Deep Neural Network for Indoor Positioning Based on Channel Impulse Response

Fingerprinting positioning aided by wireless technologies plays an important role in a variety of industrial applications, such as factory automation, warehouse automation, and underground mining, where guaranteeing a position prediction error smaller than a threshold value is necessary to meet certain functional requirements. In this paper, we firstly design a deep convolutional neural network that uses the channel impulse response measurement as an input parameter to predict the position of a mobile robot. Second, we propose a simulated annealing algorithm that finds a minimum number of access points with their respective optimal positions that satisfies an expected average distance error in terms of a mobile robot's predicted position. The obtained results show that the average distance error is significantly reduced, e.g., by half compared to the case without optimal positions of access points.