Real time implementation and assessment of spectrum sensing algorithms using small-sized data for cognitive radio

The growth of wireless communications has increased demand for RF spectrum. The FCC has already allocated most of the frequency bands to licensed primary users. This has led to spectrum scarcity. Interesting findings by the FCC show that most of the licensed bands are underutilized even in dense urban areas. For secondary users to opportunistically use the spectrum without interfering the primary users, cognitive radio technology was proposed. Spectrum sensing of cognitive radio should be quick in detection to effectively use the spectrum, and for protection of primary users. In previous works, an estimated covariance spectrum sensing method was proposed to sense the spectrum with only small size data, which means detection in short time. This thesis outlines the development of a cognitive radio testbed and the implementation of this method in a real environment. A study of its detection performance was then carried out and the results presented.;Traditional covariance based spectrum sensing algorithms use sample covariance matrix for estimating the true covariance matrix, but for small sized data and large dimension case, sample covariance matrix is a poor estimator of true covariance matrix. The estimated covariance spectrum sensing method utilizes an OAS approach for estimation of true covariance matrix and MME detection for spectrum sensing.;A cognitive radio testbed is developed with USRP2 which is easy to build and use. GNU Radio software is used along with a USRP2 to generate a primary user signal. A python program is developed which can control the secondary user USRP2 through UHD. The spectrum sensing algorithms can be programmed in this python program.;A cosine signal was generated to emulate a primary user signal for implementation in a real environment. It was observed that the estimated covariance spectrum sensing successfully detects the spectrum occupancy with small sized data in a real environment. It was also observed that the detection performance of estimated covariance spectrum sensing is better than the sample covariance spectrum sensing for small sized data in a real environment. These results validate the simulation results with the same cosine signal. Implementation issues with USRP2 are also discussed.