AY
A. Yaldız
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
1 records found
1
This thesis investigates drone detection using radar in the K-band, around 24 GHz, where detection is challenging because drones have a low radar cross section (RCS), move at relatively low radial speeds, and must be distinguished from clutter. For range-Doppler detection several Constant False Alarm Rate (CFAR) detectors have been implemented, namely Cell-Averaging CFAR (CA-CFAR), Ordered-Statistic CFAR (OS-CFAR), Greatest-Of CFAR (GO-CFAR), Smallest-Of CFAR (SO-CFAR), and Clutter-Map CFAR (CM-CFAR). For direction of arrival (DoA) estimation, MVDR and MUSIC are used to estimate the azimuth angle. Together, these stages output the range, radial velocity and azimuth of the detected drone. The trade-off between computational cost and detection performance is then studied by building a benchmark and testing the algorithms against it. At close range, from 0 to 25 m, the CFAR algorithm with the highest 𝐹2 score was CA-CFAR, while beyond 25 m CM-CFAR was generally more robust, and
among the spatial CFAR algorithms SO-CFAR remained competitive at longer ranges. For the direction of arrival a new range-Doppler guided processing pipeline is introduced, which runs between 80 and
120 times faster than the original full range-angle pipeline. This speed-up comes at the cost of detection performance. The most robust angle estimator was MUSIC combined with argmax selection, although it requires the number of sources to be known beforehand. MVDR with argmax selection achieved slightly lower results but does not carry this requirement. By contrast, in the original full range-angle pipeline, MUSIC combined with the 2D GO-CFAR consistently achieved the highest 𝐹0.5 scores. Finally, the selected configuration was evaluated on an additional validation flight. Although the detected range
trajectory broadly followed the GPS-derived trajectory, the large number of false detections showed that the selected configuration only partially transferred to this flight. ...
among the spatial CFAR algorithms SO-CFAR remained competitive at longer ranges. For the direction of arrival a new range-Doppler guided processing pipeline is introduced, which runs between 80 and
120 times faster than the original full range-angle pipeline. This speed-up comes at the cost of detection performance. The most robust angle estimator was MUSIC combined with argmax selection, although it requires the number of sources to be known beforehand. MVDR with argmax selection achieved slightly lower results but does not carry this requirement. By contrast, in the original full range-angle pipeline, MUSIC combined with the 2D GO-CFAR consistently achieved the highest 𝐹0.5 scores. Finally, the selected configuration was evaluated on an additional validation flight. Although the detected range
trajectory broadly followed the GPS-derived trajectory, the large number of false detections showed that the selected configuration only partially transferred to this flight. ...
This thesis investigates drone detection using radar in the K-band, around 24 GHz, where detection is challenging because drones have a low radar cross section (RCS), move at relatively low radial speeds, and must be distinguished from clutter. For range-Doppler detection several Constant False Alarm Rate (CFAR) detectors have been implemented, namely Cell-Averaging CFAR (CA-CFAR), Ordered-Statistic CFAR (OS-CFAR), Greatest-Of CFAR (GO-CFAR), Smallest-Of CFAR (SO-CFAR), and Clutter-Map CFAR (CM-CFAR). For direction of arrival (DoA) estimation, MVDR and MUSIC are used to estimate the azimuth angle. Together, these stages output the range, radial velocity and azimuth of the detected drone. The trade-off between computational cost and detection performance is then studied by building a benchmark and testing the algorithms against it. At close range, from 0 to 25 m, the CFAR algorithm with the highest 𝐹2 score was CA-CFAR, while beyond 25 m CM-CFAR was generally more robust, and
among the spatial CFAR algorithms SO-CFAR remained competitive at longer ranges. For the direction of arrival a new range-Doppler guided processing pipeline is introduced, which runs between 80 and
120 times faster than the original full range-angle pipeline. This speed-up comes at the cost of detection performance. The most robust angle estimator was MUSIC combined with argmax selection, although it requires the number of sources to be known beforehand. MVDR with argmax selection achieved slightly lower results but does not carry this requirement. By contrast, in the original full range-angle pipeline, MUSIC combined with the 2D GO-CFAR consistently achieved the highest 𝐹0.5 scores. Finally, the selected configuration was evaluated on an additional validation flight. Although the detected range
trajectory broadly followed the GPS-derived trajectory, the large number of false detections showed that the selected configuration only partially transferred to this flight.
among the spatial CFAR algorithms SO-CFAR remained competitive at longer ranges. For the direction of arrival a new range-Doppler guided processing pipeline is introduced, which runs between 80 and
120 times faster than the original full range-angle pipeline. This speed-up comes at the cost of detection performance. The most robust angle estimator was MUSIC combined with argmax selection, although it requires the number of sources to be known beforehand. MVDR with argmax selection achieved slightly lower results but does not carry this requirement. By contrast, in the original full range-angle pipeline, MUSIC combined with the 2D GO-CFAR consistently achieved the highest 𝐹0.5 scores. Finally, the selected configuration was evaluated on an additional validation flight. Although the detected range
trajectory broadly followed the GPS-derived trajectory, the large number of false detections showed that the selected configuration only partially transferred to this flight.