EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: DenzilPenberthy on August 19, 2014, 10:41:26 am
-
Hi All,
I'm designing a circuit to condition some fast pulses from a photon counting avalanche photodiode. The first requirement is to stretch the pulses out a bit to something like 50ns. I can do this with this circuit here using an LT1711 comparator and an RC network. I can vary the threshold of the comparator with a potentiometer and also how long to stretch the pulses for by replacing R1 with a variable resistor (and reducing C1 to something like 47pF)
http://www.planetanalog.com/document.asp?doc_id=527407 (http://www.planetanalog.com/document.asp?doc_id=527407)
This gives me a 0-5v pulse of varying length but in the order of 50ns (and it's complement on the 'notQ' output) Happy days so far.
My second requirement is I'd like to be able to vary the height of the pulses from 0-1v to 0-20v. This is outside the comparator's supply range of 2.7 to 12v so simply varying Vcc is out. Any thoughts on how to approach this? It's feeding into a 1k input which can be physically very close so I'm not overly worried about impedance matching.
I've built a similar system to this before with an output stage consisting of a 50R pull up resistor and a pull-down MOSFET but the Rds(on) of the MOSFET prevents the output going fully to zero.
-
There are lots of ways to do this but they all come down to a discrete level shifter stage and maybe some type of current switching if you want precision output levels.
How precise do the output levels need to be? You mentioned MOSFET Rds being a problem.
How fast do the transition times need to be? An LM311 or better could be used as a high voltage output stage with pulses that wide.
-
The output levels don't need to be super precise, just variable with a pot from 1 to 20v. The output does need to be pretty much 0v in between pulses though.
Transition times need to be as quick as possible but sub 10ns certainly.
Cheers.
-
I picked up a Systron-Donner 100A Pulse Generator recently for around $50. It is old (~1971), but works well. About 4ns rise time. Goes to 10V on positive, -10V on negative, that gives you 20V (at 50ohm) :)
Shaky hands photo
(https://www.eevblog.com/forum/projects/1-20v-variable-height-fast-pulse-generator/?action=dlattach;attach=106191;image)
-
I made a rather lengthy post detailing various solutions but the forum lost it. |O
-
I made a rather lengthy post detailing various solutions but the forum lost it. |O
The margins were too small to contain your marvelous proof :(
Personally, I think such a stage might be an interesting exercise in 2N3904/6 technology... not at all impossible, though if it has to be source terminated (like a proper function generator output is), you'll definitely be pushing limits, reaching for 40V and 400mA at the top end there. Be good to have something beefier than a 2N3904 to drive that, but still as fast. And of course, easier overall with things peppier than 3904s (e.g., MMBTH10/81, etc.). Or other options, like a 74AC or LVC gate to generate the pulse shape, followed by a variable gain stage (some current feedback op-amps might do that?).
Tim
-
I posted about the unsuitability of garden variety transistors when the forum misbehaved. I doubt 200 to 300 MHz transistors like the 2N3904 and 2N4401 series are going to be fast enough to make less than 10 nanosecond transitions even under the best of circumstances but it would be fun to test. Use the complementary outputs of a fast comparator to emitter switch a 2N3904/2N4401 and use a 2N3904/2N4401 cascode with a resistive load. A transistor designed for saturated switching might do it.
That however leaves a lot of options but it depends on what kind of output impedance is being driven. If it is a 50 ohm transmission line, then 20 volts is going to be a problem.
Jim Williams wrote an application note about fast level shifters.
-
About 4ns rise time.
Google avalanche pulse generator.
They can get you to the pico second territory.
-
I posted about the unsuitability of garden variety transistors when the forum misbehaved. I doubt 200 to 300 MHz transistors like the 2N3904 and 2N4401 series are going to be fast enough to make less than 10 nanosecond transitions even under the best of circumstances but it would be fun to test. Use the complementary outputs of a fast comparator to emitter switch a 2N3904/2N4401 and use a 2N3904/2N4401 cascode with a resistive load. A transistor designed for saturated switching might do it.
