HAYATİ MAMUR, Yusuf ÇOBAN
HAYATİ MAMUR, Yusuf ÇOBAN
Thermoelectric generators (TEGs) are used in small power applications to generate electrical energy fromwaste heats. Maximum power is obtained when the connected load to the ends of TEGs matches their internal resistance.However, impedance matching cannot always be ensured. Therefore, TEGs operate at lower efficiency. For this reason,maximum power point tracking (MPPT) algorithms are utilized. In this study, both TEGs and a boost converter withMPPT were modeled together. Detailed modeling, simulation, and verification of TEGs depending on the Seebeckcoefficient, the hot/cold side temperatures, and the number of modules in MATLAB/Simulink were carried out. Inaddition, a boost converter having a perturb and observation (P&O) MPPT algorithm was added to the TEG modeling.After the TEG output equations were determined, the TEG modeling was performed based on manufacturer data sheets.Thanks to the TEG model and the boost converter with P&O MPPT, the maximum power was tracked with a value of98.64% and the power derived from the TEG was nearly unaffected by the load changes. The power outputs obtainedfrom the system with and without MPPT were compared to emphasize the importance of MPPT. These simulationvalues were verified by using an experimental setup. Ultimately, the proposed modeling provides a system of TEGs anda boost converter having P&O MPPT.

Orhan KESEMEN, Özge TEZEL, Eda ÖZKUL, Buğra Kaan TİRYAKİ
Orhan KESEMEN, Özge TEZEL, Eda ÖZKUL, Buğra Kaan TİRYAKİ
Cluster analysis is widely used in data analysis. Statistical data analysis is generally performed on thelinear data. If the data has directional structure, classical statistical methods cannot be applied directly to it. Thisstudy aims to improve a new directional clustering algorithm which is based on trigonometric approximation. Thetrigonometric approximation is used for both descriptive statistics and clustering of directional data. In this paper, thefuzzy clustering algorithms (FCD and FCM4DD) improved for directional data and the proposed method are carried outon some numerical and real data examples, and the simulation results are presented. Consequently, these results indicatethat the fuzzy cmeans directional clustering algorithm gives the better results from the points of the mean square errorand the standard deviation for cluster centers.

Owais BHAT, Dawood A. KHAN
Owais BHAT, Dawood A. KHAN
Diabetic patients are quite hesitant in engaging in normal physiological activities due to difficulties associatedwith diabetes management. Over the last few decades, there have been advancements in the computational power ofembedded systems and glucose sensing technologies. These advancements have attracted the attention of researchersaround the globe developing automatic insulin delivery systems. In this paper, a method of closedloop control of diabetesbased on neural networks is proposed. These neural networks are used for making predictions based on the clinical data ofa patient. A neural network feedback controller is also designed to provide a glycemic response by regulating the insulininfusion rate. An activity recognition model based on convolutional neural networks is also proposed for predicting thepatient’s current physical activity. Predictions from this model are transformed into a sixlevel code and are fed as inputto the neural network glucose prediction model. Experimental results of the proposed system show good performance inkeeping blood glucose levels in the nondiabetic range.

Murat Cihan SORKUN, Özlem DURMAZ İNCEL, Christophe PAOLI
Murat Cihan SORKUN, Özlem DURMAZ İNCEL, Christophe PAOLI
Energy management is an emerging problem nowadays and utilization of renewable energy sources is anefficient solution. Solar radiation is an important source for electricity generation. For effective utilization, it is importantto know precisely the amount from different sources and at different horizons: minutes, hours, and days. Depending onthe horizon, two main classes of methods can be used to forecast the solar radiation: statistical time series forecastingmethods for short to midterm horizons and numerical weather prediction methods for medium to longterm horizons.Although statistical time series forecasting methods are utilized in the literature, there are a limited number of studiesthat utilize deep artificial neural networks. In this study, we focus on statistical time series forecasting methods forshortterm horizons (1 h). The aim of this study is to discover the effect of using multivariate data on solar radiationforecasting using a deep learning approach. In this context, we propose a multivariate forecast model that uses acombination of different meteorological variables, such as temperature, humidity, and nebulosity. In the proposed model,recurrent neural network (RNN) variation, namely a long shortterm memory (LSTM) unit is used. With an experimentalapproach, the effect of each meteorological variable is investigated. By hyperparameter tuning, optimal parameters arefound in order to construct the best models that fit the global solar radiation data. We compared the results with thoseof previous studies and we found that the multivariate approach performed better than the previous univariate modelsdid. In further experiments, the effect of combining the most effective parameters was investigated and, as a result, weobserved that temperature and nebulosity are the most effective parameters for predicting future solar radiance.

