Optimal-HYPnet: Hybrid deep learning technique for jointly optimizes channel estimation and robust signal detector in OFDM system

Abstract

In orthogonal frequency-division multiplexing (OFDM) systems, obtaining precise channel state information (CSI) is difficult. The number of antennas raises the pilot training and computing complexity of traditional channel estimation algorithms. Such inadequate characteristic makes the signal detection and channel estimation become challenging issues for the high-rate and reliable transmissions. Recently, several works have proposed for jointly optimized channel estimation and signal detector. Since the channel estimation performance might be degraded due to complex methodologies. This paper proposes a hybrid deep learning method for jointly optimizes channel estimation and signal detector(Optimal-HYPnet) which is used in 5G communication systems. First, we develop a reliable and scalable bi-directional convolutional neural network (Bi-CNN) to assess the channel in multipath circumstances. Then, develop an improved seagull optimization (ISO) algorithm to reduce the pilot overhead by using scheduling the pilot phase shifts optimally. Moreover, we illustrate a near-optimal buffer-aided LRT (NOLRT) detector to estimate channel information which is robust to the channel estimation error. Finally, we compare our proposed Optimal-HYPnet performance to existing approaches in terms of MSE and BER.