ASW Dipping 2018
ASW Dipping 2018
ASW Dipping 2018
Where to Dip?
Search Pattern for an Antisubmarine Helicopter Using a Dipping
Sensor
Abstract
Anti-submarine warfare (ASW) had been an important topic for military operation
research (MilOR) modelers and analysts during World War II and the Cold War. It
became however somewhat out of vogue with the collapse of the Soviet Union and the
subsequent reduction of the threat of submarine-related conflicts. In recent years, threats
of such engagement have increased, in particular in the South China Sea. The re-
emerging interest in this type of warfare, combined with new technologies and resulting
tactics, pose a renewed challenge for MilOR researchers. We study effective ways to
operate a helicopter, equipped with dipping sonar – a dipper – in ASW missions. In
particular, we examine the dipping pattern and frequency. A high rate of dipping is
desirable as search effectiveness degrades in time as the search area expands. However,
dipping too frequently results in overlap with previous dips, which may be wasteful. For
a cookie-cutter sensor and a known constant submarine velocity, we prove that disjoint
dips are optimal and generate the corresponding optimal dipping pattern. We analyze the
effect of factors, such as helicopter speed, submarine speed, sensor detection radius, and
travel time to the point of detection, on the optimal dipping pattern. We show that
temporal parameters – submarine velocity and helicopter arrival time to the datum – are
most critical. We also show that the no-overlap result is not always true; when the
submarine’s velocity is only known with probability, the optimal dipping frequency may
include overlaps.
1. INTRODUCTION
Submarines pose a major threat to naval ships and therefore submarines become prime
targets during naval operations. However, detecting and engaging these targets is
challenging due to their stealth and high endurance. A common practice in modern anti-
submarine warfare (ASW) is to send out helicopters equipped with dipping sonar, which
allows the helicopter crew to listen for underwater signals while hovering at an altitude of
50 to 300 feet above sea level (Global Security, 2016). The helicopter uses a cable to
lower the sensor to the desired depth, which can range from the just below the surface of
the sea to 2,500 ft (Global Security, 2016). The dipping sonar is primarily an active
sensor, and hence the sonar generates sound signals once lowered into position. Signal
processing algorithms process the echoes that return to the sensor to locate enemy
submarines (Global Security, 2016). In many situations, such helicopters are dispatched
to search and hunt a submarine following a cue received from some exogenous
surveillance source such as fixed-wing aircraft or towed arrays from surface ships. This
source provides the location (known as the datum) of the suspected target and the time of
detection. Given this datum, the question is what would be the optimal dipping pattern for
the search helicopter. The shape and size of this pattern can indicate if it would be
worthwhile to dispatch the helicopter. We examine a more specific question in this paper:
given the current dipping location, when and where should the next dip occur? On one
hand, the dipping frequency should be high as search effectiveness degrades in time as
the submarine moves and the search area expands. On the other hand, dipping too
frequently may result in overlap with previous dips, which may lower search efficiency.
The (mathematical) problem of search and detection has been studied for the past 70
years. The ground-breaking work of Koopman (Koopman, 1946) laid the foundation for
this area of research. Other seminal works in general search theory are (Stone, 1975),
(Haley and Stone, 1980), and (Washburn, 2002). Search models specific to ASW
operations appear in (Shephard, et al., 1988), where a helicopter, equipped with sonar
buoys and torpedoes, is out to hunt a submarine. Their model assumes a uniform
deployment of the sonar buoys in the containment circle and computes the optimal
payload of buoys and torpedoes.
3
Several papers study dipping sonar tactics. Baston and Bostock (1989) examine where a
helicopter should drop a finite number of cookie-cutter bombs to destroy a mobile
submarine. The two entities move on a one-dimensional line and this limits the spatial
impact of the increasing search area, which is crucial for our analysis of the tradeoff
between searching frequently vs. limiting search overlap. Washburn and Hohzaki (2001)
and Soto (2000) consider mechanical limitations on a submarine’s velocity. They
transform the discrete dips into a continuous search rate and examine the problem from a
random search perspective. Thus, there is no analysis of when and where to discretely dip
next.
Danskin (1968) has a very similar setup to our problem with a cookie-cutter dipping
sensor. He postulates that a discrete dipping spiral pattern may be particularly effective
(we show under certain assumptions, it is optimal). However, Danskin does not calculate
the specific time and location of individual dips. He assumes that dips will be disjoint,
which is the primary focus of our analysis. In this paper, we show that, using a similar
framework to Danskin’s, disjoint dips are not necessarily optimal. Thomas and
Washburn (1991) and Chuan (1988) also have a similar framework to our model. These
papers (as well as Danksin (1968)) consider the decreasing effectiveness of dips over
time as the search area increases. Thomas and Washburn (1991) formulate a complex
dynamic program to generate a search plan. They do account for the negative impact of
traveling too far for the next dip, but they do not explicitly consider the negative impact
of overlap as the target can move to any cell in the region between dips. Chuan (1988)
does allow for overlap in practice due to operational inefficiencies, but assumes that in
theory the dips should be disjoint.
localize their prey. In other scenarios, the searcher may move passively and activate the
sensor only at discrete times and locations to mitigate counter-detection. While we focus
on ASW dipping sonar in this paper, there are other applications where our models and
results could be useful.
