Kooperativna automatska klasifikacija signala po tipu modulacije korišćenjem mreže senzora

Abstract

U okviru disertacije razmatran je problem kooperativne automatske klasifikacije signala po tipu primenjene modulacije (Automatic Modulation Classification, AMC) koja se ostvaruje korišćenjem mreže prostorno distribuiranih senzora. Pri tome, izvršena je detaljna analiza postojećih rešenja sa stanovišta mogućnosti njihove primene u realnim uslovima rada koje definišu: nekooperativna priroda izvođenja AMC postupka, različiti prostorni raspored senzora, i realno propagaciono okruženje koje karakterše pojava fedinga usled višestruke propagacije. Sprovedenom numeričkom analizom utvrđeno je da se primenom ranije predloženih (postojećih) rešenja za kooperativnu AMC sa centralizovanom fuzijom odluka ili distribuiranim procesiranjem ne može ostvariti uspešna klasifikacija za najveći broj analiziranih realnih scenarija primene. Na osnovu uočenih nedostataka postojećih rešenja, koja ispoljavaju izuzetno veliku osetljivost na problem neusklađenosti referentnih vrednosti koje se koriste u procesu fuzije, predložen je alternativni koncept fuzije podataka. U ovom novom konceptu, umesto donošenja nezavisnih lokalnih odluka u senzorima mreže, i njihovog naknadnog kombinovanja, za potrebe fuzije koriste se vrednosti onih veličina na osnovu kojih se obavlja AMC, odnosno donose odluke u procesu klasifikacije. Na osnovu predloženog koncepta fuzije podataka definisan je određen broj novih metoda fuzije, pri čemu je kao osnov za razvoj ovih rešenja izabran AMC postupak na osnovu kumulanata četvrtog reda. Sprovođenjem numeričke analize, putem sveobuhvatnog procesa Monte-Carlo simulacija, pokazano je da se putem primene ovde predloženih metoda fuzije ostvaruju znatno bolje AMC performanse u poređenju sa postojećim rešenjima, i to kako za idealizovane scenarije primene tako i za ovde posebno razmatrane realne scenarije primene kooperativne AMC. Pri tome, najveći dobici ostvareni su u slučaju radio kanala u kojima se javlja frekvencijski-selektivan feding, dok su najlošiji rezultati dobijeni u slučaju radio kanala sa ravnim fedingom. Iz tog razloga, predložen je postupak združene korekcije kumulanata na nivou mreže, na osnovu koga su značajno poboljšanje performanse za realne scenarije primene kooperativne AMC. Osim toga, u cilju smanjivanja problema neusklađenosti referentnih veličina predložen je originalni postupak dvostepene hibridne fuzije, u kome se u prvom...

The main subject of the study portrayed in this thesis, was the proper formulation of the cooperative automatic modulation classification (AMC) solutions by using the spatially dispersed and networked sensors. A detailed analysis of the so far proposed solutions was conducted in order to evaluate their performances under the realistic application conditions defined with: the inherent uncooperativeness of the AMC process, different spatial distributions of sensors, and realistic propagation conditions characterized by the multipath fading channels. Through the comprehensive numerical analysis, it is shown that by applying the previously proposed centralized decision fusion and distributed processing under the conditions of here analyzed realistic application scenarios only a very low AMC performance can be achieved. Based on the detected deficiencies of the existing solutions for the cooperative AMC, which exhibit extremely high sensitivity to the problem of mismatched reference values that are used in decision fusion process, we have formulated a new alternative data fusion concept. In this new concept, instead of making the individual decisions at the each sensor and performing their combination, we apply the data fusion on the very features (cumulants) that are used in the classification process. In line with this new proposed data fusion concept, several new fusion methods are proposed by using the AMC algorithm based on the fourth-order cumulant. By further numerical analysis through the comprehensive Monte-Carlo trials, it was proven that here proposed cooperative AMC solutions greatly outperform the existing ones for the idealized and also for here considered realistic application scenarios. Thereby, the highest AMC performance gains were achieved for the frequency-selective fading channels, while the worst result were obtained for the flat fading channels. Therefore, additional joint cumulant estimation correction method was proposed, and it is proven that significant AMC performance enhancements are achieved by applying this solution in the realistic application scenarios. Also, in order to suppress the negative effects of using mismatched references, that is found in practice, a novel hybrid two-stage fusion (HyTSF) was proposed, with the data fusion applied in the first and the decision fusion in the second step of the proposed procedure. Through the...