Parts came in quickly, and I had everything on hand. SD card for scale.
I had to wire in a 0 ohm resistor, when I should have done that in the schematic. The AP2127's datasheet was very fuzzy on whether the enable pin was active high or active low. I left it unconnected. Turns out its high! Second design will of course include that.
On a previous batch of these, years ago, I noticed a really bad 1khz noise when recording. I think the external power supply, and single analog/digital ground merger helped fix the problem. I'm also bumped up to linux 4.9 on my desktop, so it could be driver-related. Whatever the case, it works! I'm going to make 4 of these to stress test my square inch USB hub. Supposedly, Texas Instruments uses some sort of data frame recovery for the USB clock, so even if your crystals drift out of unison, it should stay in phase. I plan on testing this with my oscope and puredata.
Taking notes from my previous attempt, hastily cobbled together:
It used a quad jeft op amp, a boost board for ~6v, then a 5v and 3v ldo for power rails. The 1khz noise persists. Looking at capacitance multipliers to filter out the rail noise, though I fear the noise is from an internal regulator.
I'm not sure why, but the noise source wasn't apparent when I was testing on a 4.x linux kernel. Only windows and 3.x kernels.