Efficient implementation of chaotic generator on FPGA using ANN

This paper presents a new method for an efficient implementation of chaotic systems, the Lorenz attractor, using Artificial Neural Networks (ANNs) on a Field Programmable Gate Array (FPGA). Chaotic systems are very complex, so we need to compute them as fast as possible and in real-time which requires a hardware program. Using ANNs to image chaotic behaviour is a credible method which offers an alluring balance between computational accuracy and resource efficiency. The solution uses a three-layer feedforward ANN which is trained with different hidden neuron configurations for optimal prediction performance. The training phase occurs within the MATLAB Neural Network Toolbox, with neuron counts ranging from one to sixteen, and utilizing the Levenberg-Marquardt, Scaled Conjugate Gradient, and Bayesian Regulation optimizations. After much testing, a less complex ANN which gives the same prediction can be implemented on hardware easier. Once training is completed, the optimized network is mapped to FPGA with Verilog and synthesized using the Xilinx Vivado tool chain targeting Xilinx Virtex 6 FPGA. Floating point arithmetic is used to improve accuracy without speed loss. The study states that ANN based FPGA based chaotic system modelling is fast and resource-effective and cost-effective as compared to traditional methods which is the significance of this research. They have a lot of implications for any secure communication or synchronization application requiring accurate, high-speed chaotic signal generation. This work shows that neural networks chaotic models can be implemented onto hardware which can be efficient when used in real-life application of a complex dynamic system.