ArticlePDF Available

An Overview of Cognitive Radar: Past, Present, and Future

Authors:

Abstract and Figures

Modern radar systems face numerous challenges due to requirements of robust, high performance across multiple missions and multiple functions in the face of dynamically changing environments. Cognitive radar has emerged over the past 15 years through synergy of cybernetics, waveform diversity, and knowledge-aided signal processing as a new vision for the future of radar systems that can address these challenges. This article provides an overview of the historical context and mechanisms for cognitive processing in engineering systems, which have driven an evolution in the definition, analysis, and design of cognitive radar. A survey of cognitive radar research trends is given to provide insight on applications and techniques, while technical and practical challenges to future progress are discussed.
Content may be subject to copyright.
An Overview of Cognitive Radar: Past, Present, and Future
S.Z. Gurbuz, H.D. Griffiths, A. Charlish, M. Rangaswamy, M.S. Greco and K. Bell
Abstract—Modern radar systems face numerous chal-
lenges due to requirements of robust, high performance
across multiple missions and multiple functions in the face
of dynamically changing environments. Cognitive radar
has emerged over the past 15 years through synergy of cy-
bernetics, waveform diversity, and knowledge-aided signal
processing as a new vision for the future of radar systems
that can address these challenges. This article provides
an overview of the historical context and mechanisms for
cognitive processing in engineering systems, which have
driven an evolution in the definition, analysis, and design
of cognitive radar. A survey of cognitive radar research
trends is given to provide insight on applications and
techniques, while technical and practical challenges to
future progress are discussed.
I. INTRODUCTION
Since the dawn of humanity, nature has inspired
creative endeavors in all facets of human intellect. In
architecture and engineering, biomimetic design has been
transformative. For hundreds of years, scientists stud-
ied birds to unlock the mysteries of flight. Artificial
neural networks have been modeled from theories and
observations on the function and structure of the neural
synapses in the brain. Additionally, nature’s masters
of echolocation - bats and dolphins - can detect and
track very small prey using sophisticated waveforms,
which are varied dynamically through the encounter
with the prey [1]. Indeed, it may well be argued that
general artificial intelligence can not just match, but
surpass, human intelligence, representing the holy grail
of biomimetics.
The integration of some form of machine learning
into engineering systems has often been referred to
using terms such as“smart” or “intelligent.” Yet, the
description of what exactly makes a sensor “smart”
is somewhat nebulous, ranging from sensors that (at
a minimum) are networked and can communicate in-
formation so as to improve operational efficiency, to
sensors that have an on-board micro-processor for some
form of embedded processing for optimal control of
the measurement process [2], and to networked sensors
that exhibit “distributed intelligence” with capabilities
of self-monitoring, making decisions to automatically
compensate for changes in their surroundings [3], [4].
Adaptive radars have the capability of changing the
processing of received data as a function of time, while
fully adaptive radars additional have the capability to
adapt on transmit. As a result, “fully adaptive” has
often been used synonymously with “cognitive” radar,
although over the past 10 years, the term “cognitive” has
become increasingly popular, perhaps in part because it
evokes a vision of biomimetic artificial intelligence fully
integrated into the sensing process, emulating human
perception.
Formally, cognition is defined in the Oxford dictionary
as ”the mental action or process of acquiring knowledge
and understanding through thought, experience, and the
senses.” Although there are many functions of the brain
that enable human cognition, the cognitive neuroscientist
Dr. Joaquin Fuster [5] has posited that there are five
essential processes: 1) the perception-action cycle (PAC),
2) attention, 3) memory, 4) language, and 5) intelligence.
The PAC is a circular flow of information - a feedback
loop - from the environment to the senses (perception),
and through motor structures back to the environment
(action). Embodied in this cycle is the key idea that
cognition is an interactive process, where the cognitive
entity must respond or change its behaviour in some
fashion as a result of external stimuli.
In traditional fore-active radar systems, the informa-
tion flow is one-way: the radar interrogates its sur-
roundings by transmitting a fixed, pre-defined waveform
regardless of any changes in the environment. Adaptive
processing may be performed on receive, but results from
such processing do not translate into the control of any
radar function on transmit (e.g. there is no ”action” only
”perception”). Current research on cognitive radar aims
not only at developing the adaptive hardware and analyt-
ical techniques necessary to enable two-way interaction
of the radar with its environment for performance opti-
mization, but also on leveraging advances in fields such
as stochastic control, optimization, machine learning,
and artificial intelligence (AI) to develop engineering
analogues to a wide range of cognitive processes.
This paper provides an overview of the evolution of
cognitive radar, focusing on the crystallization of certain
ideas that have led to a formal, technical definition of
cognitive radar in the IEEE Standards. A survey of
cognitive research trends over the past decade is provided
to give insight on the techniques being developed for
a wide range of radar applications. Finally, technical
challenges to progress in cognitive radar design are
discussed as motivation for future work in this field.
II. HISTORICAL CONTEXT
A. Early Pioneers
The foundations of cognition in an engineering context
date back to Norbert Wiener and his work in stochastic
processes, communications and control in the 1930s.
Observing that ”the nervous system and the automatic
machine are fundamentally alike in that they are devices,
which make decisions on the basis of decisions they
made in the past,” [6] Wiener first coined the term
cybernetics in 1948 as ”the scientific study of control
and communication in the animal and the machine” [7].
Whereas artificial intelligence strives for computers to
intelligently understand the world as an end goal in and
of itself, cybernetics exploits this understanding to gain
necessary feedback to achieve specific goals. In this way,
it embodies a fusion of ideas from Shannon’s information
theory, originally designed to optimize the transmission
of information through communication channels, sys-
tems theory and control, and biomimetics.
As the seeds of intelligent computing were being
sown, the groundwork for modern methods in statistical
signal processing and sequential detection was being
laid [8]. Significant advances into the optimal design
of sequences of experiments, hypothesis testing and
parameter estimation were made [9], [10]. Central to
these works was the idea that the data collection process
itself should be a closed-loop process, where a decision
on how to collect subsequent samples is determined
based on analysis of prior samples. Early examples of
such closed-loop data collections include that of Meier,
et. al. (1967) [11], where dynamic state is treated as an
adaptive measurement problem, and Athans (1972) [12],
where an optimal closed-loop selection of measurements
is determined in a Kalman filtering problem.
B. Sensor and Radar Management
A broader, more comprehensive expression of this
optimal experiment design problem is embodied by the
term ”sensor management,” which first came into use in
the late 1960s. The sensor management paradigm [13]
represented the advent of a new generation of sensors,
enabled by advancements in sensor and communication
technologies in the 1990s, where previously fixed sen-
sor operating parameters could now be adapted during
the data collection process using software commands.
Existing mathematical constructs for control of decision
processes, such as Markov decision processes and multi-
armed bandit decision processes, became important fa-
cilitators for the development of many approaches for
sensor management that remain influential today.
As active sensors, radars have experienced similar,
parallel developments, and are particularly well suited
for integration of cognition as they possess multiple
degrees of freedom via waveform agility and electroni-
cally steered antenna arrays. An early, fundamental radar
signal processing algorithm that embodies principles of
cognition is the least-mean-squares (LMS) algorithm
pioneered by Widrow, et.al in 1967 [14]. This approach
enables an antenna array to adaptively form a main lobe,
with its direction and bandwidth determined by a control
signal, as well as place nulls so as to reject any unwanted
signals or noise outside the main lobe, such that the
mean-square error is minimized.
Ideally, radar resource management can be best ac-
complished by optimal decision making and control
of degrees of freedom (transmitter, receiver, antenna,
and power) to maximize performance of multiple radar
functions (e.g. scheduling, detection, tracking, and clas-
sification) - an inherently cognitive process. In the early
1990s, a number of benchmark problems were issued
to enable comparison of techniques that used the beam
steering capabilities of phased array antennas to opti-
mize tracking in the presence of electronic countermea-
sures, while minimizing false alarms [15], [16]. This
has generated a rich body of literature, with solutions
involving a combination of multiple hypothesis testing
and multiple interacting filters for tracking (e.g., [17],
[18]) to optimize performance measures such as signal-
to-interference-plus-noise ratio (SINR), track sharpness,
and detection threshold.
Increased adaptivity was a common feature of pro-
posed solutions, which involved adaptive revisit time
scheduling, adaptive selection of detection thresholds
(e.g. constant false alarm rate (CFAR) detectors) [19],
and adaptive clutter suppression with space-time adaptive
processing (STAP) [20] for improved target detection.
Adaptive tracking techniques have varied the measure-
ment times as well as signals used for track updates,
based on measurements acquired by a tracker. This
feedback loop is used to control the radar such that
frequent measurements are made during unpredictable,
or rapid dynamic maneuvers, while infrequent measure-
ments are made during predictable periods or steady
dynamics. Experimental radar systems, such as the U.K.
