how fast can a discrete Schmitt trigger be made?

[nuno]

Could this topology, in discrete, be made to work to, say 50 - 100MHz?
(didn't breadboard this, don't know if it works)

p.s. Forum does not accept .ASC files... could this be made to work too ?

[David Hess]

Tektronix used a very similar discrete Schmitt trigger up to at least 225 MHz in their well documented DC505 and 7D15 counters.

[Marco]

They did gbit/s discrete schmitt triggers nearly half a century ago ... so with modern components probably >>10 GHz.

As for this specific circuit, no idea ... you're driving these transistors pretty hard into saturation though. If you need more speed, replacing the emitter resistor with a current source and R1 with a zener (or diode drop) gives you more flexibility in resistor values and keeps the transistors out of saturation (ignoring overdrive conditions). Maybe use MMBT5179 (still cheap as chips and high voltage/robust, but significantly faster).

Like here.

[T3sl4co1l]

Probably the fastest today is upwards of 100GHz, build on monolithic InP substrate, CMOS, sub-micron feature size.  Purposes range from electrooptical circuits to radio and line communication receiver blocks (injection locked clock dividers, ring oscillators, amplifiers, mixers..).  At least, those are some of the articles I've read.

In the 70s, Tek also used tunnel diodes, which perform the given operation in, theoretically speaking, picoseconds -- it takes longer to transmit the signal along the component leads than to switch.

If you want to go out to Digikey and pick up some parts for testing, I'd recommend something like BFR92AW (~5GHz Si NPN) for general purpose fast-snot application.  Fancier devices (PHEMPTs, MESFETs, HBTs) are available discrete with fT up to 60GHz or something like that; again, pretty much more bandwidth than you can shake a PCB at (use them carefully, they'll oscillate at frequencies you didn't know existed! ).

Tim

[David Hess]

In the 70s, Tek also used tunnel diodes, which perform the given operation in, theoretically speaking, picoseconds -- it takes longer to transmit the signal along the component leads than to switch.

The last Tektronix design I know of which used tunnel diode triggers was the 100 MHz 465B in 1980 but they were switching to integrated triggers at least as early as 1973 with the 100 MHz 7B53A.  I would like to see the details of those integrated trigger designs but their datasheets are not that detailed.

In 1964 or 1965, the 50 MHz Tektronix 547 used tunnel diode triggers but the 30 MHz 545B used a circuit almost identical to the example given using 2N2484 transistors which are pretty slow at 50 MHz to 100 MHz.  That gives tunnel diodes about 2 decades of usage in trigger circuits.

I wish tunnel diodes were still generally available for trigger and pulse generator applications.  I wonder if their esoteric mode of operation made them too difficult to engineer with leading to low demand.  Integrated circuits would have done them in anyway though.

If you want to go out to Digikey and pick up some parts for testing, I'd recommend something like BFR92AW (~5GHz Si NPN) for general purpose fast-snot application.

That would be my choice as well when it positively absolutely has to be as fast as possible in simple surface mount and it even has a complement in the PNP BFT92W but they are overkill for this.  A 600 MHz MMBTH10 which also has a PNP complement in the MMBTH81 would be easier to work with.

[Alex Eisenhut]

In the 70s, Tek also used tunnel diodes, which perform the given operation in, theoretically speaking, picoseconds -- it takes longer to transmit the signal along the component leads than to switch.

The last Tektronix design I know of which used tunnel diode triggers was the 100 MHz 465B in 1980 but they were switching to integrated triggers at least as early as 1973 with the 100 MHz 7B53A.  I would like to see the details of those integrated trigger designs but their datasheets are not that detailed.

What about the sampling plug-ins? They happily trigger on GHz signals.

[Marco]

Easier to make a trigger than a Schmitt trigger, it only has to go fast from one state to the other ... the other way around can be much slower.

Any way, this is the circuit from the book I linked with a PNP common emitter amplifier for output (needs a negative supply for the current sources BTW). The PNP is mostly responsible for the horrible output waveform.

[nuno]

Nice, thanks for the ideas and info.

Exactly at what nodes are you measuring the blue and green traces, Marco?

I replaced the resistor by the zener and replaced the 2N3904 with oldschool RF transistors, BF199, and now it gets to 30MHz. Still not that high but that's the target I had. I'm now curious to breadboard (or PCBboard) this, got to go see if I have some BF199s in stock...

update: reduced R7 to 820 Ohm for more symmetric output

[David Hess]

In the 70s, Tek also used tunnel diodes, which perform the given operation in, theoretically speaking, picoseconds -- it takes longer to transmit the signal along the component leads than to switch.

If you want to go out to Digikey and pick up some parts for testing, I'd recommend something like BFR92AW (~5GHz Si NPN) for general purpose fast-snot application.  Fancier devices (PHEMPTs, MESFETs, HBTs) are available discrete with fT up to 60GHz or something like that; again, pretty much more bandwidth than you can shake a PCB at (use them carefully, they'll oscillate at frequencies you didn't know existed! ).

