Please read the Expert System CFAR article on IEEE AESS magazine 2017 and analyze the following questions (Answers must be presented with proof, such as Matlab simulation, see individual question for more detail)
- How does Power density function affect different kind of CFAR algorithms mentioned in the article (Matlab)?
- Explain different CFAR algorithm mentioned in the article.
From the Figure 3, prototype expert system CFAR detection processor,
- Please explain what functionalities does CFAR Selection Functional Block have and its relation with respected of the neighboring functional blocks.
- What is difference between Baseline CFAR and ES-CFAR?
- Explain the relation between weighting to the inputs, blue circle (Matlab)
- From your opinion, what can we improve the performance of the Expert System CFAR detection Processor (Matlab)?
- The final report should be One Microsoft word documents (First page should be cover Letter).
- ONE Matlab routine with names .Please also use common section to identify which question that the section of code refers to.
INTRODUCTION
The genesis of the Expert System Constant
False Alarm Rate (CFAR) processing
arose in 1984 from insight gained
during earlier experiments performed
at the Air Force Research Laboratory’s
Rome Research Site, then known as
Rome Air Development Center (RADC)
(Figure 1). Measured data analysis from
the low altitude detection (LAD) experiments
conducted in RADC’s Surveillance
Laboratory was instrumental in gaining
insight into detecting weak signals (airborne
target returns) embedded in strong
nonhomogeneous clutter. This challenging
problem, investigated by Signal Processing
Chief, Mr. Clarence Silfer in the
late 1970s, was studied to improve cruise
missile detection by unattended short-range ground-based radars.
The LAD experiments were performed in support of the Enhanced
Defense Early Warning (EDEW) project [1]. Dr. Russell Brown and
Mr. David Mokry conducted measurements in the summer of 1981,
and Mr. Paul van Etten performed analysis. The first author assisted
in these early endeavors, focusing on data analysis.
A track-while-scan (TWS) instrumentation radar was used
for data collection. This coherent radar (waveform generation,
timing, and control implemented via programmable emitter coupled
logic) operated with up to 60 MHz of signal bandwidth at
L-Band, employing a 150 KW traveling wave tube for the final
stage of high-power amplification on transmit and superheterodyne
down conversion on receive. This type of radio receiver
uses stages of frequency mixing, amplification, and filtering to
convert a microwave signal of interest to a fixed intermediate
frequency (IF). The resulting IF signal, a fully coherent yet frequency-
shifted replica of the received microwave signal, is more
conveniently down-converted to baseband, digitized, processed,
and/or recorded.
This L-Band instrumentation radar, integrated in 1981 with
an AMPEX Corp HBR-3000 digital data recorder and interfaced
via a buffer memory to a Floating Point Systems AP-120B array
processor, was under the control of a Hewlett-Packard HP 1000
L-Series computer. This made data collection, quick-look calibration,
and in-depth postmission signal and data processing possible.
An embedded four-channel sidelobe canceller was instrumental in
mitigating in-band sidelobe interference in the LAD radar data collection
experiments.
Authors’ addresses: M. C. Wicks, Dept. of Electrical and
Computer Engineering, University of Dayton, 341 Kettering
Laboratory, 300 College Park, Dayton, OH 45469. E-mail:
(mwicks1@udayton.edu). W. J. Baldygo, US Air Force Research
Laboratory, 2241 Avionics Circle, Wright Patterson Air Force
Base, OH 45433-7320.
Public Approval: Approved for Public Release, Case Number
RY-16-0671.
Manuscript received November 2, 2016, revised December 21,
2016, and ready for publication December 27, 2016.
Review handled by W. Walsh.
0885/8985/17/$26.00 © 2017 IEEE
Feature Article:
Expert System CFAR: Algorithm Development,
Experimental Demonstration, and Transition to
Airborne Radar Systems
Michael C. Wicks, University of Dayton, Dayton, OH, USA
William J. Baldygo, US Air Force Research Laboratory, Wright Patterson Air Force
Base, OH, USA
DOI. No. 10.1109/MAES.2017.160243
Figure 1.
US Air Force ground-based radar test bed in Rome, NY.
