The Smart Sensing Lab (SSL) research focuses on signal processing, system design and phenomenology of mmW sensors and sensor arrays. Taking part in the autonomous driving revolution, our research addresses the smart sensing for autonomous vehicles and unmanned aerial vehicles (UAV). Since autonomous driving poses conceptually different challenges comparing to the conventional military radars, our research focuses not only on the development of innovative signal processing approaches, but also on understanding new phenomenology and its innovative modeling. The main point of our research philosophy is to understand the physical processes underlying the considered problem and to look for physical model-based solution. In other words, we are seeking to gain some prior information on a problem and to find a sophisticated way to incorporate this knowledge into the solution.
Our research focuses on array signal processing, and statistical signal processing, where we tackle some of the challenging problems caused by nonstationary environment phenomena: multipaths, clutter, and interferences. We have developed new approaches for non-stationary Doppler-spreaded clutter mitigation, snapshots deficient beamforming, adaptive beamforming without signal-free data, beamforming of spatially distributed targets, and multipath modeling and mitigation. The objective of our research in this field is to develop a physical model for the environmental behavior and to exploit diversity in environmental conditions, diversity in array location and orientation, and spatial diversity of the distributed target-induced wavefronts. Originated in the military radar applications, characterized by well-defined sparse targets, our automotive radar research focusses on the distributed targets detection within strong, dense, and Doppler-spreaded urban environment clutter. During our work on automotive sensing, we developed high-resolution imaging radars that required re-definition of the “target” and the “clutter” notions, and therefore, derived conceptually innovative signal processing methods for automotive radars.
Our research is characterized by its interdisciplinary nature, where we adopt methods from one field into another in the non-trivial way. Thus, we adopted innovative compressive sensing approach that was introduced for image processing into radar array signal processing. Similarly, inspired by various computer vision methods, we developed innovative approaches for the radar target classification, clustering, detection, and tracking. Our current research migrates deep learning tools from the computer vision domain into automotive radars. Leveraging communications advances in coding theory, we are working on the CDMA MIMO radar waveform code family selection and development of the corresponding innovative detection approaches. Introducing the information theory-based criteria into target localization problem, we formulated a novel interpretation for the positioning problem that led to alternative positioning methods. In another research, we adopted approaches from the statistics to introduce robust target classification metrics.
Our current research focusses not only on the signal processing, but also on the operation concept (e.g. distributed radar with centralized processing and widely distributed antennas), sensor fusion between radars, lidars, and optics, and phenomenology, such as radio frequency (RF) propagation through the bumper fascia, interferences, multiple reflections and others. Since automotive radar is the key component of the autonomous sensing suit along with other sensors, our research addresses sensor fusion, multi-sensors, and multi-modality sensing challenges. Moreover, my current research on the automotive sensing, provides opportunities to engage into the entire autonomous high-level system design, customer-oriented features, control, communications, localization, and applications.
