Abstract: A molecular communication channel is determined by the received signal. Received signal models form the basis for studies focused on modulation, receiver design, capacity, and coding, all of which depend on accurate signal modeling. Therefore, it is crucial to model the number of received molecules until time $t$ analytically. Modeling the diffusion-based molecular communication channel with the first-hitting process remains an open issue for spherical transmitters. In this paper, we utilize artificial neural networks (ANNs) to model the received signal for a spherical transmitter and a perfectly absorbing receiver. The proposed technique may be utilized in other studies that assume a spherical transmitter instead of a point transmitter.
Introduction
In Molecular Communication via Diffusion (MCvD), information is encoded into the release of molecules that travel through a fluidic medium. To design effective communication systems, engineers require accurate mathematical models of the "received signal"—the cumulative count of molecules that reach the receiver over time.
Historically, most models have relied on a "point transmitter" assumption for mathematical convenience, as it preserves circular symmetry and simplifies the underlying physics. However, real-world transmitters possess physical dimensions. A spherical transmitter body not only occupies space but also acts as a reflector, obstructing and redirecting molecules. This breaks the symmetry required for traditional analytical solutions of the "first-hitting process," where a molecule is absorbed upon its first contact with the receiver. This paper addresses this gap by employing machine learning to model the complex physics of spherical transmitters.
Methods
The researchers proposed a two-phase hybrid framework that integrates physical modeling with machine learning.
Phase 1: Simulation and Curve Fitting. Extensive 3D simulations were conducted to generate a dataset of molecular propagation. The study utilized 135 unique scenarios varying distance ($d$), transmitter radius ($r^{Tx}$), receiver radius ($r^{Rx}$), and diffusion coefficient ($D$), with each scenario replicated 500 times for statistical accuracy. Two analytical models were proposed: a "Primitive Model," which scales the existing point-source formula, and an "Enhanced Model," which parameterizes the diffusion coefficient and time components to account for transmitter interference. Nonlinear least squares curve fitting was used to determine the optimal parameters ($b_i$) for these models based on the simulation data.
Phase 2: ANN Training. An Artificial Neural Network (ANN) was trained using Bayesian regularization backpropagation. The ANN learns to predict the optimal $b_i$ parameters for the analytical models based on system inputs. This approach ensures the output remains physically grounded while benefiting from the predictive power of machine learning.
Key Results and Discussion
The "Enhanced Model" significantly outperformed both the "Primitive Model" and the traditional point-source formula. A critical insight is the "reflector effect": the physical body of a spherical transmitter reflects molecules that would otherwise wander away, actually increasing the probability of a hit at the receiver.
As illustrated in the performance analysis, the Enhanced Model (both fitted and ANN-predicted) closely tracks the simulation data across various distances. In contrast, the Point Source 3-D Formula consistently underestimates the received signal, particularly at closer ranges where the transmitter's physical presence is most impactful. Furthermore, the model accurately predicts the Signal-to-Interference (SIR) ratio, a vital metric for mitigating Inter-Symbol Interference (ISI) in molecular networks.
Conclusion
By combining physics-inspired analytical forms with the predictive capabilities of ANNs, this research provides a robust solution to the geometry-induced modeling challenges in molecular communication. While the current work focuses on single transmitter-receiver pairs, the framework provides a blueprint for modeling more complex environments, including multi-user scenarios and non-ideal biological receptors.
Summary updated