Improved Data Structures, Multiple Targets/Target Types and Multiple Sensors
in Discrimination Based Sensor Management
Troy Jenison
Master of Science, July 1996
ABSTRACT
INTRODUCTION
There are many applications where sensor resources need to be managed. One
of the primary applications is in the area of tactical fighter aircraft. In
this application effective sensor management reduces the pilot workload by controlling
his sensors and allocating limited sensor resources to effectively detect, classify
and track targets.
To better understand the importance of classifying targets consider typical
target classes. An abbreviated list of target classes might include friendly
fighters, bombers, and cargo aircraft; threat fighters, bombers and cargo aircraft;
and civilian airliners. A fighter must avoid engagement of friendly or civilian
aircraft, and its engagement strategy and weapons selections may vary depending
on the type of enemy aircraft.
Sensors have become increasingly important to the success of tactical air
missions. Once a pilot leaves the ground, he is almost entirely reliant on the
sensor suite for assessment of the environment. Although intelligence briefings
are provided to pilots prior to missions, this information may be inaccurate
because of the dynamic nature of the environment. Also, it is usually insufficiently
detailed. Many of the weapons on modern aircraft are capable of engaging targets
from beyond visual range. In order to take advantage of this capability, these
weapons must be deployed based on sensor measurements only. For these reasons
the sensor suite and its management play a crucial role in the effectiveness
of modern aircraft.
Many factors have contributed to the need for a sensor management system on
modern aircraft. These include, increased number and types of sensors in the
sensor suite, increased agility of sensors, increased pilot work-load, and improvements
in the tactics of threat forces.
Increased pilot work-load
Due to the high costs associated with training aircrew members, the trend has
been to reduce the size of the crew thus placing larger demands on them. This
is especially true of fighters which are often single-seat aircraft. This places
extraordinary demands on the fighter pilot because he must handle communications,
navigation, weapon systems, increased number of sensors and more complex tactics.
In addition, airframe performance has increased substantially making the whole
environment more dynamic.
The additional sensing resources have increased the amount of data collected
and the number of decisions to be made by the sensor operator. The amount of
data collected, along with an increased numbers of choices available to the
operator, make it likely that the operator will miss tactical opportunities
for become overloaded. With additional sensing resources, a pilot's success
can be improved if a sensor management system is employed.
Increased Agility/Capabilities of Sensors
One of the most significant advances in sensing capability is the Electronically
Scanned Antenna (ESA) radar system. Prior to the development of ESA, radar antennae
were mechanically gimbaled. Due to the inertia involved in moving the gimbaled
system of a Mechanically Scanned Antenna (MSA) radar system, changing the position
of the antenna is slow which makes only simple search patterns feasible. On
the other hand an ESA system can be repositioned in milliseconds, with little
of no penalty for large swings in the antenna angle, thus allowing for more
complicated search patterns.
Without the mechanical burden of an MSA system, an ESA system is able to perform
operations that are impractical with an MSA system. For example, the ESA system
can easily be backscanned, so that if an initial look at a sector where a target
is expected does not produce a detection, that sector can easily and quickly
be scanned again.
There are several newer sensors, such as infrared and optical devices, which
have been added to the sensor suite of aircraft over the years, that do not
have the agility of an ESA radar system since they are typically mechanically
gimbaled. However, because they typically have a narrow field-of-view and long
pointing time, their performance can be improved through appropriate scheduling
and management. Another sensor with an agile "aperture" is an Electronic Support
Measures (ESM) receiver. The ESM receiver is designed to detect electromagnetic
radiation from other aircraft. Due to hardware limitations of a practical ESM
receiver, only a limited band of frequencies can be monitored at a time. Thus,
selection of the frequency band can be treated as an "aperture" to be managed.
Most of these sensors also have the capability of operating in multiple modes.
For example, a radar system is usually capable of using different transmit waveforms
and pulse repetition rates, each of which has particular advantages in determining
range, range rate, or azimuth attributes of a target. Also, radar systems have
variable power levels, making it possible to alternate a high power signal to
maximize detections and a low power signal to minimize detection by enemy ESM
receivers. These additional options further complicate the task operating sensor
suite thus emphasizing the need for a sensor management system.
