arXiv Science⌕ Search

arXiv subjects

Azharul Islam

Publications and source records attributed to Azharul Islam.

2 recordsLinked to original sources

Physics-Aware Transformer-Based Beamforming Optimization for Near-Field E-MIMO ISAC Systems

This paper proposes a physics-aware Transformer-based beamforming optimization (PATBO) framework for near-field extreme massive multiple-input multiple-output integrated sensing and communication (E-MIMO-ISAC) systems. PATBO combines regularized zero-forcing (RZF) communication directions and spherical-wave near-field sensing directions with learned stream-power allocation. A token-preserving Transformer maintains the correspondence between individual users/targets and their allocated powers, while a Cramér--Rao bound (CRB)-sensitivity-aware objective emphasizes power-allocation errors with greater localization impact. Physical feasibility is explicitly verified after inference, with infeasible predictions handled through physics-constrained power repair and optimizer fallback. For a $32\times32$ E-MIMO array with $1024$ antennas, the proposed structured representation reduces the learned output from $12{,}288$ real-valued coefficients of an unstructured complex beamformer to six stream-power variables. Direct PATBO achieves $100.98\%$ reference-rate retention and $19.54\%$ joint feasibility, which increases to $87.79\%$ after repair. The complete hybrid framework achieves $99.89\%$ reference-rate retention and $100\%$ constraint satisfaction on the retained test set, while requiring optimizer fallback for only $12.21\%$ of the scenarios. Direct PATBO inference requires $1.11$~ms per sample, corresponding to approximately $858\times$ and $779\times$ lower runtime than the evaluated cold-start and PATBO-warm-start sequential quadratic programming (SQP) implementations, respectively.

eess.SP↗

An Interactive Google Earth Engine Application for Global Multi-Scale Vegetation Analysis Using NDVI Thresholding

Monitoring vegetation dynamics is crucial for addressing global environmental challenges like degradation and deforestation, but traditional remote sensing methods are often complex and resource-intensive. To overcome these barriers, we developed an interactive, cloud-based application on the Google Earth Engine (GEE) platform for few clicks on-demand global vegetation analysis without complex technical knowledge. The application automates the calculation of vegetated areas using the Normalized Difference Vegetation Index (NDVI) derived from Sentinel-2 and Landsat imagery. It utilizes a median composite of images over a selected period to create a single, robust, cloud-free image, minimizing atmospheric noise and other artifacts. It offers a flexible, global multi-scale analytical platform, allowing users to define regions of interest based on administrative boundaries, protected areas, or custom-drawn polygons. The user-friendly interface enables the selection of specific time periods and NDVI thresholds to quantify vegetation cover in real time, eliminating the need for manual and time intensive data handling and processing. A validation of the platform was conducted for two protected areas in Bangladesh which demonstrated high accuracy, with area estimates showing over 97% agreement with published reference data. By simplifying access to powerful geospatial analytics to general people, this tool provides a scalable and practical solution for researchers, land managers, policymakers, and any interested person to monitor vegetation trends, support conservation efforts, to inform decision making in spatial context where policy maker need to use insights in few clicks and inform environmental policy.

cs.HC↗