Double inverted signal

[Momchilo]

Hi,
I have a 1V_pp square wave from 0 to 1V from a function generator. I want to increase the peak-to-peak voltage to 0V to 5V and I also need the inverted signal from 0V to 5V. The frequency of the square wave will be varied between 5Hz and 200kHz.
I know there are converter and inverter ICs, who would do the job perfectly.
But I tried to do this with some components which I have at home. The inverter is built by a npn-transistor (BC547) and two resistors. It works just fine, also at 200kHz. I thought the simpliest way would be to invert the inverted signal again to get the original square wave with an increased voltage. The schematic is attached and also the LTspice simulation.
But the double inverted signal isn't useable at frequencies higher than 40kHz, I attached a picture of the scope with the 200kHz double inverted signal and the output of the function generator. You can see that the rise time of transistor Q2 is way too slow. If I lower the value of R2, the rise time will be faster, but the voltage of the inverted signal will decrease (voltage divider). In the LTspice simulation the rise time is better than enough.
Is there a simple solution for my problem? Thanks for your help.
Best regards
Momchilo 

[Zero999]

It's more likely the transistor's base charge, which is responsible for the slow off time and rise time of the double inverted signal.

Try removing R2 altogether and adding another transistor to give the inverted output. Another possibility is adding a 100pF capacitor across R2, but that might delay the rise time of the inverted signal too much.

[rstofer]

If you remove R2 the inverted output will never rise higher than Vbe of the second inverter.

[StillTrying]

  'ed

[David Hess]

A single diode Baker clamp needs the lower forward voltage drop of a schottky or germanium diode to work.  A silicon diode Baker clamp can be made with two diodes.

https://en.wikipedia.org/wiki/Baker_clamp

Instead of a Baker clamp, I would start by bypassing R2 with enough capacitance to allow fast removal of base charge like Hero999 suggests.  100 picofarads is about right but try different values.

I am not familiar with the switching characteristics of the BC547 but that storage time seems awfully long.  I would expect more like 200ns maximum.

The circuit you really want is called a "data slicer" and is commonly made with a comparator, resistor, and capacitor but a couple of transistor inverters like you have should work also.

https://www.maximintegrated.com/en/app-notes/index.mvp/id/3435

[StillTrying]

Swapping the BC547 for the much slower 2N3904 seems to show the slow TR turn off better in simulation, it takes 1.6us to turn off. There's 3 versions in the .asc, the single diode is winning!

[Zero999]

If you remove R2 the inverted output will never rise higher than Vbe of the second inverter.

I know, which is why I suggested adding another transistor for the inverted output.

[Ian.M]

If you've got supply rails above and below the logic ones (e.g. +/- 12V), it looks like a carefully biassed clamped long tailed pair with speedup caps across the base resistors and chokes in series with the collector resistors and the tail resistor could get you logic level signals with rise and fall times as fast as 10ns, and negligible timing skew between the inverted and noninverted outputs.

n.b all caps Cpn are parasitics and strays, not actual components.

[Zero999]

Interesting, for educational purposes, but might not be practical as it requires a bipolar PSU and those inductors won't be cheap.

Here's what I was suggesting above. It won't be as fast and may benefit from Baker clamps, but will probably do.

[Ian.M]

Interesting, for educational purposes, but might not be practical as it requires a bipolar PSU and those inductors won't be cheap.

Omit the inductors - they only buy you a few ns.   However back in the early days when transistors were nearly as valuable weight for weight as gold, IMHO that would have been a viable solution.

If you want to see the difference, parametrise the inductors as {x} and add:

Code: [Select]

.step param x list 1p 100uto the sim.   1pH is a very short wire!  It helps see the timing variation if you add a trace for the input pulse signal at a similar size by plotting V(pulse)/1v

Here's what I was suggesting above. It won't be as fast and may benefit from Baker clamps, but will probably do.

Yes, I came up with something similar.   Baker clamps wont help much, and if you use them on Q2 or Q3, can actually make it worse due to the junction capacitance and the Miller effect.

[David Hess]

Swapping the BC547 for the much slower 2N3904 seems to show the slow TR turn off better in simulation, it takes 1.6us to turn off. There's 3 versions in the .asc, the single diode is winning!

The simulation is just wrong.  And 2N3904s aren't any slower than BC547s.