Andreas Depold, Christian Dorn, Stefan Erhardt, Robert Weigel, Fabian Lurz
©SHUTTERSTOCK.COM/ROB ZS, DRONES— ©SHUTTERSTOCK.COM/KUBRA ASLAN ERTAP
Collapsing buildings are, while not exactly common, still an issue. The causes are diverse and often region dependent. The most common cause of the large-scale structural failure of buildings is earthquakes. In countries less affected by earthquakes, explosions caused by gas leaks are the leading cause.
In such a scenario, a quick response is paramount as the chance of survival of any victims buried under the rubble is directly correlated to the time it takes to recover them. After disaster has struck and search-and-rescue (SAR) teams arrive, the first step is to assess the situation and determine if the debris field is safe for the SAR personnel to transverse. If it is safe to traverse the site, search dogs are commonly utilized to locate victims. These dogs do an exceptional job; however, they are, of course, limited by endurance and external influences, such as weather. They also require a handler to be nearby, which forces people onto the rubble pile and therefore requires the area to be deemed safe by the safety assessment. In any case, this takes valuable time, and if the debris field is deemed unsafe, stabilizing efforts may take even longer.
Our research project, titled SORTIE and funded by the German Federal Ministry of Education and Research, aims to supplement current methods with a drone-based multifunctional system. Our subsystem targets mobile phones as an indicator of buried victims. The drone can be launched just minutes after the SAR team arrives. It then makes an initial flyover to map and photograph the stricken area, taking only a few more minutes. After mapping, the unmanned aerial vehicle (UAV) returns for a payload change, at which point the phone localization module is attached. During the next flyover, the system captures hundreds of data points for direction-of-arrival (DOA) estimation. These data are then further processed on a ground station (GS) to estimate the most promising locations for the SAR team to continue their search. All these drone measurements can happen in parallel with the traditional safety assessment of the rescue operation.
Utilizing UAVs in SAR is a topic that is currently being heavily researched. Ongoing research efforts include mapping an affected area [1] and automatic optical recognition [2].
This article is an extended version of the winning paper of the 2022 IEEE Radio and Wireless Week Student Paper Contest [3] and will focus mostly on the hardware design of the drone-based airborne station (AS) of the phone localization system. For it to function, a significant GS component is needed. A broader overview of the complete system is available in [4].
The presented system is based upon the Global System for Mobile Communication (GSMTM). While GSM is widely used around the world, it isn’t used everywhere, and considering its age, its global coverage will likely decline. This isn’t an issue for our system, though, as no existing network is required. GSM was chosen for the proof of concept because it is supported by virtually all mobile phones. In particular, the E-GSM 900 band was selected as its relatively low frequency results in lower attenuation [5] through typical building materials, and it has been widely adopted for handsets. The low channel bandwidth of 200 kHz combined with the high maximum output power of 33 dBm [6] results in a high signal-to-noise ratio (SNR), facilitating the reception through the rubble piles of collapsed buildings. There is no need to support any other radio access technology in the AS because the GS forces all phones onto our GSM network.
GSM is a time-division-based multiple access system that utilizes short 577-${\mu}$s bursts. The E-GSM 900 band offers a total uplink bandwidth of 35 MHz. The phone’s exact transmission time and frequency are controlled by a base transceiver station (BTS).
Multiple localization systems based on GSM exist, though they are generally based around multiple existing BTSs and must be supported by the carrier. These systems are not precise enough for SAR, and their need for an existing functional infrastructure makes them less than ideal for disaster-stricken areas [7].
The ground-based component consists of three elements: a smart jammer capable of intelligently disrupting any available mobile phone network; a GSM BTS; and a general-purpose server for data postprocessing and visualization. To locate a phone, we need to know exactly when and at what frequency it will be transmitting. The only way to achieve this is by having the phone connected to a BTS that is under our control. As working with the phone carriers is not feasible on short notice, and local networks may not even be available depending on the cause of the building collapse, we need to supply this network ourselves. Phones will usually connect only to their carrier’s preferred network, unless it is not available—then any network will do—to keep emergency call functionality.
