Statistical analysis as a quantitative basis for DSP architecture design

The increased demand for efficient application-specific DSP platforms encourages the utilization of extensive statistical analysis of the applications as a quantitative basis for the design of the architecture of these platforms. The paper presents the methods used for the definition of two generations of DSP cores based on both subjective and objective statistics. The subjective statistics consist of a detailed programmers' questionnaire, whereas the objective statistics are measurements of real-life software of specific applications. The analysis of these measurements led to the design of the first generation PINE DSP core. Later, similar objective and subjective statistics were gathered from the actual implementation of a few applications on PINE. Investigation of the post-design measurements confirmed the main conclusions reached in the design phase of PINE. These measurements also served for designing the second generation, high-end, OAK DSP core.