Estimating frequency characteristics of quantization noise for performance evaluation of fixed point systems

Abstract

Word-length optimization of signal processing algorithms is a necessary and crucial step for implementation. System level performance evaluation happens to be the most time consuming step during word-length optimization. Analytical techniques have been proposed as an alternative to simulation based approachto accelerate this step. The inability to handle all types of operators analytically and the increasing diversity and complexity of signal processing algorithms demand a mixed evaluation approach where both simulation and analytical techniques are used for performance evaluation of the whole system. The interoperability between simulation and analytical techniques requires study of noise sources and noise propagation characteristics. While the noise power and noise PDF have been studied, the output noise power distribution has not been studied. This paper addresses the problem of power spectral density estimation of the noise analytically. This paper also proposes to use the spectral density estimate for noise power calculation by having an approximate filter thereby accelerating the process of performance evaluation.