Our main contribution is in examining the tradeoff between dipping frequency and search
overlaps. Most work takes for granted that dips should be disjoint. While we find that to
be the case under some assumptions, disjoint dipping is not necessarily optimal under
other assumptions. In this paper, we primarily focus on a deterministic submarine
velocity. The main result is a provable optimal dipping pattern that dictates how the
search helicopter should dynamically deploy its dipping sensor. The key characteristic of
the optimal dipping pattern is that the next dip location is the closest valid dipping point
to the current location that produces a disjoint dip. Additional insights relate to the effect
of operational and physical parameters on the shape and size of the resulting search
spiral. We also consider a random submarine velocity and show that the optimal dipping
strategy may incorporate overlaps.
The rest of the paper is organized as follows. In the next section, we describe the
operational setting, which is followed with the statement of the main result for a
deterministic submarine velocity. We discuss the case of a non-deterministic distributed
submarine velocity in the fourth section. The fifth section presents sensitivity analysis
regarding some key operational and physical parameters for the deterministic case. In the
sixth section, we assume some partial knowledge about the bearing of the hunted
submarine and show how this knowledge affects the dipping pattern. Concluding remarks
appear in the seventh section.
2. OPERATIONAL SETTING
A naval task force is equipped with an antisubmarine warfare helicopter whose role is to
hunt and kill enemy submarines. The helicopter is dispatched upon receipt of information
about the location of a potential submarine target. The source of such information is
typically a long-range anti-sub patrol unit continuously surveying the operational area of
the task force (e.g., P-8 anti-sub aircraft or a surface ship equipped with a sonar device or
5
even a satellite). Launching a helicopter for an ASW mission is costly both economically
and operationally. In particular, the helicopter may have other competing missions.
Arguably, the decision to launch the ASW helicopter should depend on the probability of
mission success. This probability is affected by the distance from the launching site to the
datum, the helicopter velocity and endurance, and the submarine velocity. These factors
are manifested in the shape and size of the search spiral (see next section). Throughout
most of this paper, we assume that the searcher knows the sub’s velocity, and thus its
distance from the datum, but not its bearing.
The helicopter is equipped with dipping sonar (henceforth referred to as a dipper), which
is “a sonar transducer that is lowered into the water from a hovering antisubmarine
warfare helicopter and recovered after the search is complete” (FreeDictionary, 2016).
Depth matters for our analysis in that it affects how long it takes to deploy the dipper and
reel it back. However, we take the dipping time to be a fixed constant, and thus for this
paper we assume that the dipper has a two-dimensional circular cookie-cutter detection
function. That is, the detection range is arbitrarily deep and we ignore possible evasive
actions by the submarine going deeper or shallower. We also assume a perfect sensor: if
the submarine is present within the dipper’s circular footprint – the detection circle – the
dipper will detect the submarine with certainty. Otherwise, the submarine remains
undetected.
While the velocity of the sub is assumed to be constant and known to the searcher, its
bearing is unknown and assumed to be uniformly distributed on [0, 360o]. This
assumption is relaxed later in the paper. Thus, the location of uncertainty (LoU) – the
possible locations in which the submarine may be present – is a circumference of a circle
with a radius that is determined by the velocities of the sub and the helicopter, and the
distance the helicopter has to travel to the datum.
A dipping pattern is a series of consecutive dipping points for the dipper. A dipping
pattern is optimal if, for a given number of dipping points, it maximizes the probability of
detection, or, for an infinite number of available dipping points it minimizes the expected
time of detection. Because the sub velocity is assumed to be known, the searching
helicopter would know exactly the submarine’s location, had the searcher known the
sub’s heading. Thus, at any given time, the circumference of the circle around the datum
on which the sub is located – the location circle – is uniquely determined. The coverage
of a dip is the arc on the circumference of the location circle that is covered by a dip,
which is equal to the angle α , rooted at the datum, between the two tangents to the
detection circle (See Figure 1).
Continuous search patterns over an expanding circle are well known to be a spiral with a
shape dictated by the velocity of the sub (Washburn, 1980; 2002). As our search is
discrete, we must determine where on the spiral to next dip. There are essentially three
generic dipping patterns: overlapping, tangential and excessively disjoint (see Figure 2).
Following a dip, the helicopter can travel a short distance and dip again (Figure 2a), in
which case the coverage of the second dip is relatively large, but part of it overlaps with
the coverage of the first dip.
