496
Archived Theses
0
DOIs Assigned
0%
DOI Rate
Discipline
Gaziantep ili yapay zeka tabanlı akıllı ulaşım sistemleri ile adaftif sinyalizasyon kontrolü ve simülasyonu
Hızla gelişen bir teknoloji döneminde olduğumuz son yıllarda dünya genelinde birçok ülkenin sorunu olan trafik sorunu Türkiye'de de birçok şehirde kronik bir sorun haline gelmiştir. Gaziantep, günümüz verilerine göre 2 milyon 130 bin 432 nüfus ve 504511 araç sayısı ile ülkemizin hem araç hem de nüfus bakımından en yoğun 9. kentidir. Gaziantep'te otomobil başına düşen kişi sayısı 8,2'dir. Günden güne artan trafik hacmi özellikle kentin merkezi kavşaklarından olan İller Bankası-1, İller Bankası-2 ve Shell kavşaklarında etkili olmuştur. Bu çalışma kapsamında çalışma alanı içinde belirlenen kavşak ve koridorlarda sorunlarla ilgili doğru tespitlerin yapılması ve sonrasında oluşturulacak öneri projeler için gerekli analitik altlıkların oluşturulması amaçlanmıştır. Mevcut durum ve sorun tespitleri için sahada yapılan gözlemler ve zirve saat kavşak trafik sayımları kullanılarak PTV-Vissim ve Aimsun yazılımlarının yardımıyla mevcut durumun trafik mikro-simülasyon modelleri oluşturulmuş ve analizler gerçekleştirilmiştir.
Güneş enerjisi santrali verimini koruma ve daha iyi enerji üretmesi için destekleyeci yöntemler
Güneş enerjisi santralleri yenilenebilir enerji santralleri içerisinde önem açısından büyük bir paya sahiptir. Ayrıca Son zamanlarda üretim açısından yönelinen güneş enerjisi santralleri önem payı artmaktadır. Diğer Yenilenebilir enerji santrallerine oranla kurulumu kolay tamamlanması hızlı bakım maliyetleri de düşüktür. Güneş enerjisi santralleri kurulumu sonra bakım önem arz etmekte zamanında yapılmayan bakımlar verim düşüşüne neden olmaktadır. Gaziantep İslam Bilim ve teknoloji Üniversitesi çatısına kurulması planlanan güneş enerjisi santralinin pvsyst ortamında 3 farklı durum için yapılan analizleri ele alınmıştır. Burada havanın kirlilik oranına bağlı güneş panellerinin üzerinde oluşan kirlilik katsayısı ele alınmış. Buna bağlı olarak da verimin düşüşü ne kadar olabileceği tespit edilmiştir. Burada ki kıyaslama güneş panellerinin yüzeyinin en düşük kirlilik oranına göre belirlenmiştir. Bir diğer kayıp unsur ise hava sıcaklığı dolayısı ile ortam sıcaklığı ile güneş panellerine etkisi ele alınmış buna bağlı olarak verimin düşüşü ne kadar olabileceği tespit edilmiştir. Burada ki kıyaslama ise güneş panellerinin veriminin en yüksek optimum hava sıcaklıkları esnasında tasarıma göre bulunduğu ortamdaki maksimum enerji üretimidir. 1. durumda yapılan çalışmada panel üzeri kirlilik oranı çok düşük, hava sıcaklığının güneş panelinin maksimum verim sağlayabileceği sıcaklık aralığında aynı zamanda paneli soğutmaya sağlayacak seviyede rüzgârlı hava şartlarıyla karşımıza çıkan yüksek üretim sağlayabilme ihtimali ele alınmıştır. 2. durumda yapılan çalışmada panel üzeri kirlilik oranı belirli bir seviyede olması panel üzeri oluşan kirliliğin bir süre temizlenmemesi durumu, sıcaklığın ise ortam sıcaklığına bağlı olarak oluşan sıcaklık seviyesinde paneldeki verim kaybı ele alınarak yapılan bir simülasyon olarak değerlendirilmiştir. 3. durumda yapılan çalışmada panel üzeri oluşan kirliliğin uzun bir süre temizlenmemesi sonucu en son aşamada oluşacak verim kaybı ele alınmış, sıcaklığın ise ortam sıcaklığına bağlı verim kaybı haricinde kirlilik ile panele oluşturulan ektra sıcaklıkla da ilave verim kaybı oluştuğu ele alınmıştır. Sıcaklık ve kirlilik etmenlerini minimum seviyeye getirecek yöntem ise panel yüzeyini su sistemiyle yıkamak olarak görülmektedir. Yıkama yöntemiyle hem panel sıcaklığı düşürmekte hem de panel yüzeyindeki kirlilikler temizlenmektedir.
Derin öğrenme yöntemi ile kumaş hatalarının otomatik tespiti ve sınıflandırılması
Günümüzde, tekstil endüstrisi sürekli büyüyen ve günler geçtikçe önemi artan bir sektör haline gelmiştir. Kaliteli ve hatalardan arınmış kumaşların üretimi, hem üreticiler hem de tüketiciler için büyük bir öneme sahiptir. Ancak, kumaş üretimi sürecinde hataların ortaya çıkması kaçınılmaz olabilir ve bu da hem maliyeti artırır hem de ürünlerin kalitesini olumsuz yönde etkiler. Bu nedenle, kumaş hatalarının erken aşamada tespit edilmesi ve müdahale edilmesi, tekstil endüstrisi için kritik bir ihtiyaç haline gelmiştir. Bu tez çalışmasında, topbaşı hatası, yağ lekesi hatası, iplik hatası ve delik hatası gibi hatalardan oluşan bir kumaş veri seti için YOLO temeli bir yapay zeka yöntemi ile etkin bir hata tespit sisteminin geliştirilmesi amaçlanmıştır. Bu çalışmada, daha önce kullanılmamış bir veri seti toplanmıştır. Veri seti, kumaş hatalarının tespiti amacıyla özel olarak oluşturulmuş ve mevcut literatürde yayınlanmış başka bir veri setinden de faydalanılmıştır. Her iki veri seti birleştirilerek 5590 görüntü den yeni bir veri seti elde edilmiştir. İlk olarak veri setindeki tüm görüntüler için hata sınıfına ait etiketler belirtilmiştir. Daha sonra YOLO algoritmasının performansını iyileştirmek için sadece eğitim veri setindeki görüntüler veri arttırma yöntemleri ile sentetik olarak çoğaltılmıştır. YOLO algoritmalarının kumaş hatalarını tespit edebilme yeteneklerini değerlendirmek için bir takım deneysel çalışmalar Python programı ile gerçekleştirilmiştir. Yapılan deneysel analizlere göre modellerin ortalama hassasiyet (mAP), kesinlik (p) ve duyarlılık (R) performans ölçüleri başarı performans değerleri elde edilmiştir. Yapılan deneylerdeYOLOv5s, YOLOv5m, YOLOv5l, YOLOv5x, YOLOv7 ve YOLOv8s algoritmaları kullanılmıştır. Deneysel sonuçlara göre, YOLOv5s, YOLOv5m, YOLOv5l, YOLOv7 ve YOLOv8s modellerin verdiği sonuçları ise sırasıyla; %84, %84.5, %86, %85 ve %83 ortalama hassasiyet (mAP), %87, %87, %87.2, %86 ve %83.2 kesinlik (P), %83, %83.2, %84.6, %84 ve %83.2 duyarlılık (R) değerleri elde edilmiştir. Yapılan çalışmaların neticesinde en iyi sonucu Yolov5x modeli vermiştir %86 ortalama hassasiyet, %87.6 kesinlik ve %85.1 duyarlılık değerlerine ulaşmıştır. Çalışmanın sonuç kısımında ise YOLOv5x'in sonucu diğer modellerin sonuçları ile karşılaştırılmıştır ve iyileştirme değerleri hesaplanmıştır. Gelecekteki bölümlerde YOLO algoritması ve yapılan işlemlerden detaylı bir şekilde bahsedilmişitr.
İnsansız hava araçlarının hafif bir derin öğrenme yöntemi ile sınıflandırılması
Son yıllarda, İnsansız Hava Araçlarının (İHA) askeri ve güvenlik bölgeleri haricinde sivil alanda da kullanılmaya başlanmasıyla tarım alanlarını ilaçlamada ve gözetlemede, orman yangınlarını tespit etmede ve söndürmede, nakliye, doğal yaşamı gözlemleme, havadan fotoğraf ve görüntü çekimi yapma, deprem sonrası oluşan hasarların tespiti ve keşif yapma gibi birçok alanda yaygın olarak kullanılmaktadır. Günümüzde üretilen İHA'lar çok farklı ebat, şekil, yapılandırma ve karakterde olabilmektedir. Bu durum, önemli ve kritik görülen yerlerde İHA'ların tespit edilmesini ve sınıflandırılmasını zorunlu hale getirmektedir. Bu tez çalışmasında, İHA'ları ve kuşları sınıflandırmak için yoğun blok ve dikkat mekanizmasından oluşan bir evrişimli sinir ağı modeli önerilmiştir. Tez çalışmasında, çevirimiçi olarak elde edilen bir veri seti toplanmış ve bu veri setindeki görüntülere veri arttırma işlemi uygulanmıştır. Önerilen derin öğrenme modelinin sınıflandırma performansını değerlendirmek için karşılaştırma çalışmaları gerçekleştirilmiştir. Bu çalışmalarda, ön eğitimli modellerden olan AlexNet, ResNet-101, MobileNet-v2, ShuffleNet ve GoogleNet tercih edilmiştir. Gerçekleştirilen çalışmalarda, önerilen derin öğrenme modelinin doğruluk değeri %98.04 olarak hesaplanırken AlexNet, ResNet-101, MobileNet-v2, ShuffleNet ve GoogleNet'in doğruluk değerleri ise sırasıyla %91.50, %90.85, %92.81, %93.46 ve %94.77 olarak elde edilmiştir. Bu değerlerden önerilen yöntemin etkin bir sınıflandırma performansına sahip olmasının yanı sıra daha güvenilir sonuçları da garanti edebileceği değerlendirilmiştir.
Maksimum güç noktası takipli solar kojenerasyon sistemin geliştirilmesi
Sanayi devriminden sonra meydana gelen makineleşmeden dolayı enerjiye talep her geçen gün artarak devam etmiştir. Sanayi devrimden günümüze kadar yenilenemeyen enerji kaynakları olan fosil yakıtlar ile kontrolsüz şekilde enerji ihtiyacı karşılanmaya çalışılmış ve fosil kaynaklardan dolayı bir çok savaşlar yaşanmıştır. Yenilenemeyen enerji kaynaklarının bir gün biteceğini ön görülerek ve enerji ihtiyacını gidermek için çeşitlendirmek amacıyla yenilenebilir enerji kaynaklarına yönelik arayışlar olmuştur. Bu arayıştan sonra HES, RES, GES vb. tesislere yatırımlar yapılarak Yenilenebilir enerji kaynaklarını çeşitlendirilmiştir. Başlarda fosil yakıtlara göre pahalı olmasından dolayı yeteri kadar ilgi görmemesi zamanla küresel ısınmanın özellikle iklim konusunda etkileri hissedilecek derece artınca, ülkeler tarafından şirketler bu alana yatırım yaptırılmaya teşvik ettirilmiştir. Günümüzde Güneş enerjisinden maksimum verim alınması için hem PV panelde kullanılan hücreler üzerinde iyileştirilmeler yapılmış hem de güneş takip sistemleri yapılarak güneşten maksimum verim alınması sağlanmıştır. Bu projede yapılan işlem de ise; tek eksenli güneş takip sistemi yapılarak güneşi gün içerisinde takip edip, güneş ışınlarından maksimum verim alması amaçlanmıştır. Ayrıca Fotovoltaik panellerin sıcaklık artışından kaynaklı verimin düşmesinin önüne geçmek için de şebeke suyu yardımıyla Güneş panellerinin soğutularak verimin arttığı görülmüştür. PV panelin soğutulmasından ortaya çıkan sıcak su ile de evlerde, işletmelerde, hastanelerde, üniversitelerde vb. yerlerde kullanılarak atıl durumdaki ısı enerjisinin yeniden kullanılması sağlanmıştır.
