Lightweight Prediction-Based Tests for On-Line Min-Entropy Estimation

Health tests (on-the-fly tests) play an important role in true random number generators because they are used to assess the quality of the bits produced by entropy source and to raise an alert when failures or attacks are detected. Most of classical tests are implemented as statistical tests. A set of new health tests based on predictors was presented by National Institute of Standards and Technology in CHES 2015. These off-line tests attempt to predict the next output of the entropy source by trying to learn the patterns that the previously produced sequence of bits may possess. We provide the first integrated lightweight implementation of prediction-based tests for min-entropy estimation and verify their validity.