KSTR
Major Contributor
This!Note just blindly paralleling capacitors can also cause problems. The capacitors' ESL and ESR can work together to again peak the noise at unexpected frequencies by causing resonances among the capacitors themselves. Ironically, this is often worse with low-ESR capacitors and low-impedance power pours (planes), since it provides higher-Q resonances among the capacitors.
Most audio circuit engineers don't seem to use a network analyzer to check the supply impedance right at the component pins. The use of other methods like step exitation to check supply impedance is also not in wide use it seems. Once you do, you see the disaster that is happening there most of the time with any typical circuit that simply slaps 100nF at each opamp/component within a layout where the supply traces are long and inductive. Even worse is the staggered approach using smth. like 100nF in parallel with larger low ESR high capacitance film cap or one of those new ultra low-ESR polymer electrolytics. Anti-resonances peaking at 10's of ohms when the smaller C reacts with the ESL or phyical L!
I can only recommend to any designer: if you have money to spend on instrumentation don't aim for an audio analyzer (AP or what ever), rather buy a network analyser and learn how to use it.
The typical inverting I/V-stage draws high frequency pulse current from either supply, as much as the DAC chip can sink/source in any given moment which can be 10's of mA, way beyond the class-A range of the opamp's output stage, and that can severly exite those resonances more easily than in the class-A range because the former is half-wave rectified current with corresponding train of harmonics.
Obviously, local decoupling (in the true sense of the word, which means high series impedance and low parallel impedance) is an easy and effective way to avoid that and force the return current into the GND plane from where it can in turn return to the DAC chip's analog supply bypassing. The series impedance must be ohmic at the frequencies of interest and it doesn't need to be any higher than the value one would use for a parallel snubber, so 1...10Ohms works well. Higher, like 100Ohms doesn't hurt but requires larger parallel caps, otherwise the supply modulation can be larger than tolerable (depends on the PSRR of the opamp which will be good at DC/LF but poor at higher frequencies).
So to answer the OP's quiz: Replace the series resistor with a properly selected L//R, and use an equally well choosen cap. Eye-balling values is bound to fail, here. Get a network analyzer (and use its spectrum analyser features as well), invest in sniffer probes, etc. And sim things properly before building anything.
Last edited: