Deep Learning Based Signal Detection In NOMA Systems: A Review

This review examines the state-of-the-art techniques that utilize Deep Learning (DL) to decode non-Orthogonal multiple access signals (NOMA). NOMA achieves significant performance benefits in terms of system throughput and huge connectivity. NOMA has been regarded as a very viable contender for the Fifth Generation (5G) and future wireless communications technology. Users in NOMA have access to all subcarriers and are not separated from one another according to the orthogonality concept. In a NOMA transmission, packets from every user are superimposed to create a single signal or stream. In order to separate user data at the receiver, a more complex algorithm was needed because NOMA serves numerous users simultaneously at the same frequency, but with distinct power allocations. One of the tried-and-true NOMA detection strategies is the successive interface cancellation (SIC). Incoming packets are first sorted according to their channel quality or quality of service (QOS) requirements. Next, the user with the most stringent criteria is found, with the assumption that all other users are noise. Perfect channel state tracking is necessary for SIC-based approaches. Imperfect (SIC) causes error in channel estimation and signal detection. This paper addresses recent studies that use DL architectures rather than traditional detectors. A DL-based NOMA receiver is made to decode messages for numerous users in a single operation without explicitly estimating channels. Recurrent neural networks that can learn order dependency in sequence prediction tasks are known as Long Short-Term Memory (LSTM) networks. A challenging deep learning topic is LSTMs, which will be addressed in this work.