40 IEEE A&E SYSTEMS MAGAZINE SEPTEMBER 2017
aesm-32-09-03 Page 40 PDF Created: 2017-10-10: 8:37:AM
The goal of this measurements-driven and phenomena-oriented
research was clear and simple—to explore the impact of very
low threshold detection processing on track formation in a fully
coherent TWS radar with an embedded four-channel sidelobe canceller.
The test targets of greatest interest were small, slow moving
aircraft flying 100 m above ground level. All experiments were
conducted in close proximity (40 km) to the radar test bed. Research
focused on exploiting coherent clutter maps, autoregressive
superresolution spectral estimators, and Kalman filter-based noisy
area TWS processing. Postmission analysis of data demonstrated
that slow moving targets often “broke track” due to residual interference
in one or more Doppler filters. This was caused by strong
backscatter signals from natural terrain and close-in manmade
structures in addition to electromagnetic interference. The early
conjecture was that internal clutter motion and scanning modulation
would also contribute to degraded track performance, but it
was later determined that these effects were negligibly small by
comparison.
The weak returns from low, slow test aircraft most interested
the analysis team. We initially implemented an adaptive Doppler
processor followed by a three-dimensional cell-averaging constant
false alarm rate (CA-CFAR) detector, with a crude form of extreme
value excision. This “modified” multidimensional CA-CFAR, a
detector that spanned range, angle, and Doppler, was permanently
incorporated into the L-Band radar signal and data processing
chain (in software on the AP-120B array processor). The team decided
that a symmetric training data window was adequate only
for initial analysis, and that detector performance would indicate
the direction for future CFAR research. Analysis employing multiple
detectors immediately followed. Both the greatest-of (GO)
and trimmed mean (TM) CFAR were investigated. Topographical
maps and a coherent clutter map (using only several Doppler filters)
were used as information sources (an early, albeit self-generated
knowledge source) for suppressing outliers and to compute
clutter-plus-noise power estimates in regions of heterogeneous terrain.
It was soon verified that the effects of outliers were greatly
reduced using multiscan training data in range, angle, and Doppler.
The track processor was also utilized to excise persistent returns
from “stationary movers” spatially localized in the training data.
Furthermore, asymmetric training data windows were employed
due to the close proximity (of the slow movers) to the zero Doppler
filter and the effects of numerous dominant structures surrounding
the radar test bed.
In late 1981, the LAD experiments proved successful and Mr.
Silfer transitioned results to the sponsor. Gains in wide area surveillance
radar performance would be enabled by improvements
in track processing, with appropriate changes to Doppler filtering,
sidelobe cancellation, and especially false alarm control. This research
concluded with little fanfare, as the topic of ground-based
wide area surveillance radar was no longer of great interest to the
basic research community. However, this experience influenced
the careers of many radar engineers at RADC, as weak target detection
became the focus of signal processing experiments in the
Surveillance Laboratory for the next three decades.
EXPERT SYSTEM CF AR CONCEPT DEVELOPMENT
In 1984, as technical leader of the seedling in-house Air Defense
Initiative (ADI) at RADC, Dr. Richard Schneible requested support
from the Surveillance Laboratory staff to develop advanced
radar technology in order to improve weak signal detection in
wide area surveillance radars via advanced airborne moving target
indication (AMTI), Doppler processing, and track formation.
This futuristic ADI radar was to be a “concept car” 1 designed to
foster discussion and provide a baseline for developing a post-
AWACS (airborne warning and control system) radar capability.
Similar to the EDEW project, the mission was to develop a wide
area surveillance capability for detecting and tracking the modern
air-breathing threat under all weather conditions, albeit now from
an airborne sensor platform.
At RADC, we had examined the link budget for the airborne
radar in operation onboard AWACS. Significant performance
gains were theoretically possible through signal and data processing.
However, large gains were not practically achievable through
1 Concept cars were popular in the automotive industry during the
mid-20th century as a means of illustrating planned or potential
futuristic body styles.
SEPTEMBER 2017 IEEE A&E SYSTEMS MAGAZINE 41
aesm-32-09-03 Page 41 PDF Created: 2017-10-10: 8:37:AM
Expert System CF AR
increased radar aperture or dramatically higher average transmit
power (light-heartedly described with the graphic in Figure 2).