If the helicopter travels farther away the coverage shrinks but the overlap disappears
(Figure 2b). The tangential dip is the closest disjoint dip. If the helicopter travels even
farther, the coverage is even smaller and there are some gaps in the area searched (Figure
2c). While, evidently, excessively disjoint dips are suboptimal, it is not obvious which of
the two cases – overlapping dips or tangential dips – is better. Specifically, while dip 2 in
Figure 2a has a larger coverage than dip 2 in Figure 2b, it is not clear if the effective
coverage of dip 2 in Figure 2a, i.e., the angle between the right tangent of dip 2 and the
right tangent of dip 1 in Figure 2a, is larger or smaller than the coverage of dip 2 in
Figure 2b. We prove that the latter is true; tangential dipping is optimal.
Let U and V denote the velocities of the submarine and helicopter, respectively. The
dipper detection range is R and the time duration of a dip is τ D . Let (0,0) denote the
location of the datum, Pi = ( X i , Yi ) be the location of the i-th dipping point, and Ti is the
time, measured from the moment the external surveillance source delivered the datum,
8
the i-th dip starts. In particular, T1 is the time the helicopter arrives to the first dipping
point.
Theorem 1: For a given number of dips, tangential dips maximize the probability of
detection.
The proof of the theorem appears in Appendix A. An optimal dipping pattern appears in
Figure 3. To derive the actual expressions for the i-th dipping point, Pi , and the start time
Similarly to the framework described in Danskin (1968), suppose that immediately after
the surveillance asset detects the submarine at the datum, the submarine’s velocity and
heading are randomly initialized, and the submarine maintains these two values
throughout the search.
If the submarine’s velocity is a random variable that takes on only a finite number of
values, the analysis in Section 3 generalizes in a natural way with multiple spirals: one
corresponding to each velocity. If there are multiple searchers, then each searcher dips
on one spiral. If there is only one searcher, then we must determine the order the searcher
should process the different velocity-spirals. For more details see (Ben Yoash, 2016).
We next examine a similar tradeoff between timeliness and dipping overlaps as in the
deterministic velocity case discussed in Section 3. For simplicity, we ignore here the
dipping time τ D . Given the dipper has a cookie-cutter detection function with radius R,
and assuming the dip footprint is entirely within the containment circle, the probability
R2
the first dip detects the target is 2 . If the second dip occurs at time T2= T1 + ∆t ,
ρ (T1 )
then the contribution to the overall detection probability from the second dip is
p R 2 − overlap(∆t )
, where overlap(∆t ) is the area of overlap between two circles: the
pr 2 (T1 + ∆t )
second dip footprint and the area cleared by the first dip. For ∆t =0 , there is complete
overlap between the first two dips (overlap(0) = p R 2 ) , which results in a worthless
search effort. For a large enough ∆t , eventually there is no overlap (overlap(∆t ) =0 , for
large ∆t ). The exact expression for overlap(∆t ) is somewhat complicated and appears in
Appendix C. In general, overlap(∆t ) will decrease with ∆t , and thus both the numerator
and denominator increase in ∆t . In Section 3 we showed, for the deterministic velocity
case, that at optimality the next dip satisfies overlap(∆t ) =0 : tangential dips are optimal.
This is quite an intuitive result and it is taken for granted in other works (e.g., Chuan
(1988), Danksin (1968)). The following theorem states that this result is not necessarily
optimal when the submarine velocity is not deterministic.
Theorem 3: When the submarine heading and velocity have a bivariate uniform
distribution over the speed circle of radius U max , the optimal dipping frequency may
include overlaps.
The details of the examples demonstrating this property require tedious calculations
involving the area of the intersection of circles. We defer these examples to Appendix C.
The optimal search pattern should include overlap when the helicopter arrives to the
datum very soon after detection. The initial dip produces a relatively large detection
probability. Because the detection probability from future dips decreases quickly (
11
∝ (time) −2 ), the searcher benefits from taking the next dip soon after, even though the
second dip partially overlaps with the first dip. The amount of overlap increases with a
slow searcher as it takes longer to move to a location that produces a disjoint dip.
For most current ASW dipping scenarios, the helicopter will be much faster than the
submarine, so a disjoint dipping strategy should be near optimal for most realistic
parameters. To determine the specific times and locations for the optimal dipping pattern
in the bivariate uniform scenario is a challenging problem that requires much more
complicated machinery than we utilize in this paper. For an example of how one could
proceed, see the dynamic programming approach in Thomas and Washburn (1991). For
the remainder of the paper we return to the deterministic velocity scenario.
Next we analyse the effect of operational and physical parameters on the shape of the
optimal dipping pattern for deterministic submarine velocity. We start off with a base
case that reflects typical values of the various parameters. Specifically, helicopter speed V
= 100 knots, submarine speed U = 8 knots, time of arrival to first dipping point T1 = 2
hrs, detection range R = 2 nm and dipping time τ D = 5 min.
