Collaborative Compressive Detection With Physical Layer Secrecy Constraints

Bhavya Kailkhura, Thakshila Wimalajeewa, Pramod K. Varshney

TL;DR

This paper introduces a "Collaborative Compressive Detection" (CCD) framework that uses multiple sensors to improve signal detection performance when data is compressed, compensating for losses incurred by individual sensors. The framework also incorporates physical layer security by using cooperating nodes to inject artificial noise, aiming to maximize detection accuracy while ensuring perfect secrecy from eavesdroppers.

What Problem Does This Paper Address?

Traditional signal detection methods often require sampling at the Nyquist rate, which can be computationally expensive and generate large amounts of data, especially for high-dimensional signals. Compressive Sensing (CS) offers a way to reduce sampling requirements by acquiring fewer measurements than traditional methods, but this compression can lead to a performance loss in detection accuracy. Furthermore, in many practical scenarios, especially in wireless networks, the transmitted data might be intercepted by an eavesdropper. This paper tackles two main challenges: 1) mitigating the performance degradation caused by compression in signal detection, and 2) ensuring the security of this detection process against eavesdropping attacks, particularly in a collaborative sensing environment.

Methods

The authors propose a Collaborative Compressive Detection (CCD) framework. This framework involves multiple spatially distributed nodes that perform local compression of their observations using a universal CS-based measurement scheme. These compressed summaries are then sent to a central Fusion Center (FC) for a global decision. To compensate for the performance loss due to compression, the collaboration among nodes is key.

For security, the paper introduces an artificial noise injection model. Cooperating, trustworthy nodes inject artificial noise into their transmissions to mislead potential eavesdroppers. This noise injection is modeled probabilistically, with different noise injection strategies under different hypotheses (signal present or absent). The FC and the eavesdropper have different information: the FC knows the noise injection parameters, while the eavesdropper does not know which nodes are injecting noise.

The performance is analyzed using the modified deflection coefficient, which serves as a surrogate for the probability of error, especially in scenarios where closed-form error probability calculations are difficult. The authors also employ the concept of ε-stable embedding to derive performance bounds.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and uses synthetic data for illustrative examples, particularly in the context of spectrum sensing in cognitive radio networks.

Key Insights and Ideas

  • Collaboration as a Cure for Compression Loss: The paper demonstrates that by having multiple nodes collaborate, the performance loss incurred by individual compressive sensing nodes can be recovered, and in some cases, even surpassed traditional methods.
  • Artificial Noise for Secrecy: Injecting artificial noise is shown to be an effective strategy to confuse eavesdroppers and enhance the security of the detection system.
  • Trade-off Between Detection and Secrecy: There's an inherent trade-off between the accuracy of the detection at the Fusion Center and the level of secrecy achieved against eavesdroppers. Optimizing one can negatively impact the other.
  • Perfect Secrecy Regime: In the "perfect secrecy" regime (where the eavesdropper gains zero information), the security performance is surprisingly independent of the compression ratio, allowing for flexibility in choosing compression levels.
  • Optimal System Design: The paper formulates an optimization problem to find the best system parameters (compression ratio, noise injection strategy) that maximize detection performance while guaranteeing perfect secrecy.

Key Results

  • The paper analytically characterizes the trade-off between dimensionality reduction and achievable performance in the CCD framework.
  • It shows that for a fixed signal-to-noise ratio (SNR), if the number of collaborating nodes exceeds a certain threshold related to the compression ratio, the loss due to compression can be recovered.
  • The authors derive expressions for the modified deflection coefficient for both the Fusion Center and the eavesdropper, allowing for performance analysis under various conditions.
  • They demonstrate that in the perfect secrecy regime, the detection performance (DFC) is a monotonically increasing function of the compression ratio, while the eavesdropper's performance (DEV) can be independent of it.
  • The paper provides optimal system design strategies, suggesting how to choose compression ratios and artificial noise parameters to maximize detection performance while ensuring perfect secrecy. For instance, in the perfect secrecy regime, it's often optimal to minimize the fraction of nodes injecting noise and increase noise strength.

Limitations and Future Work

  • Deterministic vs. Random Signals: While the paper analyzes both deterministic and random signals, it notes that the deterministic signal case can be seen as a special case of the random signal analysis.
  • Assumptions: The analysis relies on assumptions such as perfect knowledge of observation and noise parameters by the FC and eavesdropper, and the ability to inject artificial noise. The identity of noise-injecting nodes is unknown to the eavesdropper.
  • Future Work: The authors suggest exploring scenarios with relaxed secrecy constraints, analyzing non-Gaussian signals in correlated noise, and investigating the impact of noisy communication channels. They also mention that the analytical methodologies developed could be applied to other detection problems.

Summary updated

Abstract

This paper considers the problem of detecting a high dimensional signal based on compressed measurements with physical layer secrecy guarantees. First, we propose a collaborative compressive detection (CCD) framework to compensate for the performance loss due to compression with a single sensor. We characterize the tradeoff between dimensionality reduction achieved by a universal compressive sensing (CS)-based measurement scheme and the achievable performance of CCD analytically. Next, we consider a scenario where the network operates in the presence of an eavesdropper who wants to discover the state of the nature being monitored by the system. To keep the data secret from the eavesdropper, we propose to use cooperating trustworthy nodes that assist the fusion center (FC) by injecting artificial noise to deceive the eavesdropper. We also design the system by determining the optimal values of parameters which maximize the CS-based collaborative detection performance at the FC while ensuring perfect secrecy at the eavesdropper. Experiments are conducted on synthetic data sets for spectrum sensing in cognitive radio networks to verify theoretical findings.
BibTeX
@article{kailkhura2017collaborative,
  author = {Bhavya Kailkhura and Thakshila Wimalajeewa and Pramod K. Varshney},
  title = {{Collaborative Compressive Detection With Physical Layer Secrecy Constraints}},
  journal = {IEEE Trans. Signal Process.},
  year = {2017},
  url = {https://openreview.net/forum?id=f6lNaZa0mz},
}
Figure