EMITTER LOCALIZATION WITH A SINGLE RECEIVER

Abstract

In a system for determining a location of an emitter, a mobile frame is configured for movement relative to the emitter. A main receiver and a frequency mixing antenna are supported on the mobile frame at different locations on the frame but so that both move with the frame relative to the emitter. The frequency mixing antenna is configured to receive an emitter signal and output a frequency-mixed signal. The main receiver is configured to directly receive the emitter signal and receive the frequency-mixed signal. A processor is configured to determine a first Doppler frequency from the direct emitter signal and a second Doppler frequency from the frequency-mixed signal and to determine the location of the emitter in a defined search area based on the first and second Doppler frequencies. Multiple Doppler frequencies can be created also using multiple frequency mixing antennas to improve the localization resolution.

Claims

1. A system for determining a location of an emitter, the system comprising: a mobile frame configured for movement relative to the emitter; a main receiver supported on the mobile frame at a first location for movement with the mobile frame relative to the emitter; and a frequency mixing antenna supported on the mobile frame at a second location for movement with the mobile frame and the main receiver relative to the emitter, the second location being spaced apart from the first location; wherein the frequency mixing antenna is configured to receive an emitter signal from the emitter at the second location and output a frequency-mixed signal; and wherein the main receiver is configured to receive the emitter signal at the first location and receive the frequency-mixed signal.

2. The system of claim 1, wherein a first Doppler frequency is derivable from the emitter signal received by the main receiver and a second Doppler frequency different than the first Doppler frequency is derivable from the frequency-mixed signal.

3. The system of claim 2, further comprising a processor connected to the main receiver and configured to determine the first Doppler frequency and the second Doppler frequency.

4. The system of claim 3, wherein the processor is configured to determine the location of the emitter based on the first Doppler frequency and the second Doppler frequency using a direct position determination approach.

5. The system of claim 3, wherein the processor is configured to use one of a random memoryless search algorithm or a random memory search algorithm to reduce the grid search computation time for a defined search area.

6. The system of claim 3, wherein the processor is configured to execute chirp RX processing techniques to improve localization performance.

7. The system of claim 1, wherein the first location is spaced apart from the second location by a distance of less than 10 m.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0055] FIG. 1 is a schematic illustration of a conventional method for EMI TX localization;

[0056] FIG. 2 is a schematic illustration similar to FIG. 1 of a method for EMI TX localization according to the present disclosure;

[0057] FIG. 3 is an implementation block and spectrum diagram corresponding to the method depicted in FIG. 2.

[0058] FIG. 4 is a schematic illustration similar to FIG. 2 but showing a method from EMI TX localization in three-dimensional space according to the present disclosure.

[0059] FIG. 5 is a three-dimensional plot of a cost function for the localization scenario in FIG. 4 where the signal is unknown, SNR is 20 dB, and RX moves along any line parallel to y-axis for any given x between Δd=1 m and 10 km and z=200 km.

[0060] FIG. 6 is a three-dimensional plot of the cost function when signal is unknown, SNR is 20 dB, and RX moves along any line parallel to x-axis for any given y between Δd=1 m and 10 km and z=200 km.

[0061] FIG. 7 is an example trace of true position search using a random memoryless search.

[0062] Corresponding parts are given corresponding reference characters throughout the drawings.

DETAILED DESCRIPTION

[0063] This disclosure aims to challenge the assumption that RXs physically separated by a larger distance can create better Doppler frequency effects and hence higher resolution in localization estimation. This paper claims that a single RX unit with multiple physically separated RX antennas can create different Doppler frequency effects and can locate the EMI emitter effectively, even if the EMI TX uses a single carrier frequency. To achieve that effect, this paper proposes using frequency-mixing (FMx) antennas or frequency-mixing intelligent reflecting surface (FMx-IRS) antennas as they appear in [15]. These antennas can create multiple carrier frequency signals at a single receiver, even if a single carrier frequency signal is transmitted by the EMI TX. The IRS has been investigated intensively and recently as a future wireless communication technology to improve channel capacity efficiency as well as secrecy channel capacity performance in wireless communications. For example, IRS studies are featured in [15]—[28].

[0064] This disclosure assumes a stationary emitter with zero velocity in the x, y, and z directions. Only the RX is mobile. The proposed method assumes no Global Positioning System (GPS) signal. Furthermore, the proposed method assumes a Rayleigh fading channel and includes the frequency instability of the EMI TX oscillator with the following uniform random variable distribution: uniform[−100, 100][Hz]. Therefore, the proposed method is considered under more practical and severe RF environments than the cUKF in [31].

[0065] FIG. 1 shows an example scenario of the existing method in [30]. Notice that two separate RXs are moving in the opposite direction with a minimum separation distance, Δd=10 km. FIG. 2 shows an example scenario of the proposed method, and FIG. 3 shows a corresponding implementation block and spectrum diagram.