We utilize this by jamming all available networks, thus forcing the phones in the area to connect to our BTS. Once they are connected, we can effectively direct them to transmit however and whenever necessary. Jamming can be accomplished in multiple ways. The simplest method saturates every band entirely with noise, reducing the phones’ and base station’s received SNR. This requires a large amount of output power to guarantee jamming success. In our case, we use a more power-efficient smart-jamming technique, which selectively disrupts the broadcast channels’ synchronization sequences. The exact method is then dependent on the type of network that is used. Our predecessor project, ILOV, demonstrated a simpler jamming technique in [8] and [9], which selectively saturated only the cell towers’ broadcast frequencies with noise.
The AS consists of an antenna array; multichannel phase-coherent receivers; synchronization hardware for the BTS; and a general-purpose processing system. A block diagram is shown in Figure 1. The processing system is permanently mounted to the drone and shared for multiple uses, while the phone localization system is a removable payload. Only the removable portion of the AS will therefore be considered for determining weight and power consumption.
Figure 1. A block diagram of the UAV-based phone localization system. RX: receiver; UL: Up Link; ADC: analog-to-digital converter; FPGA: field-programmable gate array; MCU: microcontroller unit; LVDS: low-voltage differential signaling; PCIe, Peripheral Component Interconnect Express; TRX: tranceiver.
The receive system consists of nine phase-coherent receivers; a field-programmable gate array (FPGA); and any necessary peripherals. The prototype utilizes a custom board with an Artix-7 FPGA; 512 MB of DDR3L RAM; and an FT600 USB3 SuperSpeed FIFO bridge for digital data handling as well as a local oscillator. Three triple-channel receiver boards are mounted through high-speed interconnects. The assembly is shown in Figure 2. The receivers offer a 100-million-searches-per-second sampling rate and two-stage surface acoustic wave (SAW) band selection filtering. This allows the capture of the complete 35-MHz E-GSM 900 uplink band. The SAW filters can be replaced to allow capture of the SRD860 to enable testing at 868 MHz. While we don’t require the complete band, as a single GSM channel is only 200 kHz wide, capturing the entirety allows us to extract the highest and lowest channel with a single measurement. A more in-depth look at this can be found in [10].
Figure 2. A photo of the prototype receive system mounted to the antenna reflector. LO: local oscillator.
Each of the antenna elements consists of two orthogonal dipoles. Switching between the two is accomplished by applying different levels of dc offset to the RF line through a switch and some diode logic. This allows switching between two polarities and disabling the antennas without any additional control lines. A close-up view of the feed circuit board is shown in Figure 3.
Figure 3. (a) Front and (b) back photos of a single antenna element showing the switching circuitry.
The array consists of nine of these double elements, spaced by half the wavelength. A reflector is placed a quarter of the wavelength behind the elements. The reflector is made of welded steel mesh stiffened by carbon fiber rods. This allows easy and fast prototyping, although it is relatively heavy.
Figure 4 shows an early prototype for system evaluation. As pictured, it is not mounted to a UAV but was utilized to verify the electrical system and as a mechanical pathfinder for future more integrated construction. To allow standalone operation without information from the flight computer, a GPS real-time kinematic receiver for precise relative positioning and an inertial measurement unit are mounted to the back of the array. Data are acquired by connecting a laptop.
Figure 4. The complete prototype assembly, including the GPS receiver and antenna.
The receivers must be synchronized to the transmitting phones to optimize power consumption and to differentiate among the phones as the received signal is not decoded. Fortunately, as described in the “GS” section, our own BTS controls the phones. The GS, therefore, has the exact timing and scheduling information. This information is transmitted to the airborne component via a low-speed—but robust—packet-based 433-MHz link. The exact time at which data acquisition must start is communicated by a special triggering packet. Reception of the synchronization word of this packet automatically triggers the FPGA’s capture logic without involving any further software. This same link can be used to transmit the GPS–real-time kinematic positioning correction data for standalone operation.
The average power consumption of the airborne system must be kept relatively low. This requirement does not stem from any battery capacity limitations as it will be powered by the flight battery, and any energy usage pales in comparison to the power needed to keep the UAV in flight. Instead, the requirement is related to heat load as any cooling system would either be relatively heavy or—if designed to utilize the rotor’s downwash—increase drag and therefore cause an efficiency loss in the flight system.
