A Preliminary Study of Machine-Learning-Based Ranging with LTE Channel Impulse Response in Multipath Environment

TL;DR

This paper explores a machine learning method using LTE channel impulse response data to estimate the distance between a base station and a ground vehicle, offering an alternative to GPS in urban multipath environments.

Contribution

It introduces a CNN-based ranging approach utilizing LTE CIR data extracted via SDR, demonstrating improved accuracy over RSSI-based methods in field tests.

Findings

CNN-based ranging outperforms RSSI-based methods

LTE CIR can be effectively used for UGV positioning

Field tests confirm improved accuracy

Abstract

Alternative navigation technology to global navigation satellite systems (GNSSs) is required for unmanned ground vehicles (UGVs) in multipath environments (such as urban areas). In urban areas, long-term evolution (LTE) signals can be received ubiquitously at high power without any additional infrastructure. We present a machine learning approach to estimate the range between the LTE base station and UGV based on the LTE channel impulse response (CIR). The CIR, which includes information of signal attenuation from the channel, was extracted from the LTE physical layer using a software-defined radio (SDR). We designed a convolutional neural network (CNN) that estimates ranges with the CIR as input. The proposed method demonstrated better ranging performance than a received signal strength indicator (RSSI)-based method during our field test.