The helicopter chases the submarine and therefore the faster the helicopter operates the
smaller would be the area of uncertainty and therefore also the dipping spiral, as shown in
Figure 4. While a velocity of V = 200 nm is obviously unrealistic for a helicopter, we
observe that speed has decreasing marginal effect; the decrease in the spiral radius as a
result of velocity increase from 50 knots to 100 knots is larger than the effect when the
speed increases from 100 knots to 200 knots.
12
The marginal effect of speed is demonstrated in the number of dips and the time it would
take the helicopter to complete a full (360o) spiral. See Figure 5. From the top plot in
Figure 5 we see that as the speed of the helicopter increases the flat parts of the plot
become longer. That is, the sensitivity of the number of dips to changes in speed
decreases as the helicopter travels faster. The bottom plot shows that the effect of
helicopter speed on the time to complete a full spiral is strictly monotone decreasing – as
one would expect. The discontinuities in the plot, which are aligned with the jumps in the
upper plot, correspond to unit decreases in the number of dips.
13
The decision to dispatch the search helicopter – the “go/no-go” decision – is crucially
affected by the time T1 it takes the helicopter to arrive at the first dipping point. For a
given cruising speed of the helicopter, the arrival time is determined by the distance from
the take-off site to the datum. Even if unrealistically we assume limitless endurance for
the helicopter, that is, it could always complete a full spiral, the effect of arrival time on
the shape of the spiral is quite significant, as shown in Figure 6.
14
With limited endurance the effect of lower speed becomes even more significant; slower
speeds directly create larger spirals (see Figure 4) and therefore more dips are needed for
a given coverage, but slower speeds also increase T1, which further increases the size of
the spiral.
Similar effects occur when we vary the time it takes to execute a dip, as shown in Figure
7. As one would expect, longer dipping times generate bigger spirals but, surprisingly,
while there is barely any difference between 2.5- and 5-minute dips, there is a significant
change between 5- and 10-minute dips.
15
We observe that doubling the detection range from 1.5 miles to 3 miles reduces the
number of dips by more than a factor 3. The effect of detection range on the number
of dips and duration of a complete search appears in Figure 9.
Figure 9: Number of Dips and Search Time for Varying Detection Ranges
As observed above, for small detection ranges (e.g., less than 1.5 mile) the effect of
marginal improvement in range is super-linear, which is not the case for larger
detection ranges where the marginal effect is negligible.
complete a spiral. For comparison, reducing the helicopter speed from 100 knots
(base case) to 50 knots increases the number of dips by less than 30%.
In reality the searcher will not know exactly the submarine speed, only an estimate based
on intelligence sources. The results reported above hold even when there is some
uncertainty about the actual speed as long as the velocity error produces locational errors
within the detection range of the dipper over the course of the search period.
Thus far we assume that that the searcher has no information about the sub’s bearing and
therefore each direction of movement of the sub is taken to be equally likely. In some
situations, however, additional information about the sub’s bearing may be available and
could be utilized to improve the effectiveness of the search. Suppose that the bearing of
the submarine may be in one of three possible wedges of the LoU having angular sizes
α , β and γ , α + β + γ ≤ 360α , with probabilities q, p and 1 − p − q , respectively. See
Figure 11. The direction within a wedge is uniformly distributed, which implies that the
18
optimal dipping pattern within each wedge is derived from Theorem 1, and manifested by
a partial spiral of tangential dips.
The question now is in what order to search the wedges. If the wedges are searched
sequentially – IIIIII or IIIIII – then the dipping pattern is a spiral that starts
on the left ray of wedge I or the right ray of wedge III, respectively. Otherwise, the
helicopter has to “hop” over wedges and the dipping pattern is no longer a
contiguous spiral. For example, if the search order is IIIIII then the helicopter
starts the dipping at a point on the right ray of the middle wedge (II) and spirals
towards the left ray of I. Once it reaches that ray, it flies back to a certain point,
farther away on the right ray of II and resumes the search towards the right ray of
III. See Figure 12. Note that unlike the continuous searches described for the cases
IIIIII or IIIIII, in this search pattern there is some wasted “lull” time when
the helicopter moves from wedge I to wedge III (the thin arrow in Figure 12). The
objective is to minimize the expected time to detection and therefore such
discontinuous dipping patterns are possible if, for example, p > q >> 1 − p − q .
19
V
the helicopter and the sub is =
S = 10 . Each region is labeled with the number
U
corresponding to the search order presented above. For example, region 1 contains
all the (p,q) values for which the wedges search order is I → II → III .
20
We see that the larger p is (probability that the submarine is in wedge II), the more
likely we are to start in that middle wedge (search orders 3 and 4). If q (probability
of the submarine in the wedge I) is relatively large, then it is more likely that the
search will start at wedge I (patterns 1 and 2).
Figure 14 demonstrates the effect of varying the sizes of the two side wedges – the
angles α and γ - on the optimal order. We vary them together, keeping the two
( β 30
other parameters constant= =
, S 10 ).
