Fixed-point error analysis and wordlength optimization of a distributed arithmetic based 8/spl times/8 2D-IDCT architecture

Abstract

The two dimensional discrete cosine transform (DCT) has been used widely for various image and video processing standards. Efficient implementation of the algorithm requires fixed-point arithmetic, which may result in a noticeable mismatch between the encoder and the decoder. The finite wordlength effects of a distributed arithmetic based 8/spl times/8 2D-IDCT (inverse discrete cosine transform) are analytically modeled. In order to accurately model the implementation hardware, the ensemble average of integer domain fixed-point errors after rounding is evaluated not only by calculating the mean and the variance but by considering the statistical distribution as well. Based on the error model, a set of optimum wordlengths conforming to the IEEE specifications is determined. There is a close agreement between the model and the bit-accurate simulation results.