Real-Time Passive Acoustic Tracking of Underwater Vehicles

FACULDADE DE ENGENHARIA DA UNIVERSIDADE DO PORTO Real-Time Passive Acoustic Tracking of Underwater Vehicles José Francisco Ramos Valente DISSERTATION THESIS MASTER IN ELECTRICAL AND COMPUTERS ENGINEERING Supervisor: Prof. Dr. José Carlos Santos Alves Co-Supervisor: Prof. Dr. Nuno Alexandre Lopes Moreira da Cruz July 26, 2016 c José Francisco Valente, 2016 Resumo Com o crescente interesse na exploração oceânica, os sistemas robóticos marinhos estão a ganhar uma nova importância em diversos campos científicos e domínios industriais. Neste contexto, torna-se evidente que os Veículos Submarinos Autónomos (AUV - Autonomous Underwater Vehicles) são instrumentos relevantes para atividades de investigação no mar, devido à sua versatilidade para diferentes missões, combinados com custos baixos de fabrico, operação e manutenção, quando comparados com veículos submarinos tripulados. AUVs de longo curso, como os planadores subaquáticos, são capazes de realizar missões de grande alcance com total autossuficiência energética. Durante essas missões de longo curso, esses veículos passam grande parte do tempo submersos, navegando sem conhecimento preciso da sua localização ou rota, podendo ser severamente afetados pelas correntes oceânicas. Quando estes veículos estão submersos deixam de receber as ondas eletromagnéticas emitidas por satélites e por isso os sistemas de localização e navegação baseados em sinais de rádio-frequência correntemente usados à superfície, como o GPS, tornam-se inúteis. A localização de AUVs é feita geralmente por sistemas de posicionamento acústico subaquáticos. No entanto, o funcionamento destes sistemas está limitado ao alcance relativamente reduzido dos transmissores acústicos, e requerem o apoio de uma infraestrutura localizada numa posição geográfica conhecida. Por este motivo os sistemas tradicionais de localização acústica subaquática não são adequados para a localização de AUVs a uma escala oceânica. Uma solução já explorada em outros trabalhos consiste em usar um ou mais veículos de superfície para transpostar os sistemas necessários à localização de um AUV, devendo naturalmente ter uma capacidade de autonomia e um alcance de operação similar ao AUV. O objetivo deste trabalho é o desenvolver um sistema de posicionamento subaquático de baixo custo e de baixo consumo energético, para ser usado a bordo de um veículo de superfície de longo curso. Pretende-se que este sistema permita estimar a posição de um AUV, possibilitando ao veículo de superfície acompanhar o AUV durante as missões de longo curso enquanto está submerso. O sistema de localização acústica passiva desenvolvido neste trabalho tem uma topologia idêntica à do ultra-short baseline, operando em tempo real. São usados apenas dois hidrofones para calcula a posição a duas dimensões de uma fonte acústica submersa emitindo um sinal conhecido, com base na integração de medições da direção do som ao longo do tempo. A direção de chegada da onda sonora é estimada pela diferença de tempo de chegada (TDOA) entre os dois sinais adquiridos pelos hidrofones, colocados próximos um do outro para atenuar a disparidade do canal acústico. Neste trabalho é proposto um método alternativo para o cálculo do TDOA, que consiste na determinação do início dos sinais recebidos recorrendo à deteção de uma série de transições por zero que são semelhantes em cada um dos sinais recebidos. Este processo foi implementado num dispositivo Zynq da XILINX, combinando um sistema de processamento digital dedicado na parte lógica programável com uma aplicação de software i ii executada no processador ARM embarcado. Como fonte acústica subaquática foi usado um transmissor comercial de 35 kHz, sendo os sinais adquiridos pelos dois hidrofones processados com uma frequência de amostragem de 2 Msps. Os resultados obtidos em ensaios de campo realizados numa marina permitiram obter a posição do transmissor acústico com um erro inferior a 1.5 m. Abstract With the growing interest in oceanic exploration, marine robotic systems are gaining a new importance in diverse scientific fields and industrial domains. In this context, it is clear that Autonomous Underwater Vehicles (AUV) are relevant instruments for research activities in the sea because of their adaptability for different missions, combined with a low-cost deployment and maintenance in comparison with manned vehicles. Long endurance AUVs, like the underwater gliders, are capable of performing very long range missions with complete energy self-sufficiency. During such long-term missions, these vessels usually spend most of their time submerged, navigating without a precise knowledge of their actual location and course, that may be severely affected by ocean currents. When these vehicles are submerged, they fail to receive the electromagnetic waves emitted by satellites. Therefore the localization and navigation systems based on radio frequency signals, like the GPS currently used above the surface become useless. The localization of AUVs is usually done by underwater acoustic positioning systems. However, the operation of these systems is constrained by the limited range of the acoustic transmitters, and require the support of an infrastructure located at a known geographic positioning. Because of this, the traditional underwater acoustic localization systems are not suitable for the localization of AUVs at the oceanic scale. One solution already exploited by other works uses one or more surface vehicles to carry the acoustic support infrastructure, which must have an autonomy and range of operation similar to the AUV being located. The goal of this work is to develop a low-cost and low-power underwater positioning system to be used on board of a single long-endurance surface vehicle. This system estimates the AUV position and allows the surface vehicle to accompany the AUV during long-range missions while submerged. The passive acoustic localization system developed in this work has a topology identical to the ultra-short baseline and is capable of operating in real-time. Only two hydrophones are used to determine the two-dimension position of a known underwater acoustic source, based on the integration along time of the direction of the received sound wave. The direction of arrival of the sound wave is calculated by the time difference of arrival (TDOA) between the signals acquired by the hydrophones, closely placed to attenuate the disparity of the acoustic channel. We propose an alternative method to calculate the TDOA, consisting in the detection of the beginning of the signals by discovering a series of zero crossing samples looking alike in each received signal and then calculating the time difference between them. This process was successfully implemented in a XILINX Zynq device, combining a custom designed digital signal processing system implemented in the programmable logic (PL) part, and a software application running in the embedded ARM processor. A commercial 35 kHz underwater acoustic beacon was used as the transmitter and the signals acquired by the hydrophones were processed at a 2 Msps sampling rate. The results obtained in field trials conducted in a marina allowed to determine the position of the acoustic beacon within an error of less than 1.5 m. iii iv Agradecimentos Em primeiro lugar, agradeço ao Professor José Carlos Alves, velejador, orientador e amigo, por todo o apoio não só na dissertação mas ao longo destes 5 anos, por todos os projetos, viagens, horas despendidas e piadas incompreendidas. Aos meus pais que sempre acreditaram em mim e que me apoiaram em todos os esquemas, planos e projetos desde pequeno. À Maria pela incansável revisão, assistência na piscina e no mar, apoio e paciência para me aturar, sem nunca deixar de gostar. Ao Perestrelo pelos incontáveis finos e atrasos que tornam Munique inesquecível. Ao Miguel por todos os trabalhos, relatórios e projetos nestes 5 anos e não só. A todos os colegas que passaram a amigos nos inúmeros jantares e noites sem dormir. José Francisco Valente v vi “The pessimist complains about the wind The optimist expects it to change The realist adjusts the sails” William Arthur Ward vii xiv LIST OF TABLES List of Abbreviations ADC Analog to Digital Converter ASV Autonomous Surface Vehicles AUV Autonomous Underwater Vehicles DAC Digital to Analog Converter DMA Direct Memory Access DOA Direction of Arrival DSP Digital Signal Processing FASt FEUP Autonomous Sailboat FIR Finite Impulse Response FPGA Field-Programmable Gate Array GCC General