Sezgisel algoritma yöntemleriyle kanal kestirimi
Günümüzün hızla gelişen kablosuz haberleşme teknolojileri, yüksek veri hızı, düşük gecikme ve güvenilir iletişim gibi temel gereksinimleri karşılamak üzere sürekli olarak yenilikçi çözümlere ihtiyaç duymaktadır. Bu kapsamda, Dikgen Frekans Bölmeli Çoğullama (OFDM) sistemleri, çoklu yol etkisine karşı dayanıklılık, spektral verimlilik ve modülasyon/demodülasyon kolaylığı gibi avantajları nedeniyle modern iletişim sistemlerinde yaygın olarak kullanılmaktadır. Ancak, OFDM sistemlerinin performansı, kanalın zamana ve frekansa bağlı değişken yapısından ötürü ciddi ölçüde etkilenmekte; bu da kanal kestirimi problemini sistemin başarımı açısından kritik hale getirmektedir. Bu tez çalışmasında, OFDM sistemlerinde kanal kestirimi problemi sezgisel algoritmalar kullanılarak ele alınmış ve çözüm yolları geliştirilmiştir. Geleneksel kestirim yöntemleri olan En Küçük Kareler (Least Squares - LS) ve Minimum Ortalama Kare Hatası (Minimum Mean Square Error - MMSE) algoritmaları, istatistiksel varsayımlara ve kanalın önceden bilinen özelliklerine dayanmakta olup, gerçek zamanlı ve dinamik kanal ortamlarında yeterli performansı her zaman sağlayamamaktadır. Bu eksiklikleri gidermek amacıyla doğadan esinlenen, problem çözme gücü yüksek olan sezgisel ve meta-sezgisel optimizasyon algoritmaları değerlendirilmiştir. Bu tezde; İstilacı Yabancı Ot (İYO), Karıştırılmış Kurbağa Sıçrama Algoritması (KKSA), Öğretmen-Öğrenci Tabanlı Optimizasyon (ÖÖTO) ve Mors Optimizasyon Algoritması (MOA) gibi güçlü sezgisel algoritmalar kanal kestirimi amacıyla uygulanmıştır. Her bir algoritmanın OFDM sistemine entegrasyonu detaylandırılmış ve kestirim sürecinde karşılaşılan optimizasyon problemleri, ilgili algoritmaların yapısal özelliklerine göre çözülmüştür. Simülasyon çalışmaları MATLAB ortamında gerçekleştirilmiş olup, sezgisel algoritmaların performansları Bit Hata Oranı (Bit Error Rate - BER), Ortalama Karesel Hata (Mean Square Error - MSE) ve yakınsama süreleri gibi ölçütlerle değerlendirilmiştir. Ayrıca bu sezgisel algoritmaların elde ettiği sonuçlar, klasik LS ve MMSE kestirim yöntemleriyle karşılaştırmalı olarak analiz edilmiştir. Elde edilen bulgular, önerilen sezgisel yöntemlerin özellikle karmaşık ve gürültülü kanal ortamlarında daha düşük hata oranlarıyla daha yüksek doğrulukta kestirim sağladığını ortaya koymaktadır. Bu çalışma ile hem kanal kestirimi alanında yapay zekâ tabanlı alternatif çözüm önerileri sunulmuş, hem de OFDM sistemlerinin sezgisel optimizasyon algoritmalarıyla desteklenerek daha etkin ve güvenilir hâle getirilmesi amaçlanmıştır. Bu yönüyle tez, kablosuz haberleşme sistemleri, kanal modelleme ve optimizasyon temelli kestirim alanlarında literatüre katkı sağlamayı hedeflemektedir.
Derin öğrenme yöntemlerini kullanarak akciğer görüntülerinden hastalık tespiti ve sınıflandırılması
Akciğer hastalıkları her geçen gün sağlık alanında ele alınması gereken önemli rahatsızlıkların başında gelmektedir. Hastalığın erken aşamada tespit edilmesi ve doğru tedaviye hızlı bir şekilde yanıt vermesi hastalar için hayati bir öneme sahiptir. Doktorların hastalığı erken aşamada belirlemesinde ve hastalıklı bireylerde tedaviye erken başlanması hastalığın seyrini yüksek oranda etkilemiştir. Akciğer hastalık tespiti ve sınıflandırılması uzmanlık gerektiren zor bir süreç olup gerekli müdahalenin doğru zamanda yapılması tedavinin başarısını ortaya koymaktadır. Akciğer hastalıkları her geçen gün çeşitlenmekte ve analiz edilmesi zor bir durum haline gelmektedir. Yanlış hastalık tespiti ve hastalığın doğru şekilde analiz edilememesi insan sağlığını büyük oranda etkilemiştir. Günümüzde yapılan çalışmalarda yapay zeka yönteminin kullanımı hastalığı tespit edebilmede ön plana çıkmaktadır. Görüntülerin analizi ve doğru bir şekilde sınıflandırılması hastalığın durumunu hafifletmekte ve ölüm riskini azaltmaktadır. Bu tez çalışmasında YOLO modelleri kullanılarak akciğer görüntülerinden oluşan bakterili, virüslü ve normal hastalık görüntüleri olan veri kümesi kullanılıp YOLO algoritmasının farklı sürümleri (YOLOv8, YOLOv9, YOLOv10, YOLOv11, YOLOv12) karşılaştırmalı olarak incelenmiş ve hastalık tespitinin gerçekleştirilmesi amaçlanmıştır. Yapılan deneylerde sırasıyla YOLOv8 %87,9, YOLOv9 %85, YOLOv10 %86,8, YOLOv11 %87,8 ve YOLOv12 %86,1 ortalama hassasiyet (mAP) değerine ulaşmıştır. Çalışmada YOLOv8 diğer modellere göre model mimarisindeki yapısı ve özellik çıkarımı sayesinde nesne tespitini genelleştirme yeteneğinden dolayı en iyi performansı göstermiştir. Çalışmanın hastalık tespit yeteneğini arttırmak için en iyi performans gerçekleştiren YOLOv8 modeline SE ve uzlamsal dikkat modülü entegre edilmiştir. Geliştirilen YOLOv8 modeli %88,7 mAP, %88 duyarlılık, %78,6 kesinlik ve %83 F1 skoruyla en iyi başarıyı yakalamıştır. YOLOv8 modeline entegre edilen SE ve uzlamsal dikkat modülü çalışmanın sonucunu %1,01 oranında iyileştirmiştir.
PEDOT: PSS psödo-kapasitör ve AlSb nanokristalleri tabanlı optoelektronik nöral arayüzler
Neural interfaces are the transducers that connect biological systems with artificial systems. They can read and/or write biophysical cues through electrical, mechanical, thermal and chemical mechanisms. The writing part, neural stimulation, have been widely used in biological applications such as understanding of the complex biological processes and treatment of neurological disorders. Particularly, semiconductor and metal devices were already used in biointerfaces, however, their spatial resolution, interference with recording systems and potential hazardous effects limit their efficacy. Optoelectronic stimulation with photovoltaic systems is a wireless alternative to electrode-based stimulation techniques. In these platforms, neural stimulation can be achieved by faradaic, thermal and capacitive processes. The later has gained significant attention since it is based on the perturbation of ions in the interface and resulting electrochemical gradient to induce membrane potential changes near device/cell interface. The biointerfaces that are designed for capacitive photocurrent generation rely on the double-layer capacitance at the electrode/electrolyte interface. However, this limits the total charge injection capacity. In this thesis, we developed three different approaches to increase total capacitive charge injection efficiency. Firstly, we combined a pseudocapacitive interface with an organic photovoltaic unit. Secondly, we utilized aluminum antimonide nanocrystals as the interfacial hole transport layer and showed successful neural activation on primary hippocampal neurons. Finally, we integrated PEDOT:PSS hydrogels with an organic photovoltaic device, which significantly increased the charge injection efficiency. Moreover, the potential of hydrogel integration was demonstrated with stimulation of primary hippocampal neurons and modulation of the cardiac myocyte beating frequency. Owing to their soft, organic, and biocompatible nature, organic/inorganic hybrid photovoltaic biointerfaces reduce the device-tissue mechanical mismatch, cytotoxic effects, and increase functional lifetime in biological medium. Hence, they hold high promise for future cardiac and retinal implants.
Waveguide-integrated germanium photodetector design and optimization for sensing and telecom applications
With the developing technology, electronics has become unable to reach the high speeds and high efficiency required by new applications. Silicon photonic integrated circuit technology has been proposed to solve this problem. As part of silicon photonics, photodetectors are one of the cornerstones of silicon photonics technology. In the early years of silicon photonics, photodetectors were mainly designed for fiber optical telecom applications. Silicon photonics and photodetectors can be used in many other applications such as sensors, cameras, LIDAR, and photovoltaics. In this thesis, geometric optimization of waveguide integrated Ge VPIN photodetectors has been studied. Two different photodetector designs were developed with simulations for two different application types. The first design is a photodetector having ultra-high responsivity (1.17\ AW^{-1}) and ultra-low noise (\ 48.38\ nW NEP) at 0V bias operation that can be used in optical time domain reflectometer sensor applications. This photodetector adopts a 500\ nm-thick, 2.5\ \mu m-wide and 100\ \mu m-long Ge layer. It has a wide dynamic range (55.76\ dB) and could allow for high contrast readout in long-haul OTDR applications. The second photodetector adopts a 300\ nm-thick, 2\ \mu m-wide, and 10\ \mu m-long Ge layer showing a 0.972 AW^{-1} responsivity with a bandwidth of 84.26 GHz at -2V bias. Due to having ultra-high bandwidth, high responsivity, and low dark current (8.67\ nA), it could be suitable for state-of-the-art telecom applications such as the 400G-BASE-DR4 ethernet standard. These designs and our design guidelines could pave the way for optimized silicon waveguide-integrated germanium photodetectors with superior and refined performance in sensing, telecom, and imaging applications.
Computational characterization of noise in nonlinear nanomechanical resonators
Recent breakthroughs in nanotechnology have paved the way for huge advances in the field of nano-electro-mechanical systems (NEMS). Nanomechanical resonators are used as accurate mass and force sensors in mass spectrometry and atomic force microscopy. In most current sensing schemes, the resonators operate in the linear regime, where changes in mass and force are detected by tracking the shifts in the resonance frequency. The use of resonators operating in the nonlinear regime has been recently proposed due to shrinking device sizes and some claimed advantages over linear operation. As with linear resonators, the sensor performance is ultimately limited by the inherent frequency fluctuations arising from various sources of noise. However, due to nonlinear behavior, device dynamics is intricate and noise characterization is more challenging. The fundamental sensitivity limits of resonant sensors operating in the nonlinear regime need to be determined quantitatively in order to examine their performance and assess their pros and cons over linear resonators. In this thesis, we present an efficient computational technique for characterizing the frequency fluctuations in nonlinear Duffing resonators due to noise. The method is based on repurposing and adapting a non Monte Carlo stochastic analysis technique that was originally developed for noise analysis of analog electronic circuits. The proposed technique presents the capability for noise characterization under various dynamic sensing schemes, e.g., when drive frequency is time-varying and continually updated in a closed-loop setup. We compare the results obtained with the proposed technique against extensive, carefully run Monte Carlo simulations and demonstrate that the same accuracy can be achieved with drastically reduced computation time and in a more robust manner. We furthermore compare the results of our proposed method with experimental open-loop sweep characterizations reported in the literature. We analyze the open-loop sweep method and highlight its shortcomings as a sensing scheme. Finally, we compare the noise performance in the linear and nonlinear operating regimes, and discuss the advantages of using one over the other.
Polytopic matrix factorization (PMF): A new data decomposition tool
Matrix factorization methods are widely used in signal processing and machine learning applications. These methods lay the foundation of a large number of algorithms utilized in problems of those areas. As one of the main problems, we try to discover the information hidden inside input data. As a general solution strategy for this problem, input data is modeled as a product of two factors/matrices. This is known as Blind Source Separation (Blind Data Decomposition) in the signal processing literature. In this thesis, we introduce Polytopic Matrix Factorization (PMF) as a novel data decomposition approach. We model input data as unknown linear transformations of some latent vectors drawn from a polytope. The choice of polytope determines the presumed structure of latent vectors and their relationships. We first propose the identifiability criterion and identifiability conditions of PMF regarding the latent vectors. We introduce a sufficient condition for identifiability, which requires that the maximum volume inscribed ellipsoid of the polytope is contained in the convex hull of the latent vectors with a particular tightness constraint. We propose PMF identifiability results of special polytope cases corresponding to widely utilized feature attributes in applications. Then, a generalized version of PMF is presented with the characterization of eligible polytope choices and we propose to use a decision algorithm to determine these eligible polytopes. The proposed PMF tool is extended by considering a special class of linear mappings on eligible polytopes. We further extend PMF as Bounded Matrix Factorization and provide identifiability results of some bounded sets referring to polytopes. We furthermore present a PMF algorithm and interesting examples that utilize PMF as a data decomposition tool. PMF enables us to use infinitely many polytope choices and bounded sets in characterizing latent vectors. Therefore, it is possible to define different presumed structures and relations for latent vectors such as nonnegativity, sparsity and such attributes in subvector level. In brief, we present a novel data decomposition tool that provides a high degree of flexibility in terms of presumed structures of latent vectors and offer examples illustrating this flexibility with different eligible sets and corresponding feature attributes.
