Introduction
Positioning in underground environments like mines and tunnels is critical for logistics, energy efficiency, and worker safety. However, standard radio propagation models, such as Free Space Path Loss, often fail in these settings. Tunnels act as complex waveguides where geometry, wall materials, and heavy machinery create unique electromagnetic behaviors that vary significantly between locations.
Traditionally, engineers had to choose between labor-intensive manual site surveys to tune models for every corridor or computationally expensive simulations. This paper proposes a "middle way": a self-adaptive path loss model that improves positioning accuracy by fusing Wi-Fi signal strength with RFID "anchor" data, achieving the precision of manual tuning without the associated engineering overhead.
Methodology: Self-Adaptive Two-Piece Modeling
The core of the proposed system is a self-adaptive two-piece path loss model. Unlike a single-curve model, this approach assumes two distinct propagation regions:
- Near Region: Characterized by steep signal loss, behaving similarly to free space.
- Far Region: Where the "waveguide effect" of the tunnel flattens the signal loss slope.
The system utilizes a heterogeneous data fusion strategy. While Wi-Fi infrastructure provides continuous tracking, RFID readers serve as "ground truth" anchor points. Whenever a Wi-Fi-enabled device passes an RFID reader, its exact location is captured. The system records the Wi-Fi Received Signal Strength Indication (RSSI) at that moment to calibrate the model iteratively. Using a nonlinear least-squares method, the system optimizes four key parameters: the path loss exponent ($\gamma$), a system loss constant ($C$), the "break point" distance ($d_0$), and the far-region slope ($\alpha$).
Experimental Setup
The researchers validated the model across three distinct environments:
- Simulated Environment: A 300-meter synthetic stretch used to test parameter convergence under high noise (SNR of -2 dB).
- University Corridor: An 80-meter underground corridor for initial real-world proof-of-concept.
- Kristineberg Ore Mine: A deep mine in northern Sweden. Two 300-meter test areas were used, with measurements collected via laptops and Ekahau Wi-Fi asset tags.
Key Results and Insights
A critical finding is that the placement of RFID anchors is more vital than their quantity. Effective calibration requires anchors in both the "near" and "far" propagation regions relative to the base station.
In simulations, the model parameters converged to true values even in noisy conditions. While 19 uniformly distributed anchors provided rapid convergence, the authors demonstrated that as few as 4 strategically placed anchors were sufficient. In the Kristineberg mine, the self-adaptive model's distance predictions closely matched actual distances, performing on par with models manually tuned by experts. Furthermore, by normalizing distance estimates from two base stations, the system successfully corrected errors where the sum of estimated distances deviated from the known distance between stations.
Limitations and Future Work
Currently, the model does not explicitly account for "shadow fading" caused by large temporary obstacles like moving trucks. It has also primarily been tested in static tunnel geometries. Future research may incorporate the type of moving object (e.g., human vs. metal truck) into the model, as different materials affect signal propagation differently. The authors also suggest exploring more complex piecewise linear models with more than two segments for enhanced precision.
Summary updated