UWB Indoor Positioning System Based on NLOS Recognition

In the context of indoor positioning, the utilization of Ultra-wideband (UWB) technology is extensive. However, when obstacles are present within the environment, the accuracy of target localization can be affected by non-line-of-sight (NLOS) measurements obtained from UWB. To address this issue, this paper proposes a reversible transformation method (RTM) to denoise the impulse response (CIR) data from UWB channels to improve the signal-to-noise ratio (SNR) value of the CIR dataset, and a composite network consisting of convolutional neural network (CNN) and long short-term memory (LSTM) is used to classify the NLOS/LOS signals, and the output of this model is the probability that the range measurement value as the probability of LOS, and then localized by a weighted least squares (WLS) method. The network model is first trained using all the inputs extracted from the office dataset, and then the trained model is used in subsequent localization tests in a different office environment. The experimental results show that the proposed algorithm in this paper can well identify and mitigate the localization errors caused by NLOS.