Synchronization and detection in molecular communication using a deep-learning-based approach

Abstract

The concept of Internet of Bio-Nano Things (IoBNT) has emerged due to its revolutionary possibilities that transcend traditional wireless communication systems. Molecular Communication (MC) arises as a potential centrepiece for this paradigm, enabling applications in challenging environments. However, this type of communication, which often relies on molecular diffusion, suffers from a high inter-symbol interference (ISI), which deteriorates the reliability of the transmission. To cope with the strong ISI as well as the typical short coherence time of the MC channel, this work considers the adoption of a data-driven approach to accomplish non-coherent based detection at the receiver. In particular, we investigate the performance of a low complexity one-dimensional Convolutional Neural Network (1-D CNN) based in dilated causal convolutional layers and of a Gated Recurrent Unit based Recurrent Neural Network (GRU-RNN) aimed at the tasks of symbol detection and synchronisation, comparing the results with a conventional non-coherent detection. Initially, we study the performance of the proposed Neural Networks (NNs) based detectors assuming prior synchronisation between the transmitter and the receiver and, afterwards, we extend the approach for scenarios without prior synchronisation. Furthermore, we also investigate the robustness of the proposed NNs schemes against unknown variations in the distance between the transmitter and the receiver as well as in the diffusion coefficient. Finally, the results presented in this work lead to the conclusion that the implementation of NNs for both synchronisation and non-coherent detection can be a very effective approach for the challenging MC channel, ensuring more robustness than conventional model-based approaches.