Power and Area Efficient Intelligent Hardware Design for Water Quality Applications

Abstract

The paper presents a power efficient and computationally less intensive intelligent hardware using artificial neural network for water quality applications. A compact Hardware Neural Network algorithm has been developed that takes four water quality parameters as the input vector and perform classification of the parameters using a Multilayer Perceptron Network. The computational complexity in the implementation of logistic function has been reduced at a mathematical level by use of approximation methods such as Pad===?=== approximation for exponential function and non- linear approximation for sigmoid function. The network improves accuracy of the output by learning by back-propagation of the error. Results show that non-linear approximation method is 34.13 % power efficient and utilizes 15.53 % less number of hardware resources in comparison to Pad===?===. ASIC implementation is compact and has 99 % less power consumption as compared to FPGA implementation of the same algorithm.