Several attacks on AES using far field electromagnetic (EM)emission as a side channel have been recently presented.Unlike power analysis or near filed EM analysis, far fieldEM attacks do not require a close physical proximity to thedevice under attack. However, in all previous attacks tracesfor the profiling stage are also captured at a distance (fixedor variable) from the profiling devices. This degenerates thequality of profiling traces due to noise and interference. Inthis paper, we train deep learning models on "clean" traces,captured through a coaxial cable. Our experiments show thatthe resulting models can extract the AES key from less than500 traces on average captured at 15 m from the victim devicewithout repeating each encryption more than once. This is a20-fold improvement over the previous attacks which requireabout 10K traces for the same conditions. - View it on GitHub