LVDS Optimization, Methods and Results

Document No: / RU-LEAP-025 / Page 1 of 10
Title: / LVDS Optimization, Methods and Results
Revision: / A

A / HJR / First Draft / 27 June, 2011
B / HJR / Post Downie comments / 12 Dec 2011
Rev / By / Description / CN / Date Approved

Contents

1.1Introduction

1.2Methods and Results

1.3Discussion

1.3.1Background

1.3.2Integration Time

1.3.3Threshold Level

1.4Conclusion

1.1Introduction...... 3

1.2Methods and Results...... 3

1.3Discussion...... 6

1.3.1Background...... 6

1.3.2Integration Time...... 6

1.3.3Threshold Level...... 7

LIST of Figures

Figure 1: Detection rate as a function of threshold and integration time for the YM Los Angeles using the infinite impulse response (IIR) and median background. The numbers next to the data points represent the FFT points used in the averaging.

Figure 2: Detection rate as a function of integration time for all data. Each data point represents the detection rates for the tests of the vessels shown in Table 2

Figure 3: Average detection rate as a function of integration time separated for the IIR and Median backgrounds

Figure 4: Results of sensitivity study of the YM Los Angeles Detection test. The panels on the top are for the IIR background detections per minute (left) and detection rate (right) and the panels along the bottom are for the median background detections per minute (left) and detection rate (right). The different colors represent the FFT length and the x axis in all plots is the threshold set in the detection software.

Figure 2: Detection rate as a function of integration time for all data...... 5

Figure 3: Average detection rate as a function of integration time separated for the IIR and Median backgrounds 6

Figure 4...... 87

List of Tables

Table 1: Summary of vessel detection exercises conducted under the OPT and DHS programs

Table 2: List of vessels used in this study

Table 3: List of FFT points and thresholds (dB) used in vessel detection processing

Table 4: Variation of radial Doppler resolution with integration time

Table 1: Summary of vessel detection exercises conducted under the OPT and DHS programs 3

Table 2: List of vessels used in this study...... 4

Table 3: List of FFT points and thresholds (dB) used in vessel detection processing...... 4

Table 4: Variation of radial Doppler resolution with integration time...... 7

1.1Introduction

This document defines the Rutgers and Codar contribution to the LEAP Vessel Detection System (LVDS) Optimization Report as part of the Littoral Expeditionary PowerBuoy Platform (LEAP) program.

The vessel detection research that is taking place under the LEAP program is closely aligned with the research that is being performed as part of the Department of Homeland Security Center of Excellence that Rutgers and CODAR have been contributing to since 2008. We are taking the opportunity with this report to utilize all the vessel detection tests that have been conducted under these two programs to determine the optimal threshold level, background type and integration time that will enable effective vessel detection within the New Jersey test bed. The list of vessel detection exercises is summarized in Table 1.

Table 1: Summary of vessel detection exercises conducted under the OPT and DHS programs

Sponsor / Date / Detection Type / Location / Frequency
DHS / February 26, 2009 / Monostatic / Sea Bright, NJ / 13
DHS / November 9, 2009 / Monostatic / Sea Bright, NJ / 13
OPT / May 4, 2010 / Monostatic and Bistatic / Brant Beach, NJ / 13
OPT / May 30, 2010 / Monostatic and Bistatic / Brant Beach, NJ / 13
OPT / June 22, 2010 / Monostatic / Brant Beach, NJ / 13
OPT / July 20, 2010 / Monostatic / Brant Beach, NJ / 13
DHS / April 6-13, 2011 / Monostatic / Miami, FL / 13
OPT / August 15 - / Monostatic and BistaticM/B/Multistatic / Sea Bright, NJ NJ / 13[BD1]

1.2Methods and Results

A total of seventeen vessels were used in this analysis. The information on the vessels is listed in Table 2. The average length of the vessels was190m with an average beam of 27m. The range data from the SeaSonde was run through the vessel detection software with the settings shown in Table 3. The radar was set to a 2 Hz sweep so the integration time in seconds is half the value of the FFT points.

The result of one vessel processing through the vessel detection software and association is shown in Figure 1. The detection rate is calculated by dividing the number of detections by the number of intervals where AIS data is present for the vessel. The length of an interval is 32 seconds, the update rate of the radar. For instance, the vessel used in Figure 1 is the YM Los Angeles which was detected by the Sea Bright radar station on November 9, 2009 from 06:30 to 07:45 UTC. That is a total of 75 minutes or 141 intervals (75 min * 60 s/min * 1 interval/32 s) over this time period. Using the median background, 128 point FFT and 10 dB threshold the Sea Bright radar made 66 detections of the YM Los Angeles for a detection rate of 47% (66/141). Extending this calculation to the other combinations of background, FFT length and threshold yields the plot in Figure 1.

These detection rates were compiled for the other vessels and are displayed in Figure 2. The data was then separated by background type and averaged for each integration time across all the vessels. The results of this are shown in Figure 3.