[0066] The proposed method assumes that a modulated signal of only a single carrier frequency is transmitted by the EMI TX. Then, (L−1) subcarrier frequency signals are created in addition to the received main carrier by the (L−1) FMx [15] antennas at a single RX unit (e.g., L=2) and are processed with the signal received directly from the TX to the RX. L is the total number of antennas at the RX unit, and all antennas are in proximity. The separation distances between the main RX antenna and FMx antennas are nonzero, (e.g., Δd is about 1 m or less than 10 m). Therefore, the RX antenna and (L−1) FMx antennas can be located within a single RX unit but not co-located. A massive number of antennas can be co-located for millimeter wavelength carrier frequencies in the fifth generation (5G) and future wireless communication systems. The velocity of the main RX antenna and the (L−1) FMx antennas are all equal. The objective is to localize an EMI source. The received complex equivalent lowpass signal at the lth FMx or main antenna for the kth interception interval at time t is written [30] as Equation (1) below, where Tis the observation time interval, b.sub.lk, is an unknown complex scalar representing the path attenuation (e.g., Rayleigh fading) at the kth interception interval observed by the lth FMx antenna, s.sub.k(t), which may be known or unknown depending on the application (e.g., known for search and rescue (SAR) and unknown for an intentional EMI emitter), is the observed signal envelope during the kth interception interval, f.sub.lk is the down converted frequency, and w.sub.lk(t) is a complex AWGN∈CN(0, σ.sub.n{circumflex over ( )}2), l=1, . . . , L, k=1, . . . , K.

r.sub.lk(t)=b.sub.lks.sub.k(t)e.sup.j2πf.sup.lk.sup.t+w.sub.lk(t), 0≤t≤T  (1)

[0067] Let v.sub.lk, p.sub.0, and p.sub.lk denote, respectively, the relative velocity vector between the stationary EMI TX and the mobile RX, the stationary EMI TX position vector, and the mobile lth FMx or the mobile main RX antenna position vector within a single RX at interception interval k.

[0068] The key idea of the proposed system is that the arrival frequencies at the RX are multiple frequencies, e.g., f.sub.c and f.sub.c+f.sub.r from the direct link main antenna and the indirect link via the FMx antenna, respectively, for L=2 in FIG. 2. Note that the distance between an FMx antenna and the main RX antenna is small but nonzero e.g., with Δd=1 m. Therefore, multiple Doppler frequencies can be created at the main RX antenna, as long as Δd is non-zero. The phase differences due to the wave traveling the short distance difference can be included in the Rayleigh fading coefficient, b.sub.lk(t), in Equation (1) in the analysis.

[0069] The output frequency observed by the lth receiver antenna before the down conversion during the kth interception interval is given by Equation (2):

f.sub.lk=[f.sub.cl+ν.sub.k][1+μ.sub.lk(p.sub.0)]  (2)

where:

f.sub.cl=f.sub.c+(l−1)f.sub.r, l=1, . . . , L  (3)

[0070] f.sub.c is the known transmitted signal nominal frequency, which is also the input frequency at the lth antenna, f.sub.cl is the output frequency at the lth antenna, f.sub.r=f.sub.cΔf is the FMx frequency separation, ν.sub.k is the unknown transmitted frequency shift due to source instability during the kth interception interval, and μ.sub.lk (p.sub.0) is the Doppler frequency normalized by the carrier frequency fa and written as Equation (4):

[00001]$\begin{array}{cc}{\mu }_{\mathrm{lk}}\left(p_0\right)=\frac{1}{c}\frac{v_{\mathrm{lk}}^T\left[p_0-p_{\mathrm{lk}}\right]}{.Math.p_0-p_{\mathrm{lk}}.Math.}.& \left(4\right)\end{array}$

[0071] The down-converted frequency f.sub.lk−f.sub.cl is approximated according to Equation (5) because ν.sub.kμ.sub.lk(p.sub.0)<<f.sub.cl. The emitter frequency instability ν.sub.k is modeled by a uniform random variable as Equation (6) and collects N samples per interception interval T (seconds). Hence, the sampling interval is represented by Equation (7).

f.sub.lk≃ν.sub.k+f.sub.clμ.sub.lk(p.sub.0)  (5)

ν.sub.k=Unif[−100,100]Hz,  (6)

T.sub.s=T/(N−1),0≤t≤T.  (7)

[0072] For RX signal processing, the same direct position determination approach in [30] can be used by changing f.sub.c in [30] to f.sub.cl in Equation (3) throughout. The analog RX signal in Equation (1) is sampled as r.sub.lk[n]=r.sub.lk(nT.sub.s), n={0, 1 . . . , N−1} and represented by Equation (8), where A.sub.lk is the N×N Doppler shift diagonal matrix written as Equation (9), C.sub.k is an N×N emitter frequency instability diagonal matrix written as Equation (10), s.sub.k is the N×1 TX signal vector, and w.sub.lk˜CN (0, σ.sub.n.sup.2I.sub.N) is the N×1 complex Gaussian noise vector. In Equation (9), the μ.sub.lk(p.sub.0) will be replaced by μ.sub.lk(p) for the grid search or random search, where p is a position vector and candidate for EMI TX localization, and p.sub.0 is the true location.

