The Psychoacoustic Model (PAM) is a very important role in MPEG-2/4 Advanced Audio Coding (AAC) encoding. It determines sound quality of a given encoder and influences a lot in computational complexity. This paper presents a new architecture design for MDCT-based PAM co-processor. This work is based on the dedicated hardware design which is different from traditional programmable approaches. Moreover, to reduce the complexity, we replace the calculations of spreading function with reduced fixed coefficients, and decrease the transform kernels from three to one unit.