Super-Resolution Time-of-Arrival Estimation using Neural Networks

Abstract

This paper presents a learning-based algorithm that estimates the time of arrival (ToA) of radio frequency (RF) signals from channel frequency response (CFR) measurements for wireless localization applications. A generator neural network is proposed to enhance the effective bandwidth of the narrowband CFR measurement and to produce a high-resolution estimation of channel impulse response (CIR). In addition, two regressor neural networks are introduced to perform a two-step coarsefine ToA estimation based on the enhanced CIR. For simulated channels, the proposed method achieves 9% – 58% improved root mean squared error (RMSE) for distance ranging and up to 22% improved false detection rate compared with conventional super-resolution algorithms. For real-world measured channels, the proposed method exhibits an improvement of 1.3m in distance error at 90 percentile.