I implemented the following circuit on a pcb board. enter image description here The goal of the circuit is to measure the capacitance of C1 and C2 in a differential way. C4 and C5 are the guards of C1 and C2. The comparator TL3016 is used as a Schmitt trigger creating an oscillation at around 5MHz. The voltage sources V1 and V3 are in the circuit to get the oscillation going in LTspice for the 1st few milliseconds.

I measured the jitter of the output of TL3016 with a PLL circuit at 10kHz bandwidth to be around 30ppm. I calculated the expected jitter based on the expected noise and slew rate to be 100 times smaller. What am I doing wrong? It would be great if you good give me some new insights.

enter image description here

Note that when measuring the signals with a scope I noticed that the dutycycle is not 50% of the oscillation signal, but more like 30%.

Many thanks, Chris

  • \$\begingroup\$ I think you mean "new insights". New insides are a bit out of our wheelhouse. I suggest you analyze the spectrum of the error signal to see if there is a mains or mains * 2 component, and also replace the power supply with a battery as a test. \$\endgroup\$ Commented Nov 29, 2022 at 12:26
  • \$\begingroup\$ The +5V supply adds noise, since the TL3016 outputs saturate to GND and +5V Vdd. Have you included power supply noise in your noise calc? \$\endgroup\$
    – glen_geek
    Commented Nov 29, 2022 at 16:51
  • \$\begingroup\$ @glen_geek I am using an LDO with spec 300uV rms. The TL3016 should have power supply rejection ratio of 80dB , so should add only 80nV rms noise \$\endgroup\$
    – Chris
    Commented Nov 29, 2022 at 18:35
  • \$\begingroup\$ Chris, that power supply rejection spec refers to +/- input pins, not to those complementary output pins that you feed back to those 50k resistors. \$\endgroup\$
    – glen_geek
    Commented Nov 29, 2022 at 18:44
  • \$\begingroup\$ @glen_geek Thanks a lot. I think this is the problem. The noise of the LDO has a cut-off frequency at 10kHz meaning that it does not get filtered with the bandwidth of the PLL (sqrt(fbw/fos)). If I do the math I get to 30ppm noise which is what I am measuring. \$\endgroup\$
    – Chris
    Commented Nov 30, 2022 at 6:52


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.