Coherent direct position estimation in distributed massive MIMO systems

Disertacija se bavi problemom direktne koherentne lokalizacije izvora xirokopojasnih radio signala pomou masivnih vixeantenskih sistemima u prostorno koherentnom scenariju LOS (Line-OfSight) komponenti. Ovaj scenario je tipiqan za male elije u milimetarskom opsegu u petoj generaciji 5G elijskih sistema. Lokalizacija se zasniva na obradi signala sa distribuiranih antenskih nizova koji mogu imati podnizove sa faziranim antenskim rexetkama. Ideja ove disertacije je da se infrastruktura buduih beiqnih sistema pete generacije (5G) iskoristi, pored komunikacije, i za lokalizaciju koja je predmet disertacije. Cilj je da se ostvari prezicnost procene pozicije za 2 do 3 reda veliqine bolju od talasne duine nosioca, xto klasiqne metode za dvokoraqnu i jednokoraqnu (direktnu) lokalizaciju ne omoguavaju. Da bi se to postiglo, koriste se koherentne metode { one koje pored pomaka anvelopa koriste i informacije sadrane u fazama nosioca LOS komponenti. Da bi se to moglo iskoristiti, potrebno je da u eliji postoji prostorna koherencija LOS komponenti. Zbog toga se istraivanje pre svega oslanja na male elije (sa LOS uslovima) i milimetarski (mmWave) opseg (koji ima povoljne uslove prostiranja), ali nije ograniqeno na njih dok god je prethodni uslov zadovoljen. Korixene su sledee metode istraivanja. Matematiqki je modelovan prostorno koherentni scenario i za njega su izvedene teorijske granice preciznosti lokalizacije. Zatim su predloene metode lokalizacije. Njihove performanse su analizirane simulacijama i eksperimentalno. Za eksperimente je korixen hardver koji je napravljen u sklopu istraivanja. Izvedene su Kramer-Raove granice preciznosti lokalizacije za model signala u prostorno koherentnom scenariju i pokazano je da su obrnuto srazmerne kvadratu frekvencije nosioca. Predloeno je vixe tipova metoda za lokalizaciju { nekoherentne, polukoherentne i koherentne; metode za poznatu sekvencu (kooperativan predajnik/korisnik) i za nepoznatu (nekooperativan predajnik); metode maksimalne verodostojnosti izvedene za jednokorisniqki potpuni LOS sluqaj i potprostorne metode. Predloene koherentne metode su statistiqki efikasne (njihova preciznost dostie izvedene teorijske granice) i ostvaruju preciznost za 2 do 3 reda veliqine bolju od talasne duine nosioca. Analiziran je i sluqaj vixestrukog prostiranja, a metode i tada ostvaruju preciznost za 2 reda veliqine bolju od talasne duine. Takoe funkcionixu u scenariju sa vixe (i to prostorno bliskih) predajnika pribline snage. Pritom, potprostorne metode funkcionixu i kad je interferirajui predajnik za 30 dB vee snage od ciljanog. Svi ovi rezultati su ostvareni za razumne vrednosti sistemskih parametara, kao xto su odnosi signal-xum i broj odbiraka. Pored toga, algoritmi funkcionixu i pri malim odnosima signal-xum, zahvaljujui tome xto direktno na osnovu sirovih signala procenjuju poziciju, za razliku od klasiqnih metoda koje gube informacije pravei meuprocene. Dodatna povoljnost direktne lokalizacije je xto se izbegava numeriqki zahtevan problem asocijacije. Algoritmi funkcionixu i za kratke opservacione intervale, zahvaljujui zdruenoj obradi svih spektralnih komponenti signala.The thesis deals with the problem of direct coherent localization of wideband radio signal sources, using massive MIMO (Multiple-Input-MultipleOutput) antenna systems in scenarios with spatially coherent LOS (Line-Of-Sight) signal components. These scenarios are typically found in small cells in the mmWave (millimeter wave) range in the fifth generation (5G) cellular systems. The localization is based on the processing of signals received by a distributed antenna array which may include phased antenna subarrays. In this thesis, the idea is to use the infrastructure of future 5G cellular systems for the localization in the thesis and for communication as well. The goal is to achieve position estimation accuracy by 2 to 3 orders of magnitude better than the carrier wavelength, which cannot be achieved by classical methods for two-step and one-step (direct) localization. In order to achieve that, coherent methods are used { those that, in addition to the information contained in envelope shifts, also use the information contained in carrier phase shifts of the LOS components. Spatial coherence of LOS components in the given cell is required to allow this information to be used. Thus, the research deals mostly with small cells (in LOS conditions) and the mmWave range (which has suitable propagation conditions), but is not limited to them as long as the previous condition is satisfied. The following research methods were used. A spatially coherent scenario was mathematically modelled and theoretical localization accuracy bounds were derived for it. Then, appropriate localization methods were proposed. Their performance was analyzed by simulations and experimentally. A hardware platform built as a part of the research was used in the experiments. Cram´er-Rao bounds on the localization accuracy have been derived for the signal model for the spatially coherent scenario and it has been shown that they are inversely proportional to the squared carrier frequency. Different types of localization methods have been proposed { non-coherent, semi-coherent and coherent; knownsequence methods (cooperative transmitter/user) and unknown-sequence methods (noncooperative transmitter); maximum-likelihood methods derived for the singleuser LOS-only scenario and subspace-based methods. The coherent methods are statistically efficient (their accuracy approaches the derived theoretical bounds) and an accuracy by 2 to 3 orders of magnitude better than the carrier wavelength isachieved. Multipath propagation is also analyzed and the methods achieve an accuracy by 2 orders of magnitude better than the carrier wavelength even in that case. They also perform well in a scenario with multiple (even spatially close) transmitters with comparable powers. Additionally, the subspace-based methods perform well even if the power of an interfering transmitter is 30 dB higher than that of the selected one. All of these results are achieved for reasonable values of the system parameters, such as the signal-to-noise ratio (SNR) and the number of used samples. Also, the algorithms perform well even at low SNRs, thanks to the fact that they estimate the position directly based on the raw signals, as opposed to classical methods which reduce the amount of useful information by making intermediate estimates. Another advantage of direct localization is the fact that the numerically complex data association problem is evaded. The algorithms also perform well even for short observation intervals, owing to the joint processing of all the spectral signal components