1 - Conference
NM, Aug. 28-30,
A01-39867
AIAA 2001-4654
ABSTRACT
Scanned Pattern Interferometric Radar (SPIR) has previously been proposed as a processing algorithm for space-based Ground Moving Target Indication (GMTI) radar. Performance of the algorithm is highly dependent on cluster design. Aperture placement in the cluster determines the Point Spread Function (PSF). We show that certain characteristics of the PSF can be used to identify ill- or well-conditioned systems. Using these results we develop a design algorithm to select aperture placements that will guarantee good numerical conditioning. Keywords: radar, interferometry, point spread function, aperture synthesis, sparse arrays
The algorithm synthesises a PSF matrix corresponding to the gain pattern of the cluster as it is swept across the footprint. By deconvolving the synthesised gain pattern from the received signals the true ground scene is revealed. Success of the SPIR algorithm depends on the invertibility of the PSF matrix. Using this constraint, a design algorithm is developed for one-dimensional clusters. This paper discusses the design algorithm and discusses some example arrays. We begin with a brief review of radar interferometry and the SPIR algorithm. Section 3 discusses performance parameters for the PSF, and how they are evaluated. Section 4 suggests a design algorithm by which high-performance PSFs can be developed. Examples of such PSFs are given in Section 5.
1 INTRODUCTION
Ground Moving Target Indication (GMTI) using separated spacecraft interferometry at radio wavelengths promises to be a powerful new all-weather surveillance tool. In previous work, we presented the Scanned Pattern Interferometric Radar (SPIR) algorithm as a method for obtaining high-resolution clutter-tolerant results1. SPIR uses the high angular variability of a sparse array Point Spread Function (PSF) to extract sufficient information from the signal return that the clutter and targets can be separated without an a priori assumption of the clutter statistics. Main lobe clutter can be separated from moving targets using the deterministic geometric relationship between observation direction and clutter doppler shift.
* Research Assistant, MIT Space Systems Laboratory t Research Scientist, MIT Space Systems Laboratory Copyright 2001 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
i t . ,i t- ' ro..........
FIG. 1: Calculation of the PSF using the Fourier Transform
y = Ax (4) The matrix A is referred to as the point spread function (PSF) matrix of the interferometer. In the case of a onedimensional interferometer, the elements of A are defined by A qp = E(Q)
(5)
which is the PSF of the array, focussed on a cell at angle 9^ from boresight. For a complete derivation of the above results, see reference [1],
Subsequent discussion is focussed on one-dimensional clusters, referred to as arrays. The discussion extends to the two-dimensional case in an obvious manner.
The Fourier transform of the aperture positions is referred to as the array factor, while the Fourier transform of the aperture illumination is just the individual aperture response. The array factor for a two-dimensional interferometer is the norm of the two-dimensional Fourier transform of the aperture positions
array(x,y)
S'n=\
"
dx>dy.
i>
(1)
where the individual aperture positions are given by (xn,yn). After transformation to azimuth-elevation coordinates (v|/flz,v|/e/), the array factor can be expressed as2
"M -?
n= 1
2>
(2)
Since the array factor is the sum of complex exponentials, it is periodic in vj/flz and \yel. The PSF for a two-dimensional interferometer is therefore given by2
N
The resolution is the smallest distance between two points at which they can be resolved as being separate. Points that are closer together than the angular separation of the first nulls of the PSF cannot be identified as separate points. This angular null separation defines the resolution of the aperture.
The requirement for high-resolution can be met by using a large baseline. Unfortunately, this decreases the spatial period of the array-factor, as qualitatively discussed in Section 3.2. We shall see that array factors with short spatial periods may result in PSF matrices which are not full rank. The field-of-view is determined by the width of the main lobe of the aperture response. The intersection of the field-of-view with the ground is defined as the interferometer footprint. A wide field-of-view permits coverage of a large ground area, but has higher processing requirements. In general, the field-of-view of an aperture is inversely proportional to its diameter.
Ze
(3)
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
To prevent "fading" at the edges of the footprint, an aperture response with a sharply defined main lobe is preferred. The peak side lobe level predicts the clarity of the image. Information that is located in the side lobes of the PSF overlays information in the main lobe, so that the reconstructed image is obscured. Low PSF side lobes result in clearer images. Our goal is thus to obtain high-resolution PSFs with long spatial periods, which are greater than the desired field-of-view.
