Acceleration of Monte-Carlo simulation-based quantization of DSP systems

Simulation-based methods are the most common approach to estimate the quantization noise of fixed-point implementations of Digital Signal Processing (DSP) systems. But as systems grow in complexity, solving the wordlength optimization problem using this approach leads to unaffordable execution times. Thus, it is necessary to develop new strategies that allow to obtain optimized solutions in reasonable times. In this paper, a new simulation-based technique based on the relaxation of the confidence requirements of the simulations in some stages of the evaluation is presented. Experimental results show that this process provides optimized results between 2 and 120 times faster than traditional methods with no penalty on the quality of the solutions.