Robust Long Short-Term Memory-Enabled Beamforming for Cell-Free Massive MIMOs in 6G Networks

This paper presents a performance evaluation of a long short-term memory (LSTM)-based precoder for cell-free (CF) massive multiple-input multiple-output (MIMO) systems in 6G networks operating under hardware impairments and imperfect channel state information (CSI). It also compares the proposed method with traditional Kalman, minimum mean square error (MMSE), and zero forcing (ZF) precoders. Simulations conducted at 2.4 GHz show that the LSTM-based scheme offers improved spectral efficiency (SE) and energy efficiency (EE) while remaining computationally feasible. Specifically, the LSTM precoder achieves an average per-user SE of 1.74 bps/Hz, representing gains of about 1.15% over Kalman, 3.45% over MMSE, 4.6% over ZF, and 5.75% over MRT. Under severe hardware impairments, it provides a 2.94% improvement over Kalman and a 5.88% improvement over MMSE. The total SE reaches 17.4 bps/Hz, increasing the overall system capacity by approximately 2.87% over Kalman, 4.02% over MMSE, 6.32% over ZF, and 8.05% over MRT when the number of users (K) is 10. The LSTM-based precoder also achieves the highest peak EE, indicating that its learning-driven adaptability yields higher SE for comparable power usage. Despite a slight increase in power consumption, its inference time remains shorter than both MMSE and ZF, offering a favorable balance between performance and computational complexity. Overall, the results demonstrate that a learning-driven, impairment-aware precoding approach provides significant advantages in terms of robustness and scalability for next-generation 6G CF massive MIMO networks, particularly in non-ideal hardware environments.