Remote sensing, tracking and imaging inside MRI systems
In addition to diagnostic and therapeutic uses, clinical magnetic resonance imaging (MRI) systems have also been used for interventional procedures. However, safety risks associated with increased specific absorption rate and tissue overheating should be measured in-situ and controlled during such procedures. Here, we introduce a self- resonating radio frequency (RF) sensor capable of remote temperature sensing to serve as a visual indicator of in-situ temperature changes during real-time MRI interventional operations. We propose a new sensor design that uses dielectric properties as the tuning mechanism for the sensor resonant frequency and the temperature dependence of permittivity. Using a 7 Tesla (7T) preclinical MRI, we demonstrate ex vivo feasibility and remote temperature sensing capabilities in the clinically relevant temperature range of 36- 42°C. The miniature design of the sensor allows its placement on the tip of a catheter, where it can be used for both catheter tracking and temperature sensing. The RF sensor is tuned to match the resonant frequency of 7T MRI (298 MHz), enabling a hyperintense signal on the sensor in the MR images. Hence, it can be used for three-dimensional position tracking during the steering of a given interventional medical device, such as a catheter. As temperature increases, the sensor detunes due to the change in the relative permittivity, and the hyperintense signal disappears in the MR image, serving as a direct visual indicator of the temperature change in real-time without a need for post-processing. Since this technique is based on common MR imaging sequences, the same image can be used for both device localization and temperature measurement. 0.6°C accuracy is achieved in the physiological range between 36°C and 42°C. Such RF sensors could provide safer operations in future MRI interventional procedures with potential local increased temperatures. MRI is also a heavily utilized medical diagnostic tool that usually employs a contrast agent for enhanced image sensitivity and selectivity. Superparamagnetic iron oxide nanoparticles (SPIONs) have been shown as strong MRI contrast agents in the literature but usually with a dark (T2) contrast. However, bright contrast (T1 contrast agent) is preferred by the radiologists in the clinic, yet difficult to achieve. Here, we show T1 contrast generation of polyacrylic acid-coated superparamagnetic iron oxide nanoparticles (SPION-PAA) in our 7 Tesla MRI. We showed that such T1 contrast is achievable in ex vivo mouse experiments, as well. Interaction of SPION with UV or visible light is also an exciting phenomenon that may be exploited in photopolymerization to produce polymer/SPION nanocomposites for both enhancing MRI imaging contrast and 3D printing small-scale MRI robots. Using biocompatible SPIONs of nanoscale size and high stability eliminates the need for an initiator that might be undesirable in biological applications. We show that SPION-PAA can also be used as an initiator in the photopolymerization of vinyl monomers. Differential scanning calorimeter experiments were conducted to find optimal parameters for the highest and fastest conversion conditions. Furthermore, synthesized SPION/polymer hybrids/gels were demonstrated as efficient sensitizers for hyperthermia in an alternating magnetic field.
Full-duplex relay based energy harvesting wireless network
According to the recent Ericsson mobility report, 24.6 billion sensor nodes are expected to be installed by 2025. Increasing the lifetime of this massive battery-powered installation can only be achieved by using low-power transceivers with energy harvesting capability. For this purpose, we investigate the resource allocation in full-duplex (FD) relay-based wireless powered cooperative communication networks (WPCCN), and Simultaneous Wireless Information and Power Transfer (SWIPT) networks. In the SWIPT network, the nodes are being able to receive data and harvest energy to recharge their batteries simultaneously by using the same radio frequency (RF) signal from the hybrid access point (HAP). In WPCCN, energy is transferred in the downlink from the HAP to the nodes, and information is transmitted in the uplink from the nodes to the HAP. We consider novel optimization frameworks for the optimization of time allocation, scheduling, and relay selection. In the first part of the thesis, we work on FD-WPCCN. First, we study the transmission time minimization and throughput maximization optimization problems through relay selection. Then, we incorporate the on-off transmission scheme in FD-WPCCN and investigate the transmission time minimization and throughput maximization problems using scheduling and relay selection. In the second part of the thesis, we work on the FD-SWIPT network. First, we provide the outage analysis of an FD relay-based system and present the optimization problems for the outage throughput and energy efficiency maximization. Then, we extend the system model to a more practical multi-relay-user FD multiple-input-multiple-output (MIMO) system incorporating self-energy recycling. Self-energy recycling exploits the self-interference that arises due to the FD mode. We present the resource allocation and relay selection optimization framework with the objective of maximizing the sum throughput in a SWIPT FD MIMO system.
Visible light communication and positioning for autonomous vehicles
Automotive research is currently heavily oriented towards autonomy and especially developing reliable vehicular connectivity and localization technologies for autonomous driving. Since existing technologies have failed to satisfy all requirements so far, new complementary technologies are sought. Specifically, radio frequency communications like cellular and "DSRC" suffer from reliability and security issues at mid-range (<100 m) congested driving scenarios due to heavy interference, and sensor-based localization systems (e.g., GPS, vision) fail to provide the required accuracy and rate due to sensor rate limitations and prohibitively high computational complexity. Recently, visible light communication (VLC) and positioning (VLP) technologies, based on modulated LED head/tail lights and low-cost photodiodes, were conjectured to be promising complementaries that can help solve these problems and enable challenging applications like collision avoidance and platooning. This thesis aims to prove that vehicular VLP can realize this promise. First, we describe the vehicular VLC system model and provide the problem definition for vehicular VLP considering relative positioning of vehicles using VLC signals. Then, we propose a "VLP-friendly" improvement to the vehicular VLC physical layer: A novel receiver design that enables realizing high-accuracy angle-of-arrival based vehicular VLP methods in low complexity and cost (named "QRX"). Next, we review state-of-the-art positioning algorithms used in vehicular VLP and propose new algorithms that advance the state-of-the-art, most of which are exclusively enabled by the QRX. Finally, we derive the Cramer-Rao lower bounds on positioning accuracy for all algorithms and also simulate their performances under challenging weather and noise conditions in realistic driving scenarios. The results prove the eligibility of vehicular VLP for use as a suitable complementary technology for collision avoidance and platooning scenarios in future autonomous vehicles.
Study of surface plasmon polariton interaction with 2D discontinuities: Reflection and transmission
Surface plasmon polaritons (SPPs) have been studied extensively in the past decades by virtue of their implications in nanotechnology, communications, and life sciences. Their ability to localize light beyond the limit of diffraction has given rise to different types of nano-optical devices, such as plasmonic waveguides and nano-antennas. In the analysis and design of discontinuities that may inevitably persist on such structures, complex computational tools such as full-scale electromagnetic solvers are mostly used. In order to provide a simpler design route, approximate yet efficient models, akin to the ones developed for microwave circuits in the 20th century, are needed. In this work, SPP behavior at canonical 2D discontinuities have been characterized by approximate and semi-analytical models, from which the reflection and transmission coefficients were derived. The first part of this thesis focuses on two possible simplified terminations of a 2D metallic half-space: i) dielectric and ii) metallic half-plane discontinuities. Following a review of electromagnetic theory behind plasmonics, SPP reflection coefficients from these terminations were derived through effective medium and mode-matching techniques. The validity and limits of these models were comparatively assessed through results obtained from finite-difference time-domain (FDTD) simulations in conjunction with the mode expansion method. The characteristic behavior of these terminations was associated with those of their counterparts in microwave transmission lines. The second part of this thesis involves the analysis of more complex geometries, including step and gap discontinuities, via approximate models and numerical results. The reflection and transmission coefficients of these discontinuities were contextualized within a microwave network formalism, which yielded intuitive results on the resonant behavior of SPPs at discontinuities with small dimensions.
Communication theoretical modeling and analysis of neuronal communication with synaptic plasticity
Emergence of nanoscale devices and their applications mandates the facilitation of communication between these devices, and hence, the development of nanonetworks, to overcome their computational and power limitations owing to their size. The nanonetworks need novel communication paradigms since they are fundamentally different from the traditional communication networks. The human body is a large-scale network of molecular nanonetworks composed of billions of nanomachines, i.e., cells, where molecules are used to encode, transmit and receive information. The largest and the most vital intra-body nanonetwork is the nervous nanonetwork. Thus, the aim of this thesis is to investigate the nervous nanonetwork from information and communication theoretical perspective, to lay down the foundations of a novel bio-inspired communication paradigm for nanonetworks. We focus on the communication theoretical analysis of single-input-single-output (SISO) as well as multiple-input-single-output (MISO) neuro-spike communication channels, considering memory and metabolic energy constraints. Synaptic plasticity is a ubiquitous phenomenon in central as well as peripheral nervous systems and is associated with memory and learning new behaviors and skills. Thus, we further aim to analyze the neuro-spike communication incorporating the plasticity. The results obtained from this study would allow us to select optimal parameters to realize effıcient bio-inspired nanonetworks. Moreover, since neurodegenerative diseases such as Alzheimer's Disease, Parkinson's Disease and Multiple Sclerosis are caused by the malfunction of the sub-processes of the neuro-spike communication, these results can be compared with the results obtained from diseased synapses to develop the future ICT-inspired diagnostic and treatment techniques for neural disorders.
Molecular communications: Novel relaying schemes and molecular beamforming
The communication in the small scale has received a growing interest in the recent years due to two main reasons. First of all, there is a wide variety of applications for which the traditional communication technologies are not suitable, especially those related to medical treatment and healthcare. Secondly, the communication in the nano/micro scale is feasible due to the advancements in the production of the electronic devices, both in terms of cost and scale constraints. As a prominent approach towards enabling nanonetworking, molecular communications (MC) has been proposed. First and foremost, compared to the conventional EM communication, MC solves the impediment of the size of antennas. Moreover, MC is a bio-compatible solution, which is a key aspect for its proposed usage areas. There exist numerous MC systems introduced in the literature, among which MC via diffusion (MCvD) is one of the most examined ones, since it is an efficient and practical solution that utilizes the diffusive characteristics of molecules for conveying the information. The signal arriving at the intended node features a heavy-tail shape due to the diffusion dynamics of the molecules, which leads to the MCvD systems being prone to inter-symbol interference (ISI). Alleviating ISI is one of the most investigated issues when it comes to MC. Apart from that, the communication range is another limitation that characterizes MCvD, due to the signal's amplitude rapidly decreasing as the communicating nanomachines get further from each other. In order to address these challenges, novel designs and algorithms in the communication perspective are proposed in the literature. Chapter 2 of this dissertation summarizes the single-input single-output (SISO) topology of MCvD, providing the foundations in the literature that are very useful in understanding the aforementioned challenges. Moreover, a brief literature review on relaying in MC is provided, laying the groundwork for the motivation behind two of the contributions of this dissertation. The first proposed scheme is introduced in Chapter 3. This work is inspired from the fact that most of the relay-based schemes in the literature of MC consider the relay to be positioned in the middle of the two nodes whose communication it facilitates. As this my not always be achievable in practice, an asymmetrical relaying case is considered. If the parameters of the two resulting communication links have the same values, the overall performance of an asymmetrical relaying scheme is prone to error propagation due to unequal error protection for these links. This results from the fact that the quality of the channel belonging to the node closer to the relay is higher compared to the channel of the further one. In order to overcome this, utilization of molecules with different diffusion coefficients is proposed for the uplink (UL) of the relaying system. The best performance in terms of bit error rate (BER) is obtained when the emitting points release different numbers of different types of molecules. Two parameter optimization methods are proposed and the analytical results are verified by computer simulations. The scheme proposed in Chapter 4 focuses on utilizing a relay for boosting the received signal's strength, which results in an improvement in the BER performance of the communication system. This scheme is referred to as optimal relaying. The proposed system consists of a transmitting point, a receiving node, and a relay somewhere in between them. The relay absorbs a fraction of the released molecules, until some time, and then re-directs it towards the receiver (Rx), which also absorbs from the molecules emitted from the transmitter (Tx) in the meantime. In other words, the overall absorbed molecules at the Rx come from two sources: the Tx and the relay. Overlapping these two components, such that the overall amplitude of the received signal is increased, is the main motivation of this work. In order to achieve this, optimizing the time until when the relay will be absorbing is proposed by means of maximizing a function called the signal to interference difference (SID). The system is modelled analytically as well. The BER results after optimization are compared with the BER performance of the traditional SISO system, and an improvement is shown to be attained with the incorporation of the optimal relay. The last contribution of the thesis focuses on cooperative sensing in MCvD networks. Particularly, a novel paradigm is presented, which is motivated by specific scenarios where MC can find application, such as sugar level stabilization of patients with diabetes. Inspired by the concept of beamforming in radio frequency (RF) communications, molecular beamforming is presented in Chapter 5. The proposed technique can find application in systems where the actuation of nanomachines is targeted. The typical proposed network consists of several sensors (Txs) that emit molecules with the purpose of activating one actuator (Rx) at a time, whose on or off state depends on the overall received signal. However, the sensors are prone to errors, thus they may wrongly emit/not emit. For this reason, the main motivation behind this idea is that when the number of sensors increases, the actuation error rate decreases. In order to achieve this, several design techniques are proposed, such that the overall received signal is composed of delayed versions of the components arriving from different sensors. The verification of the proposed method is achieved by the analytical derivation of the actuation error probability for several scenarios. Firstly, the system where sensors are forming a uniform linear array (ULA) is considered for presenting the scheme with less complexity. Afterwards, a more realistic scenario is considered, where the locations of the sensors are random. As for the computation of the actuation error rate, three different scenarios are presented, depending on the error source and how the actuation decision is made. The analytical results are validated with the ones obtained from computer simulations. The results reveal very interesting insights about the potential of molecular beamforming for actuation accuracy in MC networks. Finally, Chapter 6 concludes the dissertation by summarizing the main outcomes obtained from the aforementioned works. Additionally, several future research directions are provided.
