Underwater navigation using location-dependent signatures

Underwater Navigation Using Location-Dependent Signatures Di Qiu Sigtem Technology, Inc. San Mateo, CA [email protected] Robert Lynch Naval Undersea Warfare Center New Port, RI [email protected] Abstract—This paper investigates the benefits of a multisensor fusion methodology for underwater navigation using locationdependent signatures, or geotags. The proposed coordinatefree system uses both natural and man-made signals, as well as transient events to extract location-dependent signatures for navigation and guidance. Natural signals include the geomagnetic field, gravity field, bathymetric features, and naturally-occurring very low frequency radio signals. Manmade acoustic sources of opportunity include drainage outlets and pump stations in littoral zones and particularly in harbors, which can be explored to serve as underwater beacons for navigation. This paper models a multisensor coordinate-free system, characterizes various signals for underwater navigation, and evaluates the Multisensor Underwater Signature-based Navigation (MUSNav) system in terms of accuracy, availability, and continuity of the navigation solution. Keywords—underwater navigation; signature; geotag; control; MUSNav multisensor; Erik Blasch Air Force Research Lab WPAFB, OH [email protected] location accuracy. Various classifiers developed for temporal transient signal detection and classification can be applied for spatial search and localization. location Figure 1 – MUSNav in underwater environment TABLE OF CONTENTS 1. INTRODUCTION ………………………………………. 1 2. SYSTEM MODEL ……………………………………… 1 3. SIGNAL CHARACTERISTICS .….……………………… 4 4. PRELIMINARY SIMULATION RESULTS …………….…. 5 5. CONCLUSIONS ………………………………………… 7 REFERENCES ……………………………………………. 8 BIOGRAPHY …………………………………………...…. 8 I. INTRODUCTION The requirements of performance and complexity of signal processing for underwater navigation have dramatically increased over the last several decades. One important aspect of underwater navigation using location-dependent signatures, or geotags, is to build and maintain databases of location-dependent signatures, as shown in Figure 1. The geotags are compared with real-time acquired signals to estimate a position location. However, the use of a single signal for underwater navigation faces problems of poor resolution and lack of coverage. In contrast, the fusion of multiple signals has the potential to ensure the required accuracy, confidence, timeliness, and continuity of the navigation solution. A multisensor navigation system can extract more location-dependent features from received signals and provide high spatial decorrelation in the derived signatures, thus resulting in high system availability and reliability. In this paper, we study the spatial correlation (or spatial decorrelation) of signatures from multiple signals and the effect these have on search speed and position 978- -4577-0557-1/12/$26.00 ©2012 IEEE Chun Yang Sigtem Technology, Inc. San Mateo, CA [email protected] When acoustic sources of opportunity are available, we formulate the underwater navigation as a closed-loop control problem. A guidance control law is derived to navigate a user based upon the received location-dependent signatures from its current location to its destination [1]. Possible signatures derived from such sources of opportunity include differential time-of-arrival (TOA), differential angle-of-arrival (AOA), and spatial distribution (gradients) of signal strength and signal power spectra. The effects of such factors as the geometry of detectable acoustic sources, and temporal and spatial sampling rates, on navigation performance will be assessed. A trade study will help determine the optimal sampling interval and control gain, leading to an efficient fusion of multiple signals with the best spatial discrimination for navigation. The structure of the paper is organized as follows. In Section II, we first describe the geotag and matching algorithms for the underwater coordinate-free navigation system to implement MUSNav, and discuss the control laws used for system modeling. In Section III, we discuss the desirable signal characteristics for underwater navigation and study the temporal and spatial variation properties of location features from acoustic signals, the geomagnetic field, and the gravitational field. System performance is then evaluated in Section IV using simulated acoustic signals. In Section IV, the trade space between the number of measurements, position accuracy, computational complexity, and total path length is studied by varying the defined 1 parameter. For instance, the choice of wi = 1 and p = 2 leads to the Euclidean distance (2-norm). control parameters. Finally, concluding remarks are made in Section V. II. SYSTEM MODEL 1 n D(x' , x k ) One requirement of MUSNav is that location-dependent parameters at the destination, or at waypoints, must be known. Therefore, the MUSNav algorithm