Abstract
Quartz crystal resonators are a critical component in many electronic systems, providing the reference frequency source for the system’s clock. However, temperature often affect frequency stability. As a result, frequency-temperature (f-T) characteristic modeling has become an important area of research in frequency control. The traditional f-T modeling method omits system dynamics and can result in significant frequency compensation errors in the case of rapid temperature changes. To address this issue, this paper proposes a dynamic f-T modeling method with considering the thermal hysteresis. A dynamic f-T modeling method based on long short-term memory (LSTM) is presented to reflect the thermal hysteresis characteristics of quartz crystal resonators. Compared to traditional methods, LSTM are suitable for processing and predicting time-series data and consider past temperature history to make predictions. Additionally, transfer learning techniques are used during the training process of the model. Transfer learning fine-tunes the LSTM model for new/unknown crystal readout circuits using less data. Finally, the modeling and testing results on real experimental data show that the proposed method provides better frequency deviation predictions.
