Team Precoding Towards Scalable Cell-free Massive MIMO Networks

Abstract

In a recent work, we studied a novel precoding design for cell-free networks called team minimum mean-square error (TMMSE) precoding, which rigorously generalizes centralized MMSE precoding to distributed operations based on transmitter-specific channel state information (CSI). Despite its flexibility in handling different cooperation regimes at the CSI sharing level, TMMSE precoding assumes networkwide sharing of the data bearing signals, and hence it is inherently not scalable. In this work, inspired by recent advances on scalable cell-free architectures based on user-centric network clustering techniques, we address this issue by proposing a novel version of the TMMSE precoding design covering partial message sharing. The obtained framework is then successfully applied to derive a variety of novel, optimal, and efficient precoding schemes for a user-centric cell-free network deployed using multiple radio stripes. Numerical simulations of a typical industrial internet-of-things scenario corroborate the gains of TMMSE precoding over competing schemes in terms of spectral efficiency under different power constraints. Although presented in the context of downlink precoding, the results of this paper may be applied also on the uplink.

Closed-form max-min power control for some cellular and cell-free massive MIMO networks

“…Popular examples are the distributed or centralized usercentric clustered models reviewed in [5], or the special case of cellular massive MIMO [10]. Since the focus of this article is on power control, we omit the details on such constraints, and refer to [24], [25] for details. In the following, we simply assume that {Ú k } k∈K are given, and that the expectations in the following matrices and vector exist:…”

Joint optimal beamforming and power control in cell-free massive MIMO

“…More specifically, scalable cell-free networks are expected to implement some type of user-centric rule [3], which allocates to each UE a cluster of serving APs. As shown in [11], this practical limitation can be modeled by setting H N l in (1) to the set {0 N (a.s.)} if AP l is not serving UE k. Considering both impairments, we remark that V k is a linear subspace of H N L .…”

“…The solution to (7), which can be interpreted as the best distributed approximation of a regularized zero-forcing beamformer, can be obtained via the recently developed team MMSE beamforming framework given by [10]. Although presented in the context of downlink beamforming with perfect message sharing, the method in [10] directly applies to the uplink case and to user-centric clustering [11]. This method states that the solution to (7) corresponds to the unique solution to a linear feasibility problem, which, depending on the CSI structure, can be solved explicitly or approximately via efficient numerical techniques.…”

“…Asymptotic convergence to an optimal solution is guaranteed by Proposition 2, Proposition 3, and Proposition 4. Clearly, the above algorithm relies on solving Problem (7) at each iteration, which, as already mentioned, can be done following [10], [11]. Note that Ú k is a function of the CSI, and not a deterministic vector.…”

“…To relate the present study with existing approaches, we recall that iterative algorithms for long-term power control under fixed beamforming design have been extensively studied in both the cellular and cell-free massive MIMO literature (see, e.g., [1], [4], [8], [9]). In contrast, the problem of optimal distributed beamforming design without power control has been solved only recently in [10], [11] using elements of team decision theory, a rigorous control theoretical framework for decentralized decision making under asymmetry of information. However, none of these previous studies have jointly optimized beamformers and transmit powers.…”