摘要
In this paper, a robust binaural audio rendering system is proposed on the basis of a time-domain underdetermined multichannel inverse filtering approach. The celebrated multiple-input/output inverse theorem is reformulated into a general multichannel model-matching problem with the emphasis on binaural audio reproduction. Robustness with widened sweet spots is achieved by selecting multiple control points in the reproduction zones. The model-matching problem is formulated in the time domain as an underdetermined system, where the number of channels is selected in relation to the number of virtual sources and control points. Under the full-rank condition, exact solutions of inverse filters always exist to fulfill the ideal model-matching criterion. However, the gains of prefilters need to be limited in the design stage by using the Tikhonov regularization at a minor expense of matching performance. The proposed binaural audio system has been implemented on a six-element linear loudspeaker array. Three problems of binaural rendering, cross talk cancellation, source widening, and 5.1 virtual surround, are adopted to validate the proposed approach. Results of objective and subjective tests have demonstrated the efficacy of the proposed approach for binaural audio rendering.