developed here requires a calibration step, or training phase, to obtain the location-dependent parameters at the destination, which can be converted to a location signature, or geotag. n i 1 1 | x ' i xi ( k ) | p wi ( k ) 1/ p (1) Based on the calculated distances between T’ and a previously store Tk, the location of a signature that gives the minimum distance is chosen as the location for T’ as: k* (2) arg min D( x' , x k ) k {1,...,K } A. Geotag Generation It is necessary to set a threshold to guarantee that the location can be registered at the calibration phase. To emphasize relative importance of and confidence on individual location parameters and to account for correlation between the elements, a modification to NNM is made, named the weighted nearest neighbor method (WNNM) [4], which makes use of the covariance matrix Ck at xk as: An underwater coordinate-free system requires two steps. In particular, a training phase and a matching phase. The geotag-based navigation and positioning technique highly depends on the initial training phase. The training phase involves a user receiver collecting location-dependent parameters at the desired destination, or waypoints, along the path. D 2 ( x' , x k ) The geotags associated with the trained locations, indicated as grey dots in Figure 1, are computed based on the recorded location information and stored in a database for future use. The second matching phase is then employed for navigating the underwater vehicle, or vessel, to get to the destination with the assistance of the trained waypoints. In the matching phase, the underwater vehicle derives geotags using received location-dependent parameters, and matches these with the pre-computed ones in the database to determine the heading direction. Let Tk be the geotag derived from the calibration phase at a unique kth geographic location and T’ be the geotag derived during the navigating phase at the same location. x ' ) T Ck 1 ( x k x' ) (3) Parametric approach (soft metric). A parametric approach measures distance between location tags with the help of a Bayesian conditional probability to determine locations [5]. At the calibration phase, not only the location parameters but also their covariance matrix are estimated. The latter is used to help construct a more robust decision rule for verification. When the distance between location parameters is weighted using a Gaussian distribution, we use the probability density function shown in Equation (4a) to compare the likelihoods. Since Ck characterizes the location parameter xk, it only depends on xk subject to seasonal adjustment to reflect differential effects on the elements of the location parameter. The location tag Tk that gives the maximum likelihood is selected for the measured vector x’ (T’). There are many different geotag generation methods, as well as corresponding matching algorithms. The methods differ in geotag representation, computational efficiency, and ease of practical implementation. In this paper, we apply the MUSNav method [1] that considers the extracted location-dependent parameter vector xk = [x1(k), x2(k), …, xn(k)] as a geotag Tk with n elements. The MUSNav technique is similar to location fingerprinting except we also use various location-dependent parameters other than just the received signal strength [2]. There are two different approaches for the matching process, that is, a non-parametric approach and a parametric approach. Non-parametric approach (hard metric). A nonparametric approach is the nearest neighbor method (NNM) [3], which is commonly used for indoor location estimation and pattern matching. The algorithm calculates the distance between the location vector measured at a location during verification T’ and one of the previously stored vectors in the database {Tk, k = 1, …, K}. A generalized distance metric D(T’, Tk) = D(x’, xk) is defined in Equation (1) where wi(k) is an element-wise weighting factor and p is the norm (x k P( x' | x k ) 1 2 det(C k ) e 1 ( x ' x k )T C 1 ( x ' x k ) 2 (4a) When their components are equally important, the likelihood is given by: P ( x' | x k ) 1 n ni 1 1 2 exp i (k ) ( x' i xi (k )) 2 2 (4b) 2 i (k ) B. Multisensor Underwater Signature-based Navigation We formulate the MUSNav problem as a closed-loop control problem. A guidance law is derived to guide a receiver based upon the pre-computed and measured geotags. The steps to navigate a receiver from one location to another are given in the flow chart in Figure 2. Once the receiver computes the geotag associated with the current location from sensors, the user decides the heading direction and how far he will move before taking new measurements. 