SUMMARY
Passive radar systems that exploit illuminators of opportunity, such as FM radio and television broadcasts, to
detect and track airborne targets have been under development for over a decade. This paper reviews efforts
to add radar imaging and target recognition capabilities to such systems. We discuss recent developments
along two parallel threads:
1) Target recognition via radar cross section (RCS) profiles: In this approach, databases of the RCS of
targets at different incident and observed angles are created using method-of-moments computational
electromagnetics codes. The extracted RCS profiles for different targets, scaled to account for antenna
patterns and atmospheric propagation, are compared to the collected data. A coordinated flight model is used
to estimate the aircraft's orientation along its flight path. The low frequencies used in passive radar naturally
give stable features well suited for automatic target recognition.
2) Radar imaging: A traditional inverse synthetic aperture approach to forming images with passive radar
data results in severe artefacts due to the sparse and irregular Fourier sampling patterns resulting from
realistic data collection scenarios. We review the application of a recent optimization-based, regionenhancing
imaging algorithm to passive radar imaging that effectively suppresses these artefacts, and
illustrate the difficulties posed by the underlying multidimensional autofocus problem.
1.0 INTRODUCTION
Traditional active radar systems transmit waveforms and deduce information about targets by measuring and
analyzing the reflected signals. A radically different approach to radar arises when we consider that modern
civilization is already drenched in transmissions such as FM radio, television, and cell phone signals. Passive
radar systems that “hitchhike” off of such existing “illuminators of opportunity” remain covert compared to
their active brethren. Other covert sensors, like as ESM sensors employing multilateration,1 are available, but
they rely on the assumption that the objects of interest are broadcasting and don’t mind announcing their
presence. PCL sensors require no such assumption. We save on the cost of building a transmitter, since
another party has already gone through the trouble. However, communication signals were not designed with
radar applications in mind. The cost of the radar system then shifts from traditional radar hardware to the
digital signal processing know-how and horsepower required to make sense of the received signals. The price
of radar hardware remains relatively fixed, while the cost of computational power continues to plummet.
Passive radar can thus boast something active radar cannot: its further development is primarily driven by
Moore’s law. The passive radar approach has often been referred to as PCL (Passive Coherent Location.) To
our knowledge, the term PCL was first coined by Dick Lodwig and colleagues at what was then IBM, later
Loral, and currently Lockheed Martin. The term PCL is closely tied with Lockheed Martin’s Silent Sentry
series, of which Silent Sentry 3 is the latest incarnation, although the acronym PCL has evolved to refer to
passive radar systems in general. The term PCR (either “Passive Coherent Radar” or “Passive Covert Radar”)
has also become popular. Interest in PCR has skyrocketed in the past five years. International conferences on
PCR were hosted by Roke Manor in the United Kingdom in June 2002, and by the University of Washington
in Seattle in the United States in October 2003. At the time of writing, the IEE Proceedings Radar, Sonar, and
Navigation is preparing a special issue devoted to PCR guest edited by Paul Howland (NATO C3 Agency)
and Paul Gilgallon (U.S. Air Force Research Lab).
A three-year DARPA-sponsored program at the Univ. of Illinois at Urbana-Champaign on the “Design and
Optimization of Passive and Active Imaging Radar” began in the Fall of 1998. A prime thrust of this effort
was the development of ways to add automatic target recognition (ATR) and radar imaging components to
passive radar systems. Follow-up efforts (some in collaboration with colleagues at MIT and Univ. of
Michigan Ann Arbor) have continued at Georgia Tech, sponsored by NATO NC3A, AFOSR, startup funds
from Georgia Tech’s School of Electrical and Computer Engineering, and the Demetrius T. Paris
Professorship. This paper briefly reviews some of this work, and points out some avenues for further research.
2.0 AUTOMATIC TARGET RECOGNITION
One approach to target identification compares the collected data to target libraries synthesized using
electromagnetic codes. To ensure robust classification in the presence of noise and errors in estimates of
position and orientation, it is helpful if the Radar Cross Section (RCS) of the targets vary “slowly” with small
changes in these components of the state vector. The variation in RCS, as characterized by the number of nulls
encountered as a target's aspect changes, is proportional to the electrical length of the target. At FM-band
frequencies (100 MHz), a fighter-sized aircraft is approximately five wavelengths long. In contrast, at the Xband
frequencies used by many active radars (10 GHz), the same aircraft would be 500 wavelengths long.
In the late 70's and early 80's, a series of papers [1-3] illustrated that low frequencies are quite natural for
target classification. Those papers had active radars in mind, but low-frequency radar did not catch on in the
West since most of the desired spectrum has been allocated to communications. PCR systems, on the other
hand, which directly exploit existing long wavelength emissions that are convenient for target recognition,
circumvent the frequency allocation problem faced by active radars.
2.1 Joint Tracking/Recognition with Particle Filters
Target tracking and target recognition are generally considered to be separate tasks. In particular, target
tracking algorithms generally track two-dimensional or three-dimensional target positions in Cartesian
coordinates via simple constant velocity or constant acceleration models; target orientation is generally not
directly accounted for. The notion that tracking and recognition algorithms could help one another dates back
to work by Sworder and colleagues [4-6]; in particular, the authors suggest using imaging information to
detect manoeuvre changes. Sensors detailed enough to provide target recognition data generally can also
provide orientation; in fact, orientation must often be estimated as a nuisance parameter. The orientation and
position of aircraft paths are clearly coupled. Miller, Srivastava, and Grenander [7] suggest fusing the
recognition and tracking tasks into a single joint estimation problem.

Download full report
http://ftp.rta.nato.int/public//PubFullT...080-24.pdf