Potential of Passive DVB-T Radar Component Against Illegal UAV Flights

,

In a first preliminary part, a short description of the different compulsory functions will be given. These definitions are required as they may differ from one kind of sensor to another. Due to the diversity between the sensor specialties, these definitions could not be universal but will be used within this document. The main sensor types that may contribute to each function will be mentioned.
The rest of the paper will be dedicated to radar component and more especially to passive radar component and its potential interest.
In particular, the second introductory part will compare the active radar approach with the passive radar one.
Then the principle and constraints of a passive DVB-T radar component will be reminded and discussed in order to introduce its potential interest for countering UAVs.
The last chapter will illustrate passive DVB-T capabilities, using some of the most meaningful experimental results. The interest of such a sensor for the different functions required for countering UAVs will be discussed.

Compulsory functions of a counter-UAV system
The document will consider the different following compulsory functions for a counter-UAV system: • Detection: the objective is to decide, among other potential 'targets' , about the existence of an UAV within a large angular and range sector.
• Localization: the UAV (among the other detected targets) has to be localized according to the receiving system or more globally its Cartesian coordinates have to be evaluated with respect to the area of interest.
• Tracking: the UAV and its kinematic features have to be evaluated by a tracking approach using the different localization measurements.
• Extraction: the UAV has to be extracted from the other targets with no interest and this UAV has to be tagged as a potential threat.
• Identification: the UAV may be identified according to its size, its potential dangerousness, its category (fixed wing or multirotor).
• Alert: the UAV has to be classified as a real threat or not. It could be achieved, for example, by localizing an identified UAV within a forbidden area.
• Neutralization: the UAV has to be neutralized in order to avoid the attack or even the simple intrusive incursion.
Among the main types of sensor, we may mention the following: • Goniometric components: these sensors estimate the direction of arrival of an electromagnetic source within their bandwidth of survey. They generally look for AI, Computer Science and Robotics Technology 2/18 some video flux or typical links between the UAV and its pilot. Then an association of different non-collocated goniometric components will allow the localization of the UAV. Finally such sensors contribute to the detection, localization and extraction (that could even be achieved directly using the signal identification and so before the detection phase) for UAVs which are transmitting characteristic signals.
• Radar components: these sensors estimate parameters of the slow moving targets within their environment. These parameters are typically range, radial velocity and angle(s) of arrival (azimuth and elevation for two-dimensional array antennas). Then these parameters may be used for localization and tracking.
• Acoustic components: these acoustic sensors are able to detect the noise sources and estimate their azimuth and elevation directions.
• Optical components: such components are generally considered for target identification even if some emerging solutions began to be studied for detection.
Most of the existing -or under development-systems combine the advantages (and drawbacks) of the different technologies in order to reach the final objective before neutralization: detect, localize, track, extract, and identify the UAV from all the other potential 'targets' within the area of interest.
However, it is noticeable that multiplying the number of sensors is not the optimal solution in terms of cost nor efficiency. For example, a simple radar-based system for detection, localization and tracking combined with an optical component for UAV identification and extraction, may not be fully efficient in terms of reactivity. Assuming a radar with high sensitivity but unable to limit the number of tracks of no interest, this radar would systematically ask for a confirmation/ identification to the optical means, and such a systematical request is not efficient according to a reactivity criteria. This simple example outlines the interest of sensors mainly dedicated to detection and localization but with some classification capabilities.

Radar objectives
The main missions for a radar sensor in such a counter-UAV context are the following: • Detection and surveillance Radar should ensure a continuous surveillance of a wide angular sector with a high data renewal rate of the target parameters. The targets that should be detected are more specifically small targets at low altitude and low speed. The radar should also be efficient when facing multiple simultaneous attacks.
AI, Computer Science and Robotics Technology 3/18 • Localization and Tracking Radar should restitute the Cartesian coordinates of the targets with respect to the area of interest.

• Identification/extraction
As mentioned at the end of previous paragraph, a first step of identification at the radar level will lead to a substantial gain for a global system of alert.
Consequently, an efficient radar should, ideally, have some identification capabilities or at least should avoid a systematic request to the identification component for all the detected targets.
From a footprint and deployment point of view, the radar component should be compatible with a fast installation on most of the area of interest. These criteria are scenario-sensitive as, for example, for specific long-planned events, an installation within one day could be considered as fast enough. However, a system requiring huge host infrastructure due to its wide footprint and consequent weight would be more complicated to deploy.