r.sub.lk=b.sub.lkA.sub.lkC.sub.ks.sub.k+w.sub.lk  (8)

A.sub.lkdiag{1,e.sup.j2πf.sup.cl.sup.μ.sup.lk.sup.(p.sup.0.sup.)T.sup.s, . . . ,e.sup.j2πf.sup.cl.sup.μ.sup.lk.sup.(p.sup.0.sup.)(N−1)T.sup.s}   (9)

C.sub.kdiag{1,e.sup.j2πν.sup.k.sup.T.sup.s, . . . ,e.sup.j2πν.sup.k.sup.(N−1)T.sup.s}  (10)

[0073] The proposed algorithm maximizes the conditional likelihood function in (4) of [29] or the correlation cost function between the received vector including noise and the received vector excluding noise, given the observation vector r.sub.lk. Equivalently, the proposed algorithm minimizes the exponent in the likelihood function, called the minimum distance cost function, between the received signal vector including noise and the received vector excluding noise, as done in (5) of [29], [30]:

[0074] A Minimum Distance Cost function is defined by Equation (11).

[00002]$\begin{array}{cc}{L}_{\min \mathrm{distance}}=\frac{1}{{\sigma }_n^2}\underset{k=1}{\overset{K}{.Math.}}\underset{l=1}{\overset{L}{.Math.}}{.Math.r_{\mathrm{lk}}-b_{\mathrm{lk}}{A}_{\mathrm{lk}}{C}_k{s}_k.Math.}^2.& \left(11\right)\end{array}$

[0075] A Maximum Correlation Cost function is defined by Equation (12).

[00003]$\begin{array}{cc}{L}_{\max \mathrm{correlation}}=\underset{k=1}{\overset{K}{.Math.}}\underset{l=1}{\overset{L}{.Math.}}{.Math.{\left(A_{\mathrm{lk}}{C}_k{s}_k\right)}^H{r}_{\mathrm{lk}}.Math.}^2& \left(12\right)\end{array}$

[0076] Superscript H denotes the Hermitian. Equation (12) can be simplified further. The proposed algorithm will employ two simplified cost functions below from [30] for the unknown signal case and known signal case.

[0077] The objective function in Equation (12) is rewritten for the unknown signal case, and the best estimate of the EMI emitter location is found, respectively, as shown in Equations (13) and (14), where p is a possible grid point in a given search area, V.sub.k and Q.sub.k are defined by Equations (15) and (16), and λmax(Q.sub.k) is the maximum eigenvalue of Q.sub.k, which is ≥0 since Q.sub.k is positive semi-definite.

[00004]$\begin{array}{cc}{L}_{\mathrm{us}}\left(p\right)=\underset{k=1}{\overset{K}{.Math.}}{\lambda }_{\max }\left\{Q_k\right\}& \left(13\right)\end{array}$$\begin{array}{cc}{\hat{p}}_0=\arg \underset{p}{\max }\left\{L_{\mathrm{us}}\left(p\right)\right\},& \left(14\right)\end{array}$$\begin{array}{cc}{V}_k\stackrel{△}{=}\left[A_{1k}^H{r}_{1k},.Math.,A_{\mathrm{Lk}}^H{r}_{\mathrm{Lk}}\right],& \left(15\right)\end{array}$$\begin{array}{cc}{Q}_k\stackrel{△}{=}{V}_k^H{V}_k,& \left(16\right)\end{array}$

[0078] For the known signal case, the objective function in (12) is rewritten, and the best estimate of the EMI emitter location is found, respectively, as Equations (17) or (18) and Equation (19), where c.sub.k is defined according to Equation (20) and Equation (21), z.sub.k is defined according to Equation (22), C.sub.k is defined according to Equation (23), and S.sub.k is defined according to Equation (24).

[00005]$\begin{array}{cc}{L}_{\mathrm{ks}}\left(p\right)=\underset{k=1}{\overset{K}{.Math.}}\underset{l=1}{\overset{L}{.Math.}}{.Math.{\left(A_{\mathrm{lk}}{C}_k{s}_k\right)}^H{r}_{\mathrm{lk}}.Math.}^2& \left(17\right)\end{array}$$\begin{array}{cc}\stackrel{\mathrm{or}}{=}\underset{k=1}{\overset{K}{.Math.}}\underset{l=1}{\overset{L}{.Math.}}{.Math.r_{\mathrm{lk}}^H{A}_{\mathrm{lk}}{S}_k{c}_k.Math.}^2& \left(18\right)\end{array}$$\begin{array}{cc}{\hat{p}}_0=\arg \underset{p}{\max }\left\{L_{\mathrm{ks}}\left(p\right)\right\}& \left(19\right)\end{array}$$\begin{array}{cc}{c}_k\stackrel{△}{=}{\left[1,e^{j2\pi v_k{T}_s},.Math.,e^{j2\pi v_k\left(N-1\right)T_s}\right]}^T& \left(20\right)\end{array}$$\begin{array}{cc}={\left[1,{𝓏}_k,.Math.,{\left({𝓏}_k\right)}^{\left(N-1\right)}\right]}^T,& \left(21\right)\end{array}$$\begin{array}{cc}{𝓏}_k\stackrel{△}{=}{e}^{j2\pi v_k{T}_s},& \left(22\right)\end{array}$$\begin{array}{cc}{C}_k=\mathrm{diag}\left[c_k\right],& \left(23\right)\end{array}$$\begin{array}{cc}{S}_k=\mathrm{diag}\left[s_k\right].& \left(24\right)\end{array}$

