That is a good point, ejeffrey. It is not for an industrial monitor, it is a device mainly for research and development, hence the rather strange (and maybe somewhat vague) requirements. It partially works as a power metering device, hence the higher resolution requirement and higher sampling speed. Current is measured using a LEM sensor (HTFS-200 likely), but that is easy to implement and not much to improve there, apart from using a compensated LEM instead of the drifty plain hall effect based one.
I know the Silabs' cheap PWM modulator appnote, but this rather too complex solution offers rather poor performance, but sure, for low price. I still prefer to use analog optocouplers like IL300 or LOC110, as they deliver predictable performance, but still not enough for over 12b of measurement resolution.
Those isolated S-D modulators are interesting indeed, I have forgot these exist. Decoding the pulse density modulation into PCM is of no issue, I could use a MCU with a dedicated peripheral to do so. However the AMC1304x still needs an external isolated DC supply. The resulting price of not being any lower than the other solution using a digital isolator with integrated DCDC and a separate ADC.
So it seems that I am stuck again at what was the original solution to the problem: An ADUM5401 and a separate ADC, I think the original design used LTC2412. But I consider that to be way too much overkill (24bit dual channel SDADC with PGA, reference and what not) and will at least downgrade that to a rather more reasonable ADS8689 (being it the cheapest SPI ADC with 16bit resolution on Mouser), which should be more than enough for the application.
I thought that I could optimize the design a bit. But well, whatever.