The analysis begins with a discussion of the construction of the PSF matrix, which illustrates the relationship between the PSF and the invertibility of the associated PSF matrix.
Figure 3 illustrates the effect of an array factor which repeats within the field-of-view. The last row of the PSF matrix is identical to the first row. The resulting PSF matrix is not full rank and cannot be inverted.
Row 4 = Row 1
The period of the PSF therefore determines whether the PSF matrix will be invertible. In Section 3.3 we determine the relationship between the array construction and the PSF period.
The next section discusses the parameters by which different PSFs can be compared.
In this case the period of the array factor is greater than the field-of-view, therefore each row of the resulting PSF matrix is unique. Furthermore, since by construction none of the rows can be obtained by scaling another row, the rows are linearly independent. Therefore this PSF matrix has full rank and is invertible.
Since the aperture diameter is much smaller than the array baseline, the resolution of the interferometer is determined by the baseline, while the the field-of-view is determined by the individual aperture response. The width of the aperture response mainbeam is inversely proportional to the aperture diameter. Increasing the aperture diameter therefore decreases the field of view and vice versa. For a uniformly illuminated rectangular aperture, the field-of-view in one dimension is
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
(6)
The constant of proportionality can be derived from the array configuration and is determined as follows: The array factor (2) of a one-dimensional array is the norm of the sum of a sequence of complex exponential terms, which each have a maximum projection of one onto the real axis. Therefore the array factor is bounded above by the total number of apertures, TV
-2717 sin vj;
The array configuration determines the shape of the PSF and this problem is examined in detail in Section 4. The array baseline determines the resolution, and for a given configuration, the period of the PSF. As the baseline is increased the resolution is increased and the period is decreased. Figure 4 shows the array factor for two identical array and aperture configurations, one with a baseline of 500 m, the other with a baseline of 1000 m. Doubling the baseline halves the period of the array factor, and halves the width of the grating lobes.
Uw*swi}Ciwf AntyJS 5 4 4
array(\\f) =
n=1
< 1 -N
(9)
By taking the square of the array factor, the interaction between the complex exponential terms can be seen
[array(\\j)]z =
Lw = 1 n = 1
Each aperture pair spacing |*w-*w| samples a spatial frequency, which is not necessarily unique. The maxima \\fmn for a particular spatial frequency occur when
(m*n)<= {1...
where kmn is a positive integer referred to as the spatial frequency number.
FIG. 4: Effect of Array Baseline on the PSF
~"
(m*n)e{l...N}
(12)
Successive values for kmn give the positions of successive maxima corresponding to the spatial frequency sampled by the aperture pair (xm9 xn). The array factor global maxima occur when the maxima of all the sampled spatial frequencies coincide, that is, whenever all the cosines evaluate to one. The first global maximum occurs at \y = 0 ; successive global maxima occur at multiples / of the period \\ip. Setting Vmn = typ and eliminating sm(p\\jp) from (12) gives a set of relations between the aperture positions and the
A one-dimensional aperture array can be uniquely defined by specifying the relative aperture spacing (configuration) and some characteristic length / , such as baseline or minimum spacing. At a given wavelength K, a fully specified array samples a particular set of spatial frequencies. These spatial frequencies scale inversely with / and proportionally with A,. Therefore the spatial period \\fp of the array factor scales according to
(7)
(13)
In particular,
.oc^
{I...N}
(14)
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
The set {kmn} of smallest kmn which will satisfy equations (12) with \\imn = \\ip is determined by simplifying each ratio of positions to irreducible rational form, such that the greatest common divisor (GCD) for each pair (kmn, kpq) is minimised. If at least one kmn does not have a GCD of one with all the other kpq, then the entire set is reducible and the period of the array is reduced according to (12).
FOV = 2 a - _
(21)
where D is the aperture diameter and a is a shape factor such that a rectangular aperture of width
Deff =
sinx|/p =
*2| 1*1
(22)
(15)
has the same field-of-view as the actual aperture. For uniformly illuminated rectangular apertures a = 1 and for uniformly illuminated circular apertures a = 1.22 . The PSF matrix will be invertible if the period of the array factor is greater than the field-of-view of the aperture
yp>FOV
(17)
(23)
= kmax --ft
For small
nax'ft
Substituting equations (18), (21) and (22) into this relation gives
7
(18)
(24)
The effective aperture diameter De^ is selected to satisfy the field-of-view requirement.
