Optimized Design of Σ Δ $\Sigma \Delta$ Modulators Using Deep-Learning and Simulated Annealing

Abstract

This letter presents a MATLAB toolbox for the automated high-level design and optimization of analogue-to-digital converters (ADCs), using sigma-delta modulators ($\Sigma \Delta \mathrm{Ms}$) as case studies. The tool combines machine learning (ML) techniques and behavioural simulation to obtain the optimum set of building-block (amplifiers, comparators, etc.) requirements for a given set of specifications, namely resolution, signal bandwidth and power consumption. Two machine learning blocksgradient boosting classifiers and regression-type artificial neural networks—are trained, using Python libraries, to identify the best ADC architecture as well as to infer a set of design parameters which yields ADC specifications. The result from the ML blocks can be cross-checked in behavioural simulations in MATLAB and also optimized with respect to signal-to-noise ratio (SNR), power consumption, or figure-of-merit (FoM) using an embedded simulated annealing (SA) process. The toolbox is controlled through a graphical user interface (GUI) for MATLAB which guides the designer through the whole process, from specifications to obtaining an implementation that meets the required specifications.