That however leaves a lot of options but it depends on what kind of output impedance is being driven. If it is a 50 ohm transmission line, then 20 volts is going to be a problem.
Jim Williams wrote an application note about fast level shifters.
I know they're capable of the edge speeds, as I've *breadboarded* ECL style circuits that fast -- alas, I didn't get very many gates down before the whole circuit oscillated uncontrollably. :P Getting the full swing will be more of a challenge; you may have to drive the base with peaks even more than 400mA (i.e., not even cascode may be fast enough) to get it fast enough.
Other than that, I have some assorted RF and video type transistors in my junk box. 2N3866 stands out as a very likely option, say two in parallel as an emitter follower. They're still more-or-less available, but not cheap (anything that's only made by Central Semi. anymore is a bad sign..) Beyond that, not much very recognizable AFAIK. Which is kind of cheating, both myself and for purposes of being informative: hard to find specs to even use them, and impossible to reproduce the circuit (with the same obsolete parts).
To get *really* esoteric, maybe there's a distributed amplifier approach where you use each stage of transistors to generate a fraction of the voltage required (small-swing conditions where they're known to be fast), and stack them all up in the output (transformers? eww...).
Tim
-
With the SN74ACxx serie you can get a rising edge of 2ns and 5 volts output, with this you should be able to drive a RF power transistor and get 20 volts.
-
I know they're capable of the edge speeds, as I've *breadboarded* ECL style circuits that fast -- alas, I didn't get very many gates down before the whole circuit oscillated uncontrollably. :P Getting the full swing will be more of a challenge; you may have to drive the base with peaks even more than 400mA (i.e., not even cascode may be fast enough) to get it fast enough.
I would use the cascode just for the additional voltage gain and the reverse transfer isolation it provides. If necessary it would also allow a fast but low Vce transistor to drive a slower but higher Vce transistor.
Other than that, I have some assorted RF and video type transistors in my junk box. 2N3866 stands out as a very likely option, say two in parallel as an emitter follower. They're still more-or-less available, but not cheap (anything that's only made by Central Semi. anymore is a bad sign..) Beyond that, not much very recognizable AFAIK. Which is kind of cheating, both myself and for purposes of being informative: hard to find specs to even use them, and impossible to reproduce the circuit (with the same obsolete parts).
I did preliminary work on a similar pulse generator for oscilloscope calibration. One of the things I wanted to do was see how fast I could push 2N3904 and 2N4401 type transistors but I have not gotten around to it yet. I think best case with a 2N4401 into an active load, I could only barely make 10 nanoseconds if that at 20 volts if the current was high while a faster transistor could do it easily. The examples Jim Williams published using 1 GHz transistors could switch 10+ volts in 4 to 7 nanoseconds. On the other hand the pulse generators I have used with slower than 2 GHz transistors get into the 600 picosecond range.
To get *really* esoteric, maybe there's a distributed amplifier approach where you use each stage of transistors to generate a fraction of the voltage required (small-swing conditions where they're known to be fast), and stack them all up in the output (transformers? eww...).
I can't see that being necessary.
DenzilPenberthy, with a 1k load impedance I do not think you will have any problem using an emitter switched level shifter like Jim Williams shows. Biasing to produce a 0 to variable output will add complexity.
-
My second requirement is I'd like to be able to vary the height of the pulses from 0-1v to 0-20v. This is outside the comparator's supply range of 2.7 to 12v so simply varying Vcc is out. Any thoughts on how to approach this?
Although I'm not an expert on this matter, its just I do know my "ancient" :P pulse generator Tektronix PG 508 capable of generating pulse with transition time minimum at 5 ns, and with an "adjustable" pulse height from -20 Volt up to 20 Volt. See attached photo below.
