⚡ electrical engineering

In-Tunnel Single-Anchor Localization Exploiting Near-Field and Radio-Reflective Road Markings

This paper presents JAVELIN, a novel single-anchor localization framework for GNSS-denied tunnels that combines near-field propagation modeling, adaptive processing, and passive radio-reflective road markings to achieve accurate vehicular positioning without prior environmental knowledge.

Original authors: Lorenzo Italiano, Mattia Brambilla, Monica Nicoli

Published 2026-04-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Lorenzo Italiano, Mattia Brambilla, Monica Nicoli

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Problem: Getting Lost in a Concrete Tube

Imagine you are driving your car through a long, dark tunnel. You are a "smart car" (a Connected and Automated Vehicle) that relies on GPS to know exactly where it is. But as soon as you enter the tunnel, the GPS signal vanishes because the sky is blocked by concrete. You are now driving blind, relying only on your speedometer, which is like trying to drive a car in the dark by only guessing how fast you are going.

This is a huge problem for the future of self-driving cars. If the car doesn't know exactly where it is, it can't stay in its lane or avoid accidents.

The Old Way vs. The New Way

The Old Way (Multi-Anchor):
Usually, to fix this, engineers would install dozens of expensive "beacons" (like Wi-Fi routers or 5G towers) all along the tunnel walls. The car would talk to all of them to triangulate its position.

The Flaw: This is incredibly expensive and hard to build. It's like trying to light up a whole cave by installing a lightbulb every few feet.

The New Way (Single-Anchor + JAVELIN):
This paper proposes a smarter, cheaper solution called JAVELIN. Instead of needing a forest of towers, you only need one 5G base station (the "Anchor") at the entrance or middle of the tunnel.

But how does one tower know where the car is if the signal bounces off the walls?

The Analogy: Imagine you are in a large, empty cathedral and you shout. The sound bounces off the walls and comes back to you. By listening to how the echo sounds (does it sound high or low? does it come from the left or right?), a super-smart computer can figure out exactly where you are standing, even if you can't see the walls.

The Secret Sauce: Three Magic Tricks

The JAVELIN system uses three clever tricks to make this "echo location" work perfectly:

1. The "Curved Wave" Trick (Near-Field)

Usually, radio waves are treated like straight laser beams (Far-Field). But when you are close to the antenna in a tunnel, the waves look more like ripples in a pond spreading out from a stone (Near-Field).

The Metaphor: Think of a flashlight. Far away, the beam looks like a straight cylinder. But right in front of the bulb, the light spreads out in a cone.
The Innovation: The researchers realized that because the waves are "curved" in the tunnel, the antenna can measure the curvature of the wave. This gives it a 3D map of where the signal came from, not just a flat direction. It's like being able to tell how far away a sound is just by how "round" the echo feels.

2. The "Magic Mirror" Trick (Radio-Reflective Markings)

Tunnels are often empty, which makes echoes weak. To fix this, the paper suggests painting special Radio-Reflective Road Markings (RRMs) on the floor and walls.

The Metaphor: These aren't just paint; they are like tiny, invisible mirrors for radio waves. Just as a human driver uses painted lane lines to see where to drive in the fog, the car uses these "radio mirrors" to bounce signals back to the antenna.
The Result: These passive mirrors act as extra "ghost towers" without needing any electricity or cables. They make the system much more accurate.

3. The "Smart Detective" Trick (Tracking)

Signals in a tunnel bounce around wildly. Sometimes the car sees a reflection from a wall, sometimes from a ceiling, sometimes from a truck passing by.

The Metaphor: Imagine a detective trying to follow a suspect in a crowded market. The suspect (the car) is moving, and there are many people (reflections) who look similar.
The Innovation: The JAVELIN system uses a "Variable-Dimension Kalman Filter." That's a fancy way of saying it's a smart detective that:
- Keeps a running list of all the possible "ghosts" (reflections) it sees.
- Decides which ghosts are real walls and which are just noise.
- Updates its guess about the car's location every millisecond, smoothing out the bumps and jumps.

The Results: How Good Is It?

The researchers tested this in a computer simulation of a realistic tunnel.

