Look-up table and breakpoints determination for piecewise linear approximation functions using evolutionary computation

Piecewise linear approximation of nonlinear functions is a viable design approach for embedded systems without floating point capabilities. In this work, we propose a new procedure for determining the breakpoints positions and minimum memory necessary for solving this approximation problem, which meets the design's specification accuracy. The LUT memory size is minimized by using a top-down searching algorithm, which tries to find a solution for each pair of breakpoints storing resolution and number of breakpoints. An evolutionary algorithm is employed at the bottom level for searching the breakpoints positions, considering relevant practical issues. The proposed procedure optimizes the solution w.r.t the LUT size and breakpoints distribution.