The LM48901 is now discontinued, not even having a page on TI's website anymore, and was at least NRND when I initially specified it. It's still available from distributors in sufficient quantities for building a prototype or a few, but I'd rather use something that can go into production without a complete redesign. Additionally, the LM48901's programming tool doesn't seem to be online anymore either, and I was never sure whether it could even converge to a point several kilometers away. So I should probably switch to using some kind of microcontroller or FPGA for each speaker group. (I still think it's best to use this architecture rather than one central processor sending signals to each speaker directly, though I'm not entirely sure. Maybe a 3-level tree where nodes at each level do 4-speaker/4-sub-node beamforming?) This means more beamforming work for me, but it should also be far more flexible and capable of things like adaptive acoustics.