Iterative DFT-Domain Inverse Filter Determination for Adaptive Listening Room Equalization

Abstract:
Listening room equalization (LRE) is used to remove the unwanted linear distortion of a reproduced acoustic scene by the listening room. This task comprises the identification of the loudspeakerenclosure- microphone system (LEMS) and the determination of the pre-equalizing filters for the identified LEMS. For the latter subtask, filtered-X adaptive filtering algorithms allow an iterative optimization of the equalizers, whereby approximations do only affect the convergence speed, but not the optimality of the determined equalizer coefficients. Alternative approaches determine the equalizers by directly solving a system of linear equations, where the obtained equalizer coefficients may be suboptimal due to approximations which potentially facilitate an online implementation for MIMO LEMSs. In this paper, we present an alternative algorithm which allows an iterative optimization of the obtained equalizers, without any filtering of the excitation signals for the adaptation algorithm. Similar to the filtered-X generalized frequency-domain adaptive filtering (FxGFDAF), this approach uses a DFT-domain approximation with low computational cost. The evaluation results show an improved convergence behavior compared to the FxGFDAF. Index Terms — adaptive filtering, listening room equalization, inverse filtering, filtered-X, DFT domain