This invention relates to methods and apparatus for locating radio frequency emitters, and more particularly to such methods and apparatus for locating continuous waveform emitters.
In order to meet the requirements of Network Centric Warfare, existing sensing platforms, such as the E-2 early warning, command and control aircraft, will have to be leveraged through the use of emerging architectures and technologies. Specifically, they will be called upon to perform fast, accurate location of targets, including traditional threat emitters such as radars, which use waveforms that have widely separated pulses with well-defined leading edges. In addition, fast, accurate location of an emerging set of target emitters, which have more continuous waveforms, is becoming increasingly important. These non-traditional target emitters include communications terminals that might be associated with terrorists, drug dealers, urban combatants, and emergency-first-response rescue personnel. Such communications terminals may include cell phones, PDAs, laptop computers, and other devices.
To address these target emitters, legacy platforms will have to be equipped with the sensor and sensor-management capability, and the communications infrastructure to support multi-platform collaborative targeting. In a collaborative targeting system, participating sensor platforms can contribute multiple measurements that are dependant on the target emitter's location, such as the Angle of Arrival (AOA). By taking advantage of the geometries available with multiple platforms, the geometric dilution of precision (GDOP) resulting from finite measurement accuracy can be avoided, and fast yet accurate location can be obtained from measurements such as AOA.
Although AOA can generally be measured for most emitter types, other precision measurements, such as Time Difference of Arrival (TDOA), are traditionally applied to radar emitters, because radars emit easy-to-distinguish pulses with leading edges that enable time of arrival to be measured. Applying techniques to continuous waveform emitters, such as communications terminals, is less straightforward, however, due to the absence of a well-defined event, such as the leading edge of a pulse that would enable measurement of the time of arrival. There is a need, then, for a method for passively determining the location of continuous wave emitters in multi-platform network centric systems.
This invention provides an apparatus comprising a first sensor mounted on a first platform for sampling a first portion of a continuous waveform occurring in a time window and for producing a first signal sample, a second sensor mounted on a second platform for sampling a second portion of the continuous waveform occurring in the time window for producing a second signal sample, and a processor for determining time difference of arrival measurements and for applying a maximum likelihood estimation process to combine multiple time difference of arrival measurements between multiple pairs of platforms, to estimate the location of an emitter of the continuous waveform.
In another aspect, the invention provides a method comprising the steps of: using a first sensor mounted on a first platform to sample a first portion of a continuous waveform occurring in a time window to produce a first signal sample, using a second sensor mounted on a second platform to sample a second portion of the continuous waveform occurring in the time window to produce a second signal sample, time shifting the first signal sample with respect to the second signal sample, correlating the first and second signal samples to determine a time difference of arrival measurement, and applying a maximum likelihood estimation process to combine multiple time difference of arrival measurements between multiple pairs of platforms, to estimate the location of an emitter of the continuous waveform.
In yet another aspect, the invention provides an apparatus comprising a first sensor mounted on a first platform for sampling a first portion of a continuous waveform occurring in a time window and for producing a first signal sample, a second sensor mounted on a second platform for sampling a second portion of the continuous waveform occurring in the time window for producing a second signal sample, and a processor for determining frequency difference of arrival measurements and for applying a maximum likelihood estimation process to combine multiple frequency difference of arrival measurements between multiple pairs of platforms, to estimate the location of an emitter of the continuous waveform.
In still another aspect, the invention provides a method comprising the steps of using a first sensor mounted on a first platform to sample a first portion of a continuous waveform occurring in a time window to produce a first signal sample, using a second sensor mounted on a second platform to sample a second portion of the continuous waveform occurring in the time window to produce a second signal sample, determining frequency difference of arrival measurements from the first and second signal samples, and applying a maximum likelihood estimation process to combine multiple frequency difference of arrival measurements between multiple pairs of platforms, to estimate the location of an emitter of the continuous waveform.
