Reinforcement-Learning-Based Successive Approximation Algorithm

This paper presents a new approach to analog-to-digital converter (ADC) for low to medium-activity signals. We integrate the concept of reinforcement learning into the successive approximation register (SAR) ADC search methodology to reduce the comparator activity and switching energy of the digital-to-analog converter (DAC). This method is based on selecting the best solution among 8 available different solutions to digitize the first-order difference between an unknown sample and its previous digitized sample, in addition to the conventional method to digitize the sample amplitude itself. In this way, the number of comparisons needed can be smartly reduced. Our 10-bit SAR ADC simulation results show that the proposed method reduces comparator activity for low to medium-activity signals by up to 79.74%, while it operates the same as the conventional method for high-activity signals.