Exploiting the Massive MIMO Channel Structural Properties for Minimization of Channel Estimation Error and Training Overhead

Abstract

Exploiting massive multiple-input-multiple-output (MIMO) gains come at the expense of obtaining accurate channel estimates at the base station. However, conventional channel estimation techniques do not scale well with increasing number of antennas and incur an unacceptably large training overhead in many applications. This calls for training designs and channel estimation techniques that efficiently exploit the physical properties of the massive MIMO channel as captured by sophisticated system/channel models. In this paper, we present designs that exploit the sparsity of the angle and delay domain representation of the massive MIMO channel as well as the low-rank property of the channel covariance, while also providing the connection between the sparse angle-delay representation and low-rank covariance property. Numerous multiuser scenarios are investigated including uplink, downlink, and singleand multi-cell communications, with the designs aiming at minimizing the channel estimation error or maximizing achievable rates with reduced training overhead. Theoretical analysis and numerical performance results indicate significant reduction of training overhead over conventional techniques while achieving similar performance. The presented methods demonstrate the importance of exploiting fundamental channel properties and reveal important insights on the interplay/tradeoff between training overhead and performance that can serve as guidelines for the design of future massive MIMO communication systems. INDEX TERMS Channel sparsity, correlated fading, channel estimation, training design, compressive sensing, pilot contamination, performance bounds, massive MIMO.

Channel Estimation: Unified View of Optimal Performance and Pilot Sequences

“…In practice, the number of estimated paths is small (rarely more than ten), so that such sequences are short when compared to the ones required when using the least squares model (of length N t ). Note that equations (24) and (25) allow to retrieve the variance and pilot sequences proposed in [11] and [12] (without proof of optimality). The bias problem.…”

“…In [23], sequences are found by numerical optimization, minimizing a weighted sum of the channel estimation errors of each user. In [24] and [25], heuristics are proposed which amount to send pilot sequences that span the union of the spaces generated by the leading eigenvectors of the channel correlation matrices of all users.…”

Decoupling Channel Estimation for FDD Massive MIMO Systems Utilizing Joint Sparsity

“…A nonorthogonal downlink pilot design was proposed in [8], [12], [14], [28] by exploiting the spatially common sparsity in angle domain and time correlation of massive MIMO channels. The channel sparsity in angle domain with partial support information was utilized in [29], [30] to acquire compressed CSI. However, all the works including the above-mentioned ones have rarely considered the joint sparsity in the delay-angle domain for massive MIMO systems.…”

Channel estimation: unified view of optimal performance and pilot sequences

“…In [19], sequences are found by numerical optimization, minimizing a weighted sum of the channel estimation errors of each user. In [20] and [21], heuristics are proposed which amount to send pilot sequences that span the union of the spaces generated by the leading eigenvectors of the channel correlation matrices of all users.…”