You are on page 1of 7

Search & Detection Theory (OA3602) Lecture Notes Prof.

James Eagle
Text: Search & Detection, 4th Ed., Alan R. Washburn Rev: November 5, 2012

5 Lateral Range Curves and Sweepwidth

5.1 Lateral Range Curves

Definition of the lateral range curve l(x):

l(x) = P (detecting a target when the searcher’s track relative to the target
is infinitely long in both directions and with CPA x to the target) (1)

searcher track

x target

Figure 1: Lateral Range Geometry

1
1

-R 0 R 0
Cookie-Cutter Sensor Inverse Cube Law Sensor

1
1

0 -R m 0 Rm
Radar with Severe Sea Return Triangular Sensor

Figure 2: Example Lateral Range Curves

1
For an inverse cube law sensor, we have seen that
 
−2kAh
l(x) = 1 − exp ,
V (h2 + x2 )
where V here is the searcher’s speed relative to a stationary target.
Note that l(x) is neither a PDF nor a CDF. It’s rather a plot of Pd verses lateral range
(CPA) for straight-line tracks. Also note that l(x) is an aggregate measure of detection
performance in that it is immaterial where on the track the detection occurs. For l(x),
detections prior to CPA count the same as detections afterwards.
Example: A search of an area is conducted by four parallel sweeps with track spacing S,
as shown in Figure 3.

S S

1 2 3 4

Figure 3: Area Search with Parallel Sweeps

Assume a lateral range curve of l(·) and a stationary target at a distance of x from the
left-hand border of the search area. Then assuming tracks of infinite length, Pd is l(x − S)
for sweep 1, l(x − 2S) for sweep 2, l(x − 3S) for sweep 3, and l(x − 4S) for sweep 4. Now
assuming independence from sweep to sweep, the Pd for all four sweeps is
4
Y
Pd (x) = 1 − (1 − l(x − iS)) .
i=1

Now assume an inverse cube law sensor with h  x, so (x2 + h2 ) ≈ x2 . So,


2
l(x) = 1 − e−K/x ,

where K = (2kAh)/V . Then


4
Y
exp −K/(x − iS)2

Pd (x) = 1 −
i=1

2
4
!
X
2
= 1 − exp −K/(x − iS) .
i=1

In Washburn [2002][Sec. 2.4], the proof is sketched out that for an inverse cube law search
for a uniformly distributed target in a large area,
p 
Pd = 2Φ π/2(W/S) − 1, (2)
p
where W = 8πkAh/V is the sweep width (discussed in the next section) of an inverse cube
law sensor. This is an interesting result because it shows that inverse cube law searches are
better than random, but not as good as exhaustive. (See Washburn [2002][p. 2-4].)
Example: An aircraft with speed 200 kt and altitude of 1 nm searches with a track spacing
of 10 nm for a ground target of area .01 nm2 . For a target uniformly distributed over the
search area and an assumed k of 50,000 hr−1 , what is Pd after one sweep of the area?
p p
W = 8πkAh/V = 8π(50, 000)(.01)(1)/200 = 7.927 nm

So, after one complete sweep of the area,


p 
Pd = 2Φ π/2(W/S) − 1
p 
= 2Φ π/2(7.927/10) − 1
= .6795

Note that this model assumes one complete sweep of the search area, and does not require
specifying the search area size.

5.2 Sweep Width

Sweep width, W , is defined as Z ∞


W = l(x)dx. (3)
−∞

The units of W are distance. Sweep width is one of the standard measures of sensor
performance.
Note that if all the area under lateral range curve l(x) was placed in a rectangle 1 probability
unit high, the width of the rectangle would be W .
Examples:

3
1

l(x)

Figure 4: Area under l(x) is W .

1. Cookie-cutter sensor with detection range R.



1, x ∈ [−R, R]
l(x) =
0, otherwise

So, W = 2R.
2. Inverse cube law sensor.
2
l(x) = 1 − e−K/x , for K = 2kAh/V.
So,
Z ∞
2
W = 2 1 − e−K/x dx
√0
= 2 πK.
√ 2
(Use z = K/x, integrate by parts with u = 1 − e−z and dv = dz/z 2 , and
R∞ 2 √
the Gaussian integral 0 e−x dx = π/2.)
p
= 8πkAh/V .

3. Triangular sensor with maximum range Rm (see Figure 2).



1 − |x|/Rm , x ∈ [−Rm , Rm ]
l(x) =
0, otherwise

So,
Z Rm
W = 2 (1 − (x/Rm )) dx
0
= 2(Rm /2)
= Rm .

4
5.2.1 An Interpretation of Sweep Width W

Referring to Figure 5, assume a sensor with maximum detection range Rm and lateral
range curve l(x). Target CPAs are uniformly distributed over lateral ranges −L/2 to L/2,
where L/2 > Rm .

targets

W
l(x)

-L/2 -Rm searcher Rm L/2

Figure 5: Sweep Width Interpretation.

Then conditioning on the value of the random lateral range,


Z L/2
Pd = P (detection | lateral range = x) (density function for x) dx
x=−L/2
Z L/2
= l(x)(1/L)dx
x=−L/2
Z L/2
= (1/L) l(x)dx
x=−L/2
= W/L.

There are two main points of interest.


1. Pd is constant for all l(x) with the same W (i.e., area under the l(x) curve).
2. In particular, l(x) can be replaced with a cookie-cutter sensor with detection range
W/2 and the same Pd will result.
So when the target’s lateral range is uniformly distributed, a sensor with lateral range
curve l(x) can be equivalently replaced with a cookie-cutter sensor with detection range
W/2. Note, however, that if the lateral ranges are not uniform, this substitution is not
valid. Referring to Figure 5, if all targets had CPAs of either L/2 or −L/2, then Pd would
be 0.
Figure 6 illustrates how to compute the sweep width of an acoustic convergence zone
sensor.

5
Calculating the Sweep Width of a CZ Sensor

p0 = P{det| tgt crosses DP/BB region}


p1 = P{det| tgt crosses 1st CZ twice}
p2 = P{det| tgt crosses 2nd CZ twice}

R0 = radius of DP/BB region


Ri = distance to i th CZ, i = 1,2

R0 R1 R2

l2 = p2
Lateral l0 l1 = 1 - (1-p1)(1-p2)
Range l0 = 1 - (1-p0)(1-p1)(1-p2)
Curve
l1 W = 2[(l0 R0) + l1(R1-R0) + l2(R2-R1)]

l2

R0 R1 R2

Example values:
R0 = 1 nm p0 = .85 l0 = .9435
R1 = 35 nm p1= .5 ⇒ l1 = .625
R2 = 70 nm p2 = .25 l2 = .25
W= 2(30.94) = 61.88 nm

Figure 6: Calculating the Sweep Width for a CZ Sensor.

6
References
S. Christian Albright. VBA for Modelers. Thomson-Brooks/Cole, 2nd edition, 2007.

Daniel H. Wagner, W. Charles Mylander, and Thomas H. Sanders. Naval Operations


Analysis. Naval Institute Press, 3rd edition, 1999.

Alan R. Washburn. Search and Detection. INFORMS, 4th edition, 2002.

Alan R. Washburn. Notes on game theory, 2000.