Projects
Self-Supervised Contrastive Learning for Visual Representation
Implementation of SimCLR and Barlow Twins for self-supervised representation learning on STL-10, with label-efficiency evaluation across varying data regimes and comparison against a supervised baseline.
Vision Transformers for Image Classification on CIFAR-10
A Vision Transformer (ViT) built from scratch for CIFAR-10 classification, with positional embedding ablation studies and MetaFormer variant experiments (IdentityFormer, RandFormer) comparing alternative token-mixing strategies.
MLPs and ConvNets for CIFAR-10 Image Classification
Implementation and analysis of Multi-Layer Perceptrons and Convolutional Neural Networks for CIFAR-10 classification, including regularization, data augmentation, pixel-shuffling robustness studies, and label-noise experiments.
UAV Trajectory Optimization for Time-Limited Data Collection in LTE/5G Networks
An optimization-based framework for the AERPAW AADM Challenge that jointly plans UAV trajectories and base station selection using adaptive speed control and MILP-based scheduling to maximize data collection from LTE/5G base stations under mission time and geofence constraints.
LTE and 5G Communications — Simulation and Analysis
A collection of MATLAB-based simulations covering key LTE and 5G physical layer concepts including:
- Rayleigh/Rician fading channel modeling and channel impulse/frequency response visualization
- Doppler spectrum analysis and narrowband vs. wideband channel characterization
- PSS/SSS synchronization signal generation and LTE cell search
- CQI reporting and impact of CQI periodicity on throughput
- Transmit and receive diversity with Alamouti coding and maximal-ratio combining
- MIMO spatial multiplexing with pre-coding/post-coding matrix decomposition
- Rank indicator and PMI feedback under EPA/EVA channel models
- Inter-cell interference coordination and partial frequency reuse
- LTE PDSCH conformance testing and throughput analysis
- FDD/TDD resource grid comparison
- Channel estimation and MMSE equalization
- 5G NR downlink waveform generation with multi-BWP and SS Burst configuration
Digital Communication Modulation Schemes
Implementation of several common digital modulation schemes, studied and compared in terms of bandwidth efficiency and energy efficiency.
Federated Learning Model Optimization over NextG Wireless Networks — A Comprehensive Survey
A comprehensive survey on optimization techniques for federated learning over next-generation wireless networks, covering model optimization frameworks and communication-efficient strategies.
Implementation of Optimization Algorithms
Implementation of various optimization frameworks including Golden Section, Fibonacci, Gradient Descent, Steepest Descent, Quasi-Newton methods (Rank 1, DFP, BFGS), Linear Programming, and Canonical Genetic Algorithm.
Wireless System Simulation
In this project, according to different types of fading, a wireless system is simulated and some parameters such as received power, outage probability, coverage area of the base station, etc are calculated.
Narrow-Band Wireless Channel Simulation
Using different modulations or time diversity and space diversity, a wireless system is simulated and system performance in each case is analyzed.
OFDM System Implementation
A message with certain length is transmitted in a frequency selective channel via OFDM system. For this purpose, and OFDM system is designed and implemented. System performance with modules of Water-Filling, ZF & MMSE Equalizers, and ... is analyzed.
Real Time Object Detection with YOLOv3
Employing YOLOv3, a model is developed to detect 3 classes of Bus, Person, and car real time. For evaluation of the model, a test video is fed to the network and the output video has bounding boxes around the desired objects in the video.
Facial Emotion Recognition using Deep Learning Methods
Using FER2013 dataset, a CNN based deep learning method is proposed for intelligent facial emotion classification. Simulations illustrate significant improvement in comparison to baseline models.
Surface Detection using Deep Learning
A deep learning framework is developed for classification of the surface the subject is walking on using a 6 channel signal obtained from accelerometers and gyroscopes.
Developing Metric Based KNN Classifier
Using LMNN and LFDA after PCA to change the space of the datas and then implementing a KNN classifier on the mapped data.
Implementing a Deep Learning Model on CIFAR-10 Dataset
In this project, a neural network structure is developed for the purpose of classifying CIFAR-10 labels. Different hyperparameters are tuned and the optimum neural network for the classification is reached.
Naive Bayes Model Development on MINST Dataset
Initially, features are extracted from the dataset. Furthermore, a naive bayes model is developed for classification of MINST dataset.
Q-Learning Model Free Reinforcement Learning
In this project, an agent is trained for taxi-v3 game employing Q-Learning approach. The agent's intention is to pick up the fairs and dropping them at their destination.
Deep Learning Based DOA Estimation
A deep learning model based on convolutional neural networks is designed and simulated to predict direction of arrival. This model outperforms conventional methods such as MUSIC, Beamforming, etc.
BCI Signal Processing
A model based on CSP filter and LDA classifier is implemented to classify the action subject would take according to the obtained EEG signal from brain.
Implementation of Sparse Recovery Methods
Several methods for sparse recovery problem are implemented in this project and evaluated.
Independent Component Analysis
Several methods for independent component analysis are simulated in this project for the purpose of blind source separation problem in case of independent sources.
Dictionary Learning with K-SVD & MOD
Implementation of dictionary learning methods, K-SVD and MOD, for blind source separation problem in case of presence of sparse sources.
Movement Classification with Common Spatial Pattern
A model is developed based on CSP filters to map multi-channel T sampled observations to a feature vector for purpose of classification between arm movement and leg movement.
Canonical Correlation Analysis for Frequency Detection in SSVEP
A CCA model is developed for prediction of the frequency in steady state visually evoked potentials.
Implementation of Digital Communication System with 4PAM
Assuming an AWGN channel, a digital communication system is simulated which employs 4-PAM for transmitting the symbols. Each step at the transmitter and receiver is implemented in this project including generation of pulses, etc.
Compressing Image Employing 2D Fourier Transform
In this section of the project, it is aimed to compress the images via 2D fourier transform.
Mind State & Behaviour Prediction with EEG Brain Signal Analysis
By performing signal processing methods on EEG signals taken from subjects, such as separating low frequency and high frequency parts of the signal, state of one's mind is predicted.
Audio Processing using Filters
In this project, a voice which has echo and noise goes under process and using signal processing methods and filters, echo is deleted and noise is reduced.
Yagi-Yuda Antenna Design
A Yagi-Yuda antenna was designed and created with gain of 3.43dB and bandwidth of 500MHz in antenna lab using HFSS.