[0079] S.sub.k and c.sub.k are transmitted together by the EMI. So, when S.sub.kc.sub.k in (14) is treated as known, the EMI TX unstable frequency ν.sub.k and the transmitted sequence vector s.sub.k (e.g., quadrature phase shift keying (QPSK)) are known as well.

[0080] However, the known-signal case means that only the modulated symbol sequence vector s.sub.k in the received signal vector r.sub.k is known because c.sub.k is unintentionally transmitted by the EMI. The frequency instability ν.sub.k in the EMI TX signal is unknown at the RX in general, and ν.sub.k should be unknown even for the known signal case. Therefore, L.sub.ks (p) in Equation (18) should be maximized with respect to the possible grid point vector p as well as the frequency instability parameter ν.sub.k as in Equations (25)-(27), wherein B.sub.k is defined by Equation (28), G.sub.k is defined by Equation (29), and α.sub.m is the sum of elements on the m-th diagonal of matrix G.sub.k. Hence, the cost function can be rewritten as Equation (30), which finds the ν.sub.k, for each k, that maximizes the expression for β.sub.m in Equation (31). Therefore, the estimated EMI position for the known signal case, excluding the frequency instability ν.sub.k, is given by Equation (32).

[00006]$\begin{array}{cc}{L}_{\mathrm{ks}}\left(p\right)=\underset{k=1}{\overset{K}{.Math.}}\underset{l=1}{\overset{L}{.Math.}}{.Math.r_{\mathrm{lk}}^H{A}_{\mathrm{lk}}{S}_k{c}_k.Math.}^2=\underset{k=1}{\overset{K}{.Math.}}{.Math.V_k^H{S}_k{c}_k.Math.}^2=\underset{k=1}{\overset{K}{.Math.}}{.Math.B_k{c}_k.Math.}^2=\underset{k=1}{\overset{K}{.Math.}}{\left(B_k{c}_k\right)}^H\left(B_k{c}_k\right)& \left(25\right)\end{array}$$\begin{array}{cc}=\underset{k=1}{\overset{K}{.Math.}}{c}_k^H{B}_k^H{B}_k{c}_k=\underset{k=1}{\overset{K}{.Math.}}{c}_k^H{G}_k{c}_k=\underset{k=1}{\overset{K}{.Math.}}\underset{i=0}{\overset{N-1}{.Math.}}\underset{j=0}{\overset{N-1}{.Math.}}{\left({𝓏}_k^H\right)}^i{G}_k\left[i,j\right]{\left({𝓏}_k\right)}^j& \left(26\right)\end{array}$$\begin{array}{cc}=\underset{k=1}{\overset{K}{.Math.}}\underset{i=0}{\overset{N-1}{.Math.}}\underset{j=0}{\overset{N-1}{.Math.}}{G}_k\left[i,j\right]{\left({𝓏}_k\right)}^{j-i}=\underset{k=1}{\overset{K}{.Math.}}\underset{m=-\left(N-1\right)}{\overset{N-1}{.Math.}}{{\alpha }_m\left({𝓏}_k\right)}^m=\underset{k=1}{\overset{K}{.Math.}}\mathrm{Re}\left\{\underset{m=0}{\overset{N-1}{.Math.}}{{\beta }_m^{\star }\left({𝓏}_k\right)}^{-m}\right\}=\underset{k=1}{\overset{K}{.Math.}}\mathrm{Re}\left\{\mathrm{FFT}\left\{{\beta }_m^{\star }\right\}\right\}& \left(27\right)\end{array}$$\begin{array}{cc}{B}_k\stackrel{△}{=}{V}_k^H{S}_k,& \left(28\right)\end{array}$$\begin{array}{cc}{G}_k\stackrel{△}{=}{B}_k^H{B}_k,& \left(29\right)\end{array}$$\begin{array}{cc}{L}_{\mathrm{ks}}\left(p\right)=\underset{k=1}{\overset{K}{.Math.}}\underset{v_k}{\max }\left[\mathrm{Re}\left\{\mathrm{FFT}\left\{{\beta }_m^{\star }\right\}\right\}\right]& \left(30\right)\end{array}$$\begin{array}{cc}{\beta }_m\stackrel{△}{=}\{\begin{array}{cc}{\alpha }_0& \mathrm{if}m=0\\ 2{\alpha }_m& \mathrm{if}m=1,.Math.,N-1\end{array}& \left(31\right)\end{array}$$\begin{array}{cc}{\hat{p}}_0=\arg \underset{p}{\max }\left\{L_{\mathrm{ks}}\left(p\right)\right\}.& \left(32\right)\end{array}$

[0081] The same notations as the ones in [30] for Equations (1) to (33) are used in this paper because they are directly applicable for the 2D system model, 2D grid search, and even for the proposed method. The next section of this disclosure describes the proposed method for a 3D grid search. Since this paper considers localization of a stationary EMI emitter, it focuses on the unknown signal case rather than the known case.

