Performance Evaluation of Mobile Molecular Communication System Using Neural Network Detector

This letter proposes a neural network (NN) detector for mobile molecular communication (MMC). The NN uses Broyden-Fletcher-Goldfarb-Shanno (BFGS), and Levenberg-Marquardt (LM) algorithms for optimization. In the proposed work, the received signal is filtered and three different techniques for supervised training and detection are used. The three different techniques are (a) the filtered signal, (b) slope values of the filtered signal, and (c) concentration difference values of the filtered signal within a bit interval. The trained NN is used for detecting the unknown bits for time-varying parameters such as the distance between transmitter (Tx) and receiver (Rx), noise, and the number of released molecules by the Tx. Bit error rate (BER) with the signal-to-noise ratio (SNR) is shown for different parameters such as coherence time of the channel, diffusion coefficient of the Tx, and the initial distance between the Tx and the Rx. Simulation results show that the BFGS algorithm trains the NN faster compared to the LM algorithm and the trained NN can perform well in a time-varying MMC environment.