Grounded ADC pins on the CS4272-CZZR producing horrible noise?

Question

I am currently busy with an audio processing project that requires 4 output channels. I have a Nucleo F303RE running some digital filtering on audio data, which is being read by one of the two CS4272s in the system. Both are used as outputs. So one of the codecs have both differential input pairs being unused.

At the moment, one codec has one pair used, one pair grounded and the other codec has both ADC pairs grounded. The codec with one input pair grounded is working decently well, there is some static present but I assumed that my amplifier design was responsible for that.

The codec with both pairs grounded, however, produces either a loud whine/buzz sound when both DACs are unmuted or a horrible screeching sound during powerup/initialization when just one of the DACs is muted. The audio data does play, but the amount of noise is unacceptable. The PCB also draws up to 200 mA at idle, with only two codecs and the nucleo being switched on. These problems happen regardless of whether the ADC channels are muted or not. Volume control register settings also don't significantly affect the noise, though it is hard to tell.

While designing the circuit, I couldn't find what to do with unused inputs (CS4272 datasheet doesn't mention it afaik) so grounding them seemed to be the best option at the time. I've recently found datasheets from Cirrus Logic for similar ADC devices which explicitly say to leave unused analog inputs floating. I was wondering if the grounded ADCs could be responsible for the bad noise on one of the devices, and if so, is it also affecting the performance of the other one?

Answer 1

If you read the datasheet on page 32, you can see that the inputs are NOT intended to swing down to 0V. So quite likely you are upsetting the bias of the input stage by tying them to ground and this is causing problems.

I would implement the recommended dual opamp design (again page 32) and simply leave the inputs floating. It seems that the Vcom pin sets the correct bias voltage with no signal (note the circuit is DC coupled from Vcom to the inputs). The circuit given also provides a known input impedance to the inputs which might be important.

If you didn't use this circuit on your first prototype, or you want to avoide the extra opamp stages, you should start by shorting the diff inputs together (or using a low value resistor) and you could experiment with tying them to Vcom (perhaps again by a small resistor).