Referring to the drawings,
One of the functions performed in the system of
3GCT technology focuses on achieving fast, accurate geolocation of targets using passive location techniques implemented in a multi-platform collaborative configuration in which the participating sensing platforms, which have advanced Electronic Support Measures (ESM) capabilities, work in a coordinated fashion, and are connected by a broadband Internet Protocol (IP) communications infrastructure. By taking advantage of advanced ESM capabilities such as digital sampling of target emitter waveforms at each sensor platform, and broadband connectivity to transport these digitized samples to a common processing node, the application of multi-platform precision geolocation techniques such as TDOA and FDOA can be extended beyond the typical set of target emitters such as radars, which have waveforms with widely spaced pulses providing well-defined events in time, to an expanded set of target emitters, including emitters with continuous waveforms that do not contain well-defined events in time, such as communications terminals and other devices. The use of a Generalized Recursive Maximum Likelihood Estimation (GRMLE) method to combine any number of measurements and mixture of measurement types (TDOA, FDOA, AOA, etc.) further enables the application of multi-platform location to continuous waveform emitters.
In the pulse waveform example of
With continuous waveform target emitters, such as communications terminals, there is no well-defined event in time, such as a pulse leading edge, to measure. However, if two measurement platforms both take digital samples of the target emitter's waveform at the same instant in time, or if they both take digital samples and time-tag the samples using a synchronized time reference such as GPS time, and those digitized samples are transmitted to a common processing node, the two digitized samples can be time shifted with respect to one another and a correlation between the two time shifted samples can be observed. When a peak in the correlation is found, the amount of time shift used to obtain this peak is equivalent to the TDOA of the target emitter from the point of view of the two measurement platforms.
The sample waveforms can be caused to line up by applying a time delay to create a peak in the correlation. When this peak is observed, the time delay used to obtain the peak is equal to the TDOA of the sampled signal between the two measurement platforms.
While
3GCT takes advantage of advanced ESM capabilities, and broadband connectivity, used in a multi-platform collaborative architecture, to provide passive, fast, precision geolocation of all target emitters, including communications terminals that might be used by terrorists, drug dealers, and urban combatants. By knowing the TDOAs between at least two pairs of platforms, and the locations of these platforms, the location of the emitter can be determined.
Each TDOA measurement defines a curve (specifically, a hyperbola) in two dimensions (or a hyperbolic surface in three dimensions) on which the emitter lies. If two TDOA measurements are available, the location of the emitter can be uniquely determined in two dimensions, as shown in previous work. In three dimensions, three measurements are needed. If there are more than two TDOA measurements, or if there is some combination of TDOA and other measurements that is greater than two (referred to as an “over-determined” condition), (or, in three dimensions, if there are more than three TDOA measurements, or if there is some combination of TDOA measurements and other measurements that is greater than three), then an optimal combination of the measurements, such as a Maximum Likelihood Estimate (MLE), can be used.
The architectural challenge in applying TDOA to locate communications terminal targets is to be able to transport the digitized sampled emitter waveform, which could be megabytes of data, depending on the target waveform and the geometry, over the collaborative targeting network. With potentially hundreds of targets needing to be located within a short span of time, this could overwhelm a conventional collaborative targeting network. With broadband digital connectivity however, such as Transformational MILSATCOM (TSAT), TDOA can be applied to communications terminal targets.
Frequency Difference of Arrival (FDOA) relies on the Doppler effect caused by motion of the sensor platform with respect to the signal source. In principle, when two measurement platforms both measure the Frequency of Arrival (FOA) of an emitter's waveform, the measurements will differ by a small amount due to the different platform-to-target speeds. In the case of pulsed emitters, the FOA can actually be the Pulse Repetition Frequency (PRF), which can be determined fairly accurately by measuring over a wide span of pulses. By then transmitting these two measurements over the network to a common processing node, FDOA can be derived from the pair of FOA measurements obtained from the pair of platforms. However, with emitters that emit continuous waveforms, in practice, it is extremely difficult to measure FOA with sufficient precision from two separate platforms to detect the small differences in frequency that FDOA relies on. An alternative is to allow the frequency difference to be detected at a common processing node that can jointly process the signals seen by each of the two sensor platforms. As in the case of TDOA for communications terminals, this approach requires broadband airborne connectivity.
