Abstract:: Also

A Deterministic, Eigenvalue Approach to Space Time Adaptive Processing
Sheeyun Park and Tapan K. Sarkar

Department of Electrical and Computer Engineering, Syracuse University
Abstract: The problem of Space Time Adaptive Processing

(STAP) is approached in a two step process, filtering and detection. Xl(t) X2(t) XN(t)
For a certain look direction, the signal strength is estimated by a
least squares technique. The receiver scans through all possible
look directions and returns estimates of signal strength. From this
map, we can employ standard detection techniques (such as CFAR)
to extract targets.
I. Introduction
STAP techniques have arisen to address the issue of high clutter

and strong jamming environments[11. The deterministic approach
has the advantage of bypassing the problem of estimating
covariance matrices of the statistical methods. Also, it has been
shown that the deterministic approach has a lower bound [2].
dt)
11. Problem Set Up
Figure 2 Functional diagram of an N-element adaptive array
Consider the N-element uruformly spaced array of Figure 1
As in Figure 2, we can now rewrite the received processed
signal by
X1 X2 X3 XN
0 . 0
... 0
or in matrix form as
Figure 1 Array Geometry y(t) = wTx = XTW
The elements are assumed to be omnidirectional point sources in a where T denotes transpose.
uniformly spaced linear array with d as the interelement spacing. m
The signal arrives at each sensor at different times dependent on the

direction of arrival (DOA) and the spacing between the sensor
elements. Let s(t) denote the signal of interest (SOI). At each of
the N sensors, represent xk(t), the received signal, as a sum of the
SOI, sk(t), and interference, nk(t). The interference may consist of
jammers (both blinking and stationary), clutter, and thermal noise.
Therefore. x are the received voltages.
xk(t) = sk(t) + nk(t), k = 1, 2, ... , N (1)

m. Forward Solution
If 8 is the assumed DOA of the SO1 from broadside, then
we can represent the received voltage solely due to the desired Write the actual received voltages as a matrix X. Now
signal at the kth sensor element as consider the difference
x, "' s, ".
(7)
where d is the interelement spacing and h is the wavelength of the
signal located in the far field region of the array. This arises from
the fact that a far field target will exhibit a linear phase front Where the number of degrees of freedom M = (N +1)/2. Denote
(becomes a plane wave). Effects of mutual coupling are ignored in equation (7) as
this analysis. The strength of the SOI, s(t), is the desired unknown
parameter which will be estimated in the deterministic approach.
0-7803-3232-6/96/$5,00 0 1 9 9 6 IEEE 372

This contains just noise and interference which are desired to be
eliminated. Multiply the above expression by a weight vector and
set equal to zero.
[sz - as]x [E]

= [ij]
(9)' Denote this as
There is an implicit constraint in the look direction imposed by the

formulation of the S matrix. 57- ai
The structure of the pencil can be viewed as a linear

predictor. The weights, slid along the angle axis, are the Weight this and set to zero.
coefficients of the linear predictive approximation of the voltages
(see Figure 3). Equation (9) can be rewritten in a more familiar

(13).
generalized eigenvalue equation,
xw = asw (10). Equation (13) is again a generalized eigenvalue problem.
Voltage
IV. Backward Method
Given a linear array, we can either do the linear prediction

in the forward direction or in the backward direction. In the
Weights F e absence of a signal, there will still be eigenvalues associated with
noise. To determine if a signal is present we solve the backward
problem and compare the two results. The standard foward
Figure 3 Weights as Autoregressive linear predictor
formulation is illustrated in Figure 5.
Now consider two dimensions such as angle and time (PRI).
Again subtract the expected signal component from the received
vector, weight the difference, and set equal to zero. This can be
conceived as a two dimensional linear predictor. Consider a grid of
angular resolution cells and time bins. This time a two dimensional
box slides along the plane (see Figure 4).
vi v2 v3 VN
Figure 5 Voltages and Angle of Arrival
Angle To consider the backward problem, first consider the

reverse sequence of Figure 6. The voltages are rotated 180 degrees
Figure 4 2-Dimensional Extension to preserve the angle being relative to the first antenna.
As an example consider M = 2, T = 2, and the first box. The

corresponding voltages are
VN v3 v2 v1
Xvl, t2
Xvl, t l
xv2, t2
xv2, t l
0 ... 0 8
As long as the linear relationship between the positions in the box

are maintained, we can use the same weights for all boxes. The
voltages were taken across elements first then across time samples.
Collect all boxes and again consider the difference Figure 6 Reversed Voltages and Corrt:sponding Angle of Arrival
373
However, since we are reversing the time series, we are also Output SINR is defmed as
reversing the directions of arrival of the incoming signals. To
compensate for this, we conjugate the voltages as shown in Figure
7.
where a is the signal strength. The signal comes in at 0 degrees,

broadside, with complex amplitude 1 + Oj. So if the output SINR is
40 dB, there is a 1% error in the estimation of the signal strength.
AU simulations were done in single precision on a Sun Sparcstation
10. The direction of arrival of the jammer is varied from .1 degrees
to IO degrees. At each jammer arrival angle, the jammer strength is
VN* v3* v2* V I *
increased until a 40 dB output SINR is exceeded.
Figure 7 Conjugated Voltages and Associated Angle of Arrival First consider the case of a single jammer without noise
(Figures 8,9). For a 5 element array, the main lobe is
approximately 22 degrees wide. So even with the jammer in the
The eigenvalue approach can now be applied to this new sequence. main lobe and with jammer to signa1ratios of over 40 dB, the signal
can be recovered within 40 dB output SINR. A 9 element array
yields about a 6 dB improvement in performance over the 5 element
array.
V. Numerical considerations
One can either look for the minimum eigenvalue of
xw = asw
or the maximum eigenvalue of
sw=(-+
Figure 8 Single Jammer with 40 dB Output SINR
9 elemenl array. angising. ng 5tr I . 1 ~ammer.max
Numerical stability is increased for the maximum eigenvalue. For