2 The guidance process consists of a first step of coarse navigation and a second step of vernier navigation, which are equivalent to coarse acquisition and fine tracking. At the coarse navigation phase, a receiver first sweeps all directions, from 0 to 360 degrees, and computes the geotags at various directions. Figure 3 - MUSNav control parameters res, s, and p C. Guidance Law A closed-loop control is utilized to implement the navigation method to guide and control a receiver’s trajectory. The feedback control block diagram is given in Figure 4. Figure 2 - MUSNav flowchart The heading accuracy depends on the sweeping angle interval, or the number of geotags around the starting position. The initial heading direction is the one that gives the minimum distance from the target geotag. Once the initial heading is determined, the receiver enters the refined search or tracking phase, and computes geotags along the path. These geotags are used to further adjust the heading towards the destination. The number of computed geotags depends on the tracking step size, which is the distance from one measurement to the next. We define a number of control parameters, which can be specified by users and are essential to navigation performance. Figure 3 illustrates the defined control parameters. The first control parameter is the initial heading resolution, res. A smaller angular resolution produces a more accurate initial heading. As the receiver enters the tracking phase, the parameters of tracking step size rs and search angle s control the tradeoff between travel distance, convergence speed, and computational loading. A finer tracking step size requires more location-dependent measurements, resulting in a more accurate heading direction but with a higher computational burden. On the other hand, less accurate angle information resulting from a coarse tracking step size might produce a longer trajectory, or path length, but with a lower computational demand. The last parameter is the geotag threshold, p, which controls the desired location convergence. As the MUSNav technique relies on the Euclidean distance between the stored geotag and the measurements, the convergence threshold is important as a tradeoff between convergence speed and estimated position accuracy. Figure 4 - MUSNav Guidance and control block diagram The design of an error loop discriminator and a loop filter characterizes the geotag tracking phase. These functions determine two most important performance characteristics of the loop design, which are the loop thermal noise error and the maximum dynamic stress threshold. The error discriminators used in this paper are linear and piece-wise, as shown in Figure 5. The location-dependent parameter, time-of-arrival (TOA) , is chosen as an example to illustrate the implementation of the control law on MUSNav. The loop discriminator, or spatial error = i – d and , discriminator (SED), can be modeled as which respectively are the distance of the parameter between the current measurement and the target (the measurements at destination), and its power. The absolute distance between the current location and the destination, x, is estimated from the SED as well as the control gain selected. The objective of the loop filter is to reduce noise and control the convergence speed to the desired geotag. As shown in Figure 4, the output of the loop filter is fed back to the original input to produce the spatial error. There are many types of loop filters, each having different characteristics. For instance, the first order loop filter is sensitive to velocity stress while the second order filter is sensitive to acceleration stress. In this paper, we do not focus on the dynamics of the underwater vehicle. Thus, a simple first order loop gain is applied. 3 The signal propagation from a stationary transmitter remains relatively stable as location features for a given position. The resulting spatial discrimination of signal patterns is akin to standing waveforms produced by reflection, diffraction, refraction, and scattering of acoustic signals in the environment. Such a time-invariant property of localspecific location signatures, also known as fingerprinting, has been used for positioning [10, 11]. Figure 5 - Linear (left) and piece-wise loop discriminators III. SIGNAL CHARACTERISTICS A number of signal sources such as acoustic signals, the geomagnetic field, the gravity field, and bathymetric features can be used for the underwater coordinate-free navigation and guidance. Nowadays, many researches make use of these natural signals that are not designed for navigation, rely on location servers, and monitor units to calibrate the radio-based transmitter timing biases. The calibration data is provided to users via dedicated data links [6, 7]. Similar work on cooperative position location using periodic codes in broadcast digital transmissions was studied. Being cooperative, the teammates have a means to communicate to one another via a wireless data link to coordinate their activities, exchange data, and perform mutual aiding in the form of cooperative referencing and calibration [8]. However, this paper does not provide a solved position fix using conventional methods, but instead relies on the spatial distribution of a