Hybrid deep learning-based feature-augmented detection for molecular communication systems

This manuscript presents a hybrid Deep learning algorithm to detect transmitted symbols from experimental measured data in a biological molecular communication (MC) testbed. The algorithm consists of two deep neural networks capturing different aspects of the time signal, which are combined based on their relative confidence. The MC testbed uses a transfected Escherichia coli (E. coli) bacteria that express the light-driven proton pump gloeorhodopsin from Gloeobacter violaceus. Given an external controllable light stimulus, driven from a light-emitting diode (Transmitter), the bacteria secrete protons that change the pH level of the environment. A pH detector (Receiver) measures the pH of the environment. Modelling such a biological real system accurately is not feasible. Thus, in order to detect the transmitted bits we use both a convolutional and a recurrent neural network in tandem. This paper discusses the data augmentation, processing, and neural networks pertinent to a practical MC problem. The trained algorithm detects the transmitted bits with an accuracy above 99.9%.