Generalized approach for the realization of discrete cosine transform using cyclic convolutions

The authors propose a general solution for the realization of the discrete cosine transform (DCT) and the inverse discrete cosine transform (IDCT). This algorithm converts the DCT/IDCT into convolution form such that one can easily implement it with technologies that are well suited for doing convolutions. This is an efficient and effective approach as it can avoid complicated data routing and data management. For example, there is no address generation problem and a simple pipeline structure can be applied to achieve parallel processing. Compared with conventional algorithms, this algorithm is much more flexible as it can be applied to realize DCT/IDCT with any length. This algorithm also suggests an efficient approach for realizing a unified hardware for the implementation of both the DCT and the IDCT. The realization for both the DCT and the IDCT can rely on the same convolution module.<>