Abstract
Mobile Edge Computing (MEC) enhances service delivery by moving computation closer to the user, but this proximity introduces significant privacy risks. Specifically, Location-Based Services (LBS) at the edge often require raw wireless fingerprints—lists of nearby Wi-Fi Access Points (APs)—to calculate user coordinates. This paper introduces LoPEC, a privacy-preserving scheme that generates believable fake fingerprints from a locally-maintained noise pool. By leveraging graph-theoretic approaches, LoPEC provides efficient, trajectory-aware privacy protection against untrusted edge servers.
The Privacy Challenge in MEC
In traditional cloud-based systems, privacy research focuses on obfuscating final GPS coordinates. However, in MEC environments, the "Location Service" (LS) often resides at the edge and calculates the user's position based on a wireless fingerprint. This fingerprint consists of a list of nearby Wi-Fi APs and their respective signal strengths.
As shown in the architecture above, the raw signal data is the primary privacy leak. To an attacker at the edge, the fingerprint is the location. The authors identify two major gaps in existing solutions: the need for "believable" noise (using real-world APs rather than fake MAC addresses) and the inherent vulnerability of edge nodes, which may be less secure than centralized clouds.
The LoPEC Framework
To address these gaps, the authors propose LoPEC (Location Privacy in Edge Computing). The core idea is to send multiple fingerprints to the edge server—one real and several decoys—so the server cannot distinguish the true user location.
LoPEC consists of three algorithmic components:
- SCSOA (Self-Collect and Self-Organize Algorithm): The mobile device builds its own "noise pool" by recording APs encountered during daily movement. It constructs an undigraph where edges represent overlapping coverage areas.
- CSDA (Complete Subgraph Discovery Algorithm): To generate a valid fake location, the system identifies "cliques" (groups of overlapping APs) within the graph. CSDA uses clustering coefficients to find these dense areas in linear time, ensuring the decoys are geographically plausible.
- e-CSDA (Enhanced CSDA): For users in motion, e-CSDA ensures that fake locations follow a logical trajectory. By selecting noise APs adjacent to the previous noise set, it prevents "homogeneity attacks" where an attacker might spot the only location moving realistically.
Evaluation and Key Results
The researchers evaluated LoPEC using the Harbin Dataset, which contains data from 2,016 Wi-Fi routers in a business district in Harbin, China. The system was tested against two standard positioning algorithms: RADAR (deterministic) and PBL (probabilistic).
- Effectiveness: LoPEC achieved a success rate of over 85% in generating valid clique noise, even with relatively small initial datasets.
- Efficiency: Unlike brute-force clique discovery which is NP-complete, LoPEC’s computational cost scales linearly with the size of the AP graph, making it suitable for real-time mobile use.
- Trajectory Realism: Visual analysis confirmed that e-CSDA generates continuous "ghost" trajectories, making it nearly impossible for an attacker to distinguish the real user from the decoys.
Conclusion and Future Work
LoPEC demonstrates that mobile devices can serve as their own privacy providers by leveraging local knowledge and graph theory. While highly effective, the authors note that future iterations could improve kinematic realism by accounting for velocity and direction. Additionally, the system's performance is currently dependent on the maturity of the user's locally-mapped AP graph, suggesting a need for strategies to handle new or sparsely mapped environments.
Summary updated