Abstract
This paper studies training-based sparse channel estimation in massive MIMO-OFDM systems. In contrast to prior works, the focus here is on the setup in which (training) pilot tones are spread across multiple OFDM symbols. Within this setup, two training models-termed distinct block diagonal (DBD) model and repetitive block diagonal (RBD) model-are investigated. The restricted isometry property, which leads to sparse recovery guarantees, is proven for the DBD model. Further, it is established that the RBD model, through exploitation of its tensor structure, leads to computationally simpler sparse recovery algorithms. Finally, numerical experiments are provided that compare and contrast the channel estimation performance under the two models as a function of the number of pilot tones per OFDM symbol and the total number of OFDM symbols.