In the case of frequency difference of arrival (FDOA), extremely small differences in measured frequency must be detected in order for this technique to be effective, so measuring the Frequency of Arrival (FOA) independently at a pair of platforms would not likely provide the accuracy needed to derive FDOA. However, if digital samples of the target emitter's waveform detected at a pair of measurement platforms are transported to a common processing node, the two samples can be processed jointly to detect small differences in frequency. Again, the broadband connectivity in concert with the capability to perform digital sampling, is required to enable this approach.
To extract FDOA by jointly processing the two samples of the emitter waveform at the common processing node, first, both signal samples would be transformed from the time domain to the frequency domain using a digital Fourier Transform. Then, one transformed signal is frequency shifted with respect to the transform of the other signal sample while the correlation between the two transformed samples is observed. The maximum correlation occurs at a frequency shift equal to the FDOA of the continuous waveform between the two collection platforms. By knowing the FDOAs between at least two pairs of platforms, and the locations of these platforms, the location of the emitter can be determined.
Each FDOA measurement defines a curve in two dimensions (or a surface in three dimensions) on which the emitter lies. If two FDOA measurements are available, the location of the emitter can be uniquely determined in two dimensions, as shown in previous work. In three dimensions, three measurements are needed. If there are more than two FDOA measurements (referred to as an “over-determined” condition), or if there is some combination of FDOA and other measurements that is greater than two, (or, in three dimensions, if there are more than three FDOA measurements, or if there is some combination of FDOA measurements and other measurements that is greater than three), then an optimal combination of the measurements, such as a Maximum Likelihood Estimate (MLE), is used.
In another embodiment, both TDOA and FDOA measurements can be taken and a combination of TDOA and FDOA information can be used to determine the location of the target emitter.
In addition to TDOA and FDOA, other measurement types, such as AOA, range, and even location estimates, are often available from sensing platforms. By taking advantage of a broadband communications infrastructure that connects all measurement platforms to a common processing node, any and all measurement types that are available can be utilized to generate an optimal location estimation that automatically takes into account measurement accuracy and geometry. Such integration of any and all measurement types available, without having to have prior knowledge of which type of measurements will be made or how much data will be collected, can be achieved through the use of a Maximum Likelihood Estimator.
The MLE, in summary, provides the flexibility to combine any number and any combination of measurements in order to obtain the best possible, or most accurate, geolocation estimate by using as many measurements as are available, and by combining mixed measurement types. Such flexibility is important in implementing 3GCT because in a diverse multi-platform system it is impossible to predict which measurement types or how many measurements will be available in locating any given target.
Unfortunately, due to the non-linear relationship between measurements and target location, a closed form expression for the MLE generally cannot be found. To allow the application of the MLE to 3GCT in the presence of mixed measurement types and in over-determined conditions, a novel Generalized Recursive MLE method was developed. In this method, an initial guess of the target's location is made, and then an approximate MLE is found based on a “linearized” approximation of the relationship between measurements and target location about the initial guess. This solution represents an approximation to the Maximum Likelihood Estimate (MLE) based on the linearized version of the problem. It becomes the next guess at the true MLE, and the process is repeated. A detailed description of this Generalized Recursive MLE method follows.
The Generalized Recursive Maximum Likelihood Estimator approach to location estimation can be used to obtain a location estimation that is optimal in the maximum likelihood sense, and can be used with any number of measurements and any mixture of measurement types, even if the total number of measurements available is greater than a critically constrained number. The critically constrained number of measurements is equal to the number of unknown variables that are being estimated; two in the case of two-dimensional location (the X and Y positions of the target emitter), and three in the case of three-dimensional location (the X, Y, and Z positions of the emitter). This estimation technique takes into account the measurement accuracy of each individual measurement; the dependency, or correlation, if any, between measurements; and the geometry.
The explanation provided herein assumes a two-dimensional location problem. This can be easily generalized, however, to the three-dimensional problem. It will be further evident that this approach can be applied to not only location problems, but to any estimation problem in which the estimate is based on measurements that are dependant on the quantities being estimated.
As stated above, the described location estimation problem involves location in two dimensions, resulting in two unknown quantities that will be estimated; the X and Y positions of the emitter: Xe and Ye. Several measurements are made from various sensors, and each measurement is dependant on the quantities to be estimated. For example, if the measurement is the angle of arrival (AOA) of the emitter's waveform at a measurement platform, then the relationship between the measurement and the quantities to be estimated is as follows:
where xp and yp are the x and y positions of the measurement platform, which are assumed to be accurately known through some navigation system, such as GPS.
