Comprehensive comparison of the least mean square algorithm and the fast deconvolution algorithm for crosstalk cancellation

Crosstalk cancellation is an important issue in 3D audio rendering, which aims at reproduce binaural signals at listener's ears using dual or multiple loudspeakers. Among many existing algorithms, the adaptive filtering in time domain and the direct deconvolution in frequency domain have been the predominant ones. Though they have been widely used, these two methods have never been compared comprehensively as we known. In this paper, we thoroughly examine their performances from different aspects, including channel separation, timbre distortion, and sensitivity to the HRIR measurement errors. Both objective and subjective evaluations are implemented in a practical environment. The experimental results may provide a more comprehensive reference for method choice in practical applications. Additionally, the limitation of using simulation to evaluate the performance of crosstalk cancellation is highlighted.