Joint User Grouping, Sparse Beamforming, and Subcarrier Allocation for D2D Underlaid Cache-Enabled C-RANs With Rate Splitting

Abstract

We propose a Rate Splitting (RS) transmit scheme for Device-to-Device (D2D) underlaid Cache-enabled Cloud Radio Access Networks (C <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^2$</tex-math></inline-formula> -RANs). To this end, we jointly design Cellular User (CU) grouping, dynamic Remote Radio Head (RRH) clustering, beamforming, RS ratio, and subcarrier allocation to maximize the sum-rate and ensure transmit power and fronthaul cost constraints. However, the formulated problem is discrete, non-smooth, and non-convex. We thus decouple it into three subproblems, namely - (1) CU grouping, (2) D2D subcarrier allocation, and (3) joint sparse beamforming, RS ratio, and power control. We then develop a low-complexity greedy searching algorithm and a fairness-ensuring accelerated bisection searching algorithm via graph theory for the first subproblem. For the second subproblem, we use a many-to-one matching game with peer effect. We adopt the Gale-Sharply algorithm and swap operations to reach a stable matching state and propose a Two-sided Stable Subcarrier Allocation ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\text{TS}^2$</tex-math></inline-formula> A) algorithm. For the third subproblem, we develop a Quadratic Transform-based Two-Tier Alternating ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\text{QT}^3$</tex-math></inline-formula> A) algorithm. It constructs a series of accurate surrogate functions for the objective function and constraints via a Quadratic Transform (QT) technique. We then recast it as a two-tier alternating problem. We tackle the outer and inner tiers with closed-form expressions and the convex framework, respectively. By integrating these algorithms, the overall algorithm can be proved to converge to a stationary point. Simulation results show that it achieves considerable performance gains over several benchmark schemes.