Search

In this paper, the high-frequency behavior of various bondwire types is explained in terms of diameter and interconnect number up to millimeter-wave frequencies. The inductive effect of several bondwires which are used in high-frequency circuits, was calculated by considering height, angle, and length. After these calculations, 3D models and electromagnetic simulations of these interconnections were realized with a computer-aided design tool. Then, bondwires which have different characteristics, are implemented on alumina transmission lines. Furthermore, S-parameters were measured with a 2-port network analyzer using a probe station. Measured insertion loss and reflection values were compared with 3D electromagnetic simulation results, in order to obtain conceivable interconnection.

Three features are crucial for sequential forecasting and generation models: tractability, expressiveness, and theoretical backing. While neural autoregressive models are relatively tractable and offer powerful predictive and generative capabilities, they often have complex optimization landscapes, and their theoretical properties are not well understood. To address these issues, we present convex formulations of autoregressive models with one hidden layer. Specifically, we prove an exact equivalence between these models and constrained, regularized logistic regression by using semi-infinite duality to embed the data matrix onto a higher dimensional space and introducing inequality constraints. To make this formulation tractable, we approximate the constraints using a hinge loss or drop them altogether. Furthermore, we demonstrate faster training and competitive performance of these implementations compared to their neural network counterparts on a variety of data sets. Consequently, we introduce techniques to derive tractable, expressive, and theoretically-interpretable models that are nearly equivalent to neural autoregressive models.

In this paper, an efficient spectrum forecasting model is developed to estimate the required spectrum and calculate the spectrum gap in future. This developed model is essentially based on five main metrics and one constant. The five main metrics are the currently available spectrum, sites number growth, data traffic growth, average network utilization, and spectrum efficiency growth. The constant metric is considered to give a space for our model to be used in another country or when a new technology is coming. This developed model is then used to forecast the required spectrum and spectrum gap for Malaysia in 2020. The estimation is performed based on the input market data of four main mobile telecommunication operators in Malaysia: Maxis, Celcom, Digi, and U-Mobile. The input data for this model are collected from various sources, such as the Malaysian Communications and Multimedia Commission, OpenSignal, Analysys Mason, GSMA, and HUAWEI. The results indicate that by 2020, Malaysia will require around 307 MHz of additional spectrum to fulfill the enormous increase of mobile data demands. Addressing this increment can be achieved by launching additional spectrum bands, enhancing spectrum efficiency, off-loading mobile data to unlicensed bands or deploying more site numbers.

Achieving ultra low-latency and ultra-high reliable communications at millimeter-waves (mmWaves) is challenging due to the sensitivity of this frequency band to blockage. Different space diversity schemes based on either equal gain or beam-tracking algorithms, are proposed to cope with the significant propagation losses introduced by the blockage at mmWaves. However, equal gain algorithms suffer from the low throughput as they allocate considerable gain to the non-optimal links. On the other hand, beam-tracking algorithms suffer from the link outage as the beam-forming is redone after dropping a large amount of data. In this study we propose a new algorithm based on the prior collection of data to eliminate power consuming beam-tracking techniques while minimum diversity level ensures high received power, ultra-reliability, and low latency., METU, Prof. Dr. Mustafa Parlar Foundation Research Encouragement Award

The dramatic growth of machine-to-machine (M2M) communication in cellular networks brings the challenge of satisfying the quality of service (QoS) requirements of a large number of M2M devices with limited radio resources. In this paper, we propose an optimization framework for the semi-persistent scheduling of M2M transmissions based on the exploitation of their periodicity with the goal of reducing the overhead of the signaling required for connection initiation and scheduling. The goal of the optimization problem is to minimize the number of frequency bands used by the M2M devices to allow fair resource allocation of newly joining M2M and human-to-human communications. The constraints of the problem are delay and periodicity requirements of the M2M devices. We first prove that the optimization problem is NP-hard and then propose a polynomial-time heuristic algorithm employing a fixed priority assignment according to the QoS characteristics of the devices. We show that this heuristic algorithm provides an asymptotic approximation ratio of 2.33 to the optimal solution for the case where the delay tolerances of the devices are equal to their periods. Through extensive simulations, we demonstrate that the proposed algorithm performs better than the existing algorithms in terms of frequency band usage and schedulability., Turkish Academy of Sciences (TÜBA) through the Young Scientist Award Program (GEBIP); Middle East Technical University (METU)-Prof. Dr. Mustafa Parlar Foundation through the Research Encouragement Award

