Least squares channel estimation for a channel with fast time variations

It is shown how to improve the tracking properties of least squares (LS) estimation for a time-variant channel estimation (system identification) problem by introducing a parametric time-variant model instead of the conventional model, which is a constant impulse response in each iteration. A recursive LS algorithm is developed for the case that the model varies linearly with time. Simulation results are given for examples where the unknown channel has linear, sinusoidal, and stochastic variations with time. A significant improvement of the tracking properties compared to LMS and conventional LS algorithms is demonstrated for cases with high signal to noise ratio and smooth variations of the unknown channel.<
>