A Digital Hysteresis Current Controller for Three-Level Neural-Point-Clamped Inverter With Mixed-Levels and Prediction-Based Sampling

For a grid-tied three-level neural-point-clamped (NPC) inverter, fully digital hysteresis controller with simple variable hysteresis band is insufficient to stabilize switching frequency (f(s)). The challenge mainly comes from the restriction of analogue-to-digital converter (ADC) sampling frequency, which leads to inaccurate current-boundary-touching capture. As a result, f(s) jitter appears. This phenomena becomes much more serious around grid voltage zero-crossing points due to zero-approaching hysteresis band, where obvious f(s) variation happens. To stabilize f(s) in a fully digital manner, two techniques are proposed in this paper: the mixed-level scheme and the prediction-based sampling method. The former one is used to cope with the obvious f(s) variation around grid voltage zero-crossing points. It switches operating status from three-level state into two-level state around grid voltage zero-crossing points to enlarge the hysteresis band for digital implementation. The latter one is used to resolve f(s) jitter. It makes the sampling and switching to happen at the moment of current boundary-touching to achieve the most effective boundary-comparing with the lowest ADC sampling frequency. With these two techniques, f(s) of the grid-tied three-level NPC inverter could be well stabilized with a fully digital hysteresis controller, and therefore, high-quality grid-side could be achieved.