Balancing Accuracy and Privacy: Next-Generation Incentives for Indoor Localization

TL;DR

Building radio maps for indoor localization is a massive task. Crowdsourcing solves the labor problem but creates a privacy nightmare. This paper proposes two game-theoretic incentive mechanisms that use Differential Privacy to protect user trajectories while offering rewards that maximize platform profit and user participation.

The Motivation: Why Privacy Stops Crowdsourcing

Indoor localization (Wi-Fi fingerprinting) is essential for malls, airports, and stations. While researchers have moved toward Mobile Crowdsourcing (MCS) to build these maps, users hesitate for two reasons:

1. Resource Drain: Sensing RSSI and barometer data consumes battery and compute.
2. Privacy Risks: An activity trace can reveal which shops you visited, your habits, or even your identity through de-anonymization.

Existing models often treat privacy as a binary "on/off" switch. This paper treats it as a quantifiable cost in an economic game.

Methodology: The Three-Way Tug-of-War

The authors define a delicate ecosystem involving the Crowdsourcing Platform (CP), Mobile Users (MUs), and Service Customers (SCs).

1. Quantifying the "Cost" of Privacy

The paper utilizes $ϵ$-Differential Privacy. A lower $ϵ$ means more noise is added to the trajectory, providing stronger privacy but lower utility for the platform. They propose a unique utility function: $g(d_i,{ϵ}_i)=[{\pi }_1-{\pi }_2{e}^{{\pi }_3(1-{ϵ}_i)}]d_i$ This ensures that utility is proportional to the trajectory length $d$ and the privacy budget $ϵ$.

2. Mechanism A: Fixed Reward (Incomplete Information)

In reality, a platform doesn't know how much a user values their privacy.

• The Game: A two-stage Stackelberg game where the CP (Leader) sets a global reward $R$, and MUs (Followers) decide their trajectory length $d_i$ to contribute.
• The Key: Even without knowing individual privacy preferences, the platform leverages the probability distribution of user types to reach a Nash Equilibrium.

3. Mechanism B: Variable Reward (Complete Information)

If the CP knows the user's privacy sensitivity, it can offer a "price menu." This model introduces a Demand Function to account for Service Customers:

• Higher buying prices $p$ attract more users (higher quality).
• Higher service prices $q$ decrease demand from customers. The goal is to find the "sweet spot" $(p^{\ast },q^{\ast })$ using concave optimization.

Experimental Insights

The simulations yield a critical finding: Privacy protection doesn't have to hurt the bottom line.

As shown in the Results, the proposed mechanism maintains a higher profit margin for the platform compared to traditional "Platform-Centric" models that ignore privacy. By accounting for the user's privacy cost, the platform can more effectively calibrate rewards to maintain high participation rates even as user numbers scale to 1,000+.

The search for the optimal reward ($R^{\ast }$) shows that as user sensitivity to privacy loss increases, the platform must adjust its incentive strategy to prevent a mass exodus of data contributors.

Critical Analysis & Conclusion

The core contribution of this work is the marriage of Differential Privacy with Game Theory in the specific context of indoor trajectories.

• Strengths: It moves beyond generic sensing tasks and addresses "U-Turn" faking by using peak detection algorithms for trajectory verification.
• Limitations: The model assumes a trusted CP in the variable reward scenario, which may not hold in decentralised environments.
• Future Work: The logical next step is exploring Local Differential Privacy (LDP), where noise is added on the device before the platform even sees the data, removing the need for a "Trusted" central entity.

This paper provides a robust blueprint for developers of indoor LBS to build sustainable, privacy-preserving ecosystems that users can actually trust.