Battle field radar
Among the potential radar solutions, Battle field Radars [1, 2] are dedicated to terrestrial targets detection and tracking for targets ranging from pedestrians to tanks. Consequently, these radar sensors have been evaluated against UAVs. Such systems are generally X-band radar with a carrier frequency close to 10 GHz, the angular the monitored sector is close to 120°and the data renewal rate, for each target/direction, is typically close to 1.5 s. All of these active systems scan the 120°s ector by focusing the energy with a typical angular step close to 1.5°. This focalization could be achieved mechanically (rotating antenna), electronically or numerically (by applying dedicated differential phases between the antennas of the transmitting array). These parameters imply a duration of illumination per scanned direction close to the ratio between the data renewal rate (here 1.5 s) and the number of directions (here let us say eighty), so close to 20 ms per direction or per expected target. However, as it will be explained later, for radar systems, it is, sometimes, possible under some hypothesis to detect the period of modulation induced by the blade rotation. As a typical blade rotation speed is between 6000 and 9000 rotations per minute, corresponding to a frequency between 100 and 150 Hz. A coherent integration time of 20 ms leads to a frequency resolution of 50 Hz.
Consequently, such a frequency resolution is generally not sufficient for detecting the blade modulation which is only twice or three times the resolution (furthermore the UAV body contribution has a higher level than the blade modulation impact).
Consequently, the discrimination/identification capabilities of such a radar component against UAV may be limited in practice. Wi-Fi [8,9] for more local applications.

AI, Computer Science and Robotics
These broadcasters of opportunity present the following compulsory advantages for radar purpose: constant illumination, quite omnidirectional coverage, illumination ensured at the ground level. The basic information such as frequency and the main signal parameters are known. The locations of these radio-television civilian broadcasters are known, and these services generally offer a good density of transmitters. It has been evaluated that for countering UAV, among FM, DAB and DVB-T transmitters, DVB-T transmitters seem to be the most promising ones.
Furthermore as the table 1 illustrates, DVB-T has the highest carrier frequency and the highest useful bandwidth among these three candidates.
In this chapter, the principle of a passive radar will not be detailed, only the main differences between typical passive radar approaches and the passive DVB-T sensor that will be detailed in the rest of this paper will be considered.  initialization based on 'bistatic range triangulation' [10], then the track itself may be maintained with only two bistatic couples and up to one. However, as this initialization phase requires a simultaneous detection over three different bistatic couples, the domain of localization is lower than the union of the three bistatic domains of detection.
The system considered in the rest of this document, is based on a sectorial coverage due to sectorial receiving antennas. This system was studied for critical infrastructure that could not consider omnidirectional sensors due to the complexity of the buildings infrastructure of the site to be protected. Then the omnidirectional protection will be based on the association of several of these sectorial sensors. Among the advantages of such a sectorial approach, we may mention the ability of managing all the antenna resources for a given sector in order to obtain a good angular resolution for each sector (always considering 8 antennas as It is important to notice that this COFDM signal is called COFDM with guard interval (named Δ). Each symbol duration is higher than the useful duration T u , T u is the useful duration on which the transmitted sinusoids are orthogonal according to the Fourier transform over the duration T u . This guard interval goal is to absorb potential interferences between the different frequencies and the different symbols for propagation channels with a length (delay between the first path and the last contributors) lower than the guard interval duration.
For example, for a propagation channel with a length L lower than Δ, it is possible to consider after synchronization on the first path, symbols at the receiving level that may be described as follows H k m is characterizing the propagation channel at frequency k and for symbol m.
As the previous expression is valid for duration equal to the difference between However, due to the delays between the different fixed echoes, this propagation channel coefficient is highly fluctuating with the frequency k.
Finally, the signal processing after the reception of COFDM signals will be based [11,12] on the main following steps: • A sampling component that digitalizes the signals after Hilbert filtering (so I and Q components). The system is able to process (see figure 1) the signals in real time while ensuring simultaneous records of the raw data for offline analysis.
Furthermore, as the angle of arrival (azimuth) is estimated with sufficient accuracy, a direct geometrical transformation of the (bistatic range, azimuth estimation) allows a bistatic localization according to the receiver location (range, azimuth estimation) or Cartesian (X,Y) coordinates.

