Multi-level fingerprinting approach for wireless localization management using CNN and OFDM

Multi-level fingerprinting approach for wireless localization management using CNN and OFDM

연구 분야: Infrastructure

학회: International Journal of Information Technology

With the increasing demand for indoor positioning solutions that require no additional equipment and the ease of obtaining channel state information (CSI), research on indoor fingerprinting using CSI has gained significant attention. The given paper introduces a novel multi-level fingerprinting approach for enhancing accuracy of wireless systems. The approach comprises deep learning layer and filtering layer. The former makes use convolutional neural network (CNN) while the latter makes use of an optimal sub-carrier filtering method namely orthogonal frequency division multiplexing (OFDM) to illustrate the process of conversion of serial data stream into parallel streams. CNN include initialization of weights randomly, calculate predictions via forward propagation, and optimize weights using backpropagation to minimize mean absolute percentage error (MAPE). The proposed approach also performs localization and loss function process. During localization process, CNN processes the feature maps computed during training phase to produce probabilities for each access point (AP). The approach is validated and compared with conventional works based on MAPE and positioning accuracy. The results show superior performance by achieving the highest positioning accuracy (96.7%) and the lowest MAPE (4.24%) thus outperforming existing techniques on basis of received signal strength (RSS).