Next: Morphology during a Complete
Up: Experimental Results
Previous: High-Resolution Autocorrelation Functions
For the determination of regions in the F-layer where the double peaked
Doppler velocity spectra occur, an automated selection process is used. The
complete procedure can be outlined as follows.
- Select only ACFs with a SNR of or higher as determined from the noise
level and the measurement at zero lag, in order to filter out weak signals.
- Generate 256-point PSD estimates using the Burg model of order 8.
- Determine local peaks neglecting all values smaller than 15% of the
maximum in the spectrum to filter out components generated by noise.
- Select only PSDs with two peaks in them to filter out single or
multiple peaked spectra.
- Select only PSDs with a velocity separation less than .
- Select only PSDs with a velocity that is not close to zero (within ), in order to filter out signals containing ground-scatter components.
These strict rules are applied to filter out noisy, marginal data as well as
more complicated spectra. A comparison between these rules and a more
relaxed set using a maximum peak separation of and a minimum
velocity of only shows that they are sufficiently relaxed to
identify double-peaked spectral regions, yet strict enough
to avoid the misidentification of
noisy ACFs. As mentioned before, the FFT-based methods are not used for
regular analysis (exept for the comparisons) since
they lack resolution and do not perform well on short data records. In the
following figures (Figure 5.7 to
Figure 5.15)
both spectral estimation methods have been used
on the same data in a search for double-peaked spectral features using the
rules outlined above. One can
see that the FFT-based methods produce many spurious finds, especially at
the large ranges when backscatter intensities are low (16 Jan/95 and 26
Jan/95), or they do not find
anything at all, due to the lack of resolution (18 Feb/95 and 25 Feb/95).
There is good agreement between the methods when backscatter
intensities and peak separations are high (6 Jan/95, 16 Jan/95 and 26 Jan/95).
Figure 5.7: Comparison of maps of double-peak spectra derived from
different methods: Jan 1/95
Figure 5.8: Comparison of maps of double-peak spectra derived from
different methods: Jan 6/95
Figure 5.9: Comparison of maps of double-peak spectra derived from
different methods: Jan 16/95
Figure 5.10: Comparison of maps of double-peak spectra derived from
different methods: 26 Jan/95
Figure 5.11: Comparison of maps of double-peak spectra derived from
different methods: 28 Jan/95
Figure 5.12: Comparison of maps of double-peak spectra derived from
different methods: 18 Feb/95
Figure 5.13: Comparison of maps of double-peak spectra derived from
different methods: 20 Feb/95
Figure 5.14: Comparison of maps of double-peak spectra derived from
different methods: 21 Feb/95
Figure 5.15: Comparison of maps of double-peak spectra derived from
different methods: 25 Feb/95
Next: Morphology during a Complete
Up: Experimental Results
Previous: High-Resolution Autocorrelation Functions
Andreas Schiffler
Wed Oct 9 10:05:17 CST 1996