A combined artificial neural network and DSP approach to the implementation of space vector modulation techniques

Space vector modulation (SVM) in three-phase voltage source and current source converters has become the preferred PWM method for digital implementations. This paper presents an alternative SVM implementation that is based on a neural network structure. The technique reduces hardware and software complexity, and computation time, and increases the accuracy of the positioning of the switching instants. The technique exhibits the following features: (a) possibility of higher switching frequencies, (b) higher bandwidth of the control loops, (c) reduced hardware and software, and (d) reduction of parasitic harmonics in all PWM waveforms. The proposed method is compared to conventional implementations of SVM techniques in terms of hardware/software requirements, switching frequencies, harmonic spectra, and computation times. The method is applied to a 2 kVA unit and experimental results confirm theoretical and simulation results