Surprisingly simple push pull buffer

Did you build that or simulate it? Any waveforms?

Took a while for my eyes to go around it, you know 'but that's gonna zener, no, over there's clamped it... So those nicely controlled currents and voltage clamps make it how quick?

Grant.

Yes, Both. No waveforms offhand. Only quirky stuff is from the FETs' miller slope, of course.

It makes

When I started using MOSFETs last year, took me ages to work out why the gate voltage looks like it does, eventually tripped over the useful app notes. I'm a little better now ;)

No deadtime control, it's only good for full on/off complementary things. Like where you want a clock signal and its complement, but don't want the skew of a ~30ns propagation delay on one of the lines. Assuming the skew is that small anyway (I didn't measure it).

The cap sucks base charge out, minimizing storage time and sharpening the risetime. Without it, the rise time on that side is a bit skewed, since the common base gets full load current turning it off (~50mA), while the common emitter lazily turns off through the B-E resistor (~3mA).

If you used this for, say, driving complementary MOSFETs, skew doesn't matter, you'll just get corresponding dead time. You will get shoot-through when driving the gates together, when +V > 2Vgs(th), since both transistors are turned on in the middle range (the "class A" range for CMOS gates, except short circuit current might be 10-20A for the complementary SO-8 types).

With a little bootstrap, you can drive an NMOS H-bridge, but in this case, you will get shoot through equal to skew. RCD networks could be arranged to skew the rising or falling edges before the output transistors to fix this.

Current-mode commutation is easy enough to arrange, you can use a few turns on a teensy powdered iron toroid and put it in series with the H bridge supply. A diode and resistor (making an RLD snubber) is required to clamp the shoot-through-turnoff peak.

Tim

fer.png

This circuit has the same sourcing drive capability, slightly improved sinking, is faster, and simpler...

Ah, but it's not complementary output. And it draws gobs of power in the output-low state. Single ended outputs are still valuable, could you rearrange it to draw only bias current in both states?

MML* with FETs is a little different than BJTs... You could still do the gate/source input drive thing to generate the complements, but then you should have a clamp, so Vgs doesn't go substantially past Vgs(th), which would increase input-falling delay time. I'm driving this with an LM393, so the pullup would float the gates to +12V, wasting an extra 9V or so. A zener clamp could be used (or since zeners are capacitive, a diode+zener clamp would be preferred), so it ends up taking about as many components either way.

Yeah, which suggests it's miller limited. 2N3904s tend to make ~50ns edges when you drive them simple like this.

Tim

push-pull.http://myweb.msoe.edu/williamstm/Images/Inverting_Buffer.png

There is this old classic, still a cool circuit.

ftp://jjlarkin.lmi.net/TotemDriver.jpg

But really, fet gate driver ICs are plentiful and cheap and really good. It's tragic, but sooner or later most good circuits get killed by ICs, and we have to move up the abstraction stack. But playing with basic parts is still really good practice.

John

who used to make my own opamps out of transistors, and once did an all-discrete SAR ADC.

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

I use that from time to time:

The saturation towards +V is terrible (or to GND, for the PNP boost), which is a shame. It really stinks at driving transformers, but at least the NPN-to-ground inside the TL598 saturates with reasonable hFE both forward and reverse, so you can drive a transformer, with dead time, and actually have ~0V during that dead time.

That's one nice thing about complementary BJT collectors, you get arbitrarily low saturation voltage. All these ancient bipolar processes make s***ty PNPs, so they had to hack it with resistors and emitter followers.

I don't know why no one has made a FET or BiFET equivalent TL494. Or even just one with PNPs worth a damn. I suppose maybe LT has one, a voltage-mode alternating-output PWM with FET driven outputs. UC3525 has

I finally purchased a stock of SMTs, so I'm preparing a few simple circuits to put together. What better way to semi-integrate my own circuits.

Ouch, mixing first and third person tense!

Tim

Hey, I'm an engineer. I'm so bad at language things I used to get Bs in English. It was embarrassing.

My wife is a speech pathologist. She can stand by a hospital bed, with a TV going nearby, and listen to a stroke patient, with tubes in their faces, mumbling in German, and get it. But she can't figure out the simplest mechanical things, much less electrical.

John

On a sunny day (Sun, 15 Aug 2010 10:52:59 -0700) it happened John Larkin wrote in :

Yes, classic video driver :-)

A little inductance in series with the resistor can snap it up.

John

On a sunny day (Sun, 15 Aug 2010 11:47:20 -0700) it happened John Larkin wrote in :

Maybe also a little emitter resistor to linearise it.

