7. Supervised Learning Architecture-Based L&T Using RSSI

As discussed in the previous chapters, being an interconnection of artificial neurons, the ANN is capable to mimic the behavior of biological neurons with the help of activation functions. The connections between neurons are through appropriate weights, which get automatically adjusted during the off-line ANN training step. This is called as supervised learning. In off-line training step, the training vector is necessary to train the ANN. This training vector includes a set of inputs and corresponding outputs. Once trained in off-line phase, the ANN becomes ready to estimate (predict) the system output for random input vector in the online phase. The GRNN is one of the important supervised learning architecture, which we discussed and applied to target L&T domain in the previous chapter. In this chapter, we are going to discuss the applications of other important supervised learning architectures such as feed-forward neural network (FFNT), radial basis function neural network (RBFNN), and multilayer perceptron (MLP). All of these architectures will be trained with RSSI measurements and the corresponding 2-D locations of the mobile target in off-line phase. Two cases are considered during experimentation. In Case I, the impact of various training functions on FFNT-based target L&T system is analyzed in the context of localization accuracy, whereas in Case II, the target localization performance comparison is made between traditional trilateration-, FFNT-, MLP-, GRNN-, and RBFN-based target L&T system. In order to accomplish fair comparison of all of the supervised learning architectures in both the cases, all of them are fed with four RSSI measurements, and they are supposed to estimate 2-D target location, corresponding to this input vector of four RSSI measurements. In both of the cases, locations of anchor nodes as well as target locations to be estimated are kept fixed.