Function Approximation Using Polynomials

摘要

DSP designers sometimes encounter a situation, where they have to use some of the elementary functions (logarithm, square root, exponential, and trigonometric functions) on a fixed-point processor. Although optimized math libraries are widely available for floating point processors, this is not necessarily true for the fixed-point case, which means that designers must implement the elementary functions using homegrown algorithms. This article presents a very useful technique for designing algorithms for such implementations, describes range reduction methods to make the algorithms more accurate, and provides useful tips on using the proposed techniques in floating-point and fixed-point number systems.