Multi-function Electronically Scanned Adaptive Radar
(MESAR) [21], U.S. Advanced Multifunction RF Sys-
tem (AMRFS), and Royal Canadian Navy (RCN) Active
Phased Array Radar (APAR), among others, have been
used as a platform for demonstrating the new ideas
being developed. Indeed, as early as 1990, researchers
proposed future concepts for an intelligent radar that
could learn from its environment [22], fusing artificial
intelligence with prior knowledge to achieve improved
optimization and data-dependent processing for resource
management and remote sensing [23].
C. Enabling Paradigms
Simultaneously, the concept of altering the intra-pulse
waveform modulation based on the measurements pro-
vided by the tracker was also beginning to be explored.
This led to the development of methods for optimal
waveform selection (e.g. [24], [25]) and adaptive ex-
tensions thereof (e.g. [26], [27]). The term waveform
diversity [28], first introduced in 2002 by Dr. Michael
Wicks [29], has become a focal point for research into
cognitive radar and is defined in the IEEE Standard 686
as the “optimization (possibly in a dynamically adaptive
manner) of the radar waveform to maximize performance
according to particular scenarios and tasks” including
exploitation of multiple domains, such as “the antenna
radiation pattern (both on transmit and receive), time
domain, frequency domain, coding domain and polar-
ization domain.” Examples include waveform selection
from among multiple waveform classes, e.g. linear or
non-linear frequency modulation (LFM/NLFM), phase
or frequency coding, and ultrawide band waveforms.
It could also include adapting the parameters within a
waveform class, such as changing the pulse repetition
interval (PRI), bandwidth, or center frequency [30]. In
[31], Guerci showed that the optimal waveform which
maximizes the SINR arises as the solution to a general-
ized eigenvalue problem, while methods for imposing
constraints on the waveform to make it suitable for
practice are also considered.
Investigations into optimal radar waveforms date back
to the inception of waveform design in the 1930s-1940s,
and include important milestones on LFM waveform
design [32], the design of optimal coded waveforms to
reduce sidelobes, mismatch loss, mainlobe broadening
[33], [34], [35], improve pulse compression [36], and
ambiguity function design for improved detection [37]
or target matched illumination [38]. In 1953, Dr. Philip
Woodward published a seminal work [39] in which he
introduced information theory in the context of radar
detection, stating “the problem of reception is to gain in-
formation from a mixture of signal and unwanted noise,
much of the literature has been concerned with “methods
of obtaining as large a signal to-noise ratio as possible
on the grounds that noise ultimately limits sensitivity and
the less there is of it the better. [While] valid ... [this]
can be misleading, for there is no general theorem that
maximum output signal-to-noise ratio insures maximum
gain of information.” Indeed, his book ends lamenting
that ”the basic question of what to transmit remains
substantially unanswered.
Although this remains an open question, Woodward’s
work laid the foundation for subsequent information-
theoretic waveform design approaches, including that of
M. Bell [40], which proposed maximization of the mu-
tual information between a random extended target and
the received signal for optimal information extraction
[41]. Dr. David Middleton’s 1959 work on statistical
communication theory [42] took a giant leap forward by
developing a framework for “joint optimization of trans-
mission and reception by choice of signal waveform.
The approach developed was rooted in Bayesian decision
theory and provided for system optimization through
choice of waveform at the transmitter and minimization
of a cost function, which provides for a value judgement
of “error” and thus guides the decisions. Middleton’s
work thus provides a concrete analytical framework for
implementation of the perception action cycle in a radar
transceiver, which has served as groundwork for future
milestones in cognitive radar research [43], [44]. How-
ever, Middleton also notes the difficulty in selection of
optimality criteria and the assignment of costs accurately
in an objective fashion. The cost may not be unique and
reflects the “unavoidable uncertainty...that is the price
we must pay for an inevitably incomplete knowledge
of the world around us.” In this remark, recognition that
the a priori information gained from previous experience
is highly likely to be inadequate is plain. Nevertheless,
concludes Middleton, “the more uncertain our a priori
data, the greater the expected cost of operation – we
cannot avoid paying for ignorance.
Thus, complementary and equally essential as wave-
form diversity is the knowledge-aided signal processsing
paradigm, which, to put it simply, aims to exploit prior
knowledge to improve sensor performance. Originally
proposed in the context of STAP to improve adaptive
suppression of clutter in non-homogeneous clutter en-
vironments [45], prior knowledge of the interference
environment was proposed for intelligent training and
filter selection, as well as data pre-whitening [46]. Prior
knowledge in this example could take the form of a
terrain map or even images from other sensors, such
as hyperspectral imagers [47]. In 2002, the Defense
Advanced Research Projects Agency (DARPA) initiated
the Knowledge-Aided Sensor Signal Processing and Ex-
pert Reasoning (KASSPER) Program to more broadly
address challenge of minimizing sensor deficiencies
Fig. 1: Bloom’s Taxonomy
through exploitation of prior knowledge. Since then,
this concept has been applied to numerous other radar
problems, such as 2D autofocus for spotlight SAR [48],
tracking [49], ground moving target indication (GMTI)
[50], and radar identification [51].
Perhaps unsuprisingly, both waveform diversity and
knowledge-aided signal processing have analogs in the
natural world. Bats have been reported to use many
different waveforms (e.g. constant frequency, linear and
hyperbolic frequency modulation, multiple harmonics
and even other types of non-linear frequency modulation)
in pursuit of different activities, such as searching for
prey, social calls for communications with other bats, and
hunting – definitely a wonderful example of waveform
diversity in a multi-functional active sensing system!
Similarly, as humans, we are all aware of our own
capacity to learn and thereby adapt our behaviour as
a result. Children will not touch a hot cup of coffee
more than once, quickly learning that touching hot things
hurt. We consult books, our friends, and remember past
experiences when guiding ourselves to future decisions
- and we would view doing so as ”smart.
D. Vision for the Future
The crystallization of cognitive radar as a formal
concept for next-generation radar reflects a conscious
evolution in design that incorporates more and more
features of human cognitive capabilities into the radar
architecture to achieve increased autonomy and per-
formance optimization in dynamically changing envi-
ronments. It thus provides a vision for building upon
the designs of existing radar systems, some of which
may now in retrospect be recognized as having some
cognitive characteristics. One hierarchy of human cog-
nitive capabilities is given by Bloom’s taxonomy, which
originates from cognitive psychology and is shown in
Figure 1. It may be argued that existing approaches
map to only the lowest cognitive levels. While databases
or prior measurements comprise ”remembered” prior
knowledge, most signal processing and machine learning
algorithms represent methods used for ”understanding.
Knowledge-aided signal processing represents a higher
level of cognition than adaptive processing as it does,
to a certain extent, permit use of information in new
situations.
An early expression of such a vision was given in
2003 with the introduction of the Sensors as Robots [52]
concept: ”As more knowledgeable and proven techniques
are obtained, radar systems will begin to function as
robots...the final step will be autonomous operation of
these sensors under the intelligent robot paradigm.” [53]
Whereas traditionally sensors have been a means for
providing informational inputs to robots, this concept
flipped the equation: now, sensors themselves would be
autonomous, intelligent agents, ”figuring out” how best
to go about their tasks.
The term ”cognitive radar” itself was first coined by
Dr. Simon Haykin that same year [54], which drew
heavily on ideas developed by Fuster in cognitive neu-
roscience. Haykin’s work built upon past work in cyber-
netics, artificial neural networks, self-organized learning,
and Bayesian decision theory to propose engineering
analogues for implementation of four of the main cog-
nitive features identified by Fuster: the PAC, memory,
attention and intelligence [55]. The parallel’s between
the PAC envisioned by Fuster and that describing the
operation of a radar transceiver for remote sensing may
be observed from Figure 2. Both exhibit the common
feature of providing for closed-loop feedback in the radar
transceiver through interaction with the environment.
Through the perceptual hierarchy, dynamic changes in
the environment may be analyzed “bottom-up ... and
lead to the processing of further actions, top-down
through the executive hierarchy, toward motor effectors.
[56] Note that there may be interaction and feedback
along different levels of the perceptive and executive
hierarchies. Hierarchy is reflected also in the cybernetics
research of Jens Rasmussen [57], [58] in the 1980s,
in which human behavior was described in terms of
three levels: skill-based, rule-based, and knowledge-
based. Skill-based behavior described sub-conscious yet
efficient PACs, which, according to Bruggenwirth [59],
maps to basic signal processing and generation units
in a radar system. Rule-based behavior is applied by
humans in familiar situations, and although consciously
controlled, the action is reactive and thus results in
procedures that have been learned over time. In radar,
the parallel operation would be the procedures that
have been pre-stored or hard-coded, based upon offline
simulations and analysis of prior experience. The highest
layer is the knowledge-based layer, in which solutions to
Fig. 2: Perception Action Cycle for Radar Remote Sensing
problems that arise in unfamiliar situations are derived
using knowledge-based deliberation. In cognitive radar,
a similar level of behavior would be implemented by
search or inference algorithms that utilize all available
knowledge derived from sensors, mission objectives, and
memory.
