SVM-based blind beamforming of constant modulus signals

Recent work has shown how the support vector machine (SVM) framework can be used for blind equalization of constant modulus (CM) signals. The basic idea consists of exploiting the CM property of the input signals to reformulate the blind equalization problem as a regression problem. We extend this idea to encompass the problem of separating and estimating multiple CM signals mixed through an unknown matrix (i.e., blind beamforming). The quadratic inequalities derived from the CM property are transformed into linear ones, thus yielding a quadratic programming (QP) problem. Then an iterative reweighted procedure is proposed to blindly restore the CM property. Once a signal is recovered, its contribution to the original observations is removed and the iterative procedure can be applied again to extract another CM signal. Simulation results show that this SVM-based algorithm offers better performance than the algebraic constant modulus algorithm (ACMA), mainly when only a small number of snapshots is available.