A Deep-Learning Based Approach to Resource Allocation in NOMA Based Cognitive Radio Network with Heterogeneous IoT Users

In 5G mobile technology, the expansion of Internet of Things (IoT) has created a huge need for a wide variety of latency, dependability, and energy efficiency requirements, etc… Spectrum efficiency (SE) of such large scale network needs to be improved with an economical power consumption. The non-orthogonal multiple access (NOMA) technique is utilized to enhance system efficiency (SE) by merging several users in the same frequency. An energy efficient (EE) resource allocation (RA) problem has been formulated for NOMA based heterogeneous IoT networks. Using the examining technique of Cognitive Radio (CR) Network, a stepwise RA scheme is assigned for IoT Users (IoTUs) and Mobile Users (MUs) with the mutual interference management. Later, to find a solution quickly and flawlessly, a deep recurrent neural network (RNN) based mechanism has been proposed. Furthermore, to systemize the approach of heterogeneous users, a rate and precedence demands based scheduling method has been implemented. Extensive results demonstrate that the deep learning based framework performs better than traditional RA methods in terms of computational complexity. On comparing with the prevailing OFDMA technique, the NOMA system with the imperfect SIC provides an acceptable performance on the EE at the cost of low EE and high power consumption.