Flexible Autoencoder for Orthogonal Chirp Division Multiplexing Communications

This paper introduces a flexible autoencoder (AE) designed for orthogonal chirp division multiplexing (OCDM). The AE is based on a deep neural network (DNN) and a flexible structure of OCDM, which can accommodate various structures for, such as orthogonal frequency division multiplexing, chirp spread spectrum, affine frequency division multiplexing etc. The AE is designed by jointly training a mapper and a demapper made up of a DNN in the flexible structure. Comparing with traditional M-ary quadrature amplitude modulation (MQAM) systems, the trained AE shows different shapes of constellation for additive white Gaussian noise channels and outperforms the traditional MQAMs in bit error rates.