Duty-cycle detection for noisy arbitrary waveforms using artificial neural networks

A novel technique for detecting duty cycles of noisy arbitrary waveforms by employing artificial neural networks trained with empirical moments of asynchronously sampled waveforms' amplitudes is presented. The proposed technique is software-based and hence can be applied flexibly to numerous waveform types without requiring any hardware changes. Furthermore, in contrast to existing duty-cycle detection methods, this technique is capable of detecting duty cycles of noisy waveforms. Results obtained for square and sawtooth waveforms demonstrate wide duty-cycle detection range of 10-95% with mean absolute percentage errors < 0.4%.