Time difference of arrival based passive sensing and positioning system integrated into moving platforms

Abstract

The ability to locate the source of information has been a fundamental need for humanity since its earliest days. This necessity even explains why living beings have evolved to have two ears: to determine the position of a sound source, which is a form of information. As sound propagates, it reaches our ears at different distances, creating an interaural time difference (ITD). The brain uses this time difference, along with the intensity of the sound, to estimate the location of the sound source. This thesis explores the principle of localization using ITD, which has existed in nature for millions of years, from the perspective of a communications engineer. Specifically, it examines the application of this principle to locate a source emitting radio frequencies. There are various methods for locating a signal source. These methods primarily rely on the signal's strength, arrival time, frequency, phase, or a combination of these factors to determine the position. The strength of the incoming signal depends directly on the terrain, the signal's frequency, and its output power. Since these variables are often unpredictable, using signal strength for positioning usually yields low accuracy results. Measuring the signal's phase requires multiple antennas and RF stages, which can only estimate the target's angle, not its precise location. However, by determining the arrival times of signals received by multiple receivers and analyzing the time differences, the position of the target can be estimated. To solve for the unknown x, y, and z coordinates in a 3D space using only time difference of arrival (TDOA) information, at least three equations are necessary. These equations provide the minimum amount of information required for position determination. However, obtaining three linearly independent TDOA equations necessitates a minimum of four receivers. Among these receivers, one is designated as the reference, and the time differences between this reference and the other three receivers are used to create linearly independent equations. These equations are then utilized to determine the target's position. However, because the equations are typically nonlinear, achieving a quick and highly accurate solution is not always straightforward. Additionally, factors such as hardware imperfections and noise can prevent a clear solution to the equation system. Various methods can be employed to address these challenges and improve the accuracy of the results. This study compares algebraic methods such as Least Squares (LS) and heuristic methods like Particle Swarm Optimization (PSO) for signal source localization. LS methods solve the system of equations directly to estimate the target position, while PSO methods optimize a target function to find the best location. Heuristic methods, including PSO, can yield effective results even with nonlinear equations or in noisy environments. In this research, we utilized a variant of the PSO algorithm known as the Firefly Algorithm. The Firefly Algorithm begins by distributing fireflies randomly across a cost function map. The fireflies move towards the solution with the lowest cost, switching to the new best fireflies as lower-cost solutions are found. This approach is advantageous for several reasons: it uses an infinite number of TDOA measurements rather than just three equations, minimizes the likelihood of getting stuck in local minima on the cost map, and achieves high-accuracy localization. Although the Firefly Algorithm requires more computational power compared to algebraic solutions, modern computers can handle this demand effectively. While signal source localization in a 3D environment using time difference of arrival (TDOA) information has often been tested with a 4-receiver system model, successful localization can also be achieved with different system models. Unlike traditional methods, where TDOA data is collected simultaneously from fixed receivers, we propose a system where two receivers are moved to collect TDOA data at different time instances, followed by localization using the collected data. Practical issues encountered with such a system model were investigated through simulation and measurement setups. One challenge was accurately estimating the time differences of arrival of signals received by the receivers. Due to the slow variation of signals in time, time estimation is affected by noise. Another potential problem arises when the sampling frequency of the system is narrower than the signal's bandwidth, causing the cross-correlation of received signals to not yield peak values at the delay samples, making time differences difficult to discern. To address this, we decided to operate the system with the highest possible sampling rate when the bandwidth of the target signal is unknown. Ensuring reliable signal sampling, both receivers are synchronized to the same frequency and time using a GPS-disciplined oscillator. Furthermore, 1 Pulse Per Second (PPS) from GPS is used for time synchronization. Apart from these technical considerations, the trajectory of the receiver stations plays a crucial role in system performance. As the distance between the target receivers increases, so does the distance they need to cover for accurate localization. Additionally, ensuring a high-reliability and high-capacity communication network between receiver stations and the base unit is crucial during system implementation. Without this network, communication disruptions between the receivers and the base station would prevent TDOA data collection and, consequently, localization algorithms from functioning. Lastly, challenges were observed when there are multiple sources emitting signals at the same frequency or when