MADI latencies check e.g. Octamic XTC manual, it's not much, and I think it's the same for all MADI devices, so also for 12Mic:

"When multiple devices are connected serially, the MADI I/O of each OctaMic XTC causes a delay of 3 samples. Therefore, at the MADI output of the last device, the data of all upstreamed devices are delayed. At Double Speed the delay rises to 6 samples per unit, at Quad Speed to 12 samples.".

I gues you confirmed it's going to work as I expected then.

However, there is no toslink IN on the 12 mic, so this won't work I'm afraid (unless I'm mistaking):
- 1 TOSLINK connection from recording interface to 12Mic#1

(- on 12Mic#1: - routing 2 phones channels (ADAT1 ch 1+2) to its own phones output
                       - routing 2 phones channels (ADAT1 ch 3+4) to MADI 1-2 OUT towards 12Mic#2
- on 12Mic#2:- routing 2 phones channels from MADI ch 1-2 to its own phones output)

And I still would like someone confirming that there will be no added latency using MADI this way, if it is not to much to ask.

Each MADI device has its own routing.

12Mic#2 routes its Mic inputs 1-12 to MADI 1-12
12Mic#1 routes its Mic inputs 1-12 to ADAT1 ch1-8 and ADAT2 ch 1-4
12Mic#1 route MADI1-12 (Mic inputs from 12Mic#2) to ADAT2 ch5-8 and ADAT3 1-8

If you want to use the phones outputs of the two 12Mic, then you need
- 1 TOSLINK connection from recording interface to 12Mic#1
- on 12Mic#1: - routing 2 phones channels (ADAT1 ch 1+2) to its own phones output
                       - routing 2 phones channels (ADAT1 ch 3+4) to MADI 1-2 OUT towards 12Mic#2
- on 12Mic#2:- routing 2 phones channels from MADI ch 1-2 to its own phones output

All devices have digital inputs which are in use, from where they can receive a clock signal either through an ADAT or MADI input. So any of the devices can be clock master (recording interface or one of the 12Mics).

I have a question about using two RME 12 mic" units, while outputting to ADAT @ 48khz, over 3 ADAT lightpipes, to a PCIe RME digiface (3 ADAT inputs), running at 64 or 128 samples buffersize.

To be used as HQ low latency mic inputs in a live environment.

So, is it possible to merge both units, 24 channels, to a total of 3 ADAT outputs @48khz (probably by using the build in MADI coaxial connection, and DSP mixer of the 12 mic's)?

I was thinking of a setup like this:
Where the first unit outputs it's mic channel 1-8 to ADATout1_channel1-8, it's mic channel 9-12 to ADATout2 _channel 1-4.

The second unit would have to send all it's channels to the first over MADI, where they would be routed to ADATout2, channel 5-8, and ADATout3, channel 1-8.
Other options would be to send channel 1-4 of the second unit over MADI to the first, and use one of the ADAT outs on the second unit for it's channels 5-12. Or any other order that would work.

I'm aware that there has to be one clock source in this setup.
I have no experience with MADI, or the 12 mic units (yet), hence the questions.

Also I would like to know if the total AD converter latency of the setup changes when using MADI like this.

As far as I know, ADAT and MADI both do not add extra latency, while for example a DANTE network does add a few msec's.

Thanks.