Trigonometric function approximation neural network based coprocessor

Abstract

Both conventional desktop and embedded processors rely on lookup tables (LUT) and iterative interpolation/regression methods to evaluate trigonometric functions. Neural networks provide a possible medium for the development of function approximations. Typically embedded processors cannot afford the luxury of large LUTs and lack fast interpolation hardware. A neural network which performs function approximations is implemented here in hardware as a configurable coprocessor to augment an existing general purpose processor.