Deep Learning for Parametric Channel Estimation in Massive MIMO Systems

Massive Multiple-Input Multiple-Output (MIMO) communication with a low bit error rate depends upon the availability of accurate Channel State Information (CSI) at the base station. The massive MIMO systems can be either deployed using time division duplexing with channel reciprocity assumption or by availing frequency division duplexing, which requires closed-loop feedback for CSI acquisition. The channel reciprocity simplifies transmission in time division duplexing; however, it suffers a bottleneck due to pilot contamination, whereas transmission in frequency division duplexing is challenged by channel estimation complexity, CSI feedback, and overall delay in CSI transfer. This paper proposes a simplified parametric channel model, its deep neural network aided estimation along with pilot decontamination for time division duplexing and a low rate parametric feedback and improved precoding for frequency division duplexing based massive MIMO systems. This novel framework integrates the massive MIMO parametric estimation and deep learning for improved estimation and precoding. Our proposed model also offers a unified approach for CSI acquisition with a performance bound on channel correlation in fast time-varying conditions. A theoretical model has been presented using Gaussian assumptions and validated by Monte-Carlo simulations. The results show total nullification of pilot contamination and high-performance gains when the proposed technique is employed for estimation