Deep Neural Network Enhanced Compressive Sensing and Kalman Filtering for 5G Channel Estimation

Abstract: Massive MIMO systems operating at mmWave frequencies (mmWave) constitute one of the applications of 5G communication, which provides high data rate and spectral efficiency. Nevertheless, the channel estimation and tracking is still a problem because of high channel variations and sparse propagation. In this paper, I am going to suggest a joint architecture that would combine Compressive Sensing (CS), Long Short-Term Memory (LSTM) and Kalman Filtering (KF) to effectively estimate channels. CS is employed in order to recover sparsely-spread channel information with a reduced number of pilot signals. The LSTM corrects the errors and removes noise in the estimated channel. KF monitors the change in channel across the time, and maintains the accuracy. The given approach is more precise in estimations, decreases the rate of errors, and has a higher spectral efficiency than the traditional approaches.

Keywords: 5G Communication, mmWave Massive MIMO, Compressive Sensing, Deep Neural Network, Kalman Filter, Channel Estimation