Spectral efficiency and BER analysis of RNN based hybrid precoding for cell free massive MIMO under terahertz communication

In future wireless networks, integrating Terahertz (THz) communication with cell-free massive multiple-input multiple-output (CFMM) systems presents a promising approach to achieving high data rates and low latency. This paper investigates the use of recurrent neural network (RNN)-based hybrid precoding in CFMM systems operating in the THz band. The proposed method jointly designs analog and digital precoders to adapt to dynamic channel conditions and user mobility. However, THz communication is challenged by high path loss and sparse scattering, which complicate accurate channel estimation. To address this, the RNN is trained to predict optimal precoding weights by learning spatial and temporal channel patterns, thereby improving channel estimation and mitigating pilot contamination. Simulation results show that the proposed method achieves higher spectral efficiency and lower bit error rate (BER) than conventional techniques. Specifically, the RNN-based approach attains a spectral efficiency of 10 bps/Hz at a signal-to-noise ratio (SNR) of 30 dB, compared to 8.2 bps/Hz for minimum mean square error (MMSE) precoding. For 16-QAM, the RNN-based method achieves a BER of 10⁻⁶ at an SNR of 11 dB, while MMSE requires 12.5 dB to reach the same BER. Overall, the RNN-based hybrid precoding consistently outperforms traditional methods across various SNR levels, antenna configurations, and user densities, underscoring its potential in next-generation THz wireless systems.