A VLSI architecture for digital filters using complex number-theoretic transforms

In this paper a parallel architecture is developed to realize a digital filter. First a systolic array is used to compute a 248-point complex number-theoretic transform (CNT). Next an algorithm is developed to realize a digital filter that uses 248-point CNT´s and a generalization of the overlap-save method. This algorithm solves the conflict between long transform lengths and a wide dynamic range associated with the number-theoretic transform. Finally this algorithm is mapped to a parallel architecture. This architecture is simple, regular and expandable, and, hence, is suitable for VLSI implementation.