The baseline is chosen to satisfy the cross-range resolution requirement ^cross_range, according to
kmin = mm[{kmn}]
the array factor period can also be found from
(19)
B = 2K-
rground
(25)
(20)
If kmin is greater than one, the behaviour of the array factor is less well defined. Interaction between different spatial frequencies may result in low frequency terms which decrease the period of the array factor from the design value \\ip . Therefore arrays with kmin = 1 are preferred.
cross_range
(26)
In fact, the limiting case will not provide an invertible PSF matrix, since the last row of the PSF matrix will be the same as the first row. However, we can analyse the limiting case with the caveat that in the actual system by either decreasing B or increasing Dejy.
Since kmax is an integer, the baseline must be a multiple of half the effective aperture diameter. Likewise, from (15), all internal separations between apertures must also be multiples of this spacing. Therefore De^2 provides a discretisation of the design space, which can be studied non-dimensionally in terms of the minimal spatial frequency numbers kmn. The requirement that kmin = 1 means that the two most closely spaced aper-
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
tures physically overlap. This limitation can be overcome by using phased-array antennas.
Given kmax and the number of apertures, the next section shows how the remaining (N-2) apertures should be placed.
Using the design relations given in Section 4, the baseline and kmax are
(30)
(27)
The resulting array configuration, {kl',k2;...:>kN}, will have an array factor with period \\ip given by (17). We will refer to such an array as being (kmax, N) optimally irreducible. A (kmax, N) array which does not satisfy (27) will have a smaller period than the desired design value \\fp.
Arrays which have the same PSF are referred to as being similar; this is indicated by arrayl ~ array2. Arrays which are mirror images of each other
{0^2;...^_1^WflJC}~{0^WflJC-V,;--^m-^max} ( 28 )
Since kmax is not equal to one, it is impossible to design a two-aperture array which satisfies the system requirements.
With three apertures, there is one unique array, namely {0;1 ;1200}. The array factor is shown in Figure 5. The array factor has a clearly discernible envelope, with period equal to the design value \\fp . The highest spatial frequency that is sampled is A, /B and gives rise to the high frequency variation inside the envelope.
are similar and are discarded when determining the configuration solution space. With N = 2 apertures, there is only one difference pair ku = kmax and the irreducibility condition requires kmax = kmin = 1. Therefore there is one unique solution, {0;!}. The array factor has the same shape regardless of A, and the baseline. The period is given by
V\N = 2 - -B W
.06 -0.04 -0.02 0 0.02 0.04 0.06
With N = 3 apertures, one unique solution exists for each kmax, namely {0;l;fc wfljc }. It is always possible to design an array with the required period, but the shape of the array factor is defined by kmax.
For 4<N<kmax there is always at least one unique solution. The additional degrees of freedom provided by the fourth and higher apertures allows shaping of the array factor to the most desired form.
With four apertures there is one additional degree of freedom, which gives rise to a large solution space. Some examples are shown in Figure 6 through Figure 8.
In each case there is a clearly discernible envelope, with period equal to the design value \\ip. By changing the position of the third aperture in the array, the shape of the envelope is changed and additional lower frequency terms are introduced.
5 EXAMPLES
In this section we provide some examples of arrays created using the design algorithm. The system require-
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
S&ttvi* Optimally
0.02
0.02
Future work will develop additional criteria with which to select particular irreducible arrays that provide the best system-level performance. Using such arrays the performance of SPIR in the presence of clutter and noise can be accurately estimated.
ACKNOWLEDGMENT
This work was made possible in part by the support of the Air Force Office of Scientific Research Contract # F49620-99-1-0217 under the technical supervision of Dr Robert Herklotz.
0.02
REFERENCES
Marais, K., Sedwick, R. J., Space Based GMTI Using Scanned Pattern Interferometric Radar (SPIR), IEEE Aerospace Conference, Proceedings of, March 2001, Montana, 0-7803-6599-2/01.
1
By carefully choosing the position of the third aperture various characteristics of the PSF, such as the width of the envelope main lobe, can be changed.
Kong, E. M., Optimal Trajectories and Orbit Design for Separated Spacecraft Interferometry, MIT S.M. Thesis in Aeronautical and Astronautical Engineering, November 1998.