Research On GCN-TCN Based Indoor WiFi Fingerprinting Algorithm

With the increasing requirements for location-based services,the higher requirements for accuracy and real-time indoor localization accuracy have promoted the research of indoor localization technologies.In particular,fingerprint indoor localization technology based on Wi Fi devices,which makes full use of the available infrastructure and avoids non-line-of-sight interventions,has made it one of the main focuses of indoor localization research.At present,fingerprint positioning systems based on Channel State Information(CSI)are less affected by non-line-of-sight and multipath effects,and have more stable signal characteristics.At the same time,the development of deep learning has provided new research directions for indoor fingerprint localization technology.We proposed focuses on the CSI fingerprint localization algorithm based on deep learning algorithms.In this thesis,CSI data is first generated using ray-tracing method simulation based on indoor environment,and then the fingerprint data is pre-processed.A multi-scale principal component analysis algorithm is used for data reduced dimensionality,and high-dimensional CSI data is planned to a low-dimensional space,reducing computational complexity while ensuring localization accuracy.Then,this paper proposes a localization algorithm based on the spatial location of indoor Wi Fi geometric position relationship of Wi Fi transmitters,using Graph Convolution Neural Network(GCN)to extract the spatial features of the received signal.In the indoor localization process,each Wi Fi transmitter is regarded as a node in the graph neural network,and the edge connections in the network are constructed in terms of the location deployment relationship between Wi Fi devices.The spatial features of the indoor environment can be fully explored through the GCN network structure,which can then achieve high-precision and highly reliable indoor localization.Results of simulation shows that the proposed fingerprint localization algorithm has excellent indoor localization accuracy.Finally,due to the fact that the wireless signals are propagated through different paths to the receiver device,the receiving wireless signals are temporal correlated.In this paper,we propose a localization network based on spatial-temporal correlation:Graph Convolution Network-Temporal Convolution Network(GCN-TCN)for indoor localization,based on the GCN-based indoor localization algorithm.By introducing the TCN network structure,the temporal features of fingerprint data are extracted.The GCN network layer extracts the spatial features of the indoor environment and the TCN network layer extracts the temporal features of the fingerprint data.The extraction of data features both in space and time is implemented,which leads to improved indoor localization accuracy.Simulation results show that the localization accuracy of the scheme has been improved obviously compared with the traditional deep learning localization algorithm.