RSS-based Fingerprinting Localization with Artificial Neural Network

Radio Frequency (RF) based indoor is challenging due to the multipath effect in indoor signal propagation such as reflection, absorption, diffraction due to obstacles, interference and moving objects within the environments. The multipath effect phenomenon will be worsened if Received Signal Strength (RSS) is used as the localization measurement parameter. The advancement of Artificial Intelligence (AI) may hold the key for the improvement of RSS based localization. The Artificial Neural Network (ANN) in AI outperforms the traditional algorithms in indoor localization due to its capability to learn the unique features given in the training datasets. This paper discusses indoor fingerprinting localization with different architectures of ANN network to localize object of interest with the given indoor environment. The size of the experimental testbed is 14m x 8m and the testbed consists of four identical receivers that receive the signal from an active transmitter. All the tested ANN architectures have RSS as inputs and the Cartesian coordinates as outputs with different hidden layers and hidden nodes. The relationship between hidden nodes and layers in ANN and the regression losses is studied in this paper. The RSS-based fingerprinting with ANN in this paper is considered as a multi-output regression problem. The result shows that the ANN architecture with four layers with a total number of 800 hidden nodes has achieved an average of 6.01098 regression losses and a mean Euclidean Distance error of 2.54m.