Deep Learning Surrogate for Fast CIR Prediction in Reactive Molecular Diffusion Advection Channels

Abstract

Accurate channel impulse response (CIR) modeling in molecular communication (MC) often requires solving coupled reactive diffusion-advection equations, which is computationally expensive for large parameter sweeps or design loops. We develop a deep-learning surrogate for a three-dimensional duct MC channel with reactive diffusion-advection transport and reversible ligand-receptor binding on a finite ring receiver. Using a physics-based partial differential equation (PDE)-ordinary differential equation (ODE) model, we generate a large CIR dataset across broad transport, reaction, and geometric ranges and train a neural network that maps these parameters directly to the CIR. On an independent test set, the surrogate closely matches reference CIRs both qualitatively and quantitatively: the empirical cumulative distribution function (CDF) of the normalized root mean square error (NRMSE) shows that 90% of test channels are predicted with error below 0.15, with only weak dependence on individual parameters. The surrogate therefore offers an accurate and computationally efficient replacement for repeated PDE-based CIR evaluations in MC system analysis and design.