Performance Analysis of Deep Learning Detector for MIMO Cooperative Relay Communications with Imperfect CSI

Abstract

Cooperative communication is one of the promising approaches for achieving high data rates and efficient bandwidth utilization, but introducing relay nodes in the ar chitecture brings a challenge in physical layer security. Scholars propose different approaches like a secure beamforming model and a combination of beamforming and jamming using artificial noise to overcome this challenge. The channel state information (CSI) of the eavesdropper and the legitimate user is necessary for the se crecy of the transmission, but in reality, the eavesdropper is always passive, and the channel state information is difficult to obtain, and the channel state information of the legitimate user is outdated. This thesis proposes a secure multiple input multiple output (MIMO) communication system to overcome security threats during cooper ation with the relay node. A zero-forcing algorithm is used to secure leakage to the eavesdropping relay node by transmitting on null space using the beamforming tech nique. The deep convolutional neural network (DCNN) is trained with the imperfect version channel state information to produce the perfect channel state information then the input bit is recovered using a maximum likelihood detector. The Simulation was done for different performance factor parameters like imperfect correlation fac tor, doppler frequency, and the number of antennas to show the BER performance of the system. The results show that the deep convolutional neural network detector has a gain performance about 2dB in higher correlation factor and about 10.5dB in lowest imperfect correlation factor than the standard maximum likelihood detector