On dertermining mixing parameter of CC-CMA algorithm by solving semi-algebraic sets
Abstract
The global convergence of a recently proposed constant modulus (CM) and cross-correlation (CC)-based algorithm (CC-CMA) is studied in this paper. We first show the original analysis of global convergence of CC-CMA is incorrect. We then point out that the global convergent analysis of gradient stochastic algorithms including CC-CMA could be completed by solving a semi-algebraic set. By developing an optimal algorithm to examine the roots distribution a semi-algebraic set related with CC-CMA, we present that CC-CMA can converge globally if the parameter which mix the CM and CC terms is properly selected. Since our approach is quite general, it can be extended to the convergence analysis of any gradient stochastic algorithm. Simulation results confirm the theoretical analysis on the conditions of mixing parameter.