environmental factors cause signal reflections and changes in direction, affecting TDOA measurements by the receivers. Before finalizing the system setup, creating a realistic simulation environment is crucial. In the fourth section, we introduced a simulation environment designed in MATLAB to anticipate potential scenarios before the measurement setup. The simulation environment was designed to be consistent with real measurements, including terrain features using the WGS84 geolocation method. Since the system was anticipated to be tested in the TÜBİTAK-BİLGEM Gebze campus area, tests in the simulation environment were conducted accordingly. After creating the simulation environment, the first test was to examine the hyperbolas formed when different receiver paths were created. It was observed that when the target was far from the receiver paths, the hyperbolas intersected each other over a wide area. Conversely, when a receiver rotated around the target, the hyperbolas intersected each other from all directions within a small area. A small intersection area of hyperbolas is crucial for the successful operation of localization algorithms like the firefly algorithm. Secondly, the formation of hyperbolas was observed when the bandwidth of the target signal and the system's sampling frequency were reduced. With low sampling frequency, the resolution of hyperbolas was significantly reduced, spreading over a very small space. When the bandwidth dropped below a few hundred MHz, the hyperbolas generally did not pass near the target. As the sampling frequency and bandwidth increased, the hyperbolas gradually approached the target and began to intersect over it. In the third simulation, a cost function was generated for the firefly algorithm, and costs in the solution space for different receiver paths were examined. As expected, as the target moved away from the receivers, the slope of the cost function around the target decreased, allowing for wider areas to be estimated as solutions. Based on these simulations, two suitable options for receiver paths were identified for the TÜBİTAK-BİLGEM Gebze campus. Finally, the average error in position determination was investigated for different sampling frequencies of sampled target signals for the identified two paths using the firefly algorithm. As expected, errors were significantly higher at low sampling frequencies, decreasing as the sampling frequency increased. The fifth chapter illustrates the measurement setup devised from the insights gleaned from simulations, deductions, and experiences presented thus far in the thesis. It starts by delineating the hardware and software of the ground station, followed by those of the receiver units, and then narrates the measurements encompassing various scenarios. The ground station hardware comprises a simple setup, consisting of a powerful computer and a modem supporting Wi-Fi 6 for communication with the receiver unit. The user interface software enables control of the receiver units from the ground station, allowing adjustments to frequency bandwidth and gain configurations. Additionally, signals received by each unit can be individually represented in time and frequency space for adjusting gains to account for signal visibility variations. Moreover, the interface facilitates data collection, time difference calculation, and execution of the firefly algorithm, with results visualized on a map. The hardware design of the receiver unit has been the most multidisciplinary aspect, given its need for lightweight deployment on UAVs while meeting power requirements throughout the flight. The design considerations for the receiver unit include power needs for communication, computation, and GPS, with antennas strategically positioned on the UAV. Extensive efforts resulted in reducing the system weight to below 3 kilograms when integrated with the protective casing. The software running on the receiver unit operates at a lower level compared to that on the ground station, directly transferring GPS-derived position, time, and frequency information to the SDR and computer hardware, then transmitting it to the ground station via Wi-Fi 6 using the SSH interface. At the TÜBİTAK-BİLGEM Gebze campus test site, various measurements were conducted to evaluate the performance of the system hardware and software. Two receivers were mounted on drone and ground vehicle setups for different test scenarios. Different signal types and receiver paths were tested. Initially, to assess system performance under optimal conditions, a 20 MHz bandwidth high-autocorrelation M-sequence signal was transmitted from a vector signal generator, attempting to locate it with drones. Subsequently, a 20 MHz bandwidth LTE downlink signal was examined. In the third measurement, the focus shifted to existing real LTE signal sources after discontinuing the use of the signal generator. The fourth measurement pushed system boundaries by utilizing a ground vehicle and stationary receiver to locate a narrowband and intermittently available LTE uplink signal. The system performed better than expected, locating the LTE uplink signal source with a 12-meter margin of error. In the final measurement, a pulse-type modulation radar was positioned to test the system's applicability in military settings. In conclusion, this thesis demonstrated the integration of two RF receivers utilizing the TDOA principle onto drones. Simulation environments were initially created to examine system performance, followed by the implementation of the system and localization of various signal sources. These efforts illustrated that the localization accuracy varies based on the type of radio signals emitted and the trajectories followed by the drones. Moreover, the feasibility of performing localization by placing TDOA-based receivers on moving units was established.