Deep learning based non-line-of-sight identification with sub-1GHz narrow band frequency

초록

Identifying Non-Line-of-Sight (NLOS) conditions has brought new phenomena for indoor localization. NLOS Identification has increased the accuracy of indoor localization by either combining the PDR (Pedestrian Dead Reckoning) system with RSSI (Received Signal Strength Identification) or combining Wireless transmission based indoor localization with RSSI. Identifying NLOS condition techniques has been primarily investigated for wireless transmissions such as Ultra-Wide Band (UWB) and Wi-Fi transmissions. Nevertheless, in cases of emergency situations such as fire break out and power failure, these infrastructures are hard to be fully utilized. Moreover, the distance the signal can travel for UWB transmission is 5-10m and Wi-Fi transmission are able to travel up to 50m indoors. In this paper, we introduce a new and efficient methodology for indoor localization where there is no pre-installed infrastructure help such as emergency situations. We introduce the use of sub-1GHz wireless transmission for indoor localization. The sub-1Ghz are able to reach up to 200m indoors, and like other wireless transmission, to increase the accuracy of localization we have implemented NLOS identification on sub-1GHz wireless transmission. We have used deep learning for NLOS identification by measuring RSSI and using this dataset for deep neural network classification and was able to achieve 92.58 accuracy of NLOS identification. This proposed method will help enhance indoor localization during emergency situations and benefit from longer distance signal transmission. © International Research Publication House.

키워드