Application of LSTMs Recurrent Neural Networks to Characterize a Digitizer Using a Quantum Voltage Standard

This work studies an application of Long-Short Term Memory (LSTM) recurrent neural networks to characterize a digitizer using measurements performed with a quantum voltage standard based on the Josephson effect. Digitizers are fundamental instruments in electrical measurements, particularly in quantum-based metrology applications, where accuracy is important. Exploiting the LSTM's capability to learn temporal dependencies in sequential data, a methodology for characterizing a digitizer's performance is proposed. By training classification LSTM models using quantum voltage standard measurements, patterns inherent to the system behavior, like the transient that appears due to the quantum voltage steps transitions, can be detected. After eliminating these transient points, a fitting can be made using weighted least squares to calculate the digitizer's gain and offset. The results show the potential of advanced machine-learning techniques to optimize measurements in quantum-based systems.