[0082] For the existing 2D grid search in [30], the z-axis altitude is set to zero, as shown in FIG. 2. The existing 2D grid search fails for a 3D grid search scenario with nonzero altitude. As the altitude increases beyond 10 km, the RMSE, e.g., 4 km, is unacceptable. In this section, the existing 2D grid search is investigated again for a 3D grid search application. Still, an EMI emitter is assumed to be stationary on the Earth's surface, and a single mobile RX unit with multiple FMx antennas is considered.

[0083] FIG. 4 shows an example scenario of a 3D grid search for the proposed method, which uses one FMx antenna and one main antenna at the receiver, i.e., L=2. Both antennas are in a single RX unit and move together with the same velocity and direction at the same altitude, e.g., z=200 km. An LEO satellite trajectory along the x-axis or y-axis is assumed without loss of generality. A stationary EMI TX source can be located anywhere on the Earth's surface, e.g., p0=[x0, y0, z0]=[6.5, 4, R.sub.Earth] km, where R.sub.Earth=7,361 km is the Earth's radius. The altitude z.sub.0 in FIG. 4 was selected to be 0 km instead of the Earth's radius of 7,361 km without loss of generality. For simulation, z.sub.0=7,361 km is used. In fact, both z.sub.0=7,361 km and 0 km yield the same results. The altitude z.sub.0=0 km does not represent the Earth's surface constraint required in the cUKF method [31]. In this disclosure, the altitude z.sub.0 can be anywhere. The carrier frequency of the EMI TX source is f.sub.c=0.4 GHz, representing a UHF frequency, and the main antenna of a single RX moves along the x-axis from 1 km to 10 km with p.sub.l=1,k=[x.sub.l,k, y.sub.l,k=y, z.sub.l,k=R.sub.Earth+200] (km) and velocity ν.sub.l=1=6.944 km/s, and the path of the FMx antenna is p.sub.l=2,k=[x.sub.l,k, y.sub.l,k=y−0.001, z.sub.l,k=R.sub.Earth+200] (km) with the same velocity ν.sub.l=1=2,k=6.944 km/s. It moves in the same x-axis direction, x.sub.l,k=[1, 10] (km), with an LEO speed of ν=6.944 km/s. The y of the RX trajectory can be any number between Δd and 10 km. Here, y is set to 4 km without loss of generality. Note that the two antennas in the same RX are separated physically by only Δd=1 m on the y-axis. The EMI signal is intercepted every 1 km.

[0084] Since the altitude, 200 km, is much higher than one side of the search area, 10×10 (km×km), the cost functions will not be sensitive to the x-axis but only to the y-axis when the RX moves along a parallel line to the y-axis, and vice versa. FIGS. 5 and 6 show the cost function for the 3D search in FIG. 4, respectively, when the RX moves along a parallel line to the y-axis and x-axis. Here the signal is unknown, and the SNR is 20 dB. Observe that the 3D search cost function in FIGS. 5 and 6 show only the peak value near the true value of y.sub.0=6.5 km and x.sub.0=4 km, respectively, when the RX moves along a parallel line to they-axis for any given x value and a parallel line to the x-axis for any given y value in the search range. To complete the 3D search for {circumflex over ( )}p.sub.0=[{circumflex over ( )}x.sub.0, {circumflex over ( )}y.sub.0], the RX must take samples moving along any parallel line to the x-axis after moving along any parallel line to the y-axis, and vice versa.

[0085] If a meter-level RMSE is desirable, then the grid size should be small, e.g., in meters because the grid search time increases exponentially. For example, if the grid size in the x-axis and y-axis, respectively, are Δd.sub.x=Δd.sub.y=0.06 km, then the number of grid points in the x-axis and y-axis would be N.sub.x=N.sub.y=10 km/Δd.sub.x=167, and the total number of grid points would be N.sub.xN.sub.y=167{circumflex over ( )}2=27,889. At each grid point, the cost function in Equation (13) is calculated. The correlation is computed between the actual intercepted signal r.sub.lk and a hypothetical received signal (A.sub.lkC.sub.ks.sub.k) in Equation (12), which becomes Equation (13), assuming that the EMI TX is located at a hypothetical grid point, and the grid point p corresponding to the maximum correlation cost function in Equation (14) is selected. It takes 85.54 seconds for ten trials using a powerful workstation, e.g., Precision 5820 Tower. However, if Δd.sub.x=Δd.sub.y=0.001 km, then it will take more than a month for the simulation. One trial means K=10 interceptions in FIG. 1, k=1, K. Each interception takes N=100 samples of the received signal, and L=2 RX antennas, l=1, . . . , L. This prohibits using the 2D grid search in [30] when a meter-level resolution is desirable.

