Robust Training for RSSI-based Localization

We consider the localization of multiple individual moving transmitters (targets), based on Gaussian process regression (GPR) using their corresponding received signal strength indicator (RSSI) data, collected by sensors distributed at fix locations within the environment. To this GPR-based RSSI localization problem, we contribute a novel training procedure to optimize the parameters of the GPR model via mini-batch stochastic gradient descent (SGD), with gradient expressions given in closed form. Thanks to the proposed training method, robustness to imperfect RSSI knowledge is added to the localization approach. Simulation results show that the proposed method significantly outperforms the best related state-of-the-art (SotA) alternative, validating the contribution.