Noise in analog input conditioning

[bnahill]

A series of tests showed that a significant amount of input noise is coming from the analog conditioning.

The attached screenshot shows 3 distinct sets of samples. At the top (and consisting of a bunch of tests in slightly varying conditions) you can see normal samples with the microphone connected and isolated from acoustic noise sources. There is noise between 500 and 1000Hz. The series below that removes the microphone and shorts its terminals. Should that noise be at all related to the microphone or acoustic inputs, that noise should be eliminated. It isn't. To further verify, a final series at the bottom shows the result of shorting the terminals at the ADC.

Suggest working backwards through the input chain to identify the source of this. If it can't be identified, try to replicate with a different device.

[bnahill]

The above tests were with a microphone with a low battery. The blue signal below was collected in a relatively noisy environment with a single 30dB channel at 2.2kHz:

The green signal highlights a different shortcoming: apparent clipping at moderate levels. The green signal was recorded with a single 60dB channel. The below was captured by an alternate source during the recording

The signals never approached bounds of the ADC, yet the distortion is apparent.

[bnahill]

It looks like there's a decent chance the bug is in the digital parts again, which would be great.
A few options for next time re-visiting this:

1. Use current partial waveform dump capabilities to get a look at the raw data
2. Find a faster way (e.g., interrupt transfers) to dump raw waveforms to debug. This would be awesome for future stuff as well. The USB library must support this by now...
3. Trace the signals over the wire. This is probably the least work.