Robustness of virtual artifcial head topologies with respect to microphone positioning

A multimicrophone array is presented that can be used to approximate the frequency dependent directional characteristics of an arti cial head. The desired HRTFs can be realised by a set of appropriate lters. Such a setup may be referred to as a virtual arti cial head. Virtual arti cial heads are much more exible than real arti cial heads, since, e.g., the lters can be adjusted to match an individual set of HRTFs. However, virtual arti cial heads are sensitive to small errors in the characteristics and the position of the individual microphones. In the present work, the relevance of di erent microphone topologies and the robustness with respect to positioning errors of the microphones is investigated. First, a method for optimizing the microphone positions based on a Golomb array topology, which originates from number-theoretic considerations, is introduced. The method successively computes a set of microphone positions with the possibility to vary the number of microphones without changing the general topology. Second, the robustness against positioning errors is improved by applying a white noise gain regularisation constraint for the computation of the lter coe cients. It is shown by numerical simulations, for a two dimensional array, that both procedures considerably improve the robustness.