Implementation of integral transforms on the general purpose CNN neuroprocessor

The Cellular Neural Network Universal Machine (CNN-UM) is a novel neuroprocessor algorithmically programmable having real time and supercomputer power implemented in a single VLSI chip. The local CNN connectivity provides an useful computation paradigm when the problem can be reformulated as a well-defined task where the signal values are placed on a regular 2D grid (i.e., image processing), and the direct interaction between signal values are limited within a local neighborhood. This paper demonstrates how the CNN-UM architecture can be applied to perform global operations like Integral/Wavelet Transformations, in such a way that we can deliver this architecture from the use of alternative ones when nonlocal operations are needed. Lastly, examples are given to highlight the main steps of the method.