Abstract
UASs (Unmanned Air Systems) are universally used in many activities, spanning from leisure-commercial to military applications. Accordingly, as the number of UASs operating in the sky increases, so does the need to detect and identify them, in order to ensure their legitimate use. This paper introduces a continuous wave (CW) Doppler radar implementation that can be used to provide early warning for flying-by small UASs. By applying Fast Fourier Transform (FFT) to the returned signal’s Doppler frequency, estimations can be made regarding the presence of aerial bodies inside an Area of Interest (AoI). Achieving reliable detection with a low false alarm rate (FAR) while keeping the size and power demands of the system to minimum was a challenge that was successfully met. The proposed system was extensively tested in outdoor environments; measurement results are presented and parameters such as radar power, antenna gain, and noise are discussed.
1. Introduction-Related Work
The rising demands for safer, cost-effective, and autonomous solutions to complex tasks have led to the rapid development of the Unmanned Air Systems (UAS) industry and to the extensive use of drones. A large number of industrial and academic fields have benefited from this proliferation, including precision agriculture [1], marine science [2], and construction [3], to name a few. Similarly, civil, governmental, and military institutions have incorporated the use of UAS, as a vital part of their operational planning and execution, regarding privacy, security, and defence issues, respectively [4]. Despite the aforementioned lawful applications of UAS, the increasing threats caused by their potentially illegal uses have triggered the interest of both academic and industrial researchers in the means of detecting and/or mitigating these threats.
The term “LSS” (low, slow, and small) is commonly used when discussing the feasibility of detecting UAS. The main focus of traditional radar systems has been towards the detection of aircrafts, i.e., objects that fly at high altitudes and distances at least several kilometers away from the radar, moving at high speeds and comprised of large metallic surfaces that result to high returns of the radar signal [high Radar Cross Section (RCS)]. UASs on the other hand, fly at low altitudes/speeds and are made primarily out of plastic, which results in greatly reduced RCS values. The RCS value of various UASs at different frequencies has been investigated in [5] (RCS: −20 dBsm@10 GHz) and [6] (RCS: −13 dBsm@24 GHz) for a typical-sized DJI Phantom UAS such as the one that was mainly used for the purpose of the present work (shown in Figure 1). A summary of the RCS values of similar-sized quad-propeller UASs is given in [7]; for frequencies from 2.4 GHz to 15 GHz, the RCS values reported lie within −14.1 dBsm and −30 dBsm.
A wide range of counter-UAS (c-UAS) systems has been presented for use by both the civilian and military sectors [8]. These systems employ one or a combination of the available detection methods, i.e., radar, radio frequency (RF), electro-optical (EO), infrared (IR), and acoustic, to deliver high-complexity costly solutions.
Industry-grade UAS detection radar systems have been presented operating in lower frequency regions, characterized by voluminous implementations, high power demands, and considerable complexity [9–12].
On the other hand, a small number of K-band radar systems have been proposed for UAS-related applications in the literature. A distributed frequency modulation continuous wave (FMCW) radar system was tested for UAS detection in [13], using fiber-optic links instead of RF cabling, which increases complexity in favour of low distortion, low leakage between the transmitter/receiver and small propagation losses. A cadence velocity diagram (CVD) analysis using a 25 GHz K-band radar is presented in [14], for the detection of multiple microdrones while in [15], the micro-Doppler signatures of different types of drones and birds were experimentally measured by a K-band coherent radar system.
Nakamura and Hadama in [6] proposed a UWB radar system, transmitting and receiving ultrashort pulses with a bandwidth of 0.5 GHz or greater. The feasibility of UAS detection was reportedly confirmed after indoor measurements to a maximum distance of 7 meters. A detection range of 1 km ± 200 m using standard pulse-coherent short range battlefield radar with output signal (pulse) power of 200 W was achieved in [16]. A CW system was investigated in [17] for extracting micro-Doppler features of the UAS rotating parts, while an FMCW implementation in higher frequency is presented in [18].
Arguably, the use of multiple sensors operating in junction may prove to be more efficient in detecting UAS [19], given the overall system redundancy and the ability for one methods’ limitations to be compensated by another’s advantages. When detection range is of the essence, the use of radar systems is of key importance due to their (a) relatively low dependence on weather conditions, (b) manageable free-space propagation losses, and (c) availability of effective signal processing (narrowband) techniques.
Tackling the constraints of cost and complexity while considering the advantages of the radar-based detection method, in this paper, an implementation of an easily portable, quickly deployable, low-power CW Doppler radar system for UAS detection is presented. The system’s efficiency was tested in many setups, under different weather conditions, using different UASs, and other airborne vehicles.
