[Paper Review] Full-Duplex Cloud Radio Access Network: Stochastic Design and Analysis

Arman Shojaeifard, Kai‐Kit Wong+3 more|arXiv (Cornell University)|Nov 6, 2017

Full-Duplex Wireless Communications47 references3 citations

TL;DR

This paper proposes a stochastic framework for analyzing spectral efficiency (SE) in full-duplex cloud radio access networks (FD-C-RANs) with finite clustering, non-isotropic fading, and fronthaul capacity constraints. Using Poisson point process (PPP)-based modeling, it derives upper bounds on downlink and uplink SE by characterizing interference statistics via moment generating functions and probability generating functionals (PGFL), showing significant SE gains over half-duplex operation, especially with high-capacity fronthaul and advanced interference cancellation.

ABSTRACT

Full-duplex (FD) has emerged as a disruptive communications paradigm for enhancing the achievable spectral efficiency (SE), thanks to the recent major breakthroughs in self-interference (SI) mitigation. The FD versus half-duplex (HD) SE gain, in cellular networks, is however largely limited by the mutual-interference (MI) between the downlink (DL) and the uplink (UL). A potential remedy for tackling the MI bottleneck is through cooperative communications. This paper provides a stochastic design and analysis of FD enabled cloud radio access network (C-RAN) under the Poisson point process (PPP)-based abstraction model of multi-antenna radio units (RUs) and user equipments (UEs). We consider different disjoint and user-centric approaches towards the formation of finite clusters in the C-RAN. Contrary to most existing studies, we explicitly take into consideration non-isotropic fading channel conditions and finite-capacity fronthaul links. Accordingly, upper-bound expressions for the C-RAN DL and UL SEs, involving the statistics of all intended and interfering signals, are derived. The performance of the FD C-RAN is investigated through the proposed theoretical framework and Monte-Carlo (MC) simulations. The results indicate that significant FD versus HD C-RAN SE gains can be achieved, particularly in the presence of sufficient-capacity fronthaul links and advanced interference cancellation capabilities.

Motivation & Objective

• To address the limited spectral efficiency gain of full-duplex (FD) in cellular networks due to mutual-interference (MI) between uplink (UL) and downlink (DL) transmissions.
• To investigate the performance of FD-enabled cloud radio access networks (C-RANs) under realistic system constraints, including finite-capacity fronthaul links and non-isotropic fading.
• To develop a stochastic design and analysis framework for FD-C-RANs using Poisson point process (PPP)-based modeling of radio units (RUs) and user equipments (UEs).
• To compare disjoint and user-centric clustering strategies in finite C-RAN clusters and evaluate their impact on SE under interference and fronthaul constraints.
• To derive tractable upper-bound expressions for DL and UL spectral efficiency by characterizing the statistics of intended and interfering signals.

Proposed method

• Models the deployment of multi-antenna RUs and UEs using a Poisson point process (PPP) to enable tractable stochastic geometry analysis.
• Derives moment generating functions (MGFs) of interference power using the MGF of Gamma-distributed channel gains and applies the probability generating functional (PGFL) to model spatial interference from PPP-distributed interferers.
• Applies Approximation 3 to simplify the interference power expressions by modeling the sum of squared channel gains as a Gamma-distributed random variable.
• Uses integral identities $\mathscr{F}_1(.)$ and $\mathscr{F}_2(.)$ to evaluate the PGFL-based interference statistics in closed-form for both disjoint and user-centric clustering.
• Derives upper-bound expressions for DL and UL spectral efficiency by incorporating the statistics of intended signals, inter-cell interference (ICI), and cross-mode interference (CMI) under both clustering strategies.
• Validates analytical results via Monte Carlo simulations to confirm the accuracy of the derived bounds and evaluate system performance under varying fronthaul and interference cancellation conditions.

Experimental results

Research questions

• RQ1 What is the achievable spectral efficiency gain of full-duplex (FD) over half-duplex (HD) in a C-RAN with finite clustering and capacity-limited fronthaul links?
• RQ2 How do non-isotropic fading and finite-capacity fronthaul links impact the performance of FD-C-RANs compared to idealized models?
• RQ3 What is the relative performance gain of user-centric clustering versus disjoint clustering in FD-C-RANs under interference and fronthaul constraints?
• RQ4 How do advanced interference cancellation capabilities and high-capacity fronthaul links jointly affect the SE gain in FD-C-RANs?
• RQ5 Can tractable upper-bound expressions for DL and UL SE be derived in FD-C-RANs using stochastic geometry under realistic system assumptions?

Key findings

• Significant full-duplex versus half-duplex spectral efficiency (SE) gains are achievable in C-RANs, particularly when fronthaul links have sufficient capacity and interference cancellation is effective.
• The proposed stochastic framework provides tight upper bounds on DL and UL SE by accurately modeling the statistics of intended signals, inter-cell interference (ICI), and cross-mode interference (CMI) using MGFs and PGFLs.
• User-centric clustering leads to higher SE compared to disjoint clustering due to better interference management and more favorable channel statistics in the interference-limited regime.
• Non-isotropic fading channels significantly affect interference statistics, and the proposed model accounts for this by incorporating directional path loss and spatial correlation in the MGF derivation.
• Monte Carlo simulations confirm the analytical results, showing that SE gains increase with higher fronthaul capacity and improved interference cancellation at the user equipment.
• The derived expressions for interference statistics are valid under both disjoint and user-centric clustering, with the latter showing superior performance due to localized cooperation and reduced inter-cluster interference.