Molecular communication data augmentation and deep learning based detection

This manuscript presents a novel model for generating synthetic data for a biological molecular communication (MC) system to train a Neural Network (NN) for the purpose of discriminating transmitted bits. To achieve this, a deep learning algorithm was trained using the synthetic data and tested against experimentally measured data. The polynomial curve fitting coefficients are chosen as features. The featurization stage is followed by a NN that captures different aspects of the temporal correlation of the received signals. The real data was collected from an MC testbed that employed transfected Escherichia coli (E. coli) bacteria expressing the light-driven proton pump gloeorhodopsin from Gloeobacter violaceus. By stimulating the bacteria with externally controlled light, protons were secreted, which changed the pH level of the environment. A pH detector was then used to measure the pH of the environment. We propose the use of a deep convolutional neural network to detect the transmitted bits. This paper discusses the data augmentation, processing, and NNs that are pertinent to practical MC problems. The trained algorithm demonstrated an accuracy of over 99.9% in detecting transmitted bits from received signals at a bit rate of 1 bit/min, without requiring any specific knowledge of the underlying channel.