Google for it's manual, its plenty out there and has a full complete schematic in it, who knows you may get something from there.
Just a thought, not sure if this helps or not. :-//
-
The PG508 is worth studying but it uses a linear output stage so it is pretty complicated and the output is only > 20 volts peak-to-peak and not +/- 20 volts to ground which would be 40 volts peak-to-peak.
Not all of the specifications at the beginning of the manual correct. This can be seen from the output amplifier schematic which shows a +/- 15 volt supply for the common emitter class AB output stage. Bias for the output stage further limits the output to about +/- 11 volts.
-
It's feeding into a 1k input which can be physically very close so I'm not overly worried about impedance matching.
I've built a similar system to this before with an output stage consisting of a 50R pull up resistor and a pull-down MOSFET but the Rds(on) of the MOSFET prevents the output going fully to zero.
Why not use a 50 Ohm pull down and a N / P pair of transistors to pull it up then?
-
Thanks very much for the input everyone.
I've had a look at the app note for high speed comparator techniques by Jim Williams. In particular the section on level shifting (Appendix E).
http://cds.linear.com/docs/en/application-note/an13f.pdf (http://cds.linear.com/docs/en/application-note/an13f.pdf)
I've played about with the circuit in Fig E2 in LT Spice. I find it switches on much more slowly than the quoted figure of 4ns. I'm not quite sure why. It also switches off very slowly and messily once a bit of capacitance is added to the output. This is due to it not having any ability to actively pull the output low, the charge just bleeds lazily away througn the 1k resistor and back through the shottky diode etc.
Then I've played about with the circuit in Fig E4 but with RL and the power FET removed. This also takes multiple tens of ns to turn on. As above, it also turns off very slowly with some load capacitance added although in a much cleaner fashion than the previous circuit.
Is it worth me posting screenshots/LT spice files?
I can't find any references to emitter switched level shifters, especially by Jim Williams. It's not a technique I've come across before. Does anyone have any useful links?
I was having a bit of a clear out this week and was about to chuck a load of drawers full of old RF bipolar transistors. there's a drawer full of 2N3866. Glad I didn't :)
Edit: Also, I'm super grateful for everyone's help but it's a public holiday on Monday here (plus the Tuesday at the university where I work). I finish work in half an hour so probably won't be back to this thread until Weds morning :) Have a nice weekend everyone!
-
Thanks very much for the input everyone.
I've had a look at the app note for high speed comparator techniques by Jim Williams. In particular the section on level shifting (Appendix E).
http://cds.linear.com/docs/en/application-note/an13f.pdf (http://cds.linear.com/docs/en/application-note/an13f.pdf)
That is the one I was thinking of although the same discussion is in at least 2 other application notes.
I've played about with the circuit in Fig E2 in LT Spice. I find it switches on much more slowly than the quoted figure of 4ns. I'm not quite sure why. It also switches off very slowly and messily once a bit of capacitance is added to the output. This is due to it not having any ability to actively pull the output low, the charge just bleeds lazily away througn the 1k resistor and back through the shottky diode etc.
Then I've played about with the circuit in Fig E4 but with RL and the power FET removed. This also takes multiple tens of ns to turn on. As above, it also turns off very slowly with some load capacitance added although in a much cleaner fashion than the previous circuit.
Is it worth me posting screenshots/LT spice files?
I can't find any references to emitter switched level shifters, especially by Jim Williams. It's not a technique I've come across before. Does anyone have any useful links?
That is just where I would start. When I wrote "emitter switched" I was referring to the first transistor in the Jim Wiliams circuit which does the level shift. Emitter coupled might be another term for it but since the base and emitter are both driven, I prefer emitter switched.
I do not know how clean a pulse you want but switching a pair of current sources into a shunt resistor to ground will produce a clean output. Edge rates will depend on the shunt resistance and output capacitance.
Alternatively doing essentially the same thing with a class B output stage will make it insensitive to load impedance.
