부산대학교 도서관

Knowledge of propagation channel conditions enables adaptive data transmission which improves the quality and efficiency of communication system. Wireless channels are characterised by highly dynamic time-varying nature. This means that information regarding propagation channel condition obtained by channel estimation can become outdated because of delay caused by processing and feedback phases. In fast fading environments, prediction of channel based on channel states in previous moments can ensure timely information. In this paper, a comparative performance analysis of an echo state network (ESN), an extreme learning machine (ELM) and least squares support vector machines (LS-SVM) for prediction of wireless channel conditions for single-input single-output (SISO) systems in microcellular and picocellular environments is carried out. Normalised mean squared error (NMSE) and time consumption are used as performance indicators. The experimental results on measured values for signal-to-noise ratio (SNR) show that all models have better and comparable prediction accuracy in microcell environment, while prediction framework based on the ESN outperforms the others in picocell environment.