2018ApJ...864..131K

Observations of the 21 cm Epoch of Reionization signal are dominated by Galactic and extragalactic foregrounds. The need for foreground removal has led to the development of two main techniques, often referred to as "foreground avoidance" and "foreground subtraction." Avoidance is associated with filtering foregrounds in Fourier space, while subtraction uses an explicit foreground model that is removed. Using 1088 hr of data from the 64-element PAPER array, we demonstrate that subtraction of a foreground model prior to delay-space foreground filtering results in a modest but measurable improvement of the performance of the filter. This proof-of-concept result shows that improvement stems from the reduced dynamic range requirements needed for the foreground filter: subtraction of a foreground model reduces the total foreground power, so for a fixed dynamic range, the filter can push toward fainter limits. We also find that the choice of window function used in the foreground filter can have an appreciable affect on the performance near the edges of the observing band. We demonstrate these effects using a smaller 3 hr sampling of data from the MWA, and find that the hybrid filtering and subtraction removal approach provides similar improvements across the band as seen in the case with PAPER-64.