PARALLEL PROCESSING ARCHITECTURES FOR ADVANCED SIGNAL-PROCESSING

This paper describes parallel processing architectures for advanced signal processing. Advanced architectures are those that make use of nonlinear functions such as reciprocal, square root, arctangent etc. To accomplish this, a nonlinear cell has been devised which performs any one of four such nonlinear functions in two clock cycles. The paper focuses on typical advanced algorithms encountered in signal processing, such as Givens rotation, L-U decomposition, Cholesky update and direct Cholesky factorization. The use of our high-speed multi-function nonlinear cell can lead to a several-fold higher throughput over the customary software based approaches.