Each of the three receiver modules draws around 3 W of power when active, mostly dominated by the IF amplifiers. Fortunately, the duty cycle can be low. If there are 10 phones in the target area, and we consider a 2-Hz measurement rate for all phones at both ends of the band to be sufficient, this ideally requires only 2*2*10 = 40 timeslots, each of which is 577 µs long. If we assume an excessively large overhead factor of 10, caused by factors such as suboptimal scheduling or switching on-and-off time, this results in a worst-case duty cycle of slightly above 11%. Including the approximately 4 W of the digital and peripheral circuits, the total average power draw is below 5 W, which can be sufficiently sunk by the circuit board without any additional cooling as long as it isn’t fully enclosed.
As mentioned in the previous section, weight is a tight constraint. Additional mass reduces flight time and, depending on distribution, negatively affects the flight dynamics. The size of this system’s antenna array already forces efficiency loss due to the reflector being hit by the rotor’s downwash. This effect must be minimized by reducing the air resistance of the reflector, but it can’t be entirely eliminated. This further reduces the maximum weight of the system.
While the setup shown in Figure 4 is a relatively heavy early prototype, the later system has a mass target of below 1 kg, including integrated landing gear. This will be achieved by utilizing a laser-cut aluminum reflector; glass and carbon fiber rods; and dozens of 3D-printed parts. The antenna count will be reduced to eight by eliminating the center element. This not only greatly reduces the complexity of the receivers, as the analog-to-digital converters (ADCs) used are quad channel, but also frees a convenient mounting position for the 433-MHz synchronization antenna at the center of the array.
Rubble piles present a complicated environment for radio propagation. They can be extremely irregular, consisting of wood, steel, reinforced concrete, and other materials in random orientations and chunk sizes. All these effectively random materials and surface orientations cause a great deal of signal reflections. Signal attenuation can also vary greatly among different paths. According to a calculation in [11], a collapsed duplex house can cause up to 100 dB of additional path loss, without considering free space loss.
These issues can’t be fixed or nullified. Instead, we try to gather as much information as possible to work out which of our measurements correlate well and converge onto a plausible location. We fully expect many individual measurement points to be faulty. Angle estimations based on false paths show more spread in their results, so they should not converge the same way as more direct paths.
As described in the “Receive System” and “Antenna Elements” sections, diversity is achieved in two ways: through switching of the antenna polarization and by measuring at two frequencies, totaling to four somewhat decorrelated measurement points. Testing at the two intended frequencies—880.2 and 914.8 MHz—is not possible without regulatory approval as we can’t randomly broadcast in a cellular network band, so all tests thus far have been conducted at the single frequency of 868 MHz.
Figure 5(a) and (b) shows the spatial spectra of two consecutive measurements from basically the same location with the two polarization configurations. These measurements are based on a single 600-${\mu}$s snapshot and were taken at a Technisches Hilfswerk (German Federal Agency for Technical Relief) training site with the transmitter hidden in a concrete pipe covered by rubble. The array was mounted to a pole at a height of about 2 m and carried around the site. Angle estimation was done through the minimum variance distortionless response (MVDR) beamformer algorithm [12]. It is clear that both measurements show some ambiguity and don’t fully align. Combining these two results by multiplying the spatial spectra results in Figure 5(c) shows a clean ambiguity-free result. MVDR offers a higher resolution than other beamformer-based algorithms, without the computational burden of subspace-based methods like multiple signal classification (MUSIC). Further improvement is expected through using more advanced algorithms as MVDR is not particularly suited for signals with correlated noise. Exploiting the known properties of a GSM burst could also lead to great improvements, as shown in [13].
Figure 5. (a) Polarization 1. (b) Polarization 2 (c) The combined result. (a) and (b) Show the spatial spectra of two real measurements estimated through MVDR. (c) Shows the multiplicative combination of the two. Deg.: degree.
The presented system is a stepping stone to a more final drone-integrated version that can be used to locate buried victims using their cell phones. It shows the validity of the project and served as a mechanical and electrical pathfinder for further development. Future revisions will be far more refined, lighter, and more energy efficient.
We aren’t ready to go into detail yet, but Figure 6 shows a picture of the next (far more refined) prototype mounted to a UAV.
Figure 6. The next prototype with integral landing gear mounted to a UAV.
This work is funded by the Federal Ministry of Education and Research (BMBF) of Germany as part of SORTIE under Project 13N15191.
