Iterative Training for the LMS Algorithm with Extremely Short Pilot Sequences

摘要

We propose in this paper a new method of training, namely iterative training, for the least-mean-square (LMS) algorithm with an extremely small amount of training data. In this method, we apply the concept of iterative (turbo) processing into the training of the LMS algorithm. Here, the training needs only a short training data sequence and then uses it repeatedly to successively refine the mean-squared error (MSE) performance until the LMS algorithm converges. Simulation results show that the proposed training method gives the same MSE performance as the traditional training method of the LMS algorithm. In terms of the amount of training time, the proposed training method takes a much shorter time since it does not have to wait for an excessively long training data sequence to arrive. With the proposed training method, it is possible to use the LMS for real time applications where the training delay and the amount of training data are severely limited.