"----" ---
the minimal case, all the eigenvalues are squeezed into a range and
the minimum approaches zero (a compact operation) [3]. For the
maximal case, the upper limit is unbounded and precision is less of
an issue. Also the received voltage matrix, X,is less singular than
the desired look direction matrix, S. The QZ algorithm is
numerically more stable when the right matrix is non-singular.
VI. Simulation Results
The QZ algorithm was used to solve the generalized

eigenvalue equation [4]. Jammers are modeled as far field targets
with uniformly distributed amplitude and phase, arriving at different
angles.
374
Noise (modeled as a uniformly distributed random variable)
is now interjected on all channels (Figure 10). A 20 dB noise to
signal power was added. Again a signal of unit amplitude is coming
in at broadside. As the number of elements increase, the angular
separation between a relatively strong jammer and the signal (100
dB Jammer to Signal Ratio) required for a 20 dB output SINR
decreases. For a 21 element array, even when a 40 dB jammer is
within 0.5 degrees of the incoming signal, the signal strength can be
recovered with a 20 dB output SINR.
90
m
-20O I
- . . . . . L04 .
0 1
_i 2 3
,
4 5
I
6
’
7
...-----
8 9
I
10
5
p
no
Jammer Angle (Degrees) 7 0 ~
E
60
Figure 10 Single Jammer with Noisy Channels
50
Now consider the case of clutter (Figure 11). Clutter is

modeled as discrete jammers with 0.1 degree spacing between their
angles of arrival spanning from -90 to 90 degrees excluding a 1
degree pass region around the signal, representing a pre-Doppler Figure 13 STAP with Changing Jammer Angle
filter. A signal to clutter ratio of -15 dB yields an output SINR of
about 20 dB.
IV. Conclusions
The deterministic approach can give very fine resolution (20

dB output SINR) even when interfering targets are present in the
main lobe of the receiving array and mithin a degree of the desired
signal. There is some loss in the number of degrees of freedom as
compared to the statistical approaches since no secondary dlata is
used and the analysis is performed on a snapshot by snapshot basis.
However, since covariance matrices d13 not have to be inverted or
estimated, problems due to nonhomogeneities (such as blinking
jammers, multipath, and clutter edges) in the surrounding test cells
are eliminated.
[I] J. Ward, “Space-Time Adaptive Processing For Airborne

Radar,” Technical Report 1015, December 1994.
[2] Y. Hua and T. K. Sarkar, “A note on the Cramer-Rao Bound
for 2-D direction Finding Based on 2-11 Array,” IEEE May 1991.
Figure 11 Signal In Clutter [3] E. Kreyszig, “Introductory Functional Analysis with
Applications,” John Wiley and Sons, 1978.
[4] G. H. Golub and C. F. Van Loan, “Matrix Computations,” 2nd
In the STAP approach, first Doppler is kept the same as the ed, The Johns Hopkins University Press, 1989.
signal while the jammer angle is changed. The signal again arrives
at broadside. Even when the jammer is within a degree of the
signal, a 40 dB output SINR is attainable with a Jammer to Signal
ratio of over 40 dB. Next the jammer angle is kept constant at 2
degrees as the Doppler of’the jammer is varied. Now Jammer to
Signal ratios of at least 70 dB are required within 1 degree of the
signal direction of’arrival to go below an output SINR of 40 dB.
375

Abstract:: Also

Uploaded by

Copyright:

Available Formats

Abstract:: Also

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Abstract:: Also

Uploaded by

Copyright:

Available Formats

A Deterministic, Eigenvalue Approach to Space Time Adaptive Processing

Sheeyun Park and Tapan K. Sarkar

Abstract: The problem of Space Time Adaptive Processing

STAP techniques have arisen to address the issue of high clutter

Figure 1 Array Geometry y(t) = wTx = XTW

The signal arrives at each sensor at different times dependent on the

xk(t) = sk(t) + nk(t), k = 1, 2, ... , N (1)

0-7803-3232-6/96/$5,00 0 1 9 9 6 IEEE 372

[sz - as]x [E]

There is an implicit constraint in the look direction imposed by the

The structure of the pencil can be viewed as a linear

(see Figure 3). Equation (9) can be rewritten in a more familiar

xw = asw (10). Equation (13) is again a generalized eigenvalue problem.

Given a linear array, we can either do the linear prediction

Figure 5 Voltages and Angle of Arrival

Angle To consider the backward problem, first consider the

As an example consider M = 2, T = 2, and the first box. The

As long as the linear relationship between the positions in the box

where a is the signal strength. The signal comes in at 0 degrees,

One can either look for the minimum eigenvalue of

9 elemenl array. angising. ng 5tr I . 1 ~ammer.max

Numerical stability is increased for the maximum eigenvalue. For

VI. Simulation Results

The QZ algorithm was used to solve the generalized

Jammer Angle (Degrees) 7 0 ~

Now consider the case of clutter (Figure 11). Clutter is

The deterministic approach can give very fine resolution (20

[I] J. Ward, “Space-Time Adaptive Processing For Airborne

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.