A Hybrid Architecture Based on Quantized and Physics-Informed Neural Network for High-Fidelity SHC-PWM

The practical implementation of advanced modulation strategies like Selective Harmonic Control (SHC-PWM) is hindered by the prohibitive memory usage when Look-Up Tables (LUTs) are implemented. Alternatively, neural networks have been proposed to emulate the LUT behavior, using less memory resources, but at the cost of an unavoidable precision loss and large computational execution time. This paper introduces a hybrid hardware-software architecture on an FPGA-SoC platform that overcomes both challenges. The architecture features a large Quantized Neural Network (QNN) on the FPGA, trained with Quantization-Aware Training (QAT) to replicate the LUT, and complementary it proposes a compact Physics-Informed refinement network running on the CPU to mitigate quantization error. Experimental validation of SHC-PWM using 1.1 million operating points and controlling seven harmonics demonstrates a memory compression ratio exceeding 200x compared to a LUT, a total inference latency as low as 73$\mu$s and a low absolute tracking error, with a maximum error below 0.002 p.u. in dynamic tests. The proposed end-to-end methodology provides a complete and validated solution for deploying high-performance modulators on resource-constrained hardware.