This paper is organized as follows: Section 2 describes the proposed radar system, while briefly introduces the underlying Doppler signal processing principles applied to the collected signals. Section 3 presents the system’s experimental setup used in outdoor sessions while providing the respective results, summarizes the obtained outcomes, and discusses the overall performance of the proposed radar system as compared to other works. Conclusions are drawn in Section 4, indicating potential future expansions of the present work.
2. Materials and Methods
2.1. CW Doppler Radar Configuration
The developed CW Doppler radar configuration is shown in Figure 2. Operating in the K-band (24 GHz) with a total output power (PTx) calculated at 21 dBm, the system uses a homodyne receiver, down converting the received signal to baseband before separating the in-phase (I) and quadrature (Q) components.
(a)
(b)
Two standard gain horn antennas are used at the transmitter (Tx) and the receiver (Rx), providing a nominal gain of 20 dBi and a narrow beamwidth of approximately 17° (both on the horizontal and vertical planes). Α rough visualization of the estimated radar coverage volume is presented in Figure 3, indicating the space inside which a flying UAS is expected to be successfully detected.
Two setups of the specific radar system were used: (a) the static/staring setup (Figure 2(a)), where the antennas were left unobstructed, and were oriented parallel to the z-axis towards the potential Area of Interest (AoI) and (b) the full scanning mode (360°), in which a rotating reflector was installed in order to steer the radar beam by 90° around the z–axis. The hardware components involved in the implementation, are illustrated in Figure 2: power supplies, radar main board, stepper motor driver, horn antennas, reflector, and its controller (Arduino Uno).
2.2. Data Acquisition and Signal Processing
The in-phase (I) and quadrature (Q) signal components were acquired using audio channels with a sampling rate of 48 kHz. Thus, the time-domain signals can be written aswhere and represent the transmitted and received signals, respectively, and their amplitude, the center frequency and the phase difference. As it is known, the phase, as a function of the distance and the wavelength , is given by the expression:
An initial phase difference can be taken into account which can be expressed as
Considering to be the distance between the radar and the UAS at time , the UAS’s velocity and the angle between the UAS’s direction of movement and the reflected (received) signal, distance as a function of time , angle , and velocity is
The received signal given in (2), using (3)–(5) can be rewritten asor
By expressing (6b) in the form of I–Q components, the following equation is obtained for the received signal:where the UAS’s velocity is given by(representing an object moving with speed ) and represents the Doppler frequency shift. Fast Fourier Transform (FFT) is then applied to the down-converted baseband signal , allowing for a frequency domain representation of the (lower) Doppler frequencies, which is discussed in the next section.
Another important factor to be taken into account is that the main source of Doppler responses received by the CW radar is the UAS’s rotating blades. Exposed metal components, such as the camera and gimbal system, offer a negligible contribution to the UAS’s total RCS. The blades’ high rotational speed (approximately 1.500 rpm (while idling) to 3.500 rpm (during take-off)), compensates for the low RCS that plastic materials exhibit. The propeller’s size being smaller than the wavelength and of low dielectric constant, we use the Rayleigh approximation [20] to calculate the backscattering radar cross-section. For a single blade, the RCS (in m2) can be estimated aswith representing the blade’s volume and the relative permittivity of plastic. Considering the above parameters and the radar’s operating frequency at 24 GHz, the RCS calculation for a rotating blade results in or .
3. Results and Discussion
3.1. System Setup
It should be noted that before starting the main testing sessions, the whole configuration undergoes a calibration procedure to ensure the proper interconnection between subsystems and to obtain an estimation of the noise floor, as described in the following.
The radar system was deployed in a clear and obstacle-free area, with the antennas left unobstructed facing upwards. The sensitivity of the system was estimated to be at −100 dBm, i.e., the minimum signal strength the receiver is able to detect, distinguishable from the noise floor, which is a safe estimation, especially when taking into account the corresponding Doppler spectrum, indicated in Figure 4, when no UAS is flying. To avoid spectral leakage, a Hamming window is applied to the received signal. The calibration procedure refers to the process of ensuring that when no moving object is present, the radar returns are close to zero. During static mode, having estimated the noise floor, the noise profile is subtracted from the acquired signals in order to provide a more precise representation of the Doppler responses. In 360° scanning mode, full scans of a clear area are performed and corresponding Doppler frequencies (of maximum amplitude) are recorded. These frequencies correspond to radar signal returns originating from sources other than the “target” UAS. In this manner, they are disregarded (i.e., filtered out) when the radar is operating in its normal/detection mode.
