A Low Complexity Deep Learning Based Precoder Designing for Enhanced Throughput in Underlay Network

Conventional optimization methods for precoder designing exhibits significantly poor performance in terms of computational cost and higher latency rate. The reason behind this shortcoming lies in the traditionally used approaches, which are associated with complex mathematical frameworks having higher computational costs. However, due to changes in channel parameters caused by the dynamic environment, the traditional techniques is not suitable for the 5G communication system. Thus, in order to facilitate the low complexity precoder designing method, deep learning (DL) based scheme is proposed to enhance the throughput in the MIMO-cognitive relay network (CRN). Here the secondary network
$(SN)$
accesses the spectrum of the primary network
$(PN)$
using underlay mode. DL maps the non-convex problem using a non linear function and predicts the precoder weights subject to the constraints of power and interference at the relay without the need of any complex mathematical framework. We demonstrate the efficacy of the proposed scheme by simulation results where the proposed scheme gives a comparable performance as a conventional scheme with lower computational complexity, whilst, it outperforms in terms of robustness when channel statistics changes dynamically.