Cross-Correlation GIB GPS Intelligent Buoys I2C Inter-Integrated Circuit INS Inertial Navigation System LBL Long Baseline MLBL Moving Long Baseline PID Proportional Integral Derivative PL Programmable Logic ROV Remotely Operated underwater Vehicles RSSI Received Signal Strength Indicator SBL Short Baseline SLAM Simultaneous Localization and Mapping SoC System on Chip TDOA Time Difference of Arrival TOA Time of Arrival TOF Time of Flight USBL Ultra Short Baseline xv Chapter 1 Introduction On Earth the oceans cover almost 71% of its total surface, being of paramount importance for the human life in the whole planet as they influence the global climate and weather patterns. Oceans are a vast habitat for more than 230,000 known species, although much of the oceans depths still remain unexplored, hiding valuable resources in the water, on the seafloor and also beneath it. With this in mind, it is comprehensible the growing interest in oceanic exploration. Marine robotic systems are gaining a new importance in diverse scientific fields such as biology, oceanography, and meteorology. Also, in industrial domains as in offshore oil and gas exploration, seafloor mining, and in the areas of surveillance and security of marine borders and seaborne facilities. In this context, it is clear that Autonomous Underwater Vehicles (AUV) or Remotely Operated underwater Vehicles (ROVs) are relevant instruments for research activities in the sea because of their adaptability for different missions, combined with the low-cost deployment and maintenance in comparison to manned vehicles. Long endurance AUVs, such as the underwater gliders, are capable of performing very long range missions with complete energy self-sufficiency. During such long-term missions, these vessels usually spend most of their time submerged, navigating along a straight course controlled by an electronic compass or inertial navigation system, but without knowing their actual location and path that may be severely affected by ocean currents or waves. Nowadays, these vehicles are the only autonomous underwater platforms capable of collecting ocean data for several months without requiring any local assistance. Therefore, they are a powerful tool to improve the knowledge of the oceans. According to the records provided by the Everyone’s Gliding Observatories organization, since 2004 more than 350 gliders have been deployed in the oceans and presently (end of June 2016) 7 gliders are reported as being active. When these vehicles are on a mission, it is highly desirable to know their current location. While at the sea surface the global navigation satellite systems, such as the GPS, are able to provide real-time positioning with incredible accuracy. However, when these vehicles submerge, they fail to receive the electromagnetic waves emitted by satellites, and therefore these radio frequencybased navigation systems become useless. In addition to the inability to use satellite positioning, data communication underwater also 1 2Introduction suffers from the strong attenuation of electromagnetic waves in water. Because of this, the most robust way to communicate underwater is by using acoustic signals, although with very low bandwidths and ranges. Additionally, for an AUV engaged in a long-term oceanic mission, the communication of data to a shore base station can only be effectively done while at the surface using satellite communications, or relying on an acoustic communication link to a surface vehicle in range to act as a relay for satellite data networks. The localization of AUVs is usually done by underwater acoustic positioning systems. These systems consist of using acoustic transmitters and receivers to determine the relative position of the underwater vehicles, based on measurements of distance by the propagation time of acoustic signals. However, the operation of these systems is constrained in practice by the limited range of the acoustic transmitters, and because they require the support of an infrastructure located at a known geographic position, they are not suitable for the localization of AUVs at the oceanic scale. Figure 1.1: Underwater acoustic positioning system, LinkQuest Inc. These positioning systems are already available in the market, supplied by several companies such as the LINKQUEST INC., which has several products for this purpose. Figure 1.1 represents a simplified diagram of an underwater acoustic positioning system, which shows a surface vehicle as the control station and three submerged transducers that are used to locate an AUV. One solution for locating AUVs during long-range missions is to attach an underwater acoustic positioning system to one or more surface vehicles capable of similar long-range operation. To be effective, the surface vehicles must be able to accompany the AUV and stay in the range of the acoustic positioning system. Besides providing a localization infrastructure, the surface vehicles may also be used as relays to satellite data networks, thus offering a permanent communication link with the submerged vehicle, at the oceanic scale. (Papadopoulos et al.,2010) 1.1 Objectives 3 1.1 Objectives The goal of this work is to develop a low-cost and low-power underwater positioning system to be used on board of a single long-endurance surface vehicle, to estimate the AUV position and allow the surface vehicle to accompany the AUV mission while submerged. More important than determining an accurate position of the AUV, it is desirable to estimate an approximate location within a range of a few meters with low computational effort, and consequently requiring little energy for the computing tasks. Even with a rough localization data from the surface vehicle, it would be possible to make use of underwater acoustic communications to transmit data to the surface vehicle and then to a satellite network. The surface vehicle must be able to navigate faster than the AUV and with the same degree of autonomy and range of operation, what is necessary for establishing convenient navigation patterns to improve the estimation of the actual position of the AUV. With the continuous knowledge of the location of the submerged AUV, even if inaccurate, the surface vehicle can position itself to take advantage of favorable conditions for establishing acoustic communications and retrieve data from the AUV to relay to satellite data networks. For example, it is known that the acoustic communication along a vertical channel is less prone to several effects that affect the propagation of acoustic waves in the water, like multipath, reflections, and refractions at the transitions between water layers with different physical characteristics (temperature, salinity). Autonomous sailing boats, Alves and Cruz (2008), fulfill the requirements of autonomy, range and speed for the application scenario envisaged above. Although the navigation performance of a sailing boat is highly dependent on the relative wind direction and speed, their utilization in such application will require the establishment of convenient sailing patterns around the estimated location of the (moving) AUV, to improve the accuracy of localization of the AUV while staying within the range of acoustic communications. 1.2 Project Overview Long endurance autonomous underwater vehicles (AUV), such as electric underwater gliders capable of undertaking multi-month unassisted missions, suffer from the lack of real-time communications with the outside world, while submerged in remote ocean locations. In long range applications, it may be desirable to have real-time communications for retrieving data or remotely changing the mission parameters. As acoustic communications are the only practical mean to transmit data underwater, this problem can be mitigated by using a cooperative autonomous surface vehicle (ASV) capable of following at the surface the path of the submerged AUV and provide a relay to satellite data networks (Papadopoulos et al.,2010). An underwater acoustic positioning system capable of being applied to an autonomous surface vehicle is the Ultra Short Baseline (USBL) referred in 2.3.3. The USBL is a half-duplex communication system, which requires having a receiver and transmitter at both ends, with negative impact in the complexity of the acoustic systems in the two vehicles. Besides, in long endurance vehicles 4Introduction the power consumption (related directly to the demand of computation) is a critical factor that naturally affects the vehicle’s autonomy. This may be eased by attaching to the AUV a device transmitting a known acoustic signal and calculating the direction of the sound wave relative to an array of hydrophones, by measuring the time difference of arrival (TDOA) of the sound wave to the different hydrophones. Employing an array of hydrophones in a moving surface vehicle makes it possible to determine a set of relative directions between the two moving vehicles. By adopting convenient routes by the surface vehicle, and assuming a higher velocity than the AUV, it is possible to combine those directions to extrapolate the relative position of the submerged device. The tracking of the AUV can be relaxed because of the scale of the missions. A precision of a few meters in an oceanic scale is insignificant and is more important that the surface vehicle does not lose the link to the submerged vessel, granting a communication relay to the outside world. 