Estimating channel coefficients for complex topologies in 3D diffusion channel using artificial neural networks

Molecular communication via diffusion (MCvD) is one of the paradigms in nanonetworks. Finding an approximation or analytical solution for the fraction of the received molecules to analyze the channel behavior is essential in molecular communication. Current studies propose approximations to model simple channel topologies, i.e. topologies with few nodes. To model complex channel topologies, time-consuming particle-based Monte Carlo simulations are used. We propose MCvD-Transformer to avoid the time-consuming simulations and estimate the fraction of the received molecules for complex topologies. MCvD-Transformer is trained via instances containing various topologies and time-dependent estimations for a fraction of received molecules estimated by particle-based Monte Carlo simulations. Finally, MCvD-Transformer is compared with both the studies in the literature and the simulations. As a result, MCvD-Transformer performs better than literature studies in terms of root mean squared error and maximum normalized absolute error metrics on our test dataset. Therefore, the proposed model is more accurate in modeling complex MCvD topologies than the current state of the art without time-consuming simulations. Additionally, it is expected to be a benchmark for the works that focus on complex MCvD topologies.