Browsing by Author "Koçak, Taşkın"

Now showing 1 - 5 of 5

Design and implementation of high throughput bidirectional fano decoding
(Bahçeşehir Üniversitesi Fen Bilimleri Enstitüsü, 2012-09) Kakacak, Ahmet; Koçak, Taşkın
Bu çalışma çift yönlü Fano kod çözümünün donanımsal gerçeklemesini kapsamaktadır. Fano gibi ardışıl kod çözümü yöntemleri Viterbi’ye kıyasla daha düşük donanım karmaşıklığı sağlamaktadır. Bu yöntemler farklı işaret/gürültü oranlarına uyarlanabilir ve değişken hesaplama süresine sahiptir, bu nedenle yüksek işaret/gürültü oranlarında yapılan iş oldukça yüksektir. Fakat düşük işaret/gürültü oranlarında yapılan iş bir hayli düşer. Yapılan işi artırmak ve hesaplama süresindeki değişkenliği azaltmak için çift yönlü kod çözümü yöntemi uygulanabilir. İki kod çözücü ters yönlerde paralel olarak çalışır ve kod dizisinin herhangi bir yerinde birleşir. Bu birleşme iki kod çözücünün durumlarının aynı noktada eşit olmasına bağlıdır. Eğer kontrol edilen durum sayısı birden fazlaysa komşu durumlar da eşit olmalıdır. Yazılım ortamında yapılan çalışmalar çift yönlü Fano kod çözümünün, işlem yükünü azalttığını göstermektedir. Fakat söz konusu hipotez donanımsal olarak da kanıtlanmalıdır. Yazılımda sadece iterasyon sayısı gözlemlenebilir, ayrıca geleneksel işlemciler yapılan işin yüksek olmasına izin vermez. Fakat donanımda yapılan iş çok daha yüksek olabilir. Bu çalışma bu konudaki boşluğu dolduracaktır. Haberleşme sektöründe FPGA’ler yaygınca kullanıldığı için FPGA üzerindeki gerçekleme sonuçları oldukça önemli. Burada tek yönlü ve çift yönlü Fano kod çözümünün donanımsal verilerini karşılaştırma olanağı bulunacaktır. Ayrıca saat frekansı ve tüketilen FPGA kaynakları açısından Fano ve Viterbi kod çözümleri de karşılaştırılabilecektir.
GPU-Based FFT Computation for Multi-Gigabit WirelessHD Baseband Processing
(2010) Hinitt, Nicholas; Koçak, Taşkın
High-speed bidirectional fano algorithm implementation
(Bahçeşehir Üniversitesi Fen Bilimleri Enstitüsü, 2014-01) Ateş, Özgür; Koçak, Taşkın
This thesis is about the implementation of Bidirectional Fano algorithm (BFA) and Unidirectional Fano algorithm (UFA or conventional Fano algorithm) in C and CUDA. BFA is derived from the Fano algorithm used in high speed convolutional code decoding. It is a simultaneous forward and backward codeword search decoder. The high throughput demanded by the latest wireless digital interfaces like WirelessHD may benefit from a parallel computing specific baseband processing unit that runs BFA at a high speed on CUDA. In this thesis, BFA reached a throughput of 4.4Gbps in CUDA with a single thread per codeword. Its iteration is characterised by one thread decoding first forward then backward. Another BFA decoding at 3.1Gbps was performed using dual threads per codeword. One thread is a forward decoder while the other is a backward decoder. Finally, on the same board, GTX650, UFA was found to have 5.0Gbps of throughput. It was concluded that additional memory transactions and check conditions were the source of throughput loss of BFA in comparison to UFA. However, BER, NoI and TpI analysis gave improved results for BFA, meaning that it decoded more efficiently than UFA. In order to obtain such throughputs, several optimization measures were taken such as the use of look-up tables instead of metric, output bit, state calculations. Another technique was to use an indexed circular queue to hold the previous eight steps in memory instead of using a conventional array. This thesis proposes BFA implementation in CUDA application as a complement to the research done by Xu et al. in 2009, in which simulation was conducted in MATLAB. Later, it has been implemented in FPGA at 100Mbps, and here we are aiming a throughput in several Gbps. The research done by Xu and Koçak (2010) proposed the use of LUT for checking merge conditions. LUT were used in this thesis, too, but this time to pre-calculate next metric, state, output bit calculations, and this contribution was applied to both UFA and BFA.
