A new LoRa based positioning algorithm utilizing sequence based deep learning technique

• मुख्य लेखक: Suseenthiran, Kavetha
• स्वरूप: थीसिस
• भाषा: अंग्रेज़ी; अंग्रेज़ी
• प्रकाशित: 2023
• ऑनलाइन पहुंच: http://eprints.utem.edu.my/id/eprint/28326/1/A%20new%20LoRa%20based%20positioning%20algorithm%20utilizing%20sequence%20based%20deep%20learning%20technique.pdf; http://eprints.utem.edu.my/id/eprint/28326/2/A%20new%20LoRa%20based%20positioning%20algorithm%20utilizing%20sequence%20based%20deep%20learning%20technique.pdf

Positioning systems can be utilized both indoors and outdoors, however their precision varies since the environment seems to have a significant influence on localization. There are positioning system for respective environments for example, at outdoor GPS is used whereas for indoor positioning Wi-Fi and BLE are used but there is no positioning system that can be adaptive to different types of environments which leads to huge positioning error. LoRa positioning has good performance in terms of accuracy however the positioning error is high due to the Received Signal Strength Indicator (RSSI) heavy fluctuations and the selection of the parameters depending on different the type of environment. In this research, an adaptive LoRa based Positioning system is developed which consists of LoRa Transmitters and LoRa Receiver. Next, RSSI and Signal-to-Noise Ratio (SNR) that is measured is being classified whether it is LoS or NLoS environment based on the sequence-based Bi-LSTM model. Furthermore, an analysis of classification using different sequence length is done. Then, a new positioning algorithm is developed which incorporates distance estimation, Kalman Filter and trilateration technique according to the classification with different sequence length data and the positioning error is being analysed. It is concluded, having a sequence length of 100 dataset gives 100% accuracy due to the length is shorter, it is faster to be trained. The CDF gives 90% of positioning error less than 2.9m in LoS scenario whereas NLoS scenario is less than 2.41m. Comparing with the traditional trilateration method, the proposed algorithm gives higher positioning accuracy in which the estimated positions are near to the actual position. Proposed method improves the positioning error up to 28.92% for the LoS scenario meanwhile for the NLoS scenario the positioning error is improved by 32.68%. Meanwhile when the user moves from NLoS to LoS environment, the positioning error was improved to 72.16% whereas when it is was from LoS to NLoS environment, the accuracy improved 99.81%.