Of primary interest to the measurement of winds from space is the development of optimal signal processing algorithms. The Cramer-Rao lower bound is considered to be the ultimate target for algorithm performance. At this time, we expect to have an algorithm that is within 5-8 dB of that bound. As part of our ground-based lidar research, we began assessing several processing algorithms. Since a consensus approach has been adopted for simulation studies, the ground-based data were processed to evaluate that technique.
The consensus algorithm is modeled after similar processing schemes used with radar data. Basically, the lidar time series along a single line-of-sight is subdivided into range gates (approximately the same duration as the lidar pulse) that represents independent samplesof the atmosphere within a specified layer. In the 9 km case, the layer is 1 km thick and there are ~4-5 independent range gates. When three or more ofthe five LOS velocity estimates agree within 2 m s-1, a consensus is declared and the average LOS component is used in subsequent calculations of the horizontal wind components. Since the noisy bandwidth of the signal processing is plus or minus 25 m s-1, there is always a possibility that consensus may be achived randomly - such consensuses are referred to as false alarms.
The claim by the signal processing community is that the consensus algorithm should perform similar to the Capon estimates noted above.
On May 27, 1993, the MSFC lidar ran a time series of scans (10 vads) with an elevation angle of 45 degrees. In the shot pairing section, SWA provide a discussion on how we processed the ten VADS of poly-pulsed data through the SWA ground-based Doppler lidar data processor to produce wind components and errors that are calculated for a range of bi-perspective shot pairs. We also used these data to populate our consensus algorithm to further evaluate the DWL consensus signal processing technique. We have drafted a consensus curve using some very preliminary results based upon 3 VADs of 1623. To note, the curve shows a much sharper fall-off of performance (percentage of estimates that meet consensus) than that expected.
| This page managed by Sidney A. Wood | Last modified: under construction |