Accuracy: They achieved centimeter-level accuracy (about the width of a finger) when the signal was clear. Even when the signal was blocked (No Line-of-Sight), they stayed within one meter of accuracy.
Comparison: The old methods (like TeNFiLoc) got lost easily when the signal bounced too much, often being off by several meters. JAVELIN stayed on track.
The "No Paint" Test: When they removed the special radio-reflective road markings, the accuracy dropped significantly. This proves that these cheap, passive mirrors are essential for the system to work well.

Why Should We Care?

This research is a game-changer for the future of transportation because:

It's Cheap: You don't need to build a massive network of towers. One antenna and some special paint on the road are enough.
It's Safe: Self-driving cars can drive safely through tunnels, underground parking garages, and dense city canyons where GPS fails.
It's Scalable: Since the "mirrors" (RRMs) are passive, they don't need power. You can add them to existing roads easily.

In a nutshell: The paper teaches us how to turn a single 5G tower and some special road paint into a super-accurate GPS system that works even when the sky is blocked, ensuring our self-driving cars never get lost in the dark.

1. Problem Statement

Accurate vehicular localization is critical for Cooperative Intelligent Transport Systems (C-ITS) and automated driving. However, Global Navigation Satellite Systems (GNSS) fail in signal-blocked environments like road tunnels, leading to service outages or large positioning errors.

Current Limitations: Existing solutions often rely on dense infrastructure (e.g., multiple synchronized anchors for UWB or TDoA), which is costly and complex. Single-anchor approaches exist but often suffer from:
- Reliance on Line-of-Sight (LoS) components, which are frequently blocked in tunnels.
- Dependence on prior knowledge of the environment (digital maps) or reflector locations.
- Inability to effectively utilize Near-Field (NF) propagation characteristics, which contain geometric information (curvature) absent in Far-Field (FF) models.
- Lack of robustness in dynamic scenarios where multipath components change rapidly.

The paper addresses the challenge of achieving robust, high-accuracy, single-anchor localization in tunnel environments without requiring synchronized multiple anchors, digital maps, or prior knowledge of reflector positions.

2. Methodology: The JAVELIN Framework

The authors propose JAVELIN (Joint and Adaptive Virtual and Ego-user Localization In Near-field), a comprehensive framework that integrates theoretical channel modeling with recursive Bayesian tracking.

A. Theoretical Foundation: SR-NF Validity Condition

The paper first derives a geometric validity condition for the Single-Reflector Near-Field (SR-NF) channel model.

Concept: In NF propagation, wavefronts are spherical. The SR-NF model assumes that all multipath components arriving at an antenna array originate from a single specular reflection point on a planar surface (e.g., a tunnel wall).
Theorem: The authors prove that this single-reflector assumption is only valid if the antenna array size ( $M$ ) does not exceed a specific threshold ( $M_{max}$ ) relative to the propagation distance ( $R$ ) and wavelength ( $\lambda$ ).
Scaling Law: The bound follows the classical Fresnel-region scaling ( $M_{max} \propto \sqrt{R/\lambda}$ ). If the array exceeds this size, the channel must be modeled as multiple Virtual Anchors (VAs) rather than a single reflector. This provides a design guideline for array sizing in tunnel deployments.

B. Signal Processing and Parameter Estimation

Tensor Decomposition: The system uses a Single-Input Multiple-Output (SIMO) OFDM system. The channel response is modeled as a tensor and decomposed using Canonical Polyadic Decomposition (CPD).
Parameter Extraction: This decomposition extracts key parameters for each path $\ell$ $ℓ$ :
- Azimuth ( $\phi_\ell$ ) and Elevation ( $\psi_\ell$ ) Angles of Arrival (AoA).
- Wavefront radius of curvature ( $\kappa_\ell$ ), which indicates the distance to the virtual source.
- Time of Arrival (ToA) and Doppler shift.
Adaptive NF/FF Processing: The algorithm dynamically switches between Near-Field and Far-Field models based on the estimated curvature ( $\kappa$ ). If $\kappa < 0$ (indicating FF behavior), it defaults to FF estimation; otherwise, it utilizes the richer NF curvature data.

