Channel modeling for diffusion-based molecular MIMO communications using deep learning

Diffusion-based molecular communication (MC) system present immense potential and broad application prospects in the field of biomedicine, such as drug delivery. Molecular multiple-input multiple-output (MIMO) communication system can improve the reliability of communication in the environment. However, the channel modeling for diffusion-based molecular MIMO communication system is challenging. Most of the work on the modeling of molecular MIMO channels focuses on the traditional derivation of the channel impulse response (CIR). In this paper, we take into account an M x N molecular MIMO communication system with symmetric and asymmetric topologies. A deep neural networks (DNN) based model and Transformer-based model are proposed to predict the channel parameters in the CIR of this molecular MIMO system under different molecule types (DMT) and same molecule types (SMT), respectively. Simulation results show that the DNN-based model has best accuracy of prediction than the Transformer-based model and Long Short-Term Memory (LSTM) model under DMT. In particular, the Transformer-based model outperforms the DNN-based model and LSTM model under SMT.