Immuno-neural network for spectrum prediction

Abstract

The artificial neural network is an important machine learning algorithm for secondary users (SUs) in cognitive radio networks. An SU equipped with an ANN is able to perform predictive modelling using input samples acquired for a channel sensed to be idle This input samples are acquired incurring an input sample acquisition time (ISAT) that reduces the data transmission time (DTI) and throughput of SUs in the network The reduction of the IS AT is therefore important for enhanced throughput. This paper addresses this issue by proposing the addition of a Kullback Leibler divergence (KID) layer to the neural network based on inspiration from artificial immune systems theory. This layer computes the dissimilairites between previous and current inputs and reduces IS AT. We examine the performance of the neural network with the added KID layer in three scenarios that consider the achievable SU throughput. The throughput performance of SUs are examined for three scenarios for single and dual mode SUs equipped with ANN and dual mode SUs equipped with recurrent ANNs in the first, second and third scenarios respectively. Performance analysis shows that the addition of the KID layer improves the DTT and the SU throughput compared to existing scheme.