Real-time Model Predictive Control of a DC-DC Buck Converter

We propose a digitally assisted analog computing circuit for real-time model predictive control (MPC) of a DCDC buck converter. The computing circuit comprises analog elements for speed and digital components for programmability. It implements gradient-flow dynamics for a penalty-based reformulation of the quadratic optimization problem corresponding to the original MPC formulation. The MPC problem is set up to regulate output voltage while explicitly enforcing constraints on the buck converter's inductor current and duty cycle. Simulation results in a closed-loop configuration demonstrate superior dynamic response with lower settling time and overshoot compared to the Type-III controller and linear quadratic regulator (LQR). The proposed approach achieves accuracy similar to the numerically computed optimal solution from the interior-point-convex algorithm of MATLAB's quadprog solver while offering real-time implementation capability.