Projects

Simulation Results & Experimental Visuals

Numerical simulations, radiation patterns, and experimental prototypes

1 / 8

Traffic Map and Violation Detection Sensor

2 / 8

3 / 8

4 / 8

5 / 8

6 / 8

7 / 8

8 / 8

❮ ❯

MIMO Radar Systems & Sensory Arrays Master's Graduate Thesis Advisors: Dr. S. Akhavan & Prof. K. PoorAghdam

Design & Optimization of 1-D & 2-D Sensory Arrays for Tracking Offending Vehicles (Master's Thesis)

Have you ever considered any other option for tracking, managing, and controlling roads and streets rather than widely used Traffic Cameras?

In today's world with ever-increasing life speed and demanding transportation needs, managing the high volume of traffic smoothly and swiftly is vital. While optical cameras are widely deployed, their high data volume, vulnerability to adverse weather or lighting conditions, and computationally intensive computer vision pipelines pose major bottlenecks for low-latency violation detection.

To address these challenges, my Master's thesis at the University of Tehran focused on the design and mathematical optimization of 1-D and 2-D sensory arrays capable of tracking offending and violating vehicles in roads using MIMO (Multiple-Input Multiple-Output) radar systems. By synthesizing tailored radiation patterns and suppressing sidelobe levels, MIMO radar achieves high angular resolution with a reduced number of physical RF transceivers.

Key scientific and algorithmic contributions include:

1-D & 2-D Sensory Array Optimization: Formulated sensory array synthesis using both convex optimization and non-convex optimizers, alongside Gradient Descent (GD) algorithms to optimize element positions, phase weights, and excitation amplitudes for maximum directivity and sidelobe suppression.
Multi-Target Kinematic Tracking: Deployed Kalman Filtering to continuously track vehicles' kinematic state vectors (range, radial velocity, azimuth angle) across consecutive radar frames to identify speeding, lane violations, and erratic maneuvers.
Clutter & Spatial Noise Clustering: Integrated the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm to cluster dense radar point clouds and eliminate multi-path reflections and environmental ground clutter before feeding detections to the kinematic tracking filters.

The entire MIMO radar pipeline, array synthesis, and kinematic estimators were modeled and simulated in MATLAB, evaluated across 5 GHz and 77 GHz automotive millimeter-wave radar bands:

R-D Map in 5 GHz R-D Map in 5GHz

XY Plane Vehicle Tracking XY Map in 5GHz

Violation Detection & Road Tracking Violation Detection in 5GHz

GD Cost Function vs. Iterations Cost Function Convergence in Gradient Descent

Microwave Antennas & Waveguides Undergraduate Thesis

Slotted Substrate Integrated Waveguide Sensor (77 GHz)

What is the incentive to engineer such a variety of Sensors in the market?

Physical constraints, volume limitations, and manufacturing expense guide sensor design. For autonomous vehicles' RF sensors, conventional waveguides are bulky and costly. Starting from leaky-wave principles, I designed a conformal annular slotted waveguide sensor.

To transform this bulky design into a low-profile, highly integrated, and cost-effective unit, I designed a Slotted Substrate Integrated Waveguide (SSIW) Annular Sensor at 77 GHz and simulated this unit in Ansys HFSS.

77 GHz Annular Slotted SIW Sensor Design 77GHz Annular Slotted SIW Sensor Design

S11 Return Loss Parameter S11 Parameter of ASSIW Sensor

Convex Optimization Gradient Descent • Newton Method • MATLAB

Optimization: Quadratic Cost-Function

One of the most powerful optimization methods when dealing with convex and non-convex problems is the Gradient Descent method family. Here, both Steepest Descent and Newton's Method are investigated for minimizing the quadratic cost function:

Objective Function: $$ f(x) = x_{1}^{2} + x_{2}^{2} - 4x_{1} - 6x_{2} + 13 + x_{1}x_{2} $$

Analytical differentiation verifies the global minimum since the objective is strictly convex. We then apply Gradient Descent and Newton's method to numerically trace the convergence trajectory across the curvature contours:

Newton Method Trajectory

Steepest Descent Contours Steepest Descent Method

IoT & Industrial Systems Raspberry Pi • STM32 • PyQt5 • Python

Implementation of IoT-based Monitoring System in a Machine Lab

During the 4th semester at the University of Tehran, our team designed and deployed an end-to-end industrial IoT monitoring system for the Electrical Machines Laboratory.

The project encompassed real-time networking protocols, embedded telemetry with STM32 and Raspberry Pi, back-end communication in Python, and an interactive instrumentation GUI built with PyQt5.