A total of 5000 FFT points collected at a sample rate of 10,000 Hz, was found to provide sufficient frequency resolution for an effective signal processing of the proposed system.
In order to investigate and finally determine the various parameters of the returned radar signal, such as the appropriate sampling rate and the FFT points needed for sufficient frequency resolution as mentioned above, different UASs were used during the experimental sessions; these UASs varied from large, custom-made hexacopters shown in Figure 5(a) to small helicopters such as the Align T-Rex 550X DOMINATOR depicted in Figure 5(b). Those systems were significantly larger than the DJI Phantom 3 Advanced that we primarily used for testing. The DJI Phantom 3 Advanced model was the initial choice for conducting our experiments; the selection of this specific type of UAS was made due to its popularity and widespread use, especially by amateur users.
(a)
(b)
The UAS was then positioned to hover above the antennas in different altitudes. With its blades rotating on the plane defined by the x- and y-axes (perpendicular to z), the Doppler responses are minimal. An approximation of the resulting Doppler responses of the respective flying altitudes is given in Table 1.
The experimental tests were conducted at certain locations of the premises of the School of Electrical and Computer Engineering of the National Technical University of Athens, Greece, supported by the equipment and facilities of the Microwaves and Fiber Optics Lab and the Institute of Communication and Computer Systems.
3.2. Static Mode
To test the system’s detection capability in static mode, the UAS was flown over the radar following the flight paths, both in straight lines and X-shaped ones, as shown in Figure 6. The vertical distance (altitude, in z-axis) was increased by steps of 20 m to a maximum of 100 m due to the relevant flying height safety restrictions. Next, the UAS passed through the radar’s coverage volume (as shown in Figure 3); to ensure unbiased measurements, no specific care was taken for the drone’s flight path when crossing through the antennas’ main beam. By doing so, a more realistic approach of a possible detection scenario is achieved, as there is no guarantee that a target (UAS) would fly directly over the systems main beam.
(a)
(b)
The power spectrum of the received Doppler signal is plotted over time-frequency intensity charts, as illustrated in Figure 7, tracking responses over the UAS’s entire flight path. As it can be extracted from the same Figure, the main frequency content is located in the frequency region below 1.5 kHz.
A timeframe of the amplitude-frequency plot is given in Figure 8 where the captured FFT peaks of UAS flight over the radar system are shown. On the same plot, the “Sensitivity Level” and “Alarm Level” active cursors, appearing as horizontal lines, are set during the calibration phase, indicating the noise floor and the detection threshold, respectively. From Figure 8 it is seen that the dominant spectrum components are found located in the frequency region of 550 Hz. Thus, the Doppler frequency of , applied in equation (8) results in or . These values appear to be in accordance with the navigation data provided by the UAS remote control application in real time.
The UAS flight route over the radar was repeated 17 times, which is equivalent to over 2.5 hours of total flight time or in other words, an average individual session lasting approximately 9 minutes. No false alarms were recorded, despite the occasional bird presence in the area, which caused little to no disturbance, appearing in the form of clutter.
3.3. 360° Scanning Mode
In this mode, the radar was fitted with a rotating elliptical-shaped reflector, which steered the beam through a commercial stepper motor, making it parallel to the x-y plane as defined in Figure 2(b). To ensure the precise control of the reflector’s movement, the stepper motor was complemented with a zeroing switch that allowed for hardware verification of the reflector’s position after each full circle. The corresponding electronic circuits and a suitable power supply accompanied the above setup for the reflector’s rotation control. In this mode, the procedure followed to investigate the detection efficiency is outlined in Figure 9.
More specifically, using the radar position as the center, the UAS was flown following the perimeter of three circles with diameters of 100 m, 200 m, and 300 m, respectively. The relevant points are noticed as they are marked with an “X”; these were used to identify the cardinal points (S for south, E for east, N for north, W for west) which correspond to the 0°, 90°, 180°, and 270°, respectively. At each “X” point in every circle, the UAS halted its movement and hovered for 60 seconds (fixed positions). The rotation rate of the reflector was set to 0.25 Hz, thus illuminating 15 times the targeted UAS during its 60 seconds of hovering. Flight altitude was monitored via the UAS controller and ranged from 20 m to 60 m, in order to be close to the radar’s expected coverage area described in Figure 3.
