A SUBSPACE-BASED METHOD FOR SPECTRUM SENSING

A subspace-based method for gaining knowledge of the quality and usage of the spectrum is presented. The method separates so-called 'structured' and 'unstructured' components of a signal as determined by the eigenvalue decomposition. These components may be cast as carrier power and interference plus noise power (interference temperature), thus providing an estimate of the quality of a piece of spectrum even in the presence of incumbent services. Furthermore, the method may be extended to yield a unique signature for many different modulation types. Combination with a neural network leads naturally to a robust modulation recogniser which has been demonstrated on a spectrum analyser platform. Some applications for the technique are discussed.