Page 351 - Kaleidoscope Academic Conference Proceedings 2024

Innovation and Digital Transformation for a Sustainable World

e) POLICY MAKING & STRATEGIC FUND Research Organization) having a spatial resolution of 56
ALLOCATION: DoT will possess the ability to identify meters.
regions across the country that exhibit the lowest signal h) In the study by D. Madariaga et al. [20] author showed
quality, enabling them to develop relevant policies and that crowdsourced data from mobile devices can predict
allocate necessary finances to address this coverage issue. mobile signal strength while adding meteorological data
enhancing regional accuracy.
2. OTHER RELATED WORKS i) In the study by D. F. S. Fernandes et al. [21], ANN
(artificial neural networks) was used to predict path loss
a) In a study done by FRACCARO, P. et al [15] regarding in mobile telecom networks.
combining Geospatial Big Data and AI to predict Model
Mobile Signal Strength, researchers gathered open- 3. GAPS OR INCONSISTENCIES IN EXISTING
access geospatial data about several factors including RESEARCH AND NEED FOR CURRENT STUDY
weather conditions, tree coverage, land use patterns, a) Minimization of Driving Tests is proposed as a 3GPP
altitude, and telecom infrastructure. This model was standard option for gathering measurement data from real
tested across the United Kingdom. The netBravo, users and assessing coverage [22]. MDT uses end-user
crowdsourcing platform that was used didn’t provide devices to crowdsource measurements. MDT
telecom operator information, so tower characteristics implementation in present networks is difficult because
were averaged for locations with multiple operators using of imprecise positioning, limited data availability, and
different transmitters. Current research employs a similar poor indoor reporting.
methodology by considering the average distance of the b) In another study by F. Lyu et al. [22] authors examined
three nearest BTS as the measure of distance between the data from 31 cities, while the current analysis covers 711
BTS and the telecom subscriber. districts out of the total 766 districts in India, representing
b) Alimpertis, E. [17] employed a machine learning coverage of 93% of the total districts of the country.
framework and Android app data to anticipate missing c) Dense urban areas with lots of buildings are prone to
values in mobile coverage maps. Location, time, BTS errors in positioning as noted by F. Lyu et al. [4] in the
cell IDs, and device hardware were used. Current case of MDT. Precise location tracking in the mobile
research produced an Android app. Similar to the current application developed in current research minimizes
study, mobile crowdsourcing data from New York and position errors. Signal loss makes indoor measurements
Los Angeles metropolitan areas was employed. It impractical. Communication is crucial in larger buildings
developed a sophisticated Random Forests-based like airports. Thus, indoor positioning requires more
machine learning platform. The following study accuracy than outdoors [6]. The current approach uses
anonymised the dataset by assigning a random device ID the telecom subscriber's precise location for extremely
to prevent monitoring of the original users. User data was high location accuracy.
uploaded to MongoDB and the current study used d) Further MDT seems to be more useful for coverage
Cassandra. Both are NoSQL distributed databases. testing of a single telecom operator whereas the solution
c) In the study by M. F. Ahmad Fauzi et al. [5], out of proposed in the current study considers all the telecom
various machine learning models, the Random Forest operators in India.
algorithm was selected and accolade as the predominant e) To protect user privacy, a study by FRACCARO, P. et al.
machine learning approach for developing a reliable [16] consolidated data into monthly releases at different
RSRP (Reference Signal Receive Power) prediction resolutions (100 m and 1 km), resulting in less accurate
model. In the current study as well, the Random Forest results than current research. In the study by D. F. S.
algorithm demonstrated the highest level of accuracy. Fernandes et al. [22] a total of 12,194 mobile signal
d) In the study by Wang, H et al. [18], the authors collected strength measures were used. However, the current study
network measurement data from end-user's smartphones used a much larger dataset of 1.4 million mobile signal
via crowd sensing and utilized machine learning strength measurements.
techniques to create BSA (base station almanac) database. f) M. F. Ahmad Fauzi et al. [5] simulated datasets of
e) In the study by F. Lyu et al. [4], the terrain was divided 12,011,833 samples, whereas in current research actual
into 10 categories, including roads, buildings, field-level measurement data was used.
manufactured items, tracks, vehicles, crops, trees, and g) According to the study by Alimpertis E [17], previous
rivers. It analyzed telecom coverage maps using machine research often focused on evaluating the raw signal
learning. The current study used 18 terrain categories. intensity and reducing mean square error, which may not
f) In the study by I. A. Saadi et al. [19] a drone was utilized align with telecom operators' priorities. Telecom
in the study to forecast ground-level mobile signals. The subscribers' main issue is signal quality, good vs. bad
artificial neural network predicted ground signal intensity coverage, hence current research also focuses on signal
from high-altitude data. It graded signal quality as quality rather than signal strength.
excellent, good, fair, and bad, which matches the current
methodology. 4. NEED FOR CURRENT DEVELOPMENT STUDY
g) In the study by FRACCARO, P. et al [15], digital The need for current development as mentioned in this paper
elevation models from The Shuttle Radar Topography arises due to the following reasons.
Mission of NASA were used. It provides elevation data
at a 30-meter resolution in latitude and longitude i. Presently, there is no dedicated machine-learning model
coordinates. In the current research, we used terrain data for signal strength interpolation in mobile networks in
from the Space Applications Centre, ISRO (Indian Space India. However, the signal strength value and tower
position as features in ML can be used to calculate signal
strength at new sites [23]. The novel findings of the current

– 307 –