The simple goal was to look for immediate gains in filtering, false
alarm control, and track processing. At that point, management and
leadership were open to all suggestions. Mr. Fred Demma, Chief
of Surveillance Technology at RADC, heavily promoted a philosophy
of “MASS for MIPS” in which we emphasized algorithm development
and computing solutions over power-aperture, all while
remaining cognizant of the limitations due to the thermal noise
floor. An investigation into target correlation on extended coherent
dwell ensued [2].
Multiple researchers responded to Dr. Schneible’s call for technology
solutions to this challenging AMTI problem, including
members of the aforementioned LAD research team. Dr. Brown
focused on improving the linear dynamic range of the radar receiver
and improving the balance of the in-phase and quadrature
baseband signals. This proved extremely beneficial for realizing
enhanced coherent processing. This research motivated Dr. Brown
to develop novel IF sampling techniques using out-of-band noise
injection that aided in developing the Expert System CFAR. Mr.
van Etten studied the application of modern spectral estimators to
ADI, specifically focusing on autoregressive techniques to address
the problem of weak signal suppression [3]. Mr. Thomas Maggio,
also from RADC, was instrumental in applying a coherent form
of the noisy area tracker, a noncoherent track technique put into
production using analog phosphor storage in the 1960s. Mr. Maggio
investigated techniques for analyzing postdetection declarations,
separating false alarms from desired signals via “real-time”
statistical methods. At that same time, Mr. Robert Ogrodnik from
RADC sponsored Mr. Alan Corbeil, Dr. John DiDomizio, and Mr.
Lee Moyer, all from the Technology Service Corporation, to investigate
track before detect (TBD) [4], a knowledge-aided, fully
coherent, and much improved form of the aforementioned noisy
area tracker.
Dr. Schneible assigned the task of trade space analysis between
and among these and other signal and data processing techniques
to the first author. Initial results indicated that an improved CFAR
detector would yield a cost effective and immediate benefit to airborne
wide area surveillance radar, as the adaptive threshold multiplier
could be systematically reduced, weaker returns detected,
and false alarms precluded with improved exploitation of clutter
statistics. An Expert System CFAR would increase computational
complexity only modestly, and computing was a cost driver in the
mid-1980s. We pursued concept development, systems engineering
integration, and theoretical (statistical) analysis of the emerging
Expert System CFAR in addition to research in space-time
adaptive processing (STAP) for clutter rejection in airborne radar
[5]. STAP is a form of multidimensional spatial-temporal filtering
that potentially improves detection performance in airborne early
warning and ground moving target indication (GMTI) radar.
Fortuitously, Dr. Northup Fowler, a software expert and senior
research scientist at RADC, was independently pursuing intelligence,
surveillance, and reconnaissance applications of innovative
techniques in advanced computing. Dr. Fowler pursued, among
other software technology, a form of programming that emulates
the knowledge and analytic skills of a human expert. Furthermore,
basic and early applied research funds from the office of the Chief
Scientist became available to scientists and engineers conducting
research in this area, especially in speech and radar signal processing,
in order to apply expert reasoning software to enhance the
real-world performance of mathematical algorithms. Our research
in Expert System CFAR detection processing received additional
support from the office of the Chief Scientist.
The research of Dr. Hermann Rohling [6] further motivated
the first author to pursue a more rigorous systems-oriented design
of the Expert System CFAR detector. We took several months to
study the problem and propose a detector with the potential for
performance improvements but with limited demands for high-end
computing. In that era, real-time embedded parallel processing
was experimental in nature, extremely costly, and oriented towards
space applications [7]. We proposed a technology solution that
required very little advanced computing hardware. This solution,
which has since impacted several fielded systems, was designed
for simplicity of implementation. We proposed to run several different
(albeit well characterized) CFAR detector algorithms in
parallel, and to fuse these detector results using conventional algorithms.
The menu of standard CFAR detector algorithms included
cell averaging (CA), GO, TM, ordered statistic (OS), and smallest
of (SO). Additionally, several advanced CFAR detector algorithms
developed by Mr. James Sawyers from the Hughes Aircraft Company
(now Raytheon) were included. We leveraged conventional
fusion and track processing algorithm technology developed independently
by colleagues Dr. Pramod Varshney from Syracuse
University and Dr. Yaakov Bar-Shalom from the University of
Connecticut. These techniques were used to integrate the various
CFAR detector decisions and to produce a global declaration of
target present or target absent.