In another example, if the measurement is the Time Difference of Arrival (TDOA) between two measurement platforms of the emitter's waveform, then the relationship between the measurement and the quantities to be estimated is as follows:
where xp1, yp1, xp2, and yp2 are the x and y positions of platforms 1 and 2, respectively, which are again presumed to be known accurately through some navigation system, and c is the speed of light.
In yet another example, if the measurement is the Frequency Difference of Arrival (FDOA) between two measurement platforms of the emitter's waveform, then the relationship between the measurement and the quantities to be estimated is as follows:
where xp1, yp1, xp2, and yp2 are the x and y positions of platforms 1 and 2, respectively, which are again presumed to be known accurately through some navigation system; vx1, vy1, vx2, and vy2 are the x and y speeds of platforms 1 and 2, respectively, which are also presumed to be known accurately through some navigation system; and c is the speed of light.
In still another example, if the measurement is the range between the measurement platform and the emitter, then the relationship between the measurement and the quantities to be estimated is as follows:
Range=√{square root over ((xe−xp)2+(ye−yp)2)}{square root over ((xe−xp)2+(ye−yp)2)}
where xp and yp are the x and y positions of the platform, which are again presumed to be known accurately through some navigation system.
In each of the above examples, the measurement is a function of the two quantities being estimated, xe and ye. All of the other terms in each of the examples are presumed to be known quantities, thus, each measurement represents one function of two unknowns. This set of functions of the two unknowns xe and ye can be expressed in general terms, where the total number of measurements is n, as follows:
m1=ƒ1(xe,ye)
m2=ƒ2(xe,ye)
.
.
.
mn=ƒn(xe,ye)
where mi is the ith measurement, and fi is the corresponding functional relationship between mi and xe and ye. Note that these measurements are those that would be obtained if the measurement devices or sensors used were perfect, or had no measurement error.
This set of n equations and two unknowns can be written using matrix notation as:
m=F(x)
where
and
Again, F(x) is the set of measurements that would be obtained by perfect sensors, that is, sensors that have no error. Since, however, no sensor or measurement device is perfect, each measurement has some error, Δm, associated with it, and the set of actual measured values is given by:
{tilde over (m)}=m+Δ{tilde over (m)}.
By modeling the set of measurement errors as a set of random variables, then the set of measurements is also a set of random variables, and the accuracy of the measurement devices or sensors can be categorized by the covariance matrix of measurement errors, CM, which is defined as:
CM=E{({tilde over (m)}−E{{tilde over (m)}})×({tilde over (m)}−E{{tilde over (m)}})T}
If the measurement errors are unbiased, then
E{{tilde over (m)}}=m
which gives
CM=E{(Δ{tilde over (m)})×(Δ{tilde over (m)})T}.
The diagonal elements of CM are the variances, or the square of the standard deviation of each measurement device, and reflect the accuracy of each measurement device. The off-diagonal elements of CM are the covariances between measurement devices, and indicates the dependency of the value of a measurement from one sensor on the value of a measurement from another sensor. Often measurements from different sensors are independent, and the off-diagonal elements are zero.
Given the covariance matrix CM, the joint probability density of measurements can be expressed as:
This joint probability density is centered around F(x), the set of measurements that would be obtained if there were no errors in the measurement devices. The objective, however, is, given a set of actual measurements {tilde over (m)}, find some estimate of the unknown quantities x. This estimate of x is called {circumflex over (x)}, and the Maximum Likelihood Estimate (MLE) is the estimate of x that maximizes the joint probability density p({tilde over (m)}) for the set of actual measurements {tilde over (m)}. The joint probability density of measurements in terms of the estimate of x is written as:
To find {circumflex over (x)} that maximizes p({tilde over (m)}), the above expression is differentiated with respect to {circumflex over (x)}, the result is set to zero, and then {circumflex over (x)} is solved. This results in the following system of equations:
JT({circumflex over (x)})CM−1F({circumflex over (x)})−JT({circumflex over (x)})CM−1{tilde over (m)}=0
where J, the Jacobian matrix, is given by:
In general, this yields a set of l equations for the l unknown quantities in {circumflex over (x)}. In the case of two-dimensional location, l=2. This Maximum Likelihood Estimate is the estimate that provides the values of the unknown quantities {circumflex over (x)} that are the most likely to have resulted in the measurements {tilde over (m)}.