With some Vbe voltage drop compensation, the above can produce accurate output levels within 10s of millivolts anyway but will require a negative bias supply.
-
There are some ultrafast MOSFET drivers which can do what you want if a little bit of delay isn't an issue. IXRFD630 has 3ns rise time with a 1nf load (which would take some serious RF transistors to replicate) but with ~20 ns propagation delay.
-
I have a gate driver simulation that claims 1-2ns rise/fall time into some sort of load (might be 10nF + 1 ohm..?). And would probably stay under 5ns with real components. Though it's pretty wasteful on the shoot-through and all.
This is a similar circuit, stripped down a little in speed and capacity; it already does ~8ns.
(http://seventransistorlabs.com/Images/GD9.jpg)
(http://seventransistorlabs.com/Images/GD8.jpg)
Though it's not really useful for present purposes, what with the waveform being all nasty like that.
Tim
-
I have a gate driver simulation that claims 1-2ns rise/fall time into some sort of load (might be 10nF + 1 ohm..?).
Can we see it? :) (If not, with what transistors? That's >100 Amp after all at 20 V.)
-
I drew up designs a couple years ago for a more general purpose reference pulse generator with separate precision limits on the positive and negative levels. It was limited to below 20 volts peak-to-peak using a cascode output stage because fast complementary pairs are limited to a Vcbo of 20 volts.
The voltages could be doubled using a double cascode which some oscilloscope CRT amplifiers did use and some of those had 3.5 nanosecond transition times or maybe better driving a fair amount of capacitance.
MOSFETs might be better but I have not tried them in this sort of application.
-
I don't see how a cascode would help, the upper transistor in a NPN cascode would still need to be able to support the full voltage if you want a full voltage swing. Wouldn't those CRT amplifiers simply use the cascode to level shift a small swing AC voltage onto a high DC voltage?
The advantage of MOSFET is the simple driving AFAICS (no need for emitter degeneration or baker clamps to keep it out of saturation). The disadvantage is cost (seems to me you are stuck with RF LDMOS, can't abuse low end transistors like with BJTs).
-
Another way to think of it: the voltage-gain device at the 'top' of the cascode is only turned off by its load current. If the 'bottom' transistor turns off completely, the 'top' emitter (or source or whatever) current goes to zero and base current equals collector current (or whichevers). Which is pretty weak if it's a resistive load: turn-off being proportional to voltage drop.
A push-pull construction, with matched and inverse output currents, and either a resistive load (to set Vo = (Iup - Idown) * R) or a feedback circuit (to yield a constant voltage output, but performance limited by GBW) could go faster, because the turn-on of the opposite polarity delivers turn-off current for the first side. But the waveform probably isn't going to be too great, because in those few nanoseconds the one side is being turned off, it's subtracting current from the load, so it will still 'drool' a bit on turn-off (in whichever way: as a delay, like reverse recovery, or more of a ramp (of some nonlinearity) due to Miller effect, or the voltage goes most of the way, but the last 10% or whatever goes slowly due to charge storage).
The nice part about driving something common emitter (or source, or..) is, you can force far more input current than load current, and thus potentially achieve real voltage gain at h_fe < 1. (And if the voltage gain is reasonably high, you'll actually be getting power gain out of it as well, which means you're doing better than an ideal transformer.)
Tim
-
The cascode helps in three ways:
1. It allows a higher voltage swing because the Vcbo becomes the breakdown voltage instead of the Vceo.
2. It lowers the effects of the drive transistor's reverse transfer capacitance.
3. It controls the drive transistor's Vceo allowing higher Ft.
-
Another way to think of it: the voltage-gain device at the 'top' of the cascode is only turned off by its load current.
What I didn't get was the increase in voltage swing coming from Vcbo vs Vceo.