1.3 Approach The approach suggested in this work is to develop a system based on the USBL scheme but only with a transmitter in the AUV and a receiver in the surface vehicle, reducing the complexity of the system and the power consumption. This system has to be capable of measuring the direction of arrival of the underwater acoustic transmission, then, the surface vehicle has to accumulate the different directions along time and estimate the relative position of the underwater vehicle. This system does not need any type of synchronization or event trigger communication, just acting as a passive receiver that calculates the relative position of the AUV. Contrary to the conventional USBL system, it is not possible to obtain a range for each single transmission. However, as explained above, exploiting a higher velocity of the surface vehicle it will be possible to determine an approximate location by combining several directions obtained from known positions of the surface vehicle. In order to make the system more flexible and not requiring the installation of specific equipment in the AUV, we decided to use commercially available underwater acoustic beacons as the acoustic transmitter. These beacons have their own power supply, and so they do not need to be electrically connected to the AUV power or computing system; the transmitter simply needs to be mechanically attached to the AUV body. The surface vehicle needs to have two or more hydrophones to successfully estimate the direction of arrival (DOA) of an acoustic signal emitted by the AUV. The time difference of arrival (TDOA) is computed with the phase-shift between pairs of signals recorded by distinct hydrophones. Then using the multilateration technique referred in section 2.5 and the spatial distribution of the hydrophones, it becomes possible to calculate the relative angle of arrival. In this work, it was decided to employ only two hydrophones for the prototype system because with a proper alignment of the hydrophones it is sufficient to determine the direction of the received sound wave in two dimensions, thus reducing the complexity of the hardware structure. However, 1.4 Contributions 5 the system developed can be easily extended to three or even more hydrophones to improve the accuracy of the direction of arrival and of the estimated position. 1.4 Contributions A major contribution of this thesis was the development of an alternative method to determine the time difference of arrival and its implementation in an embedded programmable system on chip (SoC). This method can easily be scaled to work with more hydrophones and ported to different re-configurable digital devices. A paper was presented in the XII Jornadas Sobre Sistemas Reconfiguráveis - REC2016, University of Trás-os-Montes and Alto Douro (UTAD), on the 21st of June, 2016. Finally, an abstract was submitted to review in the OCEANS’16 Monterey Student Poster Competition and was accepted to be included in the regular technical program. The conference will be held from 19th to 23rd of September, 2016. 1.5 Document Structure Besides this introduction, the rest of the document is organized in more 5 chapters. Chapter 2, the state of the art, it is presented some of the work that is known for underwater acoustics and localization systems. We then proceed in chapter 3to describe different approaches to determine the position of an underwater vehicle, developing a distinct method for this purpose. The implementation in chapter 4, describes all the electronic systems used in this work, the digital implementation of the algorithm, and the software processing. In chapter 5is presented the initial laboratory experiments and conclusions. The analysis of the results obtained and presents all the configurations for the hardware implementation. Also describes and presents the results for the tracking of an underwater beacon in the field scenario. Finally, we conclude by summarizing the results of this work and future work to be done. 6Introduction Chapter 2 State of the Art This chapter explores the techniques for localization of underwater vehicles used nowadays, describing different vehicles, methods, and topologies to achieve underwater localization. The principal focus of this work is on underwater acoustic localization systems for vehicles and devices operating in various underwater environments. 2.1 Underwater Vehicles Since the first successful submarine built in 1620, the construction of underwater vehicles never stopped increasing. In this section, three leading technologies are considered. 1. Manned Submersibles 2. Remotely Operated Vehicles (ROVs) 3. Autonomous Underwater Vehicles (AUVs) With so many different applications like scientific, commercial, and military missions, the manned submersibles, started to be too expensive to meet all these different requirements. Therefore, it appeared a new type of vehicles: the unmanned ROV (Remote Operated Vehicle). Because they are remotely operated from a surface support vessel, they are much cheaper to operate, easier to deploy and can reach virtually any depth in the Earth’s oceans. Then, the next logic step was to make them autonomous, creating the AUVs, that are the most cost-effective alternative and versatile vehicles. Autonomous underwater vehicles (AUVs) are versatile robotic submarines that are revolutionizing the way in which researchers and industry obtain marine data. Advances in technology have resulted in competent vehicles that have made possible new discoveries in the oceans and have dramatically reduced the cost of industrial and scientific surveying of the seabed. These vehicles range in size from man-portable lightweight AUVs to large diameter vehicles of over 10 meters length. Large vehicles have advantages in terms of endurance and sensor payload capacity; the smaller vehicles benefit significantly from lower logistics and cost. 7 14 State of the Art problems in underwater acoustic channels, being less accurate in terms of positioning. Manufacturers of USBL systems include Nautronix, Sonardyne, IXSEA (GAPS pre-calibrated Ultra-Short BaseLine), Applied Acoustics (EASYTRAK USBL), LinkQuest (TrackLink USBL), Tritech (Micron Nav), EdgeTech (BATS), Kongsberg (HiPAP - High Precision Acoustic Positioning), and EvoLogics (USBL Acoustic Modem). Figure 2.6: The USBL positioning method, Sonardyne International Ltd. 2.4 Time Difference of Arrival The time difference of arrival (TDOA) method has been extensively used for passive acoustic source positioning by measuring the time difference between signals arriving in two or more hydrophones and then extrapolating the relative position of the acoustic source. One of the methods used for estimating the time difference of arrival (TDOA) is the generalized cross-correlation (GCC) (Chen et al.,2006;Liu et al.,2012). The process consists in detecting the peak position of the correlation function between the two received signals, as the delay or time shift between them. Although being computationally intensive, this method is widely used because it can weaken the impact of the ambient noise on the accuracy of the signal lag estimation. 2.5 Direction of Arrival The direction of arrival (DOA) is commonly estimated by measuring the TDOA and then applying one of the following methods. 