FPGA-Based artificial neural network driving a Stepper Motor

Abstract

This report describes the design and implementation of an FPGA-based Artificial Neural Network (ANN) for character recognition. The ANN algorithm is fully developed using VHDL in the structural modelling style. It comprises of 16 nodes in the input layer, 32 in the hidden layer and 16 in the output layer. The processing of data is done in the IEEE single precision floating-point format. In order to train the ANN, a dataset of 4×4 matrices stored in a VHDL file is used to represent the 16 letters to be recognized: A, C, D, F, H, I, J, L, N, O, P, T, U, X, Y, Z that are selected based on the feasibility of their representation in such dimensions. The weights are randomly initialized with a 16-bit Galois LFSR that has a maximum period of 65535, which are stored in an on-board SRAM unit of 2MB storage capacity. The DE2-115 board hardware platform is utilized to synthesize the overall system with the Quartus II software version 13.0. The built-in LCD display serves as an interface for the user to input the desired pattern on the two 4×4 grids, as well as to show the output class of the recognition process. We added a stepper motor circuit to test the working of the ANN with “ON” and “OF” patterns.