Variance estimations in the presence of intermittent interferences and their applications to incoherent scatter radar signal processing

Abstract. We discuss robust estimations for the variance of normally distributed random variables in the presence of interferences. The robust estimators are based on either ranking or the geometric mean. For the interference models used, estimators based on the geometric mean outperform the rank-based ones both in mitigating the effect of interferences and reducing the statistical error when there is no interference. One reason for this is that estimators using the geometric mean do not suffer from the “heavy tail” phenomenon as the rank-based estimators do. The ratio of the standard deviation over the mean of the power random variable is sensitive to interference. It can thus be used to combine the sample mean with a robust estimator to form a hybrid estimator. We apply the estimators to the Arecibo incoherent scatter radar signals to determine the total power and Doppler velocities in the ionospheric E-region altitudes. Although all the robust estimators selected work well in dealing with light contaminations, the hybrid estimator is most effective in all circumstances. It performs well in suppressing heavy contaminations and is as efficient as the sample mean in reducing the statistical error. Accurate incoherent scatter radar measurements, especially at nighttime and E-region altitudes, can improve studies of ionospheric dynamics and compositions.