variety of location features extracted from different systems and sources. A. Acoustic Signal The underwater environment, especially in sea water, induces conductivity, which results in rapid attenuation of electro-magnetic signals at high frequencies. Thus, acoustic signals are best supported at low frequencies, and in a frequency range of between 10 and 15 kHz [9]. Navigation in underwater environments presents a number of unique challenges due to the complexity of the environmental characteristics. The background noise, although often characterized as Gaussian, is not white, but has a decaying power spectral density. Surface waves, internal turbulence, fluctuations in the sound speed, and other small-scale phenomena contribute to random signal variations, as well as multipath in the received signals. The presence of Doppler spread impacts acoustic energy in the sea, due to source/receiver motion, as well as motion of the water waves that may not be well represented by a simple Doppler shift. Frequency-dependent propagation losses result in relative small available bandwidth for acoustic transmissions, and potentially large delay variations leads to strong frequency selectivity, which may be time-varying. As a result, there are no standardized models for the acoustic channel fading, and experimental measurements are often made to assess the statistical properties of the channel at particular testing sites. The usable location-dependent parameters extracted from acoustic signals are signal strength, time-of-arrival (TOA), time difference of arrival (TDOA), and angle-of-arrival (AOA). Range measurements as well as differential ranges between the user and acoustic sources (obtained from TDOA) are used to determine the user location via multilateration in conventional positioning systems. Similarly, the user location can also be estimated from AOA measurements via triangulation. A sector angle is the difference in AOA between two transmitters. Other possible features, which are not commonly used in navigation systems, are short-time energy, spectral flux, and spectral centroid [12]. The short-time energy of a frame of collected signal waveforms is defined as the sum of squares of the signal samples normalized by the same frame length. Spectral flux is a measure of how quickly the power spectrum of a signal is changing, and is location-dependent. The spectral centroid is a measure used in digital processing to characterize a spectrum. Perceptually, it has strong correlation with the ―brightness‖ of a sound, and can be calculated as the weighted mean of the frequencies. With a particular set of transmitters, the received parameters are location-dependent and have a unique geographic distribution. Figure 6 illustrates the color contours of the geographic distribution of two parameters – differential range and sector angle. A differential range is the difference in absolute ranges of two transmitters measured at a receiver, while a sector angle is the angle formed by two transmitters and a receiver. Three arbitrary signal sources were chosen and indicated as s1, s2, and s3. The color contour changes gradually from red, high amplitude, to blue, low amplitude. For instance, the receivers on the baseline between two transmitters have the highest sector angle of 180˚. As a receiver moves away from the baseline, the sector angle decreases. The geographic distribution of the parameters indicates the location-dependent uniqueness, which is essential to the underwater coordinate-free system. The use of more usable parameters, as well as a larger number of transmitters improves the spatial discrimination. B. Geomagnetic Field The Earth’s magnetic field [13], also called the geomagnetic field, is generated within its molten iron core through a combination of thermal movement, the Earth’s daily rotation, and electrical forces within the core. Many research efforts have led to models for the geomagnetic field. The geomagnetic reference model is the basis for establishing the declination and its variation across the surface of the globe. 4 Contours of Differential Ranges to 3 Sources 7000 6000 5000 y [m] 4000 s3 3000 2000 1000 0 -1000 -1000 s1 0 s2 1000 2000 3000 x [m] 4000 5000 6000 7000 Figure 7 - World magnetic chart for declination generated from 1995 Epoch IGRF model [Picture courtesy: nationalatlas.gov] Contours of Sector Angles to 3 Sources 7000 6000 C. 5000 y [m] 4000 s3 3000 2000 1000 0 -1000 -1000 s1 0 s2 1000 2000 3000 x [m] 4000 5000 6000 Figure 6 - Geographic distribution of location features: differential range (top) and sector angle (bottom) There are a number of geomagnetic field-related features that can be used for underwater coordinate-free navigation. The total magnetic field can be divided into several components: Declination (D) indicates the difference (in degrees) between the heading of the truth north and the magnetic north. Inclination (I) is the angle (in degrees) of the magnetic field above or below