We fully understood that fusion algorithms were all developed
under the assumption that each data input to the decision processor
was independent of all others, and this was clearly not true in
Expert System CFAR. In the problem under analysis, identical data
is processed using various CFAR detectors, and a global decision
made using the aforementioned fusion rules. Still, we explored this
algorithmic approach experimentally, and developed the technology
more fully as performance proved worthy. Initial demonstrations
in the Surveillance Laboratory at RADC were astonishing! In
clutter and electromagnetic interference limited environments, the
earliest Expert System CFAR outperformed all other CFAR detectors.
At that same time, Dr. Vincent Vannicola from RADC began
Figure 2.
Increased power-aperture product is not always feasible.
42 IEEE A&E SYSTEMS MAGAZINE SEPTEMBER 2017
aesm-32-09-03 Page 42 PDF Created: 2017-10-10: 8:37:AM
Wicks and Baldygo
to explore the theory of predetection fusion [8] and to demonstrate
performance using data from the L-, S-, and C-Band radars in the
Surveillance Laboratory. Dr. Hong Wang from Syracuse University
and Dr. Brown developed adaptive multiband polarization [9]
processing technology in addition to polarization STAP. We leveraged
these technology investigations to benefit the development of
Expert System CFAR.
A formal effort to quantify performance, to package, and to
transition the Expert System CFAR more widely began in 1990. In
system design and analysis, it became obvious that each detector
embedded within the Expert System CFAR was optimum only under
very restrictive conditions, and, as such, would perform below
optimum when those conditions were not met, which was most of
the time. Furthermore, each CFAR detector could perform well below
optimum when severely mismatched to real-world operating
conditions. For example, a CA-CFAR detector operating close to
a land-sea interface could exhibit unacceptably high losses (order
of magnitude) in detection performance. Alternatively, in a multitarget
environment, a highly selective (very few training samples)
TM-CFAR could dramatically out-perform all other CFAR detectors,
but perform poorly in homogeneous clutter or beyond the
clutter line-of-sight.
The accurate estimation of the underlying spatial-temporal clutter
probability density function (PDF) was an intense area of research
in the 1980s. Contributions by Drs. Donald Weiner and Aydin
Ozturk of Syracuse University dramatically affected the Expert
System CFAR. Dr. Ozturk’s research produced a mathematical technique
now known as Ozturk’s algorithm that focuses on the use of
very small data sets (i.e., minimal sample support) to estimate PDFs
accurately, including an accurate estimate of statistical tails [10].
That first Expert System CFAR was conceptually very simple,
with several different OS-CFAR and TM-CFAR detectors (with
any individual TM-CFAR detector using very few training data–5
to 10 at most) all running in parallel, each processing the same
radar data, with CA-CFAR as a baseline for comparison. Training
data sets of various sizes were employed, and analyzed via
a statistical bootstrap technique (repetitive random sampling with
replacement) for estimating the mean clutter-plus-noise power in
order to provide an improved estimate of the interference in each
cell under test. Initially, fusion was accomplished via majority rule
polling of the various detector outputs. Other, more sophisticated
fusion rules, as well as Ozturk’s algorithm, were later applied.
Many signal-processing experts told us not to pursue this line
of reasoning because the data being processed and fused were
100% dependent, and as such, results would be disappointing.
However, their analysis proved erroneous due to one simple reason.
Any one CFAR detector algorithm is matched to the environment
only under very restrictive conditions, and these conditions
are rarely met. We were fusing the results of several suboptimum,
albeit real-world detectors, and as such, fully expected to improve
overall performance.
TECHNOLOGY DEVELOPMENT AND DEMONSTR ATION
Rome Laboratory supplanted RADC upon its establishment in the
early 1990s. About this same time, the second development stage
of this technology began. The second author, chief architect and
developer of the fielded product we now call Expert System CFAR,
led this effort [11]. His team included a number of investigators,
including Dr. Paul Antonik and Dr. Gerard Capraro, researchers investigating
multiple fields: real-time processing, parametric analysis,
modeling and simulation, and software engineering, respectively.
Dr. Varshney and Dr. Weiner from Syracuse University also
formally joined the research team, as did Dr. Murali Rangaswamy
[12]. In this effort, we developed an extensive rule-based approach
to enhance performance of the aforementioned fusion processor.
In the final product, only four CFAR algorithms were selected.
