A batch least squares lattice algorithm

A fast square root batch least squares algorithm for autoregressive model structures that requires only seven floating point operations per sample per estimated parameter is derived. Memory requirements, as well as the number of floating point operations, are of order n, where n is the model order. The method is based on estimation of the top block row of the QR transform of the data regression matrix. This is used to derive the parameters using an order-recursive lattice algorithm, after all samples have been processed