That's true--from the rest of the conversation I was thinking you meant to drive a FET with it.

Of course--I anticipated that. So, if you don't mind throwing away another 1.5 cents on it, ...

| | | | | |

I've done that before. The pullup resistor looks like 10 ohms at the output. In practice, 2N4401/3 will do about 10 ohms, all the way down to short circuit. The 3904/6 shown might be closer to 20-40 ohms; ZTX651/751 will be more like 5 ohms. It's good enough for driving single power FETs, when you don't need the excess current capacity of, or are too cheap to buy, a TC4420.

Unfortunately, you get an RC rising edge, so it takes a long time to reach

So what's the ultimate driver? Low saturation voltage, high current output, low idle current (in both states). Can it be done, without the tedium of the level shifting logic typical of ICs?

Tim

It's a lot easier if you don't need DC coupling. Make the output stage just complementary mosfets. Drive the gates from a brutal TinyLogic buffer powered from +6, as many in parallel as you need. AC couple into the gates, and use appropriate resistors and DC restore diodes.

I could draw it, but I'm too lazy just now. We just finished a 4-hour hike on the Pacific Crest Trail, past Castle Peak. Careful thermal design kept two Widmers cold all the way until lunch. Spectacular. Exhausting.

I'll have to try the peak itself some day. It's the rim of an old volcano, surrounded by glacial stuff.

John

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

Yup. Might be able to DC-couple it using strategically placed zeners, assuming supply voltage isn't wide range. It could be very interesting to have seperate AC and DC coupling, so the DC is cheap and lazy, like resistors from drive to gate, and AC coupling is fast, so the edges push through, then the leakage keeps it there.

Trouble, then, is keeping drive strong in both directions, generating rail-to-rail drive (peaky or not), without wasting a bunch of bias current. It's not really elegant to drive CMOS with anything other than CMOS.

The tallest peak in Wisconsin is a quartzite mound about a billion years old. It was probably impressive back in the day, but a few hundred thousand years of glaciation have reduced it to a pimple on the landscape. Rib Mountain is only 800 feet above the surrounding plain, so it's technically not even a mountain, which is supposed to be over 1000 feet. That's Wisconsin for you: we don't even have any "mountains".

Hard to argue with the lack of earthquakes* or hurricanes, though. And tornados aren't very frequent up here, we're past the edge of tornado alley.

Tim

We did one gadget that needed p=p drive from prstty high voltage rails. We did tha cap coupling as noted, but added weak photovoltaic couplers to sustain drive all the way down to DC. Worked like a trained pig.

Some tube-type oscilloscopes coupled to the CRT grid (Z axis/unblanking) with a fast AC path, and a slow high-frequency charge pump to sustain things at DC. Cheap and easy.

Castle Peak is, like 9 grand.

Earthquakes don't kill a lot of people, so far. Hurricanes do.

John

Hmm, Porcupine Mountain is some 1594ft (and Google Earth shows spots local to it at 1610ft) and Lake Superior is 600ft, so that's pretty close to a mountain. The pine trees on top should do it. ;-)

I guess Rib Mountain is taller at 1924ft, but...

BTW, our honeymoon was in Porcupine Mountain. It's a very pretty, if incredibly depressed area.

Aw, with Global Warming, they're not even a big deal down here in Alabama anymore. We haven't even had a warning for over a year. ;-)

If the New Madrid fault rumbles again, it'll wake up a lot of people.

I don't know for sure, but I think we start calling it a mountain when it reaches maybe 3000-4000' over the local valley or plain. And cautiously, even then. Some of it is just "crumples" and "crinkles" and doesn't really count as a mountain, per se.

I'm about 15 miles due west of Mt. Hood, itself at 11,300' ASL. So I've got a back seat to more than a 10,000' elevation change.

I noticed how difficult it was to keep my bearings when traveling east of the Rockies, both north through the Dakotas and to Chicago as well as more southerly though Nebraska and the like. I suppose it means you can see the weather coming from 1000 miles away, though.

Cripes. That reminds me. How in the world do you folks get to charge TOLLs on interstate highways, for damned sake!? My taxes already paid for the damnable things. And it was like every 10 miles, or less, in some stretches.

Out here, we have had tolls (none left, now.) But when the bonds that fund the work are paid off, we get rid of the tolls. That seems like the right way to do it.

Over there, tolls seem permanent-like. Is this for the politicians' shoe boxes? (Do you remember the famous case of the Illinois Secy of State?)

Jon