21
=
Figure 14: Three Wedges Model, =
S 10, β 30, α and γ Varied.
We see that increasing the angles of the side wedges decreases the regions of (p,q)
where patterns 2 and 5 – orders in which the center wedge is searched last.
Although this seems counterintuitive, the explanation is that searching the center
wedge last means that the helicopter has to fly over a side wedge, which has already
been searched, back to the center. This wasted flying time increases as the side
wedge becomes wider. We also see that regions 3 and 4 increase, that is, starting the
search at the center wedge becomes more common, when the side wedges increase
in size. This happens because wider wedges imply lower probability per unit angle,
which make them less attractive in terms of “bang-for-the-dip” – the expected
reward from a dip. We notice however that for side wedges wide enough the
changes are marginal (e.g., see the plots for 45o and 60o.)
Last we examine the sensitivity of the search order to the speed ratio S between the
helicopter and the submarine. We assume α= β= γ= 30α throughout. Figure 15
depicts this sensitivity:
22
7. CONCLUSIONS
The US Navy MH-60R helicopter may be equipped with dipping sonar for detecting
and localizing adversaries’ submarines. This discrete search pattern is different
from more common continuous searches. In this paper, we present an analysis of
the optimal dipping frequency. We primarily focus on the deterministic submarine
velocity scenario and derive an optimal dipping pattern: the optimal next dipping
location is the closest point that produces a disjoint dip. We investigate the effect of
various operational and physical parameters on the characteristics of the dipping
pattern. We observe that temporal parameters – time to arrival to the datum and
velocity of the submarine – have significant effect on the dipping pattern and the
time to complete a full coverage (spiral) of the submarine location. We also examine
the case where the submarine velocity is a random variable. Disjoint dips are not
necessarily optimal in this scenario when the helicopter arrives on station quickly.
However, for most realistic parameter values, a disjoint dipping strategy should
perform near optimally. Future work could consider more realistic detection
functions and more complex target dynamics, such as counter-detection.
ACKNOWLEDGEMENT
This paper is based on the Master's Thesis of Roey Ben Yoash (first author).
REFERENCES
Baston, V.J. and Bostock, F.A., 1989. A one‐dimensional helicopter‐submarine game.
Naval Research Logistics, 36(4), pp. 479-490.
Ben Yoash, R.. 2016. Anti-Submarine Warfare Search Models. Naval Postgraduate
School, Monterey, CA.
Chuan, E. 1988. A helicopter submarine search game. Master thesis, Naval Postgraduate
School, Monterey, CA.
Danskin, J.M. 1968. A helicopter versus submarine search game. Operations Research,
16(3), pp. 509-517.
The Free Dictionary by Farlex, s.v. dipping sonar, retrieved June 7, 2016,
http://encyclopedia2.thefreedictionary.com/dipping+sonar
24
Haley, B.K. and Stone L.D. 1980. Search Theory and Applications. Plenum Press, New
York.
Kim, K-M. and Lee, S-H. 2012. Approximating the Poisson Scan and (λ−σ) acoustic
detection model with a Random Search formula. Computers and Industrial Engineering
62(3), pp. 777-783.
Marsden, J.E. and Hoffman, M.J., 1993. Elementary classical analysis. Macmillan.
Shephard R.W., Hartley D.A., Haysman P.J., Thorpe L. and Bathe M.R. 1988. Applied
Operations Research. Plenum Publishing Corporation, New York.
Stone, L.D. 1975. Theory of Optimal Search. Academic press, New York.
Soto, A. 2000. The Flaming Datum problem with varying speed. Master thesis, Naval
Postgraduate School, Monterey, CA.
Thomas, L.C. and Washburn, A.R., 1991. Dynamic search games. Operations Research,
39(3), pp.415-422.
Washburn, A. R., and Hohzaki, R. and 2001. The diesel submarine flaming datum
problem. Military Operations Research, 6(4), pp. 19-30.
Washburn, A. 2002. Search and Detection, 4th edition. Institute for Operations Research
and the Management Sciences, Linthicum, MD.
Given a particular dip (not necessarily the first one) that occurs at time T, we
compute the coverage angle α . See Figure A.1 for reference. The dip footprint DP in
Figure A.1 is the circle centered at the dip point with radius R (the dipper’s detection
( )
range). The coverage angle α satisfies sin α 2 =
R
T ×U
, so
R
α = 2sin −1 ,
T ×U
where we assume the inverse sin function returns degrees.
26
Looking at Figure A.2, we claim that after dipping at point Pi the best next dipping
disjoint dip. That is the dip footprint DP* is tangent to the same tangent line of
i +1
footprint DPi but “from the other side” (as shown in Figure A.2). We prove this in
Section A.2 and provide a mathematical representation for Pi +1 at the end of this section.
