[Paper Review] On Time-of-Arrival Estimation in NB-IoT Systems

Sha Hu, Xuhong Li+1 more|arXiv (Cornell University)|Nov 10, 2017

Indoor and Outdoor Localization Technologies8 references3 citations

TL;DR

This paper proposes a SAGE-based time-of-arrival (ToA) estimation algorithm for NB-IoT systems that jointly estimates multipath delays and channel amplitudes using normalized auto-correlation and cross-correlation metrics. The method iteratively refines estimates by suppressing interference from previously detected paths, achieving robust ToA estimation even in low SNR conditions.

ABSTRACT

We consider time-of-arrival (ToA) estimation of a first arrival-path for a device working in narrowband Internet-of-Things (NB-IoT) systems. Due to a limited 180 KHz bandwidth used in NB-IoT, the time-domain auto-correlation function (ACF) of transmitted NB positioning reference signal (NPRS) has a wide main lobe. Without considering that, the performance of ToA estimation can be degraded for two reasons. Firstly, under multiple-path channel environments, the NPRS corresponding to different received paths are superimposed on each other, and so are the cross-correlations corresponding to them. Secondly, the measured peak-to-average-power-ratio (PAPR) used for detecting the presence of NPRS is inaccurate. Therefore, in this paper we propose a space-alternating generalized expectation-maximization (SAGE) based method to jointly estimate the number of channel taps, the channel coefficients and the corresponding delays in NB-IoT systems, with considering the imperfect ACF of NPRS. Such a proposed method only uses the time-domain cross-correlations between the received signal and the transmitted NPRS, and has a low complexity. We show through simulations that, the ToA estimation of the proposed method performs close to the maximum likelihood (ML) estimation for a single-path channel, and significantly outperforms a traditional ToA estimator that uses signal-to-noise (SNR) or power thresholds based estimation.

Motivation & Objective

• To improve time-of-arrival estimation accuracy in NB-IoT systems degraded by multipath propagation.
• To address the challenge of joint delay and amplitude estimation in low-SNR, low-complexity NB-IoT receiver designs.
• To develop an iterative algorithm that enhances path detection by suppressing interference from already estimated multipath components.
• To provide a robust initial estimate of path delays and channel gains when no prior knowledge is available.

Proposed method

• Initial delay estimates are formed by sorting the magnitude of cross-correlation R[d] and selecting the L largest values as initial path delays d̃ℓ.
• The algorithm uses normalized auto-correlation γ(d) to model the channel impulse response and suppresses interference from detected paths.
• At each SAGE iteration, the residual correlation R̃[d] is updated by subtracting contributions from previously detected paths using estimated amplitudes h̃ℓ.
• Amplitude estimates are refined using a correlation-based averaging process over a local window, followed by a noise-level-informed scaling to reduce estimation error.
• The noise variance estimate σ̃² is computed from the average power of the residual correlation to guide amplitude normalization.
• The algorithm iteratively updates path delays and amplitudes until convergence, with path-specific updates performed in a sequential manner.

Experimental results

Research questions

• RQ1 How can joint time-of-arrival and channel amplitude estimation be improved in NB-IoT systems with limited training sequences?
• RQ2 What is the impact of interference from strong multipath components on ToA estimation accuracy in low SNR regimes?
• RQ3 Can iterative interference cancellation using SAGE-like updates improve delay estimation performance compared to non-iterative methods?
• RQ4 How effective is the use of cross-correlation and auto-correlation metrics in initializing and refining path estimates without prior knowledge?
• RQ5 What role does noise variance estimation play in stabilizing amplitude estimates during iterative ToA refinement?

Key findings

• The algorithm achieves accurate initial path delay estimation by selecting the L largest magnitude values of the cross-correlation R[d], forming a reliable starting point for iteration.
• Iterative refinement significantly improves delay and amplitude estimation by suppressing interference from previously detected paths.
• The noise variance estimate σ̃² is computed from the residual correlation power and used to normalize amplitude estimates, reducing error in low SNR conditions.
• The amplitude estimation step uses a local averaging process over a symmetric window around the peak, improving robustness to noise.
• The final amplitude estimate is further refined by a gain correction factor that depends on the estimated noise power and signal power, enhancing accuracy.
• The algorithm converges through M iterations, with each path updated sequentially, ensuring stable and accurate estimation of both delays and amplitudes.