Table 2: List of vessels used in this study

Table 3: List of FFT points and thresholds (dB) used in vessel detection processing

IIR / Median
FFT / Threshold / FFT / Threshold
16 / 6 / 32 / 8
32 / 7 / 64 / 9
64 / 8 / 128 / 10
128 / 9 / 256 / 11
256 / 10 / 512 / 12
512 / 11 / 1024 / 13

[BD2]

[BD3]

Figure 2: Detection rate as a function of integration time for all data. Each data point represents the detection rates for the tests of the vessels shown in Table 2

Figure 3: Average detection rate as a function of integration time separated for the IIR and Median backgrounds

1.3Discussion

Because of the presence of false alarms and the association algorithms are in their infancy, we must take care to make the detection process as efficient as possible in order to make the vessel detection system operational. Once the association algorithms are mature, we can relax some of the settings on the system to allow more data and hopefully detections through to the end user. We will now discuss what are the optimal settings for the background type, integration time and the threshold level for the vessel detection system.

1.3.1Background

From Figure 3, the median background outperforms the infinite impulse response (IIR) background at all integration times. The use of the median background provides an extra 5% greater detection than the iir background. One aspect to consider is which background will give you better detections, but also less false alarms[BD4]. The test of this will be discussed in Section 1.3.3.

1.3.2Integration Time

From Figure 3, the highest detection rate is obtained with an integration time of 128 seconds (256 FFT points). Table 4 gives the Doppler bin resolution for a particular integration time. From Figure 3, the integration times that resulted in the highest detection rates were 128, 256 and 512 FFT points (64, 128 and 256 seconds respectively). If the radial velocity varies by more than the resolution of the integration time, the reflected energy from the vessel will migrate into adjacent Doppler bins. It is recommended that these three integration times (for the IIR and Median background) be used in the real time detection software. For all 17 vessels the 16 point FFT never provided a valid detection. It is recommended that this integration time not be used in future work. The 32 and 64 point FFT (16 and 32 second integration time) did provide some valid detections. These integration times will be useful when the vessels make speed or course changes and the shorter integration time can account for this in the spectra over the longer integration times. The 1024 point FFT was also able to make valid detections, but the false alarm rate was high with this integration time.

Table 4: Variation of radial Doppler resolution with integration time

Doppler Resolution
FFT / [BD5]vr (knots)
16 / 2.88
32 / 1.44
64 / 0.72
128 / 0.36
256 / 0.18
512 / 0.09
1024 / 0.05

1.3.3Threshold Level

As you lower the detection threshold you increase the number of false alarms and if you increase the detection threshold you decrease the number of false alarms but you also risk rejecting potentially valid vessel detection. Where do you strike the balance? Figure 4 presents the results from a sensitivity study that we conducted using the data from the YM Los Angeles (presented in Figure 1).

From Figure 4, look at the second tick from the left on all plots (for consistency); this is 7 dB threshold for IIR and 9 dB for Median. Best probability of detection (right panels of Figure 4) is about 57% for IIR and 66% for Median. In both cases, the best FFT length is between 128 & 256. At the second (threshold) tick for the left figures, this gives a false alarm rate (per minute) of about 0.2 for IIR and 0.6 for Median. So in summary of the difference between IIR and Median, for a difference in detection probability of 9% (57% for IIR vs. 66% for Median), you triple the false alarm rate for Median over IIR. This will be valuable information as we begin providing this data to end users. As we develop the Association Algorithms in this programthe this will enable us to use both IIR & Median, and different FFTs and thresholds simultaneously, allowing us to increase probability of detection somewhat (you never get to 100%), while drastically reducing false alarm rate.

It is well known from Gaussian target detection theory going back 60 years, that small percent changes in detection probability obtained by adjusting threshold and FFT parameters lead to much larger changes in false alarm rate, as affirmed here.

Figure 4: Results of sensitivity study of the YM Los Angeles Detection test. The panels on the top are for the IIR background detections per minute (left) and detection rate (right) and the panels along the bottom are for the median background detections per minute (left) and detection rate (right). The different colors represent the FFT length and the x axis in all plots is the threshold set in the detection software. Figure description here.

1.4Conclusion

The data for the existing DHS and LEAP programs was collected and summarized to determine the optimal settings for vessel detection. FFT lengths between 128 and 512 produced the greates number of detections. The median background gave better detection rates than the IIR background but that was accompanied by a higher false alarm rate. Feedback from the Naval Research Laboratory on the efficacy of the detections in fusion algorithms like Open Mongoose will help refine these initial findings. Section?

Proprietary Information

This document & information contained herein is the sole property of Rutgers University Coastal Ocean Observation Laboratory (RUCOOL),

& may not be reproduced or used for any purpose other than which it was intended without permission from RUCOOL.

RutgersUniversity

Coastal Ocean Observation Laboratory

71 Dudley Road, New BrunswickNJ08901

Unless otherwise specified this document is CONTROLLED only when viewed in the form of electronic media from a CONTROLLED source.

Once printed, it becomes an UNCONTROLLED document & users shall verify the correct revision.

[BD1]I’d like to see a line on this table reflecting the current buoy bistatic tests, inclusive of a multistatic detection type.

[BD2]How is detection rate calculated? A small explanation on this would be beneficial (maybe this was in a prior report? But I think it would be helpful to repeat it here.

[BD3]I am not clear on what this graph shows. Does each marker represent a test set? If so does this mean that there are tests with 0% detection rates?

[BD4]Is there a metric for false alarms? Is it possible to overlay false alarms on the detection rate chart?

[BD5]Can you clarify why DeltaVr is better the higher it is, or is it better lower? Maybe a definition of DVr will explain this.