E. Current Programatic Thrusts
Haykin’s work thus marked the beginning of concerted
efforts into formulating and defining exactly what the
architecture and characteristics of future cognitive radar
systems would be like. The investigation into tangible
implementations was in part spurred by the challenges
imposed by an increasingly congested RF spectrum,
while adaptivity on transmit and controlled illumination
were being enabled by advances in electronics, embed-
ded computing, adaptable RF components (amplifiers,
filters), small, low-cost, low-power RF transceivers and
software-defined radio platforms.
In the early 2000s, two programs initiated by the
U.S. Air Force Office of Scientific Research (AFOSR)
and DARPA stimulated research that would serve as
important precursors to cognitive radar: namely, the
Multi-disciplinary University Research Initiative (MURI)
”Waveform Diversity for Full Spectral Dominance” Pro-
gram and ”Waveform Agile Sensing and Processing”
Program. The aim of these programs were to devise
methods for optimization of radar performance under
time-varying environmental conditions, including a capa-
bility to respond to unknown dynamic target parameters
through waveform agility. Together, these two programs
advanced the requisite mathematical foundations, incor-
porating the resulting theories into a systems design
perspective.
Subsequently, AFOSR would take one step further in
formulating the theories required for PAC implementa-
tion by initiating the Dynamic Data Driven Application
Systems (DDDAS) program, defining the DDDAS con-
cept as ”the ability to dynamically incorporate additional
data into an executing application, and in reverse, the
ability of an application to dynamically steer the mea-
surement (instrumentation and control) components of
the application system.” [60] Efforts are focused on 4
specific science and technology frontiers: 1) applications
modeling, 2) advances in mathematical and statistical al-
gorithms, 3) application measurement systems and meth-
ods, and 4) software infrastructures and other systems
software. This is complemented by the U.S. Air Force
Research Laboratory’s program in Fully-Adaptive Radar,
led by Dr. Muralidhar Rangaswamy, which aims to close
the loop on the radar operation at multiple levels in
an attempt to bring to bear the sense-learn-adapt (SLA)
paradigm to maximize system performance by making
adaptive and optimal use of all available degrees of free-
dom. Significant advances from this program include:
a) performance bounds for closed loop radar tracking
with controlled laboratory demonstration of this concept;
b) a powerful modeling and simulation capability for
generating training data for signal dependent interfer-
ence scenarios; c) signal processing algorithms for joint
adaptive radar processing on transmit and receive; d)
waveform design and optimization principles; e) con-
vex optimization for adaptive radar covariance matrix
estimation; f) ambiguity function analysis and Cramer-
Rao bounds for distributed passive radar (which enable
sensor geometry placement and illuminator selection
for maximizing system performance); and, g) passive
radar detection involving noisy reference channels with
analytical performance guarantees. Most importantly, the
modeling and simulation capability developed under this
program has transitioned to a program under support
from the Office of Secretary of Defense.
This intense research activity has resulted in a dra-
matic increase in publications over the past few years,
as shown by Figure 3, which surveys publications in the
IEEExplore and SPIE Digital Libraries that reference
cognitive or fully adaptive radar in their title or text.
Much of this work is theoretical in nature and substanti-
ated with simulation results; however, there are singular
works (notably at Ohio State University [61], FFI [62]
and ArmaSuisse [63]) that experimentally validate per-
formance gains due to cognitive design. Cognitive radars
face unique challenges to requirements specification and
validation, as discussed in more detail in Section V. As
enabling technologies mature, continued research will
push forward the design of next-generation radar systems
with ever increasing cognitive capabilities.
III. DEFINITION AND CLASSIFICATION OF
COGNITIVE RADAR SYSTEMS
The increasingly common use of the relatively new
term “cognitive radar” has resulted in some debate as
to how a cognitive radar actually differs from other
terms used - such as intelligent or smart radar, fully-
adaptive radar (FAR), and cognitive FAR - if at all; and
what the technical requirements for describing a radar
as ”cognitive” are. Descriptions of cognitive radar that
have been proffered include:
“Cognitive radar (CR), which differs from tradi-
tional active radar as well as fore-active radar by
virtue of the following capability: the development
of rules of behavior in a self-organized manner
through a process called learning from experience
that results from continued interactions with the
environment.” (Haykin, et al. [55] in 2012)
A system that is capable of sensing, learning, and
adapting to complex situations with performance
approaching or exceeding that achievable by a sub-
ject matter expert.” (Guerci, et al. [64] in 2014)
“While a fully adaptive radar may employ feedback
and use prior knowledge stored in memory, a cog-
nitive radar predicts the consequences of actions,
performs explicit decision-making, learns from the
environment, and uses memory to store the learned
knowledge.” (Bell, et al. [44] in 2015)
“Cognitive radar is a radar system that acquires
knowledge and understanding of its operating en-
vironment through online estimation, reasoning and
learning or from databases comprising context in-
formation. Cognitive radar then exploits this ac-
quired knowledge and understanding to enhance
information extraction, data processing and radar
management.” (Charlish, et al. [65] in 2017)
A cognitive radar system follows the four prin-
ciples of cognition: the perception-action cycle,
memory, attention, and intelligence.” (Farina et al.
[66] in 2017)
As part of the work of the NATO SET-227 Task Group
on Cognitive Radar, active between 2015 and 2019,
and of which all the authors are a member, numerous
discussions on the characteristics of cognitive radar were
conducted, which highlighted the need for and insights
on what a definition might look like. In 2017, Dr. Chris
Baker and Dr. Hugh Griffiths spearheaded efforts to
include a formal definition of cognitive radar in the IEEE
Standard Radar Definitions 686 [67]: “A radar system
that in some sense displays intelligence, adapting its
operation and its processing in response to a changing
environment and target scene. In comparison to adaptive
radar, cognitive radar learns to adapt operating param-
eters as well as processing parameters and may do so
over extended time periods.
From these definitions, it may be seen that there is
general agreement in the cognitive radar research com-
munity concerning some common elements that must
be present in a radar system for it to be classified as
cognitive. These are the same characteristics identified
by Fuster [5] within the context of cognitive psychology:
the PAC, attention and language. Firstly, the PAC, is the
framework which provides for closed-loop feedback in
the radar transceiver. If this function were not to exist,
then the ability to adapt based on the system’s perception
of the environment would be absent. Secondly, attention
enables the radar to focus its resources on critical aspects
of the observed scene. This is a characteristic that all
multi-functional radars must possess as a resource man-
agement requirement. Lastly, language may be viewed
as the ability to encode data such that it is possible for
the system to store, recall and disseminate information
both internally and externally, but more broadly can
embody any set of rules for communicating information,
including internal messaging as part of decision making.
As the storage and recall of information is essential
to learning and decision making, without it the system
would arguable be hindered from being responsive to
any perception of outside stimuli.
The remaining two elements of Fuster’s framework
for cognition includes memory and intelligence, and are
aspects regarding which designs begin to diverge. Here,
memory alludes not just to physical storage devices,
which could hold prior knowledge, but to memories of
Fig. 3: Publications on cognitive radar (2003 - March 2019).
learned experiences gained during the course of extended
periods of observations. Thus, memory can be said to
function at varying levels: 1) as a fixed internal knowl-
edge base, 2) as a dynamic knowledge base updated by
an external source, and 3) as an on-line learning capable
system. Similarly, intelligence may be characterized into
varying degrees based upon 1) the complexity of the
decision-making mechanism, and 2) capacity to plan
long-term behavior. Conceivably, a radar operating at
a high level of cognition would even be anticipative,
planning based on its predictions of future outcomes.
Thus, an overarching classification scheme that would
allow flexibility in the definition and identification of
cognitive characteristics as recently been proposed [68],
which relies on grading systems based on the degree
of planning sophistication (P), decision mechanism so-
phistication (D), and memory sophistication (M). This
taxonomy is depicted by the 3D Synthetic Classification
Space, shown in Figure 4, and provides a means for
acknowledging the cognitive aspects of existing radar
systems, while providing a scale to identify the ways
in which next-generation cognitive radar systems have
advanced. While [68] also provides for a numerical
scale to match this framework, it is not so much the
numbers used, but the recognition that systems ought
not be judged in a binary, black and white fashion, that
is significant.
