Fast convolution for binaural rendering based on HRTF spectrum

Abstract

This paper presents an efficient convolution technique in Modified Discrete Cosine Transform (MDCT) domain. It can be used to accomplish Head-Related Transform Function (HRTF) based binaural rendering. The MDCT and HRTF are firstly represented in a polyphase filter structure and further derived to a z-transform matrix representation. Based on the property of similarity between diagonal entries of filtering matrix and spectrum of HRTF, we presented a Dynamic Adjacent Elements Optimization (DAEO) for fast convolution of HRTF in the MDCT domain. Experimental results show that our algorithm greatly reduces the computational complexity, and maintains equal quality. The complexity analysis of virtual surround processing methods indicates that our filtering algorithm of HRTF consume minimal calculation, which is particularly suitable for audio compressed in MDCT format.