Optical Sensing for Fiber Spoofing Detection

This Master's Thesis focuses on the use of Optical Fiber Sensing, highlighting its exceptional efficiency in detecting fiber spoofing attempts. This technique is particularly valuable in high-security environments such as data centers, where preventing hacking attempts is critical. The Thesis is divided into two main sections. The first section explores the characterization of a new type of polarimeter, whose performance was previously unknown. The goal was to evaluate whether this polarimeter could meet the specific requirements of the Thesis while keeping costs manageable. The second section explores the application of theoretical concepts from optical fiber sensing to address the specific challenges posed by this work. This phase involves two crucial steps: first, a wide range of events typical for environments like data centers were experimentally reproduced. Three different scenarios were emulated: ideal fiber spoofing, fiber spoofing with accidental touches, and false alarms. An hacker entering a data center aims to corrupt the fiber to extract the information flowing through it. This is done bending the optical fiber so that light propagation is no longer guided. As the light exits the guiding material, the signal can be received by the hacker's device, creating a significant power drop. Next, a detection algorithm is developed to classify these events as either malicious or not. The core concept of the adopted methodology is to exploit the extreme sensitivity of optical fibers to mechanical stresses, which can be detected by tracking variations in polarization over time. The vibration characteristics were analyzed using the Poincaré Sphere and Stokes Parameters. However a more effective metric is the Angular Speed of the State of Polarization (SOPAS). SOPAS, a single parameter, is far more responsive to touches and mechanical stresses on the fiber. By assembling a large dataset of emulated events, the detection algorithm successfully differentiates between malicious spoofing incidents and harmless activities, operating in post-processing mode. The final goal was to optimize the algorithm for maximum detection accuracy while minimizing false alarms. To achieve this, particular attention was given to key parameters that are the foundation of the algorithm. Specifically, since the algorithm is threshold-based, two critical thresholds needed to be fine-tuned: one for SOPAS and another for power drops. Two types of detection maps were developed to guide this optimization. The first map focused on the joint optimization of the SOPAS threshold and the T_{mov}. Lower T_{mov} values are preferred, as they would be advantageous for potential real-time implementation of the algorithm. The second map focuses on optimizing the SOPAS threshold together with power drop values. This optimization is particularly relevant, as different levels of power drops are likely to be observed during actual fiber spoofing events. By jointly optimizing these thresholds, the algorithm becomes more robust and better equipped to detect real threats while reducing the risk of false alarms. The Thesis concludes with significant success, as the algorithm demonstrated the ability to perfectly distinguish between malicious and non-malicious events, often achieving 100 accuracy. This work lays a solid foundation for future advancements, particularly the development of real-time version. Continuous monitoring near fiber lines could trigger immediate alarms, helping in identifying suspiciuos activities.