Genome structural variation, broadly defined as alterations longer than 50 bp, are important sources forgenetic variation among humans, including those that cause complex diseases such as autism, developmental delay, andschizophrenia. Although there has been considerable progress in characterizing structural variation since the beginningsof the 1000 Genomes Project, one form of structural variation called segmental duplications (SDs) remained largelyunderstudied in large cohorts. This is mostly because SDs cannot be accurately discovered using the alignment filesgenerated with standard read mapping tools. Instead, they can only be found when multiple map locations are considered.There is still a single algorithm available for SD discovery, which includes various tools and scripts that are not portableand are difficult to use. Additionally, this algorithm relies on a priori information for regions where no structuralvariations are discovered in large number of genomes. Therefore, there is a need for fully automated, portable, anduserfriendly tools to make SD characterization a part of genome analyses. Here we introduce such an algorithm andefficient implementation, called mrCaNaVaR, that aims to fill this gap in genome analysis toolbox.

: The quality of randomness in numbers generated by true random number generators (TRNGs) dependson the source of entropy. However, in TRNGs, sources of entropy are affected by environmental changes and thiscreates a correlation between the generated bit sequences. Postprocessing is required to remove the problem created bythis correlation in TRNGs. In this study, an Sboxbased postprocessing structure is proposed as an alternative to thepostprocessing structures seen in the published literature. A ring oscillator (RO)based TRNG is used to demonstrate theuse of an Sbox for postprocessing and the removal of correlations between number sequences. The statistical propertiesof the numbers generated through postprocessing are obtained according to the entropy, autocorrelation, statisticalcomplexity measure, and the NIST 800.22 test suite. According to the results, the postprocessing successfully removedthe correlation. Moreover, the data rate of the bit sequence generated by the proposed postprocessing is reduced to 2/3of its original value at the output.

H. Seçkin DEMİR, Erdem AKAGÜNDÜZ
H. Seçkin DEMİR, Erdem AKAGÜNDÜZ
In this paper, we introduce a machine learning approach to the problem of infrared small target detectionfilter design. For this purpose, similar to a convolutional layer of a neural network, the normalizedcrosscorrelational(NCC) layer, which we utilize for designing a target detection/recognition filter bank, is proposed. By employing theNCC layer in a neural network structure, we introduce a framework, in which supervised training is used to calculatethe optimal filter shape and the optimum number of filters required for a specific target detection/recognition task oninfrared images. We also propose the meanabsolutedeviation NCC (MADNCC) layer, an efficient implementation ofthe proposed NCC layer, designed especially for FPGA systems, in which square root operations are avoided for realtimecomputation. As a case study we work on dimtarget detection on midwave infrared imagery and obtain the filters thatcan discriminate a dim target from various types of background clutter, specific to our operational concept.

Hojatollah SOLEYMANI, Sobhan ROSHANI
Hojatollah SOLEYMANI, Sobhan ROSHANI
In this paper a Wilkinson power divider (WPD) is presented with ultrawideband operation and harmonicsuppression. This WPD is designed using coupling lines and meandered open stubs at main branches. The centerfrequency of the presented WPD is 4.25 GHz, which is fabricated and measured on RT/Duroid substrate with dielectricconstant of 2.2. The proposed WPD provides good filtering band with high attenuation level. The 15 dB return lossoperational bandwidth (BW) of the WPD is obtained between 3.2 GHz and 5.3 GHz, which shows 50% operationalbandwidth.

Rasime UYGUROĞLU, Abdullah Y. ÖZTOPRAK
Rasime UYGUROĞLU, Abdullah Y. ÖZTOPRAK
Multifocal lenses have been widely used as multiple beam forming networks for linear and planar arrays, butare not suitable for large convex conformal arrays as they are not physically realizable for wide angle beams. In thisstudy, a new lens design especially suitable for convex conformal arrays is introduced. The new lens has no perfect focalpoints, but there are a certain number of correct phases in each of the beam directions. The directional patterns of theradiating elements are also taken into account to improve radiation patterns for desired directions. It has been shownthat the new lens can be designed for circular arc arrays with satisfactory radiation performance for parameters wherethe design of realizable multifocal lenses is not possible.

Guru PRASAD, Kumara SHAMA
Guru PRASAD, Kumara SHAMA
: An area efficient output capacitorfree low dropout [LDO] voltage regulator with an improved figure of meritis presented in this paper. The proposed LDO regulator consists of a novel, dynamically biased error amplifier thatreduces overshoot and undershoot voltage spikes arising from abrupt load changes. Source bulk modulation is employedto enhance the current driving capability of the pass transistor. An adaptive biasing scheme is also used along withdynamic biasing to improve the current efficiency of the system. The onchip capacitor required for proper working ofthe LDO regulator is only 35 pF. The proposed LDO regulator is designed and simulated in 180 nm standard CMOStechnology. The LDO regulator exhibits a line regulation of 1.67 mV/V and a load regulation of 100 µV /mA. Whenload changes from 0 mA to 100 mA in 1 µs, an undershoot of 148 mV and an overshoot of 172 mV are observed. Themeasured power supply rejection ratio is 25 dB at 100 kHz. The working of the proposed LDO regulator has been testedunder all process corners and MonteCarlo statistical analysis reveals that it is robust against process variations andlocal mismatch.
