If we were going for a 24 bit design and had +/-12V supplies readily available the THAT1512 chip is an excellent match. I'm very familiar with it. Its EIN (equivalent input noise) of 4.6 nV/root(Hz) @ 20dB gain is top class.
If you're intending to use the Teensy Audio Library, then with 16 bit resolution, the theoretical maximum is 96 dB S/N and with practical limitations and headroom, this comes down to 80-85dB practical S/N. There is a good discussion on this at
http://openaudio.blogspot.com/2017/03/teensy-audio-board-self-noise.html
I believe the TS472 will deliver that kind of performance, as its stated EIN is 10 nV/root(Hz), but I'll do some measurements on one to see how it performs in practice. It's only 6dB below the THAT chip's performance, and the loss is significantly less than is given away by the 24/16 bit sample depth reduction (48dB theoretical, practically more like 20dB if you assume the bottom 4 bits is mainly noise!).
Having said all that, I'm happy to design with the THAT chip if that's your desire, it will mean adding a +/- 12V (or even +/- 5V) supply. We'll need to be careful that we don't exceed the USB's 5v supply capability. The chips themselves don't consume much, but the switching supplies can have significant no-load current. I'll check.
I agree that more filtering of the 48v supply is warranted. BTW, my current phantom input configuration is based on Fig 5 from the THAT1512 datasheet.
GroupDIY tends to be a mixture of very sage advice, and folks who are keen to try out all kinds of interesting ideas to get the last iota of performance out of their designs. Capacitance multipliers are an aspect of analogue regulator design, which don't apply here. With 400kHz switching supplies, most of the noise is well-outside the audible range, and aliasing with the sampling multiplier (in this case 128Fc ~= 6 MHz) is more of a concern. Also the CMRR of the preamp (typically 60-80dB), kills any remaining output noise from phantom power.
We already have RC filters after the 48V converter, however I'll change R19/20 to ferrite beads, upgrading the existing RC pi configuration.
In all the technical discussions, we haven't any clarified non-technical specifications.
1. What's your price point expectation? I have been assuming a low-cost design. Say $20-30 for components on top of the board fabrication (say $10). If you're looking at professional assembly, that adds around $30/board for short runs (depending on complexity).
2. What level of assembly capability do we want to require? I have been assuming that we are restricting the design to simple assembly (passive components and modules, but not multi-pin SMD packages), so that others without extensive facilities (e.g. reflow ovens, and SMD rework stations) can complete the design.
3. What are the physical specifications and constraints - board size, panel layout etc?