A comparative study between pre-whitened based sign algorithms in time-varying environments

As for LMS based adaptive algorithms, the family of pre-whitened sign algorithm is introduced in order to improve performances in case of correlated inputs. In particular, two adaptive algorithms excited at pre-whitened input direction are retained. They are the elementary pre-whitened sign algorithm (equipped with the error sign) and the stop and go sign algorithm (whose adaptation is carried in some iterations and stopped on others). A comparative study between the basic sign algorithms and these two variants in case of time-varying environments, including additive impulsive noise, jumping systems and Markovian non stationarities, is carried. Different criteria dealing with transient state, steady state and complexity are used to evaluate algorithms performances. The stop and go sign algorithm is shown to be the more attractive one.