A Neural Network Approach for Spectral and Energy Efficient Multiple Antenna Systems

In this research, we investigate a low computational complexity user selection and a precoder approach that, respectively, achieves spectral/energy transmission efficiency in the multiple-antennas system utilizing supervised learning techniques. In particular, the user selection spectral efficient network employs a supervised classification approach to correspond (one to one) between each input class-label and a set of selected users where every realization of the channel is labeled with a class. On the other hand, the energy reserving precoder is a two-stage scheme. In the first one, a conventional deep-learning-based multiple antenna framework is applied to assess the uplink power allocation vector subject to service quality targets (QoS) of the individual users. In the second stage, we utilize the Lagrangian duality method to calculate the optimal precoder vector for downlink. The research outcomes recommend applying an adaptive system with a convolutional neural scheme for low target QoS users. For instance, at 1 dB target SINR signal, there is about 2.6 dBm gain in the performance than the null-steering scheme. We may, on the other hand, use a simple the null-steering strategy for users with high target QoS. Consequently, we can deduce that the proposed method provides a balance of energy efficiency and computational complexity.