2.5 Direction of Arrival 15 •Multilateration The multilateration technique is based on hyperbolic positioning, one of the most common approaches for passive source localization, uses the time delays between pairs of sensors to characterize a curve of a constant time difference, which is a hyperbola (Dalskov and Olesen,2014). Considering that the source distance is great compared to the sensor spacing, is acceptable to acknowledge the wave as a plane to the sensors. Therefore, the direction of propagation can be described by the asymptote of the hyperbola, and the slope of the asymptote will then define the direction relative to the axis of the sensors. Figure 2.7: Diagram illustrating multilateration. •Trilateration Trilateration is the process of determining relative positions of points by measuring distances, using spheres or circles (Schau and Robinson,1987;Klungmontri et al.,2015). Considering only two sensors and that the source distance is great compared to the sensor spacing, the direction of arrival can be calculated by determining the angle of the intersection of two circles. Figure 2.8: Trilateration in two dimensions. 16 State of the Art 2.6 Cooperative Localization In cooperative localization, multiple vehicles can collaborate and help each other to navigate. To achieve this cooperation between different autonomous vehicles they have to communicate with each other to share data and determine or improve their position estimate (Rui and Chitre,2010). This concept can be implemented in various manners, described below. Figure 2.9: Autonomous Surface Vehicles cooperating to localize acoustic pinger in Ferreira et al. (2012). The Moving Long Baseline (MLBL), is an inverted LBL system like GIB, but the buoys at the surface are replaced with Autonomous Surface Vehicles (ASVs) (Curcio et al.,2005;Folk et al., 2010). The surface vehicles have to know their position and transmit it to the AUV, and then it can use different estimation algorithms like the Extended Kalman Filter, Particle Filtering, or nonlinear Least Squares (Eustic et al.,2007) to estimate their own position. The work reported in Papadopoulos et al. (2010) implements a different approach by having just a single maneuvering autonomous surface vehicle (ASV) to aid the navigation of an AUV. Applying the same principle as before, the surface vehicle shares their position with an underwater vehicle, then, the submerged vessel by using acoustic range measurements combined with the shared position can determine their own position. Another approach in cooperative localization is a system that can track the underwater vehicle, but not share the position with it. Instead of the AUV having a transducer to communicate with the surface, it can have a simple beacon that will only transmit a signal repeated in time. Then, the surface vehicles can estimate the position of the AUV by acquiring the signal of the transmitter. In Ferreira et al. (2012) this topology is implemented with two ASVs tracking an autonomous underwater vehicle. The work developed in this thesis intends to build a cooperative system for the localization of a submerged beacon, but using only one surface vehicle equipped with a passive acoustic direction finder based on the USBL topology. Instead of implementing a moving baseline created by the 2.6 Cooperative Localization 17 two autonomous boats, a synthetic baseline is formed by measuring the angle of arrival in different locations. The ASV can establish known navigation paths by exploiting its higher velocity and maneuverability in comparison to the AUV. Therefore, tracking the underwater vehicle in realtime by accumulating several position estimates. 18 State of the Art Chapter 3 Algorithm As described in section 1.3, in this work we exploit the time difference of arrival between the signals acquired by two hydrophones, to measure the phase shift between them and determine the direction of arrival of the sound wave. An estimation of the position of the submerged sound source (attached to an AUV) can then be built by combining these directions with the relative movement between a (fast) ASV and a (slow) AUV. The experimental acoustic data acquired for the experiments carried out in this work were obtained using a custom acquisition and processing platform described in section 4.1. The preliminary laboratory experiments were done in a test tank of fresh water, with 4.75 m by 4.4 m and 1.7m of depth. The field tests in a more realistic environment were performed in a marina and in the sea. The acoustic transmitter we used was a commercial underwater acoustic beacon transmitting a known sequence of short pulses of a 35 kHz signal (section 4.1). One of the advantages of using passive tracking is to avoid the need for synchronization between the transmitter and the receiver. Therefore, the acoustic transmitter is not limited to any specific time slot to communicate. One objective of the system developed was to be easily adapted to other acoustic beacons, operating in different frequencies and transmitting different patterns of pulses. Consequently, in order to achieve the goal of not being bounded to a particular underwater acoustic transmitter, the method to apply in this system has to be prepared for a scope of different frequencies. 3.1 TDOA Algorithm The time difference of arrival (TDOA) technique has been extensively used for passive acoustic source positioning as discussed in section 2.4. The first approach was to analyze the pair of signals in the Matlab software and to analyze sets of real signals acquired in the test tank to identify features of the signals in the time domain that would be helpful for determining the TDOA. The duration of the signal transmitted by the beacon is 12 ms, but because of the reflection in the walls and the bottom of the tank, the significant length of the signals received was of approximately 19 20 Algorithm 80 ms. As referred in section 2.2.3, the underwater acoustic channel produces a lot of distortions, which in this confined environment are essentially due to the multiple reflections. This is a pessimistic, although possible, underwater acoustic scenario. One primary objective was that the system should be able to deal with such signals. These signals enhance the simulation for more robust algorithms that can work in rough situations. In figure 3.1 (top), the signals of the independent hydrophones are plotted. These signals were recorded with the transmitter aligned with the axis of the hydrophones, so that the time difference of arrival was the maximum possible. Considering the speed of sound in fresh water of 1482ms−1and the distance between the hydrophones of 6 cm, the delay expected is approximately 40.5µs to the hydrophone-1 receive the signal previous in hydrophone-2. Ideally, the distance between the hydrophones should be less than half of the wavelength (21.2 mm) to disambiguate the period of the transmitted signal. As this was impossible due to physical hydrophone size, the solution was to add one wavelength of the transmitted signal to the separation of the hydrophones, so the value of 6 cm. One of the most popular methods to measure the phase shift is the generalized cross-correlation (GCC) process, so this was the first method to be evaluated. The data described above is analyzed in the following section. 3.1.1 Generalized Cross-Correlation As referred in section 2.4, the generalized cross-correlation method is highly adopted to measure the time difference of arrival, because of its properties in masking the contribution of the ambient noise. The process of determining the delay consist in calculating the correlation function of the two received signals. The result is a relation between the similarity of the two series with the lag relative to one another. The peak position of the correlation function indicates the instant in time where the signals are best aligned. Figure 3.1 shows the two 35 kHz signals recorded in the test tank, where the red plot represents hydrophone-1, which is farther than hydrophone-2 (blue plot) in relation to the acoustic transmitter. The amplitude envelope is very dissimilar and highly variable with small displacements of the relative position of the two hydrophones. Because of this, the cross-correlation method is not useful in such situations. Figure 3.2 represents the correlation between the signals recorded. Measuring the time difference between the signals as the time position of the peak value of the cross-correlation function, the result is −1265µs, which is clearly wrong based on the values previously calculated for the delay. As expected, the cross-correlation in the whole duration of the signal does not allow to have the correct estimation of the delay. 