On pre-whitened sign algorithms

This paper addresses the problem of improving convergence rate of sign algorithms. For such a purpose, the idea of decorrelating signals used to pilot the adaptive algorithm is investigated. More precisely, two major modifications are carried: both input and error signals are filtered and the adaptation process is carried out using two iterations. The novel algorithm is called the "Pre-whitened Sign Algorithm". To justify the idea, we develop an analytical formulation and we consider the particular case of the dual sign algorithm. We prove that the proposed algorithm has low complexity and provides a good convergence rate, with acceptable steady state performance. Simulation results are presented to support the theoretical analysis.