1
\$\begingroup\$

I am trying to use a digital potentiometer to reduce the amplitude of a square wave. I am getting what appears to be capacitive feedthrough on the rising and falling edges. This is getting amplified by later stages and causing glitches in the output.

Looking at the scope screen shot below, the cyan trace is the ~12v/us square wave input and the yellow trace is the wiper output of a MAX5387 100K digital pot set for about a 96K : 4K divide. After recovering from the initial glitch, the yellow trace exponentially settles to the expected value after about 4uSec.

I have looked up capacitive feedthrough for digital pots and have not found anything. It is not spec'd in any datasheets that I have seen. ADI lists digital feedthrough for some of it's pots, but that's about it.

I assume that digital pot capacitive feedthrough is a known phenomenon. Any idea which particular pots minimize it or how to mitigate it in general?

At this point my only option seems to be slew rate limiting the input square wave to a slope similar to the post glitch RC slope. This takes an extra op amp stage. I don't want to just put an RC on the input, since I need the edges to be symmetric and as fast as possible. Any other ideas on how to avoid the initial glitch?

12v/us square wave into Digital Potentiometer

\$\endgroup\$
1
\$\begingroup\$

Some capacitive coupling should occur, although it depends on the internal circuitry. The datasheet of your digipot does not show frequency response curves.

Also, high resistance values make parasitic capacitance more problematic.

This one does show frequency response, and the high frequency part of the graph does not seem to do anything funny. I remember seeing digipots with a peak in the HF part of the frequency response, but can't find a datasheet illustrating this right now.

Thus I would recommend selecting a digipot with lower resistance, and one that has a specified frequency response, without a HF peak.

\$\endgroup\$
  • 1
    \$\begingroup\$ I tried the AD5258 10K digital pot and it does not show any feedthrough. It has the largest bandwidth of any pot that I could find. Changing from 100K to 10K may have also been a factor. \$\endgroup\$ – crj11 Feb 13 '18 at 12:36
0
\$\begingroup\$

Digital potentiometers are typically constructed using MOSFETS that connect to a series of resistors. Those MOSFETs do have some some capacitance associated with them.

The MAX5387N is a 100K digital potentiometer with 256 steps, so at mid scale the output-resistance of the potentiometer is 50K in parallel with 50K, which is 25K. Page 3 of the MAX5387N datasheet lists the 3dB bandwidth as being 75kHz with the code set to midscale (80h).

An equivalent capacitance can be found from the bandwidth as follows...

2*pibandwidth = 1/(RC), where bandwidth = 75kHz, and R = 25kOhm

C = 1 / (2 * pi * bandwidth * 25k ohms) = 85pF

Note that the output capacitance will vary with code, but mid-scale should be the worst case.

Just based on the MIDSCALE FREQUENCY RESPONSE graph on page 7, you probably won't want to use the MAX5387N for signals that have any components above 10kHz.

\$\endgroup\$
  • \$\begingroup\$ I think that the capacitance that affects the bandwidth is shunt capacitance. In the scope trace, that is what I assume causes the exponential decay after the initial glitch as the wiper capacitance gets charged/discharged through the tapped resistance and wiper resistance. What I think that I am seeing is series capacitance, directly from the input to the wiper. \$\endgroup\$ – crj11 Feb 5 '18 at 20:52
0
\$\begingroup\$

A 6-bit (64 tap) Digipot will have 64 MOSFETS all connected together at the output wiper. If the GATEs (64 gates) of these MOSFETS bounce around, that bouncing appears on the output wiper.

Thus you should be sure the Digipot VDD is well-bypassed right at the pin.

Keep those gates quiet. That means the logic decoder needs a hard-VDD voltage and a hard-GND voltage.

\$\endgroup\$
  • \$\begingroup\$ Note that in this case the resistance setting is static, so the gates to all of the MOSFETS should also be static. One end of the pot is driven with a 12 v/uS edge rate square wave and the other end is grounded. \$\endgroup\$ – crj11 Feb 7 '18 at 13:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.