Recursive filter algorithm for noise reduction in SLR

This report presents the concept and implementation of a recursive filter for the identification of satellite returns in laser ranging in the presence of strong noise. This project was aiming for an increased data yield of automatically filtered satellite laser ranging measurements in order to maximize the number of correctly identified returns. Furthermore the amount of false readings have to be reduced and an automatic timebias-adjustment during ranging was required. Introduction Automatic data screening of timer readouts in SLR is widely used by many laser ranging facilities of the ILRS. All of them depend on some type of histogram evaluation of short time slices of measurements throughout the ranging process. The approach uses the fact that return signals from a satellite bunch up at a specific location in the range gate window, while noise readouts caused by background light or intrinsic detector noise are far more spread out throughout the range gate. For satellite passes with reasonable or good signal to noise ratio this method is fully adequate. However, in particular for daylight passes of the GPS and GIOVE satellites, this method is often extracting much fewer returns than actually were recorded by the ranging facility. On top of that a non negligible number of false readings is usually upsetting the normal point generation process, because erratic data points prevent the fitting procedure from converging. Figure 1 shows an example of such a weak satellite pass. One can clearly see time intervals where a reasonable or good signal to noise ratio exists for the measurement. However there are also times where only sparse data is recorded. In order to extract the valid returns out of all the recorded data points in near real-time the control software examines small portions of the pass of a few seconds length. The data is then converted to a histogram and if a suitable bin contains a sufficient number of echoes, these are extracted and stored away as satellite returns. This evaluation process is fast and strictly linear in time. In the presence of very sparse data the threshold criterion is never satisfied and valid data is lost. If on the other side the threshold value is lowered too far, then randomly lumped together background noise events will accidentally be taken as good data and the postprocessing can be disrupted. By using more than one criterion at a time and introducing reprocessing of past data as well as a locally linearized look ahead strategy, one can vastly improve the robustness of the filter procedure. At the same time the data yield improves substantially in particular for passes with a low signal to noise ratio. Function of the new filter algorithm The new filter applies two distinctly different methods. A histogram-analysis is used to detect possible satellite returns in a reasonably short time interval. The results then are used to predict the likelihood of valid returns into the future, where it also successfully recovers valid data-points at a low data rate. Both methods cooperate to not only detect, but also rate identified returns during the ranging activity. Figure 1: Example for a measurement window of an ETALON pass with sparse data in daylight. From a number of verified satellite returns within a number of time slices, the actually applied time bias value for the momentarily observed satellite pass can be improved. With time bias corrected range residuals the histogram of the analysis process sharpens substantially. As a consequence the width of the rangegate can then be reduced automatically, which in turn enhances the data yield of the ranging operation. The program module works in several layers. The inner loop of the filter procedure is based on time slices of 5 seconds of observations (fig. 2). The length of the time slice is adjusted to the 10 Hz repetition rate and the background noise level typical for the Wettzell Laser Ranging System (WLRS). Other systems will have different settings. If already available a time bias correction is applied to all the data points in that time segment. Then the data is passed on to a histogram analysis routine, which has a bin width of 5 ns. This arbitrarily chosen value too has shown to work well for the WLRS operation parameters.