Indoor Position Recognition and Interference Classification with a Nested LSTM Network

Indoor position information plays an important role in a wide variety of Internet of Things (IoT) applications. In recent studies, Wi-Fi channel state information (CSI) is becoming a popular tool for indoor positioning and environmental sensing due to the richer information it contains in multiple subcarriers. Deep neural networks (DNNs) and convolutional neural networks (CNNs) have been applied in training CSI data for indoor positioning and are shown to have fine performances. However, the potential sequential correlations among the subcarriers and consecutive CSI packets are yet to be exploited in those networks. In this study, we propose a two-level nested long short-term memory (LSTM) network named IPRIC to recognize the indoor positions and classify the type of interferences of Wi-Fi CSI data by analyzing the sequential correlations among the subcarriers and consecutive packets. We apply the discrete wavelet transform (DWT) for data preprocessing to minimize noise and improve classification. As evaluated in an office environment, IPRIC achieves higher classification accuracy than the DNN and CNN-based methods.