Laboratory UI & Monitoring Dashboard Architecture Machine Laboratory UI

Neurotechnology & ML EEG Signals • CSP Filter • LDA Classifier • 90% Accuracy

Brain-Computer Interface: EEG Action Classification via ML

In this project, we tackled motor-imagery electroencephalography (EEG) signal decoding to classify 4 distinct motor intentions:

Class 1: Movement of Right Finger
Class 2: Movement of Right Arm
Class 3: Movement of Right Leg
Class 4: Rest / No Action

We engineered a processing pipeline utilizing Common Spatial Pattern (CSP) spatial filters followed by a Linear Discriminant Analysis (LDA) classifier in the mapped spatial domain. Pre-whitened signal matrices allowed the classifier to achieve approximately 90% classification accuracy.

EEG Optimal Classifier Architecture EEG Classifier

Confusion Matrix (90% Accuracy) EEG Confusion Matrix

Computational Electromagnetics FDTD • Yee Grid • Boundary Conditions • MATLAB

Finite Difference Time Domain: Wave Propagation

Numerical methods in electromagnetics form the core engine of industrial simulation suites such as Ansys HFSS, CST Studio, and COMSOL. Here, the Finite Difference Time Domain (FDTD) technique is employed to solve Maxwell's curl equations numerically on a discretized spatial-temporal grid.

Wave propagation was modeled in a closed conducting cavity containing a radiation excitation source, both with and without an obstacle in the wave path, utilizing Perfect Electric Conductor (PEC) hard-boundary conditions:

FDTD with Interior Obstacle FDTD with Obstacle

FDTD Cavity Propagation (Free) FDTD without Obstacle

Quantum Computing & AI QAOA • Adiabatic Quantum Computing • Hamiltonian • Python

Quantum Machine Learning & Parameterized Circuit Optimization

Quantum Machine Learning merges quantum mechanical principles with classical optimization paradigms to solve high-dimensional combinatorial and physics-based problems intractable for classical architectures.

The Quantum Approximate Optimization Algorithm (QAOA) numerically differentiates parameterized quantum circuits to interface with classical optimizers (e.g. Adam) to tune gate angles. Under Adiabatic Quantum Computing (AQC), Hamiltonian minimization evolves the system ground state toward the optimal problem solution:

Parameterized Quantum Circuit Cost: $$ Cost = \| F( ec{ heta}) - 0.4 \|^{2} $$

where $F( ec{ heta})$ denotes the expectation value of the quantum circuit and the objective is framed to minimize Mean Square Error (MSE):

Parameterized QML Circuit Architecture

Optimization Cost vs. Iteration QML Optimization Cost

Antenna Arrays & Beamforming Array Factor • Convex Sparsity • Side Lobe Level (SLL)

Optimization of 1-D Array to Reach Desired Pattern

Antenna array optimization synthesizes radiation patterns satisfying rigorous physical constraints and minimal element counts. We formulate 1-D array synthesis by minimizing the MSE cost function between the synthesized array factor $AF(\theta)$ and a target envelope $g(\theta)$:

Synthesis Cost Function: $$ J = \| AF(\theta) - g(\theta) \|^{2} $$

By coupling convex optimization with $\ell_1$-norm regularization, we achieve a sparse antenna arrangement with minimized elements while maintaining the desired Side Lobe Level (SLL). Below is the comparison between uniform and optimized array configurations:

Array Configuration Comparison (XZ Plane) Comparison of Array Configs

Optimized vs. Desired Radiation Pattern Radiation Pattern Synthesis

Enterprise Security & Systems Architecture Active Research (2025–Present) ISO/IEC 27001 • NIST CSF • SABSA • TOGAF

DevSecOps Fundamentals & Enterprise Architecture Implementation

How can modern automated DevSecOps pipelines successfully permeate complex organizational hierarchies without stalling under bureaucratic routines?

While continuous integration and continuous delivery (CI/CD) pipelines offer unparalleled deployment velocity, integrating robust security (Shift-Left) across enterprise-wide ecosystems often encounters organizational resistance, compliance friction, and rigid bureaucratic routines.

This active research project investigates the theoretical foundations and applied frameworks of DevSecOps, conducting an exhaustive literature review to establish a unified operational model tailored for enterprise adoption. The research develops actionable blueprints to harmonize automated security telemetry with formal governance, risk, and compliance (GRC) workflows.

The architectural synthesis bridges four internationally recognized standards:

ISO/IEC 27001: Mapping automated vulnerability assessments, artifact signing, and pipeline access controls to Annex A security control objectives and continuous audit trails.
NIST SP 800-53 & NIST CSF: Implementing risk mitigation matrices, continuous authorization (cATO), software supply chain security (NIST SSDF), and automated compliance enforcement.
SABSA (Sherwood Applied Business Security Architecture): Developing business-driven security matrices to guarantee end-to-end traceability from contextual enterprise goals down to component-level pipeline security policies.
TOGAF (The Open Group Architecture Framework): Embedding DevSecOps capabilities into enterprise architecture domains (Business, Data, Application, and Technology) via the TOGAF Architecture Development Method (ADM).