-
These guys (http://www.avtechpulse.com/medium/) seem to know a little about fast pulses and high voltage. Getting a schematic from one of their units might be very revealing. FWIW
Above link is for their medium speed this link (http://www.avtechpulse.com/speed/)is their fast stuff.
-
Where the Vcbo versus Vceo advantage really shows up is in composite power switches where a low voltage MOSFET drives the emitter of a bipolar transistor.
High voltage MOSFETs have disproportionally high Rds(on) and high reverse transfer capacitance. This is neatly solved with the bipolar cascode which also takes advantage of the bipolar transistor's high Vcbo to make a composite device which combines the best features of both.
Another advantage of the cascode connection which I forgot about is isolation of the switch capacitance from the output which encourages clean switching. Diodes can do this also though.
-
Assuming for a moment that the load capacitance is small enough to just handle the high to low transition with pull down would this kind of circuit work? (If pulse repetition is low enough you could just use a capacitor fed by a resistor/inductor instead of V5, but you'd need to add an extra diode in line with R3 to get a little more Vbe swing for the PNP on turn off ... all in simulation at least.)
PS. BFG194 == BFG31, just picked some random RF transistors simetrix had in the library with sufficient current capacity. The BAS70 was a bad pick, needs to be something like the BAT54.
-
Where the Vcbo versus Vceo advantage really shows up is in composite power switches where a low voltage MOSFET drives the emitter of a bipolar transistor.
High voltage MOSFETs have disproportionally high Rds(on) and high reverse transfer capacitance. This is neatly solved with the bipolar cascode which also takes advantage of the bipolar transistor's high Vcbo to make a composite device which combines the best features of both.
If you actually make use of Vcbo > Vceo for such a composite switch won't the emitter voltage rise above base voltage during turn off pulling the drain voltage with it? (AFAICS the transistor will start to leak when Vceo is exceeded to make that happen.)
-
Assuming for a moment that the load capacitance is small enough to just handle the high to low transition with pull down would this kind of circuit work? (If pulse repetition is low enough you could just use a capacitor fed by a resistor/inductor instead of V5, but you'd need to add an extra diode in line with R3 to get a little more Vbe swing for the PNP on turn off ... all in simulation at least.)
Sure. Many fast pulse generators which do not have to deal with high load capacitance work that way and if the 2N2907 is fast enough, this is a great way to start.
-
Where the Vcbo versus Vceo advantage really shows up is in composite power switches where a low voltage MOSFET drives the emitter of a bipolar transistor.
High voltage MOSFETs have disproportionally high Rds(on) and high reverse transfer capacitance. This is neatly solved with the bipolar cascode which also takes advantage of the bipolar transistor's high Vcbo to make a composite device which combines the best features of both.
If you actually make use of Vcbo > Vceo for such a composite switch won't the emitter voltage rise above base voltage during turn off pulling the drain voltage with it? (AFAICS the transistor will start to leak when Vceo is exceeded to make that happen.)
The emitter voltage does not rise significantly higher than the base voltage so this is not a problem. There is an interesting question about what happens to stored charge if the cascode transistor is saturated. As far as I know, it flows out of the base and sometimes provisions have to be made to handle it. This implies that a baker clamp could be used to prevent saturation but since the base bias is low impedance, that would be difficult to do and I have never seen it done.
This is not just some hypothetical configuration; it really works for high power switching and linear transconductance outputs. Sometimes you find it used with low voltage monolithic switching regulators to extend the high voltage capability of the output switch to off-line levels. When this is done, the current protection circuits in the regulator still function and protect the cascode transistor as well.
-
The emitter voltage does not rise significantly higher than the base voltage so this is not a problem. There is an interesting question about what happens to stored charge if the cascode transistor is saturated. As far as I know, it flows out of the base and sometimes provisions have to be made to handle it. This implies that a baker clamp could be used to prevent saturation but since the base bias is low impedance, that would be difficult to do and I have never seen it done.
