Abstract Due to its characteristics, the internal crystal oscillator of the clock tester blurs the boundary between abnormal and normal features, resulting in poor feature extraction accuracy. This leads to the inability of recognition methods to comprehensively and accurately capture key information about abnormal features, resulting in poor recognition performance. Therefore, an automated identification method for abnormal operation status of internal crystal oscillators in clock testers is proposed. A CNN network architecture is designed to achieve intelligent state discrimination through convolutional layer feature extraction, pooling layer dimensionality reduction, and a Softmax classifier. Confidence optimization algorithms are combined to improve recognition accuracy, ultimately achieving automated identification of abnormal crystal oscillator operation status. The experimental results show that the proposed method for automatically identifying abnormal operating states of the internal crystal oscillator of the clock tester is completely consistent with the actual operating state and can accurately identify abnormal feature points. The proposed method has a higher accuracy in identifying three different types of abnormal states: slow frequency drift, sudden frequency changes, and phase jitter anomalies, than the comparative method, with better recognition performance, stronger adaptability, and higher accuracy.
Liu et al. (Mon,) studied this question.