They were CA, GO, OS, and TM-CFAR; the first three detectors
are all examples of TM- CFAR. With no trimming, we obtain CACFAR.
With maximum trimming (all but one sample), we obtain
OS-CFAR. In addition, with range averaging before trimming (of
the weakest training data), we obtain GO-CFAR. Figure 3 depicts
an early flow diagram of the prototype Expert System CFAR processor
developed using FORTRAN and Gensym’s G2.2
Mathematically SO-CFAR employs as training data the alternative
set resulting from the logical “and” union of CA-CFAR and
GO-CFAR training data. Through these and similar mathematical
and philosophical constructs, dozens of CFAR algorithms and hundreds
of rules were reduced to a few detector algorithms, control
parameter sets, and heuristic rules, all without significant performance
degradation.
With surface maps (surface elevation data, surface cover data
[e.g. trees, grass], and cultural data [e.g. dense urban]), and only
11 rules and four CFAR detectors, a robust Expert System CFAR
was developed, and performance demonstrated using measured
AWACS data. This data, containing modern air-breathing targets
embedded in clutter, was analyzed using a variety of competing
CFAR approaches. Expert System CFAR was demonstrated to be
genuinely superior [13]. Performance margins of several dB are
reported in the literature. See the references for more details [14],
[15]. In clutter-free environments (e.g. Gaussian white noise), Expert
System CFAR performs as well as CA-CFAR due to the development
of a robust rule-base.
TECHNOLOGY TR ANSITION
The third and final stage of this effort was transition. This technology
was successfully transitioned to the user community in the
1990s. Two classes of surveillance radar were considered as transition
opportunities: initially, the E-3 AWACS airborne early warning
radar followed by the E-8 JSTARS air-to-ground imaging and
target tracking radar.
The CFAR processor used by AWACS was implemented as
a baseline for comparison to Expert System CFAR. Radar data
collected by the E-3 AWACS was analyzed using Expert System
CFAR processing and compared with the AWACS baseline (Figure
4). On a per coherent processing interval (CPI) basis, Expert System
CFAR demonstrated an order-of-magnitude reduction in false
alarms and improved detection in difficult and cluttered environments.
This represented a sensitivity improvement typically asso-
2 See www.gensym.com/products.
SEPTEMBER 2017 IEEE A&E SYSTEMS MAGAZINE 43
aesm-32-09-03 Page 43 PDF Created: 2017-10-10: 8:37:AM
Expert System CF AR
ciated with significant increases in power-aperture product. It was
instead accomplished with advanced signal processing consisting
of characterization of the underlying interference, knowledge of
the conditions where each CFAR algorithm performs optimally,
and smart but simple rules to guide in their selection. This signal
processing technique, simple in nature, produced an average performance
increase of approximately 7 dB!
By the mid-1990s, the Air Force had reorganized its four laboratories
into a consolidated organization known as the Air Force
Research Laboratory (AFRL), which remains in existence today.
The research continued within the Sensors
Directorate of AFRL into the early
2000s when Dr. Joseph Guerci, a respected
colleague, initiated technology
development at Defense Advanced
Research Projects Agency (DARPA)
under a program known as KASSPER
(Knowledge-Aided Sensor Signal Processing
and Expert Reasoning). This
program sought to maximize the payoff
from heuristic signal and data processing,
and especially to formalize the architectures
needed to field this important
technology widely and affordably.
Dr. Guerci, along with the authors and
their colleagues, integrated this technology
with waveform diversity, track
processing, and parameter estimation.
The second author elected to manage the
KASSPER program and to focus on two
key areas of intended accomplishment.
First, he sought to develop a test bed in
order to integrate various knowledge-
Figure 3.
Prototype Expert System CFAR detection processor.
Figure 4.
Expert System CFAR results using measured E-3 AWACS radar data.
44 IEEE A&E SYSTEMS MAGAZINE SEPTEMBER 2017
aesm-32-09-03 Page 44 PDF Created: 2017-10-10: 8:37:AM
Wicks and Baldygo
aided (KA) algorithms for filtering, detection, and track processing;
databases and “knowledge sources” to aid in the understanding
of the environment; sources of measured and simulated datasets to
be used in research, assessment, and demonstration; metrics tools
for assessing performance; and visualization of results [16]. Second,
building on the first key area, he sought to demonstrate performance
enhancements to the E-8 JSTARS using KA techniques,
analogous to the E-3 AWACS results demonstrated a decade earlier.
