Deep Learning Based Joint Precoding, Channel Estimation and Equalization for Millimeter-WaveMassive MIMO OFDM System

Utilizing a lens antenna array, millimeter-wave massive multiple-input multiple-output<br>(MIMO) aims to minimize the count of radio frequency (RF) chains, posing a challenge due<br>to the lower RF chain count compared to antennas. Beamspace channel estimation is defined<br>as a sparse signal recovery problem by leveraging the sparsity of beamspace channels. In<br>multi-user millimeter-wave (mmWave) MIMO-OFDM systems, achieving a cost-effective<br>and efficient hybrid precoding for optimal sum-rate is crucial. Traditional methods for hybrid<br>precoding often rely on greedy or optimization techniques, which suffer from high complexity or suboptimal performance. In this research work, optimized DL based channel estimation,<br>channel equalization and hybrid precoding for Millimetre-wave massive multiple-input multiple-output (MIMO) NOMA/OFDM systems is proposed. The proposed method consists of<br>three phases: hybrid precoding, channel equalization and Channel Estimation (CE).Sheaf attention-based Gorilla Troopsspiking neural network is proposed for CE and Patch-Based<br>Cross-Layer Non-Local Attenuation-based C-BiLSTMfor executing the hybrid precoding,<br>utilizing the estimated channel vectors for training.Eventually, the simulation results will be<br>analyzed for the devised sparse channel estimation and hybrid precoding in the mmWave<br>massive MIMO-OFDM communication system to achieve lower errors and enhanced spectral<br>efficiency compared to existing approaches.