FIS-ONE: Floor Identification System with One Label for Crowdsourced RF Signals

Weipeng Zhuo, Ka Ho Chiu, Jierun Chen, Ziqi Zhao, S.-h. Gary Chan, Sangtae Ha, Chul-Ho Lee

Introduction

Indoor localization is a cornerstone of the "smart city" vision, enabling applications ranging from autonomous delivery robots to emergency response. However, identifying the specific floor level of a device using Radio Frequency (RF) signals—such as Wi-Fi—typically requires a labor-intensive effort to "fingerprint" every floor with labeled data.

A new system, FIS-ONE, proposes a method to bypass this bottleneck. By leveraging graph neural networks and optimization theory, the researchers demonstrate that it is possible to identify floors across an entire building using just one labeled signal sample from the bottom floor.

Methodology

The FIS-ONE architecture consists of two primary phases: signal clustering and cluster indexing.

Overview of the FIS-ONE architecture, illustrating the flow from crowdsourced RF signals to final cluster indexing.

1. Signal Clustering via RF-GNN

The system models crowdsourced data as a bipartite graph, where nodes represent either signal samples or Access Points (APs). To process this graph, the authors introduced RF-GNN, an attention-based Graph Neural Network. Unlike standard GNNs, RF-GNN uses Received Signal Strength (RSS) as an attention weight. Intuitively, if a sample detects a strong signal from an AP, that AP node exerts a stronger influence on the sample's mathematical representation (embedding). Once these high-quality embeddings are learned, Hierarchical Clustering is used to group similar samples together into floor-level clusters.

2. Cluster Indexing via TSP Formulation

To determine the vertical order of these clusters (e.g., 1st, 2nd, 3rd floor), FIS-ONE exploits "signal spillover." Wi-Fi signals are not perfectly contained by floors; they bleed into adjacent levels.

Illustration of signal spillover across floors and the frequency of MAC address detection across multiple levels.

As shown in Figure 1, adjacent floors share many common APs, while distant floors share few. The system defines an Adapted Jaccard Similarity to measure this overlap. The ordering of clusters is then formulated as a Traveling Salesman Problem (TSP) variant. By finding the path with the highest cumulative similarity starting from the single labeled bottom-floor cluster, the system mathematically deduces the correct floor sequence.

Experimental Evaluation

The authors validated FIS-ONE using two major sources: the Microsoft Open Dataset (152 buildings) and real-world data collected from three large shopping malls (5–7 floors).

  • Clustering Accuracy: FIS-ONE outperformed state-of-the-art baselines (such as SDCN and METIS) by up to 23% in Adjusted Rand Index (ARI) and 25% in Normalized Mutual Information (NMI).
  • Indexing Accuracy: Using "Edit Distance" to measure floor ordering success, FIS-ONE achieved a score of ~0.88, significantly higher than the 0.6–0.8 range achieved by other methods.
  • Robustness: The system remained stable across different embedding dimensions (from 8 to 64), proving it is not overly sensitive to hyperparameter tuning.

Key Insights and Conclusion

FIS-ONE demonstrates that signal spillover, often considered a nuisance in localization, can be utilized as a vital clue for floor sequencing. By shifting the problem from classification to optimization (ordering), the system functions with virtually zero supervision. This represents a significant step toward fully unsupervised indoor localization, potentially eliminating the need for manual labeling entirely through the use of environmental cues.

Summary updated

Abstract

Floor labels of crowdsourced RF signals are crucial for many smart-city applications, such as multi-floor indoor localization, geofencing, and robot surveillance. To build a prediction model to identify the floor number of a new RF signal upon its measurement, conventional approaches using the crowdsourced RF signals assume that at least few labeled signal samples are available on each floor. In this work, we push the envelope further and demonstrate that it is technically feasible to enable such floor identification with only one floor-labeled signal sample on the bottom floor while having the rest of signal samples unlabeled. We propose FIS-ONE, a novel floor identification system with only one labeled sample. FIS-ONE consists of two steps, namely signal clustering and cluster indexing. We first build a bipartite graph to model the RF signal samples and obtain a latent representation of each node (each signal sample) using our attention-based graph neural network model so that the RF signal samples can be clustered more accurately. Then, we tackle the problem of indexing the clusters with proper floor labels, by leveraging the observation that signals from an access point can be detected on different floors, i.e., signal spillover. Specifically, we formulate a cluster indexing problem as a combinatorial optimization problem and show that it is equivalent to solving a traveling salesman problem, whose (near-)optimal solution can be found efficiently. We have implemented FIS-ONE and validated its effectiveness on the Microsoft dataset and in three large shopping malls. Our results show that FIS- ONE outperforms other baseline algorithms significantly, with up to 23 % improvement in adjusted rand index and 25% improvement in normalized mutual information using only one floor-labeled signal sample.
BibTeX
@inproceedings{zhuo2023fis,
  author = {Weipeng Zhuo and Ka Ho Chiu and Jierun Chen and Ziqi Zhao and S.-H. Gary Chan and Sangtae Ha and Chul-Ho Lee},
  title = {{FIS-ONE: Floor Identification System with One Label for Crowdsourced RF Signals}},
  booktitle = {ICDCS},
  year = {2023},
  url = {https://openreview.net/forum?id=oKmvX8YYNy},
}
Figure