C. Recursive Bayesian Tracking (EKF)

To handle the dynamic nature of vehicle movement and multipath, JAVELIN employs a Variable-Dimension Extended Kalman Filter (EKF):

State Vector: Includes the vehicle's 3D position and the positions of tracked Virtual User Equipments (VUEs) (virtual anchors representing reflectors).
Data Association: Uses a Gated Nearest-Neighbor (NN) strategy based on Mahalanobis distance to associate new measurements with existing tracks (vehicle or reflectors).
Track Management:
- Birth/Death: New tracks are initialized for new reflectors; tracks are removed if they miss association for a set number of steps (except the vehicle track).
- Static vs. Dynamic: Doppler shifts are used to distinguish static reflectors (tunnel walls, RRMs) from dynamic clutter (other vehicles), ensuring only stable geometric anchors are propagated.
LoS Handling: If a direct LoS path is unavailable, the system identifies the most likely LoS candidate based on geometric consistency between delay and curvature.

D. Radio-Reflective Road Markings (RRMs)

The paper introduces the deployment of passive Radio-Reflective Road Markings (RRMs). These are low-cost, passive metallic strips placed on road surfaces or walls.

Role: They act as "opportunistic anchors," enhancing geometric diversity without requiring active power or synchronization.
Analogy: Similar to visual lane markings for human drivers, RRMs provide geometric references for the positioning system in GNSS-denied environments.

3. Key Contributions

Geometric Validity Theorem: Formal derivation of the maximum array size ( $M_{max}$ ) for which the SR-NF model holds, linking geometric consistency to Fresnel-region scaling.
JAVELIN Framework: A novel single-BS localization method that combines tensor-based NF parameter estimation, adaptive NF/FF processing, and recursive tracking without needing digital maps.
Passive Infrastructure: Proposal of RRMs as a scalable, cost-effective solution to improve geometric diversity and robustness in tunnels.
Dynamic Tracking: Integration of variable-dimension EKF with track management to handle the birth and death of multipath components in real-time.

4. Simulation Results

The framework was evaluated in a realistic tunnel simulation (100m length, semi-cylindrical geometry) using MATLAB and a ray-tracer.

Scenarios: Tested under Line-of-Sight (LoS), Partial NLoS, and Complete NLoS conditions, with both straight and "slalom" (zig-zag) vehicle trajectories.
Baselines: Compared against TeNFiLoc, a state-of-the-art snapshot-based NF localization method.
Performance Metrics:
- JAVELIN-L (with LoS & RRMs): Achieved cm-level accuracy (2D RMSE of 0.20 m for straight trajectories).
- JAVELIN-N (NLoS): Maintained sub-meter accuracy (2D RMSE of 0.88 m) even without LoS, significantly outperforming TeNFiLoc (2.03 m).
- Impact of RRMs: Removing RRMs caused a significant degradation in accuracy (2D RMSE increased to 2.26 m), proving their critical role in geometric diversity.
- Dynamic Scenarios: In slalom trajectories, JAVELIN remained robust, whereas snapshot methods struggled with the rapid environmental changes.
CDF Analysis: JAVELIN showed a steeper Cumulative Density Function (CDF) curve, indicating a higher probability of low-error estimates compared to baselines.

5. Significance and Conclusion

This paper presents a significant advancement in GNSS-denied vehicular localization by:

Reducing Infrastructure Cost: Proving that a single anchor is sufficient for high-precision localization if Near-Field effects and passive reflectors are exploited.
Eliminating Map Dependency: The system operates without prior knowledge of the tunnel geometry or reflector locations, making it adaptable to irregular or unknown environments.
Theoretical Rigor: Providing the first formal characterization of the validity limits of single-reflector NF models, guiding future hardware design (array sizing).
Practical Viability: The proposal of RRMs offers a low-cost, scalable infrastructure upgrade for smart roads, aligning with future 5G/6G C-ITS deployment strategies.

The results demonstrate that JAVELIN is a robust, scalable solution for cooperative automated mobility in challenging tunnel environments, outperforming existing state-of-the-art methods in both accuracy and adaptability.