Abstract
This paper provides a rigorous finite-blocklength analysis of covert communication over Additive White Gaussian Noise (AWGN) channels. By utilizing Total Variation Distance (TVD) as a precise metric for adversary detection, we derive tight upper and lower bounds on maximal throughput. Our analysis identifies a second-order asymptotic term of $O(n^{1/4})$, moving beyond traditional infinite-length "Square Root Law" assumptions and providing a more accurate framework for practical system design.
Introduction
In covert communication, the objective is to hide the very occurrence of a transmission from an adversary, often referred to as "Willie." Traditionally, this field has been governed by the "Square Root Law," which posits that one can transmit $O(\sqrt{n})$ bits over $n$ channel uses with negligible detection probability. However, most existing literature relies on asymptotic analysis ($n \to \infty$) and uses Kullback-Leibler (KL) divergence as a proxy for covertness.
In practical scenarios, blocklengths are finite, and KL divergence can be a loose bound for the actual probability of detection. This paper addresses this gap by analyzing the non-asymptotic regime using Total Variation Distance (TVD), which is directly tied to the optimal hypothesis test an adversary would employ.
System Model
The communication model involves three primary actors: Alice (the transmitter), Bob (the intended receiver), and Willie (the adversary). Alice attempts to send information to Bob over an AWGN channel while ensuring that Willie cannot distinguish her signal from background noise.
As illustrated in Figure 1, Alice transmits a sequence $x_1, \dots, x_n$. Bob receives this signal plus noise $N_{B_i}$ and attempts to decode the message. Simultaneously, Willie observes the channel. His task is a hypothesis test: deciding whether he is observing pure noise $N_{W_i}$ or a signal-plus-noise sequence $z_1, \dots, z_n$.
Methods
The authors model the system as an AWGN channel where Alice employs Gaussian signaling. To analyze performance under a covertness constraint (TVD $\leq \delta$), several mathematical techniques are utilized:
- Divergence Inequalities: Pinsker’s inequality and Hellinger distance are used to establish necessary and sufficient conditions for power levels that satisfy covertness.
- Analytic TVD Formulation: A closed-form analytic expression for the TVD between the "noise-only" and "signal-plus-noise" distributions is derived using incomplete gamma functions.
- Asymptotic Expansion: Series expansions (using Stirling’s formula) characterize TVD behavior as $n$ grows, making complex gamma functions practical for evaluation.
- Second-Order Analysis: Finite-blocklength coding theorems are applied to determine the maximal number of bits that can be reliably and covertly transmitted, focusing on the $O(n^{1/4})$ term that describes the "back-off" from asymptotic capacity.
Key Insights and Results
The study yields several significant theoretical and practical insights:
- TVD vs. KL Divergence: TVD is identified as the "gold standard" for finite blocklengths because the adversary's minimum total error probability is exactly $1 - \text{TVD}$.
- The $O(n^{1/4})$ Term: The most significant contribution is the identification of the second-order asymptotic of maximal throughput. While the first-order term follows the $O(\sqrt{n})$ Square Root Law, the second-order term is $O(n^{1/4})$, providing a much more accurate picture for real-world system design.
- Phase Transition of $\tau$: If the signal-to-noise ratio ($snr$) scales as $n^{-\tau}$, the system behavior changes at $\tau = 1/2$. For $\tau < 1/2$, TVD converges to 1 exponentially fast (detection is certain), whereas for $\tau > 1/2$, TVD converges to 0 (covertness is maintained).
- Throughput Bounds: Explicit upper and lower bounds for the maximal coding rate $M^*(n, \epsilon, \delta)$ were established under decoding error probability $\epsilon$ and covertness constraint $\delta$.
Numerical simulations confirmed that the proposed analytic approximations for TVD are highly accurate even for relatively small blocklengths.
Conclusion
This work provides a foundational analysis of covert communication in the finite-blocklength regime. By moving beyond asymptotic approximations and utilizing TVD, it offers precise bounds and identifies critical second-order terms for throughput. Future research may extend this analysis to MIMO systems and channels with fading or noise uncertainty, where the fundamental limits of covertness may differ.
Summary updated