Coastal Ocean Modeling and Observation Program

Document Sample
Coastal Ocean Modeling and Observation Program Powered By Docstoc
					                               Annual Report
         Development and Demonstration of Shiptracking Capabilities
         for a Dual-Use Multi-Static Long-Range HF Radar Network

                                   Scott M. Glenn
                                  Rutgers University
                                   71 Dudley Road
                              New Brunswick, NJ 08901
       Phone: 732-932-6555 x506 Fax: 732-932-1821 email:

                                   Donald E. Barrick
                               CODAR Ocean Sensors, Ltd.
                             1000 Fremont Avenue, Suite 145
                                  Los Altos, CA 94024
             Phone: 408-773-8240 Fax: 408-773-0514 email:

                                 William J. Browning
                               Applied Mathematics, Inc.
                                1622 Route 12, Box 637
                                 Gales Ferry, CT 06335
            Phone: 860-464-7259 Fax: 860-464-6036 email:

                              Award Number: N00014-02-10917
                                    September 30, 2004


Develop a prototype dual-use HF Radar network that can be expanded to encircle most of the
U.S. coast and is capable of providing both real-time surface current fields and ship tracks to a
variety of users.


The specific objective of this project is to develop and demonstrate a ship-tracking capability
for large vessels within an existing multi-static CODAR HF Radar network deployed along the
New Jersey coast.


Rutgers University now operates a nested multi-static CODAR HF Radar network in the New
York Bight. The network was developed, installed and validated for current mapping in an
ONR-sponsored collaborative project with CODAR Ocean Sensors in 2000-2002. This
project expands the successful partnership between Rutgers University (RU) and CODAR
Ocean Sensors (COS) through the addition of Applied Mathematics, Inc (AMI), a firm with
extensive experience in the development of submarine tracking algorithms for the U.S. and
British Navies.

Short Report

The project is divided into three major tasks to be implemented over two years:
    1. Develop and demonstrate a ship detection and tracking capability for multiple
monostatic CODAR sites using the simpler radial geometry of the more common backscatter
HF radar systems.
    2. Develop and demonstrate a ship detection and tracking capability for bistatic CODAR
sites that includes the more complicated elliptical geometry of separated transmitters and
    3. Multi-static demonstration of the evolving capability in the Rutgers CODAR network.

Schedule: The schedule calls for Task 1 to be the focus of Year 1, Task 2 to begin concurrently
but extend well into Year 2, and Task 3 to take place at the end of Year 2.

Key Individuals and Expertise:
Scott Glenn, RU, Multistatic HF Radar Array Operation
Don Barrick, COS, HF Radar Ship Detection Software Development & Testing
Bill Browning, AMI, Ship Tracking Software Development & Testing


      CODAR System upgrades, operation and vessel tests. Year 1 vessel tracking tests (both
ships of opportunity and dedicated) used multiple vessels to test different signal transmission
options (waveforms, pulse repetition rates, range cell size, etc) to determine the best
transmitter characteristics for dual use (simultaneous current mapping and vessel tracking).
The received signals are processed with two FFTs, the first to convert the raw time series into
separate files for each range cell, and the second to calculate the Doppler spectra within each of
the resulting range cells. In year 1, it was determined that the current mapping and vessel
tracking processing diverged immediately after the first FFT conversion to range cell data.
Thus, during Year 2, transmitter settings were updated, the GPS timing was adjusted for multi-
static data collection from all four long-range sites and both standard sites, and data archiving
was switched to the earlier range files to enable multiple length FFT post-processing for both
current mapping and vessel tracking. Antenna beam patterns were measured at all sites using a
transponder mounted on a boat for the over water segments, and a procedure to measure the
beam pattern over the land segment was tested. Over 7 months of multi-static data at the range
file level were collected and archived. Year 2 vessel tracking tests concentrated on a long-term
study of a single known ship, the M/V Oleander, on its weekly cruises between New York and
Bermuda. GPS tracks obtained from the University of Rhode Island were used to identify
times when the Oleander was within 200 km of Sandy Hook for detection processing. Surface
wave data were obtained from NOAA.
      Detection software development and testing. Archived range files during the Oleander
transits were processed into Doppler spectra using multiple length FFT software developed for
this application. Two techniques were developed to estimate the background noise floor in the
Doppler spectra, an Infinite Impulse Response filter that averages each point back in time and a
2-D median filter that averages in space at a specific time. Two automated peak picking
algorithms were developed to identify the ship peaks that exceed the noise floor in the Doppler
spectrum of each range bin; the first technique identifies a peak in both the monopole and at

