Fast Channel Simulation via Error-Correcting Codes
Authors: Sharang Sriramu, Rochelle Barsz, Elizabeth Polito, Aaron Wagner
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 3.4 Experimental Results We run Polar Sim on the reverse (PY |X) version of three channels: (1) the BSC with a uniform input (2) the binary erasure channel, X = Z Y , where Y is uniform over { 1, 1} and Z is Bernoulli(ϵ), and (3) the binary Gaussian channel X = Y + Z, where Y is again uniform over { 1, 1} and Z is N(0, σ2). Note that the reverse of the BSC with a uniform input is the BSC itself. Fig. 3 and Fig. 4 show the rate performance of these simulations. |
| Researcher Affiliation | Academia | Sharang M. Sriramu School of ECE Cornell University Ithaca, NY 14853 sms579@cornell.edu Rochelle Barsz School of ECE Cornell University Ithaca, NY 14853 rsb359@cornell.edu Elizabeth Polito School of ECE Cornell University Ithaca, NY 14853 emp234@cornell.edu Aaron B. Wagner School of ECE Cornell University Ithaca, NY 14853 wagner@cornell.edu |
| Pseudocode | Yes | Algorithms 1 and 2 describe the complete scheme. |
| Open Source Code | Yes | We include code as part of our supplementary material with instructions on how to run it. |
| Open Datasets | No | The paper uses synthetically generated data (e.g., 'BSC with a uniform input', 'binary erasure channel', 'binary Gaussian channel') rather than pre-existing public datasets, so no external access information is provided or needed. |
| Dataset Splits | No | The paper does not specify training, validation, or test splits, as it primarily uses simulated data based on channel characteristics rather than splitting a fixed dataset. |
| Hardware Specification | No | It is noted in the paper that all experiments were run on a consumer-grade CPU. |
| Software Dependencies | No | The paper mentions using 'Dr. Henry Pfister s polar code implementation' but does not provide specific version numbers for general software dependencies (e.g., Python, PyTorch, etc.) required for replication. |
| Experiment Setup | Yes | We run Polar Sim on the reverse (PY |X) version of three channels: (1) the BSC with a uniform input (2) the binary erasure channel, X = Z Y , where Y is uniform over { 1, 1} and Z is Bernoulli(ϵ), and (3) the binary Gaussian channel X = Y + Z, where Y is again uniform over { 1, 1} and Z is N(0, σ2). ... We implement Algorithm 3 from Flamich [2024]. The proposal distribution P is chosen to be i.i.d. Bernoulli(1/2) with n = 8. Given the input Xn = xn, the target distribution Q(yn) is chosen to be Qn i=1 BSCp(yi|xi), where p ranges over (0, 1/2). ... One of the virtues of the Polar Sim is that there are no hyperparameters to choose. |