SVR Based Blind Signal Recovery for Convolutive MIMO Systems With High-Order QAM Signals

Abstract

This paper addresses the blind signal recovery for convolutive multiple-input multiple-output systems with high-order quadrature amplitude modulation (QAM) signals. First, a family of batch blind recovery algorithms is proposed. Concretely, they introduce the error function of multimodulus algorithm and the cross-correlation among different equalizer output vectors into the penalty term of the support vector regression (SVR) framework to recover all sources simultaneously. Then, the corresponding dual-mode blind recovery schemes are constructed to further decrease the interference. The new blind formulation through iterative re-weighted least square achieves low complexity optimization. The SVR framework, in essence, determines that the proposals perform better than the conventional methods in terms of data block size, total interference, and symbol error rate. Moreover, the excellent initialization provided by the first mode and the accurate error expression in the second mode ensure that the SVR-based dual-mode schemes work well with the high-order QAM signals. Finally, the efficiency of the proposals over the classical approaches is evaluated by simulations.