Blind equalization of QAM signals via extreme learning machine

In this paper, the framework of extreme learning machine (ELM) proposed for single hidden layer feedforward neural networks (SLFNs) suitable for blind equalization applications in quadrature amplitude modulation (QAM) transmission systems, together with the theoretical discussion, is presented. The QAM signals are equalized by the fully complex ELM network, and the input weight in this network generated by the QR decomposition of the receiving signal, instead of making it randomly selected. Additionally, support vector regression (SVR) which contains a Gaussian insensitive loss function is utilized to update and train the output weight. The corresponding results are verified by simulations to demonstrate the effectiveness and superiority of the proposed approach.