FPGA implementation of a neural network classifier for gas sensor array applications

A primitive gas recognition system which can discriminate limited species of industrial gas was designed and simulated. The dasiaelectronic nosepsila consists of an array of 8 micro-hotplate based SnO₂ thin film gas sensors with different selectivity patterns, signal collecting unit and a signal pattern recognition and decision part in programmable logic device chip. BP (back propagation) neural networks with Multilayer Perceptron structure was designed and implemented on FPGA (field programmable gate array), of twenty thousand gate level chip by VHDL language for processing the input signals from 8 kinds of gas sensors. The network contained eight input units, one hidden layer with 4 neurons and output with 5 regular neurons. The dasiaelectronic nosepsila system successfully discriminated 5 kinds of industrial gases in computer simulation. A small application has been tested on the APS X208 FPGA test board.