*
Before turning to the proof, we make a few observations and introduce additional
notation. The i-th dip begins at time Ti at location Pi . The datum is determined at
time 0, and time is measured since that event. We define Pi using a modified polar
the datum at the i-th dip, and θi is the angle, rooted at the datum, measured
clockwise with respect to the vertical axis. Because the submarines velocity is
known, the radial component K i +1 of Pi +1 is uniquely determined and therefore to
determine the next dipping location Pi +1 (not necessarily optimal), we only need to
valid Pi +1 along the ray defined by θi +1 from the current dipping point Pi . This
determines the actual next dipping point. More rigorous mathematical details
appear in Section A.2.
Since we only need to determine the angle of the next dip, we first define
ω=
i +1 θi +1 − θi , which is the angle created by the previous dip, the datum, and the new
dip, as illustrated in Figure A.3. We next define f (ω ) as the effective coverage of a
dip with angular difference ω .
If ωi +1 = 0 then the helicopter’s next location Pi +1 lies on the same ray as Pi . Pi +1 does
Consequently, f (0) = 0 because dipping along the same ray as the previous dip will
produce no additional coverage relative to the previous dip.
29
In most realistic scenarios the helicopter does not arrive to the datum fast enough to
find the submarine with one dip at the datum, i.e., R < K1 = T1 × U , where K1 is the
distance of the first dip from the datum and θ1 = 0 is its angle measured clockwise
from the vertical axis. Subsequent dips occur in a clockwise fashion and hence the
angles θi form a monotonically increasing series. The duration of a dip is τ D and its
location remains stationary throughout; the helicopter hovers over the dipping
point.
We conclude this section by presenting the formulas we use to compute the optimal
location of the next dip Pi +1 = ( K i +1 , θi +1 ) given current dip location Pi = ( K i , θi ) . These
* * *
two equations simultaneously solve for displacement angle ωi +1 and the time of
*
We first relate the current position Pi to the next dip position Pi +1 (not necessarily
distance from the datum to the next dipping point, Pi +1 , is K i +1= U × Ti +1 . The
helicopter departs from Pi at time Ti + τ D after finishing the dip and arrives to Pi +1 at
time Ti +1 , and therefore the distance between Pi and Pi +1 is V × (Ti +1 − (Ti + τ D )) , where
V is the helicopter’s velocity. Using the Law of Cosines (assuming ωi +1 < 180 ), we
Marsden and Hoffman(1993)). The only conditions we need are Ti , Ti +1 > 0 , which
follow by assumption, and then Ti +1 is continuous in ωi +1 for all 0 < ωi +1 < 180.
(U × Ti ) 2 + (U × Ti +1 ) 2 − (V × (Ti +1 − Ti − τ D )) 2
ωi +1 = cos −1 .
2 × U 2 × Ti × Ti +1
Thus using equation (A.3) (or (A.1)), the angle differential ωi +1 , uniquely determines
the time of the next dip such that the radial position of the helicopter corresponds to
the radial position of the submarine.
31
R
f (ωi +1 ) α=
= − g (ωi +1 ) 2 × sin −1 − g (ωi +1 ) (A.4)
Ti +1 × U
Recall that the overlap is the angle between the left tangent to DPi+1 and the right
αi R
θi + =θi + sin −1 ,
2 Ti × U
and the angle between the vertical axis to the left tangent to DPi+1 is
α i +1 R
θi +1 − =θi +1 − sin −1 .
2 Ti +1 × U
The overlap is the difference between these two angles, that is,
32
R −1 R
g (ωi +1 ) = θi + sin −1 − θi +1 + sin
Ti × U Ti +1 × U
R −1 R
=θi − θi +1 + sin −1 + sin ,
Ti × U Ti +1 × U
which simplifies to
R −1 R
g (ωi +1 ) = −ωi +1 + sin −1 + sin (A.5) .
Ti × U Ti +1 × U
Our candidate for the optimal next position Pi +1 is the closest dip to Pi with no
*
overlap, and hence g (ωi +1 ) = 0 . Therefore, to derive condition (A.2), we set (A.5) to
*
R R
Both and are positive and less than 1 by assumption. Furthermore,
Ti × U Ti +1 × U
Ti +1 is a continuous function of ωi +1 (see discussion following (A.3) above).
Now that we have derived conditions (A.1) and (A.2), and showed that f (ωi +1 ) is
which satisfies equations (A.1) and (A.2). That is Pi +1 produces a higher effective
coverage than Pi +1 . Pi +1 ”shares” a tangent with the current location Pi (see Figure
* *
A.2). Therefore, Pi +1 must be closer to the datum than Pi +1 because a location farther
*
away will obviously have a smaller coverage. We claim that ω i +1 < ωi*+1 must hold.