Vehicular visible light communication (V2LC) has recently gained popularity as a complementary technology to radio frequency (RF) based vehicular communication schemes due to its low-cost, secure and RF-interference free nature. In this paper, we propose outdoor vehicular visible light communication (V2LC) frequency domain channel sounding based channel model characterization under night, sunset and sun conditions with the usage of vector network analyzer (VNA) and commercial off-the-shelf (COTS) automotive light emitting diode (LED) light. We further bring forward a new practical system bandwidth criterion named as effective usable bandwidth (EUB) for an end-to-end V2LC system with respect to the real world measurements. We demonstrate outdoor static V2LC channel measurement results, taking into account vehicle light emitting diode (LED) response, road reflections from nearby vehicles and various daylight conditions with respect to varying inter-vehicular distances. Measurement results indicate that, sunlight decreases system effective usable bandwidth due to the limited dynamic range of avalanche photodiode (APD), nearby vehicles cause constructive interference whereas road reflections change time dispersion characteristics of the V2LC channel., NA

Heterogeneous Network (HetNet) is an architecture that contributes significantly to offering high performance and better throughput in future networks. However, ensuring a seamless connection during users’ mobility is a key challenge in HetNet. Meanwhile, securing stable wireless access with good Quality of Service (QoS) for mobile users has become more challenging due to the rapid growth of mobile data traffic. Therefore, ensuring seamless communication requires efficient mobility management techniques to be used in HetNet, especially with the inclusion of 5G networks. To the best of our knowledge, conventional techniques have been used in the previous technologies may not be capable to meet the requirements of future HetNets. Thus, this paper provides a brief overview of mobility management concepts and techniques used in 5G HetNets. It also discusses various handover solutions in order to highlight them for more investigations need in the next mobile communication systems. Meanwhile, this paper presents parts of the key challenges related o mobility management issues in 5G HetNets which need to be addressed in future mobile networks.

Potential of intrusion during smart meter data collection is an important problem for household privacy in next-generation smart homes. There are various privacy protection methods such as hiding the real usage with renewable batteries (RBs) and encrypted communication. There are privacy threats due to the detection of the energy transfer through RBs while encryption-based methods result in high computational complexity. In this study, a novel privacy improvement mechanism is proposed based on the routing of energy through multiple smart meters among multiple houses in a neighborhood. The optimization of privacy based on uniform randomization is theoretically modeled and some heuristic solutions are presented. In this method, energy transfer is randomly distributed among multiple smart meters, real usage of a house is buffered with the RB and smart meter indices are randomly selected. In numerical simulations, privacy improvement is shown by utilizing the random selection of smart meter indices for randomizing the energy routing and the buffers of RBs. Furthermore, energy privacy (EP) metric validating the simulation results is defined. Energy routing with multiple smart meters provides a practical and low complexity privacy protection mechanism for next-generation smart grid architectures.

We investigate variable-length data regression in an online setting and introduce an energy-efficient regression structure build on long short-term memory (LSTM) networks. For this structure, we also introduce highly effective online training algorithms. We first provide a generic LSTM-based regression structure for variable-length input sequences. To reduce the complexity of this structure, we then replace the regular multiplication operations with an energy-efficient operator, i.e., the ef-operator. To further reduce the complexity, we apply factorizations to the weight matrices in the LSTM network so that the total number of parameters to be trained is significantly reduced. We then introduce online training algorithms based on the stochastic gradient descent (SGD) and exponentiated gradient (EG) algorithms to learn the parameters of the introduced network. Thus, we obtain highly efficient and effective online learning algorithms based on the LSTM network. Thanks to our generic approach, we also provide and simulate an energy-efficient gated recurrent unit (GRU) network in our experiments. Through an extensive set of experiments, we illustrate significant performance gains and complexity reductions achieved by the introduced algorithms with respect to the conventional methods. This work was supported in part by the Tubitak Project under Grant 117E153.

Estimation of the time-varying optical wireless channel response is crucial in order to decode received signals coherently. In this work, we investigate symbol-error-rate and mean absolute error performance of different interpolation techniques including linear, nearest, spline, and piece-wise cubic Hermite interpolating polynomial (pchip), which are used in pilot-aided channel estimation process for visible light communication. The performance of interpolators is evaluated in realistic time-varying channel model, generated on Zemax software and compared with each other under the consideration of different modulation orders, different pilot symbol periods, and different user equipment (UE) speeds through Monte Carlo simulations. The results reveal that spline and pchip techniques are more robust to low pilot symbol transmission rate and fast time-varying channel conditions as a consequence of high UE speeds. However, low complex linear interpolation technique can be chosen for highly rated pilot signal transmission cases or when optical channel varies slowly over the time., Scientific and Technological Research Council of Turkey (TÜBİTAK); Turkish Academy of Sciences (TÜBA); Young Scientist Award Program (GEBIP); Dr. Mustata Parlar Foundation Research Encouragement Award

In the above article [1], the biographies and photos of the authors were missing.

