Maybe a tube? Just kidding. But there is a way to cheat here. You can still get buffers similar to the HA-0002 and 0033. I have sometimes used those in a "gang pack" and they'd drive almost anything. For really hardcore driver needs I stashed away a box of 0063 before they went, mainly because of their sturdy TO-3 design. Even in a non-feedback scenario you can achieve output impedances in the low single digits.
Even though I hate sockets, in a function generator it might be a good policy to socket this stuff and keep a few spares inside dummy sockets. Just in case someone accidentally transmits into the output. Not that I'd ever do that, of course, ahem ...
Being fascinated by function generator type output stages, I looked in on the AD815 and was going to ask what the problem was, as it seemed perfect for the job. First few pages showed a big output swing, big current and massive bandwidth. It was only until page 5 of the datasheet that the power bandwidth limitation was revealed. Looks like you caught that as well!.
Discrete output stages tend to end up with lots and lots of trannies and resistors. A cheap cheap x18 10MHz 40Vpp 50ohm I did a while ago, tots up
Way back, I'd drooled over using one of National's LH0033s. Most probably because they'd cheekily advertised it as "Damned fast". Never got the chance as at the time they cost way too much. Nowadays if a customer gets wind that I might be inclined to stick a high performance (=cost!) chip in their kit, they without fail, somehow miraculously discover, that they can back-off a bit on the spec' of what they want. A number of companies I'm now dealing with have been badly burned over the past 3 years by obsoleting speciality ICs. To the extent that the scanning radars of company buyers can lock-in on recognition of the usual words such as cheap, plentiful, rebate, low-cost, but now also 'discrete transistor', '324' etc!. Strange world. One day maybe, I'll have the pleasure of using something nice ... :) regards john
Yes indeed. You well understand this aspect. There's few finer pleasures than doodling with a few transistors and other components, looking for an arrangement that offers up particular properties. Don't know about the other guys out there but as time goes on, I'm getting to be more and more of a lazy bastard and won't reinvent a wheel if I can recycle some tried and tested stuff from a year previous. But then the fun aspect is lessened and I may as well have just stuck a fancy chip in, saving tailoring time, hence a quicker path through to the good stuff, ie those areas needing a bout of idle doodling, (real design? :-) regards john
The BF720 and BF721 parts come in an attractive SOT-223 package, but they're 300V parts, which means their beta and gain-bandwidth are compromised in favor of their high-voltage capability. Other good choices in the sot-223 package are the fzt2222a, fzt2907a and pzt4401, pzt4403. These BJTs have f_T in excess of 250MHz, and are rated at about 1W max each, with 1cm^2 area on the PCB.
That's an attractive aspect of this circuit. IRRC, I first saw that cute trick in an NSC note in the late 60s.
Let's evaluate the scene a bit. 10V into 100 ohms is 100mA, rising to 250mA and 3.1W if 12.5V via 50-ohms into an output short. The transistor output-stage maximum power dissipation occurs for say 7.5V out (at the 50-ohm resistor input) at 150mA into low Z load, for 1.1W at DC. Two sot-223 parts as shown may be fine, but three would be a bit safer.
It'd be more interesting if the amplifier were spec'd for +/-10V matched-impedance output into a 50-ohm load. We'd set it up for a gain of 2x, with +/-20V output and at least +/-25V supplies.
The 60V Vcbo rating of the fzt2222a and fzt2907a would suffice. One could run the input-pair current sources at 20mA and the collector sources at 10mA, to insure enough available current to drive the output emitter-follower transistors at full load. The 20mA current sources could be made from two BJTs, so I get a total transistor count of 22. The PCB would run rather hot.
Hmm, what does spice tell us about this circuit's performance? You can use 2n2222 and 2n2907 models, they're the same dies.
The buffer has the advantage that all 3 terminals of the input transistors are at the same AC-potential, because the collectors go to the output, thus a really wide bandwidth can be achieved. I agree with Win, those are high-voltage parts good for +/-150V not +/-10V. Siemens also makes some dual RF NPN and PNP with high bandwidth, but I do not know if the supply voltage is sufficient.
But at high frequencies slew rate becomes an important issue. For example, 20Vpp at 10MHz requires a better than 630V/us slew rate, while delivering a full load. That's a rarefied territory for +/-15V opamps. Generally opamps slewing that fast require a large error voltage before their slew rate approaches that speed.
One has to be sure each opamp has the same high error voltage so it slews in step with its neighbor. This could mean each opamp to be shared needs its own independent feedback network, plus the output ballast resistors shown above. Or some other trick would be needed yielding identical circuitry for all the output opamps.
I was interested in that kind of thing a few years ago because I had been using a Wavetek 142, and I wanted to design something with similar specs, using a MAX038 to generate the waveforms. The Wavetek used all discrete transistors in the output section AFAIR.
What I came up with in the end was an Elantec EL2009 1Amp 90MHz gain=1 buffer inside the feedback loop of a EL2030 amplifier run with a fairly low gain of about 3 to get good bandwidth. Then I had a second EL2030 before that to give a bit more gain. As well as all that, I had an OP-07 to null out the DC offsets of the whole arrangement. The problem is that now I can't get the EL2009 any more.
So my advice is not to rely on the continued availability of fast, high current buffer chips. Either use discrete transistors or buy more than you'll ever need of the chip.
If you come up with a good solution I would like to see it please.
Maybe cast an eyeball at THS3062. +-15 supplies, 7000 v/usec, dual,
300 mhz, isolated powerpad. I like to run one section as a gain stage, then three as pure followers (no overall fb) each with 150 ohms to the output node. This is TI's fast 30-volt process, one of the few around. They're rated at 140 mA or something, but really get droopy (and hot!) if you push them that hard, hence the paralleled sections.
Interesting arrangement. I remember Ban's original posting and look to use it. For 50ohm, 20Vpk, I started off with 30ma tails using paired BJTs. Output drive of 15ma via folded cascodes. For simplicity, the input tail currents ended up as a couple of power resistors and caps to filter supply noise. Had problems with output stage BJT heat distribution using low value emitter resistors. Went upto 22 ohms, used 6 output pairs and a +/-30 volts supply. Found additional cap' loading was limiting slew rate (feedback network), cascode current already high, so added extra base driver pairs for the o/p stage.
Fortunately, spice sim proved near to reality and allowed essential feedback shaping (x18 gain, BJTs all BC546, BC556). A whole rat pack of physical/electrical stability factors started biting at about 20-30MHz and without the spice I think I'd have been buggered!. regards john