On the Steady-State Behavior of Finite-Control-Set MPC with an Application to High-Precision Power Amplifiers

Duo Xu, Sander Damsma, Mircea Lazar

Abstract

Motivated by increasing precision requirements for switched power amplifiers, this paper addresses the problem of model predictive control (MPC) design for discrete-time linear systems with a finite control set (FCS). Typically, existing solutions for FCS-MPC penalize the output tracking error and the control input rate of change, which can lead to arbitrary switching among the available discrete control inputs and unpredictable steady-state behavior. To improve the steady-state behavior of FCS-MPC, in this paper we design a cost function that penalizes the tracking error with respect to a state and input steady-state limit cycle. We prove that if a suitable terminal cost is added to the FCS-MPC algorithm convergence to the limit cycle is ensured. The developed methodology is validated in direct switching control of a power amplifier for high-precision motion systems, where it significantly improves the steady-state output current ripple.

Introduction

In high-precision applications such as semiconductor lithography, power amplifiers must deliver extremely stable currents. Finite-Control-Set Model Predictive Control (FCS-MPC) is frequently employed for these systems because it controls switches directly, bypassing the need for a pulse-width modulator (PWM). This allows for faster response times and more effective handling of system constraints.

However, standard FCS-MPC faces a significant challenge in steady-state operation. Because it attempts to track a constant reference using a discrete set of switch positions, the controller often exhibits arbitrary switching or "chatter." This leads to unpredictable steady-state behavior and high current ripple—a critical issue for high-precision motion systems. Essentially, the controller fights the inherent physics of the switched system rather than exploiting its periodic nature.

Methodology

The authors propose a fundamental shift in perspective: instead of tracking a static reference point that the system can never truly maintain, the controller should track the best possible periodic path, known as a limit cycle.

System Topology

The proposed method is validated using a model of an industrial high-precision power amplifier. The circuit topology, featuring an H-bridge configuration and output filters, is shown in Figure 1.

Circuit diagram of the high-precision power amplifier topology.

Limit Cycle Tracking Design

The control strategy involves three primary components:

  1. Limit Cycle Computation: An optimal $p$-periodic sequence of inputs and states is pre-computed. This cycle represents the switching sequence that stays closest to the desired reference on average with minimal ripple.
  2. Time-Varying Cost Function: The MPC cost function is redesigned to penalize deviations from the specific point on the pre-computed limit cycle corresponding to the current time step.
  3. Stability Guarantee: To ensure the system settles into this cycle, a terminal cost is added to the optimization. The authors prove that if the cost matrix is derived from a specific Linear Matrix Inequality (LMI), the system is guaranteed to converge to the limit cycle asymptotically.

Results and Discussion

The methodology was evaluated through simulations of a 6A step response. The results demonstrate that the state trajectories successfully converge to the target limit cycle, as illustrated in the phase plane analysis.

State trajectory converging to the optimal limit cycle in the phase plane.

Performance Improvements

  • Ripple Reduction: Standard FCS-MPC produced a steady-state current ripple of approximately 17.88 mA. The proposed limit cycle tracking reduced this to just 4.21 mA when using a prediction horizon of $N=8$.
  • Switching Predictability: The switching pattern became a clean, repeating sequence (e.g., {3, 2, 3, 1, 1, 1}), matching the optimal theoretical limit cycle and eliminating arbitrary chatter.
  • Horizon Sensitivity: The study highlights that while standard MPC might function with short horizons, limit cycle tracking benefits significantly from longer horizons (e.g., $N=8$), which allow the controller to anticipate the full periodic path.

Conclusion

By redefining the control objective as tracking an optimal limit cycle rather than a static point, this research provides a robust framework for improving the steady-state performance of FCS-MPC in high-precision power amplifiers. The inclusion of a terminal cost provides the necessary mathematical rigor for stability, bridging the gap between control theory and industrial application. Future work will focus on real-time implementation on FPGA platforms to meet the microsecond-scale sampling requirements of modern power electronics.

Summary updated

Abstract

Motivated by increasing precision requirements for switched power amplifiers, this paper addresses the problem of model predictive control (MPC) design for discrete-time linear systems with a finite control set (FCS). Typically, existing solutions for FCS-MPC penalize the output tracking error and the control input rate of change, which can lead to arbitrary switching among the available discrete control inputs and unpredictable steady-state behavior. To improve the steady-state behavior of FCS-MPC, in this paper we design a cost function that penalizes the tracking error with respect to a state and input steady-state limit cycle. We prove that if a suitable terminal cost is added to the FCS-MPC algorithm convergence to the limit cycle is ensured. The developed methodology is validated in direct switching control of a power amplifier for high-precision motion systems, where it significantly improves the steady-state output current ripple.
BibTeX
@misc{xu2022steady,
  author = {Duo Xu and Sander Damsma and Mircea Lazar},
  title = {{On the Steady-State Behavior of Finite-Control-Set MPC with an Application to High-Precision Power Amplifiers}},
  year = {2022},
  eprint = {2205.15668},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.15668},
}
Figure