Encriptación simétrica de señales usando arquitecturas neuronales

Abstract

This project develops two encryption systems based on neural networks, the first from chaotic neural networks and the second an emulation of the DES (Data Encryption Standard) algorithm through a feedforward neural network. To develop of the chaotic neural network model, an analysis of the behavior that occurs when only a chaotic layer is found in the hidden layer was carried out, while the type of chaos and its parameters was varied. Then the results obtained were used, in order to find the values that the parameters of the chaotic signals can take so that the model presents good performance and thus measure the sensitivity of the model, to determine the space of keys for each of the different chaotic time series. Finally, and based on the results of the previous two points, an analysis of the behavior of the model was carried out with architectures of two, three and four chaotic layers in the hidden layer. The results show that as the number of chaotic layers increases performance improves. As for the model that emulates the DES algorithm, the model was generated from the original algorithm. Since no neural networks have been used before, it is desired to know if it is possible to emulate the behavior of this algorithm with a feedforward neural network. For this reason, the algorithm is used to generate a training database that will then be used to carry out the generation of the encryption model. Then the model training is carried out from the previously generated database. The results of the model achieve the proposed objective since they correctly encrypt the proposed test signals.