horizontal. Horizontal intensity (H) defines the horizontal component of the total field intensity. Vertical intensity (Z) defines the vertical component of the total field intensity. Total intensity (F) is the strength of the magnetic field. The intensity and structure of the Earth’s magnetic field vary both temporally and spatially. The temporal variation is slow but reflects influences on the flow of thermal currents within the iron core. As a result, the models of the magnetic field, as well as the location features for the MUSNav database, need to be updated periodically. The magnetic field strength, direction and change rates are predicted every five years for a 5-year period. Figure 7 illustrates the geographical distribution of declination generated from the International Geomagnetic Reference Field (IGRF) model. Gravity Gravity, and the associated acceleration produced by the Earth, varies with latitude, altitude, topography, and geology [14]. Due to the outward centrifugal force produced by the Earth’s rotation, and the Earth’s equatorial bulge, latitudes near the equator have high gravity as opposed to the polar latitudes. In addition, gravity decreases with latitude as greater latitudes indicate greater distance from the Earth’s center. Local variations in topography and geology cause fluctuation in the Earth’s gravitational field. The spatial variation of the gravitational field can benefit the design of coordinate-free navigation and add more spatial discrimination in the computed geotags or location signatures. IV. PRELIMINARY SIMULATION RESULTS A. Simulation Scenario In this paper, we use simulated acoustic signals as a case study to evaluate the performance of MUSNav. The center frequency of the signal is chosen to be 10 kHz. A simple analytical propagation model of acoustic signals is used to estimate the received signal strength at the user’s location (see Equation (5) below). The location features, which include differential range, sector angle, and signal strength, from four emitters are used to compute the geotags. There are different ways to estimate range or differential range measurements depending on the acoustic signal architecture. If a pseudorandom code or a timing sequence is embedded in the signals, correlation can be applied to detect the incoming signal and estimate the differential TOA. The differential range measurements require the different base stations to be synchronized. Without synchronization, external timing information, such as GPS, can be used to calculate the biases between different stations. A coarse way to estimate ranging information is to convert the received signal strength to the range with a proper propagation model. Sector angle measurements can be obtained using either an antenna array or multiple sensors. Propagation without obstacles is an ideal case. Several factors that include reflection, diffraction, and scattering 5 should be taken into account when acoustic signals encounter obstacles. In this paper, we assume that the path loss depends on absorption, which is the transfer of acoustic energy into heat, and spreading loss, which increases with the propagation distance. The overall path loss can be written as [9]: A(d , f ) d dr trajectory (shown as blue line in Figure 8), or path length, while the sector angle gives the longest. We next compare the performance of two different error discriminators: maximum difference and Euclidean norm. The results are given in Figure 9. b a( f ) d dr User Trajectory (5) s2 6000 Base stations Destination Diff. range, Norm Diff. range, Max Difference Sector angle, Norm Sector angle, Max Difference Signal strength, Norm Signal strength, Max Difference s1 4000 where f is the signal center frequency, and d is the transmission distance taken in reference to some dr. The path loss exponent b models the spreading loss, and its usual values are between 1 and 2, with 1 for cylindrical spreading and 2 for spherical spreading. The absorption coefficient a(f) can be obtained using an empirical formula [15]. Hence, Equation (4) is used to derive the received signal strength for geotags. y [m] 2000 -2000 -4000 s4 -8000 First, we consider the ideal case where there is no noise or other error sources added to the received location measurements. Figure 8 plots and compares the estimated trajectories of a receiver derived from the computed geotags using differential range, sector angle, and signal strength. The blue dots represent four emitters. The destination location is shown as a red star marker. Three different paths from differential range, sector angle, and signal strength are given in green, magenta, and black, respectively. y [m] -2000 0 2000 4000 6000 8000 Number of way points 200 s2 150 Diff. range, Norm Diff. range, Max Difference Sector angle, Norm Sector angle, Max Difference Signal strength, Norm Signal strength, Max Difference 100 50 s1 0 12 Base stations Destination Differential range Sector angle Signal strength