Virtually Full-duplex Cell-Free Massive MIMO with Access Point Mode Assignment

Abstract

We consider a cell-free massive multiple-input multiple-output (MIMO) network utilizing a virtually full-duplex (vFD) mode, where access points (APs) with a downlink (DL) mode and those with an uplink (UL) mode simultaneously serve DL and UL users (UEs). In order to maximize the sum spectral efficiency (SE) of the DL and UL transmissions, we formulate a mixed-integer optimization problem to jointly design the AP mode assignment and power control. This problem is subject to minimum per-UE SE requirements, per-AP power control, and per-UL UE power constraints. By employing the successive convex approximation technique, we propose an algorithm to obtain a stationary solution of the formulated problem. Numerical results show that the proposed vFD approach can provide a sum SE that is 2.5 and 1.5 times larger than the traditional half-duplex and heuristic baseline schemes, respectively, in terms of 95%-likely sum SE.

Can Dynamic TDD Enabled Half-Duplex Cell-Free Massive MIMO Outperform Full-Duplex Cellular Massive MIMO?

“…This work significantly extends our previous work [3] by considering minimum mean squared error (MMSE)/zero-forcing (ZF) precoding and combining at the APs and signal-to-interference plus noise ratio (SINR)optimal combining at the central processing unit (CPU), pilot allocation via graph coloring, joint UL-DL power allocation for sum SE maximization, and comparing the performance against an FD system. Recently, the authors in [8] investigated the joint power UL-DL power allocation problem, assuming perfect CSI, in a DTDD CF system. The power allocation policy of [8] is derived for a maximal ratio combiner and precoder, with equal weighting-based combining of the APs' signals at the CPU in the UL, which is suboptimal for CF [5].…”

“…Recently, the authors in [8] investigated the joint power UL-DL power allocation problem, assuming perfect CSI, in a DTDD CF system. The power allocation policy of [8] is derived for a maximal ratio combiner and precoder, with equal weighting-based combining of the APs' signals at the CPU in the UL, which is suboptimal for CF [5].…”

“…Specifically, in contrast to a cellular system where only the serving AP requires CSI from a given UE, in CF, accurate CSI is required at all the APs in the vicinity of the UE. The works that account for pilot allocation in DTDD CF systems consider either complex iterative algorithm [3], random allocation of pilots [7], or assume orthogonal pilots across all UEs [8]. Recent works on FD CF systems have also considered orthogonal pilots [9], [13], [14] or have analyzed the effect of pilot contamination by abstracting it as an additive channel error term in the channel estimate [11], which does not explicitly account for the pilot length, design or allocation across the UEs.…”

“…We solve this nonconvex problem using fractional programming (FP) 1 that 1 FP convexifies the non-convex cost function such that the optimal solution of the surrogate cost function and the original cost function is the same [15]. This is in contrast with the approach adopted in [8], [10], where the convex cost function is typically a lower bound of the original cost function and the algorithms optimize the lower bound. employs a series of equivalent convex reformulations.…”

“…The resulting algorithms for each sub-problem are shown to converge to local optima (see Proposition 3 and Proposition 4). Finally, our proposed FP-based algorithms are precoder/combiner scheme agnostic, unlike [8], and require fewer auxiliary variables, which makes our solutions scalable for large distributed systems. ii.…”