Advanced spatial modulation systems for future MIMO systems
Spatial modulation (SM) seems like a promising transmission technique for multiple-input multiple-output (MIMO) systems to meet the requirements of future wireless communication technologies. SM activates only one transmit antenna in each time slot and conveys additional information bits through the indices of active transmit antennas. Thus, the spectral efficiency increases without increasing the modulation order. Besides, SM requires only one radio frequency (RF) chain, which means low-level hardware complexity and inter-channel interference (ICI) cancellation due to the activation of only one transmit antenna. In recent years, the design of SM-based transmission techniques has become a general research topic because of the advantages of SM, and it is seen that clever SM-based designs can increase the spectral efficiency and bit error rate (BER) performance concerning SM. In this thesis, three novel SM-based transmission schemes are proposed for next-generation MIMO systems. Firstly, a novel transmission scheme named transmit antenna grouping quadrature SM (TAG-QSM) is proposed. The motivation of TAG-QSM is to increase the number of information bits in the spatial domain to achieve lower modulation order than the existing schemes in the literature. In TAG-QSM, transmit antenna groups with an equal number of transmit antennas are created. After that, first, the incoming bit group determines the indices of transmit antenna groups separately for the transmission of the real and imaginary parts of data symbol. The second incoming bit group chooses one transmit antenna from each of the determined transmit antenna groups. Data symbol is determined using the last bit group. To conclude, the increase in the number of additional information bits in the spatial domain results in lower modulation order and improved BER performance concerning the benchmark schemes. In this study, the theoretical upper bound is derived for BER, and Monte Carlo simulations are demonstrated for comparing the BER performance of TAG-QSM with the benchmark schemes and showing the consistency of the theoretical upper bound of BER derivation. Secondly, flexible spatial modulation with transmit antenna selection (FSM-TAS) is introduced for future multiple-input multiple-output (MIMO) systems. In this scheme, the number of active antennas varies in each time slot depending on the incoming bits. After determining the number of active antennas, channel coefficients corresponding to each possible active antenna combination are added up. Then, a certain number of antenna combinations with largest gains is selected to apply spatial modulation (SM). For the proposed system, complexity and outage probability analyses are performed. In addition, it has been shown by Monte Carlo simulations that FSM-TAS provides better bit error rate (BER) performance than the benchmark scheme, named enhanced spatial modulation with generalized antenna selection (ESM-GAS) \cite{qing2021enhanced}, under the same spectral efficiency, the same number of transmitter and receiver antennas. Thirdly, spatial modulation (SM) using signal space diversity (SM-SSD) is proposed for multiple-input multiple-output (MIMO) systems. In this scheme, consecutive time slots are processed jointly and signal and space diversity (SSD) technique is applied for the transmission of data symbols to obtain transmit diversity. In addition, a clever active antenna activation algorithm is introduced to prevent transmit diversity degradation. An upper bound BER derivation is performed and compared with Monte Carlo simulations. Besides, BER performance comparison is demonstrated with the reference schemes in the literature. Lastly, the suboptimal solution for the rotation angles is expressed to maximize minimum coding gain distance (MCGD).
Active reconfigurable intelligent surface architectures for future wireless networks
Reconfigurable intelligent surface (RIS)-assisted communication has recently attracted the attention of the wireless communication community as a potential candidate for 6th generation (6G) of wireless networks. Various studies have been carried out on the RIS technology, which is capable of enabling the control of the signal propagation environment by network operators. However, when an RIS is used in its inherently passive structure, it appears to be only a supportive technology for communications, while suffering from a multiplicative path loss. Therefore, researchers have lately begun to focus on RIS hardware designs with minimal active elements to further boost the benefits of this technology. In this thesis, first, we present a simple RIS hardware architecture including a single and variable gain amplifier for reflection amplification to confront the multiplicative path loss. The end-to-end signal model for communication systems assisted with the proposed amplifying RIS design is presented, together with an analysis focusing on the capacity maximization and theoretical bit error probability performance, which is corroborated by computer simulations. In addition, the major advantages of the proposed amplifying RIS design compared to its passive counterpart are discussed. It is shown that the proposed RIS-based wireless system significantly eliminates the double fading problem appearing in conventional passive RIS-assisted systems and improves the communication energy efficiency. This thesis also introduces an RIS-assisted grant-free non-orthogonal multiple access (GF-NOMA) scheme. We propose a joint user equipment (UE) clustering and RIS assignment/alignment approach that ensures the power reception disparity required by the power domain NOMA (PD-NOMA). The proposed approach maximizes the network sum rate by judiciously pairing UE with distinct channel gains and assigning RISs to proper clusters. To alleviate the computational complexity of the joint approach, we decouple UE clustering and RIS assignment/alignment subproblems, which reduces run times 80 times while attaining almost the same performance. Once the proposed approaches acknowledge UEs with the cluster index, UEs are allowed to access corresponding resource blocks (RBs) at any time requiring neither further grant acquisitions from the base station (BS) nor power control as all UEs are requested to transmit at the same power. In addition to passive RISs containing only passive elements and giving an 18% better performance, a fully-connected active RIS structure that enhances the performance by 37% is also used to overcome the double path loss problem. The numerical results also investigate the impact of UE density, RIS deployment, RIS hardware specifications, and the fairness among the UEs in terms of bit-per-joule energy efficiency.
Design of microheater for thermo-optic tuning of silicon photonic filters
Microheater is the small-scaled version of resistive heater and uses Joule heating for heat generation. Having a low footprint and power consumption has enabled the microheater to be used in many areas. High thermo-optic coefficient of silicon has allowed this structure to be used in silicon photonics studies. This work focuses on designing an efficient microheater for our photonic filters. First, the appropriate microheater geometry was determined because the geometry of this structure has a crucial influence on the power consumption of the microheater and the temperature uniformity of the device to be heated. After the proper geometric structure has been chosen, the determination of the properties of this microheater is divided into two parts. These features play an essential role in reducing the temperature difference in the heated device. In the first part, the separation between the heater lines was determined. In the second part, the amount of overflow of the microheater on the device was designated. After specifying all features of the microheater structure, its electrical and thermal properties were investigated in the simulation environment. Finally, it was compared with the best-in-class microheaters, and the reasons for the performance differences were examined.
Novel waveform design algorithms for pulse compression radars
Radar systems have been used widely for the detection of remote targets since World War II, and since then, they have become ubiquitous in remote sensing applications. At its core, the radar probing waveform has received considerable attention in the past decades with the advances in digital hardware and signal processing techniques. Indeed, waveforms and their synthesis methods can find diverse applications in areas, not only in active remote sensing but also in communications and medical imaging. In recent years, computational methods have been devised to synthesize arbitrary waveforms under various practical constraints, including the Low Peak-to-Average-Power-Ratio (PAPR), unimodularity, correlation constraints and spectrum allocation restrictions with the aim of improving the underlying system performance. With the new trends in radar techniques, such as noise radar and cognitive radar, the transmit waveforms and pertinent terms such as waveform diversity and waveform adaptivity have attracted more attention due to their practical benefits. In this thesis, we give an introduction to the new trends and techniques, i.e. noise radar technology and cognitive radar, and further propose two novel waveform design algorithms, particularly for noise radar technology and spectrum-aware sensing systems such as cognitive radar. The first algorithm named 'Combined Spectral Shaping and Peak-to-Average Power Reduction (COSPAR)' which embodies a parametric window function as a spectral weighting to control the autocorrelation sidelobes and a Peak-to-Average-Power-Ratio (PAPR) reduction technique inspired by a phase retrieval algorithm is proposed to synthesize tailored noise waveforms for enhanced Low Probability of Intercept (LPI) radar operation. The proposed COSPAR algorithm is recommended for the generation of infinitely many orthogonal low PAPR noise-like signals suitable to feed up noise radar waveform libraries. Furthermore, a visual analysis tool based on Spectral Kurtosis in the Time-Frequency domain is proposed to assess the noise-like behaviour and pertinent LPI characteristics of the signal. In the second algorithm, we pose the waveform design task as a nonlinear large-scale optimization problem and propose a novel computational approach utilizing a nonlinear optimization technique i.e. Limited Memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) recursion to synthesize unimodular sequences with good correlation and spectral stopband properties. The proposed algorithm, named 'L-BFGS based Sequence Design (LBSD)', includes a modified search direction and step length rule to facilitate faster convergence and aims to accelerate the overall waveform generation process. As a result, the proposed method is viable for the agile generation of spectrally compatible waveforms that are essential for cognitive radars. Finally, the benefits and good features of the proposed COSPAR signals are shown in field trials using an experimental noise radar demonstrator system, and some results from these experiments are reported in the thesis. Additionally, the performance of the proposed LBSD algorithm is shown via numerical examples in comparison to existing state-of-art waveform design algorithms.
Wireless channel modeling based on extreme values theory for ultra-reliable low latency communications
Ultra-reliable low latency communication (URLLC) is one of the most important features of the fifth generation (5G) networks with the aim of supporting mission critical applications, such as remote control of robots, remote surgery, autonomous vehicles and vehicular teleoperation applications. At URLLC, the packet error rate (PER) is guaranteed to be as low as 10−9-10−5 to address the strict reliability constraint, while the acceptable latency is on the order of a few milliseconds. A key building block in the design of ultra-reliable communication systems is a wireless channel model that captures the statistics of rare events occurring due to significant fading. Extreme value theory (EVT) is a powerful framework that characterizes the probabilistic distribution of infrequent extreme events or equivalently the tail distribution. In this thesis, we propose a novel methodology based on EVT to statistically model the behavior of extreme events in a wireless channel for ultra-reliable communication. In the first part of the thesis, we initially include techniques based on EVT for fitting the lower tail distribution of the received power to the generalized Pareto distribution (GPD), determining the optimum threshold over which the tail statistics are derived, ascertaining the optimum stopping condition on the number of samples required to estimate the tail statistics by using GPD, and finally, assessing the validity of the derived Pareto model. Second, we model the tail distribution of non-stationary channel based on EVT by including techniques for splitting the channel data sequence into multiple groups concerning the environmental factors causing non-stationarity, and fitting the lower tail distribution of the received power in each group to the GPD. The proposed approach also consists of optimally determining the time-varying threshold over which the tail statistics are derived as a function of time, and assessing the validity of the derived Pareto model. Third, we propose EVT-based framework dealing with relatively low number of data samples to estimate the optimal transmission rate and validate it by assessing the outage probability so that reliability constraints are met with a given confidence for ultra-reliable communications. Fourth, we propose a novel channel modeling methodology based on multivariate EVT (MEVT) for a system using spatial diversity in multiple input multiple output (MIMO)-URLLC to derive the lower tail statistics of the received signal power in multiple dimensions while efficiently dealing with a massive amount of corresponding data. Accordingly, we adopt EVT to determine the optimum threshold over which the tail statistics are derived by using the UGPD model, validate the final model by using probability plots, and utilize MEVT to model the tail of the joint probability distribution by using the EVT-based logistic distribution and Poisson point process approaches. In the second part of the thesis, we address the requirement of ultra-reliability at the upper communication level by proposing a new algorithm based on the prior collection of data to eliminate power-consuming beam-tracking techniques while ensuring high received power with minimum diversity level.
Affective video summarization
With the availability of video sharing and streaming services, the media consumption behavior of users has moved from TV to the internet resulting in a surge in video content. The video summarization field produces solutions for efficient video representation, retrieval, and browsing to ease the complications caused by video content and traffic surge. Inspection of the new video content shows that user-generated human-centric video production and consumption consolidates most of the surge. Conventional video summarization neglects the human content and treats all video categories similarly. In this thesis, we argue that summarization of human-centric videos requires both understanding human behavior and video summarization. We break down this complex task into serial sub-tasks to understand complex human behavior through emotion recognition and video summarization. First, we represent the human video content by affective states and propose a multi-modal, multi-task learning-based framework estimating affective states from audio-visual input. Along with predicting affective states, we define a novel problem of detecting affective bursts to capture salient regions in the affective contour accurately. We define affective video summarization which focuses on the summarization of human-centric videos and proposes a framework that integrates affective information into the video summarization process. Finally, we presented a dataset referred to as AffWild2-VS annotated for video summarization, enabling joint research on video summarization and emotion recognition.