[1] M. Mittal, R. Mohan, W. Burgard, and A. Valada, “Vision-based autonomous UAV navigation and landing for urban search and rescue,” in The International Symposium of Robotics Research, T. Asfour, E. Yoshida, J. Park, H. Christensen, and O. Khatib, Eds. Cham, Switzerland: Springer Nature Switzerland AG, Oct. 6, 2019, pp. 575–592.
[2] S. Wang, Y. Han, J. Chen, Z. Zhang, G. Wang, and N. Du, “A deep-learning-based sea search and rescue algorithm by UAV remote sensing,” in Proc. IEEE CSAA Guid., Navig. Contr. Conf. (CGNCC), 2018, pp. 1–5, doi: 10.1109/GNCC42960.2018.9019134.
[3] A. Depold, C. Dorn, S. Erhardt, R. Weigel, and F. Lurz, “A light weight 3x3 switched polarity URA antenna and receive system for direction finding,” in Proc. IEEE Topical Conf. Wireless Sens. Sens. Netw. (WiSNeT), 2022, pp. 36–39, doi: 10.1109/WiSNet53095.2022.9721370.
[4] C. Dorn, A. Depold, F. Lurz, S. Erhardt, and A. Hagelauer, “UAV-based localization of mobile phones for search and rescue applications,” in Proc. IEEE 22nd Annu. Wireless Microw. Technol. Conf. (WAMICON), 2022, pp. 1–4, doi: 10.1109/WAMICON53991.2022.9786189.
[5] L. Chen, T. Ostertag, M. Loschonsky, and L. M. Reindl, “Measurement of mobile radio propagation channel in ruins,” in Proc. IEEE Int. Conf. Wireless Commun., Netw. Inf. Security, 2010, pp. 252–256, doi: 10.1109/WCINS.2010.5541724.
[6] “GSM technical specification; digital cellular telecommunications system (Phase 2+); radio transmission and reception (GSM 05.05),” ETSI, Sophia Antipolis, France, 1996. [Online] . Available: https://www.etsi.org/deliver/etsigts/05/0505/05.00.0060/gsmts0505v050000p.pdf
[7] T. Kos, M. Grgic, and G. Sisul, “Mobile user positioning in GSM/UMTS cellular networks,” in Proc. ELMAR, 2006, pp. 185–188, doi: 10.1109/ELMAR.2006.329545.
[8] S. Zorn, M. Maser, A. Goetz, R. Rose, and R. Weigel, “A power saving jamming system for E-GSM900 and DCS1800 cellular phone networks for search and rescue applications,” in Proc. IEEE Topical Conf. Wireless Sens. Sens. Netw., 2011, pp. 33–36, doi: 10.1109/WISNET.2011.5725020.
[9] S. Zorn, M. Gardill, R. Rose, A. Goetz, R. Weigel, and A. Koelpin, “A smart jamming system for UMTS/WCDMA cellular phone networks for search and rescue applications,” in Proc. IEEE/MTT-S Int. Microw. Symp. Dig., 2012, pp. 1–3, doi: 10.1109/MWSYM.2012.6257769.
[10] A. Depold, C. Dorn, S. Erhardt, R. Weigel, and F. Lurz, “A 9-channel phase coherent receive system for direction of arrival estimation,” in Proc. 51st Eur. Microw. Conf. (EuMC), 2022, pp. 184–187, doi: 10.23919/EuMC50147.2022.9784309.
[11] S. Zorn, G. Bozsik, R. Rose, A. Goetz, R. Weigel, and A. Koelpin, “A power sensor unit for the localization of GSM mobile phones for search and rescue applications,” in Proc. IEEE Sens., 2011, pp. 1301–1304, doi: 10.1109/ICSENS.2011.6127162.
[12] H. L. Van Trees, Optimum Array Processing: Part IV of Detection, Estimation, and Modulation Theory. New York, NY, USA: Wiley, 2002, pp. 440–443.
[13] M. Cedervall and R. L. Moses, “Efficient maximum likelihood DOA estimation for signals with known waveforms in the presence of multipath,” IEEE Trans. Signal Process., vol. 45, no. 3, pp. 808–811, Mar. 1997, doi: 10.1109/78.558512.
Digital Object Identifier 10.1109/MMM.2022.3226548