The SPEAR test bed (Signal Processing Evaluation, Analysis,
and Research) was built and is illustrated in Figure 5 [17]. The
humble and simple beginnings of the Expert System CFAR processor
in 1984 gave rise to an entirely new way of approaching sensor
signal and data processing. As shown in the figure, numerous
innovations and over two decades of development were brought
together in this test bed—filtering, especially STAP, detection,
and track algorithms; modeling and simulation (M&S) tools; data
sets (simulated and measured); knowledge sources; and tools for
measures of performance (MOPS). The contributions of Mr. Todd
Cushman and Mr. Mark Novak from AFRL and Mr. Walter Szczepanski,
Mr. Robert Bozek, and Mr. Jeff Tyler of Black River Systems
Company were critical in implementing the SPEAR test bed.
An important aspect of the aforementioned MOPS tool is the linkage
it created between signal and data processing enhancements,
typically reported in terms such as test statistics, PDF, SINR, SINR
loss, or Pd, and Pfa, and corresponding tracking and exploitation
metrics more commonly employed and understood by Air Force
users and operators. The ability to “translate” and articulate signal
processing enhancements in terms more readily interpreted by operators
was a major factor in transitioning Expert System CFAR
and subsequent KA signal and data processing technology to the
Air Force and the rest of the user community. Concurrently, Mr.
Jon Jones of AFRL’s Information Directorate was very influential
in shaping the MOPS tool, elevating track processing, and especially
raising data exploitation technology to a much higher level
of sophistication and acceptance by the user community.
The SPEAR test bed enabled a series of performance assessments
to be conducted, including Expert System CFAR for
JSTARS, as well as various configurations of KA processing for
Global Hawk and other candidate transition platforms. Performance
improvements using Expert System CFAR were indeed
demonstrated for JSTARS using real-world measured data in heterogeneous
clutter environments (mountains/desert) with several
slowly moving, closely spaced ground targets (analysis and demonstration
of results presented to and lauded by the 2002 US Air
Figure 5.
AFRL Sensors Directorate SPEAR test bed.
Figure 6.
Expert System CFAR and technology development timeline.
SEPTEMBER 2017 IEEE A&E SYSTEMS MAGAZINE 45
aesm-32-09-03 Page 45 PDF Created: 2017-10-10: 8:37:AM
Expert System CF AR
Force Scientific Advisory Board). Subsequent embracing and maturation
by DARPA resulted in transition of Expert System CFAR
to JSTARS in the Block 40 tracker upgrade.
FUTURE RESEARCH DIRECTIONS
The research and development of Expert System CFAR began
over 30 years ago and built from a solid foundation across many
disciplines and into many research domains. This can be appreciated
in Figure 6. It has shaped a way of thinking about signal and
data processing that has changed the nature of US Department of
Defense (DOD) radar technology for adaptive filtering, detection
processing, as well as track formation, target identification, and
engagement [18].
As the Air Force pursues a vision for layered sensing, Expert
System CFAR and subsequent developments in knowledge-based
STAP and KA processing provided a compelling motivation to
extend the paradigm further into waveform agile, multiplatform,
multisensor, and multidomain instantiations of layered sensing in
order to meet future warfighting needs in complex environments.
Analysis of multichannel airborne radar data demonstrates the impact
of Expert Systems and knowledge-based processing on layered
sensing. In the upper and lower right of Figure 7, the impact
of heuristic processing on STAP is demonstrated via “Before” and
“After” results. In this graphic, knowledge-based STAP includes
nonhomogeneity detection [19] (analogous to excising outliers in
TM-CFAR), the application of corrected (measured) steering vectors,
the exploitation of prior knowledge sources and data from
previous flights in heuristic and algorithmic radar signal and data
processing for target detection, and interference rejection in airborne
radar.
ACKNOWLEDGMENT
The authors wish to acknowledge the management and leadership
of the US Air Force and DARPA for making this research, development,
and technology transition possible, especially Dr. Schneible
and Dr. Guerci.
REFERENCES
[1] Smith, J., and Byrd, D. Forty years of research and development at
Griffiss Air Force Base. Rome Laboratory, Griffiss AFB, NY, Tech.
Rep. RL-TR-92-25, AD-A250 435, July 1992.
