Graphs showing the results to date of the SETIEasy v1.31a
tests on the linear_drift_*.wav files
This page last updated Aug. 11, 2000 at 12:15 am
To begin to establish the performance of SETIEasy v1.31a, we are filling in the gaps in the following graphs using the weak signal test files. The following experiments were designed to test the hypothesis that SETIEasy can detect at least some of the weak frequency drifting signals in the 8-bit linear-drift test set. The results to date are:
1. the graph of signals (i.e., -1 to +1 dechirp in at least 0.1 steps) for linear_drift_1.wav and linear_drift_7.wav should produce the largest signal at the correct dechirp value(s)
2. the graph of signals (i.e., -1 to +1 dechirp in at least 0.1 steps)
for linear_drift_7.wav and linear_drift_7.wav should produce no signal
A signal is apparent, although its peak intensity is about 1/2 that of the largest signal in the test set (1 vs 7, graphed above).
It is interesting to note, however, that the nosignal.wav noise test file has a very similar dechirp profile, or "spectrum".
3. the graph of signals (i.e., -1 to +1 dechirp in at least 0.1 steps) for linear_drift_6.wav and linear_drift_7.wav should produce the smallest signal
This signal is essentially about the same size and shape as the "noise" signal graphed in 7 vs 7 above.
In practice, the ideal response will not be observed, of course, but we need to determine how close we are. All of the results to date are combined on the following graph.
The missing smallest signal dechirp results at -0.4 and -0.2 seemed interesting because the negative dechirp values appeared to detect the smallest signal. Running a "calibration curve" at dechirp = -0.5 appeared to be a good idea.
The calibration curves for dechirps = 0.2 and 0.6 are:
Clearly a signal is visible at signal level 1 (5% of p-p noise) and level 2 (2.5% of p-p noise) at dechirps 0.6 and 0.2. The smaller peak at negative dechirp (-0.5) does not perform nearly as well. Linear drift signal level 1 vs. 7 is only slightly higher than the noise peak at 6 vs. 7.
These results parallel what was observed during dechirping ("stretching") analysis of the same files with CoolEdit 2000, that is, signals above 2.5% of the noise peak-to-peak (p-p) are visible. These results also suggest that a test of 16-bit format data files would be valuable, because the weaker signals (0.08-2.5% of noise p-p) may be invisible to SETIEasy and CoolEdit only because they are so much smaller than 1/256 of the noise level (1 of 8 bits).
Finally, it is not yet clear that SETIEasy is more or less sensitive than CoolEdit to the precise setting of the dechirp parameter. In theory, it is very important to get the frequency drift correction exactly correct to integrate signals and improve S/N via the FFT. Indeed, this behavior was noted with CoolEdit. However, the harmonics that are generated by the time domain dechirp in SETIEasy, combined with the cluster analysis (SOB) subroutine in SETIEasy, may be enough to make the program tolerant of mismatches between the actual frequency drift and the dechirp.
Description of 8-Bit Linear Drift Test Files
The linear drift files are a set of seven 15-minute noise files containing a weak, narrow signal with linear frequency drift. There is a signal in the first 14 minutes and 17 seconds of each file, and then about 43 seconds of noise at the end. The signal starts at 1400 Hz and ends at 1540.8 Hz, and a drift of a bit over 9.8 Hz/min. The first file has a signal that is small but clearly visible in CoolEdit. [A signal can be seen in linear_drift_2.wav in CoolEdit only if the whole signal is Scanned in the Frequency Analysis box after the proper dechirp (Stretch) is applied.] The intensity is then cut in half repeatedly until well after it completely vanishes. The signal level is given as a % of the peak noise.
5.00% http://monroe.pharm.uky.edu/seti/linear_drift_1.wav
2.50% http://monroe.pharm.uky.edu/seti/linear_drift_2.wav
1.25% http://monroe.pharm.uky.edu/seti/linear_drift_3.wav
0.63% http://monroe.pharm.uky.edu/seti/linear_drift_4.wav
0.32% http://monroe.pharm.uky.edu/seti/linear_drift_5.wav
0.16% http://monroe.pharm.uky.edu/seti/linear_drift_6.wav
0.08% http://monroe.pharm.uky.edu/seti/linear_drift_7.wav
The optimum dechirp factor in SETIEasy for these signals is about DC = .051252, while the optimum dechirp in CoolEdit (Gliding Stretch) is about 110.8% (users must manually type decimal corrections into the Gliding Stretch box, because the slider does not allow decimal %).
The detection limit for CoolEdit is similar to SETIEasy 1.31a (2.5% of noise p-p) in that both can barely see the signal in linear_drift_2.wav. There is little doubt that the CoolEdit spectrogram is faster at detecting the signal than SETIEasy, or that CoolEdit requires more operator intervention.
In this figure, the entire 14 min, 15 sec signal at 1400 Hz in linear_drift_2 was integrated after Gliding Stretch drift correction of 110.8%. The drift corrected signal looks like:
Tweaking the spectrogram parameters enables users to see just a hint of a signal in linear_drift_3 (1.25% of noise p-p) in CoolEdit. At that level, signal artifacts and terrestrial interferences are common in the equipment and the files must be examined very carefully.
Return to the ASRG Open Source page