[Paper Review] Finite-Blocklength Bounds for Wiretap Channels

Wei Yang, Rafael F. Schaefer+1 more|arXiv (Cornell University)|Jan 22, 2016

Wireless Communication Security Techniques24 references3 citations

TL;DR

This paper presents tighter finite-blocklength bounds for wiretap channels by introducing a refined privacy amplification lemma and a novel converse bound based on binary hypothesis testing. The proposed bounds yield the tightest known second-order coding rates for both discrete memoryless and Gaussian wiretap channels, demonstrating the superiority of privacy amplification in achieving secrecy capacity with finite blocklengths.

ABSTRACT

This paper investigates the maximal secrecy rate over a wiretap channel subject to reliability and secrecy constraints at a given blocklength. New achievability and converse bounds are derived, which are shown to be tighter than existing bounds. The bounds also lead to the tightest second-order coding rate for discrete memoryless and Gaussian wiretap channels.

Motivation & Objective

• To derive tighter achievability and converse bounds on the maximal secrecy rate $ R^*(n, \\/epsilon, \delta) $ for wiretap channels at finite blocklength.
• To improve the second-order coding rate (dispersion) for discrete memoryless and Gaussian wiretap channels by refining privacy amplification techniques.
• To establish a connection between secure communication and list decoding at the eavesdropper by analyzing partition codes.
• To provide uniformly tighter bounds than existing works for the Gaussian wiretap channel, validated through numerical computation.
• To analyze the asymptotic behavior of the bounds in the $ n \to \infty $ regime to characterize the second-order coding rate.

Proposed method

• Propose a new privacy amplification lemma that refines prior results in [6, 17], enabling tighter achievability bounds.
• Develop a one-shot converse bound using binary hypothesis testing, inspired by [14, 15], to relate secrecy to statistical distinguishability.
• Apply the $ \beta_\alpha $-metric and $ E_\gamma $-metric to quantify the performance of randomized tests between joint and product distributions.
• Use spherical symmetry and Gaussian approximation to evaluate the $ E_\gamma $-metric for the Gaussian wiretap channel under uniform input on the power sphere.
• Analyze the asymptotic behavior of the bounds using central limit theorem approximations to derive second-order coding rates.
• Derive two converse bounds for partition codes, revealing a duality between secrecy and list decoding at the eavesdropper.

Experimental results

Research questions

• RQ1 How can privacy amplification be refined to yield tighter finite-blocklength bounds for secrecy capacity?
• RQ2 What is the tightest achievable second-order coding rate for discrete memoryless and Gaussian wiretap channels under finite blocklength constraints?
• RQ3 How does the proposed converse bound compare to existing ones in terms of tightness for the Gaussian wiretap channel?
• RQ4 What is the connection between secure communication and list decoding at the eavesdropper in the context of partition codes?
• RQ5 Can the proposed bounds be used to derive a strong converse for the wiretap channel in the asymptotic regime?

Key findings

• The proposed achievability bound is tighter than existing bounds, particularly improving the second-order coding rate for both discrete memoryless and Gaussian wiretap channels.
• For the Gaussian wiretap channel, both the new achievability and converse bounds are uniformly tighter than the best existing bounds.
• The second-order coding rate derived from the new achievability bound is tighter than those in [7, 10], demonstrating the advantage of privacy amplification in finite-blocklength secrecy coding.
• The asymptotic analysis confirms that the second-order coding rate is $ nC_S - \sqrt{nV_c}Q^{-1}(1 - \epsilon - \delta - \tau) + \mathcal{O}(\log n) $, with $ V_c $ given in (51).
• The converse bound for partition codes reveals that the eavesdropper’s ability to decode via list decoding is fundamentally linked to the secrecy rate.
• Numerical evaluation confirms that the new bounds are tighter than existing ones, especially in the moderate-to-short blocklength regime.