Shrinkage methods applied to adaptive filters

This paper analyzes the use of some regression shrinkage methods in adaptive signal processing. Some shrinkage strategies that render interpretable models can be solved as a linearly-constrained least squares problem and render model coefficients which are exactly zero. As a consequence, they produce estimators which may be more economical and have lower variance than those produced by ordinary least squares estimators, at the price of some bias. Economy, in this case, means less computations, consequently less battery consumption and more sustainable systems.