CSI Feedback Prediction Using UE Aware Sparse Neural Network Framework

In wireless communication systems, especially in the frequency division duplex (FDD) mode of massive MIMO, the channel state information (CSI) information is not available at the base station. This information is important for the base station to be able to provide high data rates and good spectral efficiency. Therefore, it becomes important for the user equipment (UE) to send this information to the base station (BS). With the increasing size of the channel matrix especially in MIMO systems, there is a need to use deep learning for sending the channel information. We therefore propose a sparse neural network framework model that can encode the CSI at the UE and decode it at the base station accurately even at high compression compared to conventional methods. The proposed framework can adjust the neural network model at the UE side using a novel blockchain-enabled CSI encode refinement system (CERS). CERS enables to optimization of the computational complexity and accuracy of the encoder. Simultaneously, CERS can store multiple encoder configurations in an immutable data structure using blockchain. An exhaustive evaluation of the proposed sparse neural network framework reduces the computational complexity by 57% better than previous state-of-the-art deep learning-based CSI prediction. Additionally, by using a blockchain-enabled CERS module we were able to further reduce the computational complexity to 32%, making it an ideal solution for UE devices having low computational complexity.