Formulae and multiprocessor algorithms for digital signal microprocessors
Rabin Deka · OpenGrey (Institut de l'Information Scientifique et Technique) · 1992
A taxonomy proposed for Digital Signal Microprocessors (DSMs), in this thesis, includes: a method of classification; Scheduling Theorems with the associated corollaries; a definition of MAC-Operation and an expression for the computation rate, in terms of this Operation; a few formulae, enabling it to calculate the memory speed as well as the Wait-State(s) required for an engine based on DSM(s); and a defining rule for the operation of such an engine using a Slow-Clock. Some of the formulae, presented in the taxonomy, are used for developing two algorithms. The first is engine dependent; for the computation of a Discrete Fourier Transform (DFT) at high speed, abbreviated HDFT. It is based on the inherent parallelism property of the basic DFT definition in conjunction with three new hardware (DSM) characteristics. The engine modeled for the proposed HDFT algorithm is based on commercially available DSMs and named as Algorithm-Specific Connection Machine (ACM), which best suits its characteristics. The second algorithm assists the design of a uniprocessor engine based on either the TMS32020 or the TMS320C25. It also incorporates a method to interface the engine with the MC68000. The applications of the algorithm, therefore, provide useful information about the Interface-Timing required for the development of a uniprocessor, or multiprocessor, engine based on DSM(s) and (or) a general purpose microprocessor. Finally, a Data-Format, constituting the Adaptive Instruction Formats for the TMS320C25, has been developed and implemented with a two-pass cross-assembler algorithm. In this way, the use of a host computer, for developing software for general purpose signal microprocessing, using a TMS320C25 assembler, is demonstrated.