0

As stated in this question: Antiresonance of multiple parallel decoupling capacitors: use same value or multiple values? I have been curious to the real-world effects of different capacitors and their frequency responses. Using Spice with a small signal analysis is easy to do, but it really can't be replicated on a real board to my knowledge.

If I wanted to take a board, put a high speed IC on it with it clocked at a high frequency to test various decoupling capacitor schemes, what would be the best way to quantitatively see results? I see a few options:

  1. Check the ripple voltage on the input supply to the IC. This would show how well the decoupling caps are able to supply the IC with high F voltage spikes. The issue is measuring the AC on top of the DC is not easy to do with a normal scope setup. See here: https://www.tek.com/blog/a-better-method-for-analyzing-ripple-on-power-rails It would require a high bandwidth scope and high cost probes.
  2. Use a spectrum analyzer to see the frequency content on the power rail. Similar to how conducted emissions would be checked. I really don't know much about this method, so I'm not sure it it would work.
  3. Use a spectrum analyzer to check the radiated emissions from the PCB. This would work, but requires a very noise free environment which I don't know if I can do.

Ultimately, I don't need to achieve a real-world comparison with any standard. I just need a quantitative value to compare one board to another, with the only difference being the decoupling scheme.

Kyle Hunter
  • 726
  • 4
  • 21
  • "it really can't be replicated on real boards": Of course it can! That's the whole point of modelling: once you model something sufficiently well for your purpose, the model then predicts what the real-world object does (with respect to what you modeled it about). – Marcus Müller Jan 30 '21 at 18:24
  • Also, for all your 2. and 3. you forget that you're having a measurement problem; *how* do you measure these things? "With a spectrum analyzer" doesn't actually describe how you're actually trying to pick up this information, and from where specifically. – Marcus Müller Jan 30 '21 at 18:26
  • @MarcusMüller Well... That's what I'm asking about... I'd love to be able to see a db vs freq chart of the circuit I described, from 10KHz -> 1GHz, but it's not exactly an easy thing to do. So "once you model something sufficiently well for your purpose, the model then predicts what the real-world object does" may be true, but seeing the results of it easily are not. – Kyle Hunter Jan 30 '21 at 18:30
  • I don't understand – SPICE makes exactly that easy. *Measuring* something over a range of five orders of frequency magnitude is really hard! (and the reason why your spectrum analyzer is not a cheaper option than your scope with adequate probes) – Marcus Müller Jan 30 '21 at 18:31
  • @MarcusMüller Yes and I want to do it on a real board... To accurately model with spice for this it's hard to measure the inductance/resistance/capacitance of each trace from the capacitor to IC. The reason I wish to do this, is the two different schools of thoughts on decoupling caps (same values vs different) is 100% dependent on the parasitic not easily modeled. Hence why I wish to see it on a real board... Response to your edit: I know it's hard! That's why I am asking how best to do it! – Kyle Hunter Jan 30 '21 at 18:34
  • well, then model your traces? That's not really magic, either; for your low frequencies, a simple RLC-based model depending only on ground distance and trace geometry would probably suffice. – Marcus Müller Jan 30 '21 at 18:35
  • 1
    try 50nH/cm and 5pF / cm pair for 100 ohm tracks and consider parasitic on part and traces https://tinyurl.com/yysma6r6 – Tony Stewart EE75 Jan 30 '21 at 18:38

0 Answers0