Works just fine -- you're not limiting the base voltage, but clamping the collector so it drops no less than Vf below base voltage. Which is the same thing anyway, just a different way of saying it.
You'd want base supplied from a constant voltage in this case, rather than something like a bypassed resistor divider. I suppose that goes without saying anyway, since base current rises in saturation. Not usually bothered with for linear circuits, but if you want quick recovery from saturation, or high dynamic range or whatever, a good idea.
This is not just some hypothetical configuration; it really works for high power switching and linear transconductance outputs. Sometimes you find it used with low voltage monolithic switching regulators to extend the high voltage capability of the output switch to off-line levels. When this is done, the current protection circuits in the regulator still function and protect the cascode transistor as well.
A cascode is naught but a transconductance output, so it works nicely here. At least, as long as you don't totally smash the bottom device on -- you wouldn't want to use a naked MOSFET there with no degeneration, not unless you like crispy base bondwires on the top transistor!
Tim
-
Emitter switched transistors would be quit helpful here. The upper transistor allows a lower rated, thus higher speed to be used.
st or siemens made such a device ten or so years ago.
-
Hi, Everybody, This is my first post. Hope I'm doing everything right. I just did a small series on a fast pulse generator on my youtube channel. The 2N3904 will give you a rise time in the picoseconds with very few parts. Changing the 7.5 pF cap will alter the amplitude. Anyways, if your interested, check it out. http://youtu.be/I1gfUNh5PJQ (http://youtu.be/I1gfUNh5PJQ)
-
Nice video :-+
-
Thank you!
I'm a you tube noob, so still getting used to the system. All in good time I guess :)
-
Oh Ya, You have a great channel yourself. I had to keep watching..... Because I couldn't..... Resist. 8)
-
Hi, Everybody, This is my first post. Hope I'm doing everything right. I just did a small series on a fast pulse generator on my youtube channel. The 2N3904 will give you a rise time in the picoseconds with very few parts. Changing the 7.5 pF cap will alter the amplitude. Anyways, if your interested, check it out. http://youtu.be/I1gfUNh5PJQ (http://youtu.be/I1gfUNh5PJQ)
An avalanche pulse generator might be handy for generating fast drive for the level shifter but by itself it will be unsuitable in this case because of the requirement for a controlled but variable output voltage.
-
The emitter voltage does not rise significantly higher than the base voltage so this is not a problem. There is an interesting question about what happens to stored charge if the cascode transistor is saturated. As far as I know, it flows out of the base and sometimes provisions have to be made to handle it. This implies that a baker clamp could be used to prevent saturation but since the base bias is low impedance, that would be difficult to do and I have never seen it done.
Works just fine -- you're not limiting the base voltage, but clamping the collector so it drops no less than Vf below base voltage. Which is the same thing anyway, just a different way of saying it.
You'd want base supplied from a constant voltage in this case, rather than something like a bypassed resistor divider. I suppose that goes without saying anyway, since base current rises in saturation. Not usually bothered with for linear circuits, but if you want quick recovery from saturation, or high dynamic range or whatever, a good idea.
The problem at least the way I see it is that once the baker clamp activates, any additional emitter current is drawn from the base circuit which is what happens anyway if the transistor saturates. If the base current is unlimited, then bad things will happen pretty quickly so the base supply cannot be a constant voltage if saturation or baker clamping is expected.
In the example linked below, the base impedance is controlled so a forced beta is achieved.
This is not just some hypothetical configuration; it really works for high power switching and linear transconductance outputs. Sometimes you find it used with low voltage monolithic switching regulators to extend the high voltage capability of the output switch to off-line levels. When this is done, the current protection circuits in the regulator still function and protect the cascode transistor as well.
A cascode is naught but a transconductance output, so it works nicely here. At least, as long as you don't totally smash the bottom device on -- you wouldn't want to use a naked MOSFET there with no degeneration, not unless you like crispy base bondwires on the top transistor!
