Pseudo Random Number Generator Based on Chaos Theory and Artificial Neural Networks

In recent years, developing new methods for designing Pseudo-Random Number Generators (PRNGs) has attracted the attention of researchers for their efficiency in many crypto-graphic applications such as communication and data security. In this work, an innovative hybrid design of a PRNG based on Artificial Neural Networks (ANNs) and chaos theory is proposed. The training data for fitting the proposed ANN model was generated from the numerical solution of the system of differential equations driving the dynamics of the Lorenz system. After that, this model is evaluated using the mean square error (MSE) metric. The work’s major contributions are focused on enhancing the MSE and expanding the keyspace in order to better prevent brute force attacks. In comparison to prior research, the suggested ANN model improved efficiency the MSE criteria, by reaching a value of $2.3751\times10^{-07}$, and increased the keyspace dimension by a factor up to 3 folds. Moreover, the statistical performances of the proposed PRNG have been evaluated. Generated sequences have been successfully passed the NIST SP 800-22 tests battery. Comparing to traditional chaos-based PRNGs designing methods, these results confirm that the ANN-based approach is a suitable and efficient alternative for developing future PRNGs dedicated to cryptographic applications.