IV. TECHNIQUES AND APPLICATIONS
Over the past 15 years, research into cognitive radar
design has spanned a wide range of applications, us-
ing many different techniques that draw on prior ad-
vancements in Bayesian decision theory, information
theory, decision theoretic approaches (including fuzzy
logic, rule-based systems, metaheuristic algorithms, and
Markov decision processes), dynamic programming, op-
timization (e.g. maximization of signal-to-noise ratio
(SNR), convex optimization, and minimization of the
Fig. 4: 3D Synthetic Cognitive Space [68]
Cramer-Rao Lower Bound (CRLB)), and game theory.
Figure 5 shows a histogram of applications and tech-
niques based on the 83 journal papers and 238 confer-
ence papers surveyed in Figure 3. A complete listing of
this literature may be found in the Final Report of the
NATO SET-227 Task Group on Cognitive Radar. This
histogram reveals that while many applications are being
considered, a few have been of great interest; namely,
concepts for transceiver architecture and mechanisms for
cognitive processes (ARCH), radar resource management
(RMG), target detection (DET), localization/direction-
of-arrival estimation (LOC) and tracking (TRK), radar
networks (RN), and spectrum sharing (SS). Most works
involve some form of waveform selection, optimiza-
tion and design (WD), while adaptive control of an-
tenna beam pattern, design of adaptive RF components
(ADPTHARD) as well as experimental testing (EXP)
have also been explored.
Spectrum sharing has been a topic of focus due to the
urgent challenges presented by a congested RF spectrum
to military and civilian systems alike. The availability of
Fig. 5: Techniques and applications investigated in cognitive radar publications (2003 - March 2019).
frequency spectrum for multi-function radar systems is
being continuously diminished. The growth of activities
in civil communications and the emergence of new
technologies and services that have a great demand for
spectrum allocation induce a very strong pressure upon
the frequency channels currently allocated to radars.
In the VHF (30-300 MHz) and UHF (300-1000MHz)
bands, where for instance foliage penetrating (FOPEN)
radars are active, interference can come from broadcast
and TV services. Recently, these bands have seen the
introduction of the IEEE802.11ah and IEEE802.11af
protocols for Internet of Things (IoT) and Cognitive
Radio Technology, respectively. In the U.S., the Na-
tional Telecommunications and Information Administra-
tion (NTIA) has devoted efforts on identifying frequency
bands that could be made available for wireless broad-
band service provisioning, resulting in allocation of 115
MHz of additional spectrum (1695-1710 MHz and 3550-
3650 MHz bands) and a conflict with L-band (1-2 GHz)
radars. An example is the air route surveillance radar
used by the Federal Aviation Administration (FAA) that
shares the spectral band with wireless inter-operability
microwave access (WiMAX) devices. The majority of
the LTE services, e.g. WiMAX LTE, LTE global system
for mobile (GPS) are operative in the S-band (2-4 GHz),
where they interfere with surveillance radars. In C-band,
the spectrum has been eroded by allocation of the 5 GHz
band to 802.11a/ac wireless LAN technology. X-band is
still free from communication services interference, but
when 5G systems become fully operative, even the Ka,
V and W bands will be dense.
Thus, in a near future, radars will likely be required
to share their bandwidth with communication systems,
where the latter ones, quite often, are the primary users.
Yet, this problem cannot be addressed only by traditional
modes of operation, such as antenna beamforming or
interference cancellation on receive. Future systems re-
quire the ability to anticipate the behavior of radiators in
the operational environment and to adapt its transmission
in a cognitive fashion based upon spectrum availability.
Radar cognition in this case is based on two main con-
cepts: spectrum sensing and spectrum sharing. Spectrum
sensing aims at recognizing frequencies used by other
systems occupying the same spectrum in real-time, while
spectrum sharing tries to limit interference from the radar
to other services and vice-versa.
Furthermore, battlespaces of the future will not involve
isolated geographical regions with limited technological
resources, but will require seamless integration of net-
worked ground-based, airborne, and space-based sensors
at different levels, automated to find, identify and track
threats in increasingly complex and diverse environ-
ments. The challenge of spectrum congestion is but
one dimension of this broader battlespace. Technological
advancements have not just benefited modern society,
but have also made it easier for adversaries to make
their forces both mobile and elusive, such through use
of small drones to attack a diverse set of tactical targets,
previously not exposed to any threat. Both force pro-
tection and forward operations require pervasive, robust,
and agile sensing that can optimize multiple missions
in a dynamic environment. This operational requirement
directly maps to the definition of what a cognitive radar
strives to accomplish, and indeed, the generalized notion
of a cognitive sensor network, empowered with multiple-
layers of hierarchical cognitive processing.
V. CHALLENGES
While the potential of cognitive approaches to enhance
existing radar performance in almost all respects has led
to great progress, full achievement of this potential faces
several important challenges.
A. Research
Two key challenges to the research and development
(R&D) of cognitive radars are the development of assess-
ment and evaluation tools, as well as experimental testing
methodologies. A common terminology for describing
and comparing the characteristics of cognitive radar
is needed. Although the ontology in [68] provides a
graded framework for assessment, as algorithms and
architectures advance, this will need to be further re-
vised, detailed and adapted. Furthermore, evaluation of
cognitive radar algorithm performance requires quanti-
tative metrics. This is not just vital for analysing radar
performance offline, but is also the basis for forming
cost or reward functions on which online optimisation
is based. Although system performance will still be
measured in terms of standard performance metrics –
such as probability of target detection and false alarm,
mean square error in tracking systems, and probability
of correct classification in automatic target recognition
systems – cognitive systems require additional metrics
that quantify the gain in performance achieved at the
cost of using system resources.
Common approaches fall into the categories of
information-driven [69], task-driven or quality-of-service
(QoS). Information theoretic surrogates, such as mu-
tual information and Bayesian information, can be very
valuable in optimizing the waveform to increase the
amount of information gained. However, when allocating
resources, they tend to give a bias towards information
rich tasks, such as tracking high SNR targets, which may
not be the targets of most interest to the mission. Task-
driven methods optimise each task using a performance
measure that is specific to the task. This is effective
in optimization of individual tasks; however, multiple-
task performance measures, such as in a multi-function
radar system, is a particularly difficult task. To this end,
Mitchell, et.al [70] provides some strategies for develop-
ing cost functions for executive processor optimization
by combining performance and measurement metrics.
QoS methods [65] circumvent the problem of combining
task performance measures by combining and optimizing
utilities, which represent the mission-relevant satisfaction
that is associated with a task performance level, leading
to mission relevant resource allocation. This issue of
cost and reward function design remains more of an art
than a science, and will continue to be a major research
challenge.
A related, but unique, challenge to cognitive radar
design is experimental testing, since the transmit wave-
form and settings are adapted during operation. With
more sophisticated simulations, new development and
qualification processes, including software-in-the-loop
testing, can be developed and will contribute to cognitive
radar validation. Evaluation on pre-collected data sets
is no longer possible, except in limited cases where
the data can be oversampled in some manner and then
down-selected after the fact to emulate cognitive radar
selection of parameters. For example, in [71], the pulse-
Doppler software defined radar (SDR) collected data at
a high pulse repetition frequency (PRF). The cognitive
algorithm determined the number of pulses and required
PRF (up to the actual PRF) and then downsampled the
pulses to get the correct number of pulses at the desired
PRF. A similar process was used in [72].
Thus, as of early 2015, the performance of advanced
concepts for cognitive radar were only examined through
simulation, or in the best case, using pre-recorded data.
There had been no reports of experimentally validated
concepts, largely because the necessary hardware to test
them had not been developed. However, this step is vital
to establish the true performance potential of applying
cognitive processing methods. In the last few years,
cognitive radar testbeds have been developed at the Ohio
State University (OSU) [61], Armasuisse [63], and FFI
[62] and real-time experimental evaluations have been
reported in [61], [70], [62], [73]. Challenges in real-time
experimentation involve repeatability of experiments,
determining what is truth, determining metrics that can
be obtained from the data and used for optimization,
and timely computation. Robustness to modeling and
computational errors has been largely ignored in the
research to-date, but is a critical issue that has just begun
to be investigated [74].
B. Requirements Definition
Inherent to the radar procurement process is the spec-
ification of the required radar performance. The typical
approach is to define a number of worst cases and specify
the performance that the radar should always achieve
for these example worst cases. This approach is valid
for non-cognitive radar systems that do not reconfig-
ure based on the current environment, as the single
radar configuration that matches the worst acceptable
performance may be utilized. But, for a cognitive radar,
the worst case scenario is unlikely to be selected as a
solution, because this requirement specification does not
warrant the additional development cost of cognition.
Thus, using a limited set of worst case scenarios does
not make sense for cognitive radars, and an alternative
approach is required. A related issue is that for a
cognitive radar, the performance potentially depends on
the amount of learnt or context knowledge available.
To define a processing for requirements specification
that considers learning and context, criteria weighing
the importance of performance metrics, as well as quan-
tification of the trade-offs evaluated, will be needed to
compare candidate solutions.
