A Supervised Learning Method for High‐Performance Channel State Information Estimation

ABSTRACT Channel state information (CSI) is pivotal for assuring high performances of wireless communication systems. In particular, multiple‐input multiple‐output transmission is only beneficial when CSI is known. A large number of subcarriers are desired in Orthogonal Frequency Division Multiplex (OFDM) systems to boost overall throughput, which makes channel estimation a more challenging task, especially to extract channel features in a more dynamic environment without causing a significant overhead transmission. Conventional least squares‐based methods are affected by the noise and interference that inherently exist in the acquired data for processing. We proposed the deep neural network (DNN)‐based method to estimate CSI, and one distinguishing characteristic is to adopt a Discrete Fourier Transform (DFT) operation‐based method to mitigate the impact of noise before carrying out the DNN procedure; hence, the accuracy of the learning outcome significantly improved. The effectiveness of the proposed scheme is verified with simulations under a variety of propagation scenarios. The proposed method has demonstrated a high performance for channel estimation. It has shown a particular advantage in more dynamic and noisy environments for wireless communications.