Page 151 - Kỷ yếu hội thảo quốc tế: Ứng dụng công nghệ mới trong công trình xanh - lần thứ 9 (ATiGB 2024)

142 TRƯỜNG ĐẠI HỌC SƯ PHẠM KỸ THUẬT - ĐẠI HỌC ĐÀ NẴNG

variability in both spatial and temporal distribution. the transport network that connects to the core
MOs are forced to manage traffic flows tied to network and the RAN.
various applications that have different requirements.
Therefore, they want to operate, monitor and optimize
their MBH with utmost efficiency, ensuring their
subscribers' satisfaction even under the most
demanding application prerequisites. Achieving this
demands a comprehensive understanding of their
MBH performance patterns to classify BSs that have
bad backhaul transmission quality for tuning
promptly. Unfortunately, details about discovering
unseen patterns of mobile backhaul performance have
been neglected.
Scientific literature on MBH predominantly
discusses challenges and offers new simulation-based Fig. 1. The overall mobile network architecture [8]
solutions or features to expand capacity and improve
MBH quality for compatibility with the development MBH performance is widely monitored, measured
trend of mobile radio technology [2], [3]. Another and assessed by TWAMP protocol. Fig. 2 illustrates
direction focuses on the development of protocol- the schematic diagram of TWAMP system for 5G
integrated quality monitoring systems to monitor, mobile transport network.
operate and manage the MBH [4], [5]. There are TWAMP servers record and store data with key
relatively few studies on applying ML to analyze performance indicators that consist of a set of
actual MBH performance evaluation data for MNOs common performance counters such as PL, TWAMP
to gain deep insights about their MBH characteristics. success rate. This paper investigates a factual dataset
of the operating MNO collected from these counters.
Therefore, the goal of this paper is to unveil
performance patterns of the MBH through
unsupervised learning. Investigated dataset has been
collected from TWAMP system [6] of an operating
MNOs to come up with a practical application study
which can be applied to the mobile communication
industry. The outcomes of the research highlight
classification of groups of BSs with same MBH
performance in respect of PL and TWAMP setup Fig. 2. General architecture of 5G
success rate by Elbow method combined with the k- back-haul’s TWAMP system [6]
means clustering algorithm. 3. THE PROPOSED UNSUPERVISED
The rest of the paper is organized as follows: LEARNING METHOD
section 2 gives an overview of the dataset of 3.1. Elbow method
performance of the MBH, section 3 briefly describes The Elbow method is used to determine the
the experiments of clustering collected data based on optimal number of clusters in the k-means algorithm
our proposed approach consisting of the Elbow used in this paper. This method concentrates on the
method combined with the k-means algorithm. percentage of variants as a function applied to the
Section 4 provides numerical results. Finally, number of clusters [9], [10]. The core idea aims to
conclusion is drawn in section 5. find the suitable number of clusters in the investigated
data set.
2. DATASET OF MOBILE BACKHAUL
PERFORMANCE Let's denote S as the sum square error ( SSE ).
Approximately, a new generation of mobile It is calculated as:
technologies is introduced in a 10-year-cycle, which k 2
− 
brings improvements in experience, performance, S =  x C m 2 (1)
efficiency and capability [7]. The mobile industries = m 1  x C m
have commercially beheld the mobile communication SSE is the sum of the average Euclidean Distance
evolution from 1G to 5G so far. Regardless of the of each point against its assigned centers. x and k are
mobile technology, the overall architecture has the the element of cluster C and the number of clusters,
common ground of radio access network (RAN), core m
network and MBH as shown in Fig. 1. MBH refers to respectively. When the value of k increases, the
sample partition will be more clarified and the
clustering degree of each cluster also greatly increase.
ISBN: 978-604-80-9779-0