C. Reliability of Past Knowledge and Learning from
Experiences
As the capabilities of radar transceivers advance to
jointly sense, learn, and adapt on both transmit and
receive, new opportunities and vulnerabilities will be-
come part of the changing dynamics of electronic war-
fare (EW). While boosting sensing capabilities so that
friendly systems can defend against jamming and other
counter-measures, and leaving adversaries no place left
to hide, cognitive radar nonetheless retains the risk that
it could be beguiled into poor decisions, much akin
to the human counterpart that has inspired its design.
Thus, a cognitive radar requires a means for evaluating
and ensuring the reliability of its knowledge sources:
both sources of past knowledge, provided through access
of databases, as well as knowledge learned through
operational experience. This includes considering not
only the possibility for deception, but also whether the
validity of data degrades over time. Thus, understanding
how to design a radar so that it can learn from past
mistakes caused by poor decisions, and thereby enable
it to gain the ability of making informed decisions in the
future, will be critical.
D. Laws and Regulations
Cognition in radar requires waveforms and circuits to
be reconfigurable and optimizable in real-time. However,
an often overlooked operational constraint is the national
and international laws governing operation. Naturally,
the transmissions of radars and other devices are all
regulated (e.g. ITU emission standard [75]). While radar
transmissions should not exceed the limits imposed by
regulation, unwanted emissions, due to non-linearity in
the transmitter and to the steep rise and fall times of the
radar pulses, often occur [76]. Especially in cognitive
systems, the dynamic reconfiguration of the tranmission
spectrum is not always easily implementable and may
result in out-of-band (OOB) transmissions. This is pri-
marily due to the non-linear operational regime of the
high-power radar RF circuitry (particularly for vacuum
tube amplifiers), which causes non-negligible spectral
spreading outside the assigned radar band. This makes
coexistence of communications and radar systems in
close bands with narrow guard bands difficult [76]. Mag-
netron tubes, quite often used in legacy radar systems
because they are inexpensive, have serious drawbacks
in term of spectral purity. To reduce OOB emissions,
bandpass filters are often used, though the cost of this
improvement in spectral purity means a significant loss
in the effective transmitted power. Solid-state-based am-
plifiers are much easier to control in terms of OOB,
but cannot provide the high peak power of tubes and
represent a minority of current operational systems.
An alternative short-term approach is to instead select
from a pre-defined set of waveforms or waveform param-
eters. Many modern radars already have this capability,
and a first step toward making cognitive radars a reality
could be choosing among the set of allowable wave-
forms [29]. Alternatively, cognitive algorithms could
also be implemented on passive systems [77]. Longer
term, not only are technical solutions required to ensure
cognitive radars can operate within regulatory bounds,
but also new concepts and perspectives towards legal
jurisprudence will be needed. The advent of artificial
intelligence, across all engineering disciplines, has raised
legal questions of responsibility of accountability of
engineering systems capable of autonomous or semi-
autonomous decision making. How artificial intelligence
will change jurisprudence on legal personhood and as-
signment of liability is a question that could drive the
outcome of ongoing debates regarding banning versus
regulating certain AI-based technologies, and indeed the
future of cognitive radar design.
E. Open Questions
In addition to the technical and practical challenges
described above, many open questions remain around
cognitive radar design and implementation. Related to
the issue of performance validation, for example, is the
question of whether end users will accept fielding a
sensor whose behavior is not exactly predictable. How
robust will cognitive systems be? Or will system decision
errors result in flamboyant failures far more severe than
sub-optimal performance metrics? Just as dealing with
people can be frustrating, in the long run will the
autonomy of radar systems truly lead to the benefits
as the designers intended? As advances in real-time
processing and adaptive hardware enable the physical
construction of reconfigurable systems, trends in artificial
intelligence will drive cognitive designs, necessitating
human adaptation to the new and unique challenges
posed.
VI. CONCLUSIONS
Cognitive radar is an emerging technology that has
been inspired by advancements in cybernetics, man-
machine interaction, waveform diversity, knowledge-
aided signal processing, and resource management. Al-
though the term “cognitive radar” has just been around
for about 15 years, it is perhaps best viewed, however,
not as something that has suddenly been developed,
but as the product of a steady evolution in design that
aspires towards achievement of cognition as seen in its
counterparts in nature, such as exemplified by the sensing
capabilities of bats and dolphins, or the intellectual
decision-making of humans. This article has strived to
provide a broad overview of recent progress and ideas
in cognitive radar design, including challenges that will
need to be addressed looking towards the future.
VII. ACKNOWLEDGEMENTS
The authors would like to acknowledge NATO for sup-
porting this work, as well as members of the NATO SET-
227 Task Group on Cognitive Radar for their insights and
active research in this field.
REFERENCES
[1] A. Balleri, H. Griffiths, and C. Baker, Biologically-Inspired
Radar and Sonar: Lessons from Nature. IET, 2017.
[2] S. Middelhoek, J. B. Angell, and D. J. W. Noorlag, “Micro-
processors get integrated sensors: Sensing devices and signal
processing built into one silicon chip portend a new class of
smart sensors,” IEEE Spectrum, vol. 17(2), pp. 42–46, 1980.
[3] L. Reznik, G. Von Pless, and T. Al Karim, “Distributed neural
networks for signal change detection: On the way to cognition
in sensor networks,” IEEE Sensors Journal, vol. 11, no. 3, pp.
791–798, March 2011.
[4] R. V. Kulkarni, A. Forster, and G. K. Venayagamoorthy, “Com-
putational intelligence in wireless sensor networks: A survey,”
IEEE Communications Surveys Tutorials, vol. 13, no. 1, pp.
68–96, First 2011.
[5] J. Fuster, Cortex and Mind: Unifying Cognition. Oxford, UK:
Oxford University Press, 2003.
[6] N. Wiener, The Human Use of Human Beings: Cybernetics and
Society. Houghton Mifflin, 1950.
[7] ——, Cybernetics or Control and Communication in the Animal
and the Machine. Cambridge, MA: John Wiley and Sons, 1948.
[8] C. Helstrom, Elements of Signal Estimation and Detection.
Engelwood Cliffs, NJ: Prentice-Hall, 1994.
[9] H. Chernoff, “Locally optimal designs for estimating parame-
ters,” Ann. Math. Stat., vol. 24(4), pp. 586–602, Dec. 1953.
[10] ——, “Sequential design of experiments,” Ann. Math. Stat.,
vol. 30, no. 3, pp. 755–770, Sept. 1959.
[11] L. Meier, J. Peschon, and R. Dressler, “Optimal control of
measurement subsystems,” IEEE Transactions on Automatic
Control, vol. 12, no. 5, pp. 528–536, October 1967.
[12] M. Athans, “On the determination of optimal costly mea-
surement strategies for linear stochastic systems,” Automatica,
vol. 8, no. 4, pp. 394–412, Jul. 1972.
[13] A. O. Hero and D. Cochran, “Sensor management: Past, present,
and future,” IEEE Sensors Journal, vol. 11, no. 12, pp. 3064–
3075, Dec 2011.
[14] B. Widrow, P. E. Mantey, L. J. Griffiths, and B. B. Goode,
“Adaptive antenna systems,” Proceedings of the IEEE, vol. 55,
no. 12, pp. 2143–2159, Dec 1967.
[15] W. D. Blair, G. A. Watson, and S. A. Hoffman, “Benchmark
problem for beam pointing control of phased array radar against
maneuvering targets,” in Proc. American Control Conf., vol. 2,
June 1994, pp. 2071–2075 vol.2.
[16] W. D. Blair, G. A. Watson, T. Kirubarajan, and Y. Bar-Shalom,
“Benchmark for radar allocation and tracking in ecm,” IEEE
Transactions on Aerospace and Electronic Systems, vol. 34,
no. 4, pp. 1097–1114, Oct 1998.
[17] G. van Keuk and S. S. Blackman, “On phased-array radar track-
ing and parameter control,” IEEE Transactions on Aerospace
and Electronic Systems, vol. 29, no. 1, pp. 186–194, Jan 1993.
[18] S. S. Blackman, R. J. Dempster, M. T. Busch, and R. F.
Popoli, “IMM/MHT solution to radar benchmark tracking prob-
lem,” IEEE Transactions on Aerospace and Electronic Systems,
vol. 35, no. 2, pp. 730–738, April 1999.
[19] T. Kirubarajan, Y. Bar-Shalom, W. D. Blair, and G. A. Watson,
“IMMPDAF for radar management and tracking benchmark
with ECM,” IEEE Transactions on Aerospace and Electronic
Systems, vol. 34, no. 4, pp. 1115–1134, Oct 1998.
[20] J. Ward, Space-Time Adaptive Processing for Airborne Radar.
MIT Lincoln Laboratory, Cambridge, MA, 1994.
