Spectrum Sensing of Multiple Channels Using Multiple Sensors

Thumbnail Image
Liang, Che Kang
sequential analysis , spectrum sensing
Cognitive radio (CR) is a class of wireless communication technologies that have the ability to learn from the surrounding radio environment and the intelligence to adapt communication resources to enhance quality of service. The problem of acquiring information from a CR's radio environment is called spectrum sensing, which can take on many forms. In particular, this thesis concerns the determination of whether a spectrum band (or channel) is in a busy or idle state. The binary nature of a channels availability means that spectrum sensing can be cast as a hypothesis testing problem. While an abundant literature exists on spectrum sensing as a signal detection problem, this thesis treats spectrum sensing differently, and features the following elements: 1) the system is equipped with an arbitrary number of sensors; 2) sensing is performed over multiple channels; 3) each channels availability is modelled by random periods of busy and idle times corresponding to packet transmission; and 4) the optimization criteria minimizes detection delay subject to a reliability constraint. A related spectrum sensing problem formulation based on the use of a single sensor has been proposed in the recent literature. The previous research employs an optimization framework based on modeling channel uses as an on-off process via partially observable Markov decision processes (POMDP). This thesis generalizes previous results from single-sensor to multiple-sensor spectrum sensing, i.e., detecting idle periods with multiple sensors. In addition, an alternative reduced-complexity algorithm is proposed. For both proposed detectors, the performances are evaluated based on Monte Carlo simulation with calculated confidence intervals, and the results show that 1) adding sensors generally improves the system performance by reducing detection delay (improved agility); 2) the application of previously existing quickest detection methods result in error floors complicating test design. Finally, performance assessment using a channel model derived experimentally from the wireless local area network (WLAN) traffic is conducted and compared to that obtained using a geometrically-distributed channel traffic model.
External DOI