Grant-Free NOMA with Device Activity Learning Using Long Short-Term Memory
Xiaqing Miao, Dongning Guo, Xiangming Li
Abstract
Non-orthogonal multiple access (NOMA) is a promising technique for future cellular networks. A major challenge in the uplink of grant-free NOMA is to identify all active devices as well as to decode their data. In the Internet of Things (IoT), the on-off activities of devices are predictable to various degrees. In this letter, a deep learning algorithm is employed to predict the device activities in the current slot by exploiting the history data. The prediction results are applied as input priors to a modified orthogonal matching pursuit (OMP) algorithm for joint device identification and data detection. Numerical simulation results demonstrate that the error rate is reduced to at least ten times as compared with conventional compressed sensing based algorithms at the same signal-to-noise ratio.