User Association in Cell-less 5G Networks Exploiting Particle Swarm Optimisation Tareq M. Shami, David Grace, Alister Burr Communication Technologies Research Group, Department of Electronic Engineering University of York, York, YO10 5DD, United Kingdom tareq.al-shami@york.ac.uk, david.grace@york.ac.uk, alister.burr@york.ac.uk Abstract—In heterogeneous networks (HetNets), users can by default associate with the macro base stations (BSs) while the small cell BSs are underloaded. Biasing user association is a simple and realistic approach to balance the load in HetNets, as well as creating a cell-less architecture where a user does not connect to the closest base station. Most of the existing research focuses on the static biasing scheme which is not the optimal strategy to improve the system performance. In this paper, the biasing factors are generated dynamically by the algorithm of particle swarm optimisation (PSO) with the objective of balancing the load and maximising the cell spectral efficiency (CSE). This work studies two different interference cases: the first case is when each tier uses different radio resources (typical when multiple radio access technologies are used) and a user receives interference only from same-tier base stations, whereas the second interference case is when all tiers use the same radio resources and a user receives interference from the same-tier and other tier BSs. The simulation results show that the dynamic biasing using PSO outperforms the static biasing in terms of balancing the load and maximising the CSE. Keywords—cell-less architcture; User association; hetergenoues networks; particle swarm optimisation; I. INTRODUCTION The deployment of heterogeneous networks is a promising approach towards the success of the 5G era. HetNets can improve spectral efficiency, create hot-spots, eliminate coverage holes, and reduce cost. User association in heterogeneous networks can improve the system performance by balancing the load and optimising the spectral efficiency as well as energy efficiency. Dense heterogeneous networks in 5G introduce several challenges in designing a user association scheme. As a result, user association has attracted many researchers recently. The traditional user association scheme that is based on the maximum received signal strength is not a suitable approach in heterogeneous networks since many UEs will be attracted to the macrocell due to its high transmission power while small cells will be lightly loaded [1]. To address the aforementioned problem, 3GPP Release 10 introduced the concept of Cell Range Expansion (CRE). In CRE, a bias is added to the power received by UEs from small cells which attracts more UEs to associate with small cells. CRE is a practical approach that has the ability to achieve load balancing in heterogeneous networks since it only requires a simple uncoordinated decision, i.e., adding a bias to the received power from a small cell [2, 3]. Based on the concept of CRE, the authors in [4] showed that the system capacity is improved by offloading users from macro-BS to small cells BSs. The drawback of CRE is that UEs who are encouraged to connect to small cells due to the added bias suffer from strong interference caused by the nearby macro-cell [5]. To balance between the network throughput and the load balance, the selected bias must be carefully chosen [6]. User association based on the biasing concept forms a cell- less architecture where a user does not necessarily associate with the closest BS or the BS that provides the strongest SINR. In other words, users who are out of a cell boundary can still be associated with that cell. Biasing user association can also be implemented to balance the load in coordinated multipoint transmission (CoMP) networks, which is one form of the cell- less architecture, where a user can associate with more than one BS. In CoMP, most of the users still associate with the macro BS causing the macro BS to be heavily loaded and the small cells under loaded. An easy and practical approach to solve the load imbalance in CoMP is to implement the biasing user association. A recent work has been conducted on a cell-less architecture that decouples the uplink and downlink and it showed that biasing is needed to balance the load and achieve optimal rate coverage probability [7]. It is clear that biasing user association plays an essential role in balancing the load in cell-less architectures with a simple and realistic implementation. The purpose of this paper is to show how PSO can be applied to dynamically adjust the bias to each small cell BS in order to balance the load and maximise the SE. Several research studies have been carried out on user association in multi-tier networks focusing on various performance metrics such as spectrum efficiency and energy efficiency. In [8], Q-learning was applied in order to find the bias value of each user instead of using a common bias value among all users. In the proposed technique, each user independently learns from historical experience the optimal bias value that can optimise the number of outage users. The proposed scheme outperformed the optimum common bias value in terms of network throughput as well as the number of outage users. In [9], biased user association in multi-tier heterogeneous networks for both uplink and downlink was investigated. The optimum biasing factors were derived analytically with the assistance of stochastic geometry. The obtained results showed that the optimal downlink and uplink biasing factors are not identical. Based on quantum-behaved particle swarm optimisation (QPSO), a dynamic biasing user association approach in heterogeneous networks was presented in [10] to address the problem of load balancing. The role of QPSO is to periodically find the best biasing factors that can achieve load balancing and maximise the throughput. Based on CORE Metadata, citation and similar papers at core.ac.uk Provided by ZENODO