Enhancing Prediction Entropy Estimation of RNG for On-the-Fly Test

Random number generators (RNGs) play a vital role in cryptographic applications, and ensuring the quality of the generated random numbers is crucial. At the same time, on-the-fly test plays an important role in cryptography because it is used to assess the quality of the sequences generated by entropy sources and to raise an alert when failures are detected. Moreover, environmental noise, changes in physical equipment, and other factors can introduce variations into the sequence, leading to time-varying sequences. This phenomenon is quite common in real-world scenarios, and it needs on-the-fly test. However, in terms of speed and accuracy, current methods based on mathematical formulas or deep learning algorithms for evaluating min-entropy both fail to meet the requirements of on-the-fly test. Therefore, this paper introduces a new estimator specifically designed for on-the-fly min-entropy estimation. To accurately evaluate time-varying data, we employ an appropriate change detection technology. Additionally, we introduce a new calculation method to replace the original global prediction probability calculation approach for accuracy. We evaluate the performance of our estimator using various kinds of simulated datasets, and compare our estimator with other estimators. The proposed estimator effectively meets the requirements of on-the-fly test.