Page 125 - Proceedings of the 2018 ITU Kaleidoscope

MESSAGE COLLISION IDENTIFICATION APPROACH USING MACHINE LEARNING

1
1
1
Juan Pablo, Martín ; Bruno, Marengo ; Juan Pablo, Prina ; Martín Gabriel, Riolfo 1
1
GICom, Facultad Regional San Nicolás - Universidad Tecnológica Nacional, Argentina

ABSTRACT 1.1 Considerations on ADS-B Satellite Reception

Machine learning algorithms, in particular k-nearest ADS-B is a communication system that operates at
neighbors (kNN) and support vector machine (SVM), are 1 090 MHz, and based on data from an aircraft’s
employed to estimate the potential success in decodifying on-board systems, transmits its position and status. That
ADS-B messages in highly congested areas. The main frequency range is shared with other International Civil
aim of this study is to optimize automatic dependent Aviation Organization (ICAO) and non-ICAO standardised
surveillance-broadcast (ADS-B) reception on-board low aeronautical applications. Extending the coverage of existing
Earth orbit satellites. In this first approach, simulations are terrestrial receptors of ADS-B signal by a satellite network
performed to obtain the training and testing signals. First, needs different analysis from many perspectives, using as
ADS-B communication system is described; second, machine many methods as possible.
learning, kNN and SVM are introduced. Third, the developed While it is possible to use a LEO satellites or other types
simulator is presented and the kNN and SVM algorithms are of non-geostationary satellites to receive ADS-B signals,
described with its results. Finally, the performance of these there are some operational considerations to be taken into
two is compared. account. It is needed to de-garble the ADS-B signals
from other aeronautical systems signals operating in the
same frequency band (undesired signals), such as replies
Keywords - Automatic Dependent Surveillance-Broadcast, to secondary surveillance radar (SSR) interrogations, DME
Support Vector Machine, k-Nearest Neighbours and tactical air navigation system (TACAN). Even further,
depending on the region, there could be a great number of
1. INTRODUCTION planes emitting messages at the same time, so the objective
of this study is to provide an algorithm that is able to detect
Low Earth orbit (LEO) satellites have been widely used when a message from an aircraft can be successfully decoded.
for both voice and data communications since many years
ago[1]. After the disappearance of the flight MH370, the 1.2 Detection Problem
terrestrial ADS-B system was studied in order to extend its
The general detection problem can be written as:
coverage. To achieve that, the reception of those signals
on-board LEO satellites was first studied at the International
Telecommunication Union (ITU)[2]. (
ω 0 : x = b hypothesis: noise
In 2015, ITU included the agenda item 1.10 to the (1)
ω 1 : x = b + s hypothesis: identifiable signal
World Radiocommunication Conference 2019 (WRC-19)
with the subject “Studies on spectrum needs and regulatory The objective is then to be able to detect whether the received
provisions for the introduction and use of the Global signal is only noise, or if it contains information. Therefore
Aeronautical Distress and Safety System”. Furthermore, a detector d is built such that the following error probability
the Conference Preparatory Meeting 2019 invited the ITU is minimized.
Radiocommunication Sector and Working Party 5B “... to
conduct the relevant studies, taking into account information P e (d) = p (d(X) , Y) (2)
and requirements provided by ICAO for both the terrestrial
The classical hypothesis testing problem is a method of
and satellite components, including: a) quantification and
statistical inference, which, in fact, is the process of using
characterization of radiocommunication requirements related
data analysis to deduce properties of an underlying probability
to GADSS...”[3].
distribution[5]. A dataset obtained by sampling is compared
The main purpose of this study is, with the aid of machine
against a synthetic dataset from an idealized model following:
learning techniques, to estimate if a given ADS-B message
was collided or if it is possible to decode its information (
H 0 : X ∈ ω 0 X ∼ p(X | ω 0 )
on-board satellites. (3)
H 1 : X ∈ ω 1 X ∼ p(X | ω 1 )
The study is based on a simulator that recreates the reception
of position messages on-board satellites in high air traffic The distribution of the random process X has to be known
density conditions. to use the former method, but this is usually not possible. In

978-92-61-26921-0/CFP1868P-ART @ 2018 ITU – 109 – Kaleidoscope