Long-term average proportional-fair (LTAPF) throughput optimization through power control is a popular resource allocation problem which is typically approximated by weighted sum-rate (WSR) maximization. WSR optimization is non-convex and strongly NP-hard in general. In this letter, we demonstrate that, in fact, the original sum-log-average-throughput power control problem can be recast as a convex program and thus solved to global optimality efficiently. We also generalize our result to show that the long-term average$\alpha$-fair utility maximization problem can be recast as convex for$\alpha \in (1,\infty)$. Numerical results demonstrate a substantial gain in LTAPF throughput compared to state-of-the art algorithms used to solve the max-WSR problem.

For a multi-cell, multi-user, cellular network downlink sum-rate maximization through power allocation is a nonconvex and NP-hard optimization problem. In this article, we present an effective approach to solving this problem through single- and multi-agent actor-critic deep reinforcement learning (DRL). Specifically, we use finite-horizon trust region optimization. Through extensive simulations, we show that we can simultaneously achieve higher spectral efficiency than state-of-the-art optimization algorithms like weighted minimum mean-squared error (WMMSE) and fractional programming (FP), while offering execution times more than two orders of magnitude faster than these approaches. Additionally, the proposed trust region methods demonstrate superior performance and convergence properties than the Advantage Actor-Critic (A2C) DRL algorithm. In contrast to prior approaches, the proposed decentralized DRL approaches allow for distributed optimization with limited CSI and controllable information exchange between BSs while offering competitive performance and reduced training times.

Optimizing the sum-log-utility for the downlink of multi-frequency band, multiuser, multiantenna networks requires joint solutions to the associated beamforming and user scheduling problems through the use of cloud radio access network (CRAN) architecture; optimizing such a network is, however, non-convex and NP-hard. In this paper, we present a novel iterative beamforming and scheduling strategy based on fractional programming and the Hungarian algorithm. The beamforming strategy allows us to iteratively maximize the chosen objective function in a fashion similar to block coordinate ascent. Furthermore, based on the crucial insight that, in the downlink, the interference pattern remains fixed for a given set of beamforming weights, we use the Hungarian algorithm as an efficient approach to optimally schedule users for the given set of beamforming weights. Specifically, this approach allows us to select the best subset of users (amongst the larger set of all available users). Our simulation results show that, in terms of average sum-log-utility, as well as sum-rate, the proposed scheme substantially outperforms both the state-of-the-art multicell weighted minimum mean-squared error (WMMSE) and greedy proportionally fair WMMSE schemes, as well as standard interior-point and sequential quadratic solvers. Importantly, our proposed scheme is also far more computationally efficient than the multicell WMMSE scheme.

A base-station (BS) equipped with multiple antennas can use its spatial dimensions in three different ways: 1) to serve multiple users, thereby achieving a multiplexing gain; 2) to provide spatial diversity in order to improve user rates; and 3) to null interference in neighboring cells. This paper answers the following question: What is the optimal balance between these three competing benefits? We answer this question in the context of the downlink of a cellular network, where multi-antenna BSs serve multiple single-antenna users using zero-forcing beamforming with equal power assignment, while nulling interference at a subset of out-of-cell users. Any remaining spatial dimensions provide transmit diversity for the scheduled users. Utilizing tools from stochastic geometry, we show that, surprisingly, to maximize the per-BS ergodic sum rate, with an optimal allocation of spatial resources, interference nulling does not provide a tangible benefit. The strategy of avoiding inter-cell interference nulling, reserving some fraction of spatial resources for multiplexing, and using the rest to provide diversity, is already close-to-optimal in terms of the sum-rate. However, interference nulling does bring significant benefit to cell-edge users, particularly when adopting a range-adaptive nulling strategy where the size of the cooperating BS cluster is increased for cell-edge users.