-Summary: Beamforming in mmWave compensate for the higher path loss, but it makes the topology control (e.g. interference control, base station, and beam coverage) a great challenge. A framework, termed Beamforming Oriented tOpology coNtrol (BOON), is proposed to jointly reduce downlink transmit power in multi-base stations mmWave networks, subject to min. SINR at each user.
- Key Findings: BOON smartly groups nearby user equipment into clusters, constructs sets from user equipment clusters, and associates user equipment to base stations and beams. In MATLAB simulations, it is found that BOON uses only 10%, 32%, and 25% transmit power on average, compared to the three state-of-the-art schemes in the literature, to achieve the same sum rate. This work has been published in IEEE Transaction on Mobile Computing (impact factor: 4.474).
- Summary: User association, beamforming, and scheduling tasks are intermingled and fundamentally a challenging problem in mmWave system, due to the limited power budget and beamformers, diverse user traffic loads, user quality of service requirement, etc. To solve this problem, a novel framework is proposed, termed clustering Based dOwnlink user assOciation, Scheduling, beamforming with power allocaTion (BOOST), to minimize overall DL transmission time for given user traffic loads, subject to BS transmit power budget and the min. SINR at the user.
- Key Findings: Important components of BOOST are interference controlled UE clustering, traffic balancing UE association, and beamforming and UE traffic scheduling. MATLAB simulations showed that BOOST reduced DL transmission delay by 37%, 30%, and 26% compared with the three state-of-the-art schemes in the literature, respectively. This work has been accepted in IEEE Transaction on Mobile Computing (Impact factor: 4.474).
This work investigates the impact of initial cell search overhead on achievable throughput at user equipment (UE) in a millimeter wave mmWave cellular networks system. To compensate high signal attenuation at mmWave frequency, the directional transmission is preferred at base station (BS). However, it entails a very high system overhead during the initial cell search and UE association procedure if BS transmits synchronization and channel estimation signals over a large angular space using highly directive narrow beams. Unlike LTE predetermine cell structure, which is usually hexagonal, there is no fixed cell formation structure at mmWave BS. For this reason, initial synchronization overhead highly compromises the throughput at UE. In this paper, we investigate the system overhead during exploring different BSs signals to achieve high throughput at UE. Hence, we present a stopping rule called one stage look ahead (1-SLA) for BSs signals exploration to achieve high throughput in a time frame.
The ever increasing spectrum demand can be relieved by spectrum sharing technique in cognitive radio networks. However, the most popular opportunistic spectrum sharing architecture suffers from disruption to the secondary user (SU) communications by the primary user (PU) re-emergence, large spectrum sensing overhead, and short secondary user communication cycle. In this work, at first we propose a novel spectrum sharing architecture, termed frequency division dynamic spectrum sharing (FDSS), in which the SU and the SU share spectrum simultaneously and efficiently. Here we design a wavelet transform based FDSS spectrum sensing scheme, which is entirely different from existing spectrum sensing techniques in the sense that a SU is aware of the presence (or absence) of the PU but it does not know which portion of the frequency is currently being used (or empty) by the PU. Unlike the existing spectrum sensing scheme our spectrum sensing does not rely on the PU signal signatures such as signal type, modulation type, pilot signals, cyclic prefix, etc. The simulation results indicate that the proposed approach can effectively detect the dynamic variation of PU signal bandwidth on a licensed band even when the received signal to noise ratio at SU is very low.
- Design of a Linear Model for Regression and Classification using Machine Learning
In machine learning and statistics, regression and classification are the problems of estimating the relationships among variables, and identifying and categorizing a data set based on a known categorized data sets, respectively. In this project, we design and implement a basic linear regression and classification model on the C++ platform, which is basically based on the minimum square error estimation of the training data sets. Later, we introduce a regularized coefficient with classification model to apply the model on a smaller data set. Further, we implement Bayes classifier, Nave Bayes classifier and k-nearest neighbor classifier using the Matlab.
Bag-of-Words (BoW) is a state-of-art technique for image retrieval and automatic annotation, especially in large scale image databases. On the other hand, OverFeat model which is trained by ImageNet becomes a popular choice in the image classification and extraction field. However, the performance evaluation of OverFeat and comparison with BoW for low-level visual feature image data sets was unexplored. In this project, we use OverFeat and the BoW model to represent and classify the low-level visual feature gene images from 15 classes. Popular image features extractor techniques such as Scale Invariant Feature Transform (SIFT) and Dense Scale Invariant Feature Transform (DSIFT) were extensively analyzed and exploited to implement the BoW model. Moreover, we use Support Vector Machine (SVM) classifier to run a 10-fold cross-validation experiments to obtain the accuracy of both BoW and OverFeat.
- Image Restoration from Linear and Non-linear Degradation
Low contrast and blurred gray scale image restoration becomes a challenge when an appropriate pre-processing and precise filtering is not employed and a degradation function parameters are not correctly estimated. In this project, we restore a very low contrast gray scale image which was exposed to heavy uniform motion blur, point spread function blur, and additive Gaussian noise (AGN). At first, a non-linear homomorphic filter was used to enhance the image contrast. Then, Fast Fourier Transform (FFT) of the degraded image was analyzed for locating frequency components and orientation of noise energy and blurring effect. Finally, restoration was performed by parametric Wiener filter which shows prominent result compare to other well-known approaches.
- Image Edges Detection for Noisy and Broken-Edges-Continuity Images
Image Edges detection of a noisy and broken-edges-continuity image still faces significant challenges in image processing and computer vision research field. Classical image edges detector techniques such as Roberts, Prewitt, Sobel, Canny, Laplacian of Gaussian (LoG), etc., fail to extract real image edges; misjudge the noise points as the part of real edges and miss some real edges of these images. In this project, we apply the wavelet transform method to perform the adjacent scale multiplication in order to find real image edges and compare the performance with classical techniques. We first, decompose a noisy image into multiple wavelet scales since it has a good time-frequency localization and multi-scale identity to perform the multi-resolution analysis. Therefore, adjacent two scales are multiplied as a product function to magnify the edge structures and suppress the noises along the wavelet domain. Finally, we compute an adaptive threshold using noise standard deviation and correlation coefficient of adjacent scales to remove false image edges.
- TCP Slow Start and Dynamic Window Sizing
The explosive growth of internet usage over past few decades results in severe congestion problem in the network. Limited size bu er drops many packets due to overflow but the flow of packets and congestion still exist in the network. TCP slow start and dynamic window sizing for packet transmission can significantly reduce the network flooding and congestion. In this project, we analyze a small network with two hosts and one limited size buffer for estimating some important network parameters with TCP slow start and dynamic window sizing method. Two hosts were connected to the internet. The internet was represented by limited storage size buffer with some background traffics. Important network parameters such as transmission link utilization, packet loss, re-transmission time-out, and acknowledge time-out, etc., have been investigated thoroughly to characterize the network.
Today's advances in aircraft design still remains an open issue on how automatic control systems truly behave. In this project, we investigate one main component of automatic flight control system; a pitch controller which controls the pitch angle of a 777-200 Boeing commercial aircraft. At first, multiple sinusoidal excitations, with different frequencies, are applied on the designed Matlab Simulink state-space model to achieve a Bode plot by analyzing magnitude plots and Lissajous patterns. Later, the Bode plot was decomposed to achieve the transfer function of our system. Finally, the performances of the pitch controller were evaluated for a pre-specified design specification.