[21] W. K. Stafford, “Real time control of a multifunction electron-
ically scanned adaptive radar (mesar),” in IEE Colloquium on
Real-Time Management of Adaptive Radar Systems, June 1990,
pp. 7/1–7/5.
[22] S. Haykin, “Radar vision,” in IEEE International Conference
on Radar, May 1990, pp. 585–588.
[23] S. J. Anderson, “Adaptive remote sensing with hf skywave
radar,IEE Proceedings F - Radar and Signal Processing, vol.
139, no. 2, pp. 182–192, April 1992.
[24] D. J. Kershaw and R. J. Evans, “Optimal waveform selection for
tracking systems,” IEEE Transactions on Information Theory,
vol. 40, no. 5, pp. 1536–1550, Sep. 1994.
[25] ——, “Waveform selective probabilistic data association,IEEE
Trans. Aerosp. Elec. Sys., vol. 33(4), pp. 1180–1188, 1997.
[26] B. F. La Scala, W. Moran, and R. J. Evans, “Optimal adaptive
waveform selection for target detection,” in 2003 Proceed-
ings of the International Conference on Radar (IEEE Cat.
No.03EX695), Sep. 2003, pp. 492–496.
[27] B. L. Scala, M. Rezaeian, and B. Moran, “Optimal adaptive
waveform selection for target tracking,” in 2005 7th Interna-
tional Conference on Information Fusion, vol. 1, July 2005,
pp. 6 pp.–.
[28] M. C. Wicks, “A brief history of waveform diversity,” in IEEE
Radar Conference, May 2009.
[29] S. D. Blunt and E. L. Mokole, “Overview of radar waveform
diversity,” IEEE Aerospace and Electronic Systems Magazine,
vol. 31, no. 11, pp. 2–42, November 2016.
[30] H. Kim and N. A. Goodman, “Waveform design by task-specific
information,” in IEEE Radar Conf., 2010, pp. 848–852.
[31] J. Guerci, Cognitive Radar: The Knowledge-Aided Fully Adap-
tive Approach. Artech House, 2010.
[32] S. D. J.R. Klauder, A.C. Price and W. Albersheim, “The theory
and design of chirp radars,” The Bell System Technical Journal,
vol. XXXIX, no. 4, pp. 745–808, July 1960.
[33] R. Turyn, “On barker codes of even length,” Proceedings of the
IEEE, vol. 51, no. 9, p. 1256, Sep 1963.
[34] D. DeLong and E. Hofstetter, “On the design of optimum
radar waveforms for clutter rejection,IEEE Transactions on
Information Theory, vol. 13, no. 3, pp. 454–463, July 1967.
[35] C. Stutt and L. Spafford, “A ’best’ mismatched filter response
for radar clutter discrimination,” IEEE Transactions on Infor-
mation Theory, vol. 14, no. 2, pp. 280–287, March 1968.
[36] B. L. Lewis and F. F. Kretschmer, “A new class of polyphase
pulse compression codes and techniques,” IEEE Transactions
on Aerospace and Electronic Systems, vol. AES-17, no. 3, pp.
364–372, May 1981.
[37] J. P. Costas, “A study of a class of detection waveforms having
nearly ideal range - Doppler ambiguity properties,” Proceedings
of the IEEE, vol. 72, no. 8, pp. 996–1009, Aug 1984.
[38] D. Gjessing, Target Adaptive Matched Illumination Radar:
Principles and Applications. Peter Peregrinus, Ltd., 1986.
[39] P. M. Woodward, Probability ad Information Theory, with
Applications to Radar. Artech House (reprint), 1980.
[40] M. R. Bell, “Information theory and radar waveform design,
IEEE Trans. Inf. Th., vol. 39(5), pp. 1578–1597, Sep. 1993.
[41] S. U. Pillai, H. S. Oh, D. C. Youla, and J. R. Guerci, “Optimal
transmit-receiver design in the presence of signal-dependent
interference and channel noise,” IEEE Transactions on Infor-
mation Theory, vol. 46, no. 2, pp. 577–584, March 2000.
[42] D. Middleton, “Joint optimization of transmission and reception
by choice of signal waveform,” in An Introduction to Statistical
Communication Theory. McGraw Hill Book Company, 1959.
[43] S. Haykin, “Cognitive radar: a way of the future,IEEE Signal
Processing Magazine, vol. 23, no. 1, pp. 30–40, Jan 2006.
[44] K. L. Bell, C. J. Baker, G. E. Smith, J. T. Johnson, and M. Ran-
gaswamy, “Cognitive radar framework for target detection and
tracking,” IEEE Journal of Selected Topics in Signal Processing,
vol. 9, no. 8, pp. 1427–1439, Dec 2015.
[45] P. Antonik, H. Schuman, P. Li, W. Melvin, and M. Wicks,
“Knowledge-based space-time adaptive processing,” in Pro-
ceedings of the 1997 IEEE National Radar Conference, May
1997, pp. 372–377.
[46] W. L. Melvin and J. R. Guerci, “Knowledge-aided signal
processing: a new paradigm for radar and other advanced sen-
sors,” IEEE Transactions on Aerospace and Electronic Systems,
vol. 42, no. 3, pp. 983–996, July 2006.
[47] C. T. Capraro, G. T. Capraro, I. Bradaric, D. D. Weiner, M. C.
Wicks, and W. J. Baldygo, “Implementing digital terrain data in
knowledge-aided space-time adaptive processing,IEEE Trans.
Aero. Elec. Sys., vol. 42, no. 3, pp. 1080–99, 2006.
[48] X. Mao, X. He, and D. Li, “Knowledge-aided 2-d autofocus
for spotlight sar range migration algorithm imagery,IEEE
Transactions on Geoscience and Remote Sensing, vol. 56, no. 9,
pp. 5458–5470, Sep. 2018.
[49] R. Ding, M. Yu, H. Oh, and W. Chen, “New multiple-target
tracking strategy using domain knowledge and optimization,
IEEE Transactions on Systems, Man, and Cybernetics: Systems,
vol. 47, no. 4, pp. 605–616, April 2017.
[50] W. L. Melvin, G. A. Showman, and J. R. Guerci, “A knowledge-
aided GMTI detection architecture [radar signal processing],” in
Proc. IEEE Radar Conference, April 2004, pp. 301–306.
[51] J. Matuszewski and A. Kawalec, “Knowledge-based signal pro-
cessing for radar identification,” in International Conference on
”Modern Problems of Radio Engineering, Telecommunications
and Computer Science”, Feb 2008, pp. 302–305.
[52] M. C. Wicks, “Radar the next generation - sensors as robots,
in Proc. Int. Conf. on Radar, Sep. 2003, pp. 8–14.
[53] G. T. Capraro, M. C. Wicks, and I. Bradaric, “Sensors as
intelligent robots,” in IET Int. Conf. on Radar Sys., Oct 2007.
[54] S. Haykin, “Adaptive radar: evolution to cognitive radar,” in
IEEE Int. Symp. Phased Array Sys. and Tech., Oct 2003.
[55] S. Haykin, Y. Xue, and P. Setoodeh, “Cognitive radar: Step to-
ward bridging the gap between neuroscience and engineering,
Proceedings of the IEEE, vol. 100, no. 11, pp. 3102–3130, Nov
2012.
[56] J. Fuster, “Cortical memory.” Scholarpedia, vol. 2, p. 1644, 01
2007.
[57] J. Rasmussen, “Skills, rules, and knowledge; signals, signs, and
symbols, and other distinctions in human performance models,”
IEEE Transactions on Systems, Man, and Cybernetics, vol.
SMC-13, no. 3, pp. 257–266, May 1983.
[58] ——, Information Processing and Human-Machine Interaction:
An Approach to Cognitive Eng. Elsevier Science, 1983.
[59] S. Bruggenwirth, “Design and implementation of a three-layer
cognitive radar architecture,” in 2016 50th Asilomar Conference
on Signals, Systems and Computers, Nov 2016, pp. 929–933.
[60] “DDDAS program website,” http://www.1dddas.org/, accessed:
2019-03-15.
[61] G. E. Smith, Z. Cammenga, A. Mitchell, K. L. Bell, J. Johnson,
M. Rangaswamy, and C. Baker, “Experiments with cognitive
radar,IEEE Aerospace and Electronic Systems Magazine,
vol. 31, no. 12, pp. 34–46, December 2016.
[62] J. M. Christiansen, G. E. Smith, and K. E. Olsen, “Usrp based
cognitive radar testbed,” in IEEE Radar Conf., May 2017.
[63] R. Oechslin, U. Aulenbacher, K. Rech, S. Hinrichsen,
S. Wieland, and P. Wellig, “Cognitive radar experiments with
codir,” in International Conference on Radar Systems, Oct 2017,
pp. 1–6.