Figure 7.
Expert System CFAR led to knowledge-based STAP [20].
46 IEEE A&E SYSTEMS MAGAZINE SEPTEMBER 2017
aesm-32-09-03 Page 46 PDF Created: 2017-10-10: 8:37:AM
Wicks and Baldygo
[2] Weiner, D., and Song, J. Improvements in radar detection under
scan to scan processing and scan rate reduction. Rome Laboratory,
Griffiss AFB, NY, Tech. Rep. RL-TR-91-80, AD-A247-350, June
1991.
[3] Silfer, C. et al. Proceedings of the RADC Spectrum Estimation
Workshop, Rome Air Development Center, Griffiss AFB, NY, Oct.
1979.
[4] DiDomizio, J., et al. Track-before-detect development and demonstration
program. Rome Air Development Center, Griffiss AFB, NY, Int.
Tech. Rep., Mar. 1985.
[5] Wicks, M. et al. Clutter considerations in modern radar systems.
Rome Air Development Center, Griffiss AFB, NY, Tech. Rep. RADCTR
88-252, Vol. 2, Oct. 1988.
[6] Rohling, H. Radar CFAR thresholding in clutter and multiple target
situations. IEEE Transactions on Aerospace and Electronic Systems,
Vol. AES-19, 4 (July 1983), 608–621.
[7] Dechant, D. The advanced onboard signal processor. Journal of VLSI
Signal Processing Systems, Vol. 2, 2 (Oct 1990).
[8] Willet, P. et al. The case for like-sensor pre-detection fusion. IEEE
Transactions on Aerospace and Electronic Systems, Vol. AES-30, 4
(July 1994).
[9] Brown, R., and Wang, H. Adaptive multiband polarization processing
for surveillance radar. In 1991 IEEE International Conference on
Systems Engineering, Dayton, OH, Aug. 1991.
[10] Weiner, M. (Ed.). Adaptive Antennas and Receivers. London: CRC
Press, 2005.
[11] An expert system constant false alarm rate (ES-CFAR) processor, U.S.
Patent 5,499,030, Apr. 1996.
[12] Antonik, P. et al. ES-CFAR reports intelligent use of CFAR algorithms.
Rome Laboratory, Griffiss AFB, NY, Tech. Rep. RL-TR-93-75, ADA267-
755, June 1991.
[13] Baldygo, W. et al., Artificial intelligence applications to constant false
alarm rate (CFAR) processing. In Record of the 1993 IEEE National
Radar Conference, Lynnfield, MA, Apr. 1993.
[14] Wicks, M. et al. Space-time adaptive processing: A knowledge-based
perspective for airborne radar. IEEE Signal Processing Magazine, Special
Issue on Knowledge-Based Systems, Vol. 23, 1 (Jan. 2006), 18–29.
[15] Capraro, G. et al. Knowledge-based radar signal and data processing.
IEEE Signal Processing Magazine, Special Issue on Knowledge-
Based Systems, Vol. 23, 1 (Jan. 2006), 18–29.
[16] Guerci, J. et al. The KASSPER eye-watering scenario. In Proceedings
of the 2004 DARPA/AFRL KASSPER Workshop, Clearwater Beach,
FL, Apr. 2004.
[17] Baldygo, W., et al. AFRL KASSPER algorithm test bed. In Proceedings
of the 2002 DARPA/AFRL KASSPER Workshop, Washington,
D.C., Apr. 2002.
[18] Baldygo, W. et al. Sensors technology and advanced signal processing
concepts for layered warfare/layered sensing. Air Force Research
Laboratory, Wright Patterson AFB, OH, Tech. Rep. AFRL-RY-RSTR-
2010-100, AD-A519-954, Apr. 2010.
[19] Wicks, M. et al. An efficient architecture for non-homogeneity detection
in space-time adaptive processing airborne early warning radar. In
Proceedings of the 1997 International Radar Conference, Edinburgh,
United Kingdom, Oct. 1997.
[20] Wicks, M. Knowledge-based control for space time adaptive processing.
Paris, France, Tech. Rep. RTO-EN-SET-063, Sept. 2006.
SEPTEMBER 2017 IEEE A&E SYSTEMS MAGAZINE 47
aesm-32-09-03 Page 47 PDF Created: 2017-10-10: 8:37:AM
