Nuit Blanche: CS: Is HIFT an instance of Imaging With Nature ? The data is available.

Tuesday, June 08, 2010

CS: Is HIFT an instance of Imaging With Nature ? The data is available.

I know... I know, all of you are dying to see part II of this entry on Compressed Sensing or Inpainting ? Part I but this will have to wait as there is something more inspirational today.

After talking to Raj Rao last week at the Random Matrix conference, I went ahead and asked Brian Dushaw about having access to the data of the Heard Island Feasibility Test. Brian got back to me with the following today:

Hi Igor,
The Heard Island data are on line now - you can find these data here:
http://909ers.apl.washington.edu/~dushaw/heard/data/DVD/

I still need to set up a brief navigational webpage for these directories. The place to start would be the hiftdata.pdf file on HIFT_CD1. Also the special issue of JASA from October 1994 shows all the work that was done with the data.

I can offer no help in dealing with these data - I don't know much about it and will have to thrash around as much as anyone if/when I go to do anything with it. But here it is. There are no strings attached to the data, as far as I know; the data were paid for by the U.S. Government and are in the public domain. But acknowledging the original HIFT people in any publication would be appreciated.

Cheers,

B.D.

Thanks Brian. From "The Heard Island Feasibility Test" by W. H. Munk, R. C. Spindel, A. Baggeroer, and T. G. Birdsall, one can see the type of signals being sent during these tests from the heard Island shown above and detected 20,000 km away.

A more generic presentation can be found in Signals, signal processing, and general results by Theodore G. Birdsall and Kurt Metzger, Matthew A. Dzieciuch. The abstract of that first paper reads:

Acoustic path lengths in the Heard Island Feasibility Test ranged from under 1 Mm to 18 Mm (1 Mm is 1000 km). The signal set consisted of three basic waveforms: cw, pentaline, and M-sequence-modulated carrier. This set offered the opportunity for successful measurements given the large uncertainty in prior estimates of propagation loss, stability, and arrival spread. Receivers ranged from simple sonobuoy systems to elaborate horizontal and vertical arrays. International collaborators acquired data at a variety of sites worldwide. The resulting data has been collected and subjected to a summary form of frequency domain processing. Variations in the recorded spectral phases are largely the result of nonuniformity in the speed of the source ship as determined by GPS comparison. Time domain processing has shown that at all ranges the receptions exhibit exceptional stability.

Of interest is the book on Ocean Acoustic Tomography.

The Heard Island Feasibility Test is at:

http://909ers.apl.washington.edu/~dushaw/heard/index.shtml

Why am I mentioning this test ? While the intent of the test is figuring out if one can detect a signal 10000 's miles, it also is a way of probing different parameters/landscape of the ocean. Isn't this also is a clear instance of Imaging With Nature, namely, the signal being sent (at least some of them) are sparse. The tomographic question answers "Can we infer something about the medium (i.e. the measurement matrix) when one has access to to be x and y". We know that the measurement matrix is highly underdetermined. After the measurement matrix is determined in some fashion, along with additional information such as bottom topography and sattellite determined sea temperature: can we infer something about future signals such as location, mapping and type of: