Deep Learning-Based Cluster Delay Estimation Using Prior Sparsity

Abstract

A deep learning (DL)-based cluster delay estimation method using prior sparsity is proposed. Firstly, the columns of the covariance matrix of channel frequency response in the time delay domain are formulated as undersampled noisy linear measurements of the delay spectrum. Then, a deep convolutional network (DCN) is used to recover the delay spectrum from the measurement vector. Compared with conventional model-driven methods, the proposed data-driven DCN can be used to estimate cluster delays with smaller delay intervals and also has an excellent generalization ability. Finally, numerical results show that the proposed DL-based delay estimation method has advantages in both precision and computational efficiency.