Efficient IMDCT core designs for audio signal processing

Filter bank processing techniques based on MDCT/IMDCT (modified DCT/inverse modified DCT) have been widely adopted in various audio codec standards. Most published IMDCT computing algorithms focus mainly on either the reduction of computing complexity but overlook the hardware realization issues, e.g. memory access complexity and efficient mapping of the computing kernel. By exploiting the symmetric properties in computation, we first convert an N-point IMDCT to an N/2-point DCT-II problem. A fast DCT-II computing scheme is next derived and the overall scheme is further optimized to remove redundancy. Based on the proposed fast IMDCT computing scheme, a novel design mapping is developed to minimize memory access complexity without stalling the pipelined computation. The mapping features simple address generation, small temporary storage size and low access bandwidth. Performance analyses show that, given the same hardware resource allocation, the proposed design can outperform other well known IMDCT designs in terms of memory storage size, computing latency or fixed point implementation error.