All three parametes 2000 0 -2000 -4000 s3 s4 -8000 -6000 -4000 250 User Trajectory 6000 -6000 x [m] 8000 -6000 s3 -6000 B. Noise-free Case 4000 0 -4000 -2000 0 2000 4000 6000 x [m] Figure 8. Comparison of location-dependent parameters: different range, sector angle, and signal strength Constant control gains are applied for all three cases. The angle interval for initial probing is ten degrees. The coarse tracking interval step is 1000 meters, while a fine tracking interval step of ten meters is used when the receiver approaches close to the destination. The Euclidean norm is chosen to compute the spatial error discriminator. A combination of all three parameters provides the shortest 12.5 13 13.5 14 14.5 Trajectory distance [km] 15 15.5 Figure 9. Comparison of two spatial error discriminators: maximum difference and Euclidean norm Figure 9(top) shows the trajectories of the receiver paths using the three parameters and the two error discriminators. Figure 9 (bottom) summarizes the tradeoff in performance between path distance and the number of way points, or equivalently, the computational loads. We observe that the differential range with a norm discriminator gives the shortest trajectory distance, and the sector angle with norm discriminator results in the longest distance. Further, the differential range with differencing discriminator requires the highest computational power, and the signal strength with norm discriminator has the least measurements. A combination use of different parameters would improve the spatial discrimination of the computed geotag, producing a shorter trajectory length [16, 17, 18]. The control gain plays an important role in the convergence to the target geotag or the destination. We evaluate the trade 6 space between the computational demand and trajectory distance by varying the linear gain, shown in Figure 10. User Trajectory 8000 s2 User Trajectory 6000 s2 s1 6000 y [m] 4000 Base stations Destination Control coeff.=0.01 Control coeff.=0.1 Control coeff.=0.2 Control coeff.=0.3 Control coeff.=0.4 2000 y [m] Base stations Destination No noise = 1m = 2m = 3m = 4m = 5m 4000 s1 0 -2000 2000 0 -2000 -4000 s3 -6000 -4000 s4 -1 s3 -6000 -0.5 0 0 2000 4000 6000 400 16 200 14 0 0.15 0.2 0.25 0.3 0.35 12 0.4 Linear control coefficient Figure 10. Linear control gain study In this study, we use the location-dependent feature, differential range. The linear control gain varies from 0.01 to 0.4. Similarly, the top plot in Figure 10 gives the trajectories of the selected control gains, and the bottom plot shows the tradeoff between the number of measurements and the trajectory length. The simulation results show that the linear gain is proportional to the trajectory length but inversely proportional to the computational power. A faster convergence system would aim for a smaller control gain, whereas, a more efficient system prefers a large control gain. As expected, the higher the noise floor, the slower the convergence speed. A large noise floor increases the spatial error, which delays receiver convergence to the desired geotag. A longer trajectory requires more measurements of geotags to tune the heading direction. As a result, random noise increases computational loading and trajectory length. One solution to improve the trajectory length, by minimizing the effects of random noise and other error sources, is to reduce the linear control gain as illustrated in Figure 12. The green path represents a linear control gain of 0.01, which is the smallest amongst all gains shown and gives the shortest path length. Aforementioned, the tradeoff of using a small control gain is between high computational loads or more location measurements along the path. The use of a small gain is equivalent to tightening up the noise bandwidth. Although the path length is reduced, the number of way points is increased, which leads to a longer time to reach the desired destination. User Trajectory s2 6000 s1 4000 2000 y [m] 18 Trajectory distance [km] Number of way points Tradeoff Analysis 0.1 4 Figure 11. Random noise lowers convergence speed 600 0.05 x 10 8000 x [m] 0 1 x [m] s4 -8000 -6000 -4000 -2000 0.5 Base stations Destination Control coeff.=0.01 Control coeff.=0.02 Control coeff.=0.05 Control coeff.=0.08 Control coeff.