Using Deep Learning to Demodulate Transmissions in Molecular Communication

Abstract

Molecular communication presents a new approach for data transmission between miniaturised devices, especially in the context of medical applications. A communication link is established using molecules, or other particles in the nanoscale, to modulate information. Due to a lack of data or changing physical parameters, the information channel often cannot be modelled accurately. Deep Learning provides a solution to receive a transmitted data sequence without the need for an analytical description of the channel. We present a proof-of-concept for the application of a Convolutional Neural Network to demodulate a signal using concentration shift keying. The demodulation predictor is evaluated with experimental data from a testbed using magnetic nanoparticles in an active background flow in comparison to a conventional learning approach with Linear Discriminant Analysis. The new demodulator shows a better performance for higher symbol rates than the conventional approach. Using a modulation alphabet with 8 symbols a data rate of more than 5.5 bit s <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−1</sup> can be achieved. The constructed neural network can be trained in under two minutes and can easily be adapted to changing transmission parameters.