Generalized DLMS algorithm

It is well known that a pipelined architecture helps the efficient implementation of the LMS algorithm. However, this type of architecture requires delays in the processing. In the open literature, the LMS algorithm with a single delay has been extensively studied. In this work, different types of delays are introduced into the architecture of the LMS algorithm. The influence of these delays on the algorithm behavior is analyzed theoretically and by simulations. Results lead to a number of constraints on the delays. The convergence conditions and the stationary residual error have also been obtained.