[0086] To address these limitations, this disclosure contemplates two computationally efficient methods without sacrificing the RMSE performance: (a) a random memoryless search and (b) a random memory search method to reduce the grid search computation time significantly.

[0087] The QPSK symbol transmission rate is assumed to be 10 ksps. The RX does not need to have the transmitted signal information for the unknown signal case, and it takes N=100 samples of the received signal every interception interval for K=10 interception intervals and L=2 RX antennas. In the existing 2D method, the grid point vector that maximizes the cost function is selected out of N.sub.grid=N.sub.xN.sub.y grid points per trial. Then, the average RMSE is computed over N.sub.exp=100 trials as Equation (33).

[00007]$\begin{array}{cc}\mathrm{RMSE}=\sqrt{\frac{1}{N_{\exp }}\underset{i=1}{\overset{N_{\exp }}{.Math.}}{.Math.\hat{p}\left(i\right)-p_0.Math.}^2}.& \left(33\right)\end{array}$

[0088] For a random memoryless search, a hypothetical position vector p is randomly generated in the specified search zone, e.g., 10×10 (km×km) in FIGS. 1, 2, and 4. Then, the cost function L.sub.us(p) in Equation (13) is computed with the hypothetical position vector p and the same received vector r.sub.lk obtained through k=1, K=10 interceptions via l=2, . . . , L FMx antennas and the main RX antenna l=1 for the proposed on the RX unit. Then, another hypothetical position vector p is randomly generated, and a corresponding cost function is computed. The hypothetical position vector corresponding to the higher cost function is kept. This is repeated for a certain number of hypothetical position vectors until the difference in cost functions between the current and the next hypothetical position vectors are less than ϵ, e.g., 10{circumflex over ( )}−5. Then, the algorithm concludes the present trial and proceeds to the next. FIG. 7 shows an example trace of the surviving hypothetical position vectors. After a certain number of trials N.sub.exp, the RMSE value in Equation (33) is computed for a given E.sub.s/N.sub.0 in dB. The generated number of hypothetical position vectors is typically much smaller than the total number of grid points N.sub.xN.sub.y, in the existing grid search. Here the memoryless search means that the search in the next trial is independent of that in the current trial.

[0089] For a random memory search, the final hypothetical position vector and its corresponding cost function from the previous trial are used for the current trial (unlike a random memoryless search where a hypothetical position vector is randomly generated for every trial). Once the initial starting position vector is established, the same procedure as that of the random memoryless search is conducted for the rest of the trial. Hence, there is a memory between adjacent trials.

[0090] The complexity in a grid search will be proportional to the total number of grid points, which is the square of the number of grid points per axis in a 2D search. The complexity in the proposed random memoryless or memory search is proportional to the number of randomly generated points, which can be significantly smaller than the total number of grid points. Hence, the random search can be much faster than the grid search at the same resolution. In addition, the resolution can be improved by the random search because the step size in a grid search determines the resolution; however, there is no step size in the random search. A search point in an axis can be any real number represented by a high number of bits.

[0091] An example Random Memory Search Algorithm can be conducted in the following steps:

[0092] STEP 0: Define all parameters, e.g., total number of trials=i.sub.total=10.

[0093] STEP 1: Trial i=1.

[0094] STEP 2: For each SNR value and each trial value, set the cost function L.sub.us(p).sub.init to zero. For each SNR value and each trial value, generate the signal s.sub.k(t), noise w.sub.k(t), unknown complex scalar path attenuation b.sub.lk, and channel phase ϕ.sub.lk, using the given corresponding probability density functions.

[0095] STEP 3: Generate r.sub.lk using Equation (8).

[0096] STEP 4: Define trial initial position (p.sub.x,init, p.sub.y,init) as the center position, e.g., (p.sub.x,init=5, p.sub.y,init5).

[0097] STEP 5: Define a sufficient number of iterations (e.g., i.sub.iteration=10,000) for the ith trial search run.

[0098] STEP 6: Generate a new position (p.sub.x,new, p.sub.y,new) with Gaussian distributions: p.sub.x,new˜N(p.sub.x,init=5,√{square root over (p.sub.x,init=5)}) and p.sub.y,new˜N(p.sub.y,init=5,√{square root over (p.sub.y,init=5)}).

[0099] STEP 7: Compute μ.sub.1(p.sub.0) and μ.sub.2(p.sub.0) using Equation (4).

[0100] STEP 8: Compute A.sub.1k and A.sub.2k using Equation (9).

[0101] STEP 9: Compute V.sub.1k and V.sub.2k using Equation (15).

[0102] STEP 10: Compute Q.sub.k using Equation (16), and find the max eigenvalue of the Q.sub.k matrix.

[0103] STEP 11: Compute a new cost function L.sub.us(p).sub.new.

