This paper presents a new type of improved least-squares (ILS) algorithm fo
r adaptive parameter estimation of autoregressive (AR) signals from noisy o
bservations. Unlike the previous ILS based methods, the developed algorithm
can give consistent parameter estimates in a very direct manner that it do
es not involve dealing with an augmented noisy AR model. The new algorithm
is demonstrated to outperform the previous ILS based methods in terms of it
s improved numerical efficiency.