Reconfigurable intelligent surface-based novel transceiver architectures and multiple access
Over five generations of wireless communications networks, from 1G to 5G, the wireless channel was always a given system entity and not a design parameter. The wireless channel is dictated by the physical nature of the environment that contains the transmitter and receiver, the endpoints of a wireless network, which makes it random from their perspective. Hence, it cannot be manipulated, and the only way to deal with it is to compensate for its negative impact at the endpoints of the network. Recently, a new emerging technology called reconfigurable intelligent surfaces (RISs) has appeared to provide some degree of manipulation in the random wireless channel. With unprecedented capabilities, RISs somehow offer to make (even if partially) the random wireless channel as a design parameter that can be optimized jointly with the other conventional parameters to achieve a particular unified goal for the wireless network. An RIS is a metasurface in the form of a large array of passive and low-cost elements that can be controlled electronically to reshape the electromagnetic waves impinging the surface and absorb or reflect them to the opposite side. Owing to their unique properties of reconfigurability, low cost, and intelligence, recent literature envisions that RISs can find their way to integration with almost all of the existing wireless communication systems. Therefore, RISs has drawn growing attention in the wireless communication society as a promising emerging technology that can be a potential candidate for 6G and beyond wireless networks, which motivates this thesis. In this thesis, we investigate the potential of RISs to be used in the next generation of wireless networks as an enabling technology by examining how far it will impact the network's overall performance. Our investigation is carried out by proposing new solutions that redesign existing classical wireless communications systems to include RISs in their transceiver architectures and assess their performance compared to the classical systems. Specifically, this thesis proposes the design and performance analysis of seven RIS-assisted wireless communications systems, including single user and multiple users with multiple access, which can be categorized into three parts, as follows. In the first part of this thesis, we combine RISs with multiple-input multiple-output (MIMO) systems, where we consider vertical Bell Labs space-time (VBLAST) and Alamouti's schemes as the most common and practical MIMO schemes. For the VBLAST-based new system, an RIS is used to enhance the performance of the nulling and canceling-based sub-optimal detection procedure as well as to noticeably boost the spectral efficiency by performing index modulation (IM) at the RIS side. Furthermore, we propose an RIS-based transmitter for Alamouti's scheme that replaces the two radio frequency (RF) chains at the classical transmitter with a single RF signal generator. In the second part, we integrate RISs into non-orthogonal multiple access (NOMA) systems, as follows. First, we propose a novel NOMA solution with RIS partitioning with the aim of enhancing spectrum efficiency by improving the ergodic rate of all users and maximizing user fairness. In the proposed system, we distribute the physical resources among users such that the base station (BS) and RIS are dedicated to serving different clusters of users, thus, reducing the mutual interference between users' clusters. Second, in order to increase the coverage area, we consider the generalized version of RISs, namely, simultaneous transmitting and reflecting RISs (STAR-RISs). In particular, we investigate the BER performance of a NOMA network assisted by a STAR-RIS, where the STAR-RIS serves multiple non-orthogonal users located on either side of the surface by utilizing the mode-switching protocol. We derive the closed-form BER expressions in perfect and imperfect successive interference cancellation cases. Furthermore, asymptotic analyses are also conducted to provide further insights into the BER behavior in the high signal-to-noise ratio region. Third, we consider general system design and physical resources allocation problems for STAR-RIS-assisted NOMA networks. In particular, we propose a novel STAR-RIS-assisted NOMA system, where unlike most of the related works, we target scalable phase shift design that requires a reduced channel estimation overhead. The proposed system aims to maximize the overall sum rate while guaranteeing the quality-of-service requirements (QoS) for individual users. Finally, in the third part, we consider the RIS-assisted system design considering practical implementation problems associated with RISs. The first problem is the phase shift adjustment in RISs, which is associated with many issues in hardware implementation, limiting the RIS achievable gain. Therefore, we propose a low-cost, phase shift-free and novel passive beamforming (PB) scheme by only optimizing the on/off states of the RIS elements while fixing their phase shifts. The proposed PB scheme is shown to achieve the same scaling law (quadratic growth with the RIS size) for the signal-to-noise ratio as in the classical phase shift-based PB scheme, yet, with far less sensitivity to spatial correlation and phase errors. The second problem is the fact that RISs do not selectively reflect impinging electromagnetic waves; therefore, they can also reflect electromagnetic interference (EMI) from the surrounding environment to the receiver side. To solve this problem, we propose a novel EMI cancellation scheme to mitigate the impact of the EMI by exploiting its special time-domain structure and considering a clever passive beamforming method at the RIS. For all of the proposed system designs, we assess their performance using mathematical analyses by considering different relevant performance matrices, which are validated via simulation results. Also, using comprehensive computer simulations and under different system settings, we compare the performance of the proposed systems against their state-of-the-art counterparts using the same simulation parameters to guarantee fair comparison. The obtained mathematical and computer simulation results show that RISs have a huge potential to be a promising candidate for 6G and beyond wireless communications systems. RISs are shown to offer high degrees of freedom in terms of the system design, where they can be used as a supportive technology or main module within the existing wireless systems' architectures. Accordingly, they can offer system performance enhancement over different metrics, reduce the hardware design's complexity and cost, or even offer completely new features.
Inverse design of photonic structures using topology optimization
Photonic inverse design methods are significantly more efficient than traditional photonic design methods. Unlike traditional methods, it operates on the full design area with only two simulations, forward and adjoint simulations. Using the topology optimization method in the inverse design of photonic devices enables the devices to be optimized for user-specified parameters. The topology optimization method shows better performance when compared to other methods. On the other hand, the fabrication compatibility of device geometries is the most significant challenge with structures created through topology optimization. In this thesis, fabrication-compatible photonic devices, a power splitter, and a spectral filter are designed using the topology optimization method. To make the devices more fabrication-compatible, filtering methods were used to remove small features and smooth sharp corners. 90%:10% power splitters and spectral filters were fabricated on silicon-on-insulator and devices exhibited sufficient performance.
Flexible optoelectronic biointerfaces with pseudocapacitive MnO2 nanostructures for efficient photostimulation of neurons
This thesis explores the integration of flexible optoelectronic biointerfaces with 3D manganese dioxide (MnO2) nanoflowers to achieve safe and efficient photostimulation of neurons. Optoelectronic biointerfaces have attracted significant attention for their potential in the wireless and electrical control of neurons. By utilizing 3D pseudocapacitive nanomaterials with interconnected porous structures and large surface areas, these biointerfaces can effectively transduce light into ionic currents, meeting the requirement for high electrode-electrolyte capacitance. The MnO2 nanoflowers were grown through a chemical bath deposition technique on the return electrode, which was pre-coated with a MnO2 seed layer deposited via cyclic voltammetry. The nanoflower integrated biointerfaces exhibited excellent performance with a high interfacial capacitance (greater than 10 mF cm-2) and photogenerated charge density (over 20 μC cm-2) even under low light intensity conditions (1 mW mm-2). Importantly, the MnO2 nanoflowers induced safe capacitive currents through reversible Faradaic reactions and demonstrated no toxic effects on hippocampal neurons during in vitro experiments, making them a promising material for curvature fit integration with electrogenic cells. The functionality of the optoelectronic biointerfaces was assessed using a patch-clamp electrophysiology setup in the whole-cell configuration of hippocampal neurons, revealing their ability to elicit repetitive and rapid firing of action potentials in response to light pulse trains. This research underscores the potential of electrochemically-deposited 3D pseudocapacitive nanomaterials as a robust and efficient approach for controlling neurons through optoelectronic means.
Design and optimization of broadband silicon photonic devices
The integration of photonic systems on a chip offers enhanced performance, functionality, and reliability. While there are various material platforms available for constructing photonic integrated circuits, silicon stands out due to its compatibility with existing fabrication technologies and its high refractive index contrast. These features enable sub-micrometer waveguide scaling and dense packing of optical functions, combined with cost-effective and reliable manufacturing processes. Silicon photonics has already demonstrated its effectiveness in high-speed communication applications, but its potential reaches far beyond that. It shows promise for diverse applications like LIDAR, photonic switching, computing, and sensing, making it a versatile technology with a wide range of possibilities. Consequently, there is a growing demand for efficient and innovative photonic building blocks on the silicon platform to meet the specific requirements of these applications. This thesis focuses on the development of advanced photonic devices tailored for photonic integrated circuits. The first part presents the design and experimental demonstration of compact silicon photonic filters using optimized fast adiabatic waveguides, enabling arbitrarily wide bandwidths. These devices, with the widest bandwidths reported to date, find applications in high-throughput communications and ultra-broadband sensing. Subsequently, a novel deep photonics network architecture is introduced, enabling the rapid design of high-performance, multi-functional photonic devices. A proof-of-concept device with an arbitrary dispersion profile over a wide bandwidth is demonstrated. Lastly, a fabrication variation-aware device design concept is presented, utilizing the deep photonic network architecture to account for the effects of the actual fabrication process. Overall, this work contributes to the development of high-performance and versatile silicon photonic systems for a range of applications.
The design and the implementation of a wearable medical ultrasonic transceiver for bladder volume monitoring
The manuscript presents an innovative solution for continuous bladder volume mon- itoring outside hospital settings, through a wireless and fully integrated ultrasonic transceiver. The challenges encountered during the design, engineering, and real- ization of the device have been comprehensively discussed, along with the results that fall into the performance metrics to provide such an health tracking service. The device comprises multiple ultrasonic transducers for A-mode ultrasonic mea- surements, and the raw data collected from them are processed on different media. The proposed device completes a full loop of bladder volume detection to a visual- ization on a patient's smartphone, promising accurate and continuous measurement of bladder volume for the first time. The study's in-vitro and in-vivo tests demon- strate the device's reliability and accuracy, indicating its potential for diagnosing and managing lower urinary tract dysfunctions (LUTD). The device's wireless and non-invasive nature makes it a convenient and accessible option for patients and the healthcare providers, representing a significant step towards improving the quality of life for individuals with LUTD. The manuscript also highlights the potential of wearable technology in healthcare, and the development of this device could pave the way for further advancements in this field.
Electronic system design for continuous bladder volume monitoring
Bladder volume is an important parameter assessed by healthcare professionals for the early diagnosis and treatment of lower urinary tract dysfunctions. While bladder catheterization remains the gold standard method, its invasive nature often deters patients, and it can potentially lead to various complications. To address this issue, non-invasive bladder volume measurements are conducted using ultrasound. However, this measurement method still requires a clinical setting, and patients are asked to keep a bladder diary. Unfortunately, this approach is not sufficient and efficient enough for continuous monitoring. Moreover, despite significant advancements in wearable ultrasound studies, which have shown great promise, integrating with electronic devices still requires a separate desktop electronic setup. However, our solution involves offering an ultrasonic bladder volume monitoring (UBVM) device, achieved by adding miniaturized electronic hardware to ultrasonic transducers embedded in a flexible patch. We have tested the accuracy of this device with healthy volunteers. Through a non invasive measurement method, the flexible ultrasonic patch placed on the lower abdomen, transmits signals of bladder volume measurements to a cloud system via an electronic board. By utilizing a smartphone application connected to the cloud system, both users and healthcare professionals can continuously monitor bladder volume values. This setup ensures constant and autonomous monitoring, providing convenience and ease for both users and healthcare professionals. We present an example of the application of the Internet of Things (IoT) in the healthcare domain through the utilization of ultrasonic patch and electronic systems within our current setup. This thesis also paves the way for various wearable ultrasound projects that can incorporate IoT applications. Keywords: bladder, volume, flexible, ultrasonic, monitoring, continuous, IoT
Micromagnetic modeling and demonstration of wide bandwidth and ultralow power skyrmion-based spintronic devices and circuits
The scaling of microelectronics faces two major issues: memory bottleneck and power consumption. These issues prompted the proposal of in-memory computation that encodes information in spins. Using spins allows for nonvolatile logic and data manipulation functionalities. Skyrmions, which are nanoscale, topologically protected, chiral and surface spin textures, could be used for these functions. Skyrmions can be driven by charge or spin currents and have stability against stray magnetic fields and thermal noise. Using magnetic films that stabilize skyrmions might reduce the power consumption with respect to conventional microelectronics by several orders of magnitude. Despite the progress in experimental demonstrations and theoretical studies on skyrmion initialization, detection, and manipulation, using skyrmions for their promising digital logic applications has remained elusive. A set of new device designs based on skyrmion logic processing, and the investigation of their device physics is necessary. In this thesis, we designed a comprehensive skyrmion logic gate system, which includes a skyrmion clock generator and the essential connecting blocks including the duplicator, junction, and deflector. Then, all the logic gates AND, OR, inverter, NAND, NOR, XOR, and XNOR gates have been designed and their functionalities have been verified using computational micromagnetics. Each block has been investigated thoroughly in terms of their energy consumptions, Joule heating, temporal delay and transient response, bandwidth, cascadability, stability against thermal noise and external magnetic fields as well as functional sensitivity with respect to the geometric (sidewall roughness, notch size/feature, channel width), material (Gilbert damping, non-adiabaticity of spin transfer torques) and temperature variations. Skyrmions need to be generated and propagated using charge currents for integrated ultra-wideband spintronics. We introduce a device design for initialization and generation of periodic skyrmions from 114 MHz to 21 GHz using spin-polarized direct current. We demonstrate in micromagnetic simulations that skyrmion generation frequencies can be controlled reversibly over more than seven octaves of frequencies by changing DC current density. Thus, a non-volatile current-driven digital clock source has been established. The inverter can be driven with spin polarized current pulses, operate with wide bandwidth, low energy consumption (~1350 kBT/bit at room temperature), small footprint (~300 nm), no or very limited need for external magnetic fields, cascadability and room temperature thermal stability owing to substrate's thermal conduction. Using magnetic insulators to eliminate Joule heating hints that the power consumption could be even further reduced by 3-4 orders of magnitude. These analyses suggest that the inverter block could be cascaded and operated as part of digital spintronic circuits without loading or thermal drift effects. We presented a bit-slice circuit for a digital skyrmion full adder. This circuit could in principle be scaled to larger numbers of bits. As the skyrmion logic circuit designs might be integrated to electronic design automation workflows, a new protocol for the emulation of skyrmion signals with a tiled geometry is devised. The study presents a new comprehensive emulation and simulation software, featuring different elements of micromagnetic modeling, a Python package for automated simulation script development, and a library of block sets. A user-friendly web-based version of the emulator has been developed. The findings presented in this study suggest that skyrmionics has reached a milestone enabling a new area of ultralow power and ultra-wideband digital skyrmionics for in-memory computation.
