2023-2024

Abstract: The aim of this study is to implement and test methods in event-based image processing systems using asynchronous RISC-V architecture on FPGA. During the tests, power consumption and iteration step speeds will be compared between asynchronous and synchronous architectures. A five-stage pipelined synchronous RISC-V architecture which was written using VHDL language is simulated with Xilinx Vivado. RISC-V GNU Toolchain is used to convert the instructions written in C language or MIPS assembly to coefficient file through a program written in Python language. Using this coefficient file, the instructions are integrated into the architecture and tested. Subsequently, the system is simulated and an attempt is made to be converted into an asynchronous structure. However, due to the synchronous nature of Xilinx Vivado, it has been observed that there are some problems in simulation of the asynchronous design.

Abstract: Point to point telemetry systems are used for remote monitoring of equipment in industry and defense sectors. For example, in the petroleum and gas industry, they are being used for monitoring the real time performance of drilling equipment. These systems can be used for optimization and preventing equipment failures. In point to point telemetry system applications data is transmitted from a transmitter to a receiver wirelessly. While the transmitter usually consists of a wireless transmitter and a sensor or measuring equipment parts, the receiver part receives the data that are being transmitted and for further analysis, recording and monitoring it is processed. An edge device for this telemetry will be designed and for this design software defined radio will be used. Thus the data set can be received correctly and reliably. This device, by processing the data, is going to create a data packet and will transmit the data. This system can be used for remote monitoring and control applications in the health, industry, and defense sectors.

Abstract: In this project, it is aimed to build a spiking neural network (SNN) model on breadboard and perform a logical computation. In the first phase of the project (EE491), SNN models that can be studied on a breadboard in the laboratory environment were investigated and Leaky Integrate and Fire (LIF) model was selected for these purposes. Based on this model, suitable circuit elements that can be used in the market were investigated and then the circuit was realized on the breadboard with these circuit elements. The second phase of the project (EE492) aims to create a network with the LIF model in the first phase and perform XOR calculations with this network. A suitable topology was determined for this goal. According to this topology, 2 signals of 50 Hz and 100 Hz (logic '0' and logic '1') were determined for a neuron cell and the analog circuits designed to build the network in this topology were mentioned. Some of these circuits were installed and tested on a breadboard and these tests are mentioned in the report. Accordingly, the circuit has been modified and a pcb drawing of the final version of the circuit has been made.

Abstract: This thesis includes a Mimo antenna design in addition to the written thesis representing the design of a low-cost and compact rectangular microstrip antenna design running at 28 GHz for 5G communication technology applied in the first semester. In addition, mimo offers research on the applications of structures to increase radiation efficiency and power in antennas and detailed research on the designed resonator, the main purpose of which is to create a basic working principle for 5G technologies used today. Furthermore, the designed Mimo antenna reduces the mutual connection with the resonator used and improves the performance. Resonators, known for their unique structure are used for the reduce mutual Decoupling between closely spaced antenna elements, which increases the gain of isolation and overall antenna performance. This work focuses on designing a compact MIMO antenna with integrated resonator structures to achieve high isolation and improved gain. The study aims at a comprehensive review of the current Rectangular micro ribbon patch antenna, resonator, and mimo antenna designs and their applications in antenna technology. Designed and implemented between mimo  ntennas, this resonator structure is optimized to operate in the 28 GHz frequency band, which is an important band for 5G communication. Simulations using Computer Simulation Technology (CST) Microwave Studio and verify the negative permeability and permeability of the proposed design. The antenna design was created on an FR-4 substrate and tested to validate the simulation outcomes. Adding the resonator greatly enhanced the antenna's effectiveness, lowering mutual coupling and boosting isolation by approximately -10 dB. The findings demonstrate a strong correlation with the simulations and validate the resonator's efficacy in enhancing Mimo antenna performance.

Abstract: Emotion recognition is a process with the ability to understand people's emotional states and expressions. This ability is used in many different fields today. Understanding people's emotional states is essential in areas such as human-machine interaction, marketing and security. Understanding and recognizing emotional states allows to better meet people's needs and expectations and achieve more effective results. In this project, a system that recognizes emotion from human faces is designed using Convolutional Neural Networks (CNN), one of the Deep Learning algorithms. CNN performs moderately to good when trained with a database. However, the lack of accessible, large and balanced databases for the use of deep learning methods for emotion recognition prevents high performance. In recent studies, a success of 65 ± 5% has been achieved as a result of training the CNN model with the FER2013 database, and it is aimed to increase this performance by using various methods in this project. In order to overcome the imbalance in the databases and thus increase the system success, studies were carried out on the FER2013, FER+, CK+ and KDEF databases. The number of data is increased by combining various databases; The images in the databases are subjected to various preprocessing to reduce their differences before training the neural network. These pre-processes help the neural network work better by making facial images more homogeneous and standardized. Data augmentation and synthetic rendering methods are used to reduce the imbalance in the data distribution that remains in existence despite the increasing number of data in the consolidated database and to increase the performance. With the CNN-based method developed by using database merging, image preprocessing and data augmentation, emotion recognition can be achieved with 80.44% success. 

Abstract: Quadrotor UAV usage have been gaining traction int the last few years. With increased areas of application, need for differently configured UAVs arise. In order to meet these new specifications, different components may be used in the UAV. Our project aims to build a simulation environment for quadrotor UAVs in which user can use parameters from previous experiments/simulations for every subsystem of the UAV to simulate the UAV. If done accurately this would allow decreased design and prototyping costs since effects of different component combinations can be observed without the need of building a physical prototype.