As one of my half-finished DAC projects involves a very large number of chips (72 minimum) I wanted a way to test individual DAC chips by listening before incorporating them into this design. So I designed and built a test board which accepted a single chip in a ZIF socket. I was shocked that this had more air and ambience apparent than the original PhiDAC (which also had a single DAC chip). My curiosity was piqued as previously I'd associated more ambience with more paralleled DAC chips. That is I thought I was lowering DAC noise by putting more chips in parallel, but this experimental single chip DAC threw a wrench in that hypothesis.
Turns out that the I/V stage noise is far more critical than I previously believed - it needs to be better than the DAC chip's noise by a significant margin to release the full amount of recorded bloom. In the original PhiDAC I am using AD8017 preceded by a very low impedance passive filter (~33ohm). Its this combination of low impedance with the AD8017 which contributes noise and masks the low-level ambience. AD8017 being a current-feedback opamp has significant current noise at its -ve input and that is associated with a not-very low noise corner. AD829 as on PhiDAC hex does quite a lot better - similar voltage noise but lower current noise which allows for a higher impedance passive filter. Seems with 6 chips and AD829 that 47ohm is a bit on the low impedance side, 100ohm is closer to optimum. Of course there's a trade-off in going higher impedance - more voltage swing at the DAC chip output pins. 6mA into 100ohm is 600mV but the filter 'rings up' near the corner frequency resulting in about 5dB peaking so worst case just over 1V swing. TDA1387 lacks an AC compliance spec (TDA1543 gives 25mV) - its best to keep this as low as possible, how low really can only be determined by listening. When I went up to a 240ohm filter the SQ wasn't as good so 100ohm is about the sweetspot with AD829.
An alternative chip for I/V is LT1028 which has >6dB noise improvement over AD829 in its voltage noise and so is better suited to the existing 47ohm filters. Changing the filter impedance is rather inconvenient since I have already a large stock of inductors for the existing design and would need to commission more values. Thus changing to LT1028 is the best way forward for now - the fact that it doesn't have the OPS stage bypass matters much less than the improvement gained from lower noise.