Deep-learning based optimization framework for wireless powered communication networks
Net-zero-energy networks enable communication when the connection to the electric grid or changing batteries is not feasible for the information sources by balancing harvested and consumed energy. Radio Frequency Energy Harvesting (RF-EH) is a key enabler for net-zero-energy networks as RF signals are independent from climate conditions, more predictable, and controllable. This thesis focuses on low-complexity resource allocation and optimization for RF-EH networks. In particular, the joint relay selection, scheduling, and power control problem in multiple-source-multiple-relay RF-EH network is formulated with the objective of minimizing the total schedule length and the constraints on data demand, energy causality, and maximum transmit power. The formulated problem is a mixed-integer non-linear problem and proven to be NP-hard. As a solution strategy, a bottom-up approach is followed by starting from a simpler scheduling and power control subproblem, then extending the problem with additional relay selection variables. First, the iterative algorithms for optimal and suboptimal solutions to the problem based on conventional optimization theory techniques are proposed. To address runtime concerns inherent in iterative algorithms, then, this thesis presents a novel approach integrating deep learning methods with established optimization principles. Feed-forward Deep Neural Network (DNN) architectures are presented to solve the scheduling and power control subproblem while leveraging optimality analysis for feature set extension and output layer simplification in the proposed DNN models. The remaining relay selection problem is reformulated as a classification task, successfully addressed through deep learning models, including a feed-forward DNN and two Convolutional Neural Network (CNN) based architectures. This thesis further introduces teacher-student learning, a process that transfers knowledge from a complex teacher network to a simpler student network, for an even lower complexity solution to the relay selection problem. A novel dichotomous-based neural architecture search algorithm designs the student network architecture. The results demonstrate the noteworthy reduction in runtime complexity achieved through deep learning models while maintaining optimality for the solution of resource allocation problems in RF-EH networks.
A learned post-processing model with quality-gated convlstm for video compression
Recently, significant progress has been shown in applying deep learning to video compression and enhancement in terms of coding efficiency and quality improvement. In this work, a learned post-processing network is proposed for video compression and restoration tasks, which contains Quality-Gated Convolutional Long Short-Term Memory (QG-ConvLSTM) cells to enhance the quality of the compressed video frames considering the relative quality. With the proposed QG-ConvLSTM cell-based post-processing, the network exploits and takes full advantage of the inter-frame correlation and quality fluctuation between neighboring compressed frames. Since high-quality compressed frames provide more helpful information than low-quality compressed frames, the network can adjust the input and forget gate weights in QG-ConvLSTM cells. To show the enhancement of the proposed network that uses relative quality information between frames, video frames given to the network as inputs are compressed hierarchically with different qualities using the standard VVC (H.266) codec. In the proposed network, the weights given to the QG-ConvLSTM network are determined by extracting quality-related features from compressed video frames. Additionally, there is no reference frame for the quality feature extraction, so a no-reference image quality assessment method via Transformers, relative ranking, and self-consistency is suggested. The quality enhancement performance of the proposed network is measured with frequently used metrics, namely PSNR, MS-SSIM, and VMAF.
Video deinterlacing and demosaicing by deep learning
Deinterlacing and demosaicing are commonly used techniques in the image processing pipeline for consumer video. Despite the fact that real-world video deinterlacing and demosaicing are well-suited to supervised learning from synthetically degraded data because both degradation models are known and fixed, learned video deinterlacing and demosaicing have received much less attention compared to denoising and superresolution tasks. This thesis progressively explores feature alignment, integration and reconstruction stages for both tasks tailored to their known and fixed degradation subsampling patterns. We begin by presenting our initial work of a novel multi-field deinterlacing architecture that aligns features from adjacent fields to a reference field (to be deinterlaced) by designing novel deformable residual convolution blocks with two variants of different scales. To the best of our knowledge, this work is the first to propose fusion of multi-field features that are aligned via deformable convolutions for deinterlacing. Next, based on our initial work, we propose a novel multi-field full frame-rate deinterlacing network, which adapts the state of-the-art superresolution approaches to the deinterlacing task. This model incorporates self attention mechanism with deformable convolution residual blocks to align features and additively integrate aligned features for reconstruction. In order to reconstruct odd and even fields directionally, separate reconstruction modules are utilized according to the parity of each reference. Our extensive experimental results demonstrate that the proposed method provides state-of-the-art deinterlacing results in terms of both numerical and perceptual performance. Upon all these previous work, we propose a new multi-picture architecture for both video deinterlacing or demosaicing by aligning multiple supporting pictures with missing data to a reference picture to be reconstructed, benefiting from both local and global spatio-temporal correlations in the feature space using modified deformable convolution blocks and a novel residual efficient top-$k$ self-attention (kSA) block, respectively. Separate reconstruction blocks are used to estimate different types of missing data. Our extensive experimental results demonstrate that the proposed novel architecture provides superior results that significantly exceed the state-of-the-art for both tasks in terms of PSNR, SSIM, and perceptual quality. Ablation studies are provided to justify and show the benefit of each novel modification made to the deformable convolution and residual efficient kSA blocks.
Reconfigurable intelligent surface-centric communication networks: A comprehensive exploration of channel modeling and coverage extension
Reconfigurable intelligent surface (RIS)-empowered communication has received growing interest from the wireless research community due to its undeniable potential in extending the coverage, enhancing the link capacity, mitigating interference, deep fading, and Doppler effects, and increasing the physical layer security. RISs enable the control of wireless propagation through their unique electromagnetic functionalities and provide a new degree of freedom in the system design. Delving into the potential of RISs, this thesis unfolds across four detailed chapters, each contributing distinct insights into the deployment and application of RIS in enhancing wireless communication systems. Each chapter is crafted to provide a different perspective on RIS deployment and explain how this technology can be innovatively applied to support various aspects of wireless communication networks. This thesis takes a journey from the theoretical foundations of RIS to its envisioned practical applications and provides insights into how RIS can reshape wireless communication infrastructures to meet the demanding requirements of 6G networks. In the scope of this thesis, the RIS is introduced as a groundbreaking technology, reshaping the future of wireless communications by offering enhanced network capabilities, crucial for meeting the advanced requirements of next-generation applications. First, a comprehensive discussion of the intricacies of RISs in channel modeling is embarked on, which is a cornerstone for the effective deployment of sixth-generation (6G) wireless networks. In this regard, the physical channel modeling methodologies are provided for emerging RIS-empowered networks and aimed to fill an important gap in the open literature by providing an open-source and widely applicable physical channel model for mmWaves. Moreover, the open-source SimRIS Channel Simulator MATLAB package is introduced, which can be used in channel modeling and computer simulations of RIS-based communication systems with tunable operating frequency, terminal and RIS locations, number of RIS elements, and environments. The upcoming parts go beyond merely introducing the technology, delving into how RIS can be strategically employed to address some of the most challenging aspects of modern wireless communication, such as coverage extension, signal quality improvement, and network throughput enhancement. The following chapters build upon this foundation, each focusing on specific applications and implications of RIS in the realm of wireless communications. From intricate channel modeling to practical deployment strategies for coverage extension and advanced beamforming designs, the study offers a comprehensive and nuanced understanding of the multifaceted role of RIS in future wireless networks. This work stands as a testament to the transformative impact of RIS technology within the 6G landscape, offering an extensive exploration of its applications in channel modeling, coverage extension, and more. By weaving RIS into the fabric of various network design aspects, from sophisticated beamforming to coverage optimization, the study sets a robust framework for the advancement of future wireless communications, paving the way for the realization of the 6G vision.
Computationally efficient nanophotonic design through data-driven eigenmode expansion
Silicon photonic components require rapid design procedures with state-of-the-art optical metrics as the on-chip photonic applications advance. In this dissertation, a highly efficient and flexible method is introduced for designing a variety of low-loss waveguides in compact footprints. The proposed data-driven eigenmode expansion method represents waveguides as cascading eigenmode scattering matrices and propagation matrices. This method uses parallel data processing approaches to perform electromagnetic computations for simulating the optical response of individual waveguides in tens of milliseconds, orders of magnitude faster than the conventional methods, while achieving physical accuracies respected to 3D-FDTD. This framework, coupled with nonlinear optimization algorithms, designs adiabatic tapers, power splitters, and waveguide crossings that show near-lossless state-of-the-art operation within broad bandwidths. These devices and their 3D-FDTD simulations and experimental measurements highlight this methodology's capabilities and computational efficiency. A few photonic design problems currently under development will further demonstrate its applicability.
Deep learning based resource allocation for ultra-reliable communications in wireless control systems
Wireless Networked Control Systems (WNCSs) play an important role in fifth-generation (5G) and sixth-generation (6G) networks to support mission-critical applications, such as Internet of Things (IoT), Remote Driving, and Collaborative Robots (Cobots). WNCS design demands consideration of both control and communication systems requirements to guarantee the broadcasting of reliable control commands at low latency from the controller to the actuators. In the first part of the thesis, a joint optimization of control and communication systems in the Finite Blocklength (FBL) regime is devised with the objective of minimizing the total power consumption by optimizing the sampling period of the control system, blocklength, and packet error probability of the communication system. Then, the optimization framework is simplified using optimality conditions to only one decision variable of blocklength. Then, the new optimization problem is fed to an online Deep Reinforcement Learning (DRL) algorithm to be trained, and the changing wireless environment is learned, executing optimal results. Second, a diffusion model, specifically the Denoising Diffusion Probabilistic Model (DDPM), is proposed to allocate resources for WNCSs. The optimization framework is utilized to collect a dataset of Channel State Information (CSI) and its corresponding optimal blocklength values. Then, the dataset is used to train the DDPM-based model to learn the complex distribution of the solution and the environmental parameters and generate optimal blocklength values based on the CSI as conditional information. The proposed schemes perform close to the optimization theory-based solution and outperform the previously proposed benchmarks, demonstrating superior performance in total power consumption and avoiding critical constraint violations.
Ultra-fast simulation and design of nanophotonic devices
This thesis explores advanced methodologies for the ultra-fast simulation and design of nanophotonic devices, emphasizing the efficiency and accuracy of inverse design processes. The study introduces a multi-scale hierarchical inverse design approach that optimizes the fabrication compatibility and performance of nanophotonic components. A significant contribution is the development of a factorization caching method that dramatically reduces computational overhead in the gradient computation phase of inverse design, thereby accelerating the overall optimization process. This method proves particularly effective for two-dimensional finite-difference frequency-domain (FDFD) simulations, offering substantial time savings without compromising physical accuracy. Furthermore, the integration of deep learning techniques into nanophotonic simulations is examined. A dual-stage model trained with physics-based losses is presented, achieving simulation speeds up to 3408 times faster than traditional three-dimensional finite-difference time-domain (FDTD) methods. This model ensures high fidelity in the simulation results, maintaining robustness across various device geometries and sizes. The advancements detailed in this thesis not only enhance the computational efficiency of photonic device design but also expand the feasible design space, paving the way for the development of innovative and practical nanophotonic devices.
Frequency-domain modeling and optimization of graphene FET-based molecular communication receivers
Molecular Communication (MC) is a bio-inspired communication paradigm utilizing molecules for information transfer. Research on this unconventional communication technique has recently started to transition from theoretical investigations to practical testbed implementations, primarily harnessing microfluidics and sensor technologies. Developing accurate models for input-output relationships on these platforms, which mirror real-world scenarios, is crucial for assessing modulation and detection techniques, devising optimized MC methods, and understanding the impact of physical parameters on performance. In this thesis, we consider a practical microfluidic MC system equipped with a graphene field effect transistor biosensor (bioFET)-based MC receiver as the model system, and develop an analytical end-to-end frequency-domain model. The model provides practical insights into the dispersion and distortion of received signals, informing the design of new frequency-domain MC techniques, such as modulation and detection methods. The accuracy of the developed model is verified through particle-based spatial stochastic simulations of pulse transmission in microfluidic channels and ligand-receptor binding reactions on the receiver surface. In the second part, I detail the fabrication and characterization of a graphene bioFET-based MC receiver. This micro/nanoscale receiver is integrated into a microfluidic channel and functionalized with a biorecognition layer composed of single-stranded DNA molecules-based receptors, designed to detect the target information molecules flowing through the fluidic channel. A pre-equilibrium detection method was explored to improve the data rate. The sensor's initial performance tests involved detection experiments with information encoded into ionic concentration. The fabricated MC receiver was electrically characterized in terms of transfer characteristics and hysteresis at each step of functionalization. The parasitic current and mobility of the device is obtained. After functionalization with probe DNA, the receiver's time-varying response to concentration pulses of complementary target DNA was acquired with both fixed and varying pulse widths. Additionally, an intersymbol interference (ISI) analysis was conducted to evaluate the sensor's ISI performance. Finally, binary data transmission was performed using the MC setup, exploring various data rates and system parameters. The effects of key factors such as pulse width, symbol duration, flow velocity, and target DNA concentration were investigated. As such, this experimental work refined and optimized methodologies and designs from previous research, aiming at advancing practical MC techniques.