Short Report

least one of the cross-loop receive antennas, while the second new technique combines the
three receive antenna signals to form a beam using eigenfunction analysis on their measured
patterns, much like a phased array. Several thresholds (in dB) for the peak above the
background noise were tested. Once a peak was identified, the spectral data from the monopole
and both loops were passed through the MUSIC algorithm to estimate the bearing with both
idealized and measured beam patterns. Uncertainty estimates for the resulting range, range rate
and bearing vessel detections were developed. Detections initially obtained for monostatic
backscatter data were extended to the more geometrically complicated bistatic data to produce
the associated bistatic distance, velocity and angle estimates. All peaks detected by the
automated algorithms were plotted as time series of range, rage rate and bearing as well as
spatial “pepper plots” maps were produced and compared to GPS tracks. The detection time
series and pepper plots were used to determine and reduce external noise induced false alarms
in the detections. Detections for the vessel of interest were associated by eye and assembled
into a time series of detections for each specific vessel of interest. In addition, the radar cross-
section for the Oleander was estimated to aid in the future development of an automated
association process.
      Tracking software development and testing. A constant course and speed vessel track
model that is fit to the detection time series (range, range rate, bearing and their associated
uncertainty estimates) with a Kalman filter was developed. The model includes a chi-squared
test for a potential maneuver to a second constant course and speed track. Tracking results
were compared to GPS tracks for known vessels, both as time series of range, range rate and
bearing and on track maps. The initial monostatic backscatter tracker was further modified to
accept bistatic detection data in the more complicated elliptical geometry. The tracker was
tested under numerous conditions that include single monostatic, multiple monostatic, single
bistatic, and multistatic cases. In the case of multiple radars holding contact on the same
vessel, a least squares fitting procedure to estimate bearings from the intersection of a pair of
range rings was developed. The ability to use this procedure to correct for bearing bias was


     1) The Frequency Modulated Continuous Wave (FMCW) transmitted signal format
enables the construction of waveforms with 50% on/off characteristics (i.e., the optimum for
backscatter), but with variable pulsing periods. This can enhance the detectability of certain
classes of vessels at ranges where they would normally not be seen by increasing their SNR.
These digitally synthesized waveforms are easily installed and activated remotely. For
example, the standard transmitter pulse repetition rate for current mapping results in very little
SNR near shore. Tests indicate that vessels are easier to detect near shore when the pulse
repetition rate is increased, but this results in multiple blind zones in the current mapping
region. A new unevenly pulsed transmit waveform was found to increase SNR nearshore
without creating blind zones and with only about a 20 km reduction in overall current mapping
range. The waveform is used for close in tracking of small boats when desired.
     2) Backscatter time series typically generated at the rate of 168 MB/hour can undergo the
exact same range cell processing for vessels as they do for currents (early tests indicated no
change in range cell size was required). The range files now archived for vessel detection are
much smaller but still generate about 2 MB/hour. One hour of range file data requires over 2
hours of clock time to transfer over the standard dial-up modem phone lines found at most

