Abstract
The distributed inference framework comprises a group of spatially distributed nodes which acquire observations about a phenomenon of interest. Due to bandwidth and energy constraints, the nodes often quantize their observations into a finite-bit local message before sending it to the fusion center (FC). Based on the local summary statistics transmitted by nodes, the FC makes a global decision about the presence of the phenomenon of interest. The distributed and broadcast nature of such systems makes them quite vulnerable to different types of attacks. This paper addresses the problem of secure communication in the presence of eavesdroppers. In particular, we focus on efficient mitigation schemes to mitigate the impact of eavesdropping. We present an overview of the distributed inference schemes under secrecy constraints and describe the currently available approaches in the context of distributed detection and estimation followed by a discussion on avenues for future research.
Introduction
Distributed inference networks, such as wireless sensor networks (WSNs) or cognitive radio networks (CRNs), rely on spatially distributed nodes to observe a phenomenon, compress these observations, and transmit them to a central Fusion Center (FC). The FC then performs a global task, such as detection (binary decision) or estimation (continuous value calculation). However, the wireless and broadcast nature of these transmissions makes them highly vulnerable to passive eavesdroppers ("Eve"). The core challenge in these systems is maximizing the FC's inference accuracy while strictly limiting the information leaked to Eve, ensuring confidentiality without compromising the system's primary function.
System Architecture
As illustrated in the generalized system model, the process begins with a phenomenon $\theta$ observed by $N$ sensors ($S_1, \dots, S_N$). Each sensor produces an observation $Y_i$, which is then processed into a local message $v_i$. These messages are transmitted over potentially insecure channels to the FC. Simultaneously, an eavesdropper may intercept these signals through its own channels to attempt its own inference.
Mitigation Frameworks
This survey categorizes state-of-the-art mitigation strategies into four fundamental approaches:
- Design of Sensor Quantizers and Fusion Rules: Exploits channel differences between the FC and Eve. Optimization problems are formulated to maximize FC performance (e.g., using Kullback-Leibler divergence) while keeping Eve's performance below a strict threshold.
- Stochastic Encryption: Sensors confuse Eve by randomly flipping quantized decisions before transmission. The FC uses pre-shared flipping probabilities as a cryptographic key to decode the true state, while Eve receives degraded data.
- Artificial Noise Injection: Cooperative nodes inject noise into transmissions to degrade Eve's signal-to-noise ratio. The FC, knowing the noise distribution, can filter it out.
- MIMO Beamforming: Sensors with multiple antennas physically direct energy toward the FC while placing signal "nulls" in Eve's direction.
Performance Trade-offs and Key Results
A central theme in distributed inference security is the unavoidable trade-off between inference accuracy and secrecy. While "perfect secrecy"—where information leakage to Eve is driven to zero—is theoretically attainable, tighter security constraints inevitably degrade the FC's optimal performance.
In distributed estimation tasks, the efficacy of these methods is often measured by the distortion (Mean Squared Error) at the FC. As shown in the performance analysis below, the availability of Channel State Information (CSI) and the number of antennas significantly impact the system's ability to maintain low distortion while meeting security thresholds. Systems utilizing full CSI and multiple antennas can achieve performance remarkably close to theoretical bounds, significantly outperforming those with only partial CSI.
Future Research Directions
Current literature often assumes the FC has perfect knowledge of the eavesdropper's channel state information (CSI). However, because passive eavesdroppers do not broadcast their presence, this is rarely true in practice. Future research must focus on:
- Worst-Case Optimization: Designing networks optimized for the most challenging possible eavesdropper channel conditions within a bounded set.
- Hybrid Approaches: Combining multiple strategies, such as stochastic encryption and beamforming, to achieve superior performance-secrecy trade-offs.
Summary updated