Elbir, Ahmet MusabMishra, Kumar VijayEldar, Yonina2020-04-302020-04-3020199781728116792https://dx.doi.org/10.1109/RADAR.2019.8835626https://hdl.handle.net/20.500.12684/1652019 IEEE Radar Conference, RadarConf 2019 -- 22 April 2019 through 26 April 2019 -- 152051In cognitive radar, it may be desired to select an optimal subarray from a full antenna array in each scan to reduce the cost and computational complexity. Previous works on antenna selection rely on mostly optimization or greedy search methods. In this paper, we introduce a deep learning approach for antenna selection in a cognitive radar scenario. We design a deep convolutional neural network (CNN) to select the best subarray for direction-of-arrival estimation for each scan. The CNN accepts the array covariance matrix as its input and, unlike previous works, does not require prior knowledge about the target location. The performance of the proposed CNN approach is evaluated through numerical simulations. In particular, we show that it provides more accurate results than conventional support vector machines. © 2019 IEEE.en10.1109/RADAR.2019.8835626info:eu-repo/semantics/closedAccessAntenna selection; Cognitive radar; Convolutional neural networks; Deep learning; DoA estimationConference Object