Impact of quality distortions on CNN image classifiers and restoration using deep convolutional auto-encoder
(Bahçeşehir Üniversitesi Fen Bilimleri Enstitüsü, 2018) Çiloğlu, Çağkan; Koçak, Taşkın
Deep neural networks have obtained significant performance on recognizing objects even in real-time video stream. As hardware requirements of this task decrease in cost and parts getting smaller, this technology can be used in mobile devices Most of the time, deep neural networks are trained and tested on quality picture datasets. But frames provided by camera may be distorted because of camera defects and/or weather conditions such as rain and snow. This paper provides an evaluation of four state-of-the-art deep neural network models for picture classification under quality distortions. Three types of quality distortions are considered: blur, noise and contrast. It is shown that the existing networks are susceptible to these quality distortions and architecture of the network dramatically affects the results. Using deep convolutional auto-encoder to restore picture quality is suggested and better scores have been archived utilizing it. Results enable future work in developing machine vision systems on that are more invariant to quality distortions.
Smart transportation systems
(Bahçeşehir Üniversitesi Fen Bilimleri Enstitüsü, 2012-06) Kılıç, Necati; Koçak, Taşkın
Smart or intelligent transportation systems (ITS) have been around for some time. However, they are re-emerging again with the recent advances in wireless communications. This is made possible not only by the technological breakthroughs but also the widespread usage and affordability of the mobile communications based services. Contrary to the longevity of the experiences in ITS technologies, the road traffic in metropolitan areas is far cry from acceptable levels. In order to have a deeper understanding of the aforementioned problem, the state-of-the-art in smart transportation systems has been studied. Many solutions in the literature that are offered by the range of top telecom companies, municipalities, corporations, universities and EU funded projects are reviewed. Communication technologies used in these solutions are discussed. A wide range of simulation tools in the literature are investigated which helped us choosing the right set of simulation tools. The thesis covers the traffic theory background in detail that is investigated in this project. The meaning and the importance of several parameters which are common in all traffic applications such as density, flow, speed, gap, etc. are introduced. The relationship between flow, density and velocity is given and its significance is stated. The thesis also explored the mathematical traffic flow models that are widely mentioned in the traffic engineering discipline and are listed in two categories, namely Macroscopic and Microscopic traffic flow models. A segmented freeway system is simulated with several macroscopic traffic flow models introduced by Lighthill-Whitham-Richards, Payne, Papageorgiou and Daganzo. Some of the simulation results corresponding to Payne and Daganzo’s Cell Transmission Model (CTM) which both utilize the hydrodynamic theory are given in the thesis. The fundamental diagram which is a significant concept that is used in both macroscopic and microscopic levels of traffic flow is studied, as well. The thesis primarily focused on two steps to address the undying issue of recent decades which has lately become even more irritating. The first step described our efforts to construct and introduce a short-term prediction system. A simple yet effective simulation tool called CTMsim is presented. Two different case studies modeling a portion of TEM freeway ranging from Kavacık to Maslak have been carried out to demonstrate the abilities of the tool. Firstly, a traffic accident scenario is depicted and its effects both visually and statistically are displayed in the area of influence. In addition to it, second scenario interpreted and portrayed the prediction of the traffic states fifteen minute ahead of time which resulted in a slight difference from the real-time calculation. Briefly, the tool has been able to capture the traffic dynamics and present promising results. The second step is comprised of presenting a real-time traffic monitoring system. On this sense, a simulator named BAU Traffic Simulator that is integrated with a smartphone application called TTraffic is developed from the scratch. The tool achieved to dynamically interpret the outputs of the client (android app) and represent the system in real-time. The outputs of the simulator can be fetched by the clients in order to get real-time traffic information, and hence calculate a dynamic routing plan to get to the destination within the possible minimum time. All analyses carried out during the development phase are presented in the thesis. Lastly three traffic scenarios simulating Regular Traffic, Free Flow Traffic, and Traffic Accident conditions are presented and analyzed thoroughly.