Research

Interests, selected work, and future directions

My research spans wireless communications, RF/microwave sensing, signal processing, and machine learning. I develop methods that turn noisy, incomplete measurements into useful information and evaluate them through mathematical analysis, simulation, and experiments with real data and hardware.

I view RF sensing and wireless-health monitoring as complementary parts of a sensing-to-action system: one describes the physical environment, while the other describes the connectivity supporting sensing, computation, and coordination. My longer-term goal is to bring them together through physical and agentic AI, with reliable local operation and selective use of cloud resources.

Physical and agentic AI
Edge and mobile computing
RF sensing and imaging
Wireless health and QoE
Robust and private learning
Signal processing and optimization

Selected research

Comcast · Industry research

Wi-Fi health and decision support

My team and I developed an edge/cloud model for Wi-Fi health monitoring and optimal channel/band recommendation. The work connects wireless measurements to interpretation and actionable recommendations.

My contributions also include statistical and machine-learning frameworks linking gateway/client telemetry to WebRTC audio/video and set-top-box quality. These frameworks combine temporal features, application-quality labels, and controlled network experiments to investigate how signal quality, retransmissions, channel utilization, and rate dynamics relate to user experience.

Wi-Fi health · edge/cloud models · Application quality

Worcester Polytechnic Institute · SoilX

Intelligent RF sensing for subsurface characterization

Within the USDA-funded SoilX project, I led signal-processing and machine-learning work for root-zone soil-moisture estimation from ground-penetrating radar (GPR). I helped develop radar hardware/software and data-acquisition pipelines, coordinated measurements and cross-institution collaboration, and mentored student researchers.

This work connects electromagnetic observations to estimates of physical properties for precision irrigation. Associated research received the IEEE ORSS 2023 Best Paper Award; subsequent publications investigate machine-learning-enhanced GPR analysis and wavelet scattering for soil-water-content prediction.

Ground-penetrating radar · Soil moisture · Physical-state interpretation

SoilX project ↗

Representative publications

University of Houston · Collaborative research

Secure and learning-enabled wireless systems

At the University of Houston, I developed GAN-based defenses for neural radio classifiers and evaluated them over the air on USRP software-defined radios. Our team earned first place in the constrained category of the AFRL Beyond 5G SDR University Challenge.

My collaborative work also addresses adversarial attacks on learning-based resource management, physical-layer security, UAV communications, rate-splitting multiple access, hybrid precoding, and privacy-preserving continual learning. These studies connect learning, optimization, and wireless design under reliability and security constraints.

Adversarial robustness · SDR evaluation · Secure resource allocation

Representative publications

Doctoral research · Continuing collaborations

Data-efficient computational radar imaging

My doctoral research developed joint SAR imaging and scene decomposition from undersampled measurements. Low-rank and sparse representations exploit scene structure to recover images and separate moving targets from stationary background.

My first-authored work in IEEE Transactions on Computational Imaging studies multi-feature decomposition and dynamic-scene video-SAR. Collaborative extensions address fast, robust SAR/ISAR reconstruction and real-time sparsity-driven imaging. This provides the model-based foundation for my interest in data-efficient learning and adaptive sensing.

SAR / ISAR · Inverse problems · Low-rank and sparse models

Representative publications

Proposed research agenda

Building on these contributions, I am developing a research agenda around three connected directions. The emphasis is on experimentally verifiable decisions—not simply replacing domain models with a large language model.

ObserveInterpretLearn & decideActEvaluate & adapt

Grounded edge agents for wireless health and adaptive connectivity

How can an agent turn incomplete wireless measurements into effective, verifiable interventions?

I aim to combine structured state estimates, validated diagnostic models, and constrained optimization with agent reasoning. The agent would determine what evidence is missing, recommend a channel/band or resource change, and assess its consequences. Time-sensitive tasks would remain local; higher-level reasoning would be invoked when useful.

Adaptive RF/microwave sensing for physical AI

What should a sensing system measure next, and when does it have enough information?

I plan to combine electromagnetic models, sparse/low-rank structure, learned representations, and uncertainty estimates to guide radar acquisition. Measurement location, frequency configuration, and sampling density would adapt to the downstream task, beginning with soil-moisture characterization.

Efficient and trustworthy edge/cloud computing

Which parts of a sensing-to-action system should run on a device, a gateway, or in the cloud?

I will investigate compact on-device inference, task-aware model reduction, adaptive offloading, and hardware/software co-design. Model confidence, network health, energy, and action urgency would guide execution placement, with runtime checks and bounded fallback behavior when conditions change.

Across these directions, I will evaluate complete sensing-to-action systems: the quality of the physical estimate or application experience, alongside decision latency, energy, communication overhead, and recovery from failures.