EMI Terminal Behavioral Modeling of SiC-based Power Converters

Abstract

With GaN and SiC switching devices becoming more commercially available, higher switching frequency is being applied to achieve higher efficiency and power density in power converters. However, electro-magnetic interference (EMI) becomes a more severe problem as a result. In this thesis, the switching frequency effect on conducted EMI noise is assessed. As EMI noise increases, the EMI filter plays a more important role in a power converter. As a result, an effective EMI modeling technique of the power converter system is required in order to find an optimized size and effective EMI filter. The frequency-domain model is verified to be an efficient and easy model to explore the EMI noise generation and propagation in the system. Of the various models, the unterminated behavioral model can simultaneously predict CM input and output noise of an inverter, and the prediction falls in line with the measurement around 10 MHz or higher. The DM terminated behavioral model can predict the DM input or output noise of the motor drive higher than 20 MHz. These two models are easy to extract and have high prediction capabilities; this is verified on a 10 kHz-switching-frequency Si motor drive. It is worthwhile to explore the prediction capability of the two models when they are applied to a SiC-based power inverter with switching frequency ranges from 20 kHz to 70 kHz. In this thesis, the CM unterminated behavioral model is first applied to the SiC power inverter, and results show that the model prediction capability is limited by the noise floor of the oscilloscope measurement. The proposed segmented-frequency-range measurement is developed and verified to be a good solution to the noise floor. With the improved impedance fixtures, the prediction from CM model matches the measurement to 30 MHz. To predict the DM input and output noise of the SiC inverter, the DM terminated behavioral model can be used under the condition that the CM and DM noise are decoupled. With the system noise analysis, the DM output side is verified to be independent of the CM noise and input side. The DM terminated behavioral model is extracted at the inverter output and predicts the DM output noise up to 30 MHz after solving the noise floor and DM choke saturation problem. At the DM input side, the CM and DM are seen to be coupled with each other. It is found experimentally that the mixture of the CM and DM noise results from the asymmetric impedance of the system. The mixed mode terminated behavioral model is proposed to predict the DM noise when a mixed CM effect exists. The model can capture the DM noise up to to 30 MHz when the impedance between the inverter to CM ground is not balanced. The issue often happens in extraction of the model impedance and is solved by the curving-fitting optimization described in the thesis. This thesis ends with a summary of contributions, limitations, and some future research directions.