MIMO Radar With Sparse Linear Arrays - Theory, Implementation and Applications

Multiple-input multiple-output (MIMO) radars have several advantages as compared to traditional phased arrays. They can achieve higher resolution with the same number of antennas. They can also achieve wide field of view, illuminating multiple targets at the same time, which translates to faster detection time. Reduction of the number of active antennas without hurting the radar performance would reduce the cost of the radar, while a low-cost, high resolution radar would advance the state-of-art of autonomous driving, smart environment, smart home, and IoT sensing, and would enable applications such as smart patient care, elderly monitoring, fitness assistant, etc., that rely on sensing. In an era where COVID-19 forced home isolation with limited supervision of vulnerable segments of the population, a radar device could provide information on vital signs, or detect falls without invading people's privacy in the way surveillance cameras would. MIMO radar using specially designed Sparse Linear Arrays (SLAs) can enjoy reduced hardware cost without losing the MIMO radar advantages. An SLA can be thought of as a uniform linear array with only a small number of active antennas. By careful selection of the active antennas and optimal design of transmit waveforms, one can maintain a radar performance close to that of the fully populated array. However, finding an optimal sparse array geometry in terms of the fewest antennas is a difficult combinatorial problem. The proposed project will advance the state-of-art of SLA based MIMO radar as a cost-effective imaging radar by (i) providing a novel framework for antenna selection, (ii) developing an SLA MIMO radar prototype based on frequency-scanning metamaterial (MTM) antennas, and (iii) developing real-time activity monitoring and user identification schemes that leverage the high resolution and wide field of view of MIMO SLA radar.

There are several novel aspects in the proposed work. (i) A novel machine learning approach for antenna selection is proposed, which offers a unifying framework for dealing with any performance metric. The novelty of the proposed approach lies in its ability to get multiple softmax models to work together. (ii) The use of MTM antennas brings in the added advantage of allowing for easy change of the beam elevation by varying the antenna frequency. That advantages will be exploited to look for targets in the 3-D space while still using a linear array. By varying the frequency of the MTM antennas, one can select the elevation direction of the transmit beam, and by applying the proposed SLA design method, one can design the beam pattern in the 2-D space corresponding to the selected elevation direction. The frequency scanning capability resulting from the dispersive nature of MTM allows a real time and low complexity beam scanning mechanism, whereas the SLA MIMO radar with proper waveform engineering will generate a large scale virtual array with enhanced angular resolution. As such, the combination of SLA MIMO radar with MTM antennas will enable an unprecedented radar architecture with larger field of view, finer resolution, and small number of antenna RF fronts. (iii) Low-latency signal processing algorithms will be developed for leveraging the large field of view and high angle resolution, that will have the capability to construct 3D user models and identify multiple targets simultaneously. Innovative neural network structures will be devised to enable device-free user activity monitoring. It is expected that the multi-user identification mechanisms will reveal unique user-specific activity characteristics embedded in the movements of high-resolution point clouds, facilitating a broad range of emerging mobile applications.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.