Rate-Distortion-Based Physical Layer Secrecy in Multimode Fiber

Eva C. Song, Emina Soljanin, Paul Cuff, H. Vincent Poor

Optical networks, particularly those utilizing Multimode Fiber (MMF), form the backbone of modern high-capacity communication. However, these systems are physically vulnerable; an attacker can "tap" the signal by simply bending the fiber to leak light. Traditional information-theoretic security often aims for "perfect secrecy," where zero information leaks to an eavesdropper (Eve). In complex MMF systems—which behave like Multiple-Input-Multiple-Output (MIMO) wireless channels—perfect secrecy is extremely difficult to achieve and often requires sacrificing significant transmission speed.

This paper, published in IEEE Transactions on Communications, explores a practical approach by shifting the focus from perfect secrecy to "maximum distortion." The researchers propose using distortion—the degree to which an eavesdropper’s estimate of a message is garbled—as the primary metric for physical layer secrecy.

System Model and Methodology

The authors model the MMF as a complex Gaussian MIMO broadcast channel. This model accounts for physical characteristics such as mode-dependent loss (MDL) and modal crosstalk. The physical layer interaction is represented by the channel outputs for the legitimate receiver (Bob) and the eavesdropper (Eve).

The MIMO broadcast channel model representing the physical characteristics of the multimode fiber link, where H and He represent the channel matrices for Bob and Eve respectively.

The core of the proposed method is an operational separation scheme using joint source-channel coding. The source sequence $S^k$ is encoded into $X^n$, which is transmitted over the channel. The legitimate receiver uses decoder $g_{k,n}$ to produce an estimate $\hat{S}^k$, while the eavesdropper attempts to reconstruct the source as $T^k$.

Block diagram of the source-channel coding scheme. The dashed line represents the "worst-case" scenario where the eavesdropper has access to causal information (past realizations of the source).

The framework considers two scenarios for the eavesdropper:

  1. No Causal Information: Eve only has access to her channel output $Z^n$.
  2. Causal Information: A more sophisticated attacker who also knows all past realizations of the source sequence $S^{j-1}$.

Key Insights: "Secrecy is Cheap"

One of the most striking findings of the paper (Theorem 1) is that a high-rate secure channel is not required to force maximum distortion on an attacker. If a secure channel with any positive rate exists, it is possible to transmit at Bob's full capacity while keeping Eve's estimate as distorted as if she had received no signal at all. This suggests that the natural redundancy in a data source (such as patterns in text or images) can be leveraged to mask information effectively.

While most literature uses "equivocation" (a measure of uncertainty), this paper argues that "distortion" is often more relevant for real-world applications. If an attacker's reconstructed image is sufficiently garbled, it becomes useless, regardless of whether some information theoretically leaked.

Results and Numerical Validation

The authors validated their theoretical bounds using numerical simulations of a 4-mode MMF system. Key results include:

  • Theorem 1 (The "Best Case"): Proves that for a general broadcast channel, Bob can receive data at the maximum reliable rate while Eve suffers maximum distortion.
  • Theorem 8 (The "Worst Case"): Provides a specific tradeoff curve for MMF when Eve has access to causal information, showing how Alice can maintain high distortion for the current message even against a sophisticated attacker.
  • Source Redundancy: Simulations demonstrate a "free region" for sources with redundancy. In this region, the transmission rate can be increased without Eve’s distortion dropping, effectively providing "security for free" by utilizing the source's inherent structure.

Limitations and Future Directions

The current proofs assume the source data is independent and identically distributed (i.i.d.), whereas real-world data often has complex correlations. Additionally, the model requires Bob to reconstruct the source perfectly (lossless). Future research may explore "lossy" reconstruction for the legitimate receiver and address the challenges of obtaining Channel State Information (CSI) in real-time MMF systems.

Summary updated

BibTeX
@article{song2014rate,
  author = {Eva C. Song and Emina Soljanin and Paul Cuff and H. Vincent Poor},
  title = {{Rate-Distortion-Based Physical Layer Secrecy in Multimode Fiber}},
  journal = {IEEE Trans. Communications},
  year = {2014},
  doi = {10.1109/tcomm.2014.010914.130256},
}
Figure