CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Aplicación de Machine Learning a la seguridad en CRNs

Director/a: LEÓN ABARCA, OLGA

Departament: ENTEL

Títol: Aplicación de Machine Learning a la seguridad en CRNs

Data inici oferta: 30-01-2020 Data finalització oferta: 30-09-2020

Estudis d'assignació del projecte:

GR ENG SIS TELECOMUN

Tipus: Individual

Lloc de realització: EETAC

Paraules clau:
CRNs, Machine Learning

Descripció del contingut i pla d'activitats:
Las redes cognitivas (Cognitive Radio Networks o CRNs) son redes inalámbricas que usan el espectro de forma oportunista, es decir, operan en bandas frecuenciales asignadas a servicios específicos, que constituyen los usuarios primarios de dicho espectro, siempre y cuando estos últimos no hagan uso de dichas bandas. Con el objetivo de determinar qué bandas están libres y cuando, se suelen usar mecanismos cooperativos de detección de espectro (spectrum sensing), en los que los dispositivos que forman parte de la CRN escuchan el medio y toman una decisión conjunta con respecto a la ocupación de cada canal. Los mecanismos cooperativos de detección de espectro son vulnerables a ataques de emulación de primario o a nodos maliciosos que proporcionan información, y que pretenden evitar que la CRN acceda a determinadas bandas frecuenciales que están libres. En este proyecto se aplicarán y analizarán, mediante Matlab o Python, algoritmos de Machine Learning para detectar correctamente la presencia de usuarios primarios legítimos y de posibles nodos maliciosos.

Overview (resum en anglès):
Cognitive radio networks (CRNs) are made up of smart devices that use frequency bands assigned to licensed services, as long as their legitimate users or primary users are not making use of them. Primary user detection mechanisms allow identifying these free frequency band or spectrum holes, thus allowing the cognitive network to operate on them while preventing from interfering primary communications. Cognitive network users, also known as secondary users, cooperate among them by sharing their individual detection results, which are sent to a fusion center. The fusion center is responsible for combining all results and take a decision about the occupany of a given frequency band. Despite cooperation improves the detection rate, it also opens doors to malicious users. One of the most known attack to cognitive networks is the false feedback attack or byzantine attack, in which a user or a set of users send false results to the fusion center. The goal of this attack can be preventing the network from using a free band, or the other way round, forcing the cognitive network to interfere primary communications. Another common attack is the primary emulation attack, in which an attacker impersonates a primary user, thus preventing the cognitive network from accessing a specific band. In order to detect these users that disrupt network performance, machine learning algorithms have been used. Matlab has been used to simulate a cognitive network with a set of secondary users that report power measurements to a fusion center. A dataset has been created with these measures and using Python different machine learning algorithms have been executed and their behaviour between the different scenarios has been observed. The algorithms used are K-Nearest Neighbor (KNN), Support Vector Machine (SVM) and Random Forest (RF). Different parameters of each algorithm have been modified to evaluate their performance against the input data set and find the optimal values. Also, a comparison of the different algorithms is provided.