Because Pi +1 lies closer to the datum than Pi +1 , it follows that K i +1 < K i +1 ⇒ Ti +1 < Ti +1
* * *
(see Figure A.4). Condition (A.1) ensures that pairs of (ωi +1 , Ti +1 ) produce valid
If indeed Pi +1 covers a larger angular section than Pi +1 , we next argue that there
*
must exist a valid dipping point Pj that is reachable by the helicopter in time to dip
and has the same effective coverage as Pi +1 , as shown by the middle circle in Figure
*
A.6.
a better location to dip than Pi *+1 , then f (ω i +1 ) > f (ωi*+1 ) . Finally, we showed
ω i +1 < ωi*+1 in the previous paragraph. Putting these pieces together with the
Intermediate Value Theorem, there is an ω j (and therefore Pj ) for which
f (ω j ) = f (ωi*+1 ) and 0 < ω j < ω i +1 < ωi*+1 . This implies that Pj is closer to the current
dipping point than Pi +1 and produces the same effective coverage. Figure A.7
*
is closer to DPi than DP* (See Figure A.6). Consequently, to generate the same
i +1
effective coverage, DP* and DPj must both be tangent (on the right-hand side) to the
i +1
same ray from the datum. This follows because the effective coverage is the angle
created by the right tangent to DPi and the right tangent to both DP* and DPj . We
i +1
call the later ray the “ray of coverage” (See Figure A.6). From the geometry
displayed in Figure A.6 it follows that the line through Pi +1 and Pj (dotted line in
*
Figure A.6) is parallel to the “ray of coverage” at a distance R away from the ray.
We now show that the existence of Pj leads to a contradiction. We first reintroduce
the parameter K, which is the distance from the datum to the dipping point of
interest. We next define Disti ,i +1 as the distance between Pi to Pi *+1 . We note that
K * − Ki
Disτi ,i +1 =
V × i +1 − τ D is the distance the helicopter travels while the
U
submarine moves between the two radii K i and K i +1 . Similarly, we define
*
K j − Ki
Disτi , j =
V × − τ D as the distance between Pi and Pj . The distance Dist j ,i +1
U
between Pj and Pi +1 can be found by considering the dotted line in Figure A.6,
*
observe that
35
K j − Ki
Disτi , j + Disτ j ,i +1 = V × − τ D + ( K i*+1 − K j )
U
K j − Ki ( K i*+1 − K j )
<V × +V × − Vτ D
U U
V
= [( K j − K i ) + ( K i*+1 − K j )] − Vτ D
U
V
= ( K i*+1 − K i ) − Vτ D
U
K* − K i
= V × i +1 −τ D
U
= Disτi ,i +1
The inequality part of the above expression follows from the fact that V > U ; the
helicopter moves faster than the submarine. The inequality implies that we found a
path from Pi to Pi *+1 that is shorter than Disti ,i +1 , contradicting the fact that Disti ,i +1 is
the shortest distance from Pi to Pi *+1 . We conclude that there is no location Pi +1 that
θi*+=
1 θi + ωi*+1
We assume the searcher makes two dips. The searcher arrives at time T1 and,
For the second dip, the closest disjoint dipping location is defined by K 2 = U × T2
D D
and θ 2D = θ1 + ω2D , where T2 and ω2 are the solutions to the set of simultaneous
D D
equations defined by (A.1) and (A.2). Given these parameters, the overall detection
probability for this disjoint 2-dip pattern is
36
(α1 + α 2D )q
Pdisjoint [detect ] = ,
360
where
−1 R R
=α1 2sin
= , α 2D 2sin −1 D .
T1 × U T2 × U
We contrast the disjoint 2-dip pattern with the other extreme: a complete overlap 2-
dip pattern. If the 2nd dip completely overlaps the first, the searcher wants the 2nd
dip to occur as quickly as possible (to maximize the size of the overlap). We define
T2O and ω2O to represent the position of the closest overlap dipping location. This
overlap position occurs when θ 2O = θ1 = 0 and ω2 = 0 . See Figure A.4 for reference.
O
In this case we have a closed form expression for the time of the 2nd dip as equation
V
(A.1) simplifies considerably: T2O= T1 + τ D . The detection probability using
V-U
this complete overlap 2-dip pattern is
(α1 − α 2O )q + α 2O (1 − (1 − q ) 2 )
Poverlαp [detect ] = ,
360
where
−1 R R
=α1 2sin
= , α 2O 2sin −1 O .
T1 × U T2 × U
Comparing Poverlap [detect ] to Pdisjoint [detect ] , the searcher should implement an overlap
• V = 7 kts
• U = 6 kts
• R = 4.5nm
• T1 = 1 hour
• t D = 1 minute
37
• q = 0.4
For the above parameters Poverlap [detect ] = 0.164 and Pdisjoint [detect ] = 0.129 . This is not
a realistic scenario as the searcher velocity V will likely be much higher than the
target velocity U.
In Section C.1 we derive the formulas for computing the overlap and detection
probabilities of two successive dips. In Section C.2 we present examples where
overlapping dips are optimal.
