Hybrid Recurrent Neural Network for Signal-Dependent Noise Suppression in Molecular Communication

Molecular communication (MC) employs chemical molecules for information transfer in environments where electromagnetic signals are ineffective. However, the diffusion mechanism introduces signal-dependent noise (SDN), complicating accurate signal recovery. Traditional model-based methods struggle to handle SDN’s complex dynamics and depend heavily on optimal parameter tuning, limiting their adaptability to temporal variations. To tackle these challenges, this paper introduces a hybrid recurrent neural network (RNN) model that effectively captures both short- and long-term dependencies within MC signals, surpassing the performance of single RNN models and traditional approaches. This model offers a promising data-driven solution for noise mitigation in MC, with its effectiveness validated through numerical simulation results.