[64] J. R. Guerci, R. M. Guerci, M. Ranagaswamy, J. S. Bergin,
and M. C. Wicks, “CoFAR: Cognitive fully adaptive radar,” in
IEEE Radar Conference, May 2014, pp. 0984–0989.
[65] A. Charlish and F. Hoffmann, “Cognitive radar management,” in
Novel Radar Techniques and Applications Volume 2: Waveform
Diversity and Cognitive Radar, R. Klemm and et. al, Eds.
Oxford: IET, 2017, pp. 157–193.
[66] A. D. A. Farina and S. Haykin, Eds., Cognitive Radar: the
Knowledge-Aided Fully Adaptive Approach. IET, 2017.
[67] AES, “IEEE standard for radar definitions,” IEEE Std 686-2017
(Revision of IEEE Std 686-2008), pp. 1–54, Sep. 2017.
[68] C. Horne, M. Ritchie, and H. Griffiths, “Proposed ontology for
cognitive radar systems,IET Radar, Sonar Navigation, vol. 12,
no. 12, pp. 1363–1370, 2018.
[69] C. Kreucher, A. O. Hero, and K. Kastella, “A comparison
of task driven and information driven sensor management for
target tracking,” in Proceedings of the 44th IEEE Conference
on Decision and Control, Dec 2005, pp. 4004–4009.
[70] A. E. Mitchell, G. E. Smith, K. L. Bell, A. J. Duly, and
M. Rangaswamy, “Cost function design for the fully adaptive
radar framework,IET Radar, Sonar Navigation, vol. 12, no. 12,
pp. 1380–1389, 2018.
[71] K. L. Bell, J. T. Johnson, G. E. Smith, C. J. Baker, and
M. Rangaswamy, “Cognitive radar for target tracking using
a software defined radar system,” in IEEE Radar Conference,
May 2015, pp. 1394–1399.
[72] R. Oechslin, P. Wellig, S. Hinrichsen, S. Wieland, U. Aulen-
bacher, and K. Rech, “Cognitive radar parameter optimization in
a congested spectrum environment,” in IEEE Radar Conference,
April 2018.
[73] A. E. Mitchell, G. E. Smith, K. L. Bell, A. J. Duly, and
M. Rangaswamy, “Hierarchical fully adaptive radar,IET Radar,
Sonar Navigation, vol. 12, no. 12, pp. 1371–1379, 2018.
[74] L. Ubeda-Medina, . F. Garca-Fernndez, and J. Grajal, “Robust
sensor parameter selection in fully adaptive radar using a sigma-
point gaussian approximation,” in IEEE Radar Conference,
April 2018, pp. 0263–0268.
[75] “International telecommunication union radio regulations,
http://www.itu.int/pub/R-REG-RR-2012, accessed: 2019-03-31.
[76] H. Griffiths, L. Cohen, S. Watts, E. Mokole, C. Baker,
M. Wicks, and S. Blunt, “Radar spectrum engineering and
management: Technical and regulatory issues,” Proceedings of
the IEEE, vol. 103, no. 1, pp. 85–102, Jan 2015.
[77] M. S. Greco, F. Gini, P. Stinco, and K. Bell, “Cognitive radars:
On the road to reality: Progress thus far and possibilities for
the future,” IEEE Signal Processing Magazine, vol. 35, no. 4,
pp. 112–125, July 2018.
... Radar enables systems to make informed decisions, improving performance. Cognitive radar (CR) further enhances this by actively sensing and adapting to the environment through a feedback loop between transmitter and receiver modules [1], [2]. CR systems continuously learn and adapt to their environment using a perception-action cycle, optimizing target detection and tracking in dynamic environments by responding to noise, clutter, and moving targets. ...
Preprint
Motivated by the growing interest in integrated sensing and communication for 6th generation (6G) networks, this paper presents a cognitive Multiple-Input Multiple-Output (MIMO) radar system enhanced by reinforcement learning (RL) for robust multitarget detection in dynamic environments. The system employs a planar array configuration and adapts its transmitted waveforms and beamforming patterns to optimize detection performance in the presence of unknown two-dimensional (2D) disturbances. A robust Wald-type detector is integrated with a SARSA-based RL algorithm, enabling the radar to learn and adapt to complex clutter environments modeled by a 2D autoregressive process. Simulation results demonstrate significant improvements in detection probability compared to omnidirectional methods, particularly for low Signal-to-Noise Ratio (SNR) targets masked by clutter.
... Driven by the applications for high resolution in the complex electromagnetic environment, modern radars are required to maintain the real-time detection Manusript Color versions of one or more of the figures in this article are available online at http://ieeexplore.ieee.org with high resolution while increasing their survivability in the complex environment [5][6][7][8][9][10][11][12]. Compared to the wideband radar with a single ultrawide band, the radar with multiband can impressively improve the radar's survivability under complex environment. ...
Article
Full-text available
In this paper, a novel strategy of employing microwave photonic (MWP) time-frequency limiter (TFL) for microwave photonic multiband radar is proposed to suppress the interference, achieving real-time response to the interference scenarios and high-resolution target detection. By mapping the echo signal into optical domain, the time-frequency characteristic is re-constructed through stimulated Brillouin scattering (SBS), realizing the selective suppression on high-power optical signal mapped by the interference. Based on this concept, a MWP TFL system based on the optical spectrum processing is constructed, and proof-of-concept experiments are demonstrated to verify the feasibility of the proposed strategy under different interference scenarios. Employing the proposed MWP TFL, the signal-to-noise ratio of the detection results, which is severely degraded by asynchronous interference, can be improved by 27.97dB, and the suppression ratio on the false targets generated by the synchronous interference can reach 34.10dB. The experimental results shows that the strategy can further enhance the survivability of multiband radar without compromising the range resolution for target detection. In addition, experiments are carried out to demonstrate the capability of the proposed strategy under different interference-to-signal ratios, showing a good adaptability to the complex interference scenarios.
... In the increasingly complex electronic warfare environment characterized by rapid technological advancements and sophisticated adversarial tactics, traditional radar systems that rely on a single working mechanism and simplistic beam variations are becoming inadequate [1], [2]. These conventional radars struggle to meet the diverse operational requirements of modern warfare scenarios, which demand high levels of adaptability and precision in detection capabilities . ...
Article
With the rapid advancement of phased array radar, the guidance and anti-jamming capabilities of phased array radar seekers have been further enhanced. Traditional radar jamming decision-making methods are no longer applicable in electronic warfare. Thus, radar jamming decision-making methods based on reinforcement learning have emerged, which can effectively address the issues of poor performance and low efficiency of traditional jamming approaches. Given the extremely high cost and poor repeatability of physical experiments, the construction of simulation models is frequently adopted for simulation experiments. Unlike most other studies on radar jamming decision-making that employ functional-level simulation modeling, this paper adopts signal-level simulation of radar jamming decision-making, presenting a more realistic and intuitive reflection of the entire process of interference equipment interfering with the missile terminal guidance. Through the signal-level simulation of phased array radar terminal guidance and the introduction of reinforcement learning algorithms, this study investigates how the interference equipment should act to maximize the interference benefit. Through simulation experiments, on the one hand, it is demonstrated that the reinforcement learning algorithm can improve the interference effect. On the other hand, the accuracy of the signal-level simulation model is verified.
... Any adaptation made is performed on the retrieved signals. Gurbuz et al. (2019) describe this as there is no action, only perception. Cognitive radar is defined by the Institute of Electrical and Electronics Engineers (IEEE) as "A radar system that in some sense displays intelligence, adapting its operation and its processing in response to a changing environment and target scene. ...
... Geometry optimization [12,[22][23][24], iii, EM Cancellation technology [24][25][26]. However, as artificial intelligence, new generation information technology and radar detection become increasingly connected [27], traditional methods are facing more and more constraints in reducing target radar scattering [15,28]. Therefore, investigating and examining stealth technology using novel approaches is a crucial area that warrants our focus. ...
Article
Full-text available
Recently, researchers have realized various exotic electromagnetic control devices using the coded metasurfaces, sparking a broad investigation into the phase or amplitude-based encoding method, as well as their combination, in the field of metasurface design. In this paper, to evaluate the influence of random mutual coupling between the adjacent element on the scattering performance of metasurface, and also to minimize the backward radar cross section (RCS) of metal plate targets, a novel encoding approach combining the reflection phase and element-form has been proposed. During the implementation process, an anisotropic hypocycloid inspired 3-bit digital coding metasurface was designed. It consists of 9 different element-forms, with each capable of providing 7 phase states. Simulation results demonstrate that the random mutual coupling introduced by the proposed elements does not significantly affect the RCS performance of the metasurface. With a good polarization insensitivity property for both linearly and circularly polarized waves, the designed 3-bit digital coding metasurface can achieve more than 20 dB RCS reduction at 10 GHz, while simultaneously transmitting additional information by encoding the element forms. The good consistency between theoretical simulation and sample testing unequivocally validates the precision of the design, this paper may serve as a useful reference for expanding the design methods of metasurfaces.
