An Analysis of Device-Free and Device-Based WiFi-Localization Systems

Heba Aly, Moustafa Youssef

TL;DR

This paper analyzes the performance of WiFi-based localization systems, both device-based (tracking a device) and device-free (tracking an entity without a device). It uses a novel automated radio map generation tool (AROMA) to explore various "what-if" scenarios that are difficult to test in real-world environments, providing practical insights for improving localization accuracy.

What Problem Does This Paper Address?

Indoor localization using WiFi is a promising technology due to the ubiquity of WiFi infrastructure. However, indoor environments present complex signal propagation challenges. Traditional WiFi localization methods often rely on manually constructed "fingerprint maps" (radio maps) which are time-consuming and labor-intensive to create. This paper addresses the need for a more efficient and robust understanding of factors influencing WiFi localization accuracy, particularly for both device-based and device-free systems, by simulating various environmental and configuration changes.

Methods

The core of this research utilizes AROMA, a state-of-the-art system for automatic radio map generation. AROMA employs site-specific ray tracing augmented with the Uniform Theory of Diffraction (UTD) to predict RF propagation in a 3D environment. It models human bodies as metallic cylinders. The paper then leverages AROMA to simulate and analyze various "what-if" scenarios for both device-based and device-free localization. These scenarios include:

  • Device-Based:
    • Effect of Access Point (AP) mounting location (wall vs. ceiling).
  • Impact of upgrading AP hardware (switching frequency bands from 2.4 GHz to 5.7 GHz).
  • Influence of crowd presence on calibration and operation.
  • Device-Free:
    • Effect of AP mounting location (wall vs. ceiling).
  • Impact of upgrading AP hardware (switching frequency bands from 2.4 GHz to 5.7 GHz).
  • Effect of entities moving outside the area of interest.

The analysis uses two metrics: change in Received Signal Strength (RSS) and average localization error in meters.

Datasets

The paper describes using a 3D model of a typical apartment (700 sq ft / 66 m²) for simulations. For device-based experiments, eleven radio map locations were used. For device-free experiments, a denser radio map of forty-four locations was employed. No specific external datasets were mentioned; the data was generated through simulations using the AROMA system.

Key Insights and Ideas

  • Ceiling-mounted APs are generally better for device-based localization: While wall-mounted APs might lead to more signal attenuation due to interactions with objects, ceiling-mounted APs provide a more consistent signal and thus higher accuracy, especially in crowded scenarios.
  • Higher frequencies (5.7 GHz) improve device-based accuracy but reduce coverage: While 5.7 GHz bands offer better localization accuracy, they require more APs to cover the same area due to reduced range.
  • Crowd training for device-based systems is tricky: Training a device-based system with a single person (no crowd) is generally better than training with a crowd, as crowd presence can significantly change between training and operation phases, leading to degraded accuracy.
  • Wall-mounted APs are better for device-free localization: Contrary to device-based systems, wall-mounted APs offer higher accuracy in device-free scenarios. This is because they increase the likelihood of a human entity blocking the line-of-sight (LOS) between the AP and the monitoring point, which is crucial for device-free detection.
  • Lower frequencies (2.4 GHz) are preferred for device-free accuracy and coverage: The 2.4 GHz band is preferred for device-free systems due to better accuracy and coverage compared to the 5.7 GHz band.
  • External movement doesn't affect device-free systems: People moving outside the area of interest do not impact RSS streams within the area, highlighting the potential for device-free systems in large-scale intrusion detection.

Key Results

  • Device-Based Localization:
    • Ceiling-mounted APs achieved an average localization error of 1.00m, compared to 1.84m for the base scenario (wall-mounted).
  • Using the 5.7 GHz band resulted in an average error of 1.55m, compared to 1.84m for 2.4 GHz.
  • Crowd scenarios generally increased localization error, with errors ranging from 2.02m to 3.06m.
  • Device-Free Localization:
    • Ceiling-mounted APs resulted in a significantly higher average localization error (4.48m) compared to the base scenario (wall-mounted, 1.44m).
  • Using the 5.7 GHz band increased the average localization error to 1.77m, compared to 1.44m for 2.4 GHz.
  • AROMA's validation showed a maximum average absolute error of 2.77 dB and a maximum localization error of 3.13m for both device-based and device-free cases.

Limitations and Future Work

  • Limitations: The study relies on simulations using the AROMA system, which models human bodies as metallic cylinders. While AROMA has been validated, real-world complexities might introduce further variations. The "crowd" scenarios are simplified representations.
  • Future Work: The authors suggest that their analysis provides valuable insights for both researchers and practitioners. They recommend specific configurations for better localization accuracy in different scenarios (e.g., ceiling-mounted APs for device-based, wall-mounted for device-free, and specific frequency bands). The study highlights the promise of device-free localization for large-scale intrusion detection.

Summary updated

Abstract

WiFi-based localization became one of the main indoor localization techniques due to the ubiquity of WiFi connectivity. However, indoor environments exhibit complex wireless propagation characteristics. Typically, these characteristics are captured by constructing a fingerprint map for the different locations in the area of interest. This fingerprint requires significant overhead in manual construction, and thus has been one of the major drawbacks of WiFi-based localization. In this paper, we present an automated tool for fingerprint constructions and leverage it to study novel scenarios for device-based and device-free WiFi-based localization that are difficult to evaluate in a real environment. In a particular, we examine the effect of changing the access points (AP) mounting location, AP technology upgrade, crowd effect on calibration and operation, among others; on the accuracy of the localization system. We present the analysis for the two classes of WiFi-based localization: device-based and device-free. Our analysis highlights factors affecting the localization system accuracy, how to tune it for better localization, and provides insights for both researchers and practitioners.
BibTeX
@article{aly2014analysis,
  author = {Heba Aly and Moustafa Youssef},
  title = {{An Analysis of Device-Free and Device-Based WiFi-Localization Systems}},
  journal = {Int. J. Ambient Comput. Intell.},
  year = {2014},
  url = {https://openreview.net/forum?id=bGy5CpgOYj},
}
Figure