Multi-Channel Signal Separation Challenge
In the single-channel portion of RFChallenge, the task is to design an algorithm which learns and exploits the waveform structure in cochannel RF signals to separate them. When available, multi-channel measurements offer additional leverage for signal separation in the form of low-dimensional spatial structure. Namely, manmade interference typically arises from point RF emitters, and point emitters are intrinsically low-dimensional when observed by an antenna array with a sufficient number of elements1. In this part of the Challenge, the ISM2-band downlink of an unmanned aerial vehicle (UAV) is buried in co-channel interference from another UAV at a distinct direction-of-arrival. The task in this part of the Challenge is to exploit the structure in multi-channel measurements of the mixture of the two UAV waveforms to extract the downlink signal from its interference. The lack of prior information on either the individual waveforms, or their respective array responses at the receiver, makes unsupervised machine learning an intriguing framework for solving this blind source separation problem.