Masked Neural Detection for Run-Length-Limited Channel Coding in Molecular Communication
Abstract
Molecular communication (MC) suffers from severe diffusion memory because molecules released for one symbol may arrive during later symbol intervals. Neural sequence detectors, especially sliding bidirectional recurrent neural networks (SBRNNs), substantially outperform threshold detection in such channels. This raises a central question for MC channel coding: does a code whose superiority was established under threshold detection retain it when both coded and uncoded transmission are evaluated with neural detection? This letter answers this question for run-length-limited ISI-mitigation (RLIM) codes by proposing a decoder-aware training mask that removes the positions the RLIM decoder has a high probability of deterministically overwriting, steering compact-SBRNN capacity toward the information-bearing positions. The masked RLIM2-SBRNN beats the best uncoded receiver (threshold or SBRNN) at 40 of 57 operating points; gains peak at 43× under favorable channels, while losses, confined to the most adverse, never exceed 2.7×. Masking improves the unmasked RLIM2-SBRNN in 56 of 57 matched comparisons. Finally, with storage counted equally in SBRNN weights and MLSE table entries, the masked RLIM2-SBRNN is the more accurate receiver up to a few thousand stored values despite using no channel knowledge; channel-state-aware uncoded MLSE moves ahead only beyond tens of thousands.
Related
- Litter-Masked Block z-Channel: Designing Cover Distributions Against Public-Design Observers
- Finite-blocklength classical communication over the quantum erasure channel with and without classical feedback
- Limited detection of Gulf War illness-like behavioral phenotypes in Charles River male Sprague-Dawley rats across multiple chemical and stress exposure paradigms
- Covert Communications over Enormous Fluid Antenna Systems (E-FAS): Impact of Channel Hardening on Detection
- Array Rotation in Near-Field ISAC: Communication, Detection, and Estimation Under Low-Complexity Transmit Architectures
BibTeX
@misc{sahin2026masked,
title = {{Masked Neural Detection for Run-Length-Limited Channel Coding in Molecular Communication}},
author = {Melih Şahin and Ozgur B. Akan},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.12489},
eprint = {2606.12489},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.12489},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Masked Neural Detection for Run-Length-Limited Channel Coding in Molecular Communication
AU - Melih Şahin
AU - Ozgur B. Akan
PY - 2026
DA - 2026/09/06
PB - arXiv
DO - 10.48550/arxiv.2606.12489
UR - https://arxiv.org/abs/2606.12489
AB - Molecular communication (MC) suffers from severe diffusion memory because molecules released for one symbol may arrive during later symbol intervals. Neural sequence detectors, especially sliding bidirectional recurrent neural networks (SBRNNs), substantially outperform threshold detection in such channels. This raises a central question for MC channel coding: does a code whose superiority was established under threshold detection retain it when both coded and uncoded transmission are evaluated with neural detection? This letter answers this question for run-length-limited ISI-mitigation (RLIM) codes by proposing a decoder-aware training mask that removes the positions the RLIM decoder has a high probability of deterministically overwriting, steering compact-SBRNN capacity toward the information-bearing positions. The masked RLIM_2-SBRNN beats the best uncoded receiver (threshold or SBRNN) at 40 of 57 operating points; gains peak at 43× under favorable channels, while losses, confined to the most adverse, never exceed 2.7×. Masking improves the unmasked RLIM_2-SBRNN in 56 of 57 matched comparisons. Finally, with storage counted equally in SBRNN weights and MLSE table entries, the masked RLIM_2-SBRNN is the more accurate receiver up to a few thousand stored values despite using no channel knowledge; channel-state-aware uncoded MLSE moves ahead only beyond tens of thousands.
ER -