İndüktans ve direnç değişimlerinin döner asenkron motorların vektör kontrolü üzerindeki etkileri
Hızla gelişen teknolojiyle birlikte performans, hız ve kapasiteleri artan elektronik güç anahtarları ve mikroişlemciler sayesinde elektrik motorlarının sürülmesinde ve kontrolünde kullanılan evirgeç gibi sürücü düzeneklerinin gelişmesi sağlanmıştır. Bu gelişim asenkron motorlarda hız ve moment kontrol yöntemlerinin daha rahat bir şekilde uygulanmasına olanak sağlamıştır. Asenkron motorlarda kullanılan klasik kontrol yöntemlerinde hız ve momentin kenetlenmeli bir yapıya sahip olduğu görülmektedir. Ancak bu sorunun vektör kontrol yöntemleriyle giderildiği kanıtlanmıştır.Bu kontrolör sayesinde hız ve moment birbirinden bağımsız olarak iki bileşen şeklinde kontrol edilebilir. Ancak hız ve moment kontrolünde, makine parametrelerinin doğru olarak belirlenmesi çok büyük önem kazanmaktadır. Asenkron makinenin parametrelerinin doğru olarak belirlenmesi ise çok zordur. Çalışma koşulunda parametreler; frekans, akım büyüklüğü ve sıcaklığa bağlı olarak değişimler gösterebilmektedirler.IV Bu tezde, dolaylı rotor akısı yönlendirmesi ile üç fazlı asenkron motorun vektör kontrollü benzetimi yapılarak, parametre değişimlerinin bu alan yönlendirme yöntemine olan etkisi incelenmiştir. Bu amaçla dolaylı alan yönlendirme yönteminde rotor denklemleri elde edilerek bilgisayar ortamında benzetimi yapılmıştır. Çalışmada iki farklı motor ve iki farklı evirgeç kullanılarak, benzetimlerden elde edilen sonuçlar birbirleriyle karşılaştırılıp yorumlanılmıştır. Ayrıca doğrudan alan yönlendirme yöntemlerinin de parametre değişimlerinden nasıl etkilendiği konusuna kısaca değinilmiştir. Son olarak parametre değişimlerinin düzeltilmesi için yapılmış olan öneriler sunulmuştur.
Advanced modulation techniques for next-generation wireless communications
beyond necessitates innovative approaches to enhance spectrum efficiency, reduce complexity, and improve reliability. This thesis investigates the combination of index modulation (IM) and reconfigurable intelligent surfaces (RIS) with technologies such as orthogonal frequency division multiplexing (OFDM) and non-orthogonal multiple access (NOMA) to improve bit error rate (BER), spectral efficiency, and system reliability, while ensuring low complexity for practical implementation. In Chapter 2, a coordinate interleaved OFDM with in-phase/quadrature IM (CI-OFDM-IQIM) scheme is proposed to integrate OFDM with IM, achieving enhanced error performance and spectrum efficiency by reducing decoding complexity and transmitting real and imaginary components through subblock clustering. Chapter 3 introduces the RIS-aided enhanced receive spatial modulation (RIS-ERSM) scheme, which enhances spectrum efficiency and reliability by utilizing receive antenna indices in a clever manner and $M$-ary modulation, while Chapter 4 presents the RIS code IM with quadrature RSM (RIS-CIM-QRSM) scheme that integrates QRSM with CIM to improve error performance and energy efficiency. Chapter 5 proposes two hybrid-RIS-enabled enhanced reflection modulation (Hyb-ERM) schemes that enhance BER performance and spectrum efficiency, while Chapter 6 introduces the enhanced over-the-air RIS-IM (E-OTA-RIS-IM) system, leveraging RIS partitioning for OTA transmission of additional IM bits, leading to improved spectral efficiency. Finally, Chapter 7 presents a novel downlink NOMA system with passive RIS that eliminates successive interference cancellation (SIC) to reduce complexity and improve efficiency. This thesis demonstrates the potential of these innovative communication schemes to meet the stringent requirements of 6G and beyond, including higher spectral efficiency, low complexity, and enhanced system reliability.
Annotation consensus networks for improving machine learning with human annotated data
In deep learning applications, particularly in tasks involving subjective judgments such as emotion recognition, medical diagnosis, or content moderation, it is common to collect annotations from multiple human annotators for the same data instances. This practice captures diverse perspectives but often introduces variations or disagreements due to differences in interpretation, expertise, or individual bias. Effectively modeling and reconciling these inconsistencies is a key challenge in building robust and reliable machine learning models. Annotations are typically aggregated using simple methods such as averaging or majority voting. However, these conventional methods do not effectively capture common patterns or collective annotator behavior, and instead dilute valuable information about shared understanding among annotators. To address this problem, this thesis introduces a novel Annotation Consensus Network (ACN) that explicitly models and leverages annotator consensus as a more reliable training signal. ACN extracts a Learned Annotation Consensus (LAC) that aligns with all human annotations, providing improved supervision for machine learning models within an end-to-end training framework. This consensus reduces label variance and enables backbone networks to achieve more accurate and stable learning. In this thesis, ACN is integrated into two widely studied tasks, video summarization and continuous emotion recognition from speech, both of which rely on human-annotated ground truth targets. Extensive experiments demonstrate that ACN significantly improves the quality of training annotations. Models trained with ACN consistently outperform those using traditional annotation aggregation methods, achieving higher accuracy, better generalization, and increased robustness to annotator variability. This thesis highlights that explicitly modeling consensus among annotators is critical for improving deep learning performance. By learning a common representation from multiple annotators, ACN effectively harnesses collective expertise, providing a strong foundation for more accurate and reliable AI systems. This approach is beneficial for any domain facing annotation disagreements, enabling models to better understand and represent common human judgments rather than isolated individual perspectives.
Safe DRL for resource allocation with PAoI violation guarantees
In Wireless Networked Control Systems (WNCSs), control and communication systems must be co-designed due to their strong interdependence. This thesis presents a novel optimization theory-based safe deep reinforcement learning (DRL) framework for ultra-reliable WNCSs, ensuring constraint satisfaction while optimizing performance, for the first time in the literature. The approach minimizes power consumption under key constraints, including Peak Age of Information (PAoI) violation probability, transmit power, and schedulability in the finite blocklength (FBL) regime. PAoI violation probability is uniquely derived by combining stochastic maximum allowable transfer interval (MATI) and maximum allowable packet delay (MAD) constraints in a multi-sensor network. The framework consists of two stages: optimization theory and safe DRL. The first stage derives optimality conditions to establish mathematical relationships among variables, simplifying and decomposing the problem. The second stage employs a safe DRL model where a teacher-student framework guides the DRL agent (student). The control mechanism (teacher) evaluates compliance with system constraints and suggests the nearest feasible action when needed. Extensive simulations show that the proposed framework outperforms rule-based and other optimization theory based DRL benchmarks, achieving faster convergence, higher rewards, and greater stability.
Explainable ai and optimization theory based wireless radio resource management
The International Telecommunication Union's IMT-2030 framework identifies ``Integrated artificial intelligence (AI) and Communication'' as one of the core directions for 6G networks. This shift toward AI-driven solutions necessitates greater transparency in decision-making processes, where explainability is pivotal in establishing trust in AI systems, allowing network operators and engineers to understand, validate, and troubleshoot the decisions made by deep learning (DL) models. Additionally, robustness against out-of-distribution inputs and adversarial attacks is crucial to ensure reliable performance in diverse and evolving environments. These two factors—explainability and robustness—are essential to fulfilling the broader goals of AI-native 6G networks, where AI is not just an enhancement but a foundational component integrated into communication systems. This thesis examines the application of Explainable AI (XAI) for robust and transparent radio resource management (RRM) in AI-empowered 6G networks. First, the significance of this research is presented by unveiling the merits of explainability and robustness for 6G RRM, outlining a range of core explainability and robustness techniques for efficient RRM. Then, the practical implementation of these XAI techniques is analyzed, including a quantitative evaluation of their impact on key performance indicators (KPIs) in specific 6G scenarios. Two case studies are presented, showcasing their application in model simplification and enhancing RRM robustness. The first study proposes an explainable and robust DL-based framework for beam management in a millimeter-wave (mmWave) multiple-input multiple-output (MIMO) communications system. To minimize signaling overhead for beam alignment, a novel model-agnostic, feature relevance-oriented XAI framework is designed to rank and prioritize input features, reducing input size and beam sweeping time. To incorporate transparency and resilience into the DL-based beam alignment engine (BAE), a defense mechanism against adversarial inputs is developed, examining internal representations learned by deep neural networks (DNN) to provide interpretability and robustness against malicious/out-of-distribution inputs. The second case study focuses on deep reinforcement learning (DRL)-based RRM in vehicular networks for joint transmit power and spectrum allocation. First, a model-agnostic XAI-based methodology is devised to explain the inference process of multiple trained DRL agents. Then, a novel XAI-based feature importance ranking algorithm using Shapley additive explanations (SHAP) is developed to simplify the DRL agents' input state size, followed by an automated feature selection strategy to reduce model complexity. By eliminating low-importance features, the methodology produces a simplified model with fewer network parameters and lower training time while maintaining reasonable performance. This thesis further introduces a novel DRL-based framework for joint power control and block length allocation to minimize the worst-case decoding error probability for ultra-reliable and low-latency communication (URLLC)-based vehicular networks. Initially, an algorithm grounded in optimization theory is developed based on deriving the joint convexity of the decoding error probability in the block length and transmit power variables within the region of interest. Subsequently, inspired by event-triggered control systems, an efficient event-triggered DRL-based algorithm is proposed to solve the joint optimization problem. Incorporating event-triggered learning into the DRL framework enables assessing whether to initiate the DRL process, thereby reducing the number of DRL process executions while maintaining reasonable reliability performance. The results demonstrate a noteworthy reduction in training and runtime complexity compared to the conventional optimization-based solutions.
Çok girişli çok çikişli indis modülasyonu sistemlerinde makine öğrenmesi uygulamalari
Despite the practical challenges, multiple-input multiple-output (MIMO) systems are broadly utilized in modern communication systems due to the wide range of advantages provided by the use of spatial domain. To further improve the performance of MIMO systems, index modulation (IM) is applied using the spatial domain indices. IM can offer significant improvements in MIMO systems by transmitting information both with symbols and transmit antenna indices. Therefore, it is an emerging research topic in wireless communication. However, new challenges are introduced in MIMO-IM systems. For instance, due to the increased complexity in the transmission scheme, the receiver complexity is a serious practical concern. Moreover, poor channel conditions can cause significant performance degradation in MIMO-IM systems. Finally, although the IM literature is expanded in the recent years, the practical applications and use cases are not advancing in the same pace. In this thesis, we propose system models that present machine learning-based solutions to these challenges. Firstly, we propose a hybrid receiver to the quadrature permutation matrix modulation (QPMM) system that combines neural network (NN) outputs with analytical algorithms to provide a trade-off between receiver complexity and detection performance. The proposed NN-based detector is compared with the well known maximum likelihood detector and its low-complexity alternative. Secondly, a MIMO-IM system model that utilizes movabel antennas (MAs) is presented. To the best of our knowledge, this is the study that introduces the idea of integrating MAs to the MIMO-IM systems to improve communication quality. An Euclidean distance-based algorithm is proposed to select the best MA positions. Then, a low-complexity greedy algorithm is presented for better practical implementations. Finally, a genetic algorithm (GA) model is discussed for optimizing the MA positions. Finally, we present a semantic communication system that enhances semantic information transmission by transmitting semantically more information-dense bits through robust index bits of a MIMO-IM system. We compare semantic information loss with respect to errors in different bits of the transmitted sequence. Moreover, since the input data type to the presented system is image, we compared the image reconstruction metrics that are broadly utilized in the literature as well, such as peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM).
