A Survey of Artificial Neural Network based Spectrum Inference for Occupancy Prediction in Cognitive Radio Networks

Spectrum being a natural resource is always limited in availability. Its use needs to be intelligently managed for maximum benefits, hence the idea of Cognitive Radio. If the occupied or free status of spectrum band can be predicted or inferred from existing spectrum measurement data then this technique improves spectrum efficiency. This technique called Spectrum Inference/Prediction is a valuable tool to harness the spectrum effectively. This paper gives a recent survey of existing spectrum inference techniques for spectrum occupancy prediction with focus being on Artificial Neural Network (ANN) based schemes. First the statistical spectrum occupancy modeling methods are compared with machine learning based methods and described briefly the various neural networks in the purview of ANNs. The recent trend towards the use of hybrid neural network models has been detailed by acknowledging the contribution of research in this area. The various algorithms have been weighed over performance measure metrics, dimensions, type of data and measurement set-up employed. Finally the significant research been directed towards the use of deep learning techniques proven its efficiency in this field are also noted.