Efficient implementation of a class of MDCT/IMDCT filterbanks for speech and audio coding applications

In this paper, an efficient algorithm for implementing MDCT/IMDCT of lengths N = 5 ldr 2^m (m ges 2) is presented. Transforms for such lengths are of interest for speech and audio coding applications, such as recently issued and/or emerging standards G.729.1, G.EV-VBR, and EVRC-WB. In our design we utilize a mapping of MDCT of size N into JV/2-point DCT-IV and DCT-II with isolated pre- multiplications, which are subsequently moved in the windowing stage. We show that such a modified window is piece-wise symmetric, and can be stored using N/2 words. In our algorithm we also use an efficient factorization of 5- point DCT-II which requires only 4 multiplications by irrational factors. We compare our proposed algorithm with several alternative implementations and show that our design offers practically appreciable reduction in complexity and memory usage.