Demonstrating Spectrum Sensing in Colored Noise for Signals with Partial Spectral Overlap

Wideband spectrum sensing aims to identify the occupied frequency bands. Conventional methods for single antenna spectrum sensors threshold the received power spectra to detect discrete frequency bins that are occupied. However, such methods neither group bins that are occupied by the same signal nor aggregate occupied bins over time to learn distinct frequency bands occupied by intermittently transmitting signals. This paper demonstrates a method to learn the frequency bands occupied by intermittently transmitting incumbent radios that occupy adjacent frequency bands without a guard band, such as by LTE-Advanced, or are spectrally overlapping, such as by IEEE 802.11. It formulates the wideband sensing problem as the factorization of a matrix consisting of multiple power spectrum measurements. A novel extreme ray based non-negative matrix factorization algorithm estimates the noise power spectrum and also determines the received power spectrum of the incumbent radios. Energy detection and a combinatorial algorithm is used to determine the unique supports of the received signals. Using a USRP N210 software defined radio as a receiver, we demonstrate that this algorithm can determine the frequency bands occupied by nearby transmitters in the 2.4GHz ISM band. Furthermore, we demonstrate that the algorithm learns the power spectrum of the colored noise experienced by the USRP.