Oddly enough I designed a tunnel diode replacement circuit using 3 fast bipolar transistors for the various Tektronix trigger circuits which use tunnel diodes grounded at one end but I have not had an opportunity to test it.  Tunnel diode failures are common on the 465 series of oscilloscopes.  It is similar to the example circuit but drives a cascode to simulate the grounded tunnel diode.

The last Tektronix design I know of which used tunnel diode triggers was the 100 MHz 465B in 1980 but they were switching to integrated triggers at least as early as 1973 with the 100 MHz 7B53A.  I would like to see the details of those integrated trigger designs but their datasheets are not that detailed.

What about the sampling plug-ins? They happily trigger on GHz signals.

They do not use a level trigger at the highest frequency they support.  Some high bandwidth analog sweep oscilloscopes and most or all old sampling oscilloscopes support optional synchronous triggering or countdown triggering where an oscillator operating at a frequency where the level trigger can work is injection locked to the input.  On a 7T11 sampling sweep this extends the maximum trigger frequency from 1 GHz using tunnel diodes to 12 GHz.  On a more mundane analog sweep, it may extend the frequency range or make it more sensitive.

Easier to make a trigger than a Schmitt trigger, it only has to go fast from one state to the other ... the other way around can be much slower.

Any way, this is the circuit from the book I linked with a PNP common emitter amplifier for output (needs a negative supply for the current sources BTW). The PNP is mostly responsible for the horrible output waveform.

You can find exactly this circuit without the capacitor in one of the Tektronix examples I listed.  I think the AC hysteresis was provided by the junction capacitance of the zener diode they used.

[Marco]

Exactly at what nodes are you measuring the blue and green traces, Marco?

Green is the base of the Q2, blue is the collector of Q3.

The problem with the normal emitter coupled schmitt trigger is that unless the input signal is really well behaved you crash the collector voltage saturating the transistor. I'm guessing you don't want the negative supplies for the current sources ... but that circuit you have now is a bit voodoo If you want to keep a single ended supply it might better to turn it around and do it with PNP transistors.

[Marco]

You can find exactly this circuit without the capacitor in one of the Tektronix examples I listed.  I think the AC hysteresis was provided by the junction capacitance of the zener diode they used.

The Zener sets the upper trigger voltage (Vcc - Iz*R1 - Vz) and R1 times the emitter current sets the hysteresis.

[David Hess]

You can find exactly this circuit without the capacitor in one of the Tektronix examples I listed.  I think the AC hysteresis was provided by the junction capacitance of the zener diode they used.

The Zener sets the upper trigger voltage (Vcc - Iz*R1 - Vz) and R1 times the emitter current sets the hysteresis.

I mean in addition to the DC hysteresis controlled by the zener diode, they left out the capacitor which would generate AC hysteresis to speed switching.  Where a zener diode was not used, the value of the capacitor was on the order of the junction capacitance although the one you included is much larger.  When calibrating these circuits, you can see the AC positive feedback coupling into the input signal which is handy for identifying the exact switching thresholds.

There is an alternative AC only design where emitter degeneration is used and a small capacitor feeds part of the base signal to the emitter on the other side but I think it was intended for well controlled inputs.

[Marco]

I didn't really need to include it, was just a safety feature to make sure it would work which I didn't take out ... that circuit doesn't have the current to make a dent in the voltage across that particular zener at 100 MHz, might as well be a voltage source.

[nuno]

My source is well behaved, it's a high passed square wave of fixed frequency and amplitude.
Yes, I'm trying to not have a negative supply.

[Marco]

I'd try something like this, it's quite a few components with the current sources though ... in theory the current sources aren't necessary, but with resistors and especially without using large value resistors I'd have to do too much math to determine the thresholds ... and I refuse (not going to optimize for power consumption either).

With these current sources it's easy, i1*R1 is the hysteresis, i2*R1 is the lower threshold (only approximately at 100 MHz obviously). The current sources don't have a lot of headroom with 5V supply, about 1.3V but that should be enough (2 diode drops as a reference voltage, PNP+resistor as the current source). If you need more headroom and if you can live with a lower output swing you can use only a single diode on the bases and reduce R6 (the diodes are there to give some room for the transistors to swing without saturating).

You can do level shifting on the inputs with a NPN version as well to keep the transistors out of saturation ... but the output is referenced to the supply and requires translation ... with PNP you get a ground referenced logic signal.

[nuno]

So, you would implement the current source as a traditional 2 transistor current source, Widlar or similar, right?

Meanwhile I managed to do what I wanted. It's purely a theoric exercise. Unable to do it with less transistors. The high output impedance and low amplitude of the filter output makes the job very tough (for me, that is).

[Marco]

Nah, single transistor is probably good enough. As I said, resistors too but the location of the thresholds and hysteresis isn't as easy to calculate. It all becomes interdependent.

[nuno]

It all becomes interdependent.

Yehh, I see what you mean.