A Deep-Learning Method for Device Activity Detection in mMTC Under Imperfect CSI Based on Variational-Autoencoder
Tianyu Zhao, Feng Li, Peiting Tian
Abstract
Large-scale deployment of massive device connectivity is a crucial communication challenge for Internet of Things (IoT) networks, which consist of a huge number of devices with sporadic traffic. In massive Machine Communication Scenario (mMTC), it is very important for the serving base-station (BS) to identify the active devices in each coherence block. This paper proposes a deep neural network (DNN) based on variational autoencoder (VAE) for device activity detection in mMTC under imperfect channel state information (CSI). A framework of variational optimization is constructed and the learning network structure is also designed. The derivation on the loss function for network training is presented and numerical results are provided to illustrate the accuracy of our method. The performance demonstrates the merits of the proposed method by comparison with the traditional compressed sensing algorithms, which are widely applied in multi-user detection.