Conclusions and advantages
Like the other passive systems, such a DVB-T passive sectorial component is able to cope with the main direct path and clutter limitations, which is a crucial step to detect targets at low altitude (so evolving in front of clutter contributors).
Due to the sectorial coverage with a dedicated 8-antenna array, it is possible to reach a good accuracy in azimuth domain and such a good azimuth precision combined with a good bistatic range resolution allows a direct bistatic localization by simple geometrical transformation between bistatic domain to Cartesian one.
AI, Computer Science and Robotics Technology 11/18

Short description of the bistatic configurations several ones
The different results presented hereafter were obtained under various bistatic conditions that are not mentioned for each individual result.
Generally in France, quite a limited number of powerful DVB-T transmitters may be used for a given receiver location: let say that over the different receiver locations evaluated the number of DVB-T transmitters of interest was between one and three.
According to their power and to the configuration, the range between the transmitter and the receiver varied from 30 km up to 110 km (transmitter installed on the top of a mountain.

Blade modulation detection: stationary quadrirotor and multirotor extraction
The typical coherent integration time for DVB-T component against UAV is between 0.5 s up to roughly 1 s. As mentioned previously, this is achievable as the transmitters are quite omnidirectional with a constant illumination and the sector to be protected is simply scanned numerically using our 8 antenna receiving array. All the targets within our sector of surveillance may be analyzed simultaneously (according to the digital beamforming efficiency). Such a long coherent integration time allows the detection of the blade modulation impact when the corresponding level is sufficient enough for ensuring its detection. This level depends on the blade material, the blade size,… but the conditions will not be discussed during this paper.
The blade modulation detections illustrated on figure 4 are corresponding to a stationary F450 (small quadrotor) at three kilometers away from the receiver. As the multirotor was in stationary configuration, it was impossible to detect the UAV itself while its blade modulation around 120-130 Hz were easily detected, the different modulation frequencies are corresponding to the different rotors.
The figure 5 clearly illustrates that numerous targets may be detected, but quite a few of them are presenting a behavior corresponding to target with blade modulation signature. The right part of figure 5 illustrates that the UAV had such a behavior and a small part of a 'trajectory' close to 6000 m bistatic range at the end of the record is also presenting such a behavior, this was probably due to a small aircraft with a frontal propeller.
The bistatic configuration was similar to the stationary flight illustrated on figure 4 but the small UAV was now achieving some small circles.
AI, Computer Science and Robotics Technology 12/18  The frontal propeller impact was detected as soon as the fixed wing was facing the receiver even during the two loops. Furthermore, for closest configurations, the blade modulation effect was also detected when the UAV was no longer facing the receiver.
AI, Computer Science and Robotics Technology 13/18  Furthermore, these results also illustrate the need for a first 'classification' capability at the sensor level: the 'isolated' false alarms such as the ones at low ranges have to be avoided using, for example, a tracking function and ideally the coherent 'tracks' at the end have to classified as 'none interesting targets' as soon as possible. The next paragraph will suggest some possibilities for cancelling unwanted tracks occurring on roads.

Road detections rejection
As mentioned previously, passive DVB-T radars present interesting properties for UAV detection and localization. Nevertheless, it is of great importance to avoid systematic detections due to unwanted targets, such as terrestrial vehicles on roads.
A simple way to limit a road impact is the following: each road will generally lead to numerous detections with (range, angle) histories that are characteristic of the geometrical configuration of this road according to the bistatic geometry. Once the accumulated (range, angle) domain corresponding to the numerous detections have been determined, it is possible to consider masks over these specific (range, angle) domains. Sometimes, it could also be possible to add the influence of the Doppler but this parameter may vary according to the car location along the road (traffic lights) or along the time domain (traffic jam) so managing this Doppler parameter is more complex for characterizing road detections. Figure 10 presents, on the left, the bistatic range histories of all the detected targets within 3000 and 9000 km, and on the right the remaining targets after the application of a (range, angle) mask corresponding to the accumulated plots along the road.
AI, Computer Science and Robotics Technology 16/18

Conclusion and future works
Countering illegal UAV flights is a complex challenge as the threat is in constant evolution. Furthermore, this struggle is highly asymmetric as UAV are low cost threats that could be easily deployed under various environments. Consequently it is crucial to evaluate anti-UAV system components with high efficiency and low sensitivity to the context.
The passive DVB-T radar component seems to offer interesting performances for both its detection capabilities and its Cartesian localization efficiency (even under simple bistatic configuration). Furthermore, it may offer some promising classification capabilities, for characterizing the UAV itself or for avoiding detections of unwanted targets. However up to now, this solution has mostly been evaluated for the protection of isolated infrastructures and its behavior in more complex environments needs to be confirmed.
The future works will mainly consist in the evaluation of such a DVB-T passive radar under more complex environments such as peri-urban ones. In parallel, the classification capabilities have to be enhanced in order to limit the number of systematic requests to an identification component such an optical one.