Research

This domain turns raw radar returns into trustworthy intermediate representations: SAR/ISAR formation, motion compensation, autofocus, 3D rotational compensation and cross-range scaling, super-resolution, and polarimetric or interferometric scattering analysis.

We move beyond the monostatic limit through bistatic, multistatic and MIMO interferometric ISAR geometries and W-band synchronization across distributed platforms, recovering resolution from sparse apertures via compressive sensing and deep super-resolution.

Frequency band and platform are secondary. Work belongs here when its main contribution is synchronization, correction, reconstruction or resolution — whatever the application.

• SAR/ISAR
• Autofocus
• Motion Compensation
• Super-Resolution
• Bistatic/Multistatic
• PolSAR/InSAR
• ATI/GMTI
• Compressive Sensing

Our research focuses on intelligent radar-based perception for detecting and recognizing targets in complex remote-sensing environments. Radar and synthetic aperture radar (SAR) provide reliable sensing capabilities over long distances and under diverse environmental conditions, making them well suited for large-scale surveillance and remote sensing. However, variations in sensing geometry, background clutter, and target appearance often make robust interpretation of radar imagery challenging.

To address these challenges, we study methods for object detection and feature enhancement that emphasize informative target signatures while suppressing irrelevant clutter. Our research explores deep learning-based representation learning, target-aware feature extraction, and clutter-robust processing to improve the localization and characterization of objects in radar imagery. These approaches aim to provide more discriminative and reliable features for subsequent recognition.

Based on these enhanced representations, we develop target recognition methods that accurately identify target classes across diverse sensing and operating conditions. Particular emphasis is placed on robustness and generalization, enabling recognition models to maintain reliable performance even when target signatures, backgrounds, or acquisition conditions differ from those observed during training. Ultimately, our goal is to build reliable and intelligent radar perception systems for practical remote-sensing applications.

• Remote Sensing
• Change Detection
• Radar Target Recognition
• Micro-Doppler
• Feature Enhancement
• Clutter Reduction
• Representation Learning
• Open-Set / OOD
• Domain Generalization
• Self-Supervised Learning
• Foundation Model
• Robust Radar Perception

Electronic warfare (EW) research focuses on countering anti-ship missiles equipped with monopulse radar seekers through various electronic attack (EA) techniques and expendable countermeasures such as chaff and decoys. We also investigate electronic protection (EP) techniques against hostile electronic attacks and analyze integrated EA–EP engagement scenarios for maritime defense.

We model the monopulse radar seeker installed in a threat missile and analyze its target-tracking process based on sum and difference antenna patterns. The target angle is estimated from the sum–difference ratio of the received signals, while the missile motion is modeled using six-degree-of-freedom (6-DOF) dynamics. This framework enables quantitative analysis of missile guidance and interception performance under different EA and EP conditions.

For realistic electronic attack analysis, complex radar backscattering signals and high-resolution range profiles (HRRPs) are generated from CAD models of maritime targets while considering their physical scattering characteristics. Based on these models, an engagement simulator is implemented to evaluate target tracking and interception performance. The effects of representative countermeasures—including noise jamming, range gate pull-off (RGPO), cross-eye jamming, chaff, and decoys—are analyzed by varying jamming power and operational parameters.

• Electronic Warfare
• Electronic Attack
• Electronic Protection
• Monopulse Radar Seeker
• Anti-Ship Missile
• Noise Jamming
• RGPO
• Cross-Eye Jamming
• Chaff

Radar senses people without cameras. It works in darkness and under occlusion, records no optical imagery, and stays low-cost and privacy-preserving — yet human returns are weak, buried in static clutter, corrupted by indoor multipath, and vary with posture, subject and environment. We work along the whole pipeline to close that gap, starting at the signal: with IR-UWB, MIMO FMCW and CW radars we suppress clutter, extract phase time series of millimeter-scale body motion, and separate direct paths from multipath through DoA/DoD estimation.

On top of these signals we build an understanding of people: localizing them indoors with multipath mitigation, counting occupants and estimating crowd density — evolving from scattering-center extraction through feature-based learning to spatiotemporal transformers that adapt across clutter environments — and recognizing activity and hand gestures from Doppler–angle signatures, down to digits written in the air with a single CW radar.

The same sensing extends to physiology and real spaces. We extract respiration, heart rate and blink duration from phase time series even while the subject moves, strengthen them with video–RF fusion, detect lying-down people through their respiration, and validate counting in real-world crowds. The end metric is reliable, privacy-aware information about people, not image formation itself.

• Human-Centered Sensing
• IR-UWB Radar
• MIMO FMCW Radar
• Multipath Mitigation
• People Counting
• Gesture Recognition
• Vital Sign Monitoring
• Privacy-Aware Sensing

Radar directly measures range, azimuth, elevation and radial (Doppler) velocity, works in darkness, dust and smoke, and stays low-cost and privacy-preserving — yet limited angular resolution, sparse point clouds and multipath ghost targets have kept it from becoming a primary sensor for robots. We work along the whole pipeline to close that gap, starting at the sensor: with automotive and MIMO FMCW radars we extend the maximum unambiguous range, improve angular resolution through DoA/DoD estimation, and generate dense 3D point clouds.

On top of these measurements we build scene understanding: detecting and discriminating people, robots and objects under indoor clutter, estimating wall and reflector geometry from scattering-center distributions, and turning multipath from a nuisance into information — localizing non-line-of-sight (NLOS) targets around corners by mirroring their multipath ghost targets across the estimated reflecting surfaces.

The same perception feeds real platforms. We detect vehicles and estimate their heading and length from both stationary and vehicle-mounted radars, and compare and fuse radar with LiDAR and cameras for robotics and autonomous systems. The end metric is reliable perception, localization and tracking, not image formation itself.

• Automotive Radar
• MIMO FMCW Radar
• DoA/DoD Estimation
• 3D Point Cloud
• Multipath Mitigation
• NLOS Localization
• Target Tracking
• Sensor Fusion