=0.1 0 C. Random Noise Case -2000 In practice, there is always random noise and other error sources that contaminate the received acoustic signals. In the subsection, we add random noise to the simulated signals and examine the resulting change in system performance. The result is illustrated in Figure 11. Differential range is used in the simulation for the different noise level comparison. The same location-dependent parameter, differential range, is used. The linear control gain is chosen to be 0.1. The standard deviation, , of the parameter ranges from 1 to 5 meters. -4000 -6000 s3 s4 -8000 -6000 -4000 -2000 0 2000 4000 6000 8000 x [m] Figure 12. Reducing linear control gain improves the convergence speed 7 V. CONCLUSIONS We formulated a multisensor underwater signature-based navigation (MUSNav) technique that uses locationdependent parameters from the received acoustic signals in underwater environments. Instead of providing a position fix such as longitude, latitude, and altitude, the method guides a receiver using the trained destination geotags and the location measurements along the path. A combined use of acoustic signals, the geomagnetic field, and the gravitational field can improve the spatial discrimination of derived geotags as well as navigation system performance. The proposed MUSNav algorithm supports a wide range of location-base applications, for example, animal tracking. Figure 13 shows the recorded path of migrating sea turtles, which are believed to perform long-distance navigation using geo-magnetic field based compass sensing and map sensing [19, 20]. The actual routes strongly resemble the above simulated trajectories. strength can significantly improve the spatial discrimination of the computed geotags. As a result, we can achieve better precision in the final estimated target location. However, the use of more location features increase the probability of failure due to the increased number of error sources, especially in underwater environments. Examples of error sources are random noise, biological noise, noise generated from ships, obstacles inside the water, and temporal change of water waves among others. In addition, there are errors originating from acoustic transmission system operations. Acoustic transmitters might be offline due to maintenance or other implementation issues. As a result, a geotag will not be reproducible when there is an insufficient number of location features received. Such practical issues as operational continuity and likelihood of failure to map into the desired geotag, will be further studied by developing error-tolerant algorithms to reduce the system risk, thus increasing the robustness of the computed geotags. We will evaluate such errors in both Euclidean distance and Hamming distance and develop error-tolerant algorithms that can account for various types of error sources. In addition, we will compare the parametric and non-parametric geotag generation approaches in our future simulations. Simulations provide us with the insights from an analytical point of view. The practical aspects of signal processing and system implementation are better understood using real data, which is our future plan to implement and test our approach. Moreover, we will investigate and use more location features, such as geomagnetic field, to improve the spatial discrimination of computed geotags. REFERENCE Figure 13. Trajectories of migrating sea turtles [20] A training or calibration phase is required to implement MUSNav. The geotag associated with the destination is stored in a database for future matching. A closed loop control law is used to formulate navigation and guidance. The process consists of a coarse acquisition to determine the heading direction and fine tracking to approach the destination. Both the error discriminator and loop filter are essential to reduce the noise and increase convergence speed. We evaluated the MUSNav algorithm using a simulated acoustic signal as a case study. Several trade spaces were studied and analyzed. The total path length, computational load, and position accuracy can be traded off against each other by varying the control parameters, angular resolution, tracking steps, and the threshold for convergence. In addition, we observed that the trajectory length is proportional to the linear gain whereas the computational complexity is inversely proportional to the linear gain. A combined use of various location features such as TDOA, signal strength, sector angle, gravity, and magnetic field [1] C. Yang and D. Qiu, Geo-coordinate Free Radio Navigation and Guidance Using Measured LocationSpecific Parameters, U.S. Patent Application No. US 61/39618, 2010. [2] P. Bahl and V. N. Padmanabhan, ―RADAR: An inbuilding RF-based user location and tracking system,‖ IEEE Infocom 2000, Tel Aviv, Israel, 26-30 Mar 2000, vol. 2,pp. 775-784. [3] T. Roos, P. Myllymaki, H. Tirri, P. Misikangas, and J. Sievanen, ―A probabilistic approach to WLAN user location estimation,‖ Inter. J. of Wireless Information Networks, vol. 9, no. 3, pp. 155-164, July 2002. [4] S. Saha, K. Chaudhuri, D. Sanghi, and P. Bhagwat, ―Location determination of a mobile device using IEEE 802.11b access point signals,‖ IEEE Wireless Communications & Networking Conference (WCNC), New Orleans, Louisiana, US, Mar 16-20 2003, vol.3, pp. 1987-1992. [5] Xia, H. H., "A simplified model for prediction path loss in urban and suburban environments," IEEE Trans. Veh. Technol., Vol. 46, No. 4, 1040-1046, Nov. 1997. 