Improving concurrency for cosine-modulated filterbank windowing

Abstract

A novel method exploits the time-reversal of mirror symmetric pairs of polyphase components of a linear-phase prototype filter to obtain more than 25% reduction in required MIPs to perform prototype windowing of either an analysis or synthesis cosine modulated filterbank on a suitable architecture. The architecture required to obtain this MIPs reduction is only a minor modification to many DSP architectures and is, in fact, available in the latest DSP designs. The proposed algorithm is particularly suited for large overlap factors, where the windowing computation dominates the required MIPs., one important example being the filterbank used in MPEG audio compression.