One of the problems facing wireless network planners is a perceived scarcity of spectrum. A technology that addresses this problem is cognitive radio (CR). A critical function of a CR network is spectrum sensing (SS). A secondary user (SU) in a CR network will perform SS to gather information about the radio environment within which it wishes to operate and then make decisions based on that information. While SS by individual SUs is very useful it has been found in the literature that a cooperative approach, where SUs share their individual results, may provide more accurate information about the radio environment. It has also been shown that it is beneficial for SUs to be able to make proactive decisions about spectrum resource allocation. To be able to make these proactive decisions, a SU will need to be able to make predictions about the future behaviour of other users of the same spectrum.
This research project was divided into two parts. Firstly, a measurement campaign was performed to characterise spectrum scarcity in the South African context. Detailed information, about the occupancy of various commercially utilised South African frequency bands, was collected from spectrum measurement campaigns carried at the Hatfield campus of the University of Pretoria and at Pinmill Farm in Johannesburg. These bands included the television broadcast and mobile cellular bands. On average, the television broadcast bands were found to be underutilised highlighting the existence of a number of opportunities for television white space devices. However, the mobile cellular bands were found to be much more heavily occupied, particularly for the bands around 900 MHz, suggesting that mobile operators are currently in need of additional spectrum resources.
The second part of this thesis followed a more theoretical approach and was based on the need for proactive decision making in CR networks. A single SU prediction method, of relatively cheap computational complexity, was proposed and tested under various traffic conditions. The premise that collaboration between SUs may improve the accuracy of single SU traffic predictions was then explored. Pre-fusion and post-fusion approaches to cooperative prediction were compared with the single SU prediction scenario. The prediction error for the cooperative approaches was found to be lower than for the single SU case, especially for the pre-fusion scenario. For example, for a signal-to-noise ratio of 8 dB and individual forecast probability of 0.9, the pre-fusion prediction error was found to be approximately 2% compared with 26% for single SU prediction error. The cost of this improvement, however, was added algorithm complexity.
It was then demonstrated that primary user traffic prediction could be used to improve the energy consumption associated with cooperative SS in a CR network. Combined with an optimal scheduling algorithm, this approach was shown to prolong the lifetime of a group of twenty cooperating SUs by 21.2 time samples for a uniformly distributed group of SUs when predictions were made ten time samples into the future.