Artificial Neural Network based Direct Inverse Control for a Novel 48V-1V DC/DC Converter

This paper presents an Artificial Neural-Network (ANN) based controller for a high gain step-down converter. The proposed topology adheres to a typical Point of Load (PoL) application with two active switches and minimal number of passive components. This resonant converter-based topology portrays a non-linear gain function, leading to a very high step-down factor without pushing the duty ratios to the extremums. In addition to that, so as to get superior dynamic response, an ANN based Direct Inverse Control (DIC) technique is implemented that models the inverse response of the system to generate PWM signals for maintaining a constant output voltage while rejecting load/line disturbances. This model performs offline system characterization through a training set of large number of operating points for accurate identification for plants that are analytically complex to be characterized. For concept verification, a MATLAB-based simulation of the proposed ANN-DIC control scheme for the 48V to 1V converter topology is conducted and the followings have been validated: (a) a well-regulated output voltage with ±1% ripple, (b) undershoot/overshoot of the output voltage within a band of ±3% subjected to 100% load transient as per the PoL application requirements and (c) a settling time less than 2.5ms at rated load.