Skip to Main Content
This paper investigates architectures for digital signal processors (DSP's). Typical DSP's are examined to identify their computation and communication requirements. By decomposing the general filtering function into primitive second-order digital recursive filter (2DRF) modules, an architecture is developed which contains the basic operations of every signal processor. This leads to the design of a novel, programmable arithmetic unit for high speed sum-of-product (SOP) computation which possesses hardware speed and software flexibility. For spectral decomposition of speech, a number of DRF modules must operate simultaneously. A set of MC68000 microcomputers is used to identify the interprocessor communication requirements for modules acting as a digital filter bank (DFB). A versatile bus structure permits the investigation of various data transfer and control strategies for different systems. This leads to the design of a VLSI DFB for real-time speech processing.