Improvements in Aircraft/Tactics of Threat Forces
In the cat and mouse game of fighter aircraft technology and tactics, gains
made in our aircraft can be expected to be met by improvements, counter measures
and tactics or similar advances from threat forces. Thus, a pilot can expect
to face an enemy with many , if not more, of the sensors and weapons systems
mentioned earlier.
The sensors can be divided into two classes. Active and passive sensors. The
distinction between these classes is that active sensors emit energy whereas
purely passive sensors do not. Since enemy forces will have ESM receivers, it
is important to use active sensors cautiously. Some sensor have so-called low
probability of intercept (LPI) modes. Some examples of the LPI techniques
include limiting the spatial regions where active sensors are used or using
low power or spread-spectrum waveforms to lower radiated power levels. Actual
sensor management system should support LPI capabilities. The LPI level of the
sensor manager would likely be selected by the pilot and the responsibility
of such things as determining what LPI modes the sensors should be used and
how to select the active sensors in the various spatial sectors. With the additional
restrictions imposed by the need to operate in an LPI mode, the effective use
of sensors becomes even more critical.
Previous Work
There are two categories in which the work on sensor management can be classified:
normative techniques, and descriptive or knowledge-based techniques. The normative
approaches generally use some a priori data about the environment and then use
a mathematical foundation or other formal decision making criterion to make
decisions. The descriptive techniques attempt to mimic the decisions that a
human would make in a similar situation by a rule-based approach [Popoli].
Much of the previous work in sensor management falls into the category of
descriptive techniques. Many of these approaches use an ad hoc technique but
there has also been the used of fuzzy set theory and fuzzy reasoning [Popoli].
Several normative techniques have been investigated or proposed for use in
sensor management. These include the use of Probabilistic Reasoning, Utility
Theory, Evidential Reasoning, Genetic Algorithms, Dynamic Programming, Optimal
Control Theory, and Fuzzy Logic and Control. These various techniques are evaluated
in [Fung, Horowitz], [Popoli] and [Llinas].
An Information Theory approach is also in the normative techniques category
and this forms the basis for the work in this paper. [Schmaedeke] primarily
covers an Information Theory approach to the tracking problem. This paper is
an extension of the work in [Kastella], which addresses the detection/classification
problem of sensor management.
A heuristic for sensor management based on discrimination directed search
is presented in [Kastella]. This is an example of the general problem of sensor
management which is to determine how to search a surveillance volume, which
may have multiple targets of multiple classes, with agile sensors to determine
with as few observations as possible where the targets are located as well as
which target class they belong to. [Kastella] also presents a monte carlo analysis
of this method applied to the detection problem, and compares it to a direct
search. For the small problem examined in [Kastella] (100 cells), a roughly
6 dB gain in using the discrimination directed search is observed.
This new paper extends these results in four significant ways.
(i) First, the data structures to implement the algorithm in the earlier work
were quite inefficient. As a result, only small problems could be examined in
monte carlo studies. This is improved in this work, allowing much larger problems
to be studied, and the relative performance of the algorithm as a function of
surveillance volume is more efficient. this is an important issue in the tactical
air application due to the highly dynamic nature of the problem it requires
a fast algorithm to keep pace with the environment.
(ii) Although the formalism of [Kastella] incorporated multiple targets and
target types, the monte carlo study was limited to detection of a single target
of a known type. In this paper we examine multiple targets and multiple target
types.
(iii) Additionally, a scheme for employing multiple sensors/sensor modes is
introduced and studied. Of particular interest is the case where there are two
sensors, one which is best at detection and another most useful for target classification.
(iv) Finally, we analyze the sensor dynamics generated by this sensor management
scheme. by this we mean how does the probability that a sensor is used against
a surveillance cell vary with time when each sensor dwell is selected to maximize
the expected discrimination gain. For example, an ideal sensor manager might
initially favor sensor dwells against target- containing cells. Once these cells
have been well-characterized, the sensor manager can then spend time confirming
that there are no targets in the empty cells.
The rest of this paper is organized as follows: Section 2 presents a method
of extending the results of [Kastella] to utilize multiple sensors and multiple
targets. Section 3 shows how a heap-based data structures is used to improve
the efficiency of the algorithm and thus enable larger problems to be studied.
Section 4 discusses the results that were obtained from the previous sections.
Section 5 narrates the conclusions and further work.
Research supported by the Minnesota Center for Industrial
Mathematics (MCIM)