3.1 TDOA Algorithm 21 Figure 3.1: The 35 kHz signals recorded in experimental tests (top). Close-up in the beginning of the signal (bottom). Figure 3.2: The result of the cross-correlation of the signals in figure 3.1 (top). Close-up in the peak of the correlation (bottom). Generally, the signals are correlated with their entire significant duration, which can be difficult to do in real-time on long signals with high sampling frequencies and using low performance computing systems, due to the substantial computational effort needed. Figure 3.3: The beginning of the signals, hydrophone-1 red, hydrophone-2 blue. 22 Algorithm An alternative approach is the segmentation of the signal in small portions and then applying the cross-correlation, as this requires a lighter computational effort and can produce better results. Analyzing the beginning of the signal, as shown in figure 3.3, it is easy to notice the 35kHz sine wave and the relative phase shift between the different received signals. The plot of the hydrophone-1, in red, is clearly delayed with respect to hydrophone-2 (blue) and it is clear that the delay is slightly higher than the period of the sinusoidal wave. Figure 3.4: A fraction of the recorded signals shown in figure 3.3. Figure 3.5: The result of the crosscorrelation of the signals in figure 3.4. The cross-correlation between sinusoidal waves has several peaks that correspond to delays that match the period of the wave, as shown in 3.5. Therefore, it is not plausible to determine delays above the period of the wave with cross-correlation method analyzing fractions of the signal. Figure 3.4 shows a small fraction of the beginning of the signal, the result of the crosscorrelation of the two signals is shown in 3.5. The maximum lag peak in this correlation occurs when the delay is approximately 10µs, which, when added one period of the sinusoidal wave (nearly 29 µs), the result is 39 µs, what is close to the expected delay of 40.5 µs. This process is acceptable for measuring the delay within a period, but in this case, the phaseshift is higher than the period of the wave, thus, creating an ambiguity that cannot be determined with this method. Exploring the cross-correlation in small fractions of the signal it was concluded that the phase difference is not constant, because of the echoes that are added to the direct wave, provoking distortions in amplitude and especially in phase. Consequently, in order to successfully measure measuring the delay of arrival, the analysis has to be focused only on the beginning of the signal, where it is not influenced by the replicas. Figure 3.6 shows the result of the cross-correlation of the start of the signals represented in 3.3. This function also cannot determine the true delay, because of the dissimilarity between the signals amplitude, giving a lag much higher than the expected, in this case 239µs. For that reason, the analysis of the beginning of the signal has to be independent of the amplitude. 3.1 TDOA Algorithm 23 Figure 3.6: The result of the cross-correlation of the signals in figure 3.3. 3.1.2 Zero-Crossing Methodology As discussed before, the method to measure the TDOA has to analyze the beginning of the signal, because all of the rest is affected by distortions, and it also has to be indifferent to the amplitude and the period duration of the wave. In this regard, the algorithm has to determine the instant of arrival of the two signals and then calculate the time difference between them. The focus has to be in accurately detecting only the very beginning of each received signal, analyzing them in the time domain. The process is based on continuously detecting the zero-crossings of the signal and measuring the signal period as the time difference between two consecutive zero-crossing in the same direction (either rising or falling). A series of consecutive zeros occurring within a predefined time interval, the expected signal period, indicates the beginning of the received signal. Figure 3.7: The beginning of the signal after filtering. 30 Algorithm α=           −90−arctan√4d2−r2 r,r<0 90−arctan√4d2−r2 r,r≥0 (3.6) As shown in the plot of figure 3.15, the difference between the two methods, under the assumptions taken (R d) is negligible. With these results we can conclude that it is much simpler to compute the angle of arrival using the multilateration technique. Figure 3.15: The difference between the trilateration plot in figure 3.12 and the equivalent calculated by multilateration. Summary In this chapter, several simulations were made to find the best suitable algorithm to determine the time difference of arrival, was discussed the generalized cross-correlation and a different strategy proposed in this work, the zero-crossing method. By analyzing two different methods to determine the angle of arrival with the time difference of arrival, we concluded that both methods were equivalent for the conditions considered in this work. Chapter 4 Implementation This chapter starts by describing the electronic systems used in this work, from the underwater acoustic beacon to the receiver analog front-end and embedded computing platform. Then, the implementation of the algorithm described in section 3.1.2 for determining the TDOA is presented, as a dedicated hardware peripheral. Lastly, the software implementation to calculate the direction of arrival, that runs on the embedded processor is shown. 4.1 Hardware Equipment 4.1.1 Transmission As referred in the previous chapters, we decided to use an underwater commercial acoustic beacon to serve as the transmitter because of their robustness, and easy applicability in several different scenarios. There are various different similar devices available in the market with diverse characteristics like range, depth, frequency, and autonomy. Figure 4.1: The Sonotronics Inc.,EMT-01-03, underwater acoustic beacon. The one available for this work was an EMT-01-03 Equipment Marking Transmitters, from Sonotronics, usually used to attach to submerged equipment in order to locate it in case of loss. This is generally employed with a dedicated acoustic receiver tunable in the beacon’s frequency, 31 32 Implementation with an announced maximum usable range of 4 km. As figure 4.1 shows, the beacon is a tube with 210 mm of length, 32 mm in diameter, and weights in air 223 g. The EMT-01-3 has a replaceable battery and is rated for a maximum depth of 1000 m. Figure 4.2: Recording of the beacon acoustic pattern. This beacon emits a periodic signal that consists of a series of short sinusoidal pulses with a frequency of 35 kHz, each one lasting about 12 ms. Every beacon has its own periodic pattern, the one employed as a 4-8-8 pattern that is shown in figure 4.2. The time interval between consecutive pulses is 930ms, and the other separation intervals are 1860ms and 3720 ms. One complete patterns lasts for approximately 23.5s and contains 20 identical pulses. 4.1.2 Reception Recording device The acquisition of the acoustic signals propagated in the water is made by a pair of omnidirectional hydrophones. The ones used are from the company Aquarian Audio Products, and the model is the H2a hydrophone. The interface with the recording device is a 3.5 mm microphone input. In figure 4.3, the hydrophones are shown in a structure built to fix them at a distance of 6 cm apart. The dimensions of the hydrophones are 25mm ×46 mm, which makes it impossible to be at a distance of less than 25 mm. As this distance is superior to half of the wavelength (21.2 mm) of the transmitted signal, it is not possible to measure the TDOA by the phase difference between the two signals. We decided to use a distance between the hydrophones slightly below 1.5×wave length for an easy support construction. 