PhD thesis defense to be held on February 1, 2023, at 11:00 (Tele-teaching room 1, NTUA Central Library Building / Virtually via Zoom)


Picture Credit: Alexandros Kyritsis

Thesis title: Contribution to the Development of Combined Methods for Detecting Small Unmanned Aerial Systems

Abstract: The present thesis studies the development and implementation of an original system aiming to detect small Unmanned Aerial Systems (UAS) using a combination of active and passive methods. The wide use of UAS applications in many fields of activities inevitably creates the need for systems able to detect flights over specific areas of interest. Low radar cross-section (RCS), flight altitude and speed, all render UASs difficult targets for traditional radar systems to detect. The cooperation of passive methods, that exploit signals emitted by the UAS, produces a synergy that leads to facilitating their identification.
The study begins by briefly presenting aspects of the emerging UAS technology, in order to outline the multi-layered approach necessary to address the task of UAS detection. Basic principles of EM theory involved in UAS detection using active methods are presented; respectively, for passive methods that exploit acoustic (sound) waves emitted by UASs, suitable techniques of microphone array signal processing and machine learning are reported. A detailed analysis follows, describing the proposed methods for (a) active detection using continuous wave Doppler radar, (b) estimating the direction of arrival (DOA) of incoming sound signal and (c) UAS sound identification using machine learning, discerning them from other sound sources. Next, a report of radar systems that were tested for the purposes of the thesis is presented, with emphasis given to the proposed 24 GHz CW Doppler radar and its development stages. Using custom made software created for radar control and data acquisition, the FFT of the in-phase (I) and quadrature (Q) component of its output is exploited for real-time, reliable UAS detection over significant distances. Measurement data from live test runs performed using both static mode and 360° scanning mode, confirmed the system’s detection effectiveness.
Regarding the use of passive methods for UAS detection and identification, two setups were examined: a linear microphone array and a cross-shaped array. Because of the inherent inability of the linear (one-dimensional) array for sound DOA angle estimation, the main focus shifted towards examining the cross-shaped array that consisted of 4 condenser microphones. Optimal distances between the array elements were determined through simulation and the maximum distance for DOA estimation was measured during live tests conducted at 3 different types of locations. For the task of UAS sound identification, (a) harmonic line association and (b) machine learning techniques were explored; emphasis was given in (b), by designing and training a multi-layered convolutional neural network that utilizes spectrograms of the sound signal captured by the microphones to identify UAS flights in real-time.
The proposed system integrating the abovementioned active and passive methods, has been extensively tested outdoors, performing live measurements of real-life airborne vehicles, confirming the results for reliable detection and identification.

Supervisor: Professor Emeritus Nikolaos Uzunoglou

PhD Student: Alexandros Kyritsis