Experimental Study on the Performance of RFI Detection Algorithms in Microwave Radiometry: Toward an Optimum Combined Test

Abstract

Radio-frequency interference (RFI) is probably today's most serious limitation to the accurate retrieval of geophysical parameters from microwave radiometric measurements. Strong RFI inducing a change in the detected power larger than the natural variability is simple to detect. Moderate or weak RFI can be masked by the natural variability of the measurements, passing undetected and corrupting them. A number of techniques have been devised in the past years to detect and, eventually, mitigate RFI present in microwave radiometry measurements: 1) time domain; 2) frequency domain; 3) spectrogram techniques looking for anomalously high power peaks; 4) statistical techniques testing the hypothesis of Gaussianity of the received signal; 5) polarimetric techniques looking for anomalous signatures in the third and fourth Stokes parameters; or 6) wavelet techniques to estimate the RFI signal and cancel it (if any). In this paper, the first four techniques are evaluated with real data gathered with a multifrequency microwave radiometer. It will be shown how spectrogram techniques can detect RFI signals concentrated in narrow frequency bands and/or time intervals that may pass undetected with time-domain and/or frequency-domain techniques alone or with statistical methods. A combined approach is proposed to take advantage of the best performance of each technique. On one side, for strong localized RFI, the approach is spectrogram blanking or, if it is too demanding in terms of computational resources, simple time- and frequency-domain blanking. On the other side, for weak RFI, the approach is the Kurtosis statistical test, which exhibits the best performance among the ten normality tests evaluated, in conjunction with the Anderson–Darling test to detect potential RFI in the blind spots of the Kurtosis test.