Really fast stuff

T

Tim Williams 18 years ago

An interesting afternoon at the breadboard today. Having more or less built a reasonable vertical deflection amplifer, I am turning my attention to the trigger and sweep circuits. The first step is to turn an input signal, fast or lazy, into a really sharp blip, which then triggers a one-shot timer and so on (overall, I'm headed for variable delayed sweep and retrigger holdoff).

Well, having done nice things with differential amplifiers and low impedances already, I set up a diff amp with about 9mA tail current and a pair of 330 ohm collector resistors. The pair is two 2N3904's, no emitter degeneration resistors (the more gain the better, eh!). Feeding a

10ns-edge squarewave to the input, or really anything over 5MHz, I'm getting sub-10ns edges off the collectors. Rockin'. So I put on a 2N3906 common emitter stage, gain about 1.5 (220 ohms emitter, 330 ohms collector), and I get about 10ns rising, 20ns falling. Hmm kinda pokey falling, guess it doesn't really like to turn off all the way. What if I toss in a bias resistor so it *doesn't* completely turn off? Well, that might just work! Voila, 10ns up and down. Pretty damn good for a cheapass
2N3906 I must say.

A dash of positive feedback to sharpen gain, and another diff stage with enough +FB to schmidt-trigger-ize it and I've got some gnarly

Vote

R

Robert Baer 18 years ago

Back in the 70's or so, Farichild made some ECL-like logic for AMES or some-such gov agency, using NPNs. These ICs drew a fair amount of current per pair, and were fairly fast; 700pSec or better. Been too long to remember the tail current, but i think in the region of 1-10mA, and (naturally) were basically a linear current shifter (majority one side=1 to majority other side=0). With discretes, lead length can be a killer.

Vote

J

John Larkin 18 years ago

You seem to want to do this with discrets, which is fine, but it would be interesting to go the other way, and maximize a design using ic's.

There are lots of comparators, or cheap LVDS-to-ttl converters, with differential inputs and low or sub-ns response times. One of them, driving a trigger flipflop (ecl or cmos tiny logic) could be the trigger front-end. The flop would drive a clamp diode or an open-drain gate into a ramp capacitor, in turn driven by a current source or a bootstrapped resistor.

Opamps that swing a few volts p-p can be had up to a ghz or so bandwidth, AD8009 maybe. Front one of those with a dual-gate gaasfet follower maybe, to get a high input impedance. An output deflection stage could maybe be just opamps; a THS3001 will swing 20 volts p-p to

30 MHz maybe, so a pair of them might be able to drive deflection plates. Better would be a pair of lower-voltage amps, AD8009 again, feeding fast bipolars or mosfets or GaN fets common-gate, with series-shunt inductive peaking.

I think National used to make some monstrously fast (like 100 MHz) ICs for driving the video (crt grids) of color monitors. They might be interesting, too, and should have appnotes still.

John

Vote

B

bill.sloman 18 years ago

Fun though it may be to build things with the 2N3906, Farnell also stocks the 5GHz BFR92 which I used in similar applications in the late

1980's and sells for much the same sort of price as the 2N3906, not to mention the more recent 36GHz BFP640 which costs about three times as much - around two dollars in small quantities

http://www.ortodoxism.ro/datasheets/infineon/1-bfp640.pdf

These are surface mount parts and can't take much reverse voltage across the base emitter junction, but they are a lot quicker than the

2N3906.

http://www.ortodoxism.ro/datasheets/infineon/1-bfp640.pdf

In practice would be tricky to set up a circuit where a BFP640 actually performed better than a BFR92 so this is something of a red herring, but back in the late 1980's we had a lot of fun putting together BFR92 NPN parts with their BFT92 PNP complements to make faster versions of conventional discrete circuits.

Both the BFR92 and the BFT92 seemed to need a small surface mount resistor somewhere between 22R and 33R mounted close to and in series with their base pin to prevent oscillation - at least in the cobbled- together circuits we devised - but that seemed to be enough to make them pretty well-behaved.

From time to time we tried to get hold of simulation software, but back then anything that looked like it might work was out of our price range - I had enough trouble getting hold of scope that could follow our half-nanosecond wide electron-beam-unblanking pulses.

-- Bill Sloman, Nijmegen

Vote

Really fast stuff

Join the Discussion

Didn't find your answer?