[0104] STEP 12: If L.sub.us(p).sub.new>L.sub.us(p).sub.init, then update (p.sub.x,init, p.sub.y,init) as (p.sub.x,new, p.sub.y,new). Also update the cost function with the latest value. Else, do not update position vector and cost function value. Go to STEP 6 until the iteration is complete for each trial.

[0105] STEP 13: Find the position estimate {circumflex over ( )}p(i) that gives the final value of L.sub.us(p).sub.new at the ith trial.

[0106] STEP 14: Increase i to i=i+1. Go to STEP 2 with L.sub.us(p).sub.init=L.sub.us(p).sub.new and (p.sub.x,init, p.sub.y,init)=(p.sub.x,new, p.sub.y,new) if i≤i.sub.total. Else stop.

[0107] STEP 15: Compute the distance between the last trial position estimate {circumflex over ( )}p(i.sub.total) and the true position p.sub.0.

[0108] Table 1 lists an example of computational complexity and running time using the Precision 5820 Tower workstation and MATLAB for the existing grid search, random memoryless search, and random memory search, when the scenarios in FIGS. 1 and 2 are used with SNR=20 dB, K=10 interceptions, L=2 RX antennas, and N.sub.exp=10 trials. The i.sub.rms and i.sub.rmls denote the number of randomly generated hypothetical vectors per trial until the algorithm converges for random memory and random memoryless search, respectively. For the existing grid search, the grid size is 0.06 km with N.sub.x=N.sub.y=167 and N.sub.grid=N.sub.x×N.sub.y=27,889. Observe from Table 1 that the random memory search is 1.84 times and 2.22 times faster than the random memoryless search and the grid search with grid size 0.06 km, respectively. Another grid size 0.012 km increases the computation run time 27 times that of the proposed random memory search. Furthermore, the random memory search is greater than 8,025 times faster than the grid search of grid size=0.001 km.

TABLE-US-00001 TABLE 1 Algorithm comparison based on computational complexity [SNR = 20 dB, N.sub.exp = 10 trials] Random Memory Random Memory- Search less Search 2D Grid Complexity (N.sub.exp × i.sub.rms)[ (N.sub.exp × i.sub.rmls)[ (N.sub.exp × N.sub.grid) 2L + K(2L)] 2L + K(2L)] [2L + K(2L)] Running 38.37 s 70.74 s 85.54 s (Grid Time (s) Size 0.06 km) 1038.00 s (Grid Size 0.012 km)

[0109] Table 2 lists examples of the RMSE values for the existing grid search, random memoryless search, and random memory search when the same scenarios in FIGS. 1 and 2 are used with the same parameters used for Table 1, where the grid size was 0.06 km or 0.012 km for the grid search. The grid size of the 0.001 km case could not be included because it takes more than one month of simulation time, even for one trial and one SNR value.

TABLE-US-00002 TABLE 2 Algorithm comparison based on RMSE values of search methods Random Memory Random Memory- SNR (dB) Search (m) less Search (m) 2D Grid (m) 20 105.74 121.04 104.90 30 36.46 68.64 31.60 40 11.29 59.74 8.97 (Grid Size 0.06 km) 4.00 (Grid Size 0.012 km)

[0110] Observe from Table 2 that the random memory search, the random memoryless search, and the grid search with grid size 0.06 km show a similar RMSE resolution. The quality of the random memory search and the random memoryless search over the grid search is not degraded, although the search time can be significantly reduced, as indicated in Table 1.

[0111] To improve the performance further, the inventors contemplate using chirp RX processing with a finite number of FMx antennas. This approach requires a time-dependent RX frequency at each sample time. For the case without an RX chirp, L=2 number of RX frequencies are created using L=2 number of antennas (i.e., main RX antenna and one FMx antenna). For the chirp case, multiple sample time-dependent RX frequencies are created, e.g., a different RX frequency at each sample time, by switching from an FMx antenna to another FMx antenna in a set of (N−1) FMx antennas. If N samples are taken per interception interval, then it is desirable to employ N−1 number of FMx antennas and one main antenna at the RX side for each group l=1, . . . , L. Here, l represents the RX group index for the chirp case, whereas l represents the RX frequency index for the no-chirp case. Each group consists of n=1, N number of different sample time-dependent frequencies. Here, at the first sample time n=1, the RX frequency is generated via the main RX antenna, and at sample time n, 2≤n≤N, the RX frequency is generated via the nth FMx antenna, Hence, L(N−1) number of FMx antennas are necessary. If a millimeter wave is used by the EMI TX, then the wavelength is a few millimeters, and hence, (L=2)(N−1)=254 FMx antennas can be placed at one RX unit in an area less than 10×10 (cm×cm), assuming a wavelength separation between adjacent antenna elements.

[0112] The nth element of a diagonal Doppler shift matrix in Equation (9) is rewritten for the chip case as Equation (34), where l=1, L, n=1, N, and μ.sub.lk(p.sub.0) denotes the normalized Doppler frequency fraction in Equation (4). Hence, different Doppler frequency effects can be created at each sampling time and can be applicable for the proposed method. Note again that the proposed method does not measure Doppler frequency as was done in [31] but rather uses only the received samples.

