Hardware-Efficient Decomposition Algorithm for IMDCT Computation

A new decomposition algorithm for inverse modified discrete cosine transform (IMDCT) computation is presented. The algorithm converses an N-point IMDCT to a pair of N/4-point type-IV discrete cosine transforms (DCT-IV/DCT-IV). Due to resource sharing, the implementation of DCT/DCT is hardware-efficient. Comparing with some well-known IMDCT algorithms, the proposed algorithm has higher computational efficiency (3 times higher) and requires 1 latch (20%), 4 adders (44%) and 3 multipliers (50%) less. To verify the proposed fast algorithm, a hardware accelerator based on the algorithm is designed and applied to the decoding of an AC-3 audio. The experimental results demonstrate that the AC-3 audio can be decoded in real time and therefore verify the practicability of the decomposition algorithm.