ρ1 ≤ ρ (T1 ) =
U maxT1 . Without loss of generality we assume that the first dip location
assume that T1 and ρ1 satisfy R ≤ ρ1 ≤ ( ρ (T1 ) − R) . That is, the first dip is entirely
contained within the upper part of the containment circle. See Figure C.1 for an
illustration. The solid −�− circle in Figure C.1 represents the footprint of the first
dip at time T1 , and the dotted ···�··· circle is the boundary of the containment
circle: a circle of radius ρ (T1 ) . Because the heading and velocity are uniformly
distributed within the speed circle of radius U max kts, the location of the submarine
R2
nm. Therefore, the detection probability of the first dip is .
ρ 2 (T1 )
Figure C.1: Disjoint dips (left panel) vs Overlapping dips (right panel).
The second dip occurs at time T2= T1 + ∆t . The first dip “clears” from the speed-
∆t
T2= T1 + ∆t , this cleared circle has expanded in real-space to a radius of R 1 + ,
T1
∆t
centred at point 0, ρ1 1 + . See Figure C.1 for an illustration of the cleared
T1
circle shifting and expanding in real-space and time. The solid −�− circle is the first
dip footprint at time T1 and represents the cleared circle at time T1 . As time
39
progresses to time T2 , the cleared circle expands north to the dashed − − × − − circle.
The solid − × − circle is the second dip footprint at time T2 . The second dip is disjoint
from the first if the solid − × − circle and dashed − − × − − circle do not overlap.
R2
If the second dip is disjoint, the detection probability is . The best
ρ 2 (T1 + ∆t )
disjoint strategy corresponds to the smallest ∆t that produces a disjoint dip, which
occurs when the helicopter heads due south after the first dip. To compute this best
disjoint time, which we denote ∆t ∆ , we determine when the distance between the
center of the cleared circle and the second dip location (the distance between the
two × circle centers in Figure C.1) equals the sum of the two radii:
∆t ∆∆ ∆t
ρ1 1 + − ( ρ1=− V ∆t ∆ ) R 1 + + R
T1 T1
2R
⇒ ∆t ∆ = (C.1)
( ρ − R)
V+ 1
T1
Any ∆t < ∆t ∆ will produce overlap. We define overlap(∆t ) as the area of overlap when
the next dip occurs at time ∆t < ∆t ∆ . To compute overlap(∆t ) requires calculating the
area of intersection between the following two circles:
∆t ∆t
Cleared circle (dashed − − × − − ): center = 0, ρ1 1 + , radius = R 1 +
T1 T1
The formula for this area of intersection appears in standard geometric references
(for examples, see equation (14) in Weisstein (2017)), and we provide it below for
our context:
40
2
2 ∆t
(d (∆t )) − R
2
T1
overlap(∆t ) = R 2 cos −1
2 Rd (∆t )
2
2 ∆t
2 ( d ( ∆ t )) 2
+ R
2 ∆t −1 T1
+ R 1 + cos
T1 2 R 1 + ∆t d (∆t )
T1
1 ∆t ∆t ∆t ∆t
− −d (∆t ) + R 2 + d (∆t ) − R d (∆t ) + R d (∆t ) + R 2 + ,
2 T1 T1 T1 T1
ρ
where d (∆t )= 1 + V ∆t is the distance between the center of the cleared circle
T1
and the center of the second dip. To show that overlap can be optimal, we must find
a ∆t < ∆t ∆ such that
p R 2 − overlap(∆t ) p R2
> 2 (C.2)
pr 2 (T1 + ∆t ) pr (T1 + ∆t ∆ )
In the next section we present two such examples.
• ρ1 = 3.5 nm
The detection probability on the first dip is 0.25. Substituting into Equation (C.1)
yields ∆t ∆ = 0.636 hours. The largest detection probability from a disjoint dip is
0.093 (substitute into the right-hand side of (C.2)). Numerically optimizing the left-
hand side of (C.2) yields ∆t * =0.838∆t ∆ = 0.533 hours. This generates an overlap of
41
overlap(∆t * ) =
3.25 nm2 and a detection probability of 0.097. The optimal overlap dip
occurs over 15% earlier than the disjoint dip and produces a detection probability
4% greater.
In the previous example, the differences between the disjoint dip and the optimal
overlap dip are not trivial. However, the parameter values are not realistic as the
helicopter will travel much faster than 11 kts. Below is a more realistic example
where the disjoint dip is suboptimal
• V = 82 kts
• U = 13.8 kts
• R = 3nm
• T1 = 0.45 hour
• ρ1 = 3.1 nm
The optimal time of the next dip ∆t * =0.977∆t ∆ = 0.071 is close to the time of the next
disjoint dip and the optimal detection probability (0.1732) is only slightly better
than the disjoint detection probability (0.1728).
The disjoint dip is not optimal when the helicopter arrives quickly to the datum. The
difference between the disjoint dip and optimal dip is larger for slower helicopters.
Thus, for realistic scenarios where the helicopter is much faster than the submarine,
disjoint dips should perform near optimally.