An Efficient Common-Pole Filter Design Algorithm using Balanced Model Truncation for Binaural Audio Synthesis

摘要

In this paper, an algorithm is proposed to model multiple-input single-output (MISO) system of FIR filters using 1IR filters that share a common set of poles based on balanced model approximation technique. While transforming the state-space of a system to its balanced form the approximation process often suffers from numerical inaccuracies due to the inversion of ill-conditioned balancing transformation matrices. In this paper, instead of balancing the state-space explicitly, a reduced-order model is directly formulated that is input/output equivalent to the reduced-order balanced realization of the system. The proposed algorithm constructs the loworder approximated model with significantly lower computational complexity compared to the general balanced model approximation technique. Hence the technique is suitable to model a system of large number of FIR filters. Though the algorithm is a general one that can be applied to various signal processing applications, it is particularly suitable for designing virtual audio reality systems where a large set of approximated Head-related transfer functions (HRTFs) is used for immersive rendering of a set of sound sources.