| The speed of fuzzy controllers implemented on dedicated hardware is adequate for control of any physical process, but too slow for today's high-complexity data networks. Defuzzification has been the bottleneck for fast implementations due to the large number of computationally expensive multiplication and division operations. In this thesis, we propose a high-speed fuzzy inferential system based on log-domain arithmetic, which only requires addition and subtraction operations. The system is implemented on a Xilinx Virtex-II FPGA with a processing speed of 67.6 MFLIPS having a maximum combinational path delay of 4.2 ns. It is a clear speedup compared to the reported fastest 50 MFLIPS implementation. A pipelined version of the controller is also implemented, which achieves a speed of 248.7 MFLIPS. Although a small approximation error is introduced, software simulation and hardware implementation on FPGA confirm high similarity of the outputs for control surfaces and a number of second-order plants. |