An estimate of the noise profile was again calculated prior to the UAS flights, similar to the process followed during the static mode tests described previously. The radar was left unobstructed to perform full 360° scans of the area, void of any moving aerial objects. The resulting noise profile of this calibration process is shown in Figure 10(a). In this Figure, the polar plots of FFT peaks received by the radar are shown. Each closed line represents the FFT amplitude of the Doppler responses for each one of the four highest frequencies detected, calculated in 5-degree steps.
(a)
(b)
(c)
(d)
Figures 10(b) and 10(d) represent examples of the detection during the UAS’s route over the outer circle, (b) hovering at 150 m East (90°), (c) at 150 m North-East (107°) and (d) at 150 m North (180°). While there are noticeable spurious peaks present, most likely caused by clutter, the magnitude of the “returns” the UAS produces (returned signal echoes) remains 120 mV (over 5 dB) greater than those arbitrary peaks, ensuring reliable detection. It should be noted that, considering the low RCS of the specific UAS type, as this was discussed in the introduction section, the proposed systems detection ability allows for high levels of confidence.
3.4. Discussion
In the previous Sections, it was demonstrated that the proposed CW Doppler radar, offers advantages in terms of size, low complexity, and power-detection range. The suitability of the frequency band used for this type of applications was also confirmed through quite many on-site/field tests and corresponding results (outside laboratory-controlled conditions).
Table 2 provides an overview of some key features pertaining to radar systems oriented towards UAS detection, as discussed in the introduction section; at the same time, it allows for an effortless and straightforward comparison between the proposed system and other available implementations.
Using Table 2 as a reference, it is seen that compared to staring radar implementations (as in #2 and #3), the presented light-weight design requires relatively lower dwell times of the radar beam. Moreover, it does not depend on offline processing of the data collected, offering real-time visualization of the airspace inside the AoI, avoiding at the same time increased complexity issues (as in #6).
Size and power are kept to a minimum, limited to a mere fraction of other systems (for example, #5 and #8), while maintaining significant detection range, outperforming radars with similar output power (namely, #4, #7, and #11).
It also worth noting that, not all attributes-performance measurements of the systems under discussion were made available by the respective authors (#1, #10) or are provided partially (#9).
An apparent and unavoidable nuisance while conducting live measurement sessions was the occasional bird flights in the vicinity. During our work, birds flying by were a source of clutter that had a minor impact. The effects of these occurrences are outside the scope of this paper and are investigated in other works [21, 22]. However, it is an issue that can be studied in future work, along with weather conditions’ impact on the radar’s performance; fog and/or rain are generally expected to hinder microwave transmission, and exploring this relation requires further testing.
In static mode, the proposed system attained a remarkably consistent detection rate, with virtually none false alarm been detected throughout the series of extensive open field tests. Its ability to detect a small UAS (with an RCS of −20 dBsm or similar) inside its coverage volume was verified, indicating that the only impediment for even longer detection range is its low power. Although this can be mitigated with circuits providing higher output power, the overall size and weight of the radar system should not be radically changed to maintain low complexity, low volume, and low-cost configurations.
The CW design of the radar, compensates for relatively small dwell times (or “time on target”) achieved during 360° scanning mode. The continuous wave beam requires a shorter illumination period of the moving target, compared to the pulsed radar designs. Some seemingly arbitrary peaks of considerable magnitude have been observed, possibly due to (a) multipath propagation effects of undesired random signals present or (b) imperceptible jitter of the reflector originating from the stepper motor. Whatever the case may be, the cause of these peaks needs to be addressed accordingly in future configurations, along with the investigation of the parameters involved to mitigate the undesired signal returns during 360° scanning mode operation.
4. Conclusion
This paper introduced a K-band CW Doppler radar implementation, achieving reliable LSS UAS detection using two operational modes (static and 360° scanning), while keeping size, complexity and power levels at a minimum. Its easily deployment features allow for fast installation, providing successful detection of small aerial vehicles with relatively small dwell times and with a high degree of confidence for distances up to 150 m. The feasibility of extending the detection range by increasing the radar power can be considered as long as the overall small size and volume of the device is not radically altered.
Data Availability
The data used to support the findings of this study are included within the article.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments
The authors thank Mr. Athanasios Gidas, Mr. Michalis Sofras, Mr. Anastasios Garetsos, Dr. Evangelos Groumpas, and Dr. Spyros Athanasiadis for their contributions to this research. This work was supported in part by the H2020 RESISTO project, which has received funding from the European Union’s Horizon 2020 Research and Innovation Program under grant agreement (No. 786409).