A Factor Analysis Framework for Power Spectra Separation and Multiple Emitter Localization

Spectrum sensing for cognitive radio has focused on detection and estimation of aggregate spectra, without regard for latent component identification. Unraveling the constituent power spectra and the locations of ambient transmitters can be viewed as the next step towards situational awareness, which can facilitate efficient opportunistic transmission and interference avoidance. This paper focuses on power spectra separation and multiple emitter localization using a network of multi-antenna receivers. A PARAllel FACtor analysis (PARAFAC)-based framework is proposed, which offers an array of attractive features, including identifiability guarantees, ability to work with asynchronous receivers, and low communication overhead. Dealing with corrupt receiver reports due to shadowing or jamming can be a practically important concern in this context, and addressing it requires new theory and algorithms. A robust PARAFAC formulation and a corresponding factorization algorithm are proposed for this purpose, and identifiability of the latent factors is theoretically established for this more challenging setup. In addition to pertinent simulations, real experiments with a software radio prototype are used to demonstrate the effectiveness of the proposed approach.