4.1 Hardware Equipment 33 Figure 4.3: The structure holding the two hydrophones 6cm apart. Analog front-end The hydrophones are connected to a custom designed analog front-end board with a variable gain amplifier and low-pass anti-aliasing filtering. The analog pre-processing applies two variable and digitally controllable amplifiers chains, one for each hydrophone, followed by analog low-pass anti-aliasing filters with a cutoff frequency set to, approximately, 250 kHz. This board also includes a high-speed analog switch, allowing to multiplex two additional analog signals by using a second amplifier and filtering board. The amplifier chain is divided into two stages, the first one with a gain adjusted to 10×, and the second stage with two I2C digital potentiometers that allows programming the gain of two analog amplifiers in series with an overall gain that varies from 0.1× to 2500×. Figure 4.5, shows the analog front-end board in the right side, and figure 4.4 presents a simplified block diagram of one amplifier and filtering analog path. Figure 4.4: A simplified block diagram of the analog programmable amplifier and filtering daughter board. Digital platform The digital platform used to process the signals was the RedPitaya single board computer, based on the XILINX Zynq 7010 programmable SoC (figure 4.5 left). This device integrates a dual core ARM Cortex A9 with a peripheral FPGA fabric for the integration of custom designed coprocessing or interfacing blocks. Besides, the standard on-board interfaces and devices required to 34 Implementation run a Linux operating system (DDR memory, Ethernet interface, SD flash memory and USB), this board also includes a dual high-speed ADC and a dual high-speed DAC, both capable of operating at a sample frequency of 125 Msps. This platform is usually sold as an all-in-one laboratory instrument, providing the programmable logic (PL) hardware interfaces and web applications that implements an oscilloscope, an arbitrary function generator, a spectrum analyzer and also a general purpose PID controller. The hardware design for the PL section is provided as an open source project for the XILINX Vivado design tool suite. A variant of the original hardware system also includes a DMA controller that allows the continuous acquisition and recording of the two analog input signals up to 2 Msps. Figure 4.5: The RedPitaya embedded computer (left) and the analog front-end amplifier board (right). 4.2 Digital Implementation System Integration The re-programmable hardware system was based on the original design included in the RedPitaya distribution, maintaining the parts that implement the oscilloscope and the continuous signal recording, and removing all the other unused modules to free FPGA resources and facilitate the design optimization process to reach the 125 MHz main clock frequency. The oscilloscope function is very convenient for debugging when performing field experiments, by analyzing the real signals being captured by the acoustic sensors. The input to the module implementing the oscilloscope is taken from a configurable decimator module that reduces the sampling frequency by factors equal to powers of two and averages the samples, which is used by the oscilloscope application to adjust the timebase. The oscilloscope can be accessed by an internet browser when the RedPitaya is on the same network, providing a graphic interface for analyzing the signals. 4.2 Digital Implementation 35 The TDoA calculator was integrated in the original oscilloscope interface (figure 4.6), keeping the whole original functionality. To further enhance this, a multiplexer has been added to the oscilloscope datapath, for being able to use that same circuit and software application for observing other intermediate signals in different points of the TDOA determination. Figure 4.6: Simplified block diagram of the top-level module. The ARM processor shares data with the FPGA using the AXI interface, enabling a fast communication link between the Linux OS and the re-programmable hardware. This connection is used for configuring the modules parameters in real-time and transmitting the relevant data to the post-processing unit. Figure 4.7: Simplified block diagram of the DSP module. A simplified block diagram of the top-level module on the Programmable Logic (PL) device and their interfaces is shown in figure 4.6. The output of the analog front-end for each hydrophone (H1 and H2) is connected to the analog to digital converter (ADC), then passes through the decimator (DEC) and enters the digital signal processing (DSP) unit, that contains all of the modules that implement the TDOA calculator. 36 Implementation The DSP module is represented in a simplified diagram in figure 4.7. The next subsections will describe the three different modules in more detail. High-pass filter The signal from each channel passes through a digital high-pass filter with a cutoff frequency of 34 kHz to attenuate the low-frequency components. Figure 4.8: Magnitude response of the high-pass filter. Figure 4.9: Diagram of the transposed Direct Form II Biquad. The filter was designed in Matlab to perform a type II Chebyshev infinite impulse response filter (IIR). As shown in figure 4.8, the magnitude response has no ripple in the pass-band but does 4.2 Digital Implementation 37 have equiripple in the stop-band below 28 kHz, with an attenuation of 40 dB. It was chosen an IIR topology to save resources but maintaining a high roll-off. The filter of order 8 was implemented in four biquads, direct form II transposed structure, and connected in cascade, the topology of the biquad is shown in figure 4.9. These decisions were made to prevent instability and the overflow of the signal. Using the filter design and analysis tool (FDATool) in Matlab is feasible to extract the coefficients (b0, b1, b2, a1, a2) for applying directly in the structure in figure 4.9. Amplitude Determination The AMP block in figure 4.7 is an amplitude measurement module that determines the maximum amplitude of each signal over a configurable time window. This information is then transferred to the ARM processor for later analysis, the time in which the amplitude is inspected, it is also configurable by the ARM. TDOA Calculator The DSP instantiates four TDOA calculators with different configurations that the post-processing unit can change in real-time, the delay measurements are passed to the ARM for future analysis. The filtered signal is also forward to the oscilloscope for the analysis in the web application. Figure 4.10: Simplified block diagram of one TDOA calculator. Four modules are instantiated with different parameters. The TDOA calculator module is shown in figure 4.10. Each of the four modules receives the signals of two hydrophones (H1 and H2), after passing through the decimator and high-pass filter, as shown in figure 4.7. The signal is then forwarded to the zero finder block, which is calculated continuously the time difference between two consecutive transitions by zero in the same direction (the rising edge in the block shown), measuring the period of the input signal. The sequence of periods of each signal is analyzed to detect a number of consecutive valid periods whose duration is above the threshold, as explained in section 3.1.2. This is the trigger event that indicates the 38 Implementation start of that signal. A finite state machine receives these trigger signals, analyzes the sequence of periods determined and drives the delay calculator module that estimates the TDOA, applying the method described in section 3.1.2. Two of these modules are instantiated to find the signal period using rising edge transitions and the other two, the falling edge. A pair of modules containing the two different approaches share all the same parameters. The TDOA calculator measures the time delay between detecting the arrival of the signal in each hydrophone, if this time is higher than a maximum delay configuration, then the module activates a time-out flag. When the module detects the arrival of both signals or the time-out is flagged, it waits for a configurable time until starting analyzing the signal again. This is to ensure that the system does not inspect the signal in the continuous transmission and that it is prepared for the start of the next pulse. 4.3 Implementation Results The implementation in the Zynq 7010 programmable SoC, maintaining the oscilloscope and the DMA interface for real-time recording, occupies the FPGA resources shown in table 4.1. The high utilization of the DSP slices is due to the pipelined implementation of the two high-pass IIR filters. Resource Occupancy LUT 33% (5759) FF 17% (6001) BRAM 27% (16) DSP48 83% (66) Table 4.1: FPGA resource usage (Zynq 7010). The development was done in the XILINX Vivado design tool and the simulation of the individual modules described above in the ModelSim software. The simulation of the top-level module, shown in figure 4.6, was performed with the recorded signals used in the development of the algorithm in chapter 3. Figure 4.11 shows the result of the simulation in the ModelSim, obtaining four delays that correspond to the four different TDOA calculators. The values obtained, match the expected delays simulated in MATLAB, for the signals and parameters chosen. The estimates are calculated after the modules receive 8 consecutive transitions by zero, as two of them analyze the rising edge and the other two the falling they update the delay in different instants in time. The configuration parameters were adjusted to the recorded beacon signal, as explained in section 5.1. 