Push-pull to reference and GND

[741]

I have a logic signal, about 100kHz, I'd like translating closely to 0V and a reference. It'd be nice to have a low resistance output, maybe 100 Ohms.

The reference comes from a voltage reference chip eg REF4132 etc, TBD.

I was wondering about these options

• Complementary PMOS/NMOS pair
• Parallel CMOS outputs like 4050
• Parallel CMOS outputs like 4053

The CMOS options seem easier to 2nd source and cheaper. I am wary of DG400 etc: apart from the price I recall a project where they did not work so well at that kind of frequency.

[magic]

I think a logic gate powered from the reference voltage would work. The output is just a ~10Ω RDS MOSFET to GND or VCC and it switches wickedly fast with almost zero quiescent current between transitions.

[Zero999]

How many channels?

What's the voltage?

The 74HC4053 will work up to around 10V and has three channels, which could be connected in parallel, to reduce the impedance further.

[741]

I only need 1 channel, so paralleling would make sense.

I tried with some gates I have lying around (some are rather ancient).

At 5V supply, then a MC14001B (configured as an inverter) showed about 450R to Vdd when ON, but an HCF4053 was 280R.

It might well be a newer 40xxB has lower specified channel resistance.

[mawyatt]

We did this switching between a input voltage and ground to create an ultra-precision voltage divider without precision components, patented long ago (5030848). More recently used to create a precision square-wave for DMM AC calibration switching between a reference voltage (5.000VDC) and ground. As mentioned paralleling a bunch of ordinary CMOS gates works well, as does using a discrete PMOS and NMOS to make a CMOS discrete inverter. Most CMOS gates do not have an equal valued Ron to VDD (PMOS) and Ron Ground (NMOS), the NMOS is usually lower. If this matters you can add a small "equalizing" resistor in the CMOS ground path, or use a discrete version like we did with the square-wave DMM calibrator, use PMOS and NMOS FETs that have equal Rdson at the same magnitude Vgs.

Best, 

Edit: Here's an image of the DMM calibrator using the CMOS switching concept, the discrete NMOS and PMOS SOT devices forming a CMOS inverter are under the large electrolytic capacitor.

[David Hess]

As mentioned, CMOS gates work well if their supply voltage range includes the voltage that you want to switch.  Analog multiplexers work well.

Switching currents into a parallel resistance is also a good method and can be done with diodes and bipolar transistors.

[magic]

At 5V supply, then a MC14001B (configured as an inverter) showed about 450R to Vdd when ON, but an HCF4053 was 280R.

It might well be a newer 40xxB has lower specified channel resistance.

Dinosaur technology

Try some 74 series CMOS inverter, like 74xx04, 74xx14 or 74xx1G04/14 (SOT23) for xx=HC,AC,HCT,LVC depending on availability and required input and output voltage levels.

These have low enough resistance to drive 50Ω cables.
https://www.eevblog.com/forum/projects/compact-74ac14-pulse-generator-pcb/

[Kleinstein]

The CMOS level translators (e.g. 4050) linkely have more current spike at the input / supply than a CMOS switch (e.g. 4053). Modern lower voltage chips may have less charge spike than oder higher voltage ones.  So for the 4053 I would consider LV4053 or similar if 5 V range is enough.

With CMOS the on resistance changes with the supply.   If a more constant resistance is imporaten JFETs could be used - ideally a N and P channel in parallel to compensate the gate charges.

100 kHz is quite fast for PWM modulation, so the charge injetion can get quite important.

[741]

I'd forgotten about charge injection.  Maybe this is a reason to use discrete - I can slow the turn-on and lower the spike amplitude.

[Kleinstein]

The problem with modern discrete MOSFETs is that they essentially all contain gat protection that adds extra leakage and capacitance. The choice of low voltage, low gate capacitance fets is also small, especially p-channel.
A reason to go discrete would be using JFETs, a little like Datron in the 4910.

Slow switching would cause other problems, like more jitter and noise and temperature sensitivity. The speed and delay would depend on temperature and this would change the dead time.

The ready made CMOS switches are already quite good and often include at least some compensation to keep the charge injection low. There is some choice of slower and faster types if needed - with 100 kHz modulation one would more like look at the fast ones.

[Zero999]

Another option is the '4007, but I don't see how it's any better than an analogue switch. Get the HEF, rather than CD version because it's faster so likely to have lower capacitances.

https://assets.nexperia.com/documents/data-sheet/HEF4007UB.pdf
https://www.ti.com/lit/gpn/cd4007ub

[Terry Bites]

Have you investigated fast clipping?

[David Hess]

I'd forgotten about charge injection.  Maybe this is a reason to use discrete - I can slow the turn-on and lower the spike amplitude.

Or there are various ways to compensate for charge injection by injecting the opposite charge during switching.