Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the Fragility of AI-Based Channel Decoders under Small Channel Perturbations

Haoyu Lei, Mohammad Jalali, Chin Wa Lau, Farzan Farnia

TL;DR

The paper investigates the fragility of AI-based channel decoders under small, norm-bounded perturbations of the AWGN channel output, questioning the source of nominal gains over belief-propagation. It develops a robustness framework with a smoothed loss and two universal perturbation methods (UAP-Grad and UAP-PCA), and analyzes both input-dependent and universal attacks including PGD and FGM. The results show that AI decoders like ECCT and CrossMPT suffer large performance degradations under adversarial perturbations, with universal perturbations being particularly harmful and perturbations often transferring between AI decoders but not to BP baselines. The findings highlight a robustness cost behind AI-based decoding gains and motivate robustness-aware designs and broader channel-model considerations for practical deployments.

Abstract

Recent advances in deep learning have led to AI-based error correction decoders that report empirical performance improvements over traditional belief-propagation (BP) decoding on AWGN channels. While such gains are promising, a fundamental question remains: where do these improvements come from, and what cost is paid to achieve them? In this work, we study this question through the lens of robustness to distributional shifts at the channel output. We evaluate both input-dependent adversarial perturbations (FGM and projected gradient methods under $\ell_2$ constraints) and universal adversarial perturbations that apply a single norm-bounded shift to all received vectors. Our results show that recent AI decoders, including ECCT and CrossMPT, could suffer significant performance degradation under such perturbations, despite superior nominal performance under i.i.d. AWGN. Moreover, adversarial perturbations transfer relatively strongly between AI decoders but weakly to BP-based decoders, and universal perturbations are substantially more harmful than random perturbations of equal norm. These numerical findings suggest a potential robustness cost and higher sensitivity to channel distribution underlying recent AI decoding gains.

On the Fragility of AI-Based Channel Decoders under Small Channel Perturbations

TL;DR

The paper investigates the fragility of AI-based channel decoders under small, norm-bounded perturbations of the AWGN channel output, questioning the source of nominal gains over belief-propagation. It develops a robustness framework with a smoothed loss and two universal perturbation methods (UAP-Grad and UAP-PCA), and analyzes both input-dependent and universal attacks including PGD and FGM. The results show that AI decoders like ECCT and CrossMPT suffer large performance degradations under adversarial perturbations, with universal perturbations being particularly harmful and perturbations often transferring between AI decoders but not to BP baselines. The findings highlight a robustness cost behind AI-based decoding gains and motivate robustness-aware designs and broader channel-model considerations for practical deployments.

Abstract

Recent advances in deep learning have led to AI-based error correction decoders that report empirical performance improvements over traditional belief-propagation (BP) decoding on AWGN channels. While such gains are promising, a fundamental question remains: where do these improvements come from, and what cost is paid to achieve them? In this work, we study this question through the lens of robustness to distributional shifts at the channel output. We evaluate both input-dependent adversarial perturbations (FGM and projected gradient methods under constraints) and universal adversarial perturbations that apply a single norm-bounded shift to all received vectors. Our results show that recent AI decoders, including ECCT and CrossMPT, could suffer significant performance degradation under such perturbations, despite superior nominal performance under i.i.d. AWGN. Moreover, adversarial perturbations transfer relatively strongly between AI decoders but weakly to BP-based decoders, and universal perturbations are substantially more harmful than random perturbations of equal norm. These numerical findings suggest a potential robustness cost and higher sensitivity to channel distribution underlying recent AI decoding gains.
Paper Structure (33 sections, 3 theorems, 41 equations, 2 figures, 4 tables)

This paper contains 33 sections, 3 theorems, 41 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Preliminaries
  4. Robustness of AI Decoders to Input Perturbations
  5. Sample-wise adversarial objective.
  6. Randomized smoothing of the decoder optimization objective.
  7. Sample-wise adversarial attacks
  8. PGD Attack
  9. FGM Attack
  10. Universal adversarial perturbations
  11. UAP-Grad
  12. UAP-PCA
  13. Numerical Results
  14. Experimental Setup
  15. Performance Drop after Adversarial Attack
  16. ...and 18 more sections

Key Result

Proposition 4.1

Assume the loss is bounded: for all $u\in\mathbb{R}^n$, Define $\tilde{g}(u)\triangleq \mathbb{E}_{V\sim \mathcal{N}(0,\nu^2 I_n)}\bigl[\ell\bigl(f(u+V),x^\star\bigr)\bigr]$. Then $\tilde{g}$ has a Lipschitz-continuous gradient with constant i.e., $\bigl\|\nabla \tilde{g}(u)-\nabla \tilde{g}(u')\bigr\|_2 \le \frac{C}{\nu^2}\|u-u'\|_2$ for all $u,u'$.

Figures (2)

  • Figure 1: Comparison of FER-Inverse (1/FER) under various energy constraints for ECCT on Polar(128,64).
  • Figure 2: Comparison of FER-Inverse (1/FER) under various energy constraints for ECCT on LDPC(121,60).

Theorems & Definitions (6)

  • Proposition 4.1: Dimension-free smoothness via Gaussian smoothing
  • proof
  • Proposition 4.2
  • proof
  • Theorem 4.3: Concentration for UAP objective and UAP-PCA
  • proof