RNN-based Distance Estimation using UWB Signaling
Abstract
Background/Objectives: The objective of this paper is to develop a new distance estimation method for indoor localization by using ultra-wideband (UWB) signals.The new technique is based on recurrent neural network (RNN), one of the famous deep learning techniques.
Methods/Statistical analysis: RNN is one of the most suitable methods for learning time series data. Hence, it is useful in processing data that change with time. The proposed method estimates the distance based on RNN from the received UWB signals. Specifically, from the received signals via IEEE802.15.4a indoor channels, the proposed RNN regressor estimates the distance. The proposed method is validated by computer simulation.
Findings: In order to estimate the distance using UWB signals, it is necessary to accurately detect the first arrived signal along the shortest path.To find this signal or time instance by using RNN, we convert 1-dimensional received signal into 2-dimensional signal. The converted signal is input to RNN regressor and trained so that the RNN output is the distance between the transmitter and the receiver. The conventional method needs received signal-to-noise ratio (SNR)estimation, and the threshold is determined by the estimated SNR. When the received signal exceeds the threshold, the first arrived signal is detected and the arrived time is called the time of arrival (ToA). However, the proposed method estimates the distance directly from the received signal without SNR estimation.According to the simulation results, the proposed method shows RMSE of less than 2 [m] in low SNR, and less than 0.5 [m] in high SNR.Those performancesare much better than the conventional method.
Improvements/Applications: The performance of the proposed RNN based distance estimator is examinedthrough computer simulation. To compare estimation accuracy, the root mean square error (RMSE) is measured. According to the results, the proposed estimator is superior to the conventional method.
Keywords:distance estimation, indoor localization, ultra-wideband (UWB), recurrent neural network (RNN), regression, root mean square error (RMSE)