4.4 Software Processing 39 Figure 4.11: The result of the ModelSim simulation. 4.4 Software Processing The ARM processor running Linux is the post-processing unit that connects with the FPGA. Taking advantage of the filesystem and the interfaces already developed was created programs, written in C (programming language), to control and analyze the data from the digital implementation. Automatic Gain Controller The signals recorded have amplitudes that are affected by the distance to the transmitter and can also be different between the hydrophones. To solve this was developed an automatic gain controller, which the goal was to change the gain of the amplifiers in the analog front-end to maintain the range of the signals regulated. As described in section 4.1.2, each channel as two digital configurable potentiometers that change the gain of the analog amplifiers, the input signal can be raised from 1×to 25000×the original amplitude. The C program uses the I2C digital interface to modify the gain. This is done by a proportional controller that adjusts the amplitude to 80% of the ADC maximum range. The greatest amplitude of the signal is determined by the FPGA module presented in section 4.2, the value is calculated over a period of time, then the amplitude is written in the shared memory and the module reseted. Every time the program has a new amplitude, updates the gain value for each channel. The time window used to measure the amplitude must ensure that capture of a significant part of the transmitted signal, in this case, a sine pulse from the beacon. This time can be configured by the program, and for the beacon described in section 4.1.1, it was chosen 5s towards ensuring the capture of at least one pulse. 46 Practical Experiments Figure 5.4: Number of estimates that follow specific line. This figure shows that for all the angles registered, more than 50% of all the estimates are on the theoretical line, and less than 10% do not follow any of the three lines (wrong values). This analysis allows us to conclude that is possible to correct the shifted values by adding or subtracting one period, thus merging the values in the theoretical line. 5.2.1 TDOA Post-processing As referred in section 4.4, the post-processing analysis of the TDOA estimates was enhanced with the experimental results. It was concluded that for having a good estimate of the angle was needed to compensate the TDOA estimates that are 1 period off. To achieve this in real-time we developed a moving TDOA window with the latest 16 delays that correspond to 4 pulses. Assuming that there are more values in the correct function than in the other two, the values were filtered to have a proper estimate of the delay in the right line. This value its rough estimate of the delay because enters with a lot of different pulses, so it is just used for compensating the latest 4 delays. The processing of the delay is done in two phases. The first one is the explained above, that consist of getting a rough estimate of the delay but in the correct function. The second stage is an analysis of just the last pulse received, so the latest 4 TDOA estimates. Based on the rough estimate, the four delays are rectified by adding or subtracting a period to match the correct line, after that, are removed any outlier that can be a time-out value or a delay to different from the others. The remaining values are then averaged, and if all the values are discarded is used the last valid estimate. 5.2 Laboratory experiments 47 5.2.2 Different Beacon To further test the robustness of the implementation and the developed algorithm, we decided to experiment the tracking of a different underwater commercial acoustic beacon. The transmitter used as similar characteristics to the one described in section 4.1.1, but as a pulse frequency of 76 kHz and an acoustic pattern of 4-4-6-5. As this frequency is more than the double of the previous beacon, the resolution measured in samples of the wavelength reduced to less than half of the earlier one, so it was decided to increase the sampling frequency to the double. Using the decimator factor of 32 instead of 64, the input sample rate of the DSP module changes to almost 4MHz. All of the modules stay unchanged with this modification except the high pass filter, which duplicates the cut-off frequency resulting in 68 kHz. The configurable parameters were changed to accommodate these corrections. The following table summarizes all these parameters. Parameter Value Decimation factor 32 Threshold-1 85.6% Threshold-2 93.4% Number of zeros 12 Maximum delay 51.2 µs Inactivity time 200 ms Amplitude time 10 s Table 5.2: Parameters configuration for experimental tests with a different beacon. In addition to changing the decimation factor, the zero periods thresholds were altered to the closest integer values in samples to resemble the proportional value of the last experiment. The time period in which the gain module analyzes the maximum amplitude of the signal was changed to catch the at least one pulse of the new beacon’s pattern. Figure 5.5: The angles measured in comparison with the real ones. 48 Practical Experiments The experiment was done in the test tank with the same configurations described in section 5.2. With a protractor to measure the real angle it was done a sweep from −60◦to 60◦in 10◦steps. In each position it was recorded 20 direction estimates, one for each pulse, then it was averaged to obtain one angle measure. The plot in figure 5.5 shows the result angle for each position in comparison to the real angle. The maximum absolute deviation was 3.1◦with a mean error of 1.1◦. This experiment demonstrates the versatility of the system, using a different transmitted signal and sampling frequency without changing the implementation. All the configurations can be modified in real-time allowing the system to track a specific transmitter. 5.3 Field Tests The laboratory experiments produce good results for the TDOA estimates, therefore, after having a solid delay estimate is straightforward to determine the angle of arrival, is just applying the polynomial functions calculated in section 4.4. The next logic step was to track the position of the acoustic beacon by measuring the angle of arrival in different positions. The premise was that an autonomous surface vehicle capable of better navigation performance than a long endurance AUV, in terms of speed and maneuverability, can be programmed to sail in a convenient route around an uncertain position of an AUV for improving the estimation of its location. Figure 5.6: The field experiments in the marina of Leixões. Assuming that the underwater vehicle is a lot slower than the surface vessel, a valid approximation can be that the transmitter is stationary, and the hydrophones are moving. This approach makes the tracking experiment easy to implement. 5.3 Field Tests 49 Figure 5.7: Two dimensions plot to track the underwater beacon (top). Close-up on the intersection centroid (bottom). 50 Practical Experiments To approximate the experiment to a realistic scenario was decided to track the beacon in the Leixões port, for having an ample space to move the hydrophones and noise conditions more appropriate to the ocean circumstances. Figure 5.6 shows the scenario of operations in the marina. The experiment was done with the beacon submerged at 1.5 m deep, 13 m away from any obstacle. The hydrophones were moved 22 m in a straight line along a floating pontoon maintaining a constant direction, where the closest point to the beacon was 11m, with a constant depth of 1 m. An application was made to record 20 directions of arrival per positions, as before in each location a new file was created with the 20 estimates. The recording was made in steps of 50cm, some of the positions were altered by a few centimeters because of obstacles in the floating structure. The angles recorded was the result of the conversion of individual pulses into TDOA estimates, using the real-time post analysis discussed in section 5.2.1. The files were studied in Matlab to extract the relevant information, in each position was made the average of the angles and plotted a line with the respective direction, as the plot on top of figure 5.7 shows. The bottom plot of figure 5.7 is a close-up on the intersection area, where the blue dots are the intersection of the lines; these were calculated by intersecting all the lines that start at a distance greater than 3 m, adding up to a total of 742 points. This was opted to avoid remote intersections because the small variations in position produce a minute change in the angle, which can cause the lines being almost parallel and intersect really far from the expected. The green dot is the centroid of all the intersection providing the estimated position of the acoustic transmitter. The red point is the real position of the beacon that has an error associated because was suspended on a rope that inevitably oscillated with the small waves made by moving boats. Figure 5.8: Distribution of the intersections relative to the centroid. 