Publication: Network planning for Nex Generation Networks (NGNS)
Loading...
Identifiers
Publication date
2021-10
Defense date
2021-11-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The fifth-generation (5G) of cellular communications is expected to be deployed in the
next years to support a wide range of services with different demands of peak data rates,
latency and quality of experience (QoE). To support higher data rates and latency requirements third-generation partnership project (3GPP) has introduced numerology and
bandwidth parts (BWPs) for provisioning of 5G services through new radio (NR). The
legacy 4G network has generated a lot of data which can be helpful to reveal insights
about subscriber's traffic and infrastructure of the mobile network operators (MNOs),
in order to identify 5G deployment area (5GDA) for planning new services. Given the
mission-critical nature of 5G services, QoE is a big challenge for MNOs to guarantee peak
data rates for a defined percentage of time. Apart from the access network, MNOs also
face challenges of capacity and connectivity in the backhaul network. These challenges
must be considered during the radio network planning (RNP) process to configure various
network parameters such that the defined targets are met and the cost of network
deployment is minimized. Therefore, the aim of this dissertation is to study and develop
network planning methods for 5G, considering the pragmatic aspects of new radio (NR),
capacity modelling, cost optimization, data acquisition and machine learning (ML).
Contrasting to legacy networks, where radio planning is focused on provisioning of higher
data rates, in 5G there are multiple use cases with different requirements of numerology,
BWPs and QoE, which should be an integral part of the RNP. Our proposal takes these
requirements into consideration by introducing two inter connected models, namely the
transmission and capacity model. The transmission model incorporates 5G numerology
and BWPs in service definition being offered by the MNO. It specifies each new service
in an allocated BWP according to the peak data rate requirements which corresponds
to the number of physical resource blocks (PRBs) and transmit power level of the next
generation Node-B (gNB). Each service is configured with different numerology factor
which correlates with a specific sub-carrier spacing to achieve desired latency in different
frequency ranges supported by NR. As a rule of thumb, to support ultra-low latency
services a larger sub-carrier spacing is configured and vice versa. Once the parameters of
the transmission models are defined, the maximum transmission bandwidth per service
and peak data rates per user are provided as inputs to the capacity model. The capacity
model uses this input along with probabilistic modeling of the radio resource control
(RRC) states of the 5G users to guarantee QoE by provisioning the peak data rates
for MNO's required percentage of time. The simulation results show that our proposed
model fulfils the QoE requirements for each service defined in the transmission model
with the trade-off in confining the total number of subscribers per gNB. The utilization of the available capacity per gNB is the assurance for the MNO to
a reduced cost per bit. The higher the network utilization the lower is the cost per
bit which increases the revenue stream of the MNO. The higher network utilization
can only be achieved if the planning of new deployments are assisted by real data to
determine the area of the highest traffic density. To achieve this, we propose a network
data acquisition procedure based on LTE identifiers to estimate traffic and to determine
the 5GDA where the subscribers concentration is maximum. We propose a confidence
benchmark sample (CBS) approach to take traffic information from reliable cells. We
apply different combinations of the long term evolution (LTE) identifiers on network data
to infer traffic and infrastructure information of the legacy network with a specified
country code. We use the proposed procedure to determine the traffic patterns by
visualizing the data with respect to geographical locations of the cells, though, the
demarcation of the 5GDA is performed manually. Consequently, we propose a machine
learning (ML) based framework to automate the demarcation of 5GDA by determining
the highest traffic cluster. In contrast to Elbow method of computing the number
of clusters, a cluster analysis approach is proposed to determine appropriate value of
clusters based on MNO's requirement. The results are promising as higher network
utilization is achieved with lower cost per bit values by identifying the highest traffic
region on the cluster level.
Given the highest traffic demands of the present-day networks, the cost and capacity
considerations in the network planning play a vital role in establishing the connectivity
of the backhaul segment with the core network. The new 5G deployments require higher
capacity at the distribution or gateway nodes for transmission between radio sites with
the core network. The problem of connecting the radio sites with the distribution nodes
has been investigated in the context of cost and capacity. An integer liner programing
(ILP) model has been formulated that minimizes the deployment cost by selecting the
required number of distribution nodes given that the capacity constraint is not violated.
At the same time, the model selects the distribution nodes based on centrality scores to
determine near-by nodes such that the geographical distance is minimized to save the
link cost. Moreover, we propose to co-site the distribution nodes with the already deployed
macro-sites to minimize the backhaul deployment cost. The proposed ILP model
provides the optimal topology solution at the cost of computational complexity. However,
we also proposed a backhaul connectivity algorithm with much lower computational
cost in the number of nodes compared to the optimal ILP model. The results show that
the proposed algorithm solves the backhaul connectivity and cost minimization problem
with a very small difference in the optimal topology solution obtained from ILP model
with a cost loss of 11%. However, the advantage of the lower complexity algorithm is that the near optimal solution can be obtained for real-time network applications where
immediate decision taking is essential.
Description
Keywords
5G, Network planning, New radio (NR), Capacity modeling, Cost optimization, Data acquisition, Machine learning