Performance analysis of random neural networks in LTE-UL of a cognitive radio system

In cognitive radio networks (CRNs), the cognitive Engine (CE) is responsible for decision making. This is quite a challenging task as it requires finding the balance between prediction accuracy and efficient learning for optimal configuration settings for the CRN. Artificial neural networks (ANNs) have been widely used as predictive tools in cognitive radio. In this paper, random neural networks (RNNs) have been proposed to achieve better generalization and to speed up the cognition process in LTE (Long Term Evolution) cognitive-eNodeB. The developed CE is characterizing the achievable communication performance (throughput) of available configuration settings and suggesting the optimal radio parameters for specific service demand. Furthermore, the RNN-CE is coordinating the inter-cell-interference by suggesting the acceptable transmit power of adjacent channel users. Performance evaluation has revealed 42.85% better prediction accuracy (based on MSE) and 68% better learning efficiency (based on epochs required for convergent result) of RNN as compared to ANN.