5.3 Field Tests 51 As referred before, the goal was to have a relaxed tracking system that allows a surface vehicle to follow the underwater vessel in long missions, so the high precision in the location is not a key element. The centroid of all the intersection is at 1.39 m from the real beacon position, which is not a bad estimate. The bottom plot of figure 5.7 shows that the intersections are not very dispersed from the center. The black dashed circle as a radius of 1 m and is centered in the centroid, 87.5% of all the intersection are contained in this circle. In figure 5.8 is shown the number of the intersection due to the distance to the centroid, which reveals the high density of points close to the centroid. After this analysis, the accuracy in the position tracking can be enhanced with proper calibration of the system, because the focus of the intersections are clearly away from the real location of the transmitter. This can be induced by small variations in the hydrophones axis angle, a difference in the spacing of the hydrophones or tilt. The range of the system was tested too, in the marina scenario, the maximum distance which the measurements were stable was about 20m from the transmitter, farther than that, the analog signal amplifier reached the maximum gain. The noise was so amplified that signal was almost unrecognizable, and the system did not work properly. Summary In this chapter it was presented the two principal experiments done in this work. Firstly, it is explained the hardware parameters used in the configurable modules. Afterwards, it is presented the first practical experiment for the validation of the TDOA estimates with the correspondent angle of arrival, and the conclusions are added to enhance the post-processing phase. In the laboratory experiments was tested a different beacon to demonstrate the versatility of the system. Lastly, was made the tracking experiment in the marina, with the real-time estimation of the angle of arrival, and the posterior analysis of the tracking results, culminating in the estimation of the beacon’s location. 52 Practical Experiments Chapter 6 Conclusions The approach in this thesis was to develop a low-cost and low-power underwater positioning system that could be employed in a long-endurance surface vehicle. To reduce computational effort and the infrastructure complexity was created an alternative method to determine the TDOA. In chapter 3it was analyzed the signals recorded by the hydrophones, verifying that the underwater environment introduces several distortions in amplitude and phase of the sound waves, due to reflections and the variation of the sound propagation speed with temperature, pressure and salinity. Particularly in confined spaces, the high level of reverberation makes the amplitude of the signals captured by the hydrophones very dissimilar, and because of this, the use of the cross-correlation method is not effective to identify the time difference of arrival. In this work, we propose an alternative method to calculate the TDOA consisting in the detection of the beginning of the signals by discovering a series of zero crossing samples looking alike in each received signal, and then calculating the time difference between them. The realization of this method is presented in chapter 3, as the process to determine the direction of arrival of the acoustic wave by the TDOA estimate. It was concluded that in a two-dimensional spatial configuration the trilateration and multilateration method have similar results, but with different theoretical functions. As both approaches are computationally heavy for real-time calculations, in chapter 4it was decided to fit the trilateration function with low order polynomial functions, to ease the implementation in the software processing unit. In chapter 4is presented the successful hardware implementation of the zero crossing algorithm in the RedPitaya single board computer. The hardware equipment utilized is presented, like the commercial underwater acoustic beacon used as the transmitter and the analog front-end board, where the two hydrophones are connected. The digital integration is divided into several modules, which are described individually. In this section is described the implementation of the high-pass filter, the amplitude determination module, and the TDOA calculator. The application running on the ARM processor is also defined in chapter 4, this section contains the automatic gain controller of the analog amplifiers, the clarification of all the configurable parameters, and the DOA determination. The chapter 5contains the results and conclusions of the practical experiments. Firstly, was 53 54 Conclusions determined the hardware parameters used to configure modules. Afterward, in the test tank of water was tested the prototype by measuring the time difference of arrival for several angles, then was analyzed the results, allowing the enhancement of software post-processing filters. In the laboratory was also done another experiment to demonstrate the versatility of the system, using a different transmitted signal and sampling frequency without changing the implementation. All the configurations can be modified to track a specific transmitter allowing the system to change the in real-time the beacon to track. Lastly, in the marina environment was recorded real-time estimations of the angle of arrival in different positions along a straight line, wherein after an analysis of the directions was successfully located the fixed transmitter. 6.1 Future Work Several improvements can be added to this work; these can correct some of the problems encountered and enhance the prototype characteristics. •Employing high-quality hydrophones with differential output to decrease the noise and increase the range of the system. •Apply to the structure an inertial measurement unit to compensate inclinations or changes in direction. •Testing with different acoustic transmitters with various frequencies and patterns to validate the prototype. •Employing more than two hydrophones to achieve a better angle precision and three-dimensional tracking. •Applying the system to an autonomous surface vehicle and establishing convenient navigation patterns to improve the estimation of the actual position of the transmitter. References Sonotronics Inc. (2016). Equipment marking transmitters. Available from: http://www. sonotronics.com/?page_id=119. Alcocer, A., Oliveira, P., and Pascoal, A. (2006). Underwater acoustic positioning systems based on buoys with gps. In Proceedings of the Eighth European Conference on Underwater Acoustics, volume 8, pages 1–8. Alves, J. C. and Cruz, N. A. (2008). Fast-an autonomous sailing platform for oceanographic missions. In OCEANS 2008, pages 1–7. IEEE. Chen, J., Benesty, J., and Huang, Y. (2006). Time delay estimation in room acoustic environments: an overview. EURASIP Journal on applied signal processing, 2006:170–170. Curcio, J., Leonard, J., Vaganay, J., Patrikalakis, A., Bahr, A., Battle, D., Schmidt, H., and Grund, M. (2005). Experiments in moving baseline navigation using autonomous surface craft. In OCEANS, 2005. Proceedings of MTS/IEEE, pages 730–735. IEEE. Dalskov, D. and Olesen, S. K. (2014). Locating acoustic sources with multilateration. Master’s thesis. Desert Star Systems (2015). Pilot system (accessed jan 2016. Available from: http:// desertstar.com/product/pilot-system/. Eustic, R. M., Whitcomb, L. L., Singh, H., and Grund, M. (2007). Experimental results in synchronous-clock one-way-travel-time acoustic navigation for autonomous underwater vehicles. In Robotics and Automation, 2007 IEEE International Conference on, pages 4257–4264. IEEE. Ferreira, B. M., Matos, A. C., Cruz, N. A., and Almeida, R. M. (2012). Towards cooperative localization of an acoustic pinger. In Oceans, 2012, pages 1–5. IEEE. Folk, A., Armstrong, B., Wolbrecht, E., Grip, H. F., Anderson, M., and Edwards, D. (2010). Autonomous underwater vehicle navigation using moving baseline on a target ship. In OCEANS 2010, pages 1–7. IEEE. 55