Combination and analysis of adaptive algorithms

• Grant number: 08/00773-1
• Support type: Regular Research Grants
• Duration: June 01, 2008 - May 31, 2010
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal researcher: Magno Teófilo Madeira da Silva
• Grantee: Magno Teófilo Madeira da Silva
• Home Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

The design and analysis of adaptive filters, which present a tradeoff among computational cost, convergence rate, steady-state solution, and stability, are important for several applications, in particular, for supervised or unsupervised (blind) channel equalization. Blind equalizers are employed in communication systemsto remove the interference introduced by dispersive channels without using a training sequence. The blind equalization algorithms havebeen widely studied in the last decades. However, many questions related to its theoretical analysis remain open problems in theliterature. In this research project, we will study theoretically the performance of blind equalization algorithms obtained from theconstant-modulus cost function. We intend to provide more accurate models for the steady-state mean-square error and/or for thestep-size ranges which ensure the algorithm stability. Moreover, in order to improve the performance of adaptive filters, we will studythe combinations of two or more filters operating in parallel. These combinations can be used to improve the choice of the fixed parameters, the tracking capability, the steady-state solution, the initialization, among other factors. Recently, one proved that the convex combination of adaptive algorithms is universal, i.e., using stationary signals, the combined estimate is, in the worst case, as good as the best filter-component at the steady-state. We intend to explore this property to choose combinations where it occurs more frequently. Initially, these combinations will be considered in blind equalization, and next the results can be extended to other applications. (AU)