Deep Learning-Based Spreading Sequence Design and Active User Detection for Massive Machine-Type Communications

Abstract

In this letter, we propose a deep learning-based spreading sequence design and active user detection (AUD) to support massive machine-type communications (mMTC) where a large number of devices access the base station using non-orthogonal spreading sequences. To design the whole communications system minimizing AUD error, we employ an end-to-end deep neural network (DNN) where the spreading network models the transmitter side and the AUD network estimates active devices. By using the AUD error as a loss function, network parameters including the spreading sequences are learned to minimize the AUD error. Numerical results reveal that the spreading sequences obtained from the proposed approach achieve higher AUD performance than the conventional spreading sequences in the compressive sensing-based AUD schemes, as well as in the proposed AUD scheme.