Title: / HF Radar Detection Improvement Methods
Revision: / AB
A / HJR / First Draft / 25April 2011
Rev / By / Description / CN / Date Approved
HF Radar Detection Improvement Methods
Contents
1.1Introduction
1.2Vessel Detection Improvements
Figures and Tables
Figure 1: Detections from a SeaSonde HF radar (colored circles) along with range and bearing uncertainty (colored ellipses).
1.1Introduction
This document defines efforts that Rutgers and Codar have put forth to improving the vessel detection methodologies part of the Littoral Expeditionary PowerBuoy Platform (LEAP) Vessel Detection System Testbed (VDST). The improvements to the vessel detection code have been made possible by this contract as well as the Department of Homeland Security Center of Excellence program.
1.2Vessel Detection Improvements
The vessel detection process includes several components. First the radar system is configured properly as summarized in Report 2: Shore Station Installation Report. Then the detections are generated at the shore station using the PeakPicker software package. The configuration settings for the radar and PeakPicker must be set according to the type of vessel you are trying to detect. The detection files are then retrieved from the shore station and inspected for quality. Once the operator is satisfied with the quality of the data it can be forwarded onto a vessel tracker or data fusion software.
We have made improvements to each processing step as summarized below.The following improvements were made to the vessel detection code:
Can these items be divided into sections (with an objective)? such as:
1.2.1Processing Speed Enhancements
(to allow for faster processing of data thus allowing real time processing to be attainable)
- Created C program foundation for real-time peak detection -- based on CSPro, the SeaSonde's real time application that averages raw cross spectra.
- Verified that the original MATLAB code was operating correctly and simplified its operation for offline processing.
- Designed key components of the real time C code: i.e., input parameters allowing any number of FFT sizes with the two background methods (IIR and Median); and for the range file buffering algorithm (which is important in order to feed the proper number of range points to each of the FFTs).
- For fastest execution speed, the MATLAB code was profiled to determine areas of most intensive computation. Surprisingly it was not in the FFT computation but in the Median background computation. This happened because the median method requires the data to be sorted to find its median values. Sorting is a classic time-consuming computer problem. The volume of data processed and the number of median calculations are very high.
- For best response to user and minimum computation time, the processing of the FFTs and methods of detections were designed as 'open' loops that could be exited and reentered without loss of processing continuity. Input parameters were allowed to change on-the-fly.
- Programming was adjusted to utilize array processing methods built into the Mac OS system using the native array sorting and the eigen-vectors/values computations effectively. FFT calculations built into the Mac OS were actually slower than the code transferred from the SeaSonde. This is probably because the built-ins are DFTs made for graphic processing and not true FFTs.
- Redo calculation of ideal antenna-pattern outputs to include amplitude and phase factors.
- Tested with multiple data sets run through both the MATLAB and C versions - HOMR, SEAB, and BRNT.
- Preparation, monitoring, and bug repair for real-time field-testing.
1.2.2Scalability Enhancements
We have outlined a plan (to expand the vessel detection capability of the SeaSonde HF radar to the ~ 120 radars currently deployed around the nation. This is summarized in Report 4: National LEAP Vessel Detection System Network Implementation Plan. In order to efficiently deploy this process at a regional or national scale onto the radars that are currently deployed for surface current mapping the following steps were taken: better facilitate rolling out regional or a nationwide system)
- Configured the C output file folders to be compatible with the SeaSonde system
- Converted output target detection files to a format similar to the SeaSonde tabular output files so the same output format is generated by both MATLAB and the C version.
- Facilitated meeting between LEAP program partners and the Integrated Ocean Observing System (IOOS) Program Office on May 25, 2011. The IOOS Program Office provides a large share of the operational support as well as operates the National High Frequency Radar Network. We have begun discussions with IOOS as to how the vessel detection capability could be added to the existing sites.
1.2.3Security Enhancements
The Rutgers/CODAR vessel detection software algorithms and their associated data including software documentation (i.e., source code), all information concerning the design, development, and use of the algorithm are subject to the International Traffic in Arms Regulations (ITAR). All of the above is controlled as technical data under Category XI(d) of the U.S. Munitions List. In order to protect the security of the vessel detection algorithm and software a USB key (to provide hardware and software security on the ship detection code)
- Added key-coded ITAR-related security protection.
1.2.4UsabilityUser Interface Enhancements
(to provide for easier user access, variable selection, and output processing)
- Prepared the pepper-plot MATLAB program to read and display detections from both the MATLAB and C versions.
Created C program foundation for real-time peak detection -- based on CSPro, the SeaSonde's real time application that averages raw cross spectra.
Verified that the original MATLAB code was operating correctly and simplified its operation for offline processing.
Converted output target detection files to a format similar to the SeaSonde tabular output files so the same output format is generated by both MATLAB and the C version.
Prepared the pepper-plot MATLAB program to read and display detections from both the MATLAB and C versions.
Configured the C output file folders to be compatible with the SeaSonde system.
Designed key components of the real time C code: i.e., input parameters allowing any number of FFT sizes with the two background methods (IIR and Median); and for the range file buffering algorithm (which is important in order to feed the proper number of range points to each of the FFTs).
For best response to user and minimum computation time, the processing of the FFTs and methods of detections were designed as 'open' loops that could be exited and reentered without loss of processing continuity. Input parameters were allowed to change on-the-fly.
For fastest execution speed, the MATLAB code was profiled to determine areas of most intensive computation. Surprisingly it was not in the FFT computation but in the Median background computation. This happened because the median method requires the data to be sorted to find its median values. Sorting is a classic time-consuming computer problem. The volume of data processed and the number of median calculations are very high.
Programming was adjusted to utilize array processing methods built into the Mac OS system using the native array sorting and the eigen-vectors/values computations effectively. FFT calculations built into the Mac OS were actually slower than the code transferred from the SeaSonde. This is probably because the built-ins are DFTs made for graphic processing and not true FFTs.
Redo calculation of ideal antenna-pattern outputs to include amplitude and phase factors.
Added key-coded ITAR-related security protection.
Tested with multiple data sets run through both the MATLAB and C versions - HOMR, SEAB,and BRNT.
Preparation, monitoring, and bug repair for real-time field-testing.
A visualization improvement was made to the plotting code that added the uncertainty in range and bearing to the detections placed on a map. This is shown in Figure 1.
Figure 1: Detections from a SeaSonde HF radar (colored circles) along with range and bearing uncertainty (colored ellipses).
1.3Add a summarized cConclusion
(as to the value of these refinements/improvements/enhancements)?