Deep Learning-Based Signal Detection With Denoising and Temporal Sequence Structuring for Molecular Communication in Vessel-Like Channels

Molecular communication via diffusion (MCvD) in vessel-like channels is strongly affected by laminar flow, diffusion-induced dispersion, and severe intersymbol interference (ISI). These challenges are even more pronounced in scenarios with multiple transmitters, where molecules of the same type released from multiple transmitters overlap at the receiver, giving rise to accumulated interlink interference (ILI). However, existing signal detection studies have mainly focused on the release of different types of molecules or single-symbol detection. In this article, we propose a Transformer-based sequence detector with a denoising and temporal sequence structuring (DTSS) strategy for multiple transmitters that release the same type of molecules in vessel-like MCvD channels with Poiseuille flow. The proposed approach structures raw molecular sequences into consecutive segments and uses a self-attention Transformer to model long-range ISI and ILI, thereby enhancing robustness to multiple transmitters superposition. Simulations show that the proposed Transformer-based model with the DTSS strategy converges the fastest and achieves the best detection performance among the baseline deep neural network (DNN) and Transformer models, the Transformer with Denoised and Grouped strategies, and the DNN with DTSS preprocessing.