[0113] Accordingly, it can be seen that the present disclosure provides a system for determining a location of an emitter. A system according to the present disclosure generally comprises a single mobile frame configured for movement relative to the emitter (which can emit a known emitter signal (e.g., SAR application) or an unknown emitter signal (e.g., military reconnaissance application). For example, the frame can be an air frame of a SAR aircraft or reconnaissance aircraft, the body of a satellite, a ground vehicle chassis, or a water vessel, etc.

[0114] As shown in FIG. 2-4, the system herein comprises a main receiver supported on the mobile frame at a first location for movement with the mobile frame relative to the emitter and a frequency mixing antenna supported on the mobile frame at a second location for movement with the mobile frame and the main receiver relative to the emitter. In certain embodiments, each frequency mixing antenna comprises a frequency-mixing intelligent reflecting surface antenna, but conventional frequency-mixing antennas also work for this application at significantly reduced cost and so may be preferred. The second location of the frequency mixing antenna is spaced apart from the first location of the main receiver on the common support frame by a relatively small distance (e.g., less than 10 m). As shown in FIG. 3, the frequency mixing antenna is configured to receive an emitter signal from the emitter at the second location and output a frequency-mixed signal (arrow pointing rightward and downward toward main receiver). The main receiver is configured to receive the emitter signal directly at the first location and also to receive the frequency-mixed signal.

[0115] As explained above, this system enables localization of the emitter with only one mobile frame because the main receiver receives two Doppler frequencies at offset locations. The first Doppler frequency is derivable from the direct emitter signal, and the second Doppler frequency is derivable from the frequency-mixed signal. Accordingly, the localization methods of the prior art that rely on two Doppler frequencies to establish emitter location can be adapted for use with the above described single-frame localization system.

[0116] In an exemplary embodiment, the system further comprises a processor connected to the main receiver and configured to determine the first Doppler frequency and the second Doppler frequency from the direct emitter signal and the frequency-mixed signal, respectively. The processor can be adapted use any suitable multi-Doppler frequency localization method to determine the location of the emitter in a two-dimensional grid or a three-dimensional space. In an exemplary embodiment, the processor is configured to determine the location of the emitter based on the first Doppler frequency and the second Doppler frequency using a direct position determination approach as described above and adapted from [30]. In certain embodiments, the processor is configured to use one of a random memoryless search algorithm or a random memory search algorithm as described above to reduce the grid search computation time for a defined search area. In one or more embodiments, the processor is configured to execute chirp RX processing techniques as described above to further improve localization performance.

[0117] The processor mentioned above may reside in a computer including a variety of computer hardware, including memory for storing processor-executable instructions. In general, the processor is configured to execute the programs stored in memory for carrying out the above functions. Components of the programs may reside at various times in different storage components of a computing device, and are executed by the processor of the device.

[0118] This disclosure is not limited to any particular computing environment. Examples of computing systems, environments, and/or configurations that may be suitable for use with aspects of the invention include, but are not limited to, personal computers, server computers, hand-held or laptop devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, mobile telephones, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like.

[0119] Embodiments of the aspects of the present disclosure may be described in the general context of data and/or processor-executable instructions, such as program modules, stored one or more tangible, non-transitory storage media and executed by one or more processors or other devices. Generally, program modules include, but are not limited to, routines, programs, objects, components, and data structures that perform particular tasks or implement particular abstract data types. Aspects of the present disclosure may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote storage media including memory storage devices.

[0120] In operation, processors, computers and/or servers may execute the processor-executable instructions (e.g., software, firmware, and/or hardware) such as those illustrated herein to implement aspects of the invention.

[0121] Embodiments may be implemented with processor-executable instructions. The processor-executable instructions may be organized into one or more processor-executable components or modules on a tangible processor readable storage medium. Also, embodiments may be implemented with any number and organization of such components or modules. For example, aspects of the present disclosure are not limited to the specific processor-executable instructions or the specific components or modules illustrated in the figures and described herein. Other embodiments may include different processor-executable instructions or components having more or less functionality than illustrated and described herein.

[0122] The order of execution or performance of the operations in accordance with aspects of the present disclosure illustrated and described herein is not essential, unless otherwise specified. That is, the operations may be performed in any order, unless otherwise specified, and embodiments may include additional or fewer operations than those disclosed herein. For example, it is contemplated that executing or performing a particular operation before, contemporaneously with, or after another operation is within the scope of the invention.

[0123] When introducing elements of the present disclosure or the preferred embodiment(s) thereof, the articles “a”, “an”, “the” and “said” are intended to mean that there are one or more of the elements. The terms “comprising”, “including” and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements.

[0124] In view of the above, it will be seen that the several objects of the disclosure are achieved and other advantageous results attained.

[0125] As various changes could be made in the above products and methods without departing from the scope of the disclosure, it is intended that all matter contained in the above description shall be interpreted as illustrative and not in a limiting sense.