Comparing performance of MLP and RBF neural networks for TV idle channel prediction in Cognitive Radio

The communications frequency spectrum in Nairobi city, like in most parts of the world, is statically allocated. The cellular network, due to the emerging public obsession with broadband, is overwhelmed by demand. Yet TV channels are under-used in certain geographical areas. Cognitive Radio would enable an opportunistic, un-licensed user to `borrow´ an idle band and seamlessly hand it back the moment the licensed user needs it. Cognitive Radio would conventionally sense idle channels and decide on the most technically-appropriate idle channel to allocate a secondary user. However, the process of decision is time-consuming and allocates channels erratically without due consideration of temporal appropriateness of the channel. The twin problem of time-burden and temporal in-appropriateness is solved if the decision-making were done earlier. This is only possible by predicting the occurrence of idle channels within the cell of concern. Prediction of idle channels demands that patterns of TV-viewing within that cell must be modeled successfully. This paper compares the performance of two of the most popular neural network architectures, Multi Layer Perceptron (MLP) and Radial Basis Function (RBF), in modeling the mentioned patterns. The study on patterns of TV-viewing was conducted within one of the worst-burdened mobile cells in the city, Umoja Savannah. Modeling was done using Matlab R2010a.