Digital ANFIS Architecture and Performance Analysis for Nonllinear Systems

Neuro Fuzzy systems, nowadays drawing attention of many researchers since being able to carry both neural networks and fuzzy logic's benefits. Adaptive Neuro Fuzzy Inference System (ANFIS) is such a neuro fuzzy architecture which has been widely accepted since invented. It is used in different applications like universal approximator, non-linear system realization, pattern recognition etc. However, because of variety of applications, implementation of ANFIS has been turn out to be specific and same implementation barely utilized with another one. FPGAs are potential enough to bring flexibility in hardware implementation of ANFIS so as to make it generic and application-independent. In this paper, an ANFIS model designing and implementation on FPGA is described. A unique dynamic ANFIS structure is realized with VHDL which is independent of system to be realized, membership function type used. It can be easily configurable with order of ANFIS, number of inputs/outputs, and number of membership functions. The study results illustrate that ANFIS presented in paper has been successfully realized with test nonlinear functions. Evaluations using standard error measurements revealed closer approximation of digital ANFIS to software one. The MATLAB simulation and FPGA implementation results clearly indicate that the hardware ANFIS model is acceptable in calibrate mode.