Comparative Analysis of PWM And Neural Network-Based Modulation Techniques for a 5 Level CHB Inverter

Abstract

This paper compares the performance of conventional Pulse Width Modulation (PWM) with an optimised Neural Network (NN)-based modulation strategy for inverter control. Although PWM remains one of the most widely used modulation techniques because of its simplicity and reliability, its performance is inherently limited by fixed linear switching schemes. An NN-based controller offers a different approach by learning switching patterns from the system, enabling better overall performance. To investigate this, we developed and evaluated both modulation methods in the MATLAB/Simulink environment under identical nominal operating conditions. The results indicate that the proposed NN-based controller provides a higher output RMS voltage, improved output waveform quality and lower Total Harmonic Distortion (THD) than conventional PWM. Together, these improvements lead to a cleaner output waveform and more effective use of the available DC source, without requiring any hardware modifications. The present study is limited to simulation and does not account for practical hardware non-idealities. Even so, the results suggest that the neural network's learning capability can improve performance under nonlinear operating conditions. Future work will focus on real-time implementation using DSP/FPGA platforms, further optimisation of the learning algorithm, and exploring hybrid modulation techniques for multilevel inverters.