A closed form solution to this set of equations, however, may not be easy to obtain, especially for higher values of l, given the general non-linear expressions in F({circumflex over (x)}). A novel approach is a Generalized Recursive MLE method in which an initial guess of {circumflex over (x)} is made, called {circumflex over (x)}0, and then the above system of equations is “linearized” about the “point” └{circumflex over (x)}0, y0┘, y0, where y0=F({circumflex over (x)}0), by replacing F({circumflex over (x)}) with a linear approximation of F({circumflex over (x)}) about the point └{circumflex over (x)}0, y0┘, and by replacing the dependant function J({circumflex over (x)}) with the constant J0=J({circumflex over (x)}0). This linearized set of equations is then solved for {circumflex over (x)}. This solution represents an approximation to the Maximum Likelihood Estimate (MLE) of x based on the linearized set of equations. It becomes the next guess at the true MLE of x, and the process is repeated.
A step-by-step procedure follows.
Step 1. Pick an initial guess for {circumflex over (x)} called {circumflex over (x)}0. Evaluate y0=F({circumflex over (x)}0) to give the point [{circumflex over (x)}0, y0].
Step 2. Create the constant matrix J0=J({circumflex over (x)}0).
Step 3. Create a linearized approximation to F({circumflex over (x)}) given by FL({circumflex over (x)})=J0·({circumflex over (x)}−{circumflex over (x)}0)+y0.
Step 4. “Linearize” the set of equations JT({circumflex over (x)})CM−1F({circumflex over (x)})−JT({circumflex over (x)})CM−1{tilde over (m)}=0 by replacing J({circumflex over (x)}) with J0 and F({circumflex over (x)}) with FL({circumflex over (x)}), giving
J0TCM−1FL({circumflex over (x)})−J0TCM−1{tilde over (m)}=0, or
J0TCM−1└J0·({circumflex over (x)}−{circumflex over (x)}0)+y0┘−J0TCM−1{tilde over (m)}=0.
Step 5. Solve the above linearized set of equations for {circumflex over (x)} to obtain an approximate MLE of x based on the set of equations linearized about the point [{circumflex over (x)}0, y0]:
{circumflex over (x)}={circumflex over (x)}0+[J0TCM−1J0]−1J0TCM−1[{tilde over (m)}−y0].
Step 6. This approximate MLE solution for x then becomes the next guess for {circumflex over (x)}, the true MLE of x, and the process is repeated. A general recursive expression for this approach is given by:
{circumflex over (x)}i+1={circumflex over (x)}i+[JiTCM−1Ji]−1JiTCM−1[{tilde over (m)}−yi].
While the invention has been described in terms of several embodiments, it will be apparent to those skilled in the art that various changes can be made to the described embodiments without departing from the scope of the invention as set forth in the following claims.
Number | Name | Date | Kind |
---|---|---|---|
4601025 | Lea | Jul 1986 | A |
5008679 | Effland et al. | Apr 1991 | A |
5302957 | Franzen | Apr 1994 | A |
5477230 | Tsui | Dec 1995 | A |
5568154 | Cohen | Oct 1996 | A |
5675553 | O'Brien et al. | Oct 1997 | A |
5774087 | Rose | Jun 1998 | A |
5920278 | Tyler et al. | Jul 1999 | A |
5990833 | Ahlbom et al. | Nov 1999 | A |
6021330 | Vannucci | Feb 2000 | A |
6061022 | Menegozzi et al. | May 2000 | A |
6255992 | Madden | Jul 2001 | B1 |
6577272 | Madden | Jun 2003 | B1 |
6806828 | Sparrow et al. | Oct 2004 | B1 |
7406434 | Chang et al. | Jul 2008 | B1 |
20030058924 | Darby et al. | Mar 2003 | A1 |
20050012660 | Nielsen et al. | Jan 2005 | A1 |
20050046608 | Schantz et al. | Mar 2005 | A1 |