... Modern Multi-Function Radars (MFRs) are capable of performing multiple simultaneous tasks in the radar timeline, with each task being implemented through a sequence of ordered radar pulses [1][2][3][4][5]. In addition, modern radars are equipped with the ability to flexibly select or optimize their control parameters for each individual task based on the sensing of surrounding environments [6][7][8]. ...
Article
Full-text available
Modern Multi-Function Radars (MFRs) are sophisticated sensors that are capable of flexibly adapting their control parameters in transmitted pulse sequences. In complex electromagnetic environments, efficiently and accurately recognizing the inter-pulse modulations of non-cooperative radar pulse sequences is a key step for modern Electronic Support (ES) systems. Existing recognition methods focus more on algorithmic designs, such as neural network structure designs, to improve recognition performance. However, in open electromagnetic environments with increased flexibility in radar transmission, these methods would suffer performance degradation due to domain shifts between training and testing datasets. To address this issue, this study proposes a robust radar inter-pulse modulation feature extraction and recognition method based on disentangled representation learning. At first, inspired by the Representation Learning Theory (RLT), the received radar pulse sequences can be disentangled into three explanatory factors related to (i) modulation types, (ii) modulation parameters, and (iii) measurement characteristics, such as measurement noise. Then, an explainable radar pulse sequence disentanglement network is proposed based on auto-encoding variational Bayes. The features extracted through the proposed method can effectively represent the key latent factors related to recognition tasks and maintain performance under domain shift conditions. Experiments on both ideal and non-ideal situations demonstrate the effectiveness, robustness, and superiority of the proposed method in comparison with other methods.
Article
A cognitive tracking radar continuously acquires, stores, and exploits knowledge from its target environment in order to improve kinematic tracking performance. In this work, we apply a reinforcement learning (RL) technique, API-DNN, based on approximate policy iteration (API) with a deep neural network (DNN) policy to cognitive radar tracking. API-DNN iteratively improves upon an initial base policy using repeated application of rollout and supervised learning. This approach can appropriately balance online versus offline computation in order to improve efficiency and can adapt to changes in problem specification through online replanning. Prior state of the art cognitive radar tracking approaches either rely on sophisticated search procedures with heuristics and carefully selected hyperparameters or deep reinforcement learning agents based on exotic DNN architectures with poorly understood performance guarantees. API-DNN, instead, is based on well known principles of rollout, Monte Carlo simulation, and basic DNN function approximation. We demonstrate the effectiveness of API-DNN in cognitive radar simulations based on a standard maneuvering target tracking benchmark scenario. We also show how API-DNN can implement online replanning with updated target information.
Article
In this letter, the avoiding of the powerful interference of intentional modulation (IM) information on unintentional modulation (UM) feature is primarily studied. To address this challenging issue, a novel framework for deep UM feature extraction is proposed. The ideas of decomposition reconstruction and metric learning are introduced into deep learning. Meanwhile, an objective function is designed to automatically learn the deep UM feature that is insensitive to the IM information. The experimental results show the remarkable stability and separability of the deep UM feature across measured data with variable IM parameters.
Article
Full-text available
By emulating the cognitive perception–action cycle believed to be at the core of animal cognition, cognitive radars promise to improve radar performance over standard systems. The fully adaptive radar (FAR) framework provides a generalised approach to implementing a single cognitive perception–action cycle for radar systems, but complex adaptive problems necessitate the interaction of multiple perception–action cycles. This study describes the general form of the hierarchical FAR (HFAR) framework. The HFAR framework is applied to a single-target tracking, sensor fusion problem, and real-time experimental results demonstrate the efficacy of the proposed architecture for handling problems of varying scales in a consistent, adaptive fashion.
Article
Full-text available
By emulating the neuropsychological processes underpinning animal cognition, the field of cognitive radar seeks to improve performance compared to non-adaptive systems. The fully adaptive radar (FAR) framework is an application agnostic means of implementing the perception–action cycle in radars. This work proposes a method of designing the FAR framework's component cost functions inspired by the field of multi-objective optimisation. As an illustration, the general cost functions were used to implement waveform adaptation for single target tracking. Both simulated and experimental results demonstrated how altering the cost functions can tailor the FAR performance to specific radar operating modes.
Article
Full-text available
With continuous improvement of the synthetic aperture radar (SAR) resolution, the 2-D defocus effect in SAR image resulting from uncompensated motion error has made autofocus become a new challenging problem. Conventional 2-D autofocus approaches assume that the 2-D phase error is absolutely unknown and estimate them directly. Due to high dimensionality of the unknown parameters, these approaches often suffer from high computational burden and low estimate accuracy. In this paper, we analyze the effect of range migration algorithm (RMA) processing on the 2-D echo phase, and reveal the analytical structure of the residual 2-D phase error in RMA imagery. Then, by exploiting the derived prior knowledge on the phase error structure, an accurate and efficient 2-D autofocus approach is proposed. In the new method, only 1-D error, e.g., azimuth phase error, or residual range cell migration, is required to be estimated directly, while the 2-D phase error is computed directly from the estimated 1-D error by exploiting the analytical structure of the 2-D phase error. The experimental results clearly demonstrate the effectiveness and robustness of the proposed method.
Chapter
Full-text available
Cognitive radar is a radar system that acquires knowledge and understanding of its operating environment through online estimation, reasoning and learning or from databases comprising context information. A cognitive radar then exploits this acquired knowledge and understanding to enhance information extraction, data processing and radar management. In order to make progress to this goal, the topic of cognitive radar attempts to shift the cognitive processes previously performed by an operator into automated processes in the radar system. Families of cognitive processes are well defined in cognitive psychology, such as the perceptual processes, memory processes, languages processes, and thinking processes. In this chapter, we discuss radar management techniques that enable the manifestation of one or more cognitive processes, with a particular view towards electronically steered phased array and multifunction radar systems. In particular, this chapter focuses on two cognitive processes: attention and anticipation. Attention can be manifested by effective resources management, whereby a quality of service based task management layer connects radar control parameters to mission objectives. Anticipation can be generated using stochastic control which is non-myopic, allowing the radar system to act with a consideration of how the radar system, scenario and environment will evolve in the future.
Article
Cognitive radar is a rapidly developing area of research with many opportunities for innovation. A significant obstacle to development in this discipline is the absence of a common understanding of what constitutes a cognitive radar. The proposition in this article is that radar systems should not classed as cognitive, or not cognitive, but should be graded by the degree of cognition exhibited. We introduce a new taxonomy framework for cognitive radar against which research, experimental and production systems can be benchmarked, enabling clear communication regarding the level of cognition being claimed or discussed.
Article
This article describes some key ideas and applications of cognitive radars, highlighting the limits and the path forward. Cognitive radars are systems based on the perception-action cycle of cognition that senses the environment, learns relevant information from it about the target and the background, and then adapts the radar sensor to optimally satisfy the needs of the mission according to a desired goal. The concept of cognitive radar was originally introduced only for active radar. In this article, we explain how this paradigm can also be applied to passive radar (PR).
Conference Paper
This paper presents experimental results with the Cognitive Detection, Identification and Ranging (CODIR) testbed set up in a congested spectrum and jammed environment. Band jamming is realized with a noise diode source, followed by power amplifiers and band pass filters. Special gapped linear frequency modulated (LFM) waveforms with minimal frequency content in the jammed band are designed and added to the standard LFM waveform list. Given a pre-defined performance goal in term of detection rate and track quality, the radar parameter optimization algorithm adapts bandwidth, pulse repetition frequency, integration time and pulse width such that time resources are minimized.
Conference Paper
This paper presents first experimental results from a recent outdoor campaign with the Cognitive Detection, Identification and Ranging (CODIR) testbed. First, the CODIR testbed, which consists of an adaptive radar sensor and a cognitive controller which calculates the optimized sensor parameters is presented. In a second part, we describe the campaign goals, the test design and the measurement data. Finally, the optimisation algorithms are applied to the collected data in order to get an optimized radar parameter set at each track update. Given a pre-defined performance goal in term of detection rate and track quality, the controller adapts bandwidth, pulse repetition frequency, integration time and pulse width such that time and bandwidth resources are minimized. The sensor just needs the resources to fulfil its performance goal and releases unused resources for other potential tasks.
Chapter
In the last decade, a number of researchers have used information-theoretic ideas and maximization of mutual information in the design of radar waveforms for adaptive waveform radar and multiple-input multiple-output (MIMO) radar. However, it is not clear under what circumstances these approaches lead to optimal or near-optimal results. In this chapter, we reviewthe fundamental ideas behind the use of information theory and information measures in radar waveform design. We also briefly review some of the more recent results in this area.