Collaborative spectrum sensing in cognitive radio networks
The radio frequency (RF) spectrum is a scarce natural resource, currently regulated by government agencies. With the explosive emergence of wireless applications, the demands for the RF spectrum are constantly increasing. On the other hand, it has been reported that localised temporal and geographic spectrum utilisation efficiency is extremely low. Cognitive radio is an innovative technology designed to improve spectrum utilisation by exploiting those spectrum opportunities. This ability is dependent upon spectrum sensing, which is one of most critical components in a cognitive radio system. A significant challenge is to sense the whole RF spectrum at a particular physical location in a short observation time. Otherwise, performance degrades with longer observation times since the lagging response to spectrum holes implies low spectrum utilisation efficiency. Hence, developing an efficient wideband spectrum sensing technique is prime important. In this thesis, a multirate asynchronous sub-Nyquist sampling (MASS) system that employs multiple low-rate analog-to-digital converters (ADCs) is developed that implements wideband spectrum sensing. The key features of the MASS system are, 1) low implementation complexity, 2) energy-efficiency for sharing spectrum sensing data, and 3) robustness against the lack of time synchronisation. The conditions under which recovery of the full spectrum is unique are presented using compressive sensing (CS) analysis. The MASS system is applied to both centralised and distributed cognitive radio networks. When the spectra of the cognitive radio nodes have a common spectral support, using one low-rate ADC in each cognitive radio node can successfully recover the full spectrum. This is obtained by applying a hybrid matching pursuit (HMP) algorithm - a synthesis of distributed compressive sensing simultaneous orthogonal matching pursuit (DCS-SOMP) and compressive sampling matching pursuit (CoSaMP). Moreover, a multirate spectrum detection (MSD) system is introduced to detect the primary users from a small number of measurements without ever reconstructing the full spectrum. To achieve a better detection performance, a data fusion strategy is developed for combining sensing data from all cognitive radio nodes. Theoretical bounds on detection performance are derived for distributed cognitive radio nodes suffering from additive white Gaussian noise (AWGN), Rayleigh fading, and log-normal fading channels. In conclusion, MASS and MSD both have a low implementation complexity, high energy efficiency, good data compression capability, and are applicable to distributed cognitive radio networks.