Derivation of dataflow networks for a domain specific applications

Application domain specific processors are becoming more and more appealing because of their peculiar feature of flexibility in terms of number of algorithms which one single architecture can execute and also in terms of the possible computational improvement due to the increase of parallelism. The specific domain chosen is the family of Jacobi algorithms which are often used in the field of adaptative beamforming of radar signals. The architecture chosen should be weakly programmable. The idea is to derive in a systematic way a control program to be downloaded on a simulator of the processor to obtain quantitative information about the performance on the execution of every algorithm. Moreover some techniques to improve, from a qualitative point of view, the parallelization of execution are applied, suggested by the particular structure of the Jacobi algorithms.