The following figure (click for gzipped Encapsulated Postscript file) shows a vertical profile of Doppler spectra calculated from the first 65536 pulses of a 15 second dwell (time-series file 26155055.088), together with the vertical profile of signal-to-noise ratio. The spectra at every second gate are shown, and positive velocities are downwards. Since a raw Fourier Transform has a random noise of 100%, 256-point averaging was performed on the spectra to reduce the error to 0.27 dB (theoretical value), but this also reduced the spectral resolution from 0.015 cm/s to 3.9 cm/s. A Welch window was used, and the Z-weighted I and Q values were adjusted to have a mean of zero. The peak at 0 m/s in gates 2, 20, 22 and 26 is most likely ground clutter.
There appears to be a strong drizzle component (>100 micron drops) throughout the depth of the cloud, and at 1 km there is also a cloud mode (<50 microns droplets). Since it is these smaller droplets that contain the bulk of the liquid water and are dominant in determining the radiative properties of the cloud, we need to establish how long the dwells needs to be to adequately resolve this important part of the spectrum.
In the following figure the spectra from five adjacent gates are shown, three of which show a distinct cloud component. The left plot is for a 10.5 second dwell (as above) and the right plot is for a 1.3 second dwell (from the middle of the 10.5 second time series). Both sets of spectra have been averaged to yield a velocity resolution of 7.8 cm/s. The errors for the long dwell are therefore a factor of sqrt(8) less than those for the short dwell. The cloud mode is apparent even in the spectra from the short dwell but, because of the increased error, it is much more likely that noise could be mistaken for the cloud mode.
The spectra in the left plot are a little broader than those in the right plot, indicating that variations in vertical velocity in the 10.5 second period had a `smearing' effect on the spectra. Hence there is little to be gained from extremely long dwells; any longer than 10.5 seconds is probably unnecessary. In any case, if turbulence in a short dwell is sufficient for the cloud mode to be smothered by the drizzle mode (as is probably true in gates 12 and 16 above), then even with infinite spectral resolution it is not going to be recoverable.
Hence a dwell time of 5.25 seconds, corresponding to 32768 pulses, would probably be sufficient for most purposes with the current PRF. This is based on the assumption that the only way to reduce the errors is to average in the velocity domain, but there are probably cunning ways to reduce the errors without compromising on spectral resolution, in which case a shorter dwell time would be acceptable. The current range resolution of 75 m should be retained.
Click here to see some Doppler spectra in cirrus.