Application of LSTM Network to Improve Indoor Positioning Accuracy

Area of Science:

• * Electrical Engineering
• * Computer Science
• * Robotics

Background:

• * Ultra-wideband (UWB) positioning is crucial for indoor navigation, but accuracy degrades significantly under non-line-of-sight (NLOS) conditions.
• * Traditional methods often reject NLOS signals, limiting positioning capabilities.
• * Maximizing the use of all available base stations is key to improving robustness.

Purpose of the Study:

• * To develop a novel method for enhancing UWB indoor positioning accuracy, particularly in challenging NLOS environments.
• * To leverage Long Short-Term Memory (LSTM) networks for processing Channel Impulse Response (CIR) data.
• * To improve ranging error prediction and subsequent positioning accuracy.

Main Methods:

• * Application of a Long Short-Term Memory (LSTM) network to raw Channel Impulse Response (CIR) data.
• * Utilizing the LSTM network to predict ranging errors.
• * Integration of predicted ranging errors into an improved positioning algorithm that incorporates data from all base stations.

Main Results:

• * The LSTM network achieved centimeter-level accuracy in predicting ranging errors.
• * The proposed positioning algorithm, incorporating predicted errors, demonstrated a significant improvement in accuracy.
• * Average positioning accuracy was enhanced by approximately 62% compared to baseline methods.

Conclusions:

• * LSTM networks effectively process CIR data for accurate ranging error prediction in UWB systems.
• * The proposed method significantly enhances indoor positioning accuracy, especially under NLOS conditions.
• * This approach offers a promising solution for reliable and precise indoor navigation.