Versions of SETIEasy at 1.25 and beyond read 16-bit files, and uses a new method of selecting training sets and calibrating the noise that permits dechirping (signal frequency-drift corrections) to be performed prior to long FFT signal integrations.
SETIEasy v1.24 calibrated the algorithm (i.e., trained it to recognize noise) for each data chunk time duration using the first data chunk of each duration in the downloaded file. The figure below is a CoolEdit spectrogram that shows how data chunks are compared in v1.24. Calibration is performed once on data chunk 1, which is then compared statistically to subsequent chunks of the same size in bytes, i.e., 1 to 2, 1 to 3, 1 to 4, etc. Calibration is re-performed only when the number of data bytes in chunk 1 is changed.
The big problem with this approach does not really arise until the data are dechirped. Dechirping shifts the noise distribution in the process of correcting frequency drifts of signals for Doppler shifts.
The drifting noise distribution makes comparisons of 1 to 2, 1 to 3, 1 to 4, etc. quickly trigger the unusual signal flag, even in the absence of a signal. Dramatic changes in the noise distribution are clearly observable even in the raw spectrogram, without the benefit of long FFTs in SETIEasy to increase S/N.
The SETIEasy v1.25 alpha was initiated with the intention of permitting dechirping by splitting a data file in half, and dechirping each half identically. In this case, calibration on each data chunk in one half permits testing of the corresponding chunk in the other half, without the dechirped noise triggering an unusual signal flag. Now, chunk 1 in test.wav is compared with chunk 1 in test2.wav, and chunk 2 with chunk 2, 3 with 3, etc. There is a requirement that at least some of the data chunks be separated by at least the 3 dB beamwidth of the antenna (in terms of drift-scan time, to assure that a different signal intensity is observed in two different data chunks in the presence of a real signal).
This requirement can be met by making the files at least 30 minutes long, or by making the data files smaller and downloading two separate files more than 15 minutes apart (which will probably be done in the next release of SETIEasy).
Because the first data chunk is no longer the calibration chunk for the rest of the file, the algorithm has to recalibrate for every data chunk (1, 2, 3, etc.), leading to substantially longer execution times. However, this longer running time will permit signal frequency drift correction (dechirping). In an initial test on a 200 MHz Pentium, the execution time was 60 hrs on a one hour data file, without dechirping. It seems likely that the file size will have to be reduced to something closer to that employed by S@H to keep the download and execution times reasonable.
Return to the ASRG SETI home page.