An Improved Wi-Fi RSSI-Based Indoor Localization Approach Using Deep Randomized Neural Network

Indoor localization methods based on the Wi-Fi-received signal strength indicator (RSSI) ranging technology are sensitive to noise fluctuations and signal attenuations, which could lead to a significant localization error. Therefore, this study proposes an improved indoor localization algorithm using a deep randomized neural network (RandNN) with Wi-Fi-RSSI. We have conducted a real experiment testbed for Wi-Fi RSSI data collection from a complex indoor environment. An improved adaptive unscented Kalman filter (IAUKF) method is used to minimize noise fluctuations and signal attenuations in the raw Wi-Fi RSSI data collection. Moreover, we have investigated the deep RandNN in which the weights and biases of the input hyperparameter are initially randomized to obtain the best localization performance. For convenience, the presented localization model is known as RandNN-IAUKF. Furthermore, real experiments were conducted in a room surrounding working stations, walls, patriation separations, etc., to maximize the complexity of wireless signal propagation. The performance of the presented RandNN-IAUKF algorithm is assessed and compared with other well-known conventional localization approaches. Overall, the experimental results showed that the presented RandNN-IAUKF algorithm provides significant 95% and 67% location estimation errors only at 0.79 m and 1.31 m, respectively, outperforming conventional algorithms by approximately 30% under the same test environment.