Data compression and run-length-limited ISI-mitigation (RLIM) coding for molecular communication
Molecular Communication (MC) enables information transfer at the nanoscale by encoding messages into sequences of released molecules, but its practical deployment is hindered by two fundamental constraints: the high cost of molecule releases and severe inter-symbol interference (ISI) arising from residual molecules in the diffusion channel. This thesis develops coding techniques to tackle each challenge. We first introduce source‐coding schemes that reduce both average code length and molecule usage. Building on an MC-adapted Huffman baseline (MoHuffman), we propose Optimized Molecular Prefix Coding (MoPC) to select a prefix codebook with minimal expected length and fewest number of 1-symbols. To push compression further, we derive Molecular Arithmetic Coding (MoAC) using an existing constrained arithmetic coding construction scheme and show its superior efficiency over substitution arithmetic coding (SAC), our different adaptation of arithmetic source coding to MC. Finally, we design Molecular Arithmetic with Prefix Coding (MoAPC) to ensure unique decodability under finite-precision arithmetic. Using two nucleotide alphabets, we then demonstrate that MoAPC has a better compression ratio than MoPC. Through MC simulations, the effectiveness of the proposed methods is also shown. To mitigate ISI at the channel coding level, we develop an infinite family of Run-Length-Limited ISI-Mitigation (RLIM) codes with a corresponding built-in error correction algorithm. We then demonstrate, via binomial and diffusion channel simulations, that RLIM codes reduce bit-error rate compared to prominent coding schemes. Together, these contributions lay a comprehensive foundation for reliable and efficient diffusion-based Molecular Communication.
Information and communication theoretical modeling and analysis of the gut-brain axis
The gut-brain axis (GBA) represents one of the most sophisticated and crucial communication networks in the human body. However, a quantitative understanding of its signaling mechanisms has remained elusive, as conventional approaches fail to capture the inherently stochastic, nonlinear, and multi-scale nature of intra-body communication. The most promising paradigm to address these challenges is Molecular Communication (MC). Understanding biological signaling from an information and communication theoretical perspective provides insight into the fundamental dynamics of these systems. Therefore, this thesis applies MC principles to the GBA to establish a comprehensive and quantitative communication framework. Building on this objective, the thesis constructs a multi-scale understanding of the GBA through a deliberate progression from a foundational single channel to a complete, bidirectional communication network. The research first develops a channel model for the propagation of a single microbial metabolite (p-cresol), deriving its impulse response and linking gut dysbiosis to a neurological outcome. It then advances the analysis to a complete end-to-end model of molecular-to-neural communication, modeling the short-chain fatty acid (SCFA)-driven vagal nerve pathway and quantifying its information-theoretic performance. Finally, the work culminates in a novel system-level model of the bidirectional communication within GBA that incorporates closed-loop feedback between the hypothalamic-pituitary-adrenal (HPA) axis, immune system, and gut barrier. This framework is used to analyze the system's transition into a pathological state and quantify its corresponding loss of channel capacity. Collectively, the developed communication framework provides the tools to unravel the fundamental signaling mechanisms within the GBA, allowing for a detailed analysis of how signaling disruptions contribute to disease progression. This in-depth, quantitative understanding creates a foundation for a new generation of medical technologies, providing a roadmap for ICT-inspired diagnostic tools and therapeutic strategies to detect and repair communication failures. Furthermore, it lays the essential theoretical groundwork for future Internet of Bio-Nano Things (IoBNT) applications to restore healthy information flow within the body.
Demans ile ilişkili nöro-belirteçlerin EEG ve makine öğrenmesi ile belirlenmesi
Alzheimer Hastalığı (Alzheimer Disease-AD) demansın en yaygın sebebi olup genelde yaşlılarda görülen nörolojik bir hastalıktır. Hafif Kognitif Bozukluk (Mild Cognitive Impairment, MCI), AD'nin bir önceki evresi olup erken belirteç olarak bilinir. Hastalığın ciddiyeti sebebiyle AD'nin erken evrelerdeki gelişiminin tespiti için düşük maliyetli, invaziv olmayan, ve yüksek çözünürlüklü belirteçlerin geliştirilmesi gerekmektedir. Bu tezin amacı, demans ile ilişkili nöro-belirteçlerin EEG ve makine öğrenmesi ile belirlenmesidir. Söz konusu belirteçlerin belirlenmesi amacıyla; AD/MCI gruplarına ait spontan EEG kayıtları, demansda yavaşlama, düşük karmaşıklık ve senkronizasyon örüntülerinin analizleri tez kapsamında çalışılmıştır. Tez kapsamında belirlenen nöro-belirteçler ile demansın erken evrede tespiti, literatüre önemli bir katkı sağlayacağı düşünülmektedir. Bu tez çalışmasında, göz açık-kapalı durumunda 85 AD, 85 MCI, ve 85 Sağlıklı Kontrol (Healthy Condition-HC) içeren Veri seti 1 ve göz kapalı durumda 10 MCI ve 10 HC içeren Veri seti 2 EEG kayıtları analiz edilmiştir. Permütasyon Entropi (PE) karmaşıklık nöro-belirteci olarak önerilirken, tepe güç genlikleri ve frekansları spektral belirteçler olarak seçilmiş ve elektrotlar arasındaki wPLI senkronizasyon özellikleri olarak önerilmiştir. Entopi nöro-belirteçleri için kişi tabanlı 3 sınıflı sınıflandırma yaklaşımı benimsenirken, geri kalan analizler için epok temelli ikili sınıflandırıcılar uygulanmıştır. 25 adet Lazypredict algoritması dikkate alınmış ve en uygun sonuçlar veren işlemlere devam edilmiştir. Bu çalışmada, HC/MCI gruplarını içeren Veri seti 2, 1B EEG segmentleri kullanılarak (EEGNet ve DeepConvNet ile) incelenmiş ve oluşturulan 2B EEG zaman serileri Conv2B ve ResNet mimarileri kullanılarak sınıflandırılmıştır. Ayrıca, 1B EEG kayıtları 2B zaman-frekans (ZF) görüntülerine dönüştürülmüştür. CNN tabanlı ağlar ve Vision Transformer, ZF görüntülerini sınıflandırmak için uygulanmıştır. Sonuç olarak, entropi analizinde gözler açık konumda MCI vs. AD ayrımı central, temporal ve oksipital bölgelerde %100 doğrulukla sonuçlandırılmıştır. MCI'nin bir geçiş evresi olduğu gösterilmiştir. Alfa ve beta tepe genlikleri MCI vs. HC ayrımı için en uygun belirteçlerdir. HC grubunda daha yüksek wPLI değerleri gözlemlenip gruplar %99 doğrulukla sınıflandırılmıştır. Tüm MCI ve HC grupları EEGNet, DeepConvNet ve Vision Transformer (ViT) ile tamamen doğru ayırt edilmiştir. ViT'in dikkat mekanizasması kullanarak CNN tabanlı yaklaşımlardan daha üstün olduğu görülmüştür. ResNet mimarisi de artık bloklar kullanarak CNN'den daha etkili sonuçlar üretmiştir. Mevcut tez çalışması, EEG tabanlı nöro-belirteçler ve makine öğrenme mimarileri kullanarak AD/MCI tespiti için potansiyel sonuçlar sunmaktadır.
Quantum dot-based tandem photodiodes forhigh-efficiency photo-stimulation in biomedicalapplications
Photovoltaic retinal prostheses offer a promising solution for restoring vision in indi- viduals with degenerative retinal diseases through light-mediated neural stimulation. Colloidal quantum dots (QDs), particularly silver bismuth sulfide (AgBiS2 ), provide advantages for bioelectronic interfaces due to their high NIR absorption, chemical stability, and non-toxicity. Tandem photovoltaic structures are known to enhance performance by increasing the open-circuit voltage (Voc). We introduce a novel QD-based tandem photodiode, featuring ultra-thin AgBiS2 layers in both subcells. The device generates capacitive photocurrent, and due to the higher Voc, it in- jects higher ionic charge compared to non-tandem counterparts. The biointerfaces demonstrate low in vitro toxicity and stable photoelectrical performance under vari- ous tests. These results highlight the potential of this QD-based tandem photodiode as a high-resolution, efficient, and safe platform for retinal stimulation, offering a flexible, scalable solution for bioelectronic interfaces.
Derin pekiştirmeli öğrenme kullanarak insansı robotlar için itme kurtarma kontrol sisteminin geliştirilmesi
Bu tezin amacı, iki ayaklı insansı bir robot için bir insanın eylemlerini taklit edebilecek tamamen bağımsız bir itme-kurtarma kontrol sistemi tasarlamak ve robota uygulamaktır. Bu çalışmada, dış kuvvetlerden ve itmelerden etkilenen iki ayaklı insansı robotların itme-kurtarma problemine odaklanılmıştır. İnsansı robotlar denge açısından yapısal olarak kararsız olduklarından dolayı bu problem robotlarda önemli sorun olarak ortaya çıkmaktadır. Robotlarda, itme-kurtarma kontrolörleri ayak bileği, kalça ve adım olmak üzere 3 stratejiden oluşmaktadır. Bu stratejiler, insanların denge bozukluğu durumlarında gösterdikleri biyomekanik tepkilerdir. Bu tezde, insansı robotların ayakta dururken ya da yürürken dengede kalabilmesi ve dış kuvvetlerden kaynaklanabilecek denge bozukluklarının önlenmesi için aktif bir denge kontrolü sunulmuştur. Buna ilişkin yapılan çalışmada hem simülâsyon hem de gerçek dünya testleri yapılmıştır. Çalışmanın simülâsyon testleri Webots ortamında 3 boyutlu modeller ile gerçekleştirilmiştir. Gerçek dünya testleri ise Robotis-OP2 insansı robot üzerinde yapılmıştır. Robot üzerinde bulunan sensörlerden jiroskop, ivmeölçer ve motor verileri kaydedilip, robota harici itme kuvveti uygulanmıştır. Kaydedilen bu veriler ve ayak bileği stratejisi kullanılarak robotun dengesi sağlanmıştır. Bunun için klasik kontrol yöntemi olarak PD kontrolör ve tahmine dayalı olan Model Öngörülü Kontrol (MÖK) yöntemi ile de robotun kontrolü sağlanmıştır. Bununla birlikte robotun tamamen otonom hale getirilebilmesi için Derin Pekiştirmeli Öğrenme (DPÖ) algoritmalarından Derin Q Ağı (DQA) ve Çift Derin Q Ağı (ÇDQA) yöntemleri uygulanmıştır. Bu uygulamalarda robota hem önden hem arkadan kuvvetler uygulanmıştır. Simülâsyon ortamında yapılan test çalışmalarında robotun her iki durumda da gelen itmelere karşı ayakta kaldığı gözlemlenmiştir. Uygulanan dört farklı kontrol yönteminden en iyi sonuçları, ÇDQA algoritması vermiştir. Gerçek ortam testlerinde alınan sonuçlar simülâsyon sonuçlarına paralellik göstermiştir.
İntegratörlü ve kararsız süreçler için maksimum hassasiyet kullanarak I-PD denetleyici tasarımı
Kontrol mühendisliğinde, sistemlerin istenilen şekilde çalışması için çoğu zaman bir denetleyici ile kontrol edilmesine ihtiyaç duyulmaktadır. Bu denetleyicilerden en yaygın olanı PID tip denetleyicilerdir. PID tip denetleyicilerin ayar parametrelerinin kontrol edilecek sistem için uygun seçilmesi gerekmektedir. Günlük yaşamda çok farklı türden sistemler olduğundan her birinin denetiminde kullanılacak olan PID denetleyici ayar parametrelerinin hesaplanmasında farklı stratejiler ve algoritmalar gerekmektedir. İntegratörlü ve kararsız sistemlerin kontrolü diğer sistemlere göre nispeten daha zordur. Bu zorluğu aşabilmek için PID tip denetleyicinin bazen yapısının değiştirilmesi gerekmektedir. Bu tez çalışmasında kararsız ve saf integratörlü sistemlerin I-PD tip denetleyici ile kontrol edilmesi gerçekleştirilmiştir. I-PD denetleyici, kararsız ya da saf integratörlü sistemlerin kontrolünde PID tipi denetleyicilere göre daha iyi bir başarım gösteren bir denetleyici türüdür. Yapılan bu yüksek lisans tez çalışmasında, kararsız ve integratörlü sistemler tanıtılmış ve sistemlere ait düşük dereceli transfer fonksiyonları kullanılarak I-PD denetleyicinin ayar parametreleri uygun Ms değerlerini verecek şekilde türetilen denklemlerden elde edilmiştir. İntegratörlü ve kararsız sistemlerin sırası ile saf integratörlü artı zaman gecikmeli ve birinci derecede kararsız artı zaman gecikmeli transfer fonksiyonu modelleri kullanılarak I-PD denetleyici ayar parametreleri hesaplanmıştır. Elde edilen denklemlerin, integratörlü ve kararsız süreçlerin denetimindeki performansını inceleyebilmek için Matlab Simulink programında kapalı döngü benzetim sonuçları verilmiştir. Literatürde var olan çalışmaların sonuçları ile önerilen yöntemin sonuçları karşılaştırılmıştır. Karşılaştırmalarda kullanılan denetleyici tasarımları ile önerilen I-PD denetleyici tasarımının performanslarını karşılaştırabilmek için bazı performans indeksleri (TV, ITAE, OS, Gm, Pm ve Ms) hesaplanmıştır. İntegratörlü ve kararsız süreçler için uygun modellerin elde edilebilmesi sonucunda, bu çalışmada önerilen I-PD denetleyici tasarımın literatürde bulunan mevcut PID, I-PD ve PI-PD denetleyici tasarımlarına göre daha tatmin edici bir performans sağladığı tespit edilmiştir.