Short Report

remote shore sites. Faster cable modems require close to a half hour, and a dedicated T1
requires over 2 minutes. Since T1 access to shore sites is rare, communication between
traditional backscatter radars is a significant network design issue.
     3) Simultaneous multiple FFT lengths to calculate Doppler spectra within each range cell
aids detection with the optimal length varying in time and by vessel type. For larger vessels,
128-512 second time series appear best for identifying the ship Doppler peak. For smaller go-
fasts whose speeds are less constant, 64-128 seconds are best for identifying the peak. The
advantages of calculating multiple FFT lengths on the range files highlights this as a key
network design issue for real-time applications.
     4) The background noise floor is found to be quite variable at HF, complicating the
Doppler peak detection process. The Infinite Impulse Response filter works best for slow
variations over the time period used for forming the temporal background average. The spatial
median filter works best for impulsive noise or radio interference of short duration. As with
multiple FFTs lengths, multiple background filters also appears to be a desirable design
     5) The best threshold for the detection of ship Doppler peaks appears to lie in the range 6-
9 dB above the background for the monopole signal and one (of the two) loop signals. This
admits a fair number of false alarms, but the subsequent association and tracking processor
should eliminate most of these. Combining all three receive antenna signals through
eigenfunction analysis yields the equivalent of 3-4 dB higher SNR for detection, a significant
     6) Antenna patterns matter. Measured patterns work better than ideal. This was checked
using measured and ideal patterns along with simulations and real echo data. Angle error goes
down inversely with sqrt(SNR) to zero, unless an inappropriate antenna pattern is used (e.g.,
ideal when the measured pattern shows distortion). Then angle error goes down to some
irreducible bias, and increasing SNR no longer helps.
     7) Single-angle MUSIC direction finding among the three antenna elements using
measured patterns gives more robust bearing determinations than the double-angle algorithm.
Existing detection and tracking software has identified a potential test case for dual-angle
MUSIC in which two ships appear to be approaching New York Harbor in different shipping
lanes but occasionally are passing through the same range cell at the same speed.
     8) After initial experimentation, we have found and employ good and useful uncertainty
estimates for the radar observables: range, range rate, and bearing. These reflect the statistical
uncertainties, but not biases; the latter must be identified and removed.
     9) When two radars hold contact on a single vessel, least squares fits to the intersection of
the range rings produce excellent fits to the GPS bearings, enabling a bearing bias correction to
be made if required.
     10) A significant interference problem was identified and mitigated. When strong radio
station signals are received from the back side (i.e., over land) where the antenna pattern had
not been previously measured, they lead to a confusingly large number of false alarm
detections widely scattered in range and Doppler that exceed the threshold. These have the
property that their MUSIC-produced bearings "appear" to come from the sea, but are clustered
around two or three discrete, false bearings. By using the MUSIC orthogonality measure
between "signal" and "noise" as a discriminator, nearly 95% of these interference-based false
alarms are removed.

Short Report

     11) Bistatic and multi-station looks at the same target (the Oleander) have been
confirmed. This leads to more robust tracks, and provides data to fill detection gaps when the
vessel echo falls on the strong Bragg sea-echo clutter peak at one of the sites. This confirms a
significant feature of a distributed radar approach.
     12) Detection time series and pepper plots indicate that the New York Harbor entrance is
a target-rich environment where many obvious ship echoes are seen simultaneously by multiple
radars. At any given time at Sandy Hook, for example, typically five or six vessels are
detected, and their detection trails can last for several hours and to distances beyond 100 km.
Association of these detections with a specific ship for track fitting is accomplished by eye.
     13) The RCS (radar cross section) of a detected vessel has been extracted. Backing this
quantity out of the Radar Equation was a major goal and a significant challenge. This target
information is an important aid in classification, i.e., identifying the nature of the target. The
RCS was extracted for the Oleander at various aspect angles, distances, and even for bistatic
     14) The tracker tested here does an excellent job of converting the observed detections
and their associated uncertainties into a vessel track. As expected by theory, the model fit track
is more accurate than the detection time series going into it. Similar to historical performance
in other applications, once the tracker locks on a vessel track, it continues to maintain a good
track even through noisier detection data. Use of multiple radars on the same target aid the
Kalman filter in establishing the initial track.
     15) All of the factors in the Radar Equation have been confirmed based on CODAR
system parameters and detection geometries, enabling system performance and tradeoff
studies. For example, if one were to increase the transmit power by X dB, or the receive
antenna directive gain by Y dB, or use a different signal waveform, the added surveillance
coverage expected for a target of given RCS could be calculated.