Electrocardiogram (ECG) is a biomedical signal which represents the electrical activity of the human heart. Various cardiac diseases have been detected using the outputs of ECG devices. Recently, advances in signal processing techniques bring out a new horizon for processing the ECG signals. In this scope, a novel application based on the spectrogram, which is a graphical representation of time-frequency information of the signal, and the convolutional neural network (CNN) is proposed so as to distinguish ECG signals. To this aim, a publicly available data set in Physionet was utilized. Firstly, the spectrograms of each signal were obtained. Then, these colorful spectrogram images were applied as the input to CNNs that are AlexNet, VGG-16, and ResNet-18. The transfer learning and fine-tuning approach were used for training and validation of the models. As a result, the most efficient results were provided by AlexNet with an accuracy of 83.82%. The experimental results of this study show that the proposed model ensures promising results for the ECG signal classification.

Brain hemorrhage refers to hemorrhage within the brain tissue or between the surrounding bone. Therefore, head hemorrhage can lead to many dangerous consequences, especially brain hemorrhage. Early and correct intervention by experts in such cases is important for the patient’s life. In this study, computed tomography images of brain hemorrhage are classified by AlexNet which is one of the convolutional neural network models used recently in the biomedical field. In this scope, the data set is restructured with the autoencoder network model and heat maps of each image in the data set are extracted to improve the classification success. The number of images in the data set is then increased by approximately 10 times using the data augmentation technique. The classification process is performed using support vector machines. As a result, the best success rate in the classification was 98.57%. In conclusion, the proposed approach contributed to the classification of cerebral hemorrhage images.

We consider non-convex training of shallow neural networks and introduce a convex relaxation approach with theoretical guarantees. For the single neuron case, we prove that the relaxation preserves the location of the global minimum under a planted model assumption. Therefore, a globally optimal solution can be efficiently found via a gradient method. We show that gradient descent applied on the relaxation always outperforms gradient descent on the original non-convex loss with no additional computational cost. We then characterize this relaxation as a regularizer and further introduce extensions to multineuron single hidden layer networks.

In this study, the diagnosis of some diseases in the retina of the eye by using deep learning architectures is intended to be diagnosed. Optical Coherence Tomography device from Choroidal Neovascularization, Diabetic Macular Edema, Drusen and healthy eye retinal images were examined. LeNet, AlexNet and Vgg16 architectures of deep learning were used. In each architecture, the hyper parameters were changed to diagnose these diseases. Results of the implementation showed that exhibit successful results in Vgg16 and AlexNet architecture. Dropout layer structure in AlexNet has been shown to reduce the loss by minimizing loss.

Cellular Vehicle to Everything (C - V2X) is the technology, which incorporates V2X communication and vehicular user (VUE) to network communication, and enables the system to select the communication interface, which can provide safe, reliable and energy efficient communication for VUEs. We study energy efficient VUE association problem, which aims to balance the power consumption of VUEs and cellular infrastructure with the help of the selection of the appropriate communication interface. The problem aims to minimize power consumption of the network for Uplink (UL) and Downlink (DL) decoupled UE association scheme, considers switching on/off the Small Base Stations (SBS)s and supports C-V2X link formation either over LTE-Uu conventional cellular radio interface and 5G or over Sidelink (SL), PC5 radio interface. We additionally incorporate realistic power consumption model for BSs, VUEs in UL and Fron-thaul (FH) links, centralized controller and Transmit (TX) and Receive (RX) operations of VUEs. Simulation results demonstrate that enabling the selection of communication interface provides decrease in power consumption of the network instead of forcing VUEs to form communication links over LTE-Uu conventional cellular radio interface for varying number of distance thresholds and VUEs.

Vehicular visible light communication (V2LC) has recently gained popularity as a complementary technology to radio frequency (RF) based vehicular communication schemes due to light emitting diode (LED)s’ readily availability on vehicles with its secure and RF-interference free nature. However, vehicular visible light communication (V2LC) system performance mainly depends on LED characteristics. Investigating various LED bulbs for their frequency response and optical OFDM (O-OFDM) based modulation performances, it has been observed that LED and DC-bias voltage selection is key for the V2LC system modulation performance. Experimental results indicate that, on contrary to simulation results in the literature, asymmetrically clipped optical OFDM (ACO-OFDM) is observed to perform better than unipolar OFDM (U-OFDM) as it inherits lower peak-to-average power ratio (PAPR) with lower clipping noise which is crucial for LEDs under consideration with limited linear working region.