8 [6] M. Rabinowitz and J.J. Spilker, Jr., ―A New Positioning System Using Television Synchronization Signals,‖ IEEE Trans. Broadcasting, 51(1), 51–61, March 2005. [7] T. Young, ―TVGPS, A Hybrid Location and Timing Solution for GPS-Challenged and GPS-Denied Environments,‖ Proc. of the ION-JSFE JNC 2009, June 2009, Orlando, FL. [8] Y. Zhao, Vehicle Location and Navigation Systems, Artech House, Boston, MA, 1997. [9] M. Stojanovic and J. Presisig, ―Underwater acoustic communication channels: propagation models and statistical characterization,‖ IEEE Communications Magazine, Jan 2009. [10] R. Schmidt, Andre Gagnon, and D. Allcock, ―Maritime Containers Tracking Trial Results,‖ IEEE AES Magazine, Sep 2009. [11] H.R. Everett, Sensors for Mobile Robots: Theory and Application, AK Peters, Ltd., Wellesley, MA 1995. [12] G. Padmavathi, D. Shanmugapriya, and M. Kalaivani, ―Acoustic signal based feature extraction for vehicular classification.‖ IEEE Advanced Computer Theory and Engineering (ICACTE), Aug. 2010. [13] G. Hulot, et. al., ―The magnetic field of planet Earth,‖ Space Science Reviews, Vol. 152, Issue 1-4, pp. 159222, Sep 2010. [14] R. Forsberg, M.G. Sideris, and C.K. Shum, ―The gravity field and GGOS,‖ Journal of Geodynamics 40, pp. 387-393, 2005. [15] L. M. Berkhovskikh and Y. Lysanov, Fundamentals of Ocean Acoustic, Springer, 1982. [16] D. Qiu, R. Lynch, and C. Yang, ―Location-Dependent RF Geotags for Positioning and Security,‖ Proc. SPIE, Vol. 8061, 2011. [17] D. Qiu, M. Miller, S. DeVilbiss, T. Nguyen, and C. Yang, ―Geo-Coordinate Free Guidance and Navigation,‖ ION-GNSS, Portland, OR, Sept. 2010. [18] C. Yang, T. Nguyen, M. Miller, and D. Qiu, ―Multiscale Location-Specific Features for Positioning and Navigation: Bio-Inspired Navigation,‖ Autonomous Weapons Summit & GNC Challenges for Miniature Autonomous Systems Workshop 2010, October 2010, Fort Walton Beach, FL. [19] S. Johnson and K.J. Lohmann, ―The Physics and Neurobiology of Magnetoreception,‖ Nature Review: Neuroscience, Vol. 6, Sept. 2005. [20] URL: http://www.deee.unipi.it/islameta/Sea%20Turtle%20Na vigation.html. Di Qiu received her B.S. from University of California, Los Angeles, in 2003, and her M.S. and Ph.D. both in Aeronautics and Astronautics Engineering from Stanford University in 2004 and 2009, respectively. She has worked on ionospheric threat modeling, signal authentication, locationbased security modeling and demonstration, information theory, and parametric fuzzy extraction. Dr. Qiu’s current research interests include navigation using signals of opportunity (SOOP), sensor data fusion, state estimation, and pattern classification. Robert Lynch is a senior research scientist with the Naval Undersea Warfare Center in Newport, RI. He holds BS and MS degrees, both in Electrical Engineering, from Union College, Schenectady, NY, and a Ph.D. in Electrical Engineering from the University of Connecticut, Storrs, CT. His research interests are in the areas of pattern recognition and classification, detection, data fusion, tracking, and signal processing. Dr. Lynch is Vice President Communications of the International Society of Information Fusion, and is Managing Editor of the Journal of Advances in Information Fusion. Dr. Lynch is a Senior Member of the IEEE, and is an Associate Editor of the IEEE Transactions on Systems, Man, and Cybernetics Part B, Cybernetics. He is a former recipient of the NAVSEA Excellence in Science Award, and the Federal Laboratory Consortium’s Excellence in Technology Transfer Award. Dr. Lynch is an Adjunct Lecturer in the Electrical and Computer Engineering Department at the University of Connecticut. Erik Blasch is a Fusion Evaluation Tech Lead for the Air Force Research Laboratory, Rome, NY and a Reserve Officer at the Air Force Office of Scientific Research (AFOSR). He received his MSEE (1997) and Ph.D. in Electrical Engineering (1999) from Wright State University, a MS in Mech. Eng (1994) and MS in Industrial Eng. (1995) from Georgia Tech, and a BSME from MIT in 1992 among other advanced degrees in engineering, health science, economics, and business administration. He is a past President of the International Society of Information Fusion (ISIF), a member of the IEEE AESS Board of Governors, a SPIE Fellow, and active in AIAA and ION. His research interests include target tracking, sensor and information fusion, automatic target recognition, biologically-inspired robotics, and controls. Chun Yang received his Bachelor of Engineering from Northeastern University, Shenyang, China, in 1984 and his title of Docteur en Science from Université de Paris, Orsay, France, in 1989. After two years of postdoctoral research at University of Connecticut, Storrs, CT, he has been with Sigtem Technology, Inc. since 1994. He has been working on adaptive array and baseband signal processing for GNSS receivers and radar systems as well as on nonlinear state estimation with applications in target tracking, integrated inertial navigation, and information fusion. Dr. Yang is an Adjunct Professor of Electrical and Computer Engineering at Miami University. He is the member of the ION, IEEE, ISIF, and SPIE. 9