LTE-LAA cell selection through operator data learning and numerosity reduction

Abstract

Long Term Evolution-Licensed Assisted Access (LTE-LAA) architecture is markedly different from traditional LTE HetNets. LTE-LAA deployments also have to contend with interference from coexisting Wi-Fi transmissions in the unlicensed spectrum. Hence, there is a need for innovative cell selection solutions that cater specifically to LTE-LAA. Further, the impact of cell selection on the performance of the existing LTE-LAA deployments should also be investigated through operator data analysis. This work addresses these challenges. We gather a large sample of LTE-LAA deployment data for three cellular operators, i.e., AT&T, T-Mobile, and Verizon, which is analyzed through several supervised machine learning algorithms. We study the effect of cell selection on LTE-LAA capacity and network feature relationships. Insightful inferences are drawn on the contrasting characteristics of the Licensed and Unlicensed components of an LTE-LAA system. Further, a cell-quality metric is derived from operator data and is shown to have a strong correlation with Unlicensed coexistence network performance. To validate the proposed ideas, two state-of-the-art cell association and resource allocation solutions are implemented. Validation results show that data-driven cell-selection can reduce Unlicensed association time by as much as 34.89%, and enhance Licensed network capacity by up to 90.41%. Finally, with the vision to reduce the computational overhead of data-driven cell selection in LAA and 5G New Radio Unlicensed networks, the performance of two popular numerosity reduction techniques is evaluated.

Introduction

Global mobile data demand is consistently rising and is expected to reach 77 exabytes/month by 2022 [1]. This is due to the proliferation of new services and applications such as on-the-go video streaming and Augmented/Virtual Reality applications. To cater to this surging demand, cellular service providers and standardization bodies have taken several steps. They have rapidly standardized, adopted, and deployed new technological paradigms such as Long Term Evolution/Long Term Evolution-Advanced (LTE/LTE-A) heterogeneous networks (HetNets) and LTE-WiFi coexistence networks. Spectral utilization has also been maximized by opening up the 5 GHz and 6 GHz bands for Unlicensed operation through LTE Licensed Assisted Access (LTE-LAA) and 5G New Radio (NR) in Unlicensed (NR-U) [2].

The exponential growth in data consumption has also pushed cellular operators to provide broad coverage and seamless connectivity through a dense deployment of Base Stations/eNodeBs (BS/eNB). However, increasing the BS/eNB density to provide higher data rates is expensive and time-consuming [3]. As a result, operators have chosen to deploy a large number of low-power small cells such as pico cells and femto cells. This trend is likely to continue in the next generation of HetNets and LTE-LAA/NR-U coexistence networks, giving rise to dense ($d\le$ 10 m) and ultra-dense ($d\le$ 5 m) network deployments (DNs/UDNs), where $d$ is the inter-cell distance [4].

Despite the benefits, increasing network density and the rapid growth in small cell installation have presented new architectural and optimization challenges in LTE HetNets, especially in LTE-LAA/NR-U deployments. Further, as the number and density of small cells rise dramatically, it is necessary to devise a reliable and efficient means to identify these cells for self-configuration. The need for seamless network configuration/re-configuration and uninterrupted service to the user equipments (UEs) has spawned the idea of Self-Organizing Networks (SON) [5].

The problem of physical layer identification of a cell in LTE-A/LTE-LAA/NR-U is solved through its physical cell ID (PCI). The PCI serves as the cell identity during cell selection when the UE is switched on, or during a handover [6]. A PCI value is generated using two frequency synchronization signals viz., the Primary Synchronization Signal (PSS) and the Secondary Synchronization Signal (SSS). A UE needs the two signals to achieve time domain radio frame synchronization, subframe synchronization, slot synchronization, and symbol synchronization. They are also necessary for the determination of the center of the channel bandwidth in the frequency domain. PSS can take the values 0, 1, and 2, while SSS lies in the range, 0 to 167. Together, they generate a PCI value that lies between 0 and 503, through the expression [(3$\times$SSS) $+$ (PSS)]. For example, in Fig. 1, Cells 1, 2, &3 are allocated non-conflicting PCI values “A”, “B”, & “C”, respectively, in the range of 0–503. However, given that there are only 504 unique values, PCIs are reused through strategic network planning to efficiently identify BSs and eNBs.

As a result, PCI search, selection, and attachment is a critical procedure in LTE HetNets and LTE-LAA/NR-U coexistence networks and is vital for delivering the guaranteed QoS to the end-user [5]. Improper or unplanned PCI allocation, will lead to increased interruption of the Reference Signal (RS) which will in turn adversely impact signal coverage.

Solutions to cell selection problems in LTE/LTE-A networks have been proposed by both academia [5], [7] and industry [8], [9]. However, there are several open challenges to cell selection in LTE-LAA/NR-U coexistence which are discussed ahead. The proposed work addresses some of these problems by analyzing LTE-LAA deployment data from three major operators and demonstrates the impact that PCI has on LAA network capacity and feature point relationships. The ideas and solutions presented in this work are organized as follows. Section 2 presents an outline of the research problems and highlights the major research contributions. It is followed by a literature review of relevant studies in Section 3. Section 4 discusses the data gathering exercise, the research methodology, and the learning algorithms used in this work. In Section 5, the network-level analysis and inferences are presented, followed by data-driven cell selection and network optimization in Section 6. Section 7 discusses data reduction and its impact on network feature relationships. Conclusions and future direction are presented in Section 8.