      The ability to detect and track ships with a compact multistatic coastal HF radar network
while it is simultaneously being used to map surface currents for Coast Guard sponsored
Search and Rescue tests has been demonstrated. The work has identified clear directions for
future work, several of which are already being actively pursued as indicated below. In
addition, studies of false-alarm rate vs. probability of detection are required. Although this
field is well developed for microwave radars where noise is internal and Gaussian, it is new
territory at HF where much of the noise is radio interference and/or impulsive (lightning), and
hence not Gaussian. An integration time detection model and a 'track before detect' option on
less thresholded data would help to further reduce the false alarm rate. Furthermore,
association procedures for a multi-ship environment that require development will be aided by
a better understanding of the bistatic RCS as we move to a fully automated real-time system.
Many of these options can be tested with the existing 2004 dataset which is now being archived
and distributed to all project participants.


    The Coast Guard is conducting an operational test of the Search And Rescue Optimal
Planning System (SAROPS) in November 2004 using the Rutgers CODAR surface current
maps as input to the University of Connecticut Short Term Prediction System. Real-time

Short Report

predictions will be validated with several clusters of three buoys deployed by the Coast Guard
R&D Center in the Rutgers CODAR grid.


     The multistatic CODAR array used here was developed through the ONR "Development
and Demonstration of Bistatic and Long-Range CODAR SeaSonde HF-Radar Systems" project
from 2000-2002.
      The Department of Homeland Security Advanced Research Project Office has organized
a panel of national experts coordinated by the Coast Guard to provide recommendations for
developing over-the-horizon wide area surveillance technologies. Members of the panel have
individually funded add-on work, or are now pursuing initial follow-on work, to this project
that includes (1) similar tracking tests of small fast vessels (already reported in Glenn et al.,
2003), (2) continued testing of the SIFTER vessel detection program by Mission Research
Corporation using data from this project, (3) construction and testing of a super-directive
receive antenna by CODAR to determine the additional antenna gain achieved over the
existing cross-loop configuration, (4) continuation of the long-term monitoring of the highly
variable HF background begun in this project by Rutgers, and (5) use of the existing ONR-
sponsored bistatic buoys to measure bistatic radar cross-sections similar to the Oleander but for
a small, fast vessel.
    Surface current mapping results from this network are being used by the National Science
Foundation Lagrangian Transport and Transformation Experiment for the Hudson River
Plume. Proposed future ONR projects that will use this network include data assimilation in
numerical forecast models, further optimization of the current mapping ability, and
environmental background data for the planned 2006 Shallow Water Acoustic Experiment.


Barrick, D.E., Bearing accuracy against hard targets with SeaSonde DF antennas, CODAR
      Ocean Sensors Report,
Glenn, S.M., D.E. Barrick, and W.J. Browning, Final Report to DoD Counter Narco-Terrorism
      Technology Development Office, Dec., 2003.
Kung, P. Background noise level calculation for CODAR ship detection, CODAR Ocean
      Sensors Report,
Kohut, J. T. and S. M. Glenn, 2003. Calibration of HF radar surface current measurements
      using measured antenna beam patterns. J. Atmos. Ocean. Tech., 1303-1316.
Kohut, J.T., H.J. Roarty, and S.M. Glenn. 2004. Characterizing environmental variability with
      HF Doppler radar surface current mappers and acoustic Doppler current profilers. IEEE
      Journal of Oceanic Engineering, submitted.