Hybrid Deep Learning for Channel Estimation and Power Allocation for MISO-NOMA System

In this paper, the influence of Deep Neural Network (DNN) in predicting both the channel parameters and the power factors for users in a Power Domain Multi-Input Single-Output Non-Orthogonal Multiple Access (MISO-NOMA) system is inspected. In channel prediction based Deep Learning (DL) approach, we integrate the Long Short Term Memory (LSTM) learning network into NOMA system in order that LSTM can be utilized to predict the channel coefficients. In addition, in Deep Learning based power estimation method, we introduce an algorithm based on Convolutional Neural Network (CNN) to predict and allocate the power factor for each user in MISO-NOMA cell. DNN is trained online using channel statistics in order to approximate the channel coefficients and allocate the power factors for each user, so that these parameters can be utilized by the receiver to recover the desired data. Besides, this paper demonstrates the framework where channel prediction based on LSTM layer and power approximation based on CNN can be jointly employed for multiuser detection in MISO-NOMA. In this work, Power factors are optimized analytically based on maximizing the sum-rate of users to derive the optimum power factors. Simulation outcomes for distinct metrics have verified the dominance of the channel estimation and power predication based DNN over standard approaches.