Pages 52 and 53 of Linear Technology application note 19 have a discussion of the basics involving both MOSFET and bipolar power cascode stages:
http://www.linear.com/docs/4176 (http://www.linear.com/docs/4176)
I have been meaning to wire up a test circuit for the bipolar version to get a better feel for what happens at turn off with power devices. I am more used to small signal cascode behavior.
I do not understand your comment about the MOSFET unless you are referring to turn on that is so hard that gate current destroys the device.
-
The problem at least the way I see it is that once the baker clamp activates, any additional emitter current is drawn from the base circuit which is what happens anyway if the transistor saturates. If the base current is unlimited, then bad things will happen pretty quickly so the base supply cannot be a constant voltage if saturation or baker clamping is expected.
But you're assuming the emitter current is unlimited. Which it isn't, at least in any sensible circuit -- that's what my post went on to say, quoted below:
A cascode is naught but a transconductance output, so it works nicely here. At least, as long as you don't totally smash the bottom device on -- you wouldn't want to use a naked MOSFET there with no degeneration, not unless you like crispy base bondwires on the top transistor!
Pages 52 and 53 of Linear Technology application note 19 have a discussion of the basics involving both MOSFET and bipolar power cascode stages:
http://www.linear.com/docs/4176 (http://www.linear.com/docs/4176)
I have been meaning to wire up a test circuit for the bipolar version to get a better feel for what happens at turn off with power devices. I am more used to small signal cascode behavior.
I do not understand your comment about the MOSFET unless you are referring to turn on that is so hard that gate current destroys the device.
Normally, the bottom transistor is emitter/source degenerated, so that it is no more than a transconductance amplifier, collector/drain current proportional to input voltage over the useful range. The ONLY possible way you could draw unlimited emitter current from the top BJT, is if you have unlimited collector/drain current to the bottom transistor: which means you're saturating the poor thing, not using it in the linear range.
The example in Fig.41 (p.53) in your link works by allowing the top transistor to saturate, and allowing the bottom transistor to switch Vin rather than Vload. This is a net gain because the voltage range is extended, but it is NOT a true cascode because the bottom transistor sees more than Vbe voltage swing at its collector -- indeed, it is allowed to saturate. The disadvantage of this circuit is, it costs D * (Vin - Vbe) / R1 extra supply current, and 2*Vce(sat) voltage drop. It also suffers from storage time, which isn't as bad as usual, because, thanks to R2-D2, Q1 is forced to turn off with hFE(off) as low as 1 (i.e., during turn-off, the load current leaves through the base terminal).
Also, since the V_SW terminal is not a transconductance output (or at least, you definitely wouldn't want to use the poor thing that way..), it cannot be used as a true cascode.
I used a broadly similar approach in this circuit,
http://seventransistorlabs.com/ClassD1/Images/ClassD_Full.png (http://seventransistorlabs.com/ClassD1/Images/ClassD_Full.png)
(see http://seventransistorlabs.com/ClassD1/index.html (http://seventransistorlabs.com/ClassD1/index.html) for background)
The middle drive circuitry converts from an open collector (2/2 LM393) to both a common-emitter (inverting) and a common-base (noninverting) stage, which are CCS loaded for speed, and emitter-follower'd for driving the output gates. The diode drops are arranged so that the LM393 output is pulled up to about 1.2V, or pulls down to less than 0.7V, to fully switch the two transistors. Note that the '393 has to sink full load current of the common-base branch, whose base voltage will collapse as it saturates. The same mechanism applies in Fig.41, but over a larger voltage swing (i.e., Vin).
Tim
-
I was looking for something else and ran across a good example of the type of level shifter I would on page 6 of Linear Technology application note 98:
http://www.linear.com/docs/6159 (http://www.linear.com/docs/6159)
As shown it operates into 50 ohms which limits the output voltage range.
The design shown on page 4 is similar to some other suggestions here and could be made to work also.