A Fully Differential Multi-bit MDAC Modeling with Multiple Linear Regression Calibration

Abstract

In this paper, A fully differential multi-bit multiplying digital-to-analog converter (MDAC) model for pipeline analog-to-digital converter (ADC) is presented. The proposed model considers the coupling between capacitors in differential input end, as well as other non-ideal factors such as gain mismatch, input-referred noise, resulting in a more accurate representation of the actual circuit compared to existing models. Moreover, A multiple linear regression calibration is introduced to compensate for the nonidealities in MDAC. The calibration algorithm utilizes Mini-Batch Gradient Descent (MGD) and Heavy Ball Method (HBM) to mitigate the noise amplification and accelerate the convergence rate, respectively. The presented MDAC model and calibration algorithm are validated by Simulink and S-function.