Clustering Guided Residual Neural Networks for Multi-Tx Localization in Molecular Communications

This repository contains the implementation code for multi-transmitter localization in molecular communication systems using clustering-guided residual neural networks.

Table of Contents

• Overview
• Paper Information
• Project Structure
• Dataset

---

Letter Information

Title: Clustering Guided Residual Neural Networks for Multi-Tx Localization in Molecular Communications

Focus Areas:

• Multi-transmitter localization in diffusion-based molecular communication
• Integration of clustering algorithms with deep learning
• Evaluation across K=2, 3, and 4 transmitter configurations

Key Contribution: The work demonstrates that residual neural networks can effectively correct imperfect clustering estimates by learning correction patterns from limited simulation data, leading to significant improvements in localization accuracy.

---

Project Structure

letter-repo/
├── README.md                          # This file
├── model_2tx.ipynb                    # K=2 transmitters pipeline
├── model_3tx.ipynb                    # K=3 transmitters pipeline
├── model_4tx.ipynb                    # K=4 transmitters pipeline
├── dataset/
│   ├── config.csv                     # Ground-truth TX configurations
│   ├── 0.csv                          # Particle trajectory data (scenario 0)
│   ├── 1.csv                          # Particle trajectory data (scenario 1)
│   ├── ...
│   ├── 9.csv                          # Particle trajectory data (scenario 9)
└── └── angle_results/
        └── all_results_enhanced.csv   # Wide format clustering results

---

Dataset

The dataset is located in the dataset/ directory and contains:

config.csv

Ground-truth configuration for all 10 simulation scenarios:

Column	Description	Values
step_time	Simulation timestep	1e-06 s
radius	Spherical receiver radius	5.0 μm
diffusion_coef	Diffusion coefficient	79.4 μm²/s
tx_count	Number of transmitters	K ∈ {2, 3, 4}
tx_centers	Ground-truth TX positions (3D)	List of K [x, y, z] coordinates in μm
molecule_counts	Particles emitted per TX	[10000, 10000, ...] (K values)
rx_center	Receiver position	[0, 0, 0]
from	Actual captured particle counts	Per-TX capture counts

Data CSVs (0.csv through 9.csv)

Individual particle trajectory data for each scenario:

Column	Description
x, y, z	3D particle position in μm
time	Simulation time in seconds
tx_index	Source transmitter index (0 to K-1)

Note: This dataset is a subset of the full dataset used in the paper. It provides representative scenarios for demonstrating the method. Dataset for 3 and 4 TXs are in the same format.

uniform_test/angle_results/ (Clustering Results)

Directory containing clustering and angle estimation results processed by convert_wide_to_long():

all_results_enhanced.csv (Input - Wide Format)

Raw clustering results in wide format. Each row represents a scenario with clustering method outputs (e.g., RAW, KMeans centers, ANGLE-ONLY, SIZE-ONLY, ANGLE+SIZE).

When processed by convert_wide_to_long(), this is converted to long format with the following columns:

Column	Description
file_index	Scenario index (0-9)
tx_index	Transmitter index (0 to K-1)
method	Clustering/correction method (e.g., "RAW", "KMeans centers", "ANGLE-ONLY", "SIZE-ONLY", "ANGLE+SIZE")
est_size	Estimated particle count for this TX
est_center	Estimated 3D center position [x, y, z] for this TX
true_size	Ground-truth particle count for this TX (filled from config.csv)

The long format output is saved as size_estimation_*Tx.csv for further model training and evaluation.