Deep multi-task learning-based simultaneous channel tap and coefficient estimation

Abstract

Wireless communication systems depend on accurate channel estimation to ensure efficient and reliable data transmission. The channel estimation process consists of two essential steps: channel tap and coefficient estimation. Physical layer features such as time arrival, and signal strengths are well used for the tap estimation. However, prior knowledge is required to use these methods. Recently, machine learning-based methods have been proposed. In particular, deep learning (DL)-based methods are promising because they can learn from raw data without much preprocessing, scale well with extensive and diverse datasets, and capture complex relationships. However, these methods overlook the relationship between the channel taps and coefficients. In this paper, we propose a DL-based multi-task learning method to estimate channel taps and coefficients simultaneously. Simulation results reveal that the performance of the proposed tap estimation method is superior to the traditional DL-based tap estimation. Furthermore, the proposed method removes the need to train two models to estimate channel taps and coefficients.