On Wed, Feb 29, 2012 at 3:36 PM, Gregory Maxwell <gmaxw...@gmail.com> wrote: > On Wed, Feb 29, 2012 at 6:04 PM, Aaron Heller <hel...@ai.sri.com> wrote: >> 22 speakers, 3rd-order is 352 parameters so some strategy >> is needed to guide it. > > It wouldn't work for a 3D array of 22 speakers— but for an irregular > layout with the same number of speakers as a regular one (easy in 2D, > fewer choices in 3D): > > I'd add a single term to the optimization which controls a morphing > between your desired layout and a regular one. Then start the > optimizer with 1 in that position and the analytic solution for the > rest of the matrix. The geometry error term could then just be > included in the error function. A local optimizer should be able to > then walk to a reasonable solution (if not the best one). > > (similarly, you may be able to add or eliminate speakers from a > solution in a similar way, having a gain parameter for the extra > speakers— though I'm more sceptical of that working well— it's not as > obvious to me that the solutions should be all that smooth with extra > speakers with infinitesimal gain)
Good ideas. In fact, for the 3rd-order solution for the CCRMA array (352 parameters) I had to switch to a local optimizer, PRAXIS, with initial values from the pseudoinverse modified by the rE-max per order gains to get it to converge. This is what one would do for a regular array. 1st-order (88 parameters) was fine with a global optimizer (CRS). _______________________________________________ Sursound mailing list Sursound@music.vt.edu https://mail.music.vt.edu/mailman/listinfo/sursound
