A New Key Generation Technique based on Neural Networks for Lightweight Block Ciphers

In recent years, small computing devices used in wireless sensors, radio frequency identification (RFID) tags, Internet of Things (IoT) are increasing rapidly. However, the resources and capabilities of these devices are limited. Conventional encryption ciphers are computationally expensive and not suitable for lightweight devices. Hence, research in lightweight ciphers is important. In this paper, a new key scheduling technique based on neural network (NN) is introduced for lightweight block ciphers. The proposed NN approach is based on a multilayer feedforward neural network with a single hidden layer with the concept of nonlinear activation function to satisfy the Shannon confusion properties. It is shown here that NN consisting of 4 input, 4 hidden, and 4 output neurons is the best in key scheduling process. With this architecture, 5 unique keys are generated from 64 bit input data. Nonlinear bit shuffling is applied to create enough diffusion. The 4-4-4 NN approach generates the se-cure keys with an avalanche effect of more than 50 percent and consumes less power and memory, thus ensuring better performance than that of the existing algorithms. In our experiments, the memory usage and execution cycle of the NN key scheduling technique are evaluated on the fair evaluation of lightweight cryptographic systems (FELICS) tool that runs on the Linux operating system. The proposed NN approach is also implemented using MATLAB 2021a to test the key sensitivity by the histogram and correlation graphs of several encrypted and decrypted images. Results also show that compared to the existing algorithms, the proposed NN-cipher algorithm has lower number of execution cycles and hence less power consumption.

@article{Rana2021,
  title     = {A New Key Generation Technique based on Neural Networks for Lightweight Block Ciphers},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {12},
  number    = {6},
  year      = {2021},
  publisher = {The Science and Information Organization},
  author    = {Sohel Rana and M. Rubaiyat Hossain Mondal and A. H. M. Shahariar Parvez},
  doi       = {10.14569/IJACSA.2021.0120623},
  url       = {https://doi.org/10.14569/IJACSA.2021.0120623}
}