Unit:
Κατεύθυνση Ηλεκτρονική και Ραδιοηλεκτρολογία (Ρ/Η)Library of the School of Science
Author:
Ntokos Alexandros
Supervisors info:
Άννα Τζανακάκη, Αναπληρώτρια Καθηγήτρια, Τμήμα Φυσικής, Ε.Κ.Π.Α.
Μάρκος Αναστασόπουλος, Αναπληρωτής Καθηγητής, Τμήμα Φυσικής, Ε.Κ.Π.Α.
Διονύσιος Ρεΐσης, Καθηγητής, Τμήμα Φυσικής, Ε.Κ.Π.Α.
Original Title:
Εφαρμογή Νευρωνικών Δικτύων στην αποκωδικοποίηση τηλεπικοινωνιακων σημάτων
Translated title:
Application of Neural Networks in the decoding of telecommunication signals
Summary:
Convolutional codes are a popular method of encoding in modern digital communication systems.
For the decoding of the received sequence, various statistical methods can be employed. An optimal
and widely used method for decoding convolutionally encoded signals is the maximum likelihood
method, specifically the Viterbi algorithm.
Artificial neural networks are contemporary models of machine learning. Their structure and basic
principles of operation are inspired by the neural networks of living organisms. They consist of
artificial neurons that are interconnected, and the way these neurons are connected is determined
by the chosen architecture. Their operation is divided into two phases: the training or learning phase
and the recall phase. During the learning process, the neural network learns the appropriate values
for the connection weights to approximate the desired functionality. In the recall phase, the neural
network is used by the designer to make predictions based on what it learned during the training
phase.
In the context of this thesis, the ability of neural networks to act as signal decoders encoded using
convolutional codes was explored. A neural network was designed and trained to successfully
decode messages sent by a digital transmitter using convolutional codes. To simulate real-world
conditions under which a communication system operates, Gaussian noise was introduced into the
transmitted sequence.
Various popular architectures of neural networks were tested and evaluated as decoders in terms of
their performances. The widely used Viterbi algorithm served as a comparison measure. Both the
bit error rate (BER) and decoding time were examined. Additionally, various ways of formatting the
received sequence for input to the neural network were analyzed and investigated.
Main subject category:
Science
Keywords:
Digital Communications, Convolutional Codes, Signal Decoding, Viterbi Algorithm, Artificial Intelligence, Machine Learning, Neural Networks
