Technically, Nebula 2/3 do not use IR's or "dynamic convolution" as described (for example) by Kemp (USPTO 7,039,194). Rather, they use truncated Volterra series expansions, apparently cast as multidimensional convolution, which explains the exorbitant CPU usage for most models as well as possibly explaining the existing confusion over what is actually being done inside Nebula by Acustica Audio. A good background reference is for the use of truncated Volterra series expansions is: V. John Mathews and Giovanni L. Sicuranza, "Polynomial Signal Processing," Wiley, 2000, ISBN 0471-03414-2. Chapters 2-4 provide a fairly good picture of how to implement this kind of thing, while chapters 5 and 6 provide parameter estimation information (i.e. to help understand what NAT is doing). This is a graduate-level text, so beware.... Regards, Dave Clark