Friendly Attacks to Improve Channel Coding Reliability

TL;DR

This work introduces a gradient-based 'friendly attack' that perturbs the transmitted, modulated codeword $S^N$ by an attack vector $a^N$ (added as $S^N + a^N$) and norm-normalizes to satisfy the power constraint $||C(S^N + a^N)||_2^2 \leq NP$, with updates $a^N_{t} \leftarrow a^N_{t-1} - \epsilon \nabla_S J(f(S + a^N_{t-1}), m)$. The attack is optimized to reduce decoding errors on average over the stochastic channel while preserving average transmitted power, and it is shown to improve reliability across LDPC, polar, and convolutional codes using diverse decoders (BP, NBP, ECCT, NBCJR) and modulation schemes (e.g., BPSK, 4-QAM) in AWGN, Rayleigh fading, and bursty-noise channels. The method leverages batch averaging over channel realizations and can incorporate clustering or multiple runs to obtain robust perturbations; the results indicate meaningful BER/BLER gains, particularly for suboptimal decoders and short block lengths, without altering the decoding algorithms themselves. Overall, the paper demonstrates a practical pathway to boost communication reliability by modifying transmitted signals in a hardware-agnostic way, with potential applications in downlink reliability or broadcasting where receiver upgrades are impractical.

Abstract

This paper introduces a novel approach called "friendly attack" aimed at enhancing the performance of error correction channel codes. Inspired by the concept of adversarial attacks, our method leverages the idea of introducing slight perturbations to the neural network input, resulting in a substantial impact on the network's performance. By introducing small perturbations to fixed-point modulated codewords before transmission, we effectively improve the decoder's performance without violating the input power constraint. The perturbation design is accomplished by a modified iterative fast gradient method. This study investigates various decoder architectures suitable for computing gradients to obtain the desired perturbations. Specifically, we consider belief propagation (BP) for LDPC codes; the error correcting code transformer, BP and neural BP (NBP) for polar codes, and neural BCJR for convolutional codes. We demonstrate that the proposed friendly attack method can improve the reliability across different channels, modulations, codes, and decoders. This method allows us to increase the reliability of communication with a legacy receiver by simply modifying the transmitted codeword appropriately.

Figures (6)

• Figure 1: The studied coded communication scenario with AWGN channel.
• Figure 2: The communication scheme with a friendly attack $a^N$.
• Figure 3: BER for LDPC code $(64, 32)$ with and without attack using BPSK (top) and $4$-QAM (bottom) modulations. 'BP-$n$' refers to BP decoder with $n$ iterations.
• Figure 4: BLER (top) and BER (bottom) for $(64, 32)$ and $(512, 256)$ polar codes with BPSK and $4$-QAM modulations.
• Figure 5: BLER (top) and BER (bottom) for $(64, 32)$ polar code over a fading channel with BPSK.