Friendly Attacks to Improve Channel Coding Reliability

Authors: Anastasiia Kurmukova, Deniz Gunduz

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide the results of successful friendly attacks for following scenarios: LDPC code n = 64, k = 32 for BP decoder with BPSK and 4-QAM modulations; polar code n = 64, k = 32 for BP, NBP and ECCT decoders with BPSK and 4-QAM modulations and also for fading channel with BPSK; long block length polar code with n = 512, k = 256 for BP and NBP with BPSK modulation; convolutional code k = 100, R = 1 / 2 with NBCJR decoder over AWGN and bursty channels. We show that the proposed friendly attack method can improve the reliability across different channels, modulations, codes, and decoders. All the simulations were run on Nvidia RTX A6000 GPUs with 48GB of memory.
Researcher Affiliation Academia Anastasiia Kurmukova and Deniz Gunduz Department of Electrical and Electronic Engineering, Imperial College London, London, UK a.kurmukova22@imperial.ac.uk, d.gunduz@imperial.ac.uk
Pseudocode Yes Algorithm 1: Search for Friendly Attack
Open Source Code No The paper mentions using the 'sionna library' (Hoydis et al. 2022) but does not provide a link to its own source code for the described methodology or state that it will be made publicly available.
Open Datasets No The paper uses generated data based on channel codes (LDPC, polar, convolutional) and simulated channels (AWGN, Rayleigh fading), rather than a named public dataset. No concrete access information for a public dataset is provided.
Dataset Splits No The paper does not provide specific dataset split information (percentages, sample counts, or citations to predefined splits) for training, validation, or testing.
Hardware Specification Yes All the simulations were run on Nvidia RTX A6000 GPUs with 48GB of memory.
Software Dependencies No The paper mentions using 'sionna library' and 'scikit-learn library implementation' but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes The choice of parameters I, B, and a scheduler for ϵ defines an attack. We can highlight four different approaches: 1. In the first approach, we employ a rather large batch size B 1000 10000 and relatively small number of iterations I 1 100. 2. The second approach needs an average batch size B 100 500 and relatively high number of iterations I 500 5000.