On the Colluding Attack to Compromise Physical Layer Key Generation Through a Large-Scale Fading Estimation

Yang Lu, Yansong Gao, Seyit Camtepe, Dhammika JayalatharXiv 2019 · arXiv

Abstract

Physical-layer key generation (PKG) offers a lightweight security solution for the Internet of Things (IoT) by leveraging wireless channel randomness. However, in long-range (LoRa) scenarios, large-scale fading—which is spatially correlated—introduces a significant vulnerability. This paper characterizes a colluding-eavesdropping attack where multiple adversaries estimate large-scale fading to predict secret keys. We propose a countermeasure using Discrete Cosine Transform (DCT)-based high-pass filtering to remove predictable signal components. Real-world experiments demonstrate that while colluding eavesdroppers can significantly reduce key entropy, our filtering method restores security, achieving an eavesdropper Key Disagreement Rate (KDR) of approximately 0.5.

1. Introduction

The proliferation of IoT devices necessitates efficient cryptographic solutions. PKG exploits the reciprocity and spatial decorrelation of wireless channels to generate shared secrets without traditional key exchange. While effective in short-range environments dominated by small-scale fading, the security of PKG in long-range systems like LoRa is challenged by large-scale fading (path loss and shadowing), which remains correlated over larger distances.

2. The Colluding-Eavesdropping Attack Model

The core threat identified is the "channel perceiving attack." In this scenario, multiple eavesdroppers (Eves) surround a legitimate user (Alice). By colluding and averaging their received signal strength (RSSI) measurements, they can effectively filter out local small-scale fading and isolate the large-scale fading component shared with the legitimate link between Alice and Bob.

The proposed colluding-eavesdropping attack model where multiple Eves surround Alice to estimate the channel state.

3. Experimental Methodology

To validate this threat and test countermeasures, a LoRa testbed was constructed using SX1276 modules and Arduino Nanos. Experiments were conducted in diverse environments, including indoor (non-line-of-sight) and outdoor (line-of-sight) settings.

3.1 Data Collection

RSSI samples were collected across four primary scenarios: static channels, environments with moving scatterers, and scenarios with a moving user. Figure 5 illustrates the raw RSSI traces obtained from these experiments, showing the high correlation between Alice and Bob's measurements—a feature that both enables PKG and creates the vulnerability to large-scale fading estimation.

RSSI measurements for Alice and Bob across four experimental scenarios (a-d), demonstrating the inherent correlation used for key generation.

4. Proposed Countermeasure: DCT-Based High-Pass Filtering

The researchers developed a signal-processing fix to address the predictability of large-scale fading. By applying a Discrete Cosine Transform (DCT-II) to the RSSI sequences, the signal is decomposed into frequency components. A high-pass filter then removes the low-frequency components associated with large-scale fading, leaving only the rapidly fluctuating small-scale fading.

This approach contrasts with the commonly used Moving Window Average (MWA), which the authors identified as a security risk in long-range settings; MWA smooths out the very randomness required for secret keys, inadvertently aiding the attacker. Our adaptive algorithm instead uses conditional entropy to determine the optimal number of low-frequency components to discard.

5. Key Results and Discussion

The attack's impact was significant: without filtering, colluding Eves obtained 15% more channel information than a single attacker, reducing the eavesdropper KDR to approximately 0.25.

Implementation of the DCT-based filter successfully mitigated the attack. The eavesdropper KDR increased to 0.49 (near-perfect uncertainty), while the legitimate KDR remained below 0.2, manageable by standard error correction. Furthermore, the resulting keys passed the NIST randomness test suite, confirming their cryptographic suitability.

6. Conclusion

This work highlights a critical oversight in applying short-range PKG assumptions to long-range LoRa communications. By accounting for large-scale fading through DCT-based filtering, the security of PKG can be preserved even against sophisticated colluding adversaries. Future work will focus on optimizing these algorithms for highly dynamic environments and resource-constrained hardware.

Summary updated

BibTeX
@misc{lu2019colluding,
  author = {Yang Lu and Yansong Gao and Seyit Camtepe and Dhammika Jayalath},
  title = {{On the Colluding Attack to Compromise Physical Layer Key Generation Through a Large-Scale Fading Estimation}},
  year = {2019},
  eprint = {1910.08770},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1910.08770},
}
Figure