A parallel least-squares linear prediction method based on the circular lattice filter

In this correspondence, we propose a parallel least-squares method for linear prediction and estimation based on the circular lattice filter. Instead of minimizing the total sum of prediction errors, the method proceeds by dividing the sum into several parts and then minimizing each part in parallel, and then averaging the resulting estimates to form a final estimate. It is shown that this estimator is asymptotically efficient in the same way as the traditional least-squares estimator.