In this reported work, radio frequency interferences (RFI) in synthetic aperture radar (SAR) are suppressed using novel tools. Introduced first are the random sampling techniques that are commonly used in large-scale machine learning areas to obtain the RFI subspace via the approximated spectral decomposition. Thus, one can reconstruct the RFI signals efficiently and then remove them from the original received SAR pulses one by one. The performance of the algorithm was tested on a P-band real dataset.