In recent years, deep learning models have been widely used on remote sensing images. Deep learning is held on remote sensing images as well as in every area; is to be able to perform better performance classification than existing approaches and to perceive the feature inferences on its own. In remote sensing, more studies are made especially on hyperspectral images. The most important reason for this is that it can carry a large number of data features. The large number of data features means that there are a large number of attributes for that image. The most important disadvantage of hyperspectral images is; due to the influence of the environment of the device which is shooting the image, various noises may occur. There may be a variety of information loss on this image. Various algorithms techniques have been developed to prevent these losses, while hyperspectral images have been better classified by deep learning models. Recent advances in deep learning models in technological firms and the creation and development of their own deep learning model are evidence of how intense this interest is in this area. Our aim is to examine recent developments in deep learning activities on remote sensing images in this article; to compare the performances obtained by deep learning model and to give brief information about the methods used in this area.

Breast cancer is one of the most common cancer types diagnosed in the ladies worldwide. Statistics indicate that breast cancer rate is about 12 % in all cancer cases in the world. Also, approximately 25% of women have breast cancer. Therefore, rapid and accurate analysis of medical images obtained with breast cancer is extremely important for diagnosis. Many methods have been used to classify breast cancer. One of the most important methods among these methods is deep learning-based methods. The most important reason for choosing the deep learning model for breast cancer diagnosis is that it can give faster and more accurate results than the existing methods. Because early diagnosis is always important. In this study, a total of 700 images, including benign and malign variants of breast cancer images, were analyzed using the convolutional neural network method. The data set used in this study is publicly accessible. As a result, the histopathologic images of the breast cancer features were extracted using the AlexNet method which is one of the deep learning approaches. The classification process was performed with Support Vector Machines and accuracy of 93.4% was achieved.

Social networks are an increasingly popular medias today. On these medias, users are able to interact with each other or share their views on an up-to-date topic, enabling them to have information about their own characters and worldviews. In this study, using the twitter data, the political tendency or parties supported by the users were predicted using data mining techniques.

Date of Conference: 2-5 May 2018 In this paper, we introduce a Long Short Term Memory (LSTM) networks based anomaly detection algorithm, which works in an unsupervised framework. We first introduce LSTM based structure for variable length data sequences to obtain fixed length sequences. Then, we propose One Class Support Vector Machines (OC-SVM) algorithm based scoring function for anomaly detection. For training, we propose a gradient based algorithm to find the optimal parameters for both LSTM architecture and the OC-SVM formulation. Since we modify the original OC-SVM formulation, we also provide the convergence results of the modified formulation to the original one. Thus, the algorithm that we proposed is able to process data with variable length sequences. Also, the algorithm provides high performance for time series data. In our experiments, we illustrate significant performance improvements with respect to the conventional methods.

Technology brings autonomous vehicles into a reality where vehicles cruise themselves without human input. Vehicular platoon, on the other hand, is a group of autonomous vehicles that are organized into close proximity through wireless communication. In an autonomous platoon, vehicles cooperatively send data to each other to adjust their speed and distance to the leader, the first vehicle in the platoon. However, this cooperative data exchange can lead to security risks. A misbehaving platoon member could alter the data packets which may cause platoon instability. Therefore, identifying the modified packets has become an important requirement. In this paper, we investigate data-driven anomaly detection mechanisms for the autonomous platoon. We propose a novel statistical learning based technique to detect the modified packets and misbehaving vehicles. We demonstrate that the distance change to the leader would be sufficient to detect anomalies and misbehavior.

In this study, a deep learning method was used to detect sleep states of the drivers in the driving environment. A convolutional neural network (CNN) model has been proposed to determine whether the eyes of certain constant face images of drivers are closed. The proposed model has a wide potential application area such as human­computer interface design, facial expression recognition, driver fatigue-sleepiness determination. This method, which was developed on driver sleepiness data, has been applied on 4,846 real eye images in the Closed Eyes In The Wild (CEW) database. Commonly used CNN models are used on the same data to compare performances of the prepared model. According to the classification results obtained, 96.5% and 92.99% of the designed model achieved success and it is seen that this structure can be used in this problem area.

Date of Conference: 2-5 May 2018 In this paper, we study online nonlinear data regression and propose a highly efficient long short term memory (LSTM) network based architecture. Here, we also introduce on-line training algorithms to learn the parameters of the introduced architecture. We first propose an LSTM based architecture for data regression. To diminish the complexity of this architecture, we use an energy efficient operator (ef-operator) instead of the multiplication operation. We then factorize the matrices of the LSTM network to reduce the total number of parameters to be learned. In order to train the parameters of this structure, we introduce online learning methods based on the exponentiated gradient (EG) and stochastic gradient descent (SGD) algorithms. Experimental results demonstrate considerable performance and efficiency improvements provided by the introduced architecture.