Aligning or not aligning in massive MIMO downlink?

Downlink (DL) pilot design for a multi-user massive multiple-input multiple-output (MIMO) system is a challenging problem. It is relatively simple to have a pilot design working well for a particular mobile station (MS). However, it is difficult to ensure that design will also provide good channel estimation performance at the other MSs. In our previous works, we have proposed a scalable framework for channel state information (CSI) acquisition in frequency-division duplexing (FDD) massive MIMO through aligning the DL channel paths judiciously. However, the total number of aligning patterns is limited and the channel estimation quality is poor when some channel paths always overlap with the channel paths with similar angles of departure (AoDs). In this paper, instead of aligning the DL channel paths, we propose a random pilot scheme for the DL CSI acquisition in FDD massive MIMO. In particular, the serving BS transmits pilots with random phase shifts and each MS compresses the observations with a compression matrix. Then the BS relies on the compressed feedback to recover the DL CSI. The optimal compression matrix for each MS is derived in closed form. Numerical simulations are then carried out and demonstrate the random pilot scheme can deliver better channel estimation performance than the scheme relying on aligning the DL channels paths.