Robust adaptive filters: a comprehensive comparison

In this study, we conduct a detailed comparison of popular gradient-based robust adaptive filtering algorithms. Their robustness against different noise distributions, including Gaussian and non-Gaussian noise, is systematically evaluated in the system identification scenario in terms of mean square deviation (MSD) (dB), runtime, and computational complexity. This study is constructed on two key objectives. The first objective is to determine which algorithms, based on their existing weight update rules, exhibit inherent robustness to light- and heavy-tailed non-Gaussian noises. The second is to determine algorithms that need to be improved in terms of robustness to non-Gaussian noises. With the above objectives, the findings in this work aim to fill possible gaps in the development and evaluation of gradient-based robust adaptive filtering techniques.