Time Domain GMAP Clutter Filter for Weather Radars

Clutter filtering is a long-standing problem but always an important aspect to improve the radar data quality and ensure the correct functioning of any weather radar system. There are many filters have been used such as the FIR, IIR filter, and regression filter. These filters are simple and fast. However, when precipitation overlaps ground clutter, the use of these filters removes the clutter power but also cancels part of signal and makes no attempt to repair the filter bias. Recently, a Gaussian Model Adaptive Processing (GMAP) method (Siggia and Passarelli 2004) have been developed to overcome the drawback of traditional IIR/FIR filters. GMAP is a frequency domain method that combines both filtering and interpolating to minimize estimate's errors. GMAP is the current state of the art method and is being deployed in the NEXRAD. The interpolation procedure in GMAP is the biggest advance when compared to previously developed methods. However, the major limitation of any spectral processing methods such as GMAP is the window effect that associates with the calculation of discrete Fourier transform (DFT). This problem restricts the applicability of spectral clutter filtering method to cases of moderate clutter-to-signal ratio (CSR). Even in the case of moderate CSR, finite length of the data introduces clutter leakage problem that results large error in sensitive parameters' estimation such as co-polar correlation coefficient and differential propagation phase. Besides, by applying a window to data, it naturally increases signal standard deviations. Last, GMAP can not be applied directly to non-uniform sampling techniques like staggered pulse repetition time (PRT).

In this work, we introduced a new filter, namely Time Domain GMAP (GMAP-TD), which is based on the concept of GMAP. In GMAP-TD, the clutter removal is implemented in time domain to obtain signal moment parameters. The proposed time domain filter addresses many important aspects: 1) it does not bear the disadvantages of spectral processing method, as in GMAP; 2) the filter design is independent of sampling scheme; therefore it can be applied to staggered PRT transmission techniques; 3) GMAP-TD can work with less samples compared to original GMAP. Finally, GMAP is not too costly computationally and can be implemented for real time application using general computer processors.

This work is illustrated on both simulation and CSU-CHILL observations for uniform and staggered PRT 2/3 scheme. The results show that GMAP-TD performance is as good as the original GMAP. In addition, GMAP-TD works well at very high CSR.