LTC7060, it's rated only 6V input; no biggie, 12V supply for everything should be fine, plus a bit of a divider on the 494 output.  Apparently the sim doesn't mind, either because they didn't model internal ESD diodes, or the emitter outputs drive into the short circuit and it's just whatever (the sim doesn't include temp rise or magic smoke, sadly; not that it would necessarily run for long enough to notice anyway.)

Have you considered something newer, like a LM25119 or the like?  (Or the single version, which, I forget what number that is offhand..)  494 is very old, and it shows (all the examples are voltage mode, yeech!)

Also, those transistors are WAY too big for a "hundred watts" converter; they'd be fine for a few kW.  Plain old IRFZ46Ns are more than overkill at this level, though those exact parts are more suitable for 12/24V operation than 48V.  Well, there are zillions of types in this range, easy to find something suitable.

Even with those transistors operated closer to ratings (if you're planning something bigger), you might want smaller transistors anyway, just because at currents that high, the lead inductance is significant, both for turn-off time as well as commutation.  And with that being only a few nH (for the D2PAK version, or ~7nH for TO-220 with shortest leads), it's just about impossible to snub*.  And so, rather than trying to deal with something that can't be dealt with, just use multiple transistors in parallel; or even better, multiple converters in parallel, with phase shift between them to save on input/output ripple (bypass caps) as well.

*You can't make any connections "inside" of those ~nH, so the excess peak voltage or ringing due to that inductance, will be unclamped to that extent.  It doesn't sound like much, but say 20A turning off in 10ns, implies a voltage drop of 14V over 7nH (V = L dI/dt) -- this is comparable to the gate drive voltage, thus acting to oppose it, preventing any faster turn-off.  Any R+C or diode you add across the transistor, will have several nH of its own, at best acting in parallel with device inductance, but not dominating over it.  Snubbers can really only help against circuit inductance.  But circuit inductance is the easiest to minimize, say by using a 4-layer board with the three connections on planes (+V, OUT, GND), the supplies being stacked with nearby bypass caps.


For reference, some years ago I made a little "modified sine" of comparable power -- easier than doing a full adjustable sine output, and yeh, a bit noisy, the hard-switching output creates an obvious buzz (partly from the output filter chokes, partly from the caps -- the effect of magnetostriction and electrostriction respectively).  Most things behave with such a source, so it's alright.
Prototyped, filters on breadboard: https://www.seventransistorlabs.com/Images/InverterProto1.jpg
DC-DC converter (the important part; there's also a UC3843 for isolated control power): https://www.seventransistorlabs.com/Images/InverterProto2.jpg
Output stage: https://www.seventransistorlabs.com/Images/InverterProto3.jpg
oh, those aren't marked, that's IR2101 bootstrap drivers and FS12KM MOSFETs (salvaged; IRF644 is basically equivalent).  Output filter is an LC with, I think around 100uH series, 0.47uF shunt, and some R+C to dampen it.

Note the output is current limited; the gates are, well, gated by a hysteretic comparator each, to implement a crude current limit.  So the reactive load of the filter is no problem, nor is any kind of capacitive or harmonic load such as an SMPS.

Tim