Temporal Convolutional Network-Based Signal Detection for Magnetotactic Bacteria Communication System.

Area of Science:

• Biomedical Engineering
• Communication Systems
• Machine Learning

Background:

• Molecular communication (MC) utilizes signaling molecules for information transfer between biological entities.
• MC systems face challenges like inter-symbol interference (ISI) and external noise, hindering accurate mathematical modeling.
• Intractable channel state information (CSI) in MC necessitates advanced signal detection methods, particularly deep learning (DL).

Purpose of the Study:

• To propose a novel deep learning-based signal detection method for MC systems.
• To evaluate the performance of a modified Temporal Convolutional Network (TCN) for MC signal detection using magnetotactic bacteria (MTB) as information carriers.
• To compare the TCN detector's efficacy against other detection techniques, including Maximum A Posteriori (MAP), Deep Neural Networks (DNN), and Bidirectional Long Short-Term Memory (BiLSTM).

Main Methods:

• Development and implementation of a modified Temporal Convolutional Network (TCN) architecture.
• Signal detection within a specialized MC system employing magnetotactic bacteria (MTB).
• Comparative performance analysis against sub-optimal MAP, DNN, and BiLSTM detectors, considering bit error rate (BER) and computational complexity.

Main Results:

• The TCN-based detector achieved superior overall performance in MC signal detection.
• It demonstrated better bit error rate (BER) performance than sub-optimal MAP and DNN detectors.
• The TCN detector exhibited comparable BER to BiLSTM but with reduced computational complexity, and superior robustness against other DL-based detectors.

Conclusions:

• The modified TCN presents a highly effective and robust solution for signal detection in MC systems, particularly those using MTB.
• The TCN-based detector offers a favorable balance between performance (BER) and computational efficiency, outperforming BiLSTM in practical considerations.
• This deep learning approach significantly enhances the reliability and feasibility of molecular communication systems.