Jointly Optimized Extreme Learning Machine for Short-Term Prediction of Fading Channel

Abstract

Accurately evaluating channel state information (CSI) is an extremely important precondition for wireless communication to effectively obtain exact sending data. In order to fast obtain CSI, channel prediction tends to get its popularity in obtaining CSI as a result of the advantages of lightweight calculation and negligible feedback delay. In this paper, a jointly optimized extreme learning machine (JOELM) is proposed for the short-term prediction of fading channel. The JOELM scheme consists of two key steps: intelligent optimization and targeted repair. First, in order to gain high prediction accuracy, the firefly algorithm is imported to intelligently optimize the traditional extreme learning machine. Second, the Savitzky-Golay filter is innovatively adopted for reducing potential prediction errors. Extensive experiments about computational complexity, influences of repair coefficient and weight coefficient, contributing degree of two key steps, amplitude, transmission bit/symbol error rates, root-mean-square errors, and four typical statistical properties are given in final simulation section. The analyzed results indicate that the proposed JOELM can more accurately and efficiently deal with the short-term prediction of fading channel.