Learning strategies in cognitive radio involving soft computing techniques

Abstract

Over the last era, the world of wireless communication has been undergoing tremendous changes. This has resulted in the advent of a whole range of innovative technologies such as Wi-Fi, Wi-Max, 802.22, wireless mesh networks, and software-defined radio. In wireless communication domain, with increasing demand for radio spectrum, management of the valuable but natural resource of radio spectrum is a Herculean task. The available static allocation techniques result in underutilized spectrum bands. To handle this problem, an efficient and emerging technology aimed toward dynamic spectrum allocation is Cognitive Radio. A Cognitive Radio (CR) can alter its communication process in line with its understanding of the context in which it is present. The intelligence and smartness of the Cognitive Radio is mainly due to the presence of cognitive engine. This cognitive engine primarily consists of knowledge base, reasoning block, and learning phase. As part of cognitive 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 and functional parameters of Cognitive Radio. The proposed work in this chapter contributes in this direction toward building predictive learning schemes based on soft computing techniques for Cognitive Radio. Predictive schemes toward forecast of key functional parameters of data rates and throughput are built. The different learning schemes used in the proposed work range from basic supervised algorithms like feedforward network, focused time delay neural networks, and recurrent networks to unsupervised algorithms based on self-organizing maps. Hybrid network of adaptive neuro-fuzzy inference is developed toward improvement of prediction accuracy. It has been found that ANFIS approaches have high prediction accuracy up to 97%. Subsequently self-organized map-based learning scheme is used to investigate improvement in design flexibility. It has been found that these learning schemes aid in adding more input parameters without altering the network design with prediction accuracy up to 85%.These learning schemes form useful inputs which result in improved cognitive engine, leading to enhanced dynamic spectrum allocation in Cognitive Radio. In the future, these algorithms are to be an integral part of cognitive engine in large scale, leading to intelligent spectrum management and allocation and hence a smart radio.