Access and alignment of arrays for a bidimensional parallel memory

Describes the use of a parallel memory system for a SIMD architecture for signal processing. This paper develops the Chinese linear skewing scheme in order: (1) to have conflict-free access to vectors of interest in signal processing; (2) to allow a simple computation of local addresses; and (3) to use 100% of the memory capacity. With a linear skewing scheme, the vectors fetched from the parallel memory belong to a class of vectors called p-ordered vectors. For an odd number of memory banks, we present a new multidimensional alignment network which is able to unscramble all p-ordered vectors and which has a topology that is easy to implement