Investigation of Bi-Directional LSTM deep learning-based ubiquitous MIMO uplink NOMA detection for military application considering Robust channel conditions

Abstract

Ubiquitous multiple-input multiple-output (MIMO) non-orthogonal multiple access (NOMA) networks (UMNs) have emerged as an important technology for enabling security and other applications that need continuous monitoring. Their implementation, however, could be obstructed by the limited bandwidth available due to many wireless users. In this paper, bidirectional long short-term memory (LSTM)-based MIMO-NOMA detector is analyzed considering imperfect successive interference cancelation (SIC). Simulation results demonstrate that the traditional SIC MIMO-NOMA scheme achieves 15 dB, and the deep learning (DL) MIMO-NOMA scheme achieves 11 dB for ${10}^5$ number of iterations. There is a gap of 4 dB which means that the DL-based MIMO-NOMA performs better than the traditional SIC MIMO-NOMA techniques. It has been observed that when the channel error factor increases from 0 to 1, the performance of DL decreases significantly. For the channel error factor value less than 0.07, the DL detector performance much better than the SIC detector even though the perfect channel state information (CSI) is considered. The DL detector’s performance decreases significantly where variations between the actual and expected channel states occurred, although the DL-based detectors’ performance was able to sustain its predominance within a specified tolerance range.