Comparative Performance Investigation of Supervised and Unsupervised Learning Outlines applied in Cognitive Radio Systems

Abstract

A Cognitive Radio is an intelligent wireless communication arrangement that is conscious of its surrounding environment, and has ability to learn from the situation and adapt its internal states to deviation in the incoming RF stimuli by making corresponding changes in certain operating parameters (e.g., transmit-power, carrier frequency, and modulation strategy) in real-time. The cognitive engine greatly contributes towards a smart radio. As part of the process, the radio observes, orients, takes decisions and evaluates the outcomes of decisions taken which is part of the learning phase. There are a variety of learning techniques enabling prediction of various operating parameters. This paper presents investigation of different learning algorithms travelling from supervised to unsupervised, applied to the problem of prediction of throughput and data rate and a comparison amongst them paving way towards predictive modeling. These learning algorithms can be incorporated into the cognitive cycle enabling spectrum allocation according to the learning done. It is shown that the unsupervised algorithms are most flexible, adaptable while supervised are more reliable and accurate. All algorithms are compared on basis of percentage of correct predictions